Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * raid1.c : Multiple Devices driver for Linux
   4 *
   5 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
   6 *
   7 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
   8 *
   9 * RAID-1 management functions.
  10 *
  11 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
  12 *
  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 support
  17 * bitmapped intelligence in resync:
  18 *
  19 *      - bitmap marked during normal i/o
  20 *      - bitmap used to skip nondirty blocks during sync
  21 *
  22 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
  23 * - persistent bitmap code
 
 
 
 
 
 
 
 
 
  24 */
  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-bio
 121 * '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 32
 138#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 writing
 164	 */
 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 to
 174	 * the first bio.
 175	 * If this is a user-requested check/repair, allocate
 176	 * RESYNC_PAGES for each bio.
 177	 */
 178	if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
 179		need_pages = pi->raid_disks;
 180	else
 181		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 mirrored
 295 * operation and are ready to return a success/failure code to the buffer
 296 * 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 device
 327	 * 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 bio
 345 */
 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 definitely
 379		 * want to retry */
 380		;
 381	else {
 382		/* If all other devices have failed, we want to return
 383		 * 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	if (test_bit(R1BIO_BehindIO, &r1_bio->state))
 424		mddev->bitmap_ops->end_behind_write(mddev);
 425	md_write_end(mddev);
 
 
 
 
 
 426}
 427
 428static void r1_bio_write_done(struct r1bio *r1_bio)
 429{
 430	if (!atomic_dec_and_test(&r1_bio->remaining))
 431		return;
 432
 433	if (test_bit(R1BIO_WriteError, &r1_bio->state))
 434		reschedule_retry(r1_bio);
 435	else {
 436		close_write(r1_bio);
 437		if (test_bit(R1BIO_MadeGood, &r1_bio->state))
 438			reschedule_retry(r1_bio);
 439		else
 440			raid_end_bio_io(r1_bio);
 441	}
 442}
 443
 444static void raid1_end_write_request(struct bio *bio)
 445{
 446	struct r1bio *r1_bio = bio->bi_private;
 447	int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
 448	struct r1conf *conf = r1_bio->mddev->private;
 449	struct bio *to_put = NULL;
 450	int mirror = find_bio_disk(r1_bio, bio);
 451	struct md_rdev *rdev = conf->mirrors[mirror].rdev;
 452	bool discard_error;
 453	sector_t lo = r1_bio->sector;
 454	sector_t hi = r1_bio->sector + r1_bio->sectors;
 455
 456	discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD;
 457
 458	/*
 459	 * 'one mirror IO has finished' event handler:
 460	 */
 461	if (bio->bi_status && !discard_error) {
 462		set_bit(WriteErrorSeen,	&rdev->flags);
 463		if (!test_and_set_bit(WantReplacement, &rdev->flags))
 464			set_bit(MD_RECOVERY_NEEDED, &
 465				conf->mddev->recovery);
 466
 467		if (test_bit(FailFast, &rdev->flags) &&
 468		    (bio->bi_opf & MD_FAILFAST) &&
 469		    /* We never try FailFast to WriteMostly devices */
 470		    !test_bit(WriteMostly, &rdev->flags)) {
 471			md_error(r1_bio->mddev, rdev);
 472		}
 473
 474		/*
 475		 * When the device is faulty, it is not necessary to
 476		 * handle write error.
 477		 */
 478		if (!test_bit(Faulty, &rdev->flags))
 
 
 
 
 
 479			set_bit(R1BIO_WriteError, &r1_bio->state);
 480		else {
 481			/* Finished with this branch */
 482			r1_bio->bios[mirror] = NULL;
 483			to_put = bio;
 484		}
 485	} else {
 486		/*
 487		 * Set R1BIO_Uptodate in our master bio, so that we
 488		 * will return a good error code for to the higher
 489		 * levels even if IO on some other mirrored buffer
 490		 * fails.
 491		 *
 492		 * The 'master' represents the composite IO operation
 493		 * to user-side. So if something waits for IO, then it
 494		 * will wait for the 'master' bio.
 495		 */
 
 
 
 496		r1_bio->bios[mirror] = NULL;
 497		to_put = bio;
 498		/*
 499		 * Do not set R1BIO_Uptodate if the current device is
 500		 * rebuilding or Faulty. This is because we cannot use
 501		 * such device for properly reading the data back (we could
 502		 * potentially use it, if the current write would have felt
 503		 * before rdev->recovery_offset, but for simplicity we don't
 504		 * check this here.
 505		 */
 506		if (test_bit(In_sync, &rdev->flags) &&
 507		    !test_bit(Faulty, &rdev->flags))
 508			set_bit(R1BIO_Uptodate, &r1_bio->state);
 509
 510		/* Maybe we can clear some bad blocks. */
 511		if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) &&
 512		    !discard_error) {
 513			r1_bio->bios[mirror] = IO_MADE_GOOD;
 514			set_bit(R1BIO_MadeGood, &r1_bio->state);
 515		}
 516	}
 517
 518	if (behind) {
 519		if (test_bit(CollisionCheck, &rdev->flags))
 520			remove_serial(rdev, lo, hi);
 521		if (test_bit(WriteMostly, &rdev->flags))
 522			atomic_dec(&r1_bio->behind_remaining);
 523
 524		/*
 525		 * In behind mode, we ACK the master bio once the I/O
 526		 * has safely reached all non-writemostly
 527		 * disks. Setting the Returned bit ensures that this
 528		 * gets done only once -- we don't ever want to return
 529		 * -EIO here, instead we'll wait
 530		 */
 531		if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
 532		    test_bit(R1BIO_Uptodate, &r1_bio->state)) {
 533			/* Maybe we can return now */
 534			if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
 535				struct bio *mbio = r1_bio->master_bio;
 536				pr_debug("raid1: behind end write sectors"
 537					 " %llu-%llu\n",
 538					 (unsigned long long) mbio->bi_iter.bi_sector,
 539					 (unsigned long long) bio_end_sector(mbio) - 1);
 540				call_bio_endio(r1_bio);
 541			}
 542		}
 543	} else if (rdev->mddev->serialize_policy)
 544		remove_serial(rdev, lo, hi);
 545	if (r1_bio->bios[mirror] == NULL)
 546		rdev_dec_pending(rdev, conf->mddev);
 547
 548	/*
 549	 * Let's see if all mirrored write operations have finished
 550	 * already.
 551	 */
 552	r1_bio_write_done(r1_bio);
 553
 554	if (to_put)
 555		bio_put(to_put);
 556}
 557
 558static sector_t align_to_barrier_unit_end(sector_t start_sector,
 559					  sector_t sectors)
 560{
 561	sector_t len;
 562
 563	WARN_ON(sectors == 0);
 564	/*
 565	 * len is the number of sectors from start_sector to end of the
 566	 * barrier unit which start_sector belongs to.
 567	 */
 568	len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) -
 569	      start_sector;
 570
 571	if (len > sectors)
 572		len = sectors;
 573
 574	return len;
 575}
 576
 577static void update_read_sectors(struct r1conf *conf, int disk,
 578				sector_t this_sector, int len)
 579{
 580	struct raid1_info *info = &conf->mirrors[disk];
 581
 582	atomic_inc(&info->rdev->nr_pending);
 583	if (info->next_seq_sect != this_sector)
 584		info->seq_start = this_sector;
 585	info->next_seq_sect = this_sector + len;
 586}
 587
 588static int choose_first_rdev(struct r1conf *conf, struct r1bio *r1_bio,
 589			     int *max_sectors)
 590{
 591	sector_t this_sector = r1_bio->sector;
 592	int len = r1_bio->sectors;
 
 
 
 593	int disk;
 
 
 
 
 
 594
 595	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
 596		struct md_rdev *rdev;
 597		int read_len;
 598
 599		if (r1_bio->bios[disk] == IO_BLOCKED)
 600			continue;
 601
 602		rdev = conf->mirrors[disk].rdev;
 603		if (!rdev || test_bit(Faulty, &rdev->flags))
 604			continue;
 605
 606		/* choose the first disk even if it has some bad blocks. */
 607		read_len = raid1_check_read_range(rdev, this_sector, &len);
 608		if (read_len > 0) {
 609			update_read_sectors(conf, disk, this_sector, read_len);
 610			*max_sectors = read_len;
 611			return disk;
 612		}
 613	}
 614
 615	return -1;
 616}
 617
 618static bool rdev_in_recovery(struct md_rdev *rdev, struct r1bio *r1_bio)
 619{
 620	return !test_bit(In_sync, &rdev->flags) &&
 621	       rdev->recovery_offset < r1_bio->sector + r1_bio->sectors;
 622}
 623
 624static int choose_bb_rdev(struct r1conf *conf, struct r1bio *r1_bio,
 625			  int *max_sectors)
 626{
 627	sector_t this_sector = r1_bio->sector;
 628	int best_disk = -1;
 629	int best_len = 0;
 630	int disk;
 631
 632	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
 633		struct md_rdev *rdev;
 634		int len;
 635		int read_len;
 
 
 636
 637		if (r1_bio->bios[disk] == IO_BLOCKED)
 
