brintos

brintos / linux-shallow public Read only

0
0
Text · 93.5 KiB · 6c9d242 Raw
3506 lines · c
1// SPDX-License-Identifier: GPL-2.0-or-later2/*3 * raid1.c : Multiple Devices driver for Linux4 *5 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat6 *7 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman8 *9 * RAID-1 management functions.10 *11 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 200012 *13 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>14 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>15 *16 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support17 * bitmapped intelligence in resync:18 *19 *      - bitmap marked during normal i/o20 *      - bitmap used to skip nondirty blocks during sync21 *22 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:23 * - persistent bitmap code24 */25 26#include <linux/slab.h>27#include <linux/delay.h>28#include <linux/blkdev.h>29#include <linux/module.h>30#include <linux/seq_file.h>31#include <linux/ratelimit.h>32#include <linux/interval_tree_generic.h>33 34#include <trace/events/block.h>35 36#include "md.h"37#include "raid1.h"38#include "md-bitmap.h"39 40#define UNSUPPORTED_MDDEV_FLAGS		\41	((1L << MD_HAS_JOURNAL) |	\42	 (1L << MD_JOURNAL_CLEAN) |	\43	 (1L << MD_HAS_PPL) |		\44	 (1L << MD_HAS_MULTIPLE_PPLS))45 46static void allow_barrier(struct r1conf *conf, sector_t sector_nr);47static void lower_barrier(struct r1conf *conf, sector_t sector_nr);48 49#define RAID_1_10_NAME "raid1"50#include "raid1-10.c"51 52#define START(node) ((node)->start)53#define LAST(node) ((node)->last)54INTERVAL_TREE_DEFINE(struct serial_info, node, sector_t, _subtree_last,55		     START, LAST, static inline, raid1_rb);56 57static int check_and_add_serial(struct md_rdev *rdev, struct r1bio *r1_bio,58				struct serial_info *si, int idx)59{60	unsigned long flags;61	int ret = 0;62	sector_t lo = r1_bio->sector;63	sector_t hi = lo + r1_bio->sectors;64	struct serial_in_rdev *serial = &rdev->serial[idx];65 66	spin_lock_irqsave(&serial->serial_lock, flags);67	/* collision happened */68	if (raid1_rb_iter_first(&serial->serial_rb, lo, hi))69		ret = -EBUSY;70	else {71		si->start = lo;72		si->last = hi;73		raid1_rb_insert(si, &serial->serial_rb);74	}75	spin_unlock_irqrestore(&serial->serial_lock, flags);76 77	return ret;78}79 80static void wait_for_serialization(struct md_rdev *rdev, struct r1bio *r1_bio)81{82	struct mddev *mddev = rdev->mddev;83	struct serial_info *si;84	int idx = sector_to_idx(r1_bio->sector);85	struct serial_in_rdev *serial = &rdev->serial[idx];86 87	if (WARN_ON(!mddev->serial_info_pool))88		return;89	si = mempool_alloc(mddev->serial_info_pool, GFP_NOIO);90	wait_event(serial->serial_io_wait,91		   check_and_add_serial(rdev, r1_bio, si, idx) == 0);92}93 94static void remove_serial(struct md_rdev *rdev, sector_t lo, sector_t hi)95{96	struct serial_info *si;97	unsigned long flags;98	int found = 0;99	struct mddev *mddev = rdev->mddev;100	int idx = sector_to_idx(lo);101	struct serial_in_rdev *serial = &rdev->serial[idx];102 103	spin_lock_irqsave(&serial->serial_lock, flags);104	for (si = raid1_rb_iter_first(&serial->serial_rb, lo, hi);105	     si; si = raid1_rb_iter_next(si, lo, hi)) {106		if (si->start == lo && si->last == hi) {107			raid1_rb_remove(si, &serial->serial_rb);108			mempool_free(si, mddev->serial_info_pool);109			found = 1;110			break;111		}112	}113	if (!found)114		WARN(1, "The write IO is not recorded for serialization\n");115	spin_unlock_irqrestore(&serial->serial_lock, flags);116	wake_up(&serial->serial_io_wait);117}118 119/*120 * for resync bio, r1bio pointer can be retrieved from the per-bio121 * 'struct resync_pages'.122 */123static inline struct r1bio *get_resync_r1bio(struct bio *bio)124{125	return get_resync_pages(bio)->raid_bio;126}127 128static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)129{130	struct pool_info *pi = data;131	int size = offsetof(struct r1bio, bios[pi->raid_disks]);132 133	/* allocate a r1bio with room for raid_disks entries in the bios array */134	return kzalloc(size, gfp_flags);135}136 137#define RESYNC_DEPTH 32138#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)139#define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)140#define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)141#define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)142#define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)143 144static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)145{146	struct pool_info *pi = data;147	struct r1bio *r1_bio;148	struct bio *bio;149	int need_pages;150	int j;151	struct resync_pages *rps;152 153	r1_bio = r1bio_pool_alloc(gfp_flags, pi);154	if (!r1_bio)155		return NULL;156 157	rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages),158			    gfp_flags);159	if (!rps)160		goto out_free_r1bio;161 162	/*163	 * Allocate bios : 1 for reading, n-1 for writing164	 */165	for (j = pi->raid_disks ; j-- ; ) {166		bio = bio_kmalloc(RESYNC_PAGES, gfp_flags);167		if (!bio)168			goto out_free_bio;169		bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0);170		r1_bio->bios[j] = bio;171	}172	/*173	 * Allocate RESYNC_PAGES data pages and attach them to174	 * the first bio.175	 * If this is a user-requested check/repair, allocate176	 * RESYNC_PAGES for each bio.177	 */178	if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))179		need_pages = pi->raid_disks;180	else181		need_pages = 1;182	for (j = 0; j < pi->raid_disks; j++) {183		struct resync_pages *rp = &rps[j];184 185		bio = r1_bio->bios[j];186 187		if (j < need_pages) {188			if (resync_alloc_pages(rp, gfp_flags))189				goto out_free_pages;190		} else {191			memcpy(rp, &rps[0], sizeof(*rp));192			resync_get_all_pages(rp);193		}194 195		rp->raid_bio = r1_bio;196		bio->bi_private = rp;197	}198 199	r1_bio->master_bio = NULL;200 201	return r1_bio;202 203out_free_pages:204	while (--j >= 0)205		resync_free_pages(&rps[j]);206 207out_free_bio:208	while (++j < pi->raid_disks) {209		bio_uninit(r1_bio->bios[j]);210		kfree(r1_bio->bios[j]);211	}212	kfree(rps);213 214out_free_r1bio:215	rbio_pool_free(r1_bio, data);216	return NULL;217}218 219static void r1buf_pool_free(void *__r1_bio, void *data)220{221	struct pool_info *pi = data;222	int i;223	struct r1bio *r1bio = __r1_bio;224	struct resync_pages *rp = NULL;225 226	for (i = pi->raid_disks; i--; ) {227		rp = get_resync_pages(r1bio->bios[i]);228		resync_free_pages(rp);229		bio_uninit(r1bio->bios[i]);230		kfree(r1bio->bios[i]);231	}232 233	/* resync pages array stored in the 1st bio's .bi_private */234	kfree(rp);235 236	rbio_pool_free(r1bio, data);237}238 239static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)240{241	int i;242 243	for (i = 0; i < conf->raid_disks * 2; i++) {244		struct bio **bio = r1_bio->bios + i;245		if (!BIO_SPECIAL(*bio))246			bio_put(*bio);247		*bio = NULL;248	}249}250 251static void free_r1bio(struct r1bio *r1_bio)252{253	struct r1conf *conf = r1_bio->mddev->private;254 255	put_all_bios(conf, r1_bio);256	mempool_free(r1_bio, &conf->r1bio_pool);257}258 259static void put_buf(struct r1bio *r1_bio)260{261	struct r1conf *conf = r1_bio->mddev->private;262	sector_t sect = r1_bio->sector;263	int i;264 265	for (i = 0; i < conf->raid_disks * 2; i++) {266		struct bio *bio = r1_bio->bios[i];267		if (bio->bi_end_io)268			rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);269	}270 271	mempool_free(r1_bio, &conf->r1buf_pool);272 273	lower_barrier(conf, sect);274}275 276static void reschedule_retry(struct r1bio *r1_bio)277{278	unsigned long flags;279	struct mddev *mddev = r1_bio->mddev;280	struct r1conf *conf = mddev->private;281	int idx;282 283	idx = sector_to_idx(r1_bio->sector);284	spin_lock_irqsave(&conf->device_lock, flags);285	list_add(&r1_bio->retry_list, &conf->retry_list);286	atomic_inc(&conf->nr_queued[idx]);287	spin_unlock_irqrestore(&conf->device_lock, flags);288 289	wake_up(&conf->wait_barrier);290	md_wakeup_thread(mddev->thread);291}292 293/*294 * raid_end_bio_io() is called when we have finished servicing a mirrored295 * operation and are ready to return a success/failure code to the buffer296 * cache layer.297 */298static void call_bio_endio(struct r1bio *r1_bio)299{300	struct bio *bio = r1_bio->master_bio;301 302	if (!test_bit(R1BIO_Uptodate, &r1_bio->state))303		bio->bi_status = BLK_STS_IOERR;304 305	bio_endio(bio);306}307 308static void raid_end_bio_io(struct r1bio *r1_bio)309{310	struct bio *bio = r1_bio->master_bio;311	struct r1conf *conf = r1_bio->mddev->private;312	sector_t sector = r1_bio->sector;313 314	/* if nobody has done the final endio yet, do it now */315	if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {316		pr_debug("raid1: sync end %s on sectors %llu-%llu\n",317			 (bio_data_dir(bio) == WRITE) ? "write" : "read",318			 (unsigned long long) bio->bi_iter.bi_sector,319			 (unsigned long long) bio_end_sector(bio) - 1);320 321		call_bio_endio(r1_bio);322	}323 324	free_r1bio(r1_bio);325	/*326	 * Wake up any possible resync thread that waits for the device327	 * to go idle.  All I/Os, even write-behind writes, are done.328	 */329	allow_barrier(conf, sector);330}331 332/*333 * Update disk head position estimator based on IRQ completion info.334 */335static inline void update_head_pos(int disk, struct r1bio *r1_bio)336{337	struct r1conf *conf = r1_bio->mddev->private;338 339	conf->mirrors[disk].head_position =340		r1_bio->sector + (r1_bio->sectors);341}342 343/*344 * Find the disk number which triggered given bio345 */346static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)347{348	int mirror;349	struct r1conf *conf = r1_bio->mddev->private;350	int raid_disks = conf->raid_disks;351 352	for (mirror = 0; mirror < raid_disks * 2; mirror++)353		if (r1_bio->bios[mirror] == bio)354			break;355 356	BUG_ON(mirror == raid_disks * 2);357	update_head_pos(mirror, r1_bio);358 359	return mirror;360}361 362static void raid1_end_read_request(struct bio *bio)363{364	int uptodate = !bio->bi_status;365	struct r1bio *r1_bio = bio->bi_private;366	struct r1conf *conf = r1_bio->mddev->private;367	struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;368 369	/*370	 * this branch is our 'one mirror IO has finished' event handler:371	 */372	update_head_pos(r1_bio->read_disk, r1_bio);373 374	if (uptodate)375		set_bit(R1BIO_Uptodate, &r1_bio->state);376	else if (test_bit(FailFast, &rdev->flags) &&377		 test_bit(R1BIO_FailFast, &r1_bio->state))378		/* This was a fail-fast read so we definitely379		 * want to retry */380		;381	else {382		/* If all other devices have failed, we want to return383		 * the error upwards rather than fail the last device.384		 * Here we redefine "uptodate" to mean "Don't want to retry"385		 */386		unsigned long flags;387		spin_lock_irqsave(&conf->device_lock, flags);388		if (r1_bio->mddev->degraded == conf->raid_disks ||389		    (r1_bio->mddev->degraded == conf->raid_disks-1 &&390		     test_bit(In_sync, &rdev->flags)))391			uptodate = 1;392		spin_unlock_irqrestore(&conf->device_lock, flags);393	}394 395	if (uptodate) {396		raid_end_bio_io(r1_bio);397		rdev_dec_pending(rdev, conf->mddev);398	} else {399		/*400		 * oops, read error:401		 */402		pr_err_ratelimited("md/raid1:%s: %pg: rescheduling sector %llu\n",403				   mdname(conf->mddev),404				   rdev->bdev,405				   (unsigned long long)r1_bio->sector);406		set_bit(R1BIO_ReadError, &r1_bio->state);407		reschedule_retry(r1_bio);408		/* don't drop the reference on read_disk yet */409	}410}411 412static void close_write(struct r1bio *r1_bio)413{414	struct mddev *mddev = r1_bio->mddev;415 416	/* it really is the end of this request */417	if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {418		bio_free_pages(r1_bio->behind_master_bio);419		bio_put(r1_bio->behind_master_bio);420		r1_bio->behind_master_bio = NULL;421	}422 423	/* clear the bitmap if all writes complete successfully */424	mddev->bitmap_ops->endwrite(mddev, r1_bio->sector, r1_bio->sectors,425				    !test_bit(R1BIO_Degraded, &r1_bio->state),426				    test_bit(R1BIO_BehindIO, &r1_bio->state));427	md_write_end(mddev);428}429 430static void r1_bio_write_done(struct r1bio *r1_bio)431{432	if (!atomic_dec_and_test(&r1_bio->remaining))433		return;434 435	if (test_bit(R1BIO_WriteError, &r1_bio->state))436		reschedule_retry(r1_bio);437	else {438		close_write(r1_bio);439		if (test_bit(R1BIO_MadeGood, &r1_bio->state))440			reschedule_retry(r1_bio);441		else442			raid_end_bio_io(r1_bio);443	}444}445 446static void raid1_end_write_request(struct bio *bio)447{448	struct r1bio *r1_bio = bio->bi_private;449	int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);450	struct r1conf *conf = r1_bio->mddev->private;451	struct bio *to_put = NULL;452	int mirror = find_bio_disk(r1_bio, bio);453	struct md_rdev *rdev = conf->mirrors[mirror].rdev;454	bool discard_error;455	sector_t lo = r1_bio->sector;456	sector_t hi = r1_bio->sector + r1_bio->sectors;457 458	discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD;459 460	/*461	 * 'one mirror IO has finished' event handler:462	 */463	if (bio->bi_status && !discard_error) {464		set_bit(WriteErrorSeen,	&rdev->flags);465		if (!test_and_set_bit(WantReplacement, &rdev->flags))466			set_bit(MD_RECOVERY_NEEDED, &467				