brintos

brintos / linux-shallow public Read only

0
0
Text · 54.2 KiB · 12bbe48 Raw
2224 lines · c
1// SPDX-License-Identifier: GPL-2.0-only2/*3 * Copyright (C) 2019 Arrikto, Inc. All Rights Reserved.4 */5 6#include <linux/mm.h>7#include <linux/bio.h>8#include <linux/err.h>9#include <linux/hash.h>10#include <linux/list.h>11#include <linux/log2.h>12#include <linux/init.h>13#include <linux/slab.h>14#include <linux/wait.h>15#include <linux/dm-io.h>16#include <linux/mutex.h>17#include <linux/atomic.h>18#include <linux/bitops.h>19#include <linux/blkdev.h>20#include <linux/kdev_t.h>21#include <linux/kernel.h>22#include <linux/module.h>23#include <linux/jiffies.h>24#include <linux/mempool.h>25#include <linux/spinlock.h>26#include <linux/blk_types.h>27#include <linux/dm-kcopyd.h>28#include <linux/workqueue.h>29#include <linux/backing-dev.h>30#include <linux/device-mapper.h>31 32#include "dm.h"33#include "dm-clone-metadata.h"34 35#define DM_MSG_PREFIX "clone"36 37/*38 * Minimum and maximum allowed region sizes39 */40#define MIN_REGION_SIZE (1 << 3)  /* 4KB */41#define MAX_REGION_SIZE (1 << 21) /* 1GB */42 43#define MIN_HYDRATIONS 256 /* Size of hydration mempool */44#define DEFAULT_HYDRATION_THRESHOLD 1 /* 1 region */45#define DEFAULT_HYDRATION_BATCH_SIZE 1 /* Hydrate in batches of 1 region */46 47#define COMMIT_PERIOD HZ /* 1 sec */48 49/*50 * Hydration hash table size: 1 << HASH_TABLE_BITS51 */52#define HASH_TABLE_BITS 1553 54DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(clone_hydration_throttle,55	"A percentage of time allocated for hydrating regions");56 57/* Slab cache for struct dm_clone_region_hydration */58static struct kmem_cache *_hydration_cache;59 60/* dm-clone metadata modes */61enum clone_metadata_mode {62	CM_WRITE,		/* metadata may be changed */63	CM_READ_ONLY,		/* metadata may not be changed */64	CM_FAIL,		/* all metadata I/O fails */65};66 67struct hash_table_bucket;68 69struct clone {70	struct dm_target *ti;71 72	struct dm_dev *metadata_dev;73	struct dm_dev *dest_dev;74	struct dm_dev *source_dev;75 76	unsigned long nr_regions;77	sector_t region_size;78	unsigned int region_shift;79 80	/*81	 * A metadata commit and the actions taken in case it fails should run82	 * as a single atomic step.83	 */84	struct mutex commit_lock;85 86	struct dm_clone_metadata *cmd;87 88	/* Region hydration hash table */89	struct hash_table_bucket *ht;90 91	atomic_t ios_in_flight;92 93	wait_queue_head_t hydration_stopped;94 95	mempool_t hydration_pool;96 97	unsigned long last_commit_jiffies;98 99	/*100	 * We defer incoming WRITE bios for regions that are not hydrated,101	 * until after these regions have been hydrated.102	 *103	 * Also, we defer REQ_FUA and REQ_PREFLUSH bios, until after the104	 * metadata have been committed.105	 */106	spinlock_t lock;107	struct bio_list deferred_bios;108	struct bio_list deferred_discard_bios;109	struct bio_list deferred_flush_bios;110	struct bio_list deferred_flush_completions;111 112	/* Maximum number of regions being copied during background hydration. */113	unsigned int hydration_threshold;114 115	/* Number of regions to batch together during background hydration. */116	unsigned int hydration_batch_size;117 118	/* Which region to hydrate next */119	unsigned long hydration_offset;120 121	atomic_t hydrations_in_flight;122 123	/*124	 * Save a copy of the table line rather than reconstructing it for the125	 * status.126	 */127	unsigned int nr_ctr_args;128	const char **ctr_args;129 130	struct workqueue_struct *wq;131	struct work_struct worker;132	struct delayed_work waker;133 134	struct dm_kcopyd_client *kcopyd_client;135 136	enum clone_metadata_mode mode;137	unsigned long flags;138};139 140/*141 * dm-clone flags142 */143#define DM_CLONE_DISCARD_PASSDOWN 0144#define DM_CLONE_HYDRATION_ENABLED 1145#define DM_CLONE_HYDRATION_SUSPENDED 2146 147/*---------------------------------------------------------------------------*/148 149/*150 * Metadata failure handling.151 */152static enum clone_metadata_mode get_clone_mode(struct clone *clone)153{154	return READ_ONCE(clone->mode);155}156 157static const char *clone_device_name(struct clone *clone)158{159	return dm_table_device_name(clone->ti->table);160}161 162static void __set_clone_mode(struct clone *clone, enum clone_metadata_mode new_mode)163{164	static const char * const descs[] = {165		"read-write",166		"read-only",167		"fail"168	};169 170	enum clone_metadata_mode old_mode = get_clone_mode(clone);171 172	/* Never move out of fail mode */173	if (old_mode == CM_FAIL)174		new_mode = CM_FAIL;175 176	switch (new_mode) {177	case CM_FAIL:178	case CM_READ_ONLY:179		dm_clone_metadata_set_read_only(clone->cmd);180		break;181 182	case CM_WRITE:183		dm_clone_metadata_set_read_write(clone->cmd);184		break;185	}186 187	WRITE_ONCE(clone->mode, new_mode);188 189	if (new_mode != old_mode) {190		dm_table_event(clone->ti->table);191		DMINFO("%s: Switching to %s mode", clone_device_name(clone),192		       descs[(int)new_mode]);193	}194}195 196static void __abort_transaction(struct clone *clone)197{198	const char *dev_name = clone_device_name(clone);199 200	if (get_clone_mode(clone) >= CM_READ_ONLY)201		return;202 203	DMERR("%s: Aborting current metadata transaction", dev_name);204	if (dm_clone_metadata_abort(clone->cmd)) {205		DMERR("%s: Failed to abort metadata transaction", dev_name);206		__set_clone_mode(clone, CM_FAIL);207	}208}209 210static void __reload_in_core_bitset(struct clone *clone)211{212	const char *dev_name = clone_device_name(clone);213 214	if (get_clone_mode(clone) == CM_FAIL)215		return;216 217	/* Reload the on-disk bitset */218	DMINFO("%s: Reloading on-disk bitmap", dev_name);219	if (dm_clone_reload_in_core_bitset(clone->cmd)) {220		DMERR("%s: Failed to reload on-disk bitmap", dev_name);221		__set_clone_mode(clone, CM_FAIL);222	}223}224 225static void __metadata_operation_failed(struct clone *clone, const char *op, int r)226{227	DMERR("%s: Metadata operation `%s' failed: error = %d",228	      clone_device_name(clone), op, r);229 230	__abort_transaction(clone);231	__set_clone_mode(clone, CM_READ_ONLY);232 233	/*234	 * dm_clone_reload_in_core_bitset() may run concurrently with either235	 * dm_clone_set_region_hydrated() or dm_clone_cond_set_range(), but236	 * it's safe as we have already set the metadata to read-only mode.237	 */238	__reload_in_core_bitset(clone);239}240 241/*---------------------------------------------------------------------------*/242 243/* Wake up anyone waiting for region hydrations to stop */244static inline void wakeup_hydration_waiters(struct clone *clone)245{246	wake_up_all(&clone->hydration_stopped);247}248 249static inline void wake_worker(struct clone *clone)250{251	queue_work(clone->wq, &clone->worker);252}253 254/*---------------------------------------------------------------------------*/255 256/*257 * bio helper functions.258 */259static inline void remap_to_source(struct clone *clone, struct bio *bio)260{261	bio_set_dev(bio, clone->source_dev->bdev);262}263 264static inline void remap_to_dest(struct clone *clone, struct bio *bio)265{266	bio_set_dev(bio, clone->dest_dev->bdev);267}268 269static bool bio_triggers_commit(struct clone *clone, struct bio *bio)270{271	return op_is_flush(bio->bi_opf) &&272		dm_clone_changed_this_transaction(clone->cmd);273}274 275/* Get the address of the region in sectors */276static inline sector_t region_to_sector(struct clone *clone, unsigned long region_nr)277{278	return ((sector_t)region_nr << clone->region_shift);279}280 281/* Get the region number of the bio */282static inline unsigned long bio_to_region(struct clone *clone, struct bio *bio)283{284	return (bio->bi_iter.bi_sector >> clone->region_shift);285}286 287/* Get the region