 
 
 638			continue;
 639
 640		rdev = conf->mirrors[disk].rdev;
 641		if (!rdev || test_bit(Faulty, &rdev->flags) ||
 642		    rdev_in_recovery(rdev, r1_bio) ||
 643		    test_bit(WriteMostly, &rdev->flags))
 644			continue;
 645
 646		/* keep track of the disk with the most readable sectors. */
 647		len = r1_bio->sectors;
 648		read_len = raid1_check_read_range(rdev, this_sector, &len);
 649		if (read_len > best_len) {
 650			best_disk = disk;
 651			best_len = read_len;
 652		}
 653	}
 654
 655	if (best_disk != -1) {
 656		*max_sectors = best_len;
 657		update_read_sectors(conf, best_disk, this_sector, best_len);
 658	}
 659
 660	return best_disk;
 661}
 662
 663static int choose_slow_rdev(struct r1conf *conf, struct r1bio *r1_bio,
 664			    int *max_sectors)
 665{
 666	sector_t this_sector = r1_bio->sector;
 667	int bb_disk = -1;
 668	int bb_read_len = 0;
 669	int disk;
 670
 671	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
 672		struct md_rdev *rdev;
 673		int len;
 674		int read_len;
 675
 676		if (r1_bio->bios[disk] == IO_BLOCKED)
 677			continue;
 678
 679		rdev = conf->mirrors[disk].rdev;
 680		if (!rdev || test_bit(Faulty, &rdev->flags) ||
 681		    !test_bit(WriteMostly, &rdev->flags) ||
 682		    rdev_in_recovery(rdev, r1_bio))
 683			continue;
 684
 685		/* there are no bad blocks, we can use this disk */
 686		len = r1_bio->sectors;
 687		read_len = raid1_check_read_range(rdev, this_sector, &len);
 688		if (read_len == r1_bio->sectors) {
 689			*max_sectors = read_len;
 690			update_read_sectors(conf, disk, this_sector, read_len);
 691			return disk;
 692		}
 693
 694		/*
 695		 * there are partial bad blocks, choose the rdev with largest
 696		 * read length.
 697		 */
 698		if (read_len > bb_read_len) {
 699			bb_disk = disk;
 700			bb_read_len = read_len;
 701		}
 702	}
 703
 704	if (bb_disk != -1) {
 705		*max_sectors = bb_read_len;
 706		update_read_sectors(conf, bb_disk, this_sector, bb_read_len);
 707	}
 708
 709	return bb_disk;
 710}
 711
 712static bool is_sequential(struct r1conf *conf, int disk, struct r1bio *r1_bio)
 713{
 714	/* TODO: address issues with this check and concurrency. */
 715	return conf->mirrors[disk].next_seq_sect == r1_bio->sector ||
 716	       conf->mirrors[disk].head_position == r1_bio->sector;
 717}
 718
 719/*
 720 * If buffered sequential IO size exceeds optimal iosize, check if there is idle
 721 * disk. If yes, choose the idle disk.
 722 */
 723static bool should_choose_next(struct r1conf *conf, int disk)
 724{
 725	struct raid1_info *mirror = &conf->mirrors[disk];
 726	int opt_iosize;
 727
 728	if (!test_bit(Nonrot, &mirror->rdev->flags))
 729		return false;
 730
 731	opt_iosize = bdev_io_opt(mirror->rdev->bdev) >> 9;
 732	return opt_iosize > 0 && mirror->seq_start != MaxSector &&
 733	       mirror->next_seq_sect > opt_iosize &&
 734	       mirror->next_seq_sect - opt_iosize >= mirror->seq_start;
 735}
 736
 737static bool rdev_readable(struct md_rdev *rdev, struct r1bio *r1_bio)
 738{
 739	if (!rdev || test_bit(Faulty, &rdev->flags))
 740		return false;
 741
 742	if (rdev_in_recovery(rdev, r1_bio))
 743		return false;
 744
 745	/* don't read from slow disk unless have to */
 746	if (test_bit(WriteMostly, &rdev->flags))
 747		return false;
 748
 749	/* don't split IO for bad blocks unless have to */
 750	if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors))
 751		return false;
 752
 753	return true;
 754}
 755
 756struct read_balance_ctl {
 757	sector_t closest_dist;
 758	int closest_dist_disk;
 759	int min_pending;
 760	int min_pending_disk;
 761	int sequential_disk;
 762	int readable_disks;
 763};
 764
 765static int choose_best_rdev(struct r1conf *conf, struct r1bio *r1_bio)
 766{
 767	int disk;
 768	struct read_balance_ctl ctl = {
 769		.closest_dist_disk      = -1,
 770		.closest_dist           = MaxSector,
 771		.min_pending_disk       = -1,
 772		.min_pending            = UINT_MAX,
 773		.sequential_disk	= -1,
 774	};
 775
 776	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
 777		struct md_rdev *rdev;
 778		sector_t dist;
 779		unsigned int pending;
 780
 781		if (r1_bio->bios[disk] == IO_BLOCKED)
 782			continue;
 783
 784		rdev = conf->mirrors[disk].rdev;
 785		if (!rdev_readable(rdev, r1_bio))
 786			continue;
 
 
 787
 788		/* At least two disks to choose from so failfast is OK */
 789		if (ctl.readable_disks++ == 1)
 790			set_bit(R1BIO_FailFast, &r1_bio->state);
 791
 
 
 792		pending = atomic_read(&rdev->nr_pending);
 793		dist = abs(r1_bio->sector - conf->mirrors[disk].head_position);
 794
 
 
 
 795		/* Don't change to another disk for sequential reads */
 796		if (is_sequential(conf, disk, r1_bio)) {
 797			if (!should_choose_next(conf, disk))
 798				return disk;
 
 799
 
 800			/*
 801			 * Add 'pending' to avoid choosing this disk if
 802			 * there is other idle disk.
 803			 */
 804			pending++;
 805			/*
 806			 * If there is no other idle disk, this disk
 807			 * will be chosen.
 
 
 
 
 808			 */
 809			ctl.sequential_disk = disk;
 
 
 
 
 
 
 
 
 810		}
 811
 812		if (ctl.min_pending > pending) {
 813			ctl.min_pending = pending;
 814			ctl.min_pending_disk = disk;
 
 
 
 815		}
 816
 817		if (ctl.closest_dist > dist) {
 818			ctl.closest_dist = dist;
 819			ctl.closest_dist_disk = disk;
 820		}
 821	}
 822
 823	/*
 824	 * sequential IO size exceeds optimal iosize, however, there is no other
 825	 * idle disk, so choose the sequential disk.
 826	 */
 827	if (ctl.sequential_disk != -1 && ctl.min_pending != 0)
 828		return ctl.sequential_disk;
 829
 830	/*
 831	 * If all disks are rotational, choose the closest disk. If any disk is
 832	 * non-rotational, choose the disk with less pending request even the
 833	 * disk is rotational, which might/might not be optimal for raids with
 834	 * mixed ratation/non-rotational disks depending on workload.
 835	 */
 836	if (ctl.min_pending_disk != -1 &&
 837	    (READ_ONCE(conf->nonrot_disks) || ctl.min_pending == 0))
 838		return ctl.min_pending_disk;
 839	else
 840		return ctl.closest_dist_disk;
 841}
 842
 843/*
 844 * This routine returns the disk from which the requested read should be done.
 845 *
 846 * 1) If resync is in progress, find the first usable disk and use it even if it
 847 * has some bad blocks.
 848 *
 849 * 2) Now that there is no resync, loop through all disks and skipping slow
 850 * disks and disks with bad blocks for now. Only pay attention to key disk
 851 * choice.
 852 *
 853 * 3) If we've made it this far, now look for disks with bad blocks and choose
 854 * the one with most number of sectors.
 855 *
 856 * 4) If we are all the way at the end, we have no choice but to use a disk even
 857 * if it is write mostly.
 858 *
 859 * The rdev for the device selected will have nr_pending incremented.
 860 */
 861static int read_balance(struct r1conf *conf, struct r1bio *r1_bio,
 862			int *max_sectors)
 863{
 864	int disk;
 865
 866	clear_bit(R1BIO_FailFast, &r1_bio->state);
 
 867
 868	if (raid1_should_read_first(conf->mddev, r1_bio->sector,
 869				    r1_bio->sectors))
 870		return choose_first_rdev(conf, r1_bio, max_sectors);
 871
 872	disk = choose_best_rdev(conf, r1_bio);
 873	if (disk >= 0) {
 874		*max_sectors = r1_bio->sectors;
 875		update_read_sectors(conf, disk, r1_bio->sector,
 876				    r1_bio->sectors);
 877		return disk;
 878	}
 