conf->mddev->recovery);468 469		if (test_bit(FailFast, &rdev->flags) &&470		    (bio->bi_opf & MD_FAILFAST) &&471		    /* We never try FailFast to WriteMostly devices */472		    !test_bit(WriteMostly, &rdev->flags)) {473			md_error(r1_bio->mddev, rdev);474		}475 476		/*477		 * When the device is faulty, it is not necessary to478		 * handle write error.479		 */480		if (!test_bit(Faulty, &rdev->flags))481			set_bit(R1BIO_WriteError, &r1_bio->state);482		else {483			/* Fail the request */484			set_bit(R1BIO_Degraded, &r1_bio->state);485			/* Finished with this branch */486			r1_bio->bios[mirror] = NULL;487			to_put = bio;488		}489	} else {490		/*491		 * Set R1BIO_Uptodate in our master bio, so that we492		 * will return a good error code for to the higher493		 * levels even if IO on some other mirrored buffer494		 * fails.495		 *496		 * The 'master' represents the composite IO operation497		 * to user-side. So if something waits for IO, then it498		 * will wait for the 'master' bio.499		 */500		r1_bio->bios[mirror] = NULL;501		to_put = bio;502		/*503		 * Do not set R1BIO_Uptodate if the current device is504		 * rebuilding or Faulty. This is because we cannot use505		 * such device for properly reading the data back (we could506		 * potentially use it, if the current write would have felt507		 * before rdev->recovery_offset, but for simplicity we don't508		 * check this here.509		 */510		if (test_bit(In_sync, &rdev->flags) &&511		    !test_bit(Faulty, &rdev->flags))512			set_bit(R1BIO_Uptodate, &r1_bio->state);513 514		/* Maybe we can clear some bad blocks. */515		if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) &&516		    !discard_error) {517			r1_bio->bios[mirror] = IO_MADE_GOOD;518			set_bit(R1BIO_MadeGood, &r1_bio->state);519		}520	}521 522	if (behind) {523		if (test_bit(CollisionCheck, &rdev->flags))524			remove_serial(rdev, lo, hi);525		if (test_bit(WriteMostly, &rdev->flags))526			atomic_dec(&r1_bio->behind_remaining);527 528		/*529		 * In behind mode, we ACK the master bio once the I/O530		 * has safely reached all non-writemostly531		 * disks. Setting the Returned bit ensures that this532		 * gets done only once -- we don't ever want to return533		 * -EIO here, instead we'll wait534		 */535		if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&536		    test_bit(R1BIO_Uptodate, &r1_bio->state)) {537			/* Maybe we can return now */538			if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {539				struct bio *mbio = r1_bio->master_bio;540				pr_debug("raid1: behind end write sectors"541					 " %llu-%llu\n",542					 (unsigned long long) mbio->bi_iter.bi_sector,543					 (unsigned long long) bio_end_sector(mbio) - 1);544				call_bio_endio(r1_bio);545			}546		}547	} else if (rdev->mddev->serialize_policy)548		remove_serial(rdev, lo, hi);549	if (r1_bio->bios[mirror] == NULL)550		rdev_dec_pending(rdev, conf->mddev);551 552	/*553	 * Let's see if all mirrored write operations have finished554	 * already.555	 */556	r1_bio_write_done(r1_bio);557 558	if (to_put)559		bio_put(to_put);560}561 562static sector_t align_to_barrier_unit_end(sector_t start_sector,563					  sector_t sectors)564{565	sector_t len;566 567	WARN_ON(sectors == 0);568	/*569	 * len is the number of sectors from start_sector to end of the570	 * barrier unit which start_sector belongs to.571	 */572	len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) -573	      start_sector;574 575	if (len > sectors)576		len = sectors;577 578	return len;579}580 581static void update_read_sectors(struct r1conf *conf, int disk,582				sector_t this_sector, int len)583{584	struct raid1_info *info = &conf->mirrors[disk];585 586	atomic_inc(&info->rdev->nr_pending);587	if (info->next_seq_sect != this_sector)588		info->seq_start = this_sector;589	info->next_seq_sect = this_sector + len;590}591 592static int choose_first_rdev(struct r1conf *conf, struct r1bio *r1_bio,593			     int *max_sectors)594{595	sector_t this_sector = r1_bio->sector;596	int len = r1_bio->sectors;597	int disk;598 599	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {600		struct md_rdev *rdev;601		int read_len;602 603		if (r1_bio->bios[disk] == IO_BLOCKED)604			continue;605 606		rdev = conf->mirrors[disk].rdev;607		if (!rdev || test_bit(Faulty, &rdev->flags))608			continue;609 610		/* choose the first disk even if it has some bad blocks. */611		read_len = raid1_check_read_range(rdev, this_sector, &len);612		if (read_len > 0) {613			update_read_sectors(conf, disk, this_sector, read_len);614			*max_sectors = read_len;615			return disk;616		}617	}618 619	return -1;620}621 622static bool rdev_in_recovery(struct md_rdev *rdev, struct r1bio *r1_bio)623{624	return !test_bit(In_sync, &rdev->flags) &&625	       rdev->recovery_offset < r1_bio->sector + r1_bio->sectors;626}627 628static int choose_bb_rdev(struct r1conf *conf, struct r1bio *r1_bio,629			  int *max_sectors)630{631	sector_t this_sector = r1_bio->sector;632	int best_disk = -1;633	int best_len = 0;634	int disk;635 636	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {637		struct md_rdev *rdev;638		int len;639		int read_len;640 641		if (r1_bio->bios[disk] == IO_BLOCKED)642			continue;643 644		rdev = conf->mirrors[disk].rdev;645		if (!rdev || test_bit(Faulty, &rdev->flags) ||646		    rdev_in_recovery(rdev, r1_bio) ||647		    test_bit(WriteMostly, &rdev->flags))648			continue;649 650		/* keep track of the disk with the most readable sectors. */651		len = r1_bio->sectors;652		read_len = raid1_check_read_range(rdev, this_sector, &len);653		if (read_len > best_len) {654			best_disk = disk;655			best_len = read_len;656		}657	}658 659	if (best_disk != -1) {660		*max_sectors = best_len;661		update_read_sectors(conf, best_disk, this_sector, best_len);662	}663 664	return best_disk;665}666 667static int choose_slow_rdev(struct r1conf *conf, struct r1bio *r1_bio,668			    int *max_sectors)669{670	sector_t this_sector = r1_bio->sector;671	int bb_disk = -1;672	int bb_read_len = 0;673	int disk;674 675	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {676		struct md_rdev *rdev;677		int len;678		int read_len;679 680		if (r1_bio->bios[disk] == IO_BLOCKED)681			continue;682 683		rdev = conf->mirrors[disk].rdev;684		if (!rdev || test_bit(Faulty, &rdev->flags) ||685		    !test_bit(WriteMostly, &rdev->flags) ||686		    rdev_in_recovery(rdev, r1_bio))687			continue;688 689		/* there are no bad blocks, we can use this disk */690		len = r1_bio->sectors;691		read_len = raid1_check_read_range(rdev, this_sector, &len);692		if (read_len == r1_bio->sectors) {693			*max_sectors = read_len;694			update_read_sectors(conf, disk, this_sector, read_len);695			return disk;696		}697 698		/*699		 * there are partial bad blocks, choose the rdev with largest700		 * read length.701		 */702		if (read_len > bb_read_len) {703			bb_disk = disk;704			bb_read_len = read_len;705		}706	}707 708	if (bb_disk != -1) {709		*max_sectors = bb_read_len;710		update_read_sectors(conf, bb_disk, this_sector, bb_read_len);711	}712 713	return bb_disk;714}715 716static bool is_sequential(struct r1conf *conf, int disk, struct r1bio *r1_bio)717{718	/* TODO: address issues with this check and concurrency. */719	return conf->mirrors[disk].next_seq_sect == r1_bio->sector ||720	       conf->mirrors[disk].head_position == r1_bio->sector;721}722 723/*724 * If buffered sequential IO size exceeds optimal iosize, check if there is idle725 * disk. If yes, choose the idle disk.726 */727static bool should_choose_next(struct r1conf *conf, int disk)728{729	struct raid1_info *mirror = &conf->mirrors[disk];730	int opt_iosize;731 732	if (!test_bit(Nonrot, &mirror->rdev->flags))733		return false;734 735	opt_iosize = bdev_io_opt(mirror->rdev->bdev) >> 9;736	return opt_iosize > 0 && mirror->seq_start != MaxSector &&737	       mirror->next_seq_sect > opt_iosize &&738	       mirror->next_seq_sect - opt_iosize >= mirror->seq_start;739}740 741static bool rdev_readable(struct md_rdev *rdev, struct r1bio *r1_bio)742{743	if (!rdev || test_bit(Faulty, &rdev->flags))744		return false;745 746	if (rdev_in_recovery(rdev, r1_bio))747		return false;748 749	/* don't read from slow disk unless have to */750	if (test_bit(WriteMostly, &rdev->flags))751		return false;752 753	/* don't split IO for bad blocks unless have to */754	if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors))755		return false;756 757	return true;758}759 760struct read_balance_ctl {761	sector_t closest_dist;762	int closest_dist_disk;763	int min_pending;764	int min_pending_disk;765	int sequential_disk;766	int readable_disks;767};768 769static int choose_best_rdev(struct r1conf *conf, struct r1bio *r1_bio)770{771	int disk;772	struct read_balance_ctl ctl = {773		.closest_dist_disk      = -1,774		.closest_dist           = MaxSector,775		.min_pending_disk       = -1,776		.min_pending            = UINT_MAX,777		.sequential_disk	= -1,778	};779 780	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {781		struct md_rdev *rdev;782		sector_t dist;783		unsigned int pending;784 785		if (r1_bio->bios[disk] == IO_BLOCKED)786			continue;787 788		rdev = conf->mirrors[disk].rdev;789		if (!rdev_readable(rdev, r1_bio))790			continue;791 792		/* At least two disks to choose from so failfast is OK */793		if (ctl.readable_disks++ == 1)794			set_bit(R1BIO_FailFast, &r1_bio->state);795 796		pending = atomic_read(&rdev->nr_pending);797		dist = abs(r1_bio->sector - conf->mirrors[disk].head_position);798 799		/* Don't change to another disk for sequential reads */800		if (is_sequential(conf, disk, r1_bio)) {801			if (!should_choose_next(conf, disk))802				return disk;803 804			/*805			 * Add 'pending' to avoid choosing this disk if806			 * there is other idle disk.807			 */808			pending++;809			/*810			 * If there is no other idle disk, this disk811			 * will be chosen.812			 */813			ctl.sequential_disk = disk;814		}815 816		if (ctl.min_pending > pending) {817			ctl.min_pending = pending;818			ctl.min_pending_disk = disk;819		}820 821		if (ctl.closest_dist > dist) {822			ctl.closest_dist = dist;823			ctl.closest_dist_disk = disk;824		}825	}826 827	/*828	 * sequential IO size exceeds optimal iosize, however, there is no other829	 * idle disk, so choose the sequential disk.830	 */831	if (ctl.sequential_disk != -1 && ctl.min_pending != 0)832		return ctl.sequential_disk;833 834	/*835	 * If all disks are rotational, choose the closest disk. If any disk is836	 * non-rotational, choose the disk with less pending request even the837	 * disk is rotational, which might/might not be optimal for raids with838	 * mixed ratation/non-rotational disks depending on workload.839	 */840	if (ctl.min_pending_disk != -1 &&841	    (READ_ONCE(conf->nonrot_disks) || ctl.min_pending == 0))842		return ctl.min_pending_disk;843	else844		return ctl.closest_dist_disk;845}846 847/*848 * This routine returns the disk from which the requested read should be done.849 *850 * 1) If resync is in progress, find the first usable disk and use it even if it851 * has some bad blocks.852 *853 * 2) Now that there is no resync, loop through all disks and skipping slow854 * disks and disks with bad blocks for now. Only pay attention to key disk855 * choice.856 *857 * 3) If we've made it this far, now look for disks with bad blocks and choose858 * the one with most number of sectors.859 *860 * 4) If we are all the way at the end, we have no choice but to use a disk even861 * if it is write mostly.862 *863 * The rdev for the device selected will have nr_pending incremented.864 */865static int read_balance(struct r1conf *conf, struct r1bio *r1_bio,866			int *max_sectors)867{868	int disk;869 870	clear_bit(R1BIO_FailFast, &r1_bio->state);871 872	if (raid1_should_read_first(conf->mddev, r1_bio->sector,873				    r1_bio->sectors))874		return choose_first_rdev(conf, r1_bio, max_sectors);875 876	disk = choose_best_rdev(conf, r1_bio);877	if (disk >= 0) {878		*max_sectors = r1_bio->sectors;879		update_read_sectors(conf, disk, r1_bio->sector,880				    r1_bio->sectors);881		return disk;882	}883 884	/*885	 * If we are here it means we didn't find a perfectly good disk so886	 * now spend a bit more time trying to find one with the most good887	 * sectors.888	 */889	disk = choose_bb_rdev(conf, r1_bio, max_sectors);890	if (disk >= 0)891		return disk;892 893	return choose_slow_rdev(conf, r1_bio, max_sectors);894}895 896static void wake_up_barrier(struct r1conf *conf)897{898	if (wq_has_sleeper(&conf->wait_barrier))899		wake_up(&conf->wait_barrier);900}901 902static void flush_bio_list(struct r1conf *conf, struct bio *bio)903{904	/* flush any pending bitmap writes to disk before proceeding w/ I/O */905	raid1_prepare_flush_writes(conf->mddev);906	wake_up_barrier(conf);907 908	while (bio) { /* submit pending writes */909		struct bio *next = bio->bi_next;910 911		raid1_submit_write(bio);912		bio = next;913		cond_resched();914	}915}916 917static void flush_pending_writes(struct r1conf *conf)918{919	/* Any writes that have been queued but are awaiting920	 * bitmap updates get flushed here.921	 */922	spin_lock_irq(&conf->device_lock);923 924	if (conf->pending_bio_list.head) {925		struct blk_plug plug;926		struct bio *bio;927 928		bio = bio_list_get(&conf->pending_bio_list);929		spin_unlock_irq(&conf->device_lock);930 931		/*932		 * As this is called in a wait_event() loop (see freeze_array),933		 * current->state might be TASK_UNINTERRUPTIBLE which will934		 * cause a warning when we prepare to wait again.  