range covered by the bio */288static void bio_region_range(struct clone *clone, struct bio *bio,289			     unsigned long *rs, unsigned long *nr_regions)290{291	unsigned long end;292 293	*rs = dm_sector_div_up(bio->bi_iter.bi_sector, clone->region_size);294	end = bio_end_sector(bio) >> clone->region_shift;295 296	if (*rs >= end)297		*nr_regions = 0;298	else299		*nr_regions = end - *rs;300}301 302/* Check whether a bio overwrites a region */303static inline bool is_overwrite_bio(struct clone *clone, struct bio *bio)304{305	return (bio_data_dir(bio) == WRITE && bio_sectors(bio) == clone->region_size);306}307 308static void fail_bios(struct bio_list *bios, blk_status_t status)309{310	struct bio *bio;311 312	while ((bio = bio_list_pop(bios))) {313		bio->bi_status = status;314		bio_endio(bio);315	}316}317 318static void submit_bios(struct bio_list *bios)319{320	struct bio *bio;321	struct blk_plug plug;322 323	blk_start_plug(&plug);324 325	while ((bio = bio_list_pop(bios)))326		submit_bio_noacct(bio);327 328	blk_finish_plug(&plug);329}330 331/*332 * Submit bio to the underlying device.333 *334 * If the bio triggers a commit, delay it, until after the metadata have been335 * committed.336 *337 * NOTE: The bio remapping must be performed by the caller.338 */339static void issue_bio(struct clone *clone, struct bio *bio)340{341	if (!bio_triggers_commit(clone, bio)) {342		submit_bio_noacct(bio);343		return;344	}345 346	/*347	 * If the metadata mode is RO or FAIL we won't be able to commit the348	 * metadata, so we complete the bio with an error.349	 */350	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY)) {351		bio_io_error(bio);352		return;353	}354 355	/*356	 * Batch together any bios that trigger commits and then issue a single357	 * commit for them in process_deferred_flush_bios().358	 */359	spin_lock_irq(&clone->lock);360	bio_list_add(&clone->deferred_flush_bios, bio);361	spin_unlock_irq(&clone->lock);362 363	wake_worker(clone);364}365 366/*367 * Remap bio to the destination device and submit it.368 *369 * If the bio triggers a commit, delay it, until after the metadata have been370 * committed.371 */372static void remap_and_issue(struct clone *clone, struct bio *bio)373{374	remap_to_dest(clone, bio);375	issue_bio(clone, bio);376}377 378/*379 * Issue bios that have been deferred until after their region has finished380 * hydrating.381 *382 * We delegate the bio submission to the worker thread, so this is safe to call383 * from interrupt context.384 */385static void issue_deferred_bios(struct clone *clone, struct bio_list *bios)386{387	struct bio *bio;388	unsigned long flags;389	struct bio_list flush_bios = BIO_EMPTY_LIST;390	struct bio_list normal_bios = BIO_EMPTY_LIST;391 392	if (bio_list_empty(bios))393		return;394 395	while ((bio = bio_list_pop(bios))) {396		if (bio_triggers_commit(clone, bio))397			bio_list_add(&flush_bios, bio);398		else399			bio_list_add(&normal_bios, bio);400	}401 402	spin_lock_irqsave(&clone->lock, flags);403	bio_list_merge(&clone->deferred_bios, &normal_bios);404	bio_list_merge(&clone->deferred_flush_bios, &flush_bios);405	spin_unlock_irqrestore(&clone->lock, flags);406 407	wake_worker(clone);408}409 410static void complete_overwrite_bio(struct clone *clone, struct bio *bio)411{412	unsigned long flags;413 414	/*415	 * If the bio has the REQ_FUA flag set we must commit the metadata416	 * before signaling its completion.417	 *418	 * complete_overwrite_bio() is only called by hydration_complete(),419	 * after having successfully updated the metadata. This means we don't420	 * need to call dm_clone_changed_this_transaction() to check if the421	 * metadata has changed and thus we can avoid taking the metadata spin422	 * lock.423	 */424	if (!(bio->bi_opf & REQ_FUA)) {425		bio_endio(bio);426		return;427	}428 429	/*430	 * If the metadata mode is RO or FAIL we won't be able to commit the431	 * metadata, so we complete the bio with an error.432	 */433	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY)) {434		bio_io_error(bio);435		return;436	}437 438	/*439	 * Batch together any bios that trigger commits and then issue a single440	 * commit for them in process_deferred_flush_bios().441	 */442	spin_lock_irqsave(&clone->lock, flags);443	bio_list_add(&clone->deferred_flush_completions, bio);444	spin_unlock_irqrestore(&clone->lock, flags);445 446	wake_worker(clone);447}448 449static void trim_bio(struct bio *bio, sector_t sector, unsigned int len)450{451	bio->bi_iter.bi_sector = sector;452	bio->bi_iter.bi_size = to_bytes(len);453}454 455static void complete_discard_bio(struct clone *clone, struct bio *bio, bool success)456{457	unsigned long rs, nr_regions;458 459	/*460	 * If the destination device supports discards, remap and trim the461	 * discard bio and pass it down. Otherwise complete the bio462	 * immediately.463	 */464	if (test_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags) && success) {465		remap_to_dest(clone, bio);466		bio_region_range(clone, bio, &rs, &nr_regions);467		trim_bio(bio, region_to_sector(clone, rs),468			 nr_regions << clone->region_shift);469		submit_bio_noacct(bio);470	} else471		bio_endio(bio);472}473 474static void process_discard_bio(struct clone *clone, struct bio *bio)475{476	unsigned long rs, nr_regions;477 478	bio_region_range(clone, bio, &rs, &nr_regions);479	if (!nr_regions) {480		bio_endio(bio);481		return;482	}483 484	if (WARN_ON(rs >= clone->nr_regions || (rs + nr_regions) < rs ||485		    (rs + nr_regions) > clone->nr_regions)) {486		DMERR("%s: Invalid range (%lu + %lu, total regions %lu) for discard (%llu + %u)",487		      clone_device_name(clone), rs, nr_regions,488		      clone->nr_regions,489		      (unsigned long long)bio->bi_iter.bi_sector,490		      bio_sectors(bio));491		bio_endio(bio);492		return;493	}494 495	/*496	 * The covered regions are already hydrated so we just need to pass497	 * down the discard.498	 */499	if (dm_clone_is_range_hydrated(clone->cmd, rs, nr_regions)) {500		complete_discard_bio(clone, bio, true);501		return;502	}503 504	/*505	 * If the metadata mode is RO or FAIL we won't be able to update the506	 * metadata for the regions covered by the discard so we just ignore507	 * it.508	 */509	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY)) {510		bio_endio(bio);511		return;512	}513 514	/*515	 * Defer discard processing.516	 */517	spin_lock_irq(&clone->lock);518	bio_list_add(&clone->deferred_discard_bios, bio);519	spin_unlock_irq(&clone->lock);520 521	wake_worker(clone);522}523 524/*---------------------------------------------------------------------------*/525 526/*527 * dm-clone region hydrations.528 */529struct dm_clone_region_hydration {530	struct clone *clone;531	unsigned long region_nr;532 533	struct bio *overwrite_bio;534	bio_end_io_t *overwrite_bio_end_io;535 536	struct bio_list deferred_bios;537 538	blk_status_t status;539 540	/* Used by hydration batching */541	struct list_head list;542 543	/* Used by hydration hash table */544	struct hlist_node h;545};546 547/*548 * Hydration hash table implementation.549 *550 * Ideally we would like to use list_bl, which uses bit spin locks and employs551 * the least significant bit of the list head to lock the corresponding bucket,552 * reducing the memory overhead for the locks. But, currently, list_bl and bit553 * spin locks don't support IRQ safe versions. Since we have to take the lock554 * in both process and interrupt context, we must fall back to using regular555 * spin locks; one per hash table bucket.556 */557struct hash_table_bucket {558	struct hlist_head head;559 560	/* Spinlock protecting the bucket */561	spinlock_t lock;562};563 564#define bucket_lock_irqsave(bucket, flags) \565	spin_lock_irqsave(&(bucket)->lock, flags)566 