 
 879
 880	/*
 881	 * If we are here it means we didn't find a perfectly good disk so
 882	 * now spend a bit more time trying to find one with the most good
 883	 * sectors.
 884	 */
 885	disk = choose_bb_rdev(conf, r1_bio, max_sectors);
 886	if (disk >= 0)
 887		return disk;
 888
 889	return choose_slow_rdev(conf, r1_bio, max_sectors);
 890}
 891
 892static void wake_up_barrier(struct r1conf *conf)
 893{
 894	if (wq_has_sleeper(&conf->wait_barrier))
 895		wake_up(&conf->wait_barrier);
 896}
 897
 898static void flush_bio_list(struct r1conf *conf, struct bio *bio)
 899{
 900	/* flush any pending bitmap writes to disk before proceeding w/ I/O */
 901	raid1_prepare_flush_writes(conf->mddev);
 902	wake_up_barrier(conf);
 903
 904	while (bio) { /* submit pending writes */
 905		struct bio *next = bio->bi_next;
 
 
 
 906
 907		raid1_submit_write(bio);
 908		bio = next;
 909		cond_resched();
 
 
 
 
 
 
 
 910	}
 
 
 911}
 912
 913static void flush_pending_writes(struct r1conf *conf)
 914{
 915	/* Any writes that have been queued but are awaiting
 916	 * bitmap updates get flushed here.
 917	 */
 918	spin_lock_irq(&conf->device_lock);
 919
 920	if (conf->pending_bio_list.head) {
 921		struct blk_plug plug;
 922		struct bio *bio;
 923
 924		bio = bio_list_get(&conf->pending_bio_list);
 
 925		spin_unlock_irq(&conf->device_lock);
 
 
 
 
 926
 927		/*
 928		 * As this is called in a wait_event() loop (see freeze_array),
 929		 * current->state might be TASK_UNINTERRUPTIBLE which will
 930		 * cause a warning when we prepare to wait again.  As it is
 931		 * rare that this path is taken, it is perfectly safe to force
 932		 * us to go around the wait_event() loop again, so the warning
 933		 * is a false-positive.  Silence the warning by resetting
 934		 * thread state
 935		 */
 936		__set_current_state(TASK_RUNNING);
 937		blk_start_plug(&plug);
 938		flush_bio_list(conf, bio);
 939		blk_finish_plug(&plug);
 
 
 
 940	} else
 941		spin_unlock_irq(&conf->device_lock);
 942}
 943
 944/* Barriers....
 945 * Sometimes we need to suspend IO while we do something else,
 946 * either some resync/recovery, or reconfigure the array.
 947 * To do this we raise a 'barrier'.
 948 * The 'barrier' is a counter that can be raised multiple times
 949 * to count how many activities are happening which preclude
 950 * normal IO.
 951 * We can only raise the barrier if there is no pending IO.
 952 * i.e. if nr_pending == 0.
 953 * We choose only to raise the barrier if no-one is waiting for the
 954 * barrier to go down.  This means that as soon as an IO request
 955 * is ready, no other operations which require a barrier will start
 956 * until the IO request has had a chance.
 957 *
 958 * So: regular IO calls 'wait_barrier'.  When that returns there
 959 *    is no backgroup IO happening,  It must arrange to call
 960 *    allow_barrier when it has finished its IO.
 961 * backgroup IO calls must call raise_barrier.  Once that returns
 962 *    there is no normal IO happeing.  It must arrange to call
 963 *    lower_barrier when the particular background IO completes.
 964 *
 965 * If resync/recovery is interrupted, returns -EINTR;
 966 * Otherwise, returns 0.
 967 */
 968static int raise_barrier(struct r1conf *conf, sector_t sector_nr)
 969{
 970	int idx = sector_to_idx(sector_nr);
 971
 972	spin_lock_irq(&conf->resync_lock);
 973
 974	/* Wait until no block IO is waiting */
 975	wait_event_lock_irq(conf->wait_barrier,
 976			    !atomic_read(&conf->nr_waiting[idx]),
 977			    conf->resync_lock);
 978
 979	/* block any new IO from starting */
 980	atomic_inc(&conf->barrier[idx]);
 981	/*
 982	 * In raise_barrier() we firstly increase conf->barrier[idx] then
 983	 * check conf->nr_pending[idx]. In _wait_barrier() we firstly
 984	 * increase conf->nr_pending[idx] then check conf->barrier[idx].
 985	 * A memory barrier here to make sure conf->nr_pending[idx] won't
 986	 * be fetched before conf->barrier[idx] is increased. Otherwise
 987	 * there will be a race between raise_barrier() and _wait_barrier().
 988	 */
 989	smp_mb__after_atomic();
 990
 991	/* For these conditions we must wait:
 992	 * A: while the array is in frozen state
 993	 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O
 994	 *    existing in corresponding I/O barrier bucket.
 995	 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches
 996	 *    max resync count which allowed on current I/O barrier bucket.
 
 
 997	 */
 998	wait_event_lock_irq(conf->wait_barrier,
 999			    (!conf->array_frozen &&
1000			     !atomic_read(&conf->nr_pending[idx]) &&
1001			     atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) ||
1002				test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery),
 
1003			    conf->resync_lock);
1004
1005	if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
1006		atomic_dec(&conf->barrier[idx]);
1007		spin_unlock_irq(&conf->resync_lock);
1008		wake_up(&conf->wait_barrier);
1009		return -EINTR;
1010	}
1011
1012	atomic_inc(&conf->nr_sync_pending);
1013	spin_unlock_irq(&conf->resync_lock);
1014
1015	return 0;
1016}
1017
1018static void lower_barrier(struct r1conf *conf, sector_t sector_nr)
1019{
1020	int idx = sector_to_idx(sector_nr);
1021
1022	BUG_ON(atomic_read(&conf->barrier[idx]) <= 0);
1023
1024	atomic_dec(&conf->barrier[idx]);
1025	atomic_dec(&conf->nr_sync_pending);
1026	wake_up(&conf->wait_barrier);
1027}
1028
1029static bool _wait_barrier(struct r1conf *conf, int idx, bool nowait)
1030{
1031	bool ret = true;
1032
1033	/*
1034	 * We need to increase conf->nr_pending[idx] very early here,
1035	 * then raise_barrier() can be blocked when it waits for
1036	 * conf->nr_pending[idx] to be 0. Then we can avoid holding
1037	 * conf->resync_lock when there is no barrier raised in same
1038	 * barrier unit bucket. Also if the array is frozen, I/O
1039	 * should be blocked until array is unfrozen.
1040	 */
1041	atomic_inc(&conf->nr_pending[idx]);
1042	/*
1043	 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then
1044	 * check conf->barrier[idx]. In raise_barrier() we firstly increase
1045	 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory
1046	 * barrier is necessary here to make sure conf->barrier[idx] won't be
1047	 * fetched before conf->nr_pending[idx] is increased. Otherwise there
1048	 * will be a race between _wait_barrier() and raise_barrier().
1049	 */
1050	smp_mb__after_atomic();
1051
1052	/*
1053	 * Don't worry about checking two atomic_t variables at same time
1054	 * here. If during we check conf->barrier[idx], the array is
1055	 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is
1056	 * 0, it is safe to return and make the I/O continue. Because the
1057	 * array is frozen, all I/O returned here will eventually complete
1058	 * or be queued, no race will happen. See code comment in
1059	 * frozen_array().
1060	 */
1061	if (!READ_ONCE(conf->array_frozen) &&
1062	    !atomic_read(&conf->barrier[idx]))
1063		return ret;
1064
1065	/*
1066	 * After holding conf->resync_lock, conf->nr_pending[idx]
1067	 * should be decreased before waiting for barrier to drop.
1068	 * Otherwise, we may encounter a race condition because
1069	 * raise_barrer() might be waiting for conf->nr_pending[idx]
1070	 * to be 0 at same time.
1071	 */
1072	spin_lock_irq(&conf->resync_lock);
1073	atomic_inc(&conf->nr_waiting[idx]);
1074	atomic_dec(&conf->nr_pending[idx]);
1075	/*
1076	 * In case freeze_array() is waiting for
1077	 * get_unqueued_pending() == extra
1078	 */
1079	wake_up_barrier(conf);
1080	/* Wait for the barrier in same barrier unit bucket to drop. */
1081
1082	/* Return false when nowait flag is set */
1083	if (nowait) {
1084		ret = false;
1085	} else {
1086		wait_event_lock_irq(conf->wait_barrier,
1087				!conf->array_frozen &&
1088				!atomic_read(&conf->barrier[idx]),
1089				conf->resync_lock);
1090		atomic_inc(&conf->nr_pending[idx]);
1091	}
1092
1093	atomic_dec(&conf->nr_waiting[idx]);
1094	spin_unlock_irq(&conf->resync_lock);
1095	return ret;
1096}
1097
1098static bool wait_read_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait)
1099{
1100	int idx = sector_to_idx(sector_nr);
1101	bool ret = true;
1102
1103	/*
1104	 * Very similar to _wait_barrier(). The difference is, for read
1105	 * I/O we don't need wait for sync I/O, but if the whole array
1106	 * is frozen, the read I/O still has to wait until the array is
1107	 * unfrozen. Since there is no ordering requirement with
1108	 * conf->barrier[idx] here, memory barrier is unnecessary as well.
1109	 */
1110	atomic_inc(&conf->nr_pending[idx]);
1111
1112	if (!READ_ONCE(conf->array_frozen))
1113		return ret;
1114
1115	spin_lock_irq(&conf->resync_lock);
1116	atomic_inc(&conf->nr_waiting[idx]);
1117	atomic_dec(&conf->nr_pending[idx]);
1118	/*
1119	 * In case freeze_array() is waiting for
1120	 * get_unqueued_pending() == extra
1121	 */
1122	wake_up_barrier(conf);
1123	/* Wait for array to be unfrozen */
1124
1125	/* Return false when nowait flag is set */
1126	if (nowait) {
1127		/* Return false when nowait flag is set */
1128		ret = false;
1129	} else {
1130		wait_event_lock_irq(conf->wait_barrier,
1131				!conf->array_frozen,
1132				conf->resync_lock);
1133		atomic_inc(&conf->nr_pending[idx]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1134	}
1135
1136	atomic_dec(&conf->nr_waiting[idx]);
1137	spin_unlock_irq(&conf->resync_lock);
1138	return ret;
1139}
1140
1141static bool wait_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait)
 