As it is935		 * rare that this path is taken, it is perfectly safe to force936		 * us to go around the wait_event() loop again, so the warning937		 * is a false-positive.  Silence the warning by resetting938		 * thread state939		 */940		__set_current_state(TASK_RUNNING);941		blk_start_plug(&plug);942		flush_bio_list(conf, bio);943		blk_finish_plug(&plug);944	} else945		spin_unlock_irq(&conf->device_lock);946}947 948/* Barriers....949 * Sometimes we need to suspend IO while we do something else,950 * either some resync/recovery, or reconfigure the array.951 * To do this we raise a 'barrier'.952 * The 'barrier' is a counter that can be raised multiple times953 * to count how many activities are happening which preclude954 * normal IO.955 * We can only raise the barrier if there is no pending IO.956 * i.e. if nr_pending == 0.957 * We choose only to raise the barrier if no-one is waiting for the958 * barrier to go down.  This means that as soon as an IO request959 * is ready, no other operations which require a barrier will start960 * until the IO request has had a chance.961 *962 * So: regular IO calls 'wait_barrier'.  When that returns there963 *    is no backgroup IO happening,  It must arrange to call964 *    allow_barrier when it has finished its IO.965 * backgroup IO calls must call raise_barrier.  Once that returns966 *    there is no normal IO happeing.  It must arrange to call967 *    lower_barrier when the particular background IO completes.968 *969 * If resync/recovery is interrupted, returns -EINTR;970 * Otherwise, returns 0.971 */972static int raise_barrier(struct r1conf *conf, sector_t sector_nr)973{974	int idx = sector_to_idx(sector_nr);975 976	spin_lock_irq(&conf->resync_lock);977 978	/* Wait until no block IO is waiting */979	wait_event_lock_irq(conf->wait_barrier,980			    !atomic_read(&conf->nr_waiting[idx]),981			    conf->resync_lock);982 983	/* block any new IO from starting */984	atomic_inc(&conf->barrier[idx]);985	/*986	 * In raise_barrier() we firstly increase conf->barrier[idx] then987	 * check conf->nr_pending[idx]. In _wait_barrier() we firstly988	 * increase conf->nr_pending[idx] then check conf->barrier[idx].989	 * A memory barrier here to make sure conf->nr_pending[idx] won't990	 * be fetched before conf->barrier[idx] is increased. Otherwise991	 * there will be a race between raise_barrier() and _wait_barrier().992	 */993	smp_mb__after_atomic();994 995	/* For these conditions we must wait:996	 * A: while the array is in frozen state997	 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O998	 *    existing in corresponding I/O barrier bucket.999	 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches1000	 *    max resync count which allowed on current I/O barrier bucket.1001	 */1002	wait_event_lock_irq(conf->wait_barrier,1003			    (!conf->array_frozen &&1004			     !atomic_read(&conf->nr_pending[idx]) &&1005			     atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) ||1006				test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery),1007			    conf->resync_lock);1008 1009	if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {1010		atomic_dec(&conf->barrier[idx]);1011		spin_unlock_irq(&conf->resync_lock);1012		wake_up(&conf->wait_barrier);1013		return -EINTR;1014	}1015 1016	atomic_inc(&conf->nr_sync_pending);1017	spin_unlock_irq(&conf->resync_lock);1018 1019	return 0;1020}1021 1022static void lower_barrier(struct r1conf *conf, sector_t sector_nr)1023{1024	int idx = sector_to_idx(sector_nr);1025 1026	BUG_ON(atomic_read(&conf->barrier[idx]) <= 0);1027 1028	atomic_dec(&conf->barrier[idx]);1029	atomic_dec(&conf->nr_sync_pending);1030	wake_up(&conf->wait_barrier);1031}1032 1033static bool _wait_barrier(struct r1conf *conf, int idx, bool nowait)1034{1035	bool ret = true;1036 1037	/*1038	 * We need to increase conf->nr_pending[idx] very early here,1039	 * then raise_barrier() can be blocked when it waits for1040	 * conf->nr_pending[idx] to be 0. Then we can avoid holding1041	 * conf->resync_lock when there is no barrier raised in same1042	 * barrier unit bucket. Also if the array is frozen, I/O1043	 * should be blocked until array is unfrozen.1044	 */1045	atomic_inc(&conf->nr_pending[idx]);1046	/*1047	 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then1048	 * check conf->barrier[idx]. In raise_barrier() we firstly increase1049	 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory1050	 * barrier is necessary here to make sure conf->barrier[idx] won't be1051	 * fetched before conf->nr_pending[idx] is increased. Otherwise there1052	 * will be a race between _wait_barrier() and raise_barrier().1053	 */1054	smp_mb__after_atomic();1055 1056	/*1057	 * Don't worry about checking two atomic_t variables at same time1058	 * here. If during we check conf->barrier[idx], the array is1059	 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is1060	 * 0, it is safe to return and make the I/O continue. Because the1061	 * array is frozen, all I/O returned here will eventually complete1062	 * or be queued, no race will happen. See code comment in1063	 * frozen_array().1064	 */1065	if (!READ_ONCE(conf->array_frozen) &&1066	    !atomic_read(&conf->barrier[idx]))1067		return ret;1068 1069	/*1070	 * After holding conf->resync_lock, conf->nr_pending[idx]1071	 * should be decreased before waiting for barrier to drop.1072	 * Otherwise, we may encounter a race condition because1073	 * raise_barrer() might be waiting for conf->nr_pending[idx]1074	 * to be 0 at same time.1075	 */1076	spin_lock_irq(&conf->resync_lock);1077	atomic_inc(&conf->nr_waiting[idx]);1078	atomic_dec(&conf->nr_pending[idx]);1079	/*1080	 * In case freeze_array() is waiting for1081	 * get_unqueued_pending() == extra1082	 */1083	wake_up_barrier(conf);1084	/* Wait for the barrier in same barrier unit bucket to drop. */1085 1086	/* Return false when nowait flag is set */1087	if (nowait) {1088		ret = false;1089	} else {1090		wait_event_lock_irq(conf->wait_barrier,1091				!conf->array_frozen &&1092				!atomic_read(&conf->barrier[idx]),1093				conf->resync_lock);1094		atomic_inc(&conf->nr_pending[idx]);1095	}1096 1097	atomic_dec(&conf->nr_waiting[idx]);1098	spin_unlock_irq(&conf->resync_lock);1099	return ret;1100}1101 1102static bool wait_read_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait)1103{1104	int idx = sector_to_idx(sector_nr);1105	bool ret = true;1106 1107	/*1108	 * Very similar to _wait_barrier(). The difference is, for read1109	 * I/O we don't need wait for sync I/O, but if the whole array1110	 * is frozen, the read I/O still has to wait until the array is1111	 * unfrozen. Since there is no ordering requirement with1112	 * conf->barrier[idx] here, memory barrier is unnecessary as well.1113	 */1114	atomic_inc(&conf->nr_pending[idx]);1115 1116	if (!READ_ONCE(conf->array_frozen))1117		return ret;1118 1119	spin_lock_irq(&conf->resync_lock);1120	atomic_inc(&conf->nr_waiting[idx]);1121	atomic_dec(&conf->nr_pending[idx]);1122	/*1123	 * In case freeze_array() is waiting for1124	 * get_unqueued_pending() == extra1125	 */1126	wake_up_barrier(conf);1127	/* Wait for array to be unfrozen */1128 1129	/* Return false when nowait flag is set */1130	if (nowait) {1131		/* Return false when nowait flag is set */1132		ret = false;1133	} else {1134		wait_event_lock_irq(conf->wait_barrier,1135				!conf->array_frozen,1136				conf->resync_lock);1137		atomic_inc(&conf->nr_pending[idx]);1138	}1139 1140	atomic_dec(&conf->nr_waiting[idx]);1141	spin_unlock_irq(&conf->resync_lock);1142	return ret;1143}1144 1145static bool wait_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait)1146{1147	int idx = sector_to_idx(sector_nr);1148 1149	return _wait_barrier(conf, idx, nowait);1150}1151 1152static void _allow_barrier(struct r1conf *conf, int idx)1153{1154	atomic_dec(&conf->nr_pending[idx]);1155	wake_up_barrier(conf);1156}1157 1158static void allow_barrier(struct r1conf *conf, sector_t sector_nr)1159{1160	int idx = sector_to_idx(sector_nr);1161 1162	_allow_barrier(conf, idx);1163}1164 1165/* conf->resync_lock should be held */1166static int get_unqueued_pending(struct r1conf *conf)1167{1168	int idx, ret;1169 1170	ret = atomic_read(&conf->nr_sync_pending);1171	for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++)1172		ret += atomic_read(&conf->nr_pending[idx]) -1173			atomic_read(&conf->nr_queued[idx]);1174 1175	return ret;1176}1177 1178static void freeze_array(struct r1conf *conf, int extra)1179{1180	/* Stop sync I/O and normal I/O and wait for everything to1181	 * go quiet.1182	 * This is called in two situations:1183	 * 1) management command handlers (reshape, remove disk, quiesce).1184	 * 2) one normal I/O request failed.1185 1186	 * After array_frozen is set to 1, new sync IO will be blocked at1187	 * raise_barrier(), and new normal I/O will blocked at _wait_barrier()1188	 * or wait_read_barrier(). The flying I/Os will either complete or be1189	 * queued. When everything goes quite, there are only queued I/Os left.1190 1191	 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the1192	 * barrier bucket index which this I/O request hits. When all sync and1193	 * normal I/O are queued, sum of all conf->nr_pending[] will match sum1194	 * of all conf->nr_queued[]. But normal I/O failure is an exception,1195	 * in handle_read_error(), we may call freeze_array() before trying to1196	 * fix the read error. In this case, the error read I/O is not queued,1197	 * so get_unqueued_pending() == 1.1198	 *1199	 * Therefore before this function returns, we need to wait until1200	 * get_unqueued_pendings(conf) gets equal to extra. For1201	 * normal I/O context, extra is 1, in rested situations extra is 0.1202	 */1203	spin_lock_irq(&conf->resync_lock);1204	conf->array_frozen = 1;1205	mddev_add_trace_msg(conf->mddev, "raid1 wait freeze");1206	wait_event_lock_irq_cmd(1207		conf->wait_barrier,1208		get_unqueued_pending(conf) == extra,1209		conf->resync_lock,1210		flush_pending_writes(conf));1211	spin_unlock_irq(&conf->resync_lock);1212}1213static void unfreeze_array(struct r1conf *conf)1214{1215	/* reverse the effect of the freeze */1216	spin_lock_irq(&conf->resync_lock);1217	conf->array_frozen = 0;1218	spin_unlock_irq(&conf->resync_lock);1219	wake_up(&conf->wait_barrier);1220}1221 1222static void alloc_behind_master_bio(struct r1bio *r1_bio,1223					   struct bio *bio)1224{1225	int size = bio->bi_iter.bi_size;1226	unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;1227	int i = 0;1228	struct bio *behind_bio = NULL;1229 1230	behind_bio = bio_alloc_bioset(NULL, vcnt, 0, GFP_NOIO,1231				      &r1_bio->mddev->bio_set);1232 1233	/* discard op, we don't support writezero/writesame yet */1234	if (!bio_has_data(bio)) {1235		behind_bio->bi_iter.bi_size = size;1236		goto skip_copy;1237	}1238 1239	while (i < vcnt && size) {1240		struct page *page;1241		int len = min_t(int, PAGE_SIZE, size);1242 1243		page = alloc_page(GFP_NOIO);1244		if (unlikely(!page))1245			goto free_pages;1246 1247		if (!bio_add_page(behind_bio, page, len, 0)) {1248			put_page(page);1249			goto free_pages;1250		}1251 1252		size -= len;1253		i++;1254	}1255 1256	bio_copy_data(behind_bio, bio);1257skip_copy:1258	r1_bio->behind_master_bio = behind_bio;1259	set_bit(R1BIO_BehindIO, &r1_bio->state);1260 1261	return;1262 1263free_pages:1264	pr_debug("%dB behind alloc failed, doing sync I/O\n",1265		 bio->bi_iter.bi_size);1266	bio_free_pages(behind_bio);1267	bio_put(behind_bio);1268}1269 1270static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)1271{1272	struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,1273						  cb);1274	struct mddev *mddev = plug->cb.data;1275	struct r1conf *conf = mddev->private;1276	struct bio *bio;1277 1278	if (from_schedule) {1279		spin_lock_irq(&conf->device_lock);1280		bio_list_merge(&conf->pending_bio_list, &plug->pending);1281		spin_unlock_irq(&conf->device_lock);1282		wake_up_barrier(conf);1283		md_wakeup_thread(mddev->thread);1284		kfree(plug);1285		return;1286	}1287 1288	/* we aren't scheduling, so we can do the write-out directly. */1289	bio = bio_list_get(&plug->pending);1290	flush_bio_list(conf, bio);1291	kfree(plug);1292}1293 1294static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio)1295{1296	r1_bio->master_bio = bio;1297	r1_bio->sectors = bio_sectors(bio);1298	r1_bio->state = 0;1299	r1_bio->mddev = mddev;1300	r1_bio->sector = bio->bi_iter.bi_sector;1301}1302 1303static inline struct r1bio *1304alloc_r1bio(struct mddev *mddev, struct bio *bio)1305{1306	struct r1conf *conf = mddev->private;1307	struct r1bio *r1_bio;1308 1309	r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO);1310	/* Ensure no bio records IO_BLOCKED */1311	memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0]));1312	init_r1bio(r1_bio, mddev, bio);1313	return r1_bio;1314}1315 1316static void raid1_read_request(struct mddev *mddev, struct bio *bio,1317			       int max_read_sectors, struct r1bio *r1_bio)1318{1319	struct r1conf *conf = mddev->private;1320	struct raid1_info *mirror;1321	struct bio *read_bio;1322	const enum req_op op = bio_op(bio);1323	const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC;1324	int max_sectors;1325	int rdisk;1326	bool r1bio_existed = !!r1_bio;1327 1328	/*1329	 * If r1_bio is set, we are blocking the raid1d thread1330	 * so there is a tiny risk of deadlock.  