567#define bucket_unlock_irqrestore(bucket, flags) \568	spin_unlock_irqrestore(&(bucket)->lock, flags)569 570#define bucket_lock_irq(bucket) \571	spin_lock_irq(&(bucket)->lock)572 573#define bucket_unlock_irq(bucket) \574	spin_unlock_irq(&(bucket)->lock)575 576static int hash_table_init(struct clone *clone)577{578	unsigned int i, sz;579	struct hash_table_bucket *bucket;580 581	sz = 1 << HASH_TABLE_BITS;582 583	clone->ht = kvmalloc_array(sz, sizeof(struct hash_table_bucket), GFP_KERNEL);584	if (!clone->ht)585		return -ENOMEM;586 587	for (i = 0; i < sz; i++) {588		bucket = clone->ht + i;589 590		INIT_HLIST_HEAD(&bucket->head);591		spin_lock_init(&bucket->lock);592	}593 594	return 0;595}596 597static void hash_table_exit(struct clone *clone)598{599	kvfree(clone->ht);600}601 602static struct hash_table_bucket *get_hash_table_bucket(struct clone *clone,603						       unsigned long region_nr)604{605	return &clone->ht[hash_long(region_nr, HASH_TABLE_BITS)];606}607 608/*609 * Search hash table for a hydration with hd->region_nr == region_nr610 *611 * NOTE: Must be called with the bucket lock held612 */613static struct dm_clone_region_hydration *__hash_find(struct hash_table_bucket *bucket,614						     unsigned long region_nr)615{616	struct dm_clone_region_hydration *hd;617 618	hlist_for_each_entry(hd, &bucket->head, h) {619		if (hd->region_nr == region_nr)620			return hd;621	}622 623	return NULL;624}625 626/*627 * Insert a hydration into the hash table.628 *629 * NOTE: Must be called with the bucket lock held.630 */631static inline void __insert_region_hydration(struct hash_table_bucket *bucket,632					     struct dm_clone_region_hydration *hd)633{634	hlist_add_head(&hd->h, &bucket->head);635}636 637/*638 * This function inserts a hydration into the hash table, unless someone else639 * managed to insert a hydration for the same region first. In the latter case640 * it returns the existing hydration descriptor for this region.641 *642 * NOTE: Must be called with the hydration hash table lock held.643 */644static struct dm_clone_region_hydration *645__find_or_insert_region_hydration(struct hash_table_bucket *bucket,646				  struct dm_clone_region_hydration *hd)647{648	struct dm_clone_region_hydration *hd2;649 650	hd2 = __hash_find(bucket, hd->region_nr);651	if (hd2)652		return hd2;653 654	__insert_region_hydration(bucket, hd);655 656	return hd;657}658 659/*---------------------------------------------------------------------------*/660 661/* Allocate a hydration */662static struct dm_clone_region_hydration *alloc_hydration(struct clone *clone)663{664	struct dm_clone_region_hydration *hd;665 666	/*667	 * Allocate a hydration from the hydration mempool.668	 * This might block but it can't fail.669	 */670	hd = mempool_alloc(&clone->hydration_pool, GFP_NOIO);671	hd->clone = clone;672 673	return hd;674}675 676static inline void free_hydration(struct dm_clone_region_hydration *hd)677{678	mempool_free(hd, &hd->clone->hydration_pool);679}680 681/* Initialize a hydration */682static void hydration_init(struct dm_clone_region_hydration *hd, unsigned long region_nr)683{684	hd->region_nr = region_nr;685	hd->overwrite_bio = NULL;686	bio_list_init(&hd->deferred_bios);687	hd->status = 0;688 689	INIT_LIST_HEAD(&hd->list);690	INIT_HLIST_NODE(&hd->h);691}692 693/*---------------------------------------------------------------------------*/694 695/*696 * Update dm-clone's metadata after a region has finished hydrating and remove697 * hydration from the hash table.698 */699static int hydration_update_metadata(struct dm_clone_region_hydration *hd)700{701	int r = 0;702	unsigned long flags;703	struct hash_table_bucket *bucket;704	struct clone *clone = hd->clone;705 706	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY))707		r = -EPERM;708 709	/* Update the metadata */710	if (likely(!r) && hd->status == BLK_STS_OK)711		r = dm_clone_set_region_hydrated(clone->cmd, hd->region_nr);712 713	bucket = get_hash_table_bucket(clone, hd->region_nr);714 715	/* Remove hydration from hash table */716	bucket_lock_irqsave(bucket, flags);717	hlist_del(&hd->h);718	bucket_unlock_irqrestore(bucket, flags);719 720	return r;721}722 723/*724 * Complete a region's hydration:725 *726 *	1. Update dm-clone's metadata.727 *	2. Remove hydration from hash table.728 *	3. Complete overwrite bio.729 *	4. Issue deferred bios.730 *	5. If this was the last hydration, wake up anyone waiting for731 *	   hydrations to finish.732 */733static void hydration_complete(struct dm_clone_region_hydration *hd)734{735	int r;736	blk_status_t status;737	struct clone *clone = hd->clone;738 739	r = hydration_update_metadata(hd);740 741	if (hd->status == BLK_STS_OK && likely(!r)) {742		if (hd->overwrite_bio)743			complete_overwrite_bio(clone, hd->overwrite_bio);744 745		issue_deferred_bios(clone, &hd->deferred_bios);746	} else {747		status = r ? BLK_STS_IOERR : hd->status;748 749		if (hd->overwrite_bio)750			bio_list_add(&hd->deferred_bios, hd->overwrite_bio);751 752		fail_bios(&hd->deferred_bios, status);753	}754 755	free_hydration(hd);756 757	if (atomic_dec_and_test(&clone->hydrations_in_flight))758		wakeup_hydration_waiters(clone);759}760 761static void hydration_kcopyd_callback(int read_err, unsigned long write_err, void *context)762{763	blk_status_t status;764 765	struct dm_clone_region_hydration *tmp, *hd = context;766	struct clone *clone = hd->clone;767 768	LIST_HEAD(batched_hydrations);769 770	if (read_err || write_err) {771		DMERR_LIMIT("%s: hydration failed", clone_device_name(clone));772		status = BLK_STS_IOERR;773	} else {774		status = BLK_STS_OK;775	}776	list_splice_tail(&hd->list, &batched_hydrations);777 778	hd->status = status;779	hydration_complete(hd);780 781	/* Complete batched hydrations */782	list_for_each_entry_safe(hd, tmp, &batched_hydrations, list) {783		hd->status = status;784		hydration_complete(hd);785	}786 787	/* Continue background hydration, if there is no I/O in-flight */788	if (test_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags) &&789	    !atomic_read(&clone->ios_in_flight))790		wake_worker(clone);791}792 793static void hydration_copy(struct dm_clone_region_hydration *hd, unsigned int nr_regions)794{795	unsigned long region_start, region_end;796	sector_t tail_size, region_size, total_size;797	struct dm_io_region from, to;798	struct clone *clone = hd->clone;799 800	if (WARN_ON(!nr_regions))801		return;802 803	region_size = clone->region_size;804	region_start = hd->region_nr;805	region_end = region_start + nr_regions - 1;806 807	total_size = region_to_sector(clone, nr_regions - 1);808 809	if (region_end == clone->nr_regions - 1) {810		/*811		 * The last region of the target might be smaller than812		 * region_size.813		 */814		tail_size = clone->ti->len & (region_size - 1);815		if (!tail_size)816			tail_size = region_size;817	} else {818		tail_size = region_size;819	}820 821	total_size += tail_size;822 823	from.bdev = clone->source_dev->bdev;824	from.sector = region_to_sector(clone, region_start);825	from.count = total_size;826 827	to.bdev = clone->dest_dev->bdev;828	to.sector = from.sector;829	to.count = from.count;830 831	/* Issue copy */832	atomic_add(nr_regions, &clone->hydrations_in_flight);833	dm_kcopyd_copy(clone->kcopyd_client, &from, 1, &to, 0,834		       hydration_kcopyd_callback, hd);835}836 837static void overwrite_endio(struct bio *bio)838{839	struct dm_clone_region_hydration *hd = bio->bi_private;840 841	bio->bi_end_io = hd->overwrite_bio_end_io;842	hd->status = bio->bi_status;843 844	hydration_complete(hd);845}846 847static void hydration_overwrite(struct dm_clone_region_hydration *hd, struct bio *bio)848{849	/*850	 * We don't need to save and restore bio->bi_private because device851	 * mapper core generates a new bio for us to use, with clean852	 * bi_private.853	 */854	hd->overwrite_bio = bio;855	hd->overwrite_bio_end_io = bio->bi_end_io;856 857	bio->bi_end_io = overwrite_endio;858	bio->bi_private = hd;859 860	atomic_inc(&hd->clone->hydrations_in_flight);861	submit_bio_noacct(bio);862}863 864/*865 * Hydrate bio's region.866 *867 * This function starts the hydration of the bio's region and puts the bio in868 * the list of deferred bios for this region. In case, by the time this869 * function is called, the region has finished hydrating it's submitted to the870 * destination device.871 *872 * NOTE: The bio remapping must be performed by the caller.873 */874static void hydrate_bio_region(struct clone *clone, struct bio *bio)875{876	unsigned long region_nr;877	struct hash_table_bucket *bucket;878	struct dm_clone_region_hydration *hd, *hd2;879 880	region_nr = bio_to_region(clone, bio);881	bucket = get_hash_table_bucket(clone, region_nr);882 883	bucket_lock_irq(bucket);884 885	hd = __hash_find(bucket, region_nr);886	if (hd) {887		/* Someone else is hydrating the region */888		bio_list_add(&hd->deferred_bios, bio);889		bucket_unlock_irq(bucket);890		return;891	}892 893	if (dm_clone_is_region_hydrated(clone->cmd, region_nr)) {894		/* The region has been hydrated */895		bucket_unlock_irq(bucket);896		issue_bio(clone, bio);897		return;898	}899 900	/*901	 * We must allocate a hydration descriptor and start the hydration of902	 * the corresponding region.903	 */904	bucket_unlock_irq(bucket);905 906	hd = alloc_hydration(clone);907	hydration_init(hd, region_nr);908 909	bucket_lock_irq(bucket);910 911	/* Check if the region has been hydrated in the meantime. */912	if (dm_clone_is_region_hydrated(clone->cmd, region_nr)) {913		bucket_unlock_irq(bucket);914		free_hydration(hd);915		issue_bio(clone, bio);916		return;917	}918 919	hd2 = __find_or_insert_region_hydration(bucket, hd);920	if (hd2 != hd) {921		/* Someone else started the region's hydration. */922		bio_list_add(&hd2->deferred_bios, bio);923		bucket_unlock_irq(bucket);924		free_hydration(hd);925		return;926	}927 928	/*929	 * If the metadata mode is RO or FAIL then there is no point starting a930	 * hydration, since we will not be able to update the metadata when the931	 * hydration finishes.932	 */933	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY)) {934		hlist_del(&hd->h);935		bucket_unlock_irq(bucket);936		free_hydration(hd);937		bio_io_error(bio);938		return;939	}940 941	/*942	 * Start region hydration.943	 *944	 * If a bio overwrites a region, i.e., its size is equal to the945	 * region's size, then we don't need to copy the region from the source946	 * to the destination device.947	 */948	if (is_overwrite_bio(clone, bio)) {949		bucket_unlock_irq(bucket);950		hydration_overwrite(hd, bio);951	} else {952		bio_list_add(&hd->deferred_bios, bio);953		bucket_unlock_irq(bucket);954		hydration_copy(hd, 1);955	}956}957 958/*---------------------------------------------------------------------------*/959 960/*961 * Background hydrations.962 */963 964/*965 * Batch region hydrations.966 *967 * To better utilize device bandwidth we batch together the hydration of968 * adjacent regions. This allows us to use small region sizes, e.g., 4KB, which969 * is good for small, random write performance (because of the overwriting of970 * un-hydrated regions) and at the same time issue big copy requests to kcopyd971 * to achieve high hydration bandwidth.972 */973struct batch_info {974	struct dm_clone_region_hydration *head;975	unsigned int nr_batched_regions;976};977 978static void __batch_hydration(struct batch_info *batch,979			      struct dm_clone_region_hydration *hd)980{981	struct clone *clone = hd->clone;982	unsigned int max_batch_size = READ_ONCE(clone->hydration_batch_size);983 984	if (batch->head) {985		/* Try to extend the current batch */986		if (batch->nr_batched_regions < max_batch_size &&987		    (batch->head->region_nr + batch->nr_batched_regions) == hd->region_nr) {988			list_add_tail(&hd->list, &batch->head->list);989			batch->nr_batched_regions++;990			hd = NULL;991		}992 993		/* Check if we should issue the current batch */994		if (batch->nr_batched_regions >= max_batch_size || hd) {995			hydration_copy(batch->head, batch->nr_batched_regions);996			batch->head = NULL;997			batch->nr_batched_regions = 0;998		}999	}1000 1001	if (!hd)1002		return;1003 1004	/* We treat max batch sizes of zero and one equivalently */1005	if (max_batch_size <= 1) {1006		hydration_copy(hd, 1);1007		return;1008	}1009 1010	/* Start a new batch */1011	BUG_ON(!list_empty(&hd->list));1012	batch->head = hd;1013	batch->nr_batched_regions = 1;1014}1015 1016static unsigned long __start_next_hydration(struct clone *clone,1017					    unsigned long offset,1018					    struct batch_info *batch)1019{1020	struct hash_table_bucket *bucket;1021	struct dm_clone_region_hydration *hd;1022	unsigned long nr_regions = clone->nr_regions;1023 1024	hd = alloc_hydration(clone);1025 1026	/* Try to find a region to hydrate. */1027	do {1028		offset = dm_clone_find_next_unhydrated_region(clone->cmd, offset);1029		if (offset == nr_regions)1030			break;1031 1032		bucket = get_hash_table_bucket(clone, offset);1033		bucket_lock_irq(bucket);1034 1035		if (!dm_clone_is_region_hydrated(clone->cmd, offset) &&1036		    !__hash_find(bucket, offset)) {1037			hydration_init(hd, offset);1038			__insert_region_hydration(bucket, hd);1039			bucket_unlock_irq(bucket);1040 1041			/* Batch hydration */1042			__batch_hydration(batch, hd);1043 1044			return (offset + 1);1045		}1046 1047		bucket_unlock_irq(bucket);1048 1049	} while (++offset < nr_regions);1050 1051	if (hd)1052		free_hydration(hd);1053 1054	return offset;1055}1056 1057/*1058 * This function searches for regions that still reside in the source device1059 * and starts their hydration.1060 */1061static void do_hydration(struct clone *clone)1062{1063	unsigned int current_volume;1064	unsigned long offset, nr_regions = clone->nr_regions;1065 1066	struct batch_info batch = {1067		.head = NULL,1068		.nr_batched_regions = 0,1069	};1070 1071	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY))1072		return;1073 1074	if (dm_clone_is_hydration_done(clone->cmd))1075		return;1076 1077	/*1078	 * Avoid race with device suspension.1079	 */1080	atomic_inc(&clone->hydrations_in_flight);1081 1082	/*1083	 * Make sure atomic_inc() is ordered before test_bit(), otherwise we1084	 * might race with clone_postsuspend() and start a region hydration1085	 * after the target has been suspended.1086	 *1087	 * This is paired with the smp_mb__after_atomic() in1088	 * clone_postsuspend().1089	 */1090	smp_mb__after_atomic();1091 1092	offset = clone->hydration_offset;1093	while (likely(!test_bit(DM_CLONE_HYDRATION_SUSPENDED, &clone->flags)) &&1094	       !atomic_read(&clone->ios_in_flight) &&1095	       test_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags) &&1096	       offset < nr_regions) {1097		current_volume = atomic_read(&clone->hydrations_in_flight);1098		current_volume += batch.nr_batched_regions;1099 1100		if (current_volume > READ_ONCE(clone->hydration_threshold))1101			break;1102 1103		offset = __start_next_hydration(clone, offset, &batch);1104	}1105 1106	if (batch.head)1107		hydration_copy(batch.head, batch.nr_batched_regions);1108 1109	if (offset >= nr_regions)1110		offset = 0;1111 1112	clone->hydration_offset = offset;1113 1114	if (atomic_dec_and_test(&clone->hydrations_in_flight))1115		wakeup_hydration_waiters(clone);1116}1117 1118/*---------------------------------------------------------------------------*/1119 1120static bool need_commit_due_to_time(struct clone *clone)1121{1122	return !time_in_range(jiffies, clone->last_commit_jiffies,1123			      clone->last_commit_jiffies + COMMIT_PERIOD);1124}1125 1126/*1127 * A non-zero return indicates read-only or fail mode.1128 */1129static int commit_metadata(struct clone *clone, bool *dest_dev_flushed)1130{1131	int r = 0;1132 1133	if (dest_dev_flushed)1134		*dest_dev_flushed = false;1135 1136	mutex_lock(&clone->commit_lock);1137 1138	if (!dm_clone_changed_this_transaction(clone->cmd))1139		goto out;1140 1141	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY)) {1142		r = -EPERM;1143		goto out;1144	}1145 1146	r = dm_clone_metadata_pre_commit(clone->cmd);1147	if (unlikely(r)) {1148		__metadata_operation_failed(clone, "dm_clone_metadata_pre_commit", r);1149		goto out;1150	}1151 1152	r = blkdev_issue_flush(clone->dest_dev->bdev);1153	if (unlikely(r)) {1154		__metadata_operation_failed(clone, "flush destination device", r);1155		goto out;1156	}1157 1158	if (dest_dev_flushed)1159		*dest_dev_flushed = true;1160 1161	r = dm_clone_metadata_commit(clone->cmd);1162	if (unlikely(r)) {1163		__metadata_operation_failed(clone, "dm_clone_metadata_commit", r);1164		goto out;1165	}1166 1167	if (dm_clone_is_hydration_done(clone->cmd))1168		dm_table_event(clone->ti->table);1169out:1170	mutex_unlock(&clone->commit_lock);1171 1172	return r;1173}1174 1175static void process_deferred_discards(struct clone *clone)1176{1177	int r = -EPERM;1178	struct bio *bio;1179	struct blk_plug plug;1180	unsigned long rs, nr_regions;1181	struct bio_list discards = BIO_EMPTY_LIST;1182 1183	spin_lock_irq(&clone->lock);1184	bio_list_merge_init(&discards, &clone->deferred_discard_bios);1185	spin_unlock_irq(&clone->lock);1186 1187	if (bio_list_empty(&discards))1188		return;1189 1190	if (unlikely(get_clone_mode(clone) >= CM_READ_ONLY))1191		goto out;1192 1193	/* Update the metadata */1194	bio_list_for_each(bio, &discards) {1195		bio_region_range(clone, bio, &rs, &nr_regions);1196		/*1197		 * A discard request might cover regions that have been already1198		 * hydrated. There is no need to update the metadata for these1199		 * regions.1200		 */1201		r = dm_clone_cond_set_range(clone->cmd, rs, nr_regions);1202		if (unlikely(r))1203			break;1204	}1205out:1206	blk_start_plug(&plug);1207	while ((bio = bio_list_pop(&discards)))1208		complete_discard_bio(clone, bio, r == 0);1209	blk_finish_plug(&plug);1210}1211 1212static void process_deferred_bios(struct clone *clone)1213{1214	struct bio_list bios = BIO_EMPTY_LIST;1215 1216	spin_lock_irq(&clone->lock);1217	bio_list_merge_init(&bios, &clone->deferred_bios);1218	spin_unlock_irq(&clone->lock);1219 1220	if (bio_list_empty(&bios))1221		return;1222 1223	submit_bios(&bios);1224}1225 1226static void process_deferred_flush_bios(struct clone *clone)1227{1228	struct bio *bio;1229	bool dest_dev_flushed;1230	struct bio_list bios = BIO_EMPTY_LIST;1231	struct bio_list bio_completions = BIO_EMPTY_LIST;1232 1233	/*1234	 * If there are any deferred flush bios, we must commit the metadata1235	 * before issuing them or signaling their completion.1236	 */1237	spin_lock_irq(&clone->lock);1238	bio_list_merge_init(&bios, &clone->deferred_flush_bios);1239	bio_list_merge_init(&bio_completions,1240			    &clone->deferred_flush_completions);1241	spin_unlock_irq(&clone->lock);1242 1243	if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&1244	    !(dm_clone_changed_this_transaction(clone->cmd) && need_commit_due_to_time(clone)))1245		return;1246 1247	if (commit_metadata(clone, &dest_dev_flushed)) {1248		bio_list_merge(&bios, &bio_completions);1249 1250		while ((bio = bio_list_pop(&bios)))1251			bio_io_error(bio);1252 1253		return;1254	}1255 1256	clone->last_commit_jiffies = jiffies;1257 1258	while ((bio = bio_list_pop(&bio_completions)))1259		bio_endio(bio);1260 1261	while ((bio = bio_list_pop(&bios))) {1262		if ((bio->bi_opf & REQ_PREFLUSH) && dest_dev_flushed) {1263			/* We just flushed the destination device as part of1264			 * the metadata commit, so there is no reason to send1265			 * another flush.1266			 */1267			bio_endio(bio);1268		} else {1269			submit_bio_noacct(bio);1270		}1271	}1272}1273 1274static void do_worker(struct work_struct *work)1275{1276	struct clone *clone = container_of(work, typeof(*clone), worker);1277 1278	process_deferred_bios(clone);1279	process_deferred_discards(clone);1280 1281	/*1282	 * process_deferred_flush_bios():1283	 *1284	 *   - Commit metadata1285	 *1286	 *   - Process deferred REQ_FUA completions1287	 *1288	 *   - Process deferred REQ_PREFLUSH bios1289	 */1290	process_deferred_flush_bios(clone);1291 1292	/* Background hydration */1293	do_hydration(clone);1294}1295 1296/*1297 * Commit periodically so that not too much unwritten data builds up.1298 *1299 * Also, restart background hydration, if it has been stopped by in-flight I/O.1300 */1301static void do_waker(struct work_struct *work)1302{1303	struct clone *clone = container_of(to_delayed_work(work), struct clone, waker);1304 1305	wake_worker(clone);1306	queue_delayed_work(clone->wq, &clone->waker, COMMIT_PERIOD);1307}1308 1309/*---------------------------------------------------------------------------*/1310 1311/*1312 * Target methods1313 */1314static int clone_map(struct dm_target *ti, struct bio *bio)1315{1316	struct clone *clone = ti->private;1317	unsigned long region_nr;1318 1319	atomic_inc(&clone->ios_in_flight);1320 1321	if (unlikely(get_clone_mode(clone) == CM_FAIL))1322		return DM_MAPIO_KILL;1323 1324	/*1325	 * REQ_PREFLUSH bios carry no data:1326	 *1327	 * - Commit metadata, if changed1328	 *1329	 * - Pass down to destination device1330	 */1331	if (bio->bi_opf & REQ_PREFLUSH) {1332		remap_and_issue(clone, bio);1333		return DM_MAPIO_SUBMITTED;1334	}1335 1336	bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);1337 1338	/*1339	 * dm-clone interprets discards and performs a fast hydration of the1340	 * discarded regions, i.e., we skip the copy from the source device and1341	 * just mark the regions as hydrated.1342	 */1343	if (bio_op(bio) == REQ_OP_DISCARD) {1344		process_discard_bio(clone, bio);1345		return DM_MAPIO_SUBMITTED;1346	}1347 1348	/*1349	 * If the bio's region is hydrated, redirect it to the destination1350	 * device.1351	 *1352	 * If the region is not hydrated and the bio is a READ, redirect it to1353	 * the source device.1354	 *1355	 * Else, defer WRITE bio until after its region has been hydrated and1356	 * start the region's hydration immediately.1357	 */1358	region_nr = bio_to_region(clone, bio);1359	if (dm_clone_is_region_hydrated(clone->cmd, region_nr)) {1360		remap_and_issue(clone, bio);1361		return DM_MAPIO_SUBMITTED;1362	} else if (bio_data_dir(bio) == READ) {1363		remap_to_source(clone, bio);1364		return DM_MAPIO_REMAPPED;1365	}1366 1367	remap_to_dest(clone, bio);1368	hydrate_bio_region(clone, bio);1369 1370	return DM_MAPIO_SUBMITTED;1371}1372 1373static int clone_endio(struct dm_target *ti, struct bio *bio, blk_status_t *error)1374{1375	struct clone *clone = ti->private;1376 1377	atomic_dec(&clone->ios_in_flight);1378 1379	return DM_ENDIO_DONE;1380}1381 1382static void emit_flags(struct clone *clone, char *result, unsigned int maxlen,1383		       ssize_t *sz_ptr)1384{1385	ssize_t sz = *sz_ptr;1386	unsigned int count;1387 1388	count = !test_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags);1389	count += !test_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags);1390 1391	DMEMIT("%u ", count);1392 1393	if (!test_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags))1394		DMEMIT("no_hydration ");1395 1396	if (!test_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags))1397		DMEMIT("no_discard_passdown ");1398 1399	*sz_ptr = sz;1400}1401 1402static void emit_core_args(struct clone *clone, char *result,1403			   