1142{
1143	int idx = sector_to_idx(sector_nr);
1144
1145	return _wait_barrier(conf, idx, nowait);
1146}
1147
1148static void _allow_barrier(struct r1conf *conf, int idx)
1149{
1150	atomic_dec(&conf->nr_pending[idx]);
1151	wake_up_barrier(conf);
1152}
1153
1154static void allow_barrier(struct r1conf *conf, sector_t sector_nr)
1155{
1156	int idx = sector_to_idx(sector_nr);
1157
1158	_allow_barrier(conf, idx);
1159}
1160
1161/* conf->resync_lock should be held */
1162static int get_unqueued_pending(struct r1conf *conf)
1163{
1164	int idx, ret;
1165
1166	ret = atomic_read(&conf->nr_sync_pending);
1167	for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++)
1168		ret += atomic_read(&conf->nr_pending[idx]) -
1169			atomic_read(&conf->nr_queued[idx]);
 
 
 
 
 
 
 
1170
1171	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
1172}
1173
1174static void freeze_array(struct r1conf *conf, int extra)
1175{
1176	/* Stop sync I/O and normal I/O and wait for everything to
1177	 * go quiet.
1178	 * This is called in two situations:
1179	 * 1) management command handlers (reshape, remove disk, quiesce).
1180	 * 2) one normal I/O request failed.
1181
1182	 * After array_frozen is set to 1, new sync IO will be blocked at
1183	 * raise_barrier(), and new normal I/O will blocked at _wait_barrier()
1184	 * or wait_read_barrier(). The flying I/Os will either complete or be
1185	 * queued. When everything goes quite, there are only queued I/Os left.
1186
1187	 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the
1188	 * barrier bucket index which this I/O request hits. When all sync and
1189	 * normal I/O are queued, sum of all conf->nr_pending[] will match sum
1190	 * of all conf->nr_queued[]. But normal I/O failure is an exception,
1191	 * in handle_read_error(), we may call freeze_array() before trying to
1192	 * fix the read error. In this case, the error read I/O is not queued,
1193	 * so get_unqueued_pending() == 1.
1194	 *
1195	 * Therefore before this function returns, we need to wait until
1196	 * get_unqueued_pendings(conf) gets equal to extra. For
1197	 * normal I/O context, extra is 1, in rested situations extra is 0.
1198	 */
1199	spin_lock_irq(&conf->resync_lock);
1200	conf->array_frozen = 1;
1201	mddev_add_trace_msg(conf->mddev, "raid1 wait freeze");
1202	wait_event_lock_irq_cmd(
1203		conf->wait_barrier,
1204		get_unqueued_pending(conf) == extra,
1205		conf->resync_lock,
1206		flush_pending_writes(conf));
1207	spin_unlock_irq(&conf->resync_lock);
1208}
1209static void unfreeze_array(struct r1conf *conf)
1210{
1211	/* reverse the effect of the freeze */
1212	spin_lock_irq(&conf->resync_lock);
1213	conf->array_frozen = 0;
1214	spin_unlock_irq(&conf->resync_lock);
1215	wake_up(&conf->wait_barrier);
 
1216}
1217
1218static void alloc_behind_master_bio(struct r1bio *r1_bio,
1219					   struct bio *bio)
 
1220{
1221	int size = bio->bi_iter.bi_size;
1222	unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1223	int i = 0;
1224	struct bio *behind_bio = NULL;
1225
1226	behind_bio = bio_alloc_bioset(NULL, vcnt, 0, GFP_NOIO,
1227				      &r1_bio->mddev->bio_set);
1228
1229	/* discard op, we don't support writezero/writesame yet */
1230	if (!bio_has_data(bio)) {
1231		behind_bio->bi_iter.bi_size = size;
1232		goto skip_copy;
1233	}
1234
1235	while (i < vcnt && size) {
1236		struct page *page;
1237		int len = min_t(int, PAGE_SIZE, size);
1238
1239		page = alloc_page(GFP_NOIO);
1240		if (unlikely(!page))
1241			goto free_pages;
1242
1243		if (!bio_add_page(behind_bio, page, len, 0)) {
1244			put_page(page);
1245			goto free_pages;
1246		}
1247
1248		size -= len;
1249		i++;
 
 
 
 
 
 
 
1250	}
1251
1252	bio_copy_data(behind_bio, bio);
1253skip_copy:
1254	r1_bio->behind_master_bio = behind_bio;
1255	set_bit(R1BIO_BehindIO, &r1_bio->state);
1256
1257	return;
1258
1259free_pages:
 
 
 
 
1260	pr_debug("%dB behind alloc failed, doing sync I/O\n",
1261		 bio->bi_iter.bi_size);
1262	bio_free_pages(behind_bio);
1263	bio_put(behind_bio);
1264}
1265
 
 
 
 
 
 
1266static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
1267{
1268	struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1269						  cb);
1270	struct mddev *mddev = plug->cb.data;
1271	struct r1conf *conf = mddev->private;
1272	struct bio *bio;
1273
1274	if (from_schedule) {
1275		spin_lock_irq(&conf->device_lock);
1276		bio_list_merge(&conf->pending_bio_list, &plug->pending);
 
1277		spin_unlock_irq(&conf->device_lock);
1278		wake_up_barrier(conf);
1279		md_wakeup_thread(mddev->thread);
1280		kfree(plug);
1281		return;
1282	}
1283
1284	/* we aren't scheduling, so we can do the write-out directly. */
1285	bio = bio_list_get(&plug->pending);
1286	flush_bio_list(conf, bio);
1287	kfree(plug);
1288}
1289
1290static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio)
1291{
1292	r1_bio->master_bio = bio;
1293	r1_bio->sectors = bio_sectors(bio);
1294	r1_bio->state = 0;
1295	r1_bio->mddev = mddev;
1296	r1_bio->sector = bio->bi_iter.bi_sector;
1297}
1298
1299static inline struct r1bio *
1300alloc_r1bio(struct mddev *mddev, struct bio *bio)
1301{
1302	struct r1conf *conf = mddev->private;
1303	struct r1bio *r1_bio;
1304
1305	r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO);
1306	/* Ensure no bio records IO_BLOCKED */
1307	memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0]));
1308	init_r1bio(r1_bio, mddev, bio);
1309	return r1_bio;
 
 
 
 
 
 
 
 
 
 
 
 
1310}
1311
1312static void raid1_read_request(struct mddev *mddev, struct bio *bio,
1313			       int max_read_sectors, struct r1bio *r1_bio)
1314{
1315	struct r1conf *conf = mddev->private;
1316	struct raid1_info *mirror;
1317	struct bio *read_bio;
1318	const enum req_op op = bio_op(bio);
1319	const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC;
 
 
1320	int max_sectors;
1321	int rdisk, error;
1322	bool r1bio_existed = !!r1_bio;
1323
1324	/*
1325	 * If r1_bio is set, we are blocking the raid1d thread
1326	 * so there is a tiny risk of deadlock.  So ask for
1327	 * emergency memory if needed.
1328	 */
1329	gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO;
1330
1331	/*
1332	 * Still need barrier for READ in case that whole
1333	 * array is frozen.
1334	 */
1335	if (!wait_read_barrier(conf, bio->bi_iter.bi_sector,
1336				bio->bi_opf & REQ_NOWAIT)) {
1337		bio_wouldblock_error(bio);
1338		return;
1339	}
1340
1341	if (!r1_bio)
1342		r1_bio = alloc_r1bio(mddev, bio);
1343	else
1344		init_r1bio(r1_bio, mddev, bio);
1345	r1_bio->sectors = max_read_sectors;
1346
1347	/*
1348	 * make_request() can abort the operation when read-ahead is being
1349	 * used and no empty request is available.
1350	 */
1351	rdisk = read_balance(conf, r1_bio, &max_sectors);
 