So ask for1331	 * emergency memory if needed.1332	 */1333	gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO;1334 1335	/*1336	 * Still need barrier for READ in case that whole1337	 * array is frozen.1338	 */1339	if (!wait_read_barrier(conf, bio->bi_iter.bi_sector,1340				bio->bi_opf & REQ_NOWAIT)) {1341		bio_wouldblock_error(bio);1342		return;1343	}1344 1345	if (!r1_bio)1346		r1_bio = alloc_r1bio(mddev, bio);1347	else1348		init_r1bio(r1_bio, mddev, bio);1349	r1_bio->sectors = max_read_sectors;1350 1351	/*1352	 * make_request() can abort the operation when read-ahead is being1353	 * used and no empty request is available.1354	 */1355	rdisk = read_balance(conf, r1_bio, &max_sectors);1356	if (rdisk < 0) {1357		/* couldn't find anywhere to read from */1358		if (r1bio_existed)1359			pr_crit_ratelimited("md/raid1:%s: %pg: unrecoverable I/O read error for block %llu\n",1360					    mdname(mddev),1361					    conf->mirrors[r1_bio->read_disk].rdev->bdev,1362					    r1_bio->sector);1363		raid_end_bio_io(r1_bio);1364		return;1365	}1366	mirror = conf->mirrors + rdisk;1367 1368	if (r1bio_existed)1369		pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n",1370				    mdname(mddev),1371				    (unsigned long long)r1_bio->sector,1372				    mirror->rdev->bdev);1373 1374	if (test_bit(WriteMostly, &mirror->rdev->flags)) {1375		/*1376		 * Reading from a write-mostly device must take care not to1377		 * over-take any writes that are 'behind'1378		 */1379		mddev_add_trace_msg(mddev, "raid1 wait behind writes");1380		mddev->bitmap_ops->wait_behind_writes(mddev);1381	}1382 1383	if (max_sectors < bio_sectors(bio)) {1384		struct bio *split = bio_split(bio, max_sectors,1385					      gfp, &conf->bio_split);1386		bio_chain(split, bio);1387		submit_bio_noacct(bio);1388		bio = split;1389		r1_bio->master_bio = bio;1390		r1_bio->sectors = max_sectors;1391	}1392 1393	r1_bio->read_disk = rdisk;1394	if (!r1bio_existed) {1395		md_account_bio(mddev, &bio);1396		r1_bio->master_bio = bio;1397	}1398	read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp,1399				   &mddev->bio_set);1400 1401	r1_bio->bios[rdisk] = read_bio;1402 1403	read_bio->bi_iter.bi_sector = r1_bio->sector +1404		mirror->rdev->data_offset;1405	read_bio->bi_end_io = raid1_end_read_request;1406	read_bio->bi_opf = op | do_sync;1407	if (test_bit(FailFast, &mirror->rdev->flags) &&1408	    test_bit(R1BIO_FailFast, &r1_bio->state))1409	        read_bio->bi_opf |= MD_FAILFAST;1410	read_bio->bi_private = r1_bio;1411	mddev_trace_remap(mddev, read_bio, r1_bio->sector);1412	submit_bio_noacct(read_bio);1413}1414 1415static void raid1_write_request(struct mddev *mddev, struct bio *bio,1416				int max_write_sectors)1417{1418	struct r1conf *conf = mddev->private;1419	struct r1bio *r1_bio;1420	int i, disks;1421	unsigned long flags;1422	struct md_rdev *blocked_rdev;1423	int first_clone;1424	int max_sectors;1425	bool write_behind = false;1426	bool is_discard = (bio_op(bio) == REQ_OP_DISCARD);1427 1428	if (mddev_is_clustered(mddev) &&1429	     md_cluster_ops->area_resyncing(mddev, WRITE,1430		     bio->bi_iter.bi_sector, bio_end_sector(bio))) {1431 1432		DEFINE_WAIT(w);1433		if (bio->bi_opf & REQ_NOWAIT) {1434			bio_wouldblock_error(bio);1435			return;1436		}1437		for (;;) {1438			prepare_to_wait(&conf->wait_barrier,1439					&w, TASK_IDLE);1440			if (!md_cluster_ops->area_resyncing(mddev, WRITE,1441							bio->bi_iter.bi_sector,1442							bio_end_sector(bio)))1443				break;1444			schedule();1445		}1446		finish_wait(&conf->wait_barrier, &w);1447	}1448 1449	/*1450	 * Register the new request and wait if the reconstruction1451	 * thread has put up a bar for new requests.1452	 * Continue immediately if no resync is active currently.1453	 */1454	if (!wait_barrier(conf, bio->bi_iter.bi_sector,1455				bio->bi_opf & REQ_NOWAIT)) {1456		bio_wouldblock_error(bio);1457		return;1458	}1459 1460 retry_write:1461	r1_bio = alloc_r1bio(mddev, bio);1462	r1_bio->sectors = max_write_sectors;1463 1464	/* first select target devices under rcu_lock and1465	 * inc refcount on their rdev.  Record them by setting1466	 * bios[x] to bio1467	 * If there are known/acknowledged bad blocks on any device on1468	 * which we have seen a write error, we want to avoid writing those1469	 * blocks.1470	 * This potentially requires several writes to write around1471	 * the bad blocks.  Each set of writes gets it's own r1bio1472	 * with a set of bios attached.1473	 */1474 1475	disks = conf->raid_disks * 2;1476	blocked_rdev = NULL;1477	max_sectors = r1_bio->sectors;1478	for (i = 0;  i < disks; i++) {1479		struct md_rdev *rdev = conf->mirrors[i].rdev;1480 1481		/*1482		 * The write-behind io is only attempted on drives marked as1483		 * write-mostly, which means we could allocate write behind1484		 * bio later.1485		 */1486		if (!is_discard && rdev && test_bit(WriteMostly, &rdev->flags))1487			write_behind = true;1488 1489		if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {1490			atomic_inc(&rdev->nr_pending);1491			blocked_rdev = rdev;1492			break;1493		}1494		r1_bio->bios[i] = NULL;1495		if (!rdev || test_bit(Faulty, &rdev->flags)) {1496			if (i < conf->raid_disks)1497				set_bit(R1BIO_Degraded, &r1_bio->state);1498			continue;1499		}1500 1501		atomic_inc(&rdev->nr_pending);1502		if (test_bit(WriteErrorSeen, &rdev->flags)) {1503			sector_t first_bad;1504			int bad_sectors;1505			int is_bad;1506 1507			is_bad = is_badblock(rdev, r1_bio->sector, max_sectors,1508					     &first_bad, &bad_sectors);1509			if (is_bad < 0) {1510				/* mustn't write here until the bad block is1511				 * acknowledged*/1512				set_bit(BlockedBadBlocks, &rdev->flags);1513				blocked_rdev = rdev;1514				break;1515			}1516			if (is_bad && first_bad <= r1_bio->sector) {1517				/* Cannot write here at all */1518				bad_sectors -= (r1_bio->sector - first_bad);1519				if (bad_sectors < max_sectors)1520					/* mustn't write more than bad_sectors1521					 * to other devices yet1522					 */1523					max_sectors = bad_sectors;1524				rdev_dec_pending(rdev, mddev);1525				/* We don't set R1BIO_Degraded as that1526				 * only applies if the disk is1527				 * missing, so it might be re-added,1528				 * and we want to know to recover this1529				 * chunk.1530				 * In this case the device is here,1531				 * and the fact that this chunk is not1532				 * in-sync is recorded in the bad1533				 * block log1534				 */1535				continue;1536			}1537			if (is_bad) {1538				int good_sectors = first_bad - r1_bio->sector;1539				if (good_sectors < max_sectors)1540					max_sectors = good_sectors;1541			}1542		}1543		r1_bio->bios[i] = bio;1544	}1545 1546	if (unlikely(blocked_rdev)) {1547		/* Wait for this device to become unblocked */1548		int j;1549 1550		for (j = 0; j < i; j++)1551			if (r1_bio->bios[j])1552				rdev_dec_pending(conf->mirrors[j].rdev, mddev);1553		mempool_free(r1_bio, &conf->r1bio_pool);1554		allow_barrier(conf, bio->bi_iter.bi_sector);1555 1556		if (bio->bi_opf & REQ_NOWAIT) {1557			bio_wouldblock_error(bio);1558			return;1559		}1560		mddev_add_trace_msg(mddev, "raid1 wait rdev %d blocked",1561				blocked_rdev->raid_disk);1562		md_wait_for_blocked_rdev(blocked_rdev, mddev);1563		wait_barrier(conf, bio->bi_iter.bi_sector, false);1564		goto retry_write;1565	}1566 1567	/*1568	 * When using a bitmap, we may call alloc_behind_master_bio below.1569	 * alloc_behind_master_bio allocates a copy of the data payload a page1570	 * at a time and thus needs a new bio that can fit the whole payload1571	 * this bio in page sized chunks.1572	 */1573	if (write_behind && mddev->bitmap)1574		max_sectors = min_t(int, max_sectors,1575				    BIO_MAX_VECS * (PAGE_SIZE >> 9));1576	if (max_sectors < bio_sectors(bio)) {1577		struct bio *split = bio_split(bio, max_sectors,1578					      GFP_NOIO, &conf->bio_split);1579		bio_chain(split, bio);1580		submit_bio_noacct(bio);1581		bio = split;1582		r1_bio->master_bio = bio;1583		r1_bio->sectors = max_sectors;1584	}1585 1586	md_account_bio(mddev, &bio);1587	r1_bio->master_bio = bio;1588	atomic_set(&r1_bio->remaining, 1);1589	atomic_set(&r1_bio->behind_remaining, 0);1590 1591	first_clone = 1;1592 1593	for (i = 0; i < disks; i++) {1594		struct bio *mbio = NULL;1595		struct md_rdev *rdev = conf->mirrors[i].rdev;1596		if (!r1_bio->bios[i])1597			continue;1598 1599		if (first_clone) {1600			unsigned long max_write_behind =1601				mddev->bitmap_info.max_write_behind;1602			struct md_bitmap_stats stats;1603			int err;1604 1605			/* do behind I/O ?1606			 * Not if there are too many, or cannot1607			 * allocate memory, or a reader on WriteMostly1608			 * is waiting for behind writes to flush */1609			err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);1610			if (!err && write_behind && !stats.behind_wait &&1611			    stats.behind_writes < max_write_behind)1612				alloc_behind_master_bio(r1_bio, bio);1613 1614			mddev->bitmap_ops->startwrite(1615				mddev, r1_bio->sector, r1_bio->sectors,1616				test_bit(R1BIO_BehindIO, &r1_bio->state));1617			first_clone = 0;1618		}1619 1620		if (r1_bio->behind_master_bio) {1621			mbio = bio_alloc_clone(rdev->bdev,1622					       r1_bio->behind_master_bio,1623					       GFP_NOIO, &mddev->bio_set);1624			if (test_bit(CollisionCheck, &rdev->flags))1625				wait_for_serialization(rdev, r1_bio);1626			if (test_bit(WriteMostly, &rdev->flags))1627				atomic_inc(&r1_bio->behind_remaining);1628		} else {1629			mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO,1630					       &mddev->bio_set);1631 1632			if (mddev->serialize_policy)1633				wait_for_serialization(rdev, r1_bio);1634		}1635 1636		r1_bio->bios[i] = mbio;1637 1638		mbio->bi_iter.bi_sector	= (r1_bio->sector + rdev->data_offset);1639		mbio->bi_end_io	= raid1_end_write_request;1640		mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA));1641		if (test_bit(FailFast, &rdev->flags) &&1642		    !test_bit(WriteMostly, &rdev->flags) &&1643		    conf->raid_disks - mddev->degraded > 1)1644			mbio->bi_opf |= MD_FAILFAST;1645		mbio->bi_private = r1_bio;1646 1647		atomic_inc(&r1_bio->remaining);1648		mddev_trace_remap(mddev, mbio, r1_bio->sector);1649		/* flush_pending_writes() needs access to the rdev so...*/1650		mbio->bi_bdev = (void *)rdev;1651		if (!raid1_add_bio_to_plug(mddev, mbio, raid1_unplug, disks)) {1652			spin_lock_irqsave(&conf->device_lock, flags);1653			bio_list_add(&conf->pending_bio_list, mbio);1654			spin_unlock_irqrestore(&conf->device_lock, flags);1655			md_wakeup_thread(mddev->thread);1656		}1657	}1658 1659	r1_bio_write_done(r1_bio);1660 1661	/* In case raid1d snuck in to freeze_array */1662	wake_up_barrier(conf);1663}1664 1665static bool raid1_make_request(struct mddev *mddev, struct bio *bio)1666{1667	sector_t sectors;1668 1669	if (unlikely(bio->bi_opf & REQ_PREFLUSH)1670	    && md_flush_request(mddev, bio))1671		return true;1672 1673	/*1674	 * There is a limit to the maximum size, but1675	 * the read/write handler might find a lower limit1676	 * due to bad blocks.  To avoid multiple splits,1677	 * we pass the maximum number of sectors down1678	 * and let the lower level perform the split.1679	 */1680	sectors = align_to_barrier_unit_end(1681		bio->bi_iter.bi_sector, bio_sectors(bio));1682 1683	if (bio_data_dir(bio) == READ)1684		raid1_read_request(mddev, bio, sectors, NULL);1685	else {1686		md_write_start(mddev,bio);1687		raid1_write_request(mddev, bio, sectors);1688	}1689	return true;1690}1691 1692static void raid1_status(struct seq_file *seq, struct mddev *mddev)1693{1694	struct r1conf *conf = mddev->private;1695	int i;1696 1697	lockdep_assert_held(&mddev->lock);1698 1699	seq_printf(seq, " [%d/%d] [", conf->raid_disks,1700		   conf->raid_disks - mddev->degraded);1701	for (i = 0; i < conf->raid_disks; i++) {1702		struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev);1703 1704		seq_printf(seq, "%s",1705			   rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");1706	}1707	seq_printf(seq, "]");1708}1709 1710/**1711 * raid1_error() - RAID1 error handler.1712 * @mddev: affected md device.1713 * @rdev: member device to fail.1714 *1715 * The routine acknowledges &rdev failure and determines new @mddev state.1716 * If it failed, then:1717 *	- &MD_BROKEN flag is set in &mddev->flags.1718 *	- recovery is disabled.1719 * Otherwise, it must be degraded:1720 *	- recovery is interrupted.1721 *	- &mddev->degraded is bumped.1722 *1723 * @rdev is marked as &Faulty excluding case when array is failed and1724 * &mddev->fail_last_dev is off.1725 */1726static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)1727{1728	struct r1conf *conf = mddev->private;1729	unsigned long flags;1730 1731	spin_lock_irqsave(&conf->device_lock, flags);1732 1733	if (test_bit(In_sync, &rdev->flags) &&1734	    (conf->raid_disks - mddev->degraded) == 1) {1735		set_bit(MD_BROKEN, &mddev->flags);1736 1737		if (!mddev->fail_last_dev) {1738			conf->recovery_disabled = mddev->recovery_disabled;1739			spin_unlock_irqrestore(&conf->device_lock, flags);1740			return;1741		}1742	}1743	set_bit(Blocked, &rdev->flags);1744	if (test_and_clear_bit(In_sync, &rdev->flags))1745		mddev->degraded++;1746	set_bit(Faulty, &rdev->flags);1747	spin_unlock_irqrestore(&conf->device_lock, flags);1748	/*1749	 * if recovery is running, make sure it aborts.1750	 */1751	set_bit(MD_RECOVERY_INTR, &mddev->recovery);1752	set_mask_bits(&mddev->sb_flags, 0,1753		      BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));1754	pr_crit("md/raid1:%s: Disk failure on %pg, disabling device.