unsigned int maxlen, ssize_t *sz_ptr)1404{1405	ssize_t sz = *sz_ptr;1406	unsigned int count = 4;1407 1408	DMEMIT("%u hydration_threshold %u hydration_batch_size %u ", count,1409	       READ_ONCE(clone->hydration_threshold),1410	       READ_ONCE(clone->hydration_batch_size));1411 1412	*sz_ptr = sz;1413}1414 1415/*1416 * Status format:1417 *1418 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>1419 * <clone region size> <#hydrated regions>/<#total regions> <#hydrating regions>1420 * <#features> <features>* <#core args> <core args>* <clone metadata mode>1421 */1422static void clone_status(struct dm_target *ti, status_type_t type,1423			 unsigned int status_flags, char *result,1424			 unsigned int maxlen)1425{1426	int r;1427	unsigned int i;1428	ssize_t sz = 0;1429	dm_block_t nr_free_metadata_blocks = 0;1430	dm_block_t nr_metadata_blocks = 0;1431	char buf[BDEVNAME_SIZE];1432	struct clone *clone = ti->private;1433 1434	switch (type) {1435	case STATUSTYPE_INFO:1436		if (get_clone_mode(clone) == CM_FAIL) {1437			DMEMIT("Fail");1438			break;1439		}1440 1441		/* Commit to ensure statistics aren't out-of-date */1442		if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))1443			(void) commit_metadata(clone, NULL);1444 1445		r = dm_clone_get_free_metadata_block_count(clone->cmd, &nr_free_metadata_blocks);1446 1447		if (r) {1448			DMERR("%s: dm_clone_get_free_metadata_block_count returned %d",1449			      clone_device_name(clone), r);1450			goto error;1451		}1452 1453		r = dm_clone_get_metadata_dev_size(clone->cmd, &nr_metadata_blocks);1454 1455		if (r) {1456			DMERR("%s: dm_clone_get_metadata_dev_size returned %d",1457			      clone_device_name(clone), r);1458			goto error;1459		}1460 1461		DMEMIT("%u %llu/%llu %llu %u/%lu %u ",1462		       DM_CLONE_METADATA_BLOCK_SIZE,1463		       (unsigned long long)(nr_metadata_blocks - nr_free_metadata_blocks),1464		       (unsigned long long)nr_metadata_blocks,1465		       (unsigned long long)clone->region_size,1466		       dm_clone_nr_of_hydrated_regions(clone->cmd),1467		       clone->nr_regions,1468		       atomic_read(&clone->hydrations_in_flight));1469 1470		emit_flags(clone, result, maxlen, &sz);1471		emit_core_args(clone, result, maxlen, &sz);1472 1473		switch (get_clone_mode(clone)) {1474		case CM_WRITE:1475			DMEMIT("rw");1476			break;1477		case CM_READ_ONLY:1478			DMEMIT("ro");1479			break;1480		case CM_FAIL:1481			DMEMIT("Fail");1482		}1483 1484		break;1485 1486	case STATUSTYPE_TABLE:1487		format_dev_t(buf, clone->metadata_dev->bdev->bd_dev);1488		DMEMIT("%s ", buf);1489 1490		format_dev_t(buf, clone->dest_dev->bdev->bd_dev);1491		DMEMIT("%s ", buf);1492 1493		format_dev_t(buf, clone->source_dev->bdev->bd_dev);1494		DMEMIT("%s", buf);1495 1496		for (i = 0; i < clone->nr_ctr_args; i++)1497			DMEMIT(" %s", clone->ctr_args[i]);1498		break;1499 1500	case STATUSTYPE_IMA:1501		*result = '\0';1502		break;1503	}1504 1505	return;1506 1507error:1508	DMEMIT("Error");1509}1510 1511static sector_t get_dev_size(struct dm_dev *dev)1512{1513	return bdev_nr_sectors(dev->bdev);1514}1515 1516/*---------------------------------------------------------------------------*/1517 1518/*1519 * Construct a clone device mapping:1520 *1521 * clone <metadata dev> <destination dev> <source dev> <region size>1522 *	[<#feature args> [<feature arg>]* [<#core args> [key value]*]]1523 *1524 * metadata dev: Fast device holding the persistent metadata1525 * destination dev: The destination device, which will become a clone of the1526 *                  source device1527 * source dev: The read-only source device that gets cloned1528 * region size: dm-clone unit size in sectors1529 *1530 * #feature args: Number of feature arguments passed1531 * feature args: E.g. no_hydration, no_discard_passdown1532 *1533 * #core arguments: An even number of core arguments1534 * core arguments: Key/value pairs for tuning the core1535 *		   E.g. 'hydration_threshold 256'1536 */1537static int parse_feature_args(struct dm_arg_set *as, struct clone *clone)1538{1539	int r;1540	unsigned int argc;1541	const char *arg_name;1542	struct dm_target *ti = clone->ti;1543 1544	const struct dm_arg args = {1545		.min = 0,1546		.max = 2,1547		.error = "Invalid number of feature arguments"1548	};1549 1550	/* No feature arguments supplied */1551	if (!as->argc)1552		return 0;1553 1554	r = dm_read_arg_group(&args, as, &argc, &ti->error);1555	if (r)1556		return r;1557 1558	while (argc) {1559		arg_name = dm_shift_arg(as);1560		argc--;1561 1562		if (!strcasecmp(arg_name, "no_hydration")) {1563			__clear_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags);1564		} else if (!strcasecmp(arg_name, "no_discard_passdown")) {1565			__clear_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags);1566		} else {1567			ti->error = "Invalid feature argument";1568			return -EINVAL;1569		}1570	}1571 1572	return 0;1573}1574 1575static int parse_core_args(struct dm_arg_set *as, struct clone *clone)1576{1577	int r;1578	unsigned int argc;1579	unsigned int value;1580	const char *arg_name;1581	struct dm_target *ti = clone->ti;1582 1583	const struct dm_arg args = {1584		.min = 0,1585		.max = 4,1586		.error = "Invalid number of core arguments"1587	};1588 1589	/* Initialize core arguments */1590	clone->hydration_batch_size = DEFAULT_HYDRATION_BATCH_SIZE;1591	clone->hydration_threshold = DEFAULT_HYDRATION_THRESHOLD;1592 1593	/* No core arguments supplied */1594	if (!as->argc)1595		return 0;1596 1597	r = dm_read_arg_group(&args, as, &argc, &ti->error);1598	if (r)1599		return r;1600 1601	if (argc & 1) {1602		ti->error = "Number of core arguments must be even";1603		return -EINVAL;1604	}1605 1606	while (argc) {1607		arg_name = dm_shift_arg(as);1608		argc -= 2;1609 1610		if (!strcasecmp(arg_name, "hydration_threshold")) {1611			if (kstrtouint(dm_shift_arg(as), 10, &value)) {1612				ti->error = "Invalid value for argument `hydration_threshold'";1613				return -EINVAL;1614			}1615			clone->hydration_threshold = value;1616		} else if (!strcasecmp(arg_name, "hydration_batch_size")) {1617			if (kstrtouint(dm_shift_arg(as), 10, &value)) {1618				ti->error = "Invalid value for argument `hydration_batch_size'";1619				return -EINVAL;1620			}1621			clone->hydration_batch_size = value;1622		} else {1623			ti->error = "Invalid core argument";1624			return -EINVAL;1625		}1626	}1627 1628	return 0;1629}1630 1631static int parse_region_size(struct clone *clone, struct dm_arg_set *as, char **error)1632{1633	int r;1634	unsigned int region_size;1635	struct dm_arg arg;1636 1637	arg.min = MIN_REGION_SIZE;1638	arg.max = MAX_REGION_SIZE;1639	arg.error = "Invalid region size";1640 1641	r = dm_read_arg(&arg, as, &region_size, error);1642	if (r)1643		return r;1644 1645	/* Check region size is a power of 2 */1646	if (!is_power_of_2(region_size)) {1647		*error = "Region size is not a power of 2";1648		return -EINVAL;1649	}1650 1651	/* Validate the region size against the device logical block size */1652	if (region_size % (bdev_logical_block_size(clone->source_dev->bdev) >> 9) ||1653	    region_size % (bdev_logical_block_size(clone->dest_dev->bdev) >> 9)) {1654		*error = "Region size is not a multiple of device logical block size";1655		return -EINVAL;1656	}1657 1658	clone->region_size = region_size;1659 1660	return 0;1661}1662 1663static int validate_nr_regions(unsigned long n, char **error)1664{1665	/*1666	 * dm_bitset restricts us to 2^32 regions. test_bit & co. restrict us1667	 * further to 2^31 regions.1668	 */1669	if (n > (1UL << 31)) {1670		*error = "Too many regions. Consider increasing the region size";1671		return -EINVAL;1672	}1673 1674	return 0;1675}1676 1677static int parse_metadata_dev(struct clone *clone, struct dm_arg_set *as, char **error)1678{1679	int r;1680	sector_t metadata_dev_size;1681 1682	r = dm_get_device(clone->ti, dm_shift_arg(as),1683			  BLK_OPEN_READ | BLK_OPEN_WRITE, &clone->metadata_dev);1684	if (r) {1685		*error = "Error opening metadata device";1686		return r;1687	}1688 1689	metadata_dev_size = get_dev_size(clone->metadata_dev);1690	if (metadata_dev_size > DM_CLONE_METADATA_MAX_SECTORS_WARNING)1691		DMWARN("Metadata device %pg is larger than %u sectors: excess space will not be used.",1692		       clone->metadata_dev->bdev, DM_CLONE_METADATA_MAX_SECTORS);1693 1694	return 0;1695}1696 1697static int parse_dest_dev(struct clone *clone, struct dm_arg_set *as, char **error)1698{1699	int r;1700	sector_t dest_dev_size;1701 1702	r = dm_get_device(clone->ti, dm_shift_arg(as),1703			  BLK_OPEN_READ | BLK_OPEN_WRITE, &clone->dest_dev);1704	if (r) {1705		*error = "Error opening destination device";1706		return r;1707	}1708 1709	dest_dev_size = get_dev_size(clone->dest_dev);1710	if (dest_dev_size < clone->ti->len) {1711		dm_put_device(clone->ti, clone->dest_dev);1712		*error = "Device size larger than destination device";1713		return -EINVAL;1714	}1715 1716	return 0;1717}1718 1719static int parse_source_dev(struct clone *clone, struct dm_arg_set *as, char **error)1720{1721	int r;1722	sector_t source_dev_size;1723 1724	r = dm_get_device(clone->ti, dm_shift_arg(as), BLK_OPEN_READ,1725			  &clone->source_dev);1726	if (r) {1727		*error = "Error opening source device";1728		return r;1729	}1730 1731	source_dev_size = get_dev_size(clone->source_dev);1732	if (source_dev_size < clone->ti->len) {1733		dm_put_device(clone->ti, clone->source_dev);1734		*error = "Device size larger than source device";1735		return -EINVAL;1736	}1737 1738	return 0;1739}1740 1741static int copy_ctr_args(struct clone *clone, int argc, const char **argv, char **error)1742{1743	unsigned int i;1744	const char **copy;1745 1746	copy = kcalloc(argc, sizeof(*copy), GFP_KERNEL);1747	if (!copy)1748		goto error;1749 1750	for (i = 0; i < argc; i++) {1751		copy[i] = kstrdup(argv[i], GFP_KERNEL);1752 1753		if (!copy[i]) {1754			while (i--)1755				kfree(copy[i]);1756			kfree(copy);1757			goto error;1758		}1759	}1760 1761	clone->nr_ctr_args = argc;1762	clone->ctr_args = copy;1763	return 0;1764 1765error:1766	*error = "Failed to allocate memory for table line";1767	return -ENOMEM;1768}1769 1770static int clone_ctr(struct dm_target *ti, unsigned int argc, char **argv)1771{1772	int r;1773	sector_t nr_regions;1774	struct clone *clone;1775	struct dm_arg_set as;1776 1777	if (argc < 4) {1778		ti->error = "Invalid number of arguments";1779		return -EINVAL;1780	}1781 1782	as.argc = argc;1783	as.argv = argv;1784 1785	clone = kzalloc(sizeof(*clone), GFP_KERNEL);1786	if (!clone) {1787		ti->error = "Failed to allocate clone structure";1788		return -ENOMEM;1789	}1790 1791	clone->ti = ti;1792 1793	/* Initialize dm-clone flags */1794	__set_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags);1795	__set_bit(DM_CLONE_HYDRATION_SUSPENDED, &clone->flags);1796	__set_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags);1797 1798	r = parse_metadata_dev(clone, &as, &ti->error);1799	if (r)1800		goto out_with_clone;1801 1802	r = parse_dest_dev(clone, &as, &ti->error);1803	if (r)1804		goto out_with_meta_dev;1805 1806	r = parse_source_dev(clone, &as, &ti->error);1807	if (r)1808		goto out_with_dest_dev;1809 1810	r = parse_region_size(clone, &as, &ti->error);1811	if (r)1812		goto out_with_source_dev;1813 1814	clone->region_shift = __ffs(clone->region_size);1815	nr_regions = dm_sector_div_up(ti->len, clone->region_size);1816 1817	/* Check for overflow */1818	if (nr_regions != (unsigned long)nr_regions) {1819		ti->error = "Too many regions. Consider increasing the region size";1820		r = -EOVERFLOW;1821		goto out_with_source_dev;1822	}1823 1824	clone->nr_regions = nr_regions;1825 1826	r = validate_nr_regions(clone->nr_regions, &ti->error);1827	if (r)1828		goto out_with_source_dev;1829 1830	r = dm_set_target_max_io_len(ti, clone->region_size);1831	if (r) {1832		ti->error = "Failed to set max io len";1833		goto out_with_source_dev;1834	}1835 1836	r = parse_feature_args(&as, clone);1837	if (r)1838		goto out_with_source_dev;1839 1840	r = parse_core_args(&as, clone);1841	if (r)1842		goto out_with_source_dev;1843 1844	/* Load metadata */1845	clone->cmd = dm_clone_metadata_open(clone->metadata_dev->bdev, ti->len,1846					    clone->region_size);1847	if (IS_ERR(clone->cmd)) {1848		ti->error = "Failed to load metadata";1849		r = PTR_ERR(clone->cmd);1850		goto out_with_source_dev;1851	}1852 1853	__set_clone_mode(clone, CM_WRITE);1854 1855	if (get_clone_mode(clone) != CM_WRITE) {1856		ti->error = "Unable to get write access to metadata, please check/repair metadata";1857		r = -EPERM;1858		goto out_with_metadata;1859	}1860 1861	clone->last_commit_jiffies = jiffies;1862 1863	/* Allocate hydration hash table */1864	r = hash_table_init(clone);1865	if (r) {1866		ti->error = "Failed to allocate hydration hash table";1867		goto out_with_metadata;1868	}1869 1870	atomic_set(&clone->ios_in_flight, 0);1871	init_waitqueue_head(&clone->hydration_stopped);1872	spin_lock_init(&clone->lock);1873	bio_list_init(&clone->deferred_bios);1874	bio_list_init(&clone->deferred_discard_bios);1875	bio_list_init(&clone->deferred_flush_bios);1876	bio_list_init(&clone->deferred_flush_completions);1877	clone->hydration_offset = 0;1878	atomic_set(&clone->hydrations_in_flight, 0);1879 1880	clone->wq = alloc_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM, 0);1881	if (!clone->wq) {1882		ti->error = "Failed to allocate workqueue";1883		r = -ENOMEM;1884		goto out_with_ht;1885	}1886 1887	INIT_WORK(&clone->worker, do_worker);1888	INIT_DELAYED_WORK(&clone->waker, do_waker);1889 1890	clone->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);1891	if (IS_ERR(clone->kcopyd_client)) {1892		r = PTR_ERR(clone->kcopyd_client);1893		goto out_with_wq;1894	}1895 1896	r = mempool_init_slab_pool(&clone->hydration_pool, MIN_HYDRATIONS,1897				   _hydration_cache);1898	if (r) {1899		ti->error = "Failed to create dm_clone_region_hydration memory pool";1900		goto out_with_kcopyd;1901	}1902 1903	/* Save a copy of the table line */1904	r = copy_ctr_args(clone, argc - 3, (const char **)argv + 3, &ti->error);1905	if (r)1906		goto out_with_mempool;1907 1908	mutex_init(&clone->commit_lock);1909 1910	/* Enable flushes */1911	ti->num_flush_bios = 1;1912	ti->flush_supported = true;1913 1914	/* Enable discards */1915	ti->discards_supported = true;1916	ti->num_discard_bios = 1;1917 1918	ti->private = clone;1919 1920	return 0;1921 1922out_with_mempool:1923	mempool_exit(&clone->hydration_pool);1924out_with_kcopyd:1925	dm_kcopyd_client_destroy(clone->kcopyd_client);1926out_with_wq:1927	destroy_workqueue(clone->wq);1928out_with_ht:1929	hash_table_exit(clone);1930out_with_metadata:1931	dm_clone_metadata_close(clone->cmd);1932out_with_source_dev:1933	dm_put_device(ti, clone->source_dev);1934out_with_dest_dev:1935	dm_put_device(ti, clone->dest_dev);1936out_with_meta_dev:1937	dm_put_device(ti, clone->metadata_dev);1938out_with_clone:1939	kfree(clone);1940 1941	return r;1942}1943 1944static void clone_dtr(struct dm_target *ti)1945{1946	unsigned int i;1947	struct clone *clone = ti->private;1948 1949	mutex_destroy(&clone->commit_lock);1950 1951	for (i = 0; i < clone->nr_ctr_args; i++)1952		kfree(clone->ctr_args[i]);1953	kfree(clone->ctr_args);1954 1955	mempool_exit(&clone->hydration_pool);1956	dm_kcopyd_client_destroy(clone->kcopyd_client);1957	cancel_delayed_work_sync(&clone->waker);1958	destroy_workqueue(clone->wq);1959	hash_table_exit(clone);1960	dm_clone_metadata_close(clone->cmd);1961	dm_put_device(ti, clone->source_dev);1962	