1352	if (rdisk < 0) {
1353		/* couldn't find anywhere to read from */
1354		if (r1bio_existed)
1355			pr_crit_ratelimited("md/raid1:%s: %pg: unrecoverable I/O read error for block %llu\n",
1356					    mdname(mddev),
1357					    conf->mirrors[r1_bio->read_disk].rdev->bdev,
1358					    r1_bio->sector);
1359		raid_end_bio_io(r1_bio);
1360		return;
1361	}
1362	mirror = conf->mirrors + rdisk;
1363
1364	if (r1bio_existed)
1365		pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n",
1366				    mdname(mddev),
1367				    (unsigned long long)r1_bio->sector,
1368				    mirror->rdev->bdev);
1369
1370	if (test_bit(WriteMostly, &mirror->rdev->flags)) {
1371		/*
1372		 * Reading from a write-mostly device must take care not to
1373		 * over-take any writes that are 'behind'
1374		 */
1375		mddev_add_trace_msg(mddev, "raid1 wait behind writes");
1376		mddev->bitmap_ops->wait_behind_writes(mddev);
1377	}
1378
1379	if (max_sectors < bio_sectors(bio)) {
1380		struct bio *split = bio_split(bio, max_sectors,
1381					      gfp, &conf->bio_split);
1382
1383		if (IS_ERR(split)) {
1384			error = PTR_ERR(split);
1385			goto err_handle;
1386		}
1387		bio_chain(split, bio);
1388		submit_bio_noacct(bio);
1389		bio = split;
1390		r1_bio->master_bio = bio;
1391		r1_bio->sectors = max_sectors;
1392	}
1393
1394	r1_bio->read_disk = rdisk;
1395	if (!r1bio_existed) {
1396		md_account_bio(mddev, &bio);
1397		r1_bio->master_bio = bio;
1398	}
1399	read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp,
1400				   &mddev->bio_set);
1401
1402	r1_bio->bios[rdisk] = read_bio;
1403
1404	read_bio->bi_iter.bi_sector = r1_bio->sector +
1405		mirror->rdev->data_offset;
 
1406	read_bio->bi_end_io = raid1_end_read_request;
1407	read_bio->bi_opf = op | do_sync;
1408	if (test_bit(FailFast, &mirror->rdev->flags) &&
1409	    test_bit(R1BIO_FailFast, &r1_bio->state))
1410	        read_bio->bi_opf |= MD_FAILFAST;
1411	read_bio->bi_private = r1_bio;
1412	mddev_trace_remap(mddev, read_bio, r1_bio->sector);
1413	submit_bio_noacct(read_bio);
1414	return;
1415
1416err_handle:
1417	atomic_dec(&mirror->rdev->nr_pending);
1418	bio->bi_status = errno_to_blk_status(error);
1419	set_bit(R1BIO_Uptodate, &r1_bio->state);
1420	raid_end_bio_io(r1_bio);
1421}
1422
1423static bool wait_blocked_rdev(struct mddev *mddev, struct bio *bio)
1424{
1425	struct r1conf *conf = mddev->private;
1426	int disks = conf->raid_disks * 2;
1427	int i;
1428
1429retry:
1430	for (i = 0; i < disks; i++) {
1431		struct md_rdev *rdev = conf->mirrors[i].rdev;
1432
1433		if (!rdev)
1434			continue;
 
 
1435
1436		/* don't write here until the bad block is acknowledged */
1437		if (test_bit(WriteErrorSeen, &rdev->flags) &&
1438		    rdev_has_badblock(rdev, bio->bi_iter.bi_sector,
1439				      bio_sectors(bio)) < 0)
1440			set_bit(BlockedBadBlocks, &rdev->flags);
1441
1442		if (rdev_blocked(rdev)) {
1443			if (bio->bi_opf & REQ_NOWAIT)
1444				return false;
1445
1446			mddev_add_trace_msg(rdev->mddev, "raid1 wait rdev %d blocked",
1447					    rdev->raid_disk);
1448			atomic_inc(&rdev->nr_pending);
1449			md_wait_for_blocked_rdev(rdev, rdev->mddev);
1450			goto retry;
1451		}
1452	}
1453
1454	return true;
 
 
 
 
 
 
 
1455}
1456
1457static void raid1_write_request(struct mddev *mddev, struct bio *bio,
1458				int max_write_sectors)
1459{
1460	struct r1conf *conf = mddev->private;
1461	struct r1bio *r1_bio;
1462	int i, disks, k, error;
1463	unsigned long flags;
 
 
 
 
 
 
 
1464	int first_clone;
 
1465	int max_sectors;
1466	bool write_behind = false;
1467	bool is_discard = (bio_op(bio) == REQ_OP_DISCARD);
1468
1469	if (mddev_is_clustered(mddev) &&
 
 
 
 
 
 
 
 
 
 
1470	     md_cluster_ops->area_resyncing(mddev, WRITE,
1471		     bio->bi_iter.bi_sector, bio_end_sector(bio))) {
1472
 
 
 
 
1473		DEFINE_WAIT(w);
1474		if (bio->bi_opf & REQ_NOWAIT) {
1475			bio_wouldblock_error(bio);
1476			return;
1477		}
1478		for (;;) {
 
1479			prepare_to_wait(&conf->wait_barrier,
1480					&w, TASK_IDLE);
1481			if (!md_cluster_ops->area_resyncing(mddev, WRITE,
1482							bio->bi_iter.bi_sector,
1483							bio_end_sector(bio)))
 
 
 
1484				break;
1485			schedule();
1486		}
1487		finish_wait(&conf->wait_barrier, &w);
1488	}
 
1489
1490	/*
1491	 * Register the new request and wait if the reconstruction
1492	 * thread has put up a bar for new requests.
1493	 * Continue immediately if no resync is active currently.
1494	 */
1495	if (!wait_barrier(conf, bio->bi_iter.bi_sector,
1496				bio->bi_opf & REQ_NOWAIT)) {
1497		bio_wouldblock_error(bio);
1498		return;
1499	}
1500
1501	if (!wait_blocked_rdev(mddev, bio)) {
1502		bio_wouldblock_error(bio);
1503		return;
1504	}
1505
1506	r1_bio = alloc_r1bio(mddev, bio);
1507	r1_bio->sectors = max_write_sectors;
1508
1509	/* first select target devices under rcu_lock and
1510	 * inc refcount on their rdev.  Record them by setting
1511	 * bios[x] to bio
1512	 * If there are known/acknowledged bad blocks on any device on
1513	 * which we have seen a write error, we want to avoid writing those
1514	 * blocks.
1515	 * This potentially requires several writes to write around
1516	 * the bad blocks.  Each set of writes gets it's own r1bio
1517	 * with a set of bios attached.
1518	 */
1519
1520	disks = conf->raid_disks * 2;
 
 
 
 
1521	max_sectors = r1_bio->sectors;
1522	for (i = 0;  i < disks; i++) {
1523		struct md_rdev *rdev = conf->mirrors[i].rdev;
1524
1525		/*
1526		 * The write-behind io is only attempted on drives marked as
1527		 * write-mostly, which means we could allocate write behind
1528		 * bio later.
1529		 */
1530		if (!is_discard && rdev && test_bit(WriteMostly, &rdev->flags))
1531			write_behind = true;
1532
1533		r1_bio->bios[i] = NULL;
1534		if (!rdev || test_bit(Faulty, &rdev->flags))
 
 
1535			continue;
 
1536
1537		atomic_inc(&rdev->nr_pending);
1538		if (test_bit(WriteErrorSeen, &rdev->flags)) {
1539			sector_t first_bad;
1540			int bad_sectors;
1541			int is_bad;
1542
1543			is_bad = is_badblock(rdev, r1_bio->sector, max_sectors,
1544					     &first_bad, &bad_sectors);
 
 
 
 
 
 
 
1545			if (is_bad && first_bad <= r1_bio->sector) {
1546				/* Cannot write here at all */
1547				bad_sectors -= (r1_bio->sector - first_bad);
1548				if (bad_sectors < max_sectors)
1549					/* mustn't write more than bad_sectors
1550					 * to other devices yet
1551					 */
1552					max_sectors = bad_sectors;
1553				rdev_dec_pending(rdev, mddev);
 
 
 
 
 
 
 
 
 
 
1554				continue;
1555			}
1556			if (is_bad) {
1557				int good_sectors;
1558
1559				/*
1560				 * We cannot atomically write this, so just
1561				 * error in that case. It could be possible to
1562				 * atomically write other mirrors, but the
1563				 * complexity of supporting that is not worth
1564				 * the benefit.
1565				 */
1566				if (bio->bi_opf & REQ_ATOMIC) {
1567					error = -EIO;
1568					goto err_handle;
1569				}
1570
1571				good_sectors = first_bad - r1_bio->sector;
1572				if (good_sectors < max_sectors)
1573					max_sectors = good_sectors;
1574			}
1575		}
1576		r1_bio->bios[i] = bio;
1577	}
 