\n"1755		"md/raid1:%s: Operation continuing on %d devices.\n",1756		mdname(mddev), rdev->bdev,1757		mdname(mddev), conf->raid_disks - mddev->degraded);1758}1759 1760static void print_conf(struct r1conf *conf)1761{1762	int i;1763 1764	pr_debug("RAID1 conf printout:\n");1765	if (!conf) {1766		pr_debug("(!conf)\n");1767		return;1768	}1769	pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,1770		 conf->raid_disks);1771 1772	lockdep_assert_held(&conf->mddev->reconfig_mutex);1773	for (i = 0; i < conf->raid_disks; i++) {1774		struct md_rdev *rdev = conf->mirrors[i].rdev;1775		if (rdev)1776			pr_debug(" disk %d, wo:%d, o:%d, dev:%pg\n",1777				 i, !test_bit(In_sync, &rdev->flags),1778				 !test_bit(Faulty, &rdev->flags),1779				 rdev->bdev);1780	}1781}1782 1783static void close_sync(struct r1conf *conf)1784{1785	int idx;1786 1787	for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) {1788		_wait_barrier(conf, idx, false);1789		_allow_barrier(conf, idx);1790	}1791 1792	mempool_exit(&conf->r1buf_pool);1793}1794 1795static int raid1_spare_active(struct mddev *mddev)1796{1797	int i;1798	struct r1conf *conf = mddev->private;1799	int count = 0;1800	unsigned long flags;1801 1802	/*1803	 * Find all failed disks within the RAID1 configuration1804	 * and mark them readable.1805	 * Called under mddev lock, so rcu protection not needed.1806	 * device_lock used to avoid races with raid1_end_read_request1807	 * which expects 'In_sync' flags and ->degraded to be consistent.1808	 */1809	spin_lock_irqsave(&conf->device_lock, flags);1810	for (i = 0; i < conf->raid_disks; i++) {1811		struct md_rdev *rdev = conf->mirrors[i].rdev;1812		struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;1813		if (repl1814		    && !test_bit(Candidate, &repl->flags)1815		    && repl->recovery_offset == MaxSector1816		    && !test_bit(Faulty, &repl->flags)1817		    && !test_and_set_bit(In_sync, &repl->flags)) {1818			/* replacement has just become active */1819			if (!rdev ||1820			    !test_and_clear_bit(In_sync, &rdev->flags))1821				count++;1822			if (rdev) {1823				/* Replaced device not technically1824				 * faulty, but we need to be sure1825				 * it gets removed and never re-added1826				 */1827				set_bit(Faulty, &rdev->flags);1828				sysfs_notify_dirent_safe(1829					rdev->sysfs_state);1830			}1831		}1832		if (rdev1833		    && rdev->recovery_offset == MaxSector1834		    && !test_bit(Faulty, &rdev->flags)1835		    && !test_and_set_bit(In_sync, &rdev->flags)) {1836			count++;1837			sysfs_notify_dirent_safe(rdev->sysfs_state);1838		}1839	}1840	mddev->degraded -= count;1841	spin_unlock_irqrestore(&conf->device_lock, flags);1842 1843	print_conf(conf);1844	return count;1845}1846 1847static bool raid1_add_conf(struct r1conf *conf, struct md_rdev *rdev, int disk,1848			   bool replacement)1849{1850	struct raid1_info *info = conf->mirrors + disk;1851 1852	if (replacement)1853		info += conf->raid_disks;1854 1855	if (info->rdev)1856		return false;1857 1858	if (bdev_nonrot(rdev->bdev)) {1859		set_bit(Nonrot, &rdev->flags);1860		WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks + 1);1861	}1862 1863	rdev->raid_disk = disk;1864	info->head_position = 0;1865	info->seq_start = MaxSector;1866	WRITE_ONCE(info->rdev, rdev);1867 1868	return true;1869}1870 1871static bool raid1_remove_conf(struct r1conf *conf, int disk)1872{1873	struct raid1_info *info = conf->mirrors + disk;1874	struct md_rdev *rdev = info->rdev;1875 1876	if (!rdev || test_bit(In_sync, &rdev->flags) ||1877	    atomic_read(&rdev->nr_pending))1878		return false;1879 1880	/* Only remove non-faulty devices if recovery is not possible. */1881	if (!test_bit(Faulty, &rdev->flags) &&1882	    rdev->mddev->recovery_disabled != conf->recovery_disabled &&1883	    rdev->mddev->degraded < conf->raid_disks)1884		return false;1885 1886	if (test_and_clear_bit(Nonrot, &rdev->flags))1887		WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks - 1);1888 1889	WRITE_ONCE(info->rdev, NULL);1890	return true;1891}1892 1893static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)1894{1895	struct r1conf *conf = mddev->private;1896	int err = -EEXIST;1897	int mirror = 0, repl_slot = -1;1898	struct raid1_info *p;1899	int first = 0;1900	int last = conf->raid_disks - 1;1901 1902	if (mddev->recovery_disabled == conf->recovery_disabled)1903		return -EBUSY;1904 1905	if (rdev->raid_disk >= 0)1906		first = last = rdev->raid_disk;1907 1908	/*1909	 * find the disk ... but prefer rdev->saved_raid_disk1910	 * if possible.1911	 */1912	if (rdev->saved_raid_disk >= 0 &&1913	    rdev->saved_raid_disk >= first &&1914	    rdev->saved_raid_disk < conf->raid_disks &&1915	    conf->mirrors[rdev->saved_raid_disk].rdev == NULL)1916		first = last = rdev->saved_raid_disk;1917 1918	for (mirror = first; mirror <= last; mirror++) {1919		p = conf->mirrors + mirror;1920		if (!p->rdev) {1921			err = mddev_stack_new_rdev(mddev, rdev);1922			if (err)1923				return err;1924 1925			raid1_add_conf(conf, rdev, mirror, false);1926			/* As all devices are equivalent, we don't need a full recovery1927			 * if this was recently any drive of the array1928			 */1929			if (rdev->saved_raid_disk < 0)1930				conf->fullsync = 1;1931			break;1932		}1933		if (test_bit(WantReplacement, &p->rdev->flags) &&1934		    p[conf->raid_disks].rdev == NULL && repl_slot < 0)1935			repl_slot = mirror;1936	}1937 1938	if (err && repl_slot >= 0) {1939		/* Add this device as a replacement */1940		clear_bit(In_sync, &rdev->flags);1941		set_bit(Replacement, &rdev->flags);1942		raid1_add_conf(conf, rdev, repl_slot, true);1943		err = 0;1944		conf->fullsync = 1;1945	}1946 1947	print_conf(conf);1948	return err;1949}1950 1951static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)1952{1953	struct r1conf *conf = mddev->private;1954	int err = 0;1955	int number = rdev->raid_disk;1956	struct raid1_info *p = conf->mirrors + number;1957 1958	if (unlikely(number >= conf->raid_disks))1959		goto abort;1960 1961	if (rdev != p->rdev) {1962		number += conf->raid_disks;1963		p = conf->mirrors + number;1964	}1965 1966	print_conf(conf);1967	if (rdev == p->rdev) {1968		if (!raid1_remove_conf(conf, number)) {1969			err = -EBUSY;1970			goto abort;1971		}1972 1973		if (number < conf->raid_disks &&1974		    conf->mirrors[conf->raid_disks + number].rdev) {1975			/* We just removed a device that is being replaced.1976			 * Move down the replacement.  We drain all IO before1977			 * doing this to avoid confusion.1978			 */1979			struct md_rdev *repl =1980				conf->mirrors[conf->raid_disks + number].rdev;1981			freeze_array(conf, 0);1982			if (atomic_read(&repl->nr_pending)) {1983				/* It means that some queued IO of retry_list1984				 * hold repl. Thus, we cannot set replacement1985				 * as NULL, avoiding rdev NULL pointer1986				 * dereference in sync_request_write and1987				 * handle_write_finished.1988				 */1989				err = -EBUSY;1990				unfreeze_array(conf);1991				goto abort;1992			}1993			clear_bit(Replacement, &repl->flags);1994			WRITE_ONCE(p->rdev, repl);1995			conf->mirrors[conf->raid_disks + number].rdev = NULL;1996			unfreeze_array(conf);1997		}1998 1999		clear_bit(WantReplacement, &rdev->flags);2000		err = md_integrity_register(mddev);2001	}2002abort:2003 2004	print_conf(conf);2005	return err;2006}2007 2008static void end_sync_read(struct bio *bio)2009{2010	struct r1bio *r1_bio = get_resync_r1bio(bio);2011 2012	update_head_pos(r1_bio->read_disk, r1_bio);2013 2014	/*2015	 * we have read a block, now it needs to be re-written,2016	 * or re-read if the read failed.2017	 * We don't do much here, just schedule handling by raid1d2018	 */2019	if (!bio->bi_status)2020		set_bit(R1BIO_Uptodate, &r1_bio->state);2021 2022	if (atomic_dec_and_test(&r1_bio->remaining))2023		reschedule_retry(r1_bio);2024}2025 2026static void abort_sync_write(struct mddev *mddev, struct r1bio *r1_bio)2027{2028	sector_t sync_blocks = 0;2029	sector_t s = r1_bio->sector;2030	long sectors_to_go = r1_bio->sectors;2031 2032	/* make sure these bits don't get cleared. */2033	do {2034		mddev->bitmap_ops->end_sync(mddev, s, &sync_blocks);2035		s += sync_blocks;2036		sectors_to_go -= sync_blocks;2037	} while (sectors_to_go > 0);2038}2039 2040static void put_sync_write_buf(struct r1bio *r1_bio, int uptodate)2041{2042	if (atomic_dec_and_test(&r1_bio->remaining)) {2043		struct mddev *mddev = r1_bio->mddev;2044		int s = r1_bio->sectors;2045 2046		if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||2047		    test_bit(R1BIO_WriteError, &r1_bio->state))2048			reschedule_retry(r1_bio);2049		else {2050			put_buf(r1_bio);2051			md_done_sync(mddev, s, uptodate);2052		}2053	}2054}2055 2056static void end_sync_write(struct bio *bio)2057{2058	int uptodate = !bio->bi_status;2059	struct r1bio *r1_bio = get_resync_r1bio(bio);2060	struct mddev *mddev = r1_bio->mddev;2061	struct r1conf *conf = mddev->private;2062	struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;2063 2064	if (!uptodate) {2065		abort_sync_write(mddev, r1_bio);2066		set_bit(WriteErrorSeen, &rdev->flags);2067		if (!test_and_set_bit(WantReplacement, &rdev->flags))2068			set_bit(MD_RECOVERY_NEEDED, &2069				mddev->recovery);2070		set_bit(R1BIO_WriteError, &r1_bio->state);2071	} else if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) &&2072		   !rdev_has_badblock(conf->mirrors[r1_bio->read_disk].rdev,2073				      r1_bio->sector, r1_bio->sectors)) {2074		set_bit(R1BIO_MadeGood, &r1_bio->state);2075	}2076 2077	put_sync_write_buf(r1_bio, uptodate);2078}2079 2080static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,2081			   int sectors, struct page *page, blk_opf_t rw)2082{2083	if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))2084		/* success */2085		return 1;2086	if (rw == REQ_OP_WRITE) {2087		set_bit(WriteErrorSeen, &rdev->flags);2088		if (!test_and_set_bit(WantReplacement,2089				      &rdev->flags))2090			set_bit(MD_RECOVERY_NEEDED, &2091				rdev->mddev->recovery);2092	}2093	/* need to record an error - either for the block or the device */2094	if (!rdev_set_badblocks(rdev, sector, sectors, 0))2095		md_error(rdev->mddev, rdev);2096	return 0;2097}2098 2099static int fix_sync_read_error(struct r1bio *r1_bio)2100{2101	/* Try some synchronous reads of other devices to get2102	 * good data, much like with normal read errors.  Only2103	 * read into the pages we already have so we don't2104	 * need to re-issue the read request.2105	 * We don't need to freeze the array, because being in an2106	 * active sync request, there is no normal IO, and2107	 * no overlapping syncs.2108	 * We don't need to check is_badblock() again as we2109	 * made sure that anything with a bad block in range2110	 * will have bi_end_io clear.2111	 */2112	struct mddev *mddev = r1_bio->mddev;2113	struct r1conf *conf = mddev->private;2114	struct bio *bio = r1_bio->bios[r1_bio->read_disk];2115	struct page **pages = get_resync_pages(bio)->pages;2116	sector_t sect = r1_bio->sector;2117	int sectors = r1_bio->sectors;2118	int idx = 0;2119	struct md_rdev *rdev;2120 2121	rdev = conf->mirrors[r1_bio->read_disk].rdev;2122	if (test_bit(FailFast, &rdev->flags)) {2123		/* Don't try recovering from here - just fail it2124		 * ... unless it is the last working device of course */2125		md_error(mddev, rdev);2126		if (test_bit(Faulty, &rdev->flags))2127			/* Don't try to read from here, but make sure2128			 * put_buf does it's thing2129			 */2130			bio->bi_end_io = end_sync_write;2131	}2132 2133	while(sectors) {2134		int s = sectors;2135		int d = r1_bio->read_disk;2136		int success = 0;2137		int start;2138 2139		if (s > (PAGE_SIZE>>9))2140			s = PAGE_SIZE >> 9;2141		do {2142			if (r1_bio->bios[d]->bi_end_io == end_sync_read) {2143				/* No rcu protection needed here devices2144				 * can only be removed when no resync is2145				 * active, and resync is currently active2146				 */2147				rdev = conf->mirrors[d].rdev;2148				if (sync_page_io(rdev, sect, s<<9,2149						 pages[idx],2150						 REQ_OP_READ, false)) {2151					success = 1;2152					break;2153				}2154			}2155			d++;2156			if (d == conf->raid_disks * 2)2157				d = 0;2158		} while (!success && d != r1_bio->read_disk);2159 2160		if (!success) {2161			int abort = 0;2162			/* Cannot read from anywhere, this block is lost.2163			 * Record a bad block on each device.  