dm_put_device(ti, clone->dest_dev);1963	dm_put_device(ti, clone->metadata_dev);1964 1965	kfree(clone);1966}1967 1968/*---------------------------------------------------------------------------*/1969 1970static void clone_postsuspend(struct dm_target *ti)1971{1972	struct clone *clone = ti->private;1973 1974	/*1975	 * To successfully suspend the device:1976	 *1977	 *	- We cancel the delayed work for periodic commits and wait for1978	 *	  it to finish.1979	 *1980	 *	- We stop the background hydration, i.e. we prevent new region1981	 *	  hydrations from starting.1982	 *1983	 *	- We wait for any in-flight hydrations to finish.1984	 *1985	 *	- We flush the workqueue.1986	 *1987	 *	- We commit the metadata.1988	 */1989	cancel_delayed_work_sync(&clone->waker);1990 1991	set_bit(DM_CLONE_HYDRATION_SUSPENDED, &clone->flags);1992 1993	/*1994	 * Make sure set_bit() is ordered before atomic_read(), otherwise we1995	 * might race with do_hydration() and miss some started region1996	 * hydrations.1997	 *1998	 * This is paired with smp_mb__after_atomic() in do_hydration().1999	 */2000	smp_mb__after_atomic();2001 2002	wait_event(clone->hydration_stopped, !atomic_read(&clone->hydrations_in_flight));2003	flush_workqueue(clone->wq);2004 2005	(void) commit_metadata(clone, NULL);2006}2007 2008static void clone_resume(struct dm_target *ti)2009{2010	struct clone *clone = ti->private;2011 2012	clear_bit(DM_CLONE_HYDRATION_SUSPENDED, &clone->flags);2013	do_waker(&clone->waker.work);2014}2015 2016/*2017 * If discard_passdown was enabled verify that the destination device supports2018 * discards. Disable discard_passdown if not.2019 */2020static void disable_passdown_if_not_supported(struct clone *clone)2021{2022	struct block_device *dest_dev = clone->dest_dev->bdev;2023	struct queue_limits *dest_limits = &bdev_get_queue(dest_dev)->limits;2024	const char *reason = NULL;2025 2026	if (!test_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags))2027		return;2028 2029	if (!bdev_max_discard_sectors(dest_dev))2030		reason = "discard unsupported";2031	else if (dest_limits->max_discard_sectors < clone->region_size)2032		reason = "max discard sectors smaller than a region";2033 2034	if (reason) {2035		DMWARN("Destination device (%pg) %s: Disabling discard passdown.",2036		       dest_dev, reason);2037		clear_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags);2038	}2039}2040 2041static void set_discard_limits(struct clone *clone, struct queue_limits *limits)2042{2043	struct block_device *dest_bdev = clone->dest_dev->bdev;2044	struct queue_limits *dest_limits = &bdev_get_queue(dest_bdev)->limits;2045 2046	if (!test_bit(DM_CLONE_DISCARD_PASSDOWN, &clone->flags)) {2047		/* No passdown is done so we set our own virtual limits */2048		limits->discard_granularity = clone->region_size << SECTOR_SHIFT;2049		limits->max_hw_discard_sectors = round_down(UINT_MAX >> SECTOR_SHIFT,2050							    clone->region_size);2051		return;2052	}2053 2054	/*2055	 * clone_iterate_devices() is stacking both the source and destination2056	 * device limits but discards aren't passed to the source device, so2057	 * inherit destination's limits.2058	 */2059	limits->max_hw_discard_sectors = dest_limits->max_hw_discard_sectors;2060	limits->discard_granularity = dest_limits->discard_granularity;2061	limits->discard_alignment = dest_limits->discard_alignment;2062	limits->max_discard_segments = dest_limits->max_discard_segments;2063}2064 2065static void clone_io_hints(struct dm_target *ti, struct queue_limits *limits)2066{2067	struct clone *clone = ti->private;2068	u64 io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;2069 2070	/*2071	 * If the system-determined stacked limits are compatible with2072	 * dm-clone's region size (io_opt is a factor) do not override them.2073	 */2074	if (io_opt_sectors < clone->region_size ||2075	    do_div(io_opt_sectors, clone->region_size)) {2076		limits->io_min = clone->region_size << SECTOR_SHIFT;2077		limits->io_opt = clone->region_size << SECTOR_SHIFT;2078	}2079 2080	disable_passdown_if_not_supported(clone);2081	set_discard_limits(clone, limits);2082}2083 2084static int clone_iterate_devices(struct dm_target *ti,2085				 iterate_devices_callout_fn fn, void *data)2086{2087	int ret;2088	struct clone *clone = ti->private;2089	struct dm_dev *dest_dev = clone->dest_dev;2090	struct dm_dev *source_dev = clone->source_dev;2091 2092	ret = fn(ti, source_dev, 0, ti->len, data);2093	if (!ret)2094		ret = fn(ti, dest_dev, 0, ti->len, data);2095	return ret;2096}2097 2098/*2099 * dm-clone message functions.2100 */2101static void set_hydration_threshold(struct clone *clone, unsigned int nr_regions)2102{2103	WRITE_ONCE(clone->hydration_threshold, nr_regions);2104 2105	/*2106	 * If user space sets hydration_threshold to zero then the hydration2107	 * will stop. If at a later time the hydration_threshold is increased2108	 * we must restart the hydration process by waking up the worker.2109	 */2110	wake_worker(clone);2111}2112 2113static void set_hydration_batch_size(struct clone *clone, unsigned int nr_regions)2114{2115	WRITE_ONCE(clone->hydration_batch_size, nr_regions);2116}2117 2118static void enable_hydration(struct clone *clone)2119{2120	if (!test_and_set_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags))2121		wake_worker(clone);2122}2123 2124static void disable_hydration(struct clone *clone)2125{2126	clear_bit(DM_CLONE_HYDRATION_ENABLED, &clone->flags);2127}2128 2129static int clone_message(struct dm_target *ti, unsigned int argc, char **argv,2130			 char *result, unsigned int maxlen)2131{2132	struct clone *clone = ti->private;2133	unsigned int value;2134 2135	if (!argc)2136		return -EINVAL;2137 2138	if (!strcasecmp(argv[0], "enable_hydration")) {2139		enable_hydration(clone);2140		return 0;2141	}2142 2143	if (!strcasecmp(argv[0], "disable_hydration")) {2144		disable_hydration(clone);2145		return 0;2146	}2147 2148	if (argc != 2)2149		return -EINVAL;2150 2151	if (!strcasecmp(argv[0], "hydration_threshold")) {2152		if (kstrtouint(argv[1], 10, &value))2153			return -EINVAL;2154 2155		set_hydration_threshold(clone, value);2156 2157		return 0;2158	}2159 2160	if (!strcasecmp(argv[0], "hydration_batch_size")) {2161		if (kstrtouint(argv[1], 10, &value))2162			return -EINVAL;2163 2164		set_hydration_batch_size(clone, value);2165 2166		return 0;2167	}2168 2169	DMERR("%s: Unsupported message `%s'", clone_device_name(clone), argv[0]);2170	return -EINVAL;2171}2172 2173static struct target_type clone_target = {2174	.name = "clone",2175	.version = {1, 0, 0},2176	.module = THIS_MODULE,2177	.ctr = clone_ctr,2178	.dtr =  clone_dtr,2179	.map = clone_map,2180	.end_io = clone_endio,2181	.postsuspend = clone_postsuspend,2182	.resume = clone_resume,2183	.status = clone_status,2184	.message = clone_message,2185	.io_hints = clone_io_hints,2186	.iterate_devices = clone_iterate_devices,2187};2188 2189/*---------------------------------------------------------------------------*/2190 2191/* Module functions */2192static int __init dm_clone_init(void)2193{2194	int r;2195 2196	_hydration_cache = KMEM_CACHE(dm_clone_region_hydration, 0);2197	if (!_hydration_cache)2198		return -ENOMEM;2199 2200	r = dm_register_target(&clone_target);2201	if (r < 0) {2202		kmem_cache_destroy(_hydration_cache);2203		return r;2204	}2205 2206	return 0;2207}2208 2209static void __exit dm_clone_exit(void)2210{2211	dm_unregister_target(&clone_target);2212 2213	kmem_cache_destroy(_hydration_cache);2214	_hydration_cache = NULL;2215}2216 2217/* Module hooks */2218module_init(dm_clone_init);2219module_exit(dm_clone_exit);2220 2221MODULE_DESCRIPTION(DM_NAME " clone target");2222MODULE_AUTHOR("Nikos Tsironis <ntsironis@arrikto.com>");2223MODULE_LICENSE("GPL");2224