1578
1579	/*
1580	 * When using a bitmap, we may call alloc_behind_master_bio below.
1581	 * alloc_behind_master_bio allocates a copy of the data payload a page
1582	 * at a time and thus needs a new bio that can fit the whole payload
1583	 * this bio in page sized chunks.
1584	 */
1585	if (write_behind && mddev->bitmap)
1586		max_sectors = min_t(int, max_sectors,
1587				    BIO_MAX_VECS * (PAGE_SIZE >> 9));
1588	if (max_sectors < bio_sectors(bio)) {
1589		struct bio *split = bio_split(bio, max_sectors,
1590					      GFP_NOIO, &conf->bio_split);
1591
1592		if (IS_ERR(split)) {
1593			error = PTR_ERR(split);
1594			goto err_handle;
1595		}
1596		bio_chain(split, bio);
1597		submit_bio_noacct(bio);
1598		bio = split;
1599		r1_bio->master_bio = bio;
 
 
 
 
 
 
 
 
1600		r1_bio->sectors = max_sectors;
 
 
 
 
 
 
1601	}
 
1602
1603	md_account_bio(mddev, &bio);
1604	r1_bio->master_bio = bio;
1605	atomic_set(&r1_bio->remaining, 1);
1606	atomic_set(&r1_bio->behind_remaining, 0);
1607
1608	first_clone = 1;
1609
1610	for (i = 0; i < disks; i++) {
1611		struct bio *mbio = NULL;
1612		struct md_rdev *rdev = conf->mirrors[i].rdev;
1613		if (!r1_bio->bios[i])
1614			continue;
1615
1616		if (first_clone) {
1617			unsigned long max_write_behind =
1618				mddev->bitmap_info.max_write_behind;
1619			struct md_bitmap_stats stats;
1620			int err;
1621
 
1622			/* do behind I/O ?
1623			 * Not if there are too many, or cannot
1624			 * allocate memory, or a reader on WriteMostly
1625			 * is waiting for behind writes to flush */
1626			err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
1627			if (!err && write_behind && !stats.behind_wait &&
1628			    stats.behind_writes < max_write_behind)
1629				alloc_behind_master_bio(r1_bio, bio);
1630
1631			if (test_bit(R1BIO_BehindIO, &r1_bio->state))
1632				mddev->bitmap_ops->start_behind_write(mddev);
 
 
 
1633			first_clone = 0;
1634		}
 
 
 
1635
1636		if (r1_bio->behind_master_bio) {
1637			mbio = bio_alloc_clone(rdev->bdev,
1638					       r1_bio->behind_master_bio,
1639					       GFP_NOIO, &mddev->bio_set);
1640			if (test_bit(CollisionCheck, &rdev->flags))
1641				wait_for_serialization(rdev, r1_bio);
1642			if (test_bit(WriteMostly, &rdev->flags))
1643				atomic_inc(&r1_bio->behind_remaining);
1644		} else {
1645			mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO,
1646					       &mddev->bio_set);
1647
1648			if (mddev->serialize_policy)
1649				wait_for_serialization(rdev, r1_bio);
1650		}
1651
1652		r1_bio->bios[i] = mbio;
1653
1654		mbio->bi_iter.bi_sector	= (r1_bio->sector + rdev->data_offset);
 
 
1655		mbio->bi_end_io	= raid1_end_write_request;
1656		mbio->bi_opf = bio_op(bio) |
1657			(bio->bi_opf & (REQ_SYNC | REQ_FUA | REQ_ATOMIC));
1658		if (test_bit(FailFast, &rdev->flags) &&
1659		    !test_bit(WriteMostly, &rdev->flags) &&
1660		    conf->raid_disks - mddev->degraded > 1)
1661			mbio->bi_opf |= MD_FAILFAST;
1662		mbio->bi_private = r1_bio;
1663
1664		atomic_inc(&r1_bio->remaining);
1665		mddev_trace_remap(mddev, mbio, r1_bio->sector);
 
 
 
 
1666		/* flush_pending_writes() needs access to the rdev so...*/
1667		mbio->bi_bdev = (void *)rdev;
1668		if (!raid1_add_bio_to_plug(mddev, mbio, raid1_unplug, disks)) {
1669			spin_lock_irqsave(&conf->device_lock, flags);
 
 
 
 
 
 
 
 
 
1670			bio_list_add(&conf->pending_bio_list, mbio);
1671			spin_unlock_irqrestore(&conf->device_lock, flags);
1672			md_wakeup_thread(mddev->thread);
1673		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1674	}
1675
1676	r1_bio_write_done(r1_bio);
1677
1678	/* In case raid1d snuck in to freeze_array */
1679	wake_up_barrier(conf);
1680	return;
1681err_handle:
1682	for (k = 0; k < i; k++) {
1683		if (r1_bio->bios[k]) {
1684			rdev_dec_pending(conf->mirrors[k].rdev, mddev);
1685			r1_bio->bios[k] = NULL;
1686		}
1687	}
1688
1689	bio->bi_status = errno_to_blk_status(error);
1690	set_bit(R1BIO_Uptodate, &r1_bio->state);
1691	raid_end_bio_io(r1_bio);
1692}
1693
1694static bool raid1_make_request(struct mddev *mddev, struct bio *bio)
1695{
1696	sector_t sectors;
 
1697
1698	if (unlikely(bio->bi_opf & REQ_PREFLUSH)
1699	    && md_flush_request(mddev, bio))
1700		return true;
 
 
 
 
 
 
 
 
 
1701
1702	/*
1703	 * There is a limit to the maximum size, but
1704	 * the read/write handler might find a lower limit
1705	 * due to bad blocks.  To avoid multiple splits,
1706	 * we pass the maximum number of sectors down
1707	 * and let the lower level perform the split.
1708	 */
1709	sectors = align_to_barrier_unit_end(
1710		bio->bi_iter.bi_sector, bio_sectors(bio));
1711
1712	if (bio_data_dir(bio) == READ)
1713		raid1_read_request(mddev, bio, sectors, NULL);
1714	else {
1715		md_write_start(mddev,bio);
1716		raid1_write_request(mddev, bio, sectors);
1717	}
1718	return true;
1719}
1720
1721static void raid1_status(struct seq_file *seq, struct mddev *mddev)
1722{
1723	struct r1conf *conf = mddev->private;
1724	int i;
1725
1726	lockdep_assert_held(&mddev->lock);
1727
1728	seq_printf(seq, " [%d/%d] [", conf->raid_disks,
1729		   conf->raid_disks - mddev->degraded);
 
1730	for (i = 0; i < conf->raid_disks; i++) {
1731		struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev);
1732
1733		seq_printf(seq, "%s",
1734			   rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1735	}
 
1736	seq_printf(seq, "]");
1737}
1738
1739/**
1740 * raid1_error() - RAID1 error handler.
1741 * @mddev: affected md device.
1742 * @rdev: member device to fail.
1743 *
1744 * The routine acknowledges &rdev failure and determines new @mddev state.
1745 * If it failed, then:
1746 *	- &MD_BROKEN flag is set in &mddev->flags.
1747 *	- recovery is disabled.
1748 * Otherwise, it must be degraded:
1749 *	- recovery is interrupted.
1750 *	- &mddev->degraded is bumped.
1751 *
1752 * @rdev is marked as &Faulty excluding case when array is failed and
1753 * &mddev->fail_last_dev is off.
1754 */
1755static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
1756{
 
1757	struct r1conf *conf = mddev->private;
1758	unsigned long flags;
1759
 
 
 
 
 
 
1760	spin_lock_irqsave(&conf->device_lock, flags);
1761
1762	if (test_bit(In_sync, &rdev->flags) &&
1763	    (conf->raid_disks - mddev->degraded) == 1) {
1764		set_bit(MD_BROKEN, &mddev->flags);
1765
1766		if (!mddev->fail_last_dev) {
1767			conf->recovery_disabled = mddev->recovery_disabled;
1768			spin_unlock_irqrestore(&conf->device_lock, flags);
1769			return;
1770		}
 
1771	}
1772	set_bit(Blocked, &rdev->flags);
1773	if (test_and_clear_bit(In_sync, &rdev->flags))
1774		mddev->degraded++;
1775	set_bit(Faulty, &rdev->flags);
 
 
1776	spin_unlock_irqrestore(&conf->device_lock, flags);
1777	/*
1778	 * if recovery is running, make sure it aborts.
1779	 */
1780	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1781	set_mask_bits(&mddev->sb_flags, 0,
1782		      BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
1783	pr_crit("md/raid1:%s: Disk failure on %pg, disabling device.\n"
1784		"md/raid1:%s: Operation continuing on %d devices.\n",
1785		mdname(mddev), rdev->bdev,
1786		mdname(mddev), conf->raid_disks - mddev->degraded);
1787}
1788
1789static void print_conf(struct r1conf *conf)
1790{
1791	int i;
1792
1793	pr_debug("RAID1 conf printout:\n");
1794	if (!conf) {
1795		pr_debug("(!conf)\n");
1796		return;
1797	}
1798	pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1799		 conf->raid_disks);
1800
1801	lockdep_assert_held(&conf->mddev->reconfig_mutex);
1802	for (i = 0; i < conf->raid_disks; i++) {
1803		struct md_rdev *rdev = conf->mirrors[i].rdev;
 