If that doesn't2164			 * work just disable and interrupt the recovery.2165			 * Don't fail devices as that won't really help.2166			 */2167			pr_crit_ratelimited("md/raid1:%s: %pg: unrecoverable I/O read error for block %llu\n",2168					    mdname(mddev), bio->bi_bdev,2169					    (unsigned long long)r1_bio->sector);2170			for (d = 0; d < conf->raid_disks * 2; d++) {2171				rdev = conf->mirrors[d].rdev;2172				if (!rdev || test_bit(Faulty, &rdev->flags))2173					continue;2174				if (!rdev_set_badblocks(rdev, sect, s, 0))2175					abort = 1;2176			}2177			if (abort) {2178				conf->recovery_disabled =2179					mddev->recovery_disabled;2180				set_bit(MD_RECOVERY_INTR, &mddev->recovery);2181				md_done_sync(mddev, r1_bio->sectors, 0);2182				put_buf(r1_bio);2183				return 0;2184			}2185			/* Try next page */2186			sectors -= s;2187			sect += s;2188			idx++;2189			continue;2190		}2191 2192		start = d;2193		/* write it back and re-read */2194		while (d != r1_bio->read_disk) {2195			if (d == 0)2196				d = conf->raid_disks * 2;2197			d--;2198			if (r1_bio->bios[d]->bi_end_io != end_sync_read)2199				continue;2200			rdev = conf->mirrors[d].rdev;2201			if (r1_sync_page_io(rdev, sect, s,2202					    pages[idx],2203					    REQ_OP_WRITE) == 0) {2204				r1_bio->bios[d]->bi_end_io = NULL;2205				rdev_dec_pending(rdev, mddev);2206			}2207		}2208		d = start;2209		while (d != r1_bio->read_disk) {2210			if (d == 0)2211				d = conf->raid_disks * 2;2212			d--;2213			if (r1_bio->bios[d]->bi_end_io != end_sync_read)2214				continue;2215			rdev = conf->mirrors[d].rdev;2216			if (r1_sync_page_io(rdev, sect, s,2217					    pages[idx],2218					    REQ_OP_READ) != 0)2219				atomic_add(s, &rdev->corrected_errors);2220		}2221		sectors -= s;2222		sect += s;2223		idx ++;2224	}2225	set_bit(R1BIO_Uptodate, &r1_bio->state);2226	bio->bi_status = 0;2227	return 1;2228}2229 2230static void process_checks(struct r1bio *r1_bio)2231{2232	/* We have read all readable devices.  If we haven't2233	 * got the block, then there is no hope left.2234	 * If we have, then we want to do a comparison2235	 * and skip the write if everything is the same.2236	 * If any blocks failed to read, then we need to2237	 * attempt an over-write2238	 */2239	struct mddev *mddev = r1_bio->mddev;2240	struct r1conf *conf = mddev->private;2241	int primary;2242	int i;2243	int vcnt;2244 2245	/* Fix variable parts of all bios */2246	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);2247	for (i = 0; i < conf->raid_disks * 2; i++) {2248		blk_status_t status;2249		struct bio *b = r1_bio->bios[i];2250		struct resync_pages *rp = get_resync_pages(b);2251		if (b->bi_end_io != end_sync_read)2252			continue;2253		/* fixup the bio for reuse, but preserve errno */2254		status = b->bi_status;2255		bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ);2256		b->bi_status = status;2257		b->bi_iter.bi_sector = r1_bio->sector +2258			conf->mirrors[i].rdev->data_offset;2259		b->bi_end_io = end_sync_read;2260		rp->raid_bio = r1_bio;2261		b->bi_private = rp;2262 2263		/* initialize bvec table again */2264		md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9);2265	}2266	for (primary = 0; primary < conf->raid_disks * 2; primary++)2267		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&2268		    !r1_bio->bios[primary]->bi_status) {2269			r1_bio->bios[primary]->bi_end_io = NULL;2270			rdev_dec_pending(conf->mirrors[primary].rdev, mddev);2271			break;2272		}2273	r1_bio->read_disk = primary;2274	for (i = 0; i < conf->raid_disks * 2; i++) {2275		int j = 0;2276		struct bio *pbio = r1_bio->bios[primary];2277		struct bio *sbio = r1_bio->bios[i];2278		blk_status_t status = sbio->bi_status;2279		struct page **ppages = get_resync_pages(pbio)->pages;2280		struct page **spages = get_resync_pages(sbio)->pages;2281		struct bio_vec *bi;2282		int page_len[RESYNC_PAGES] = { 0 };2283		struct bvec_iter_all iter_all;2284 2285		if (sbio->bi_end_io != end_sync_read)2286			continue;2287		/* Now we can 'fixup' the error value */2288		sbio->bi_status = 0;2289 2290		bio_for_each_segment_all(bi, sbio, iter_all)2291			page_len[j++] = bi->bv_len;2292 2293		if (!status) {2294			for (j = vcnt; j-- ; ) {2295				if (memcmp(page_address(ppages[j]),2296					   page_address(spages[j]),2297					   page_len[j]))2298					break;2299			}2300		} else2301			j = 0;2302		if (j >= 0)2303			atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);2304		if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)2305			      && !status)) {2306			/* No need to write to this device. */2307			sbio->bi_end_io = NULL;2308			rdev_dec_pending(conf->mirrors[i].rdev, mddev);2309			continue;2310		}2311 2312		bio_copy_data(sbio, pbio);2313	}2314}2315 2316static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)2317{2318	struct r1conf *conf = mddev->private;2319	int i;2320	int disks = conf->raid_disks * 2;2321	struct bio *wbio;2322 2323	if (!test_bit(R1BIO_Uptodate, &r1_bio->state))2324		/* ouch - failed to read all of that. */2325		if (!fix_sync_read_error(r1_bio))2326			return;2327 2328	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))2329		process_checks(r1_bio);2330 2331	/*2332	 * schedule writes2333	 */2334	atomic_set(&r1_bio->remaining, 1);2335	for (i = 0; i < disks ; i++) {2336		wbio = r1_bio->bios[i];2337		if (wbio->bi_end_io == NULL ||2338		    (wbio->bi_end_io == end_sync_read &&2339		     (i == r1_bio->read_disk ||2340		      !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))2341			continue;2342		if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) {2343			abort_sync_write(mddev, r1_bio);2344			continue;2345		}2346 2347		wbio->bi_opf = REQ_OP_WRITE;2348		if (test_bit(FailFast, &conf->mirrors[i].rdev->flags))2349			wbio->bi_opf |= MD_FAILFAST;2350 2351		wbio->bi_end_io = end_sync_write;2352		atomic_inc(&r1_bio->remaining);2353		md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));2354 2355		submit_bio_noacct(wbio);2356	}2357 2358	put_sync_write_buf(r1_bio, 1);2359}2360 2361/*2362 * This is a kernel thread which:2363 *2364 *	1.	Retries failed read operations on working mirrors.2365 *	2.	Updates the raid superblock when problems encounter.2366 *	3.	Performs writes following reads for array synchronising.2367 */2368 2369static void fix_read_error(struct r1conf *conf, struct r1bio *r1_bio)2370{2371	sector_t sect = r1_bio->sector;2372	int sectors = r1_bio->sectors;2373	int read_disk = r1_bio->read_disk;2374	struct mddev *mddev = conf->mddev;2375	struct md_rdev *rdev = conf->mirrors[read_disk].rdev;2376 2377	if (exceed_read_errors(mddev, rdev)) {2378		r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;2379		return;2380	}2381 2382	while(sectors) {2383		int s = sectors;2384		int d = read_disk;2385		int success = 0;2386		int start;2387 2388		if (s > (PAGE_SIZE>>9))2389			s = PAGE_SIZE >> 9;2390 2391		do {2392			rdev = conf->mirrors[d].rdev;2393			if (rdev &&2394			    (test_bit(In_sync, &rdev->flags) ||2395			     (!test_bit(Faulty, &rdev->flags) &&2396			      rdev->recovery_offset >= sect + s)) &&2397			    rdev_has_badblock(rdev, sect, s) == 0) {2398				atomic_inc(&rdev->nr_pending);2399				if (sync_page_io(rdev, sect, s<<9,2400					 conf->tmppage, REQ_OP_READ, false))2401					success = 1;2402				rdev_dec_pending(rdev, mddev);2403				if (success)2404					break;2405			}2406 2407			d++;2408			if (d == conf->raid_disks * 2)2409				d = 0;2410		} while (d != read_disk);2411 2412		if (!success) {2413			/* Cannot read from anywhere - mark it bad */2414			struct md_rdev *rdev = conf->mirrors[read_disk].rdev;2415			if (!rdev_set_badblocks(rdev, sect, s, 0))2416				md_error(mddev, rdev);2417			break;2418		}2419		/* write it back and re-read */2420		start = d;2421		while (d != read_disk) {2422			if (d==0)2423				d = conf->raid_disks * 2;2424			d--;2425			rdev = conf->mirrors[d].rdev;2426			if (rdev &&2427			    !test_bit(Faulty, &rdev->flags)) {2428				atomic_inc(&rdev->nr_pending);2429				r1_sync_page_io(rdev, sect, s,2430						conf->tmppage, REQ_OP_WRITE);2431				rdev_dec_pending(rdev, mddev);2432			}2433		}2434		d = start;2435		while (d != read_disk) {2436			if (d==0)2437				d = conf->raid_disks * 2;2438			d--;2439			rdev = conf->mirrors[d].rdev;2440			if (rdev &&2441			    !test_bit(Faulty, &rdev->flags)) {2442				atomic_inc(&rdev->nr_pending);2443				if (r1_sync_page_io(rdev, sect, s,2444						conf->tmppage, REQ_OP_READ)) {2445					atomic_add(s, &rdev->corrected_errors);2446					pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %pg)\n",2447						mdname(mddev), s,2448						(unsigned long long)(sect +2449								     rdev->data_offset),2450						rdev->bdev);2451				}2452				rdev_dec_pending(rdev, mddev);2453			}2454		}2455		sectors -= s;2456		sect += s;2457	}2458}2459 2460static int narrow_write_error(struct r1bio *r1_bio, int i)2461{2462	struct mddev *mddev = r1_bio->mddev;2463	struct r1conf *conf = mddev->private;2464	struct md_rdev *rdev = conf->mirrors[i].rdev;2465 2466	/* bio has the data to be written to device 'i' where2467	 * we just recently had a write error.2468	 * We repeatedly clone the bio and trim down to one block,2469	 * then try the write.  Where the write fails we record2470	 * a bad block.2471	 * It is conceivable that the bio doesn't exactly align with2472	 * blocks.  We must handle this somehow.2473	 *2474	 * We currently own a reference on the rdev.2475	 */2476 2477	int block_sectors;2478	sector_t sector;2479	int sectors;2480	int sect_to_write = r1_bio->sectors;2481	int ok = 1;2482 2483	if (rdev->badblocks.shift < 0)2484		return 0;2485 2486	block_sectors = roundup(1 << rdev->badblocks.shift,2487				bdev_logical_block_size(rdev->bdev) >> 9);2488	sector = r1_bio->sector;2489	sectors = ((sector + block_sectors)2490		   & ~(sector_t)(block_sectors - 1))2491		- sector;2492 2493	while (sect_to_write) {2494		struct bio *wbio;2495		if (sectors > sect_to_write)2496			sectors = sect_to_write;2497		/* Write at 'sector' for 'sectors'*/2498 2499		if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {2500			wbio = bio_alloc_clone(rdev->bdev,2501					       r1_bio->behind_master_bio,2502					       GFP_NOIO, &mddev->bio_set);2503		} else {2504			wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio,2505					       GFP_NOIO, &mddev->bio_set);2506		}2507 2508		wbio->bi_opf = REQ_OP_WRITE;2509		wbio->bi_iter.bi_sector = r1_bio->sector;2510		wbio->bi_iter.bi_size = r1_bio->sectors << 9;2511 2512		bio_trim(wbio, sector - r1_bio->sector, sectors);2513		wbio->bi_iter.bi_sector += rdev->data_offset;2514 2515		if (submit_bio_wait(wbio) < 0)2516			/* failure! */2517			ok = rdev_set_badblocks(rdev, sector,2518						sectors, 0)2519				&& ok;2520 2521		bio_put(wbio);2522		sect_to_write -= sectors;2523		sector += sectors;2524		sectors = block_sectors;2525	}2526	return ok;2527}2528 2529static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)2530{2531	int m;2532	int s = r1_bio->sectors;2533	for (m = 0; m < conf->raid_disks * 2 ; m++) {2534		struct md_rdev *rdev = conf->mirrors[m].rdev;2535		struct bio *bio = r1_bio->bios[m];2536		if (bio->bi_end_io == NULL)2537			continue;2538		if (!bio->bi_status &&2539		    test_bit(R1BIO_MadeGood, &r1_bio->state)) {2540			rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);2541		}2542		if (bio->bi_status &&2543		    test_bit(R1BIO_WriteError, &r1_bio->state)) {2544			if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))2545				md_error(conf->mddev, rdev);2546		}2547	}2548	put_buf(r1_bio);2549	md_done_sync(conf->mddev, s, 1);2550}2551 2552static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)2553{2554	int m, idx;2555	bool fail = false;2556 2557	for (m = 0; m < conf->raid_disks * 2 ; m++)2558		if (r1_bio->bios[m] == IO_MADE_GOOD) {2559			struct md_rdev *rdev = conf->mirrors[m].rdev;2560			rdev_clear_badblocks(rdev,2561					     r1_bio->sector,2562					     r1_bio->sectors, 0);2563			rdev_dec_pending(rdev, conf->mddev);2564		} else if (r1_bio->bios[m] != NULL) {2565			/* This drive got a write error.  We need to2566			 * narrow down and record precise write2567			 * errors.2568			 */2569			fail = true;2570			if (!narrow_write_error(r1_bio, m)) {2571				md_error(conf->mddev,2572					 conf->mirrors[m].rdev);2573				/* an I/O failed, we can't clear the bitmap */2574				set_bit(R1BIO_Degraded, &r1_bio->state);2575			}2576			rdev_dec_pending(conf->mirrors[m].rdev,2577					 conf->mddev);2578		}2579	if (fail) {2580		spin_lock_irq(&conf->device_lock);2581		list_add(&r1_bio->retry_list, &conf->bio_end_io_list);2582		idx = sector_to_idx(r1_bio->sector);2583		atomic_inc(&conf->nr_queued[idx]);2584		spin_unlock_irq(&conf->device_lock);2585		/*2586		 * In case freeze_array() is waiting for condition2587		 * get_unqueued_pending() == extra to be true.2588		 */2589		wake_up(&conf->wait_barrier);2590		md_wakeup_thread(conf->mddev->thread);2591	} else {2592		if (test_bit(R1BIO_WriteError, &r1_bio->state))2593			close_write(r1_bio);2594		raid_end_bio_io(r1_bio);2595	}2596}2597 2598static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)2599{2600	struct mddev *mddev = conf->mddev;2601	struct bio *bio;2602	struct md_rdev *rdev;2603	sector_t sector;2604 2605	clear_bit(R1BIO_ReadError, &r1_bio->state);2606	/* we got a read error. Maybe the drive is bad.  Maybe just2607	 * the block and we can fix it.2608	 * We freeze all other IO, and try reading the block from2609	 * other devices.  