1804		if (rdev)
1805			pr_debug(" disk %d, wo:%d, o:%d, dev:%pg\n",
1806				 i, !test_bit(In_sync, &rdev->flags),
1807				 !test_bit(Faulty, &rdev->flags),
1808				 rdev->bdev);
1809	}
 
1810}
1811
1812static void close_sync(struct r1conf *conf)
1813{
1814	int idx;
 
1815
1816	for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) {
1817		_wait_barrier(conf, idx, false);
1818		_allow_barrier(conf, idx);
1819	}
1820
1821	mempool_exit(&conf->r1buf_pool);
 
 
 
 
 
 
1822}
1823
1824static int raid1_spare_active(struct mddev *mddev)
1825{
1826	int i;
1827	struct r1conf *conf = mddev->private;
1828	int count = 0;
1829	unsigned long flags;
1830
1831	/*
1832	 * Find all failed disks within the RAID1 configuration
1833	 * and mark them readable.
1834	 * Called under mddev lock, so rcu protection not needed.
1835	 * device_lock used to avoid races with raid1_end_read_request
1836	 * which expects 'In_sync' flags and ->degraded to be consistent.
1837	 */
1838	spin_lock_irqsave(&conf->device_lock, flags);
1839	for (i = 0; i < conf->raid_disks; i++) {
1840		struct md_rdev *rdev = conf->mirrors[i].rdev;
1841		struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1842		if (repl
1843		    && !test_bit(Candidate, &repl->flags)
1844		    && repl->recovery_offset == MaxSector
1845		    && !test_bit(Faulty, &repl->flags)
1846		    && !test_and_set_bit(In_sync, &repl->flags)) {
1847			/* replacement has just become active */
1848			if (!rdev ||
1849			    !test_and_clear_bit(In_sync, &rdev->flags))
1850				count++;
1851			if (rdev) {
1852				/* Replaced device not technically
1853				 * faulty, but we need to be sure
1854				 * it gets removed and never re-added
1855				 */
1856				set_bit(Faulty, &rdev->flags);
1857				sysfs_notify_dirent_safe(
1858					rdev->sysfs_state);
1859			}
1860		}
1861		if (rdev
1862		    && rdev->recovery_offset == MaxSector
1863		    && !test_bit(Faulty, &rdev->flags)
1864		    && !test_and_set_bit(In_sync, &rdev->flags)) {
1865			count++;
1866			sysfs_notify_dirent_safe(rdev->sysfs_state);
1867		}
1868	}
1869	mddev->degraded -= count;
1870	spin_unlock_irqrestore(&conf->device_lock, flags);
1871
1872	print_conf(conf);
1873	return count;
1874}
1875
1876static bool raid1_add_conf(struct r1conf *conf, struct md_rdev *rdev, int disk,
1877			   bool replacement)
1878{
1879	struct raid1_info *info = conf->mirrors + disk;
1880
1881	if (replacement)
1882		info += conf->raid_disks;
1883
1884	if (info->rdev)
1885		return false;
1886
1887	if (bdev_nonrot(rdev->bdev)) {
1888		set_bit(Nonrot, &rdev->flags);
1889		WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks + 1);
1890	}
1891
1892	rdev->raid_disk = disk;
1893	info->head_position = 0;
1894	info->seq_start = MaxSector;
1895	WRITE_ONCE(info->rdev, rdev);
1896
1897	return true;
1898}
1899
1900static bool raid1_remove_conf(struct r1conf *conf, int disk)
1901{
1902	struct raid1_info *info = conf->mirrors + disk;
1903	struct md_rdev *rdev = info->rdev;
1904
1905	if (!rdev || test_bit(In_sync, &rdev->flags) ||
1906	    atomic_read(&rdev->nr_pending))
1907		return false;
1908
1909	/* Only remove non-faulty devices if recovery is not possible. */
1910	if (!test_bit(Faulty, &rdev->flags) &&
1911	    rdev->mddev->recovery_disabled != conf->recovery_disabled &&
1912	    rdev->mddev->degraded < conf->raid_disks)
1913		return false;
1914
1915	if (test_and_clear_bit(Nonrot, &rdev->flags))
1916		WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks - 1);
1917
1918	WRITE_ONCE(info->rdev, NULL);
1919	return true;
1920}
1921
1922static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1923{
1924	struct r1conf *conf = mddev->private;
1925	int err = -EEXIST;
1926	int mirror = 0, repl_slot = -1;
1927	struct raid1_info *p;
1928	int first = 0;
1929	int last = conf->raid_disks - 1;
1930
1931	if (mddev->recovery_disabled == conf->recovery_disabled)
1932		return -EBUSY;
1933
 
 
 
1934	if (rdev->raid_disk >= 0)
1935		first = last = rdev->raid_disk;
1936
1937	/*
1938	 * find the disk ... but prefer rdev->saved_raid_disk
1939	 * if possible.
1940	 */
1941	if (rdev->saved_raid_disk >= 0 &&
1942	    rdev->saved_raid_disk >= first &&
1943	    rdev->saved_raid_disk < conf->raid_disks &&
1944	    conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1945		first = last = rdev->saved_raid_disk;
1946
1947	for (mirror = first; mirror <= last; mirror++) {
1948		p = conf->mirrors + mirror;
1949		if (!p->rdev) {
1950			err = mddev_stack_new_rdev(mddev, rdev);
1951			if (err)
1952				return err;
1953
1954			raid1_add_conf(conf, rdev, mirror, false);
 
 
 
 
 
 
1955			/* As all devices are equivalent, we don't need a full recovery
1956			 * if this was recently any drive of the array
1957			 */
1958			if (rdev->saved_raid_disk < 0)
1959				conf->fullsync = 1;
 
1960			break;
1961		}
1962		if (test_bit(WantReplacement, &p->rdev->flags) &&
1963		    p[conf->raid_disks].rdev == NULL && repl_slot < 0)
1964			repl_slot = mirror;
1965	}
1966
1967	if (err && repl_slot >= 0) {
1968		/* Add this device as a replacement */
1969		clear_bit(In_sync, &rdev->flags);
1970		set_bit(Replacement, &rdev->flags);
1971		raid1_add_conf(conf, rdev, repl_slot, true);
1972		err = 0;
1973		conf->fullsync = 1;
1974	}
1975
 
1976	print_conf(conf);
1977	return err;
1978}
1979
1980static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1981{
1982	struct r1conf *conf = mddev->private;
1983	int err = 0;
1984	int number = rdev->raid_disk;
1985	struct raid1_info *p = conf->mirrors + number;
1986
1987	if (unlikely(number >= conf->raid_disks))
1988		goto abort;
1989
1990	if (rdev != p->rdev) {
1991		number += conf->raid_disks;
1992		p = conf->mirrors + number;
1993	}
1994
1995	print_conf(conf);
1996	if (rdev == p->rdev) {
1997		if (!raid1_remove_conf(conf, number)) {
 
1998			err = -EBUSY;
1999			goto abort;
2000		}
2001
2002		if (number < conf->raid_disks &&
2003		    conf->mirrors[conf->raid_disks + number].rdev) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2004			/* We just removed a device that is being replaced.
2005			 * Move down the replacement.  We drain all IO before
2006			 * doing this to avoid confusion.
2007			 */
2008			struct md_rdev *repl =
2009				conf->mirrors[conf->raid_disks + number].rdev;
2010			freeze_array(conf, 0);
2011			if (atomic_read(&repl->nr_pending)) {
2012				/* It means that some queued IO of retry_list
2013				 * hold repl. Thus, we cannot set replacement
2014				 * as NULL, avoiding rdev NULL pointer
2015				 * dereference in sync_request_write and
2016				 * handle_write_finished.
2017				 */
2018				err = -EBUSY;
2019				unfreeze_array(conf);
2020				goto abort;
2021			}
2022			clear_bit(Replacement, &repl->flags);
2023			WRITE_ONCE(p->rdev, repl);
2024			conf->mirrors[conf->raid_disks + number].rdev = NULL;
2025			unfreeze_array(conf);
2026		}
2027
2028		clear_bit(WantReplacement, &rdev->flags);
2029		err = md_integrity_register(mddev);
2030	}
2031abort:
2032
2033	print_conf(conf);
2034	return err;
2035}
2036
2037static void end_sync_read(struct bio *bio)
2038{
2039	struct r1bio *r1_bio = get_resync_r1bio(bio);
2040
2041	update_head_pos(r1_bio->read_disk, r1_bio);
2042
2043	/*
2044	 * we have read a block, now it needs to be re-written,
2045	 * or re-read if the read failed.
2046	 * We don't do much here, just schedule handling by raid1d
2047	 */
2048	if (!bio->bi_status)
2049		set_bit(R1BIO_Uptodate, &r1_bio->state);
2050
2051	if (atomic_dec_and_test(&r1_bio->remaining))
2052		reschedule_retry(r1_bio);
2053}
2054
2055static void abort_sync_write(struct mddev *mddev, struct r1bio *r1_bio)
2056{
2057	sector_t sync_blocks = 0;
2058	sector_t s = r1_bio->sector;
2059	long sectors_to_go = r1_bio->sectors;
2060
2061	/* make sure these bits don't get cleared. */
2062	do {
2063		mddev->bitmap_ops->end_sync(mddev, s, &sync_blocks);
2064		s += sync_blocks;
2065		sectors_to_go -= sync_blocks;
2066	} while (sectors_to_go > 0);
2067}
2068
2069static void put_sync_write_buf(struct r1bio *r1_bio, int uptodate)
2070{
2071	if (atomic_dec_and_test(&r1_bio->remaining)) {
2072		struct mddev *mddev = r1_bio->mddev;
2073		int s = r1_bio->sectors;
2074
2075		if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2076		    test_bit(R1BIO_WriteError, &r1_bio->state))
2077			reschedule_retry(r1_bio);
2078		else {
2079			put_buf(r1_bio);
2080			md_done_sync(mddev, s, uptodate);
2081		}
2082	}
2083}
2084
2085static void end_sync_write(struct bio *bio)
2086{
2087	int uptodate = !bio->bi_status;
2088	struct r1bio *r1_bio = get_resync_r1bio(bio);
2089	struct mddev *mddev = r1_bio->mddev;
2090	struct r1conf *conf = mddev->private;
 