When we find one, we re-write2610	 * and check it that fixes the read error.2611	 * This is all done synchronously while the array is2612	 * frozen2613	 */2614 2615	bio = r1_bio->bios[r1_bio->read_disk];2616	bio_put(bio);2617	r1_bio->bios[r1_bio->read_disk] = NULL;2618 2619	rdev = conf->mirrors[r1_bio->read_disk].rdev;2620	if (mddev->ro == 02621	    && !test_bit(FailFast, &rdev->flags)) {2622		freeze_array(conf, 1);2623		fix_read_error(conf, r1_bio);2624		unfreeze_array(conf);2625	} else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) {2626		md_error(mddev, rdev);2627	} else {2628		r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;2629	}2630 2631	rdev_dec_pending(rdev, conf->mddev);2632	sector = r1_bio->sector;2633	bio = r1_bio->master_bio;2634 2635	/* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */2636	r1_bio->state = 0;2637	raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio);2638	allow_barrier(conf, sector);2639}2640 2641static void raid1d(struct md_thread *thread)2642{2643	struct mddev *mddev = thread->mddev;2644	struct r1bio *r1_bio;2645	unsigned long flags;2646	struct r1conf *conf = mddev->private;2647	struct list_head *head = &conf->retry_list;2648	struct blk_plug plug;2649	int idx;2650 2651	md_check_recovery(mddev);2652 2653	if (!list_empty_careful(&conf->bio_end_io_list) &&2654	    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {2655		LIST_HEAD(tmp);2656		spin_lock_irqsave(&conf->device_lock, flags);2657		if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))2658			list_splice_init(&conf->bio_end_io_list, &tmp);2659		spin_unlock_irqrestore(&conf->device_lock, flags);2660		while (!list_empty(&tmp)) {2661			r1_bio = list_first_entry(&tmp, struct r1bio,2662						  retry_list);2663			list_del(&r1_bio->retry_list);2664			idx = sector_to_idx(r1_bio->sector);2665			atomic_dec(&conf->nr_queued[idx]);2666			if (mddev->degraded)2667				set_bit(R1BIO_Degraded, &r1_bio->state);2668			if (test_bit(R1BIO_WriteError, &r1_bio->state))2669				close_write(r1_bio);2670			raid_end_bio_io(r1_bio);2671		}2672	}2673 2674	blk_start_plug(&plug);2675	for (;;) {2676 2677		flush_pending_writes(conf);2678 2679		spin_lock_irqsave(&conf->device_lock, flags);2680		if (list_empty(head)) {2681			spin_unlock_irqrestore(&conf->device_lock, flags);2682			break;2683		}2684		r1_bio = list_entry(head->prev, struct r1bio, retry_list);2685		list_del(head->prev);2686		idx = sector_to_idx(r1_bio->sector);2687		atomic_dec(&conf->nr_queued[idx]);2688		spin_unlock_irqrestore(&conf->device_lock, flags);2689 2690		mddev = r1_bio->mddev;2691		conf = mddev->private;2692		if (test_bit(R1BIO_IsSync, &r1_bio->state)) {2693			if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||2694			    test_bit(R1BIO_WriteError, &r1_bio->state))2695				handle_sync_write_finished(conf, r1_bio);2696			else2697				sync_request_write(mddev, r1_bio);2698		} else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||2699			   test_bit(R1BIO_WriteError, &r1_bio->state))2700			handle_write_finished(conf, r1_bio);2701		else if (test_bit(R1BIO_ReadError, &r1_bio->state))2702			handle_read_error(conf, r1_bio);2703		else2704			WARN_ON_ONCE(1);2705 2706		cond_resched();2707		if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING))2708			md_check_recovery(mddev);2709	}2710	blk_finish_plug(&plug);2711}2712 2713static int init_resync(struct r1conf *conf)2714{2715	int buffs;2716 2717	buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;2718	BUG_ON(mempool_initialized(&conf->r1buf_pool));2719 2720	return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc,2721			    r1buf_pool_free, conf->poolinfo);2722}2723 2724static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf)2725{2726	struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO);2727	struct resync_pages *rps;2728	struct bio *bio;2729	int i;2730 2731	for (i = conf->poolinfo->raid_disks; i--; ) {2732		bio = r1bio->bios[i];2733		rps = bio->bi_private;2734		bio_reset(bio, NULL, 0);2735		bio->bi_private = rps;2736	}2737	r1bio->master_bio = NULL;2738	return r1bio;2739}2740 2741/*2742 * perform a "sync" on one "block"2743 *2744 * We need to make sure that no normal I/O request - particularly write2745 * requests - conflict with active sync requests.2746 *2747 * This is achieved by tracking pending requests and a 'barrier' concept2748 * that can be installed to exclude normal IO requests.2749 */2750 2751static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,2752				   sector_t max_sector, int *skipped)2753{2754	struct r1conf *conf = mddev->private;2755	struct r1bio *r1_bio;2756	struct bio *bio;2757	sector_t nr_sectors;2758	int disk = -1;2759	int i;2760	int wonly = -1;2761	int write_targets = 0, read_targets = 0;2762	sector_t sync_blocks;2763	bool still_degraded = false;2764	int good_sectors = RESYNC_SECTORS;2765	int min_bad = 0; /* number of sectors that are bad in all devices */2766	int idx = sector_to_idx(sector_nr);2767	int page_idx = 0;2768 2769	if (!mempool_initialized(&conf->r1buf_pool))2770		if (init_resync(conf))2771			return 0;2772 2773	if (sector_nr >= max_sector) {2774		/* If we aborted, we need to abort the2775		 * sync on the 'current' bitmap chunk (there will2776		 * only be one in raid1 resync.2777		 * We can find the current addess in mddev->curr_resync2778		 */2779		if (mddev->curr_resync < max_sector) /* aborted */2780			mddev->bitmap_ops->end_sync(mddev, mddev->curr_resync,2781						    &sync_blocks);2782		else /* completed sync */2783			conf->fullsync = 0;2784 2785		mddev->bitmap_ops->close_sync(mddev);2786		close_sync(conf);2787 2788		if (mddev_is_clustered(mddev)) {2789			conf->cluster_sync_low = 0;2790			conf->cluster_sync_high = 0;2791		}2792		return 0;2793	}2794 2795	if (mddev->bitmap == NULL &&2796	    mddev->recovery_cp == MaxSector &&2797	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&2798	    conf->fullsync == 0) {2799		*skipped = 1;2800		return max_sector - sector_nr;2801	}2802	/* before building a request, check if we can skip these blocks..2803	 * This call the bitmap_start_sync doesn't actually record anything2804	 */2805	if (!mddev->bitmap_ops->start_sync(mddev, sector_nr, &sync_blocks, true) &&2806	    !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {2807		/* We can skip this block, and probably several more */2808		*skipped = 1;2809		return sync_blocks;2810	}2811 2812	/*2813	 * If there is non-resync activity waiting for a turn, then let it2814	 * though before starting on this new sync request.2815	 */2816	if (atomic_read(&conf->nr_waiting[idx]))2817		schedule_timeout_uninterruptible(1);2818 2819	/* we are incrementing sector_nr below. To be safe, we check against2820	 * sector_nr + two times RESYNC_SECTORS2821	 */2822 2823	mddev->bitmap_ops->cond_end_sync(mddev, sector_nr,2824		mddev_is_clustered(mddev) &&2825		(sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));2826 2827	if (raise_barrier(conf, sector_nr))2828		return 0;2829 2830	r1_bio = raid1_alloc_init_r1buf(conf);2831 2832	/*2833	 * If we get a correctably read error during resync or recovery,2834	 * we might want to read from a different device.  So we2835	 * flag all drives that could conceivably be read from for READ,2836	 * and any others (which will be non-In_sync devices) for WRITE.2837	 * If a read fails, we try reading from something else for which READ2838	 * is OK.2839	 */2840 2841	r1_bio->mddev = mddev;2842	r1_bio->sector = sector_nr;2843	r1_bio->state = 0;2844	set_bit(R1BIO_IsSync, &r1_bio->state);2845	/* make sure good_sectors won't go across barrier unit boundary */2846	good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors);2847 2848	for (i = 0; i < conf->raid_disks * 2; i++) {2849		struct md_rdev *rdev;2850		bio = r1_bio->bios[i];2851 2852		rdev = conf->mirrors[i].rdev;2853		if (rdev == NULL ||2854		    test_bit(Faulty, &rdev->flags)) {2855			if (i < conf->raid_disks)2856				still_degraded = true;2857		} else if (!test_bit(In_sync, &rdev->flags)) {2858			bio->bi_opf = REQ_OP_WRITE;2859			bio->bi_end_io = end_sync_write;2860			write_targets ++;2861		} else {2862			/* may need to read from here */2863			sector_t first_bad = MaxSector;2864			int bad_sectors;2865 2866			if (is_badblock(rdev, sector_nr, good_sectors,2867					&first_bad, &bad_sectors)) {2868				if (first_bad > sector_nr)2869					good_sectors = first_bad - sector_nr;2870				else {2871					bad_sectors -= (sector_nr - first_bad);2872					if (min_bad == 0 ||2873					    min_bad > bad_sectors)2874						min_bad = bad_sectors;2875				}2876			}2877			if (sector_nr < first_bad) {2878				if (test_bit(WriteMostly, &rdev->flags)) {2879					if (wonly < 0)2880						wonly = i;2881				} else {2882					if (disk < 0)2883						disk = i;2884				}2885				bio->bi_opf = REQ_OP_READ;2886				bio->bi_end_io = end_sync_read;2887				read_targets++;2888			} else if (!test_bit(WriteErrorSeen, &rdev->flags) &&2889				test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&2890				!test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {2891				/*2892				 * The device is suitable for reading (InSync),2893				 * but has bad block(s) here. Let's try to correct them,2894				 * if we are doing resync or repair. Otherwise, leave2895				 * this device alone for this sync request.2896				 */2897				bio->bi_opf = REQ_OP_WRITE;2898				bio->bi_end_io = end_sync_write;2899				write_targets++;2900			}2901		}2902		if (rdev && bio->bi_end_io) {2903			atomic_inc(&rdev->nr_pending);2904			bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;2905			bio_set_dev(bio, rdev->bdev);2906			if (test_bit(FailFast, &rdev->flags))2907				bio->bi_opf |= MD_FAILFAST;2908		}2909	}2910	if (disk < 0)2911		disk = wonly;2912	r1_bio->read_disk = disk;2913 2914	if (read_targets == 0 && min_bad > 0) {2915		/* These sectors are bad on all InSync devices, so we2916		 * need to mark them bad on all write targets2917		 */2918		int ok = 1;2919		for (i = 0 ; i < conf->raid_disks * 2 ; i++)2920			if (r1_bio->bios[i]->bi_end_io == end_sync_write) {2921				struct md_rdev *rdev = conf->mirrors[i].rdev;2922				ok = rdev_set_badblocks(rdev, sector_nr,2923							min_bad, 02924					) && ok;2925			}2926		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);2927		*skipped = 1;2928		put_buf(r1_bio);2929 2930		if (!ok) {2931			/* Cannot record the badblocks, so need to2932			 * abort the resync.2933			 * If there are multiple read targets, could just2934			 * fail the really bad ones ???2935			 */2936			conf->recovery_disabled = mddev->recovery_disabled;2937			set_bit(MD_RECOVERY_INTR, &mddev->recovery);2938			return 0;2939		} else2940			return min_bad;2941 2942	}2943	if (min_bad > 0 && min_bad < good_sectors) {2944		/* only resync enough to reach the next bad->good2945		 * transition */2946		good_sectors = min_bad;2947	}2948 2949	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)2950		/* extra read targets are also write targets */2951		write_targets += read_targets-1;2952 2953	if (write_targets == 0 || read_targets == 0) {2954		/* There is nowhere to write, so all non-sync2955		 * drives must be failed - so we are finished2956		 */2957		sector_t rv;2958		if (min_bad > 0)2959			max_sector = sector_nr + min_bad;2960		rv = max_sector - sector_nr;2961		*skipped = 1;2962		put_buf(r1_bio);2963		return rv;2964	}2965 2966	if (max_sector > mddev->resync_max)2967		max_sector = mddev->resync_max; /* Don't do IO beyond here */2968	if (max_sector > sector_nr + good_sectors)2969		max_sector = sector_nr + good_sectors;2970	nr_sectors = 0;2971	sync_blocks = 0;2972	do {2973		struct page *page;2974		int len = PAGE_SIZE;2975		if (sector_nr + (len>>9) > max_sector)2976			len = (max_sector - sector_nr) << 9;2977		if (len == 0)2978			break;2979		if (sync_blocks == 0) {2980			if (!mddev->bitmap_ops->start_sync(mddev, sector_nr,2981						&sync_blocks, still_degraded) &&2982			    !conf->fullsync &&2983			    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))2984				break;2985			if ((len >> 9) > sync_blocks)2986				len = sync_blocks<<9;2987		}2988 2989		for (i = 0 ; i < conf->raid_disks * 2; i++) {2990			struct resync_pages *rp;2991 2992			bio = r1_bio->bios[i];2993			rp = get_resync_pages(bio);2994			if (bio->bi_end_io) {2995				page = resync_fetch_page(rp, page_idx);2996 2997				/*2998				 * won't fail because the vec table is big2999				 * enough to hold all these pages3000				 */3001				__bio_add_page(bio, page, len, 0);3002			}3003		}3004		nr_sectors += len>>9;3005		sector_nr += len>>9;3006		sync_blocks -= (len>>9);3007	} while (++page_idx < RESYNC_PAGES);3008 3009	r1_bio->sectors = nr_sectors;3010 3011	if (mddev_is_clustered(mddev) &&3012			conf->cluster_sync_high < sector_nr + nr_sectors) {3013		conf->cluster_sync_low = mddev->curr_resync_completed;3014		conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;3015		/* Send resync message */3016		md_cluster_ops->resync_info_update(mddev,3017				conf->cluster_sync_low,3018				conf->cluster_sync_high);3019	}3020 3021	/* For a user-requested sync, we read all readable devices and do a3022	 * compare3023	 */3024	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {3025		atomic_set(&r1_bio->remaining, read_targets);3026		for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {3027			bio = r1_bio->bios[i];3028			if (bio->bi_end_io == end_sync_read) {3029				read_targets--;3030				md_sync_acct_bio(bio, nr_sectors);3031				if (read_targets == 1)3032					bio->bi_opf &= ~MD_FAILFAST;3033				submit_bio_noacct(bio);3034			}3035		}3036	} else {3037		atomic_set(&r1_bio->remaining, 1);3038		bio = r1_bio->bios[r1_bio->read_disk];3039		md_sync_acct_bio(bio, nr_sectors);3040		if (read_targets == 1)3041			bio->bi_opf &= ~MD_FAILFAST;3042		submit_bio_noacct(bio);3043	}3044	return nr_sectors;3045}3046 3047static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)3048{3049	if (sectors)3050		return sectors;3051 3052	return mddev->dev_sectors;3053}3054 3055static struct r1conf *setup_conf(struct mddev *mddev)3056{3057	struct r1conf *conf;3058	int i;3059	struct raid1_info *disk;3060	struct md_rdev *rdev;3061	int err = -ENOMEM;3062 3063	conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);3064	if (!conf)3065		goto abort;3066 3067	conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR,3068				   sizeof(atomic_t), GFP_KERNEL);3069	if (!conf->nr_pending)3070		goto abort;3071 3072	conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR,3073				   sizeof(atomic_t), GFP_KERNEL);3074	if (!conf->nr_waiting)3075		goto abort;3076 3077	conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR,3078				  sizeof(atomic_t), GFP_KERNEL);3079	if (!conf->nr_queued)3080		goto abort;3081 3082	conf->barrier = kcalloc(BARRIER_BUCKETS_NR,3083				sizeof(atomic_t), GFP_KERNEL);3084	if (!conf->barrier)3085		goto abort;3086 3087	conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info),3088					    mddev->raid_disks, 2),3089				GFP_KERNEL);3090	if (!conf->mirrors)3091		goto abort;3092 3093	conf->tmppage = alloc_page(GFP_KERNEL);3094	if (!conf->tmppage)3095		goto abort;3096 3097	conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);3098	if (!conf->poolinfo)3099		goto abort;3100	conf->poolinfo->raid_disks = mddev->raid_disks * 2;3101	err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc,3102			   rbio_pool_free, conf->poolinfo);3103	if (err)3104		goto abort;3105 3106	err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0);3107	if (err)3108		goto abort;3109 3110	conf->poolinfo->mddev = mddev;3111 3112	err = -EINVAL;3113	spin_lock_init(&conf->device_lock);3114	conf->raid_disks = mddev->raid_disks;3115	rdev_for_each(rdev, mddev) {3116		int disk_idx = rdev->raid_disk;3117 3118		if (disk_idx >= conf->raid_disks || disk_idx < 0)3119			continue;3120 3121		if (!raid1_add_conf(conf, rdev, disk_idx,3122				    test_bit(Replacement, &rdev->flags)))3123			goto abort;3124	}3125	conf->mddev = mddev;3126	INIT_LIST_HEAD(&conf->retry_list);3127	INIT_LIST_HEAD(&conf->bio_end_io_list);3128 3129	spin_lock_init(&conf->resync_lock);3130	init_waitqueue_head(&conf->wait_barrier);3131 3132	bio_list_init(&conf->pending_bio_list);3133	conf->recovery_disabled = mddev->recovery_disabled - 1;3134 3135	err = -EIO;3136	for (i = 0; i < conf->raid_disks * 2; i++) {3137 3138		disk = conf->mirrors + i;3139 3140		if (i < conf->raid_disks &&3141		    disk[conf->raid_disks].rdev) {3142			/* This slot has a replacement. */3143			if (!disk->rdev) {3144				/* No original, just make the replacement3145				 * a recovering spare3146				 */3147				disk->rdev =3148					disk[conf->raid_disks].rdev;3149				disk[conf->raid_disks].rdev = NULL;3150			} else if (!test_bit(In_sync, &disk->rdev->flags))3151				/* Original is not in_sync - bad */3152				goto abort;3153		}3154 3155		if (!disk->rdev ||3156		    !test_bit(In_sync, &disk->rdev->flags)) {3157			disk->head_position = 0;3158			if (disk->rdev &&3159			    (disk->rdev->saved_raid_disk < 0))3160				conf->fullsync = 1;3161		}3162	}3163 3164	err = -ENOMEM;3165	rcu_assign_pointer(conf->thread,3166			   md_register_thread(raid1d, mddev, "raid1"));3167	if (!conf->thread)3168		goto abort;3169 3170	return conf;3171 3172 abort:3173	if (conf) {3174		mempool_exit(&conf->r1bio_pool);3175		kfree(conf->mirrors);3176		safe_put_page(conf->tmppage);3177		kfree(conf->poolinfo);3178		kfree(conf->nr_pending);3179		kfree(conf->nr_waiting);3180		kfree(conf->nr_queued);3181		kfree(conf->barrier);3182		bioset_exit(&conf->bio_split);3183		kfree(conf);3184	}3185	return ERR_PTR(err);3186}3187 3188static int raid1_set_limits(struct mddev *mddev)3189{3190	struct queue_limits lim;3191	int err;3192 3193	md_init_stacking_limits(&lim);3194	lim.max_write_zeroes_sectors = 0;3195	err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);3196	if (err) {3197		queue_limits_cancel_update(mddev->gendisk->queue);3198		return err;3199	}3200	return queue_limits_set(mddev->gendisk->queue, &lim);3201}3202 3203static int raid1_run(struct mddev *mddev)3204{3205	struct r1conf *conf;3206	int i;3207	int ret;3208 3209	if (mddev->level != 1) {3210		pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n",3211			mdname(mddev), mddev->level);3212		return -EIO;3213	}3214	if (mddev->reshape_position != MaxSector) {3215		pr_warn("md/raid1:%s: reshape_position set but not supported\n",3216			mdname(mddev));3217		return -EIO;3218	}3219 3220	/*3221	 * copy the already verified devices into our private RAID13222	 * bookkeeping area. [whatever we allocate in run(),3223	 * should be freed in raid1_free()]3224	 */3225	if (mddev->private == NULL)3226		conf = setup_conf(mddev);3227	else3228		conf = mddev->private;3229 3230	if (IS_ERR(conf))3231		return PTR_ERR(conf);3232 3233	if (!mddev_is_dm(mddev)) {3234		ret = raid1_set_limits(mddev);3235		if (ret)3236			return ret;3237	}3238 3239	mddev->degraded = 0;3240	for (i = 0; i < conf->raid_disks; i++)3241		if (conf->mirrors[i].rdev == NULL ||3242		    !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||3243		    test_bit(Faulty, &conf->mirrors[i].rdev->flags))3244			mddev->degraded++;3245	/*3246	 * RAID1 needs at least one disk in active3247	 */3248	if (conf->raid_disks - mddev->degraded < 1) {3249		md_unregister_thread(mddev, &conf->thread);3250		return -EINVAL;3251	}3252 3253	if (conf->raid_disks - mddev->degraded == 1)3254		mddev->recovery_cp = MaxSector;3255 3256	if (mddev->recovery_cp != MaxSector)3257		pr_info("md/raid1:%s: not clean -- starting background reconstruction\n",3258			mdname(mddev));3259	pr_info("md/raid1:%s: active with %d out of %d mirrors\n",3260		mdname(mddev), mddev->raid_disks - mddev->degraded,3261		mddev->raid_disks);3262 3263	/*3264	 * Ok, everything is just fine now3265	 */3266	rcu_assign_pointer(mddev->thread, conf->thread);3267	rcu_assign_pointer(conf->thread, NULL);3268	mddev->private = conf;3269	set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);3270 3271	md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));3272 3273	ret = md_integrity_register(mddev);3274	if (ret)3275		md_unregister_thread(mddev, &mddev->thread);3276	return ret;3277}3278 3279static void raid1_free(struct mddev *mddev, void *priv)3280{3281	struct r1conf *conf = priv;3282 3283	mempool_exit(&conf->r1bio_pool);3284	kfree(conf->mirrors);3285	safe_put_page(conf->tmppage);3286	kfree(conf->poolinfo);3287	kfree(conf->nr_pending);3288	kfree(conf->nr_waiting);3289	kfree(conf->nr_queued);3290	kfree(conf->barrier);3291	bioset_exit(&conf->bio_split);3292	kfree(conf);3293}3294 3295static int raid1_resize(struct mddev *mddev, sector_t sectors)3296{3297	/* no resync is happening, and there is enough space3298	 * on all devices, so we can resize.3299	 * We need to make sure resync covers any new space.3300	 * If the array is shrinking we should possibly wait until3301	 * any io in the removed space completes, but it hardly seems3302	 * worth it.3303	 */3304	sector_t newsize = raid1_size(mddev, sectors, 0);3305	int ret;3306 3307	if (mddev->external_size &&3308	    mddev->array_sectors > newsize)3309		return -EINVAL;3310 3311	ret = mddev->bitmap_ops->resize(mddev, newsize, 0, false);3312	if (ret)3313		return ret;3314 3315	md_set_array_sectors(mddev, newsize);3316	if (sectors > mddev->dev_sectors &&3317	    mddev->recovery_cp > mddev->dev_sectors) {3318		mddev->recovery_cp = mddev->dev_sectors;3319		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);3320	}3321	mddev->dev_sectors = sectors;3322	mddev->resync_max_sectors = sectors;3323	return 0;3324}3325 3326static int raid1_reshape(struct mddev *mddev)3327{3328	/* We need to:3329	 * 1/ resize the r1bio_pool3330	 * 2/ resize conf->mirrors3331	 *3332	 * We allocate a new r1bio_pool if we can.3333	 * Then raise a device barrier and wait until all IO stops.3334	 * Then resize conf->mirrors and swap in the new r1bio pool.3335	 *3336	 * At the same time, we "pack" the devices so that all the missing3337	 * devices have the higher raid_disk numbers.3338	 */3339	mempool_t newpool, oldpool;3340	struct pool_info *newpoolinfo;3341	struct raid1_info *newmirrors;3342	struct r1conf *conf = mddev->private;3343	int cnt, raid_disks;3344	unsigned long flags;3345	int d, d2;3346	int ret;3347 3348	memset(&newpool, 0, sizeof(newpool));3349	memset(&oldpool, 0, sizeof(oldpool));3350 3351	/* Cannot change chunk_size, layout, or level */3352	if (mddev->chunk_sectors != mddev->new_chunk_sectors ||3353	    mddev->layout != mddev->new_layout ||3354	    mddev->level != mddev->new_level) {3355		mddev->new_chunk_sectors = mddev->chunk_sectors;3356		mddev->new_layout = mddev->layout;3357		mddev->new_level = mddev->level;3358		return -EINVAL;3359	}3360 3361	if (!mddev_is_clustered(mddev))3362		md_allow_write(mddev);3363 3364	raid_disks = mddev->raid_disks + mddev->delta_disks;3365 3366	if (raid_disks < conf->raid_disks) {3367		cnt=0;3368		for (d= 0; d < conf->raid_disks; d++)3369			if (conf->mirrors[d].rdev)3370				cnt++;3371		if (cnt > raid_disks)3372			return -EBUSY;3373	}3374 3375	newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);3376	if (!newpoolinfo)3377		return -ENOMEM;3378	newpoolinfo->mddev = mddev;3379	newpoolinfo->raid_disks = raid_disks * 2;3380 3381	ret = mempool_init(&newpool, NR_RAID_BIOS, r1bio_pool_alloc,3382			   rbio_pool_free, newpoolinfo);3383	if (ret) {3384		kfree(newpoolinfo);3385		return ret;3386	}3387	newmirrors = kzalloc(array3_size(sizeof(struct raid1_info),3388					 raid_disks, 2),3389			     GFP_KERNEL);3390	if (!newmirrors) {3391		kfree(newpoolinfo);3392		mempool_exit(&newpool);3393		return -ENOMEM;3394	}3395 3396	freeze_array(conf, 0);3397 3398	/* ok, everything is stopped */3399	oldpool = conf->r1bio_pool;3400	conf->r1bio_pool = newpool;3401 3402	for (d = d2 = 0; d < conf->raid_disks; d++) {3403		struct md_rdev *rdev = conf->mirrors[d].rdev;3404		if (rdev && rdev->raid_disk != d2) {3405			sysfs_unlink_rdev(mddev, rdev);3406			rdev->raid_disk = d2;3407			sysfs_unlink_rdev(mddev, rdev);3408			if (sysfs_link_rdev(mddev, rdev))3409				pr_warn("md/raid1:%s: cannot register rd%d\n",3410					mdname(mddev), rdev->raid_disk);3411		}3412		if (rdev)3413			newmirrors[d2++].rdev = rdev;3414	}3415	kfree(conf->mirrors);3416	conf->mirrors = newmirrors;3417	kfree(conf->poolinfo);3418	conf->poolinfo = newpoolinfo;3419 3420	spin_lock_irqsave(&conf->device_lock, flags);3421	mddev->degraded += (raid_disks - conf->raid_disks);3422	spin_unlock_irqrestore(&conf->device_lock, flags);3423	conf->raid_disks = mddev->raid_disks = raid_disks;3424	mddev->delta_disks = 0;3425 3426	unfreeze_array(conf);3427 3428	set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);3429	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);3430	md_wakeup_thread(mddev->thread);3431 3432	mempool_exit(&oldpool);3433	return 0;3434}3435 3436static void raid1_quiesce(struct mddev *mddev, int quiesce)3437{3438	struct r1conf *conf = mddev->private;3439 3440	if (quiesce)3441		freeze_array(conf, 0);3442	else3443		unfreeze_array(conf);3444}3445 3446static void *raid1_takeover(struct mddev *mddev)3447{3448	/* raid1 can take over:3449	 *  raid5 with 2 devices, any layout or chunk size3450	 */3451	if (mddev->level == 5 && mddev->raid_disks == 2) {3452		struct r1conf *conf;3453		mddev->new_level = 1;3454		mddev->new_layout = 0;3455		mddev->new_chunk_sectors = 0;3456		conf = setup_conf(mddev);3457		if (!IS_ERR(conf)) {3458			/* Array must appear to be quiesced */3459			conf->array_frozen = 1;3460			mddev_clear_unsupported_flags(mddev,3461				UNSUPPORTED_MDDEV_FLAGS);3462		}3463		return conf;3464	}3465	return ERR_PTR(-EINVAL);3466}3467 3468static struct md_personality raid1_personality =3469{3470	.name		= "raid1",3471	.level		= 1,3472	.owner		= THIS_MODULE,3473	.make_request	= raid1_make_request,3474	.run		= raid1_run,3475	.free		= raid1_free,3476	.status		= raid1_status,3477	.error_handler	= raid1_error,3478	.hot_add_disk	= raid1_add_disk,3479	.hot_remove_disk= raid1_remove_disk,3480	.spare_active	= raid1_spare_active,3481	.sync_request	= raid1_sync_request,3482	.resize		= raid1_resize,3483	.size		= raid1_size,3484	.check_reshape	= raid1_reshape,3485	.quiesce	= raid1_quiesce,3486	.takeover	= raid1_takeover,3487};3488 3489static int __init raid_init(void)3490{3491	return register_md_personality(&raid1_personality);3492}3493 3494static void raid_exit(void)3495{3496	unregister_md_personality(&raid1_personality);3497}3498 3499module_init(raid_init);3500module_exit(raid_exit);3501MODULE_LICENSE("GPL");3502MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");3503MODULE_ALIAS("md-personality-3"); /* RAID1 */3504MODULE_ALIAS("md-raid1");3505MODULE_ALIAS("md-level-1");3506