 
2091	struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
2092
2093	if (!uptodate) {
2094		abort_sync_write(mddev, r1_bio);
 
 
 
 
 
 
 
 
 
2095		set_bit(WriteErrorSeen, &rdev->flags);
2096		if (!test_and_set_bit(WantReplacement, &rdev->flags))
2097			set_bit(MD_RECOVERY_NEEDED, &
2098				mddev->recovery);
2099		set_bit(R1BIO_WriteError, &r1_bio->state);
2100	} else if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) &&
2101		   !rdev_has_badblock(conf->mirrors[r1_bio->read_disk].rdev,
2102				      r1_bio->sector, r1_bio->sectors)) {
 
 
 
 
2103		set_bit(R1BIO_MadeGood, &r1_bio->state);
2104	}
2105
2106	put_sync_write_buf(r1_bio, uptodate);
 
 
 
 
 
 
 
 
 
2107}
2108
2109static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
2110			   int sectors, struct page *page, blk_opf_t rw)
2111{
2112	if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
2113		/* success */
2114		return 1;
2115	if (rw == REQ_OP_WRITE) {
2116		set_bit(WriteErrorSeen, &rdev->flags);
2117		if (!test_and_set_bit(WantReplacement,
2118				      &rdev->flags))
2119			set_bit(MD_RECOVERY_NEEDED, &
2120				rdev->mddev->recovery);
2121	}
2122	/* need to record an error - either for the block or the device */
2123	if (!rdev_set_badblocks(rdev, sector, sectors, 0))
2124		md_error(rdev->mddev, rdev);
2125	return 0;
2126}
2127
2128static int fix_sync_read_error(struct r1bio *r1_bio)
2129{
2130	/* Try some synchronous reads of other devices to get
2131	 * good data, much like with normal read errors.  Only
2132	 * read into the pages we already have so we don't
2133	 * need to re-issue the read request.
2134	 * We don't need to freeze the array, because being in an
2135	 * active sync request, there is no normal IO, and
2136	 * no overlapping syncs.
2137	 * We don't need to check is_badblock() again as we
2138	 * made sure that anything with a bad block in range
2139	 * will have bi_end_io clear.
2140	 */
2141	struct mddev *mddev = r1_bio->mddev;
2142	struct r1conf *conf = mddev->private;
2143	struct bio *bio = r1_bio->bios[r1_bio->read_disk];
2144	struct page **pages = get_resync_pages(bio)->pages;
2145	sector_t sect = r1_bio->sector;
2146	int sectors = r1_bio->sectors;
2147	int idx = 0;
2148	struct md_rdev *rdev;
2149
2150	rdev = conf->mirrors[r1_bio->read_disk].rdev;
2151	if (test_bit(FailFast, &rdev->flags)) {
2152		/* Don't try recovering from here - just fail it
2153		 * ... unless it is the last working device of course */
2154		md_error(mddev, rdev);
2155		if (test_bit(Faulty, &rdev->flags))
2156			/* Don't try to read from here, but make sure
2157			 * put_buf does it's thing
2158			 */
2159			bio->bi_end_io = end_sync_write;
2160	}
2161
2162	while(sectors) {
2163		int s = sectors;
2164		int d = r1_bio->read_disk;
2165		int success = 0;
2166		int start;
2167
2168		if (s > (PAGE_SIZE>>9))
2169			s = PAGE_SIZE >> 9;
2170		do {
2171			if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
2172				/* No rcu protection needed here devices
2173				 * can only be removed when no resync is
2174				 * active, and resync is currently active
2175				 */
2176				rdev = conf->mirrors[d].rdev;
2177				if (sync_page_io(rdev, sect, s<<9,
2178						 pages[idx],
2179						 REQ_OP_READ, false)) {
2180					success = 1;
2181					break;
2182				}
2183			}
2184			d++;
2185			if (d == conf->raid_disks * 2)
2186				d = 0;
2187		} while (!success && d != r1_bio->read_disk);
2188
2189		if (!success) {
 
2190			int abort = 0;
2191			/* Cannot read from anywhere, this block is lost.
2192			 * Record a bad block on each device.  If that doesn't
2193			 * work just disable and interrupt the recovery.
2194			 * Don't fail devices as that won't really help.
2195			 */
2196			pr_crit_ratelimited("md/raid1:%s: %pg: unrecoverable I/O read error for block %llu\n",
2197					    mdname(mddev), bio->bi_bdev,
 
2198					    (unsigned long long)r1_bio->sector);
2199			for (d = 0; d < conf->raid_disks * 2; d++) {
2200				rdev = conf->mirrors[d].rdev;
2201				if (!rdev || test_bit(Faulty, &rdev->flags))
2202					continue;
2203				if (!rdev_set_badblocks(rdev, sect, s, 0))
2204					abort = 1;
2205			}
2206			if (abort) {
2207				conf->recovery_disabled =
2208					mddev->recovery_disabled;
2209				set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2210				md_done_sync(mddev, r1_bio->sectors, 0);
2211				put_buf(r1_bio);
2212				return 0;
2213			}
2214			/* Try next page */
2215			sectors -= s;
2216			sect += s;
2217			idx++;
2218			continue;
2219		}
2220
2221		start = d;
2222		/* write it back and re-read */
2223		while (d != r1_bio->read_disk) {
2224			if (d == 0)
2225				d = conf->raid_disks * 2;
2226			d--;
2227			if (r1_bio->bios[d]->bi_end_io != end_sync_read)
2228				continue;
2229			rdev = conf->mirrors[d].rdev;
2230			if (r1_sync_page_io(rdev, sect, s,
2231					    pages[idx],
2232					    REQ_OP_WRITE) == 0) {
2233				r1_bio->bios[d]->bi_end_io = NULL;
2234				rdev_dec_pending(rdev, mddev);
2235			}
2236		}
2237		d = start;
2238		while (d != r1_bio->read_disk) {
2239			if (d == 0)
2240				d = conf->raid_disks * 2;
2241			d--;
2242			if (r1_bio->bios[d]->bi_end_io != end_sync_read)
2243				continue;
2244			rdev = conf->mirrors[d].rdev;
2245			if (r1_sync_page_io(rdev, sect, s,
2246					    pages[idx],
2247					    REQ_OP_READ) != 0)
2248				atomic_add(s, &rdev->corrected_errors);
2249		}
2250		sectors -= s;
2251		sect += s;
2252		idx ++;
2253	}
2254	set_bit(R1BIO_Uptodate, &r1_bio->state);
2255	bio->bi_status = 0;
2256	return 1;
2257}
2258
2259static void process_checks(struct r1bio *r1_bio)
2260{
2261	/* We have read all readable devices.  If we haven't
2262	 * got the block, then there is no hope left.
2263	 * If we have, then we want to do a comparison
2264	 * and skip the write if everything is the same.
2265	 * If any blocks failed to read, then we need to
2266	 * attempt an over-write
2267	 */
2268	struct mddev *mddev = r1_bio->mddev;
2269	struct r1conf *conf = mddev->private;
2270	int primary;
2271	int i;
2272	int vcnt;
2273
2274	/* Fix variable parts of all bios */
2275	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
2276	for (i = 0; i < conf->raid_disks * 2; i++) {
2277		blk_status_t status;
 
 
2278		struct bio *b = r1_bio->bios[i];
2279		struct resync_pages *rp = get_resync_pages(b);
2280		if (b->bi_end_io != end_sync_read)
2281			continue;
2282		/* fixup the bio for reuse, but preserve errno */
2283		status = b->bi_status;
2284		bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ);
2285		b->bi_status = status;
 
 
2286		b->bi_iter.bi_sector = r1_bio->sector +
2287			conf->mirrors[i].rdev->data_offset;
 
2288		b->bi_end_io = end_sync_read;
2289		rp->raid_bio = r1_bio;
2290		b->bi_private = rp;
2291
2292		/* initialize bvec table again */
2293		md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9);
 
 
 
 
 
 
 
 
 
2294	}
2295	for (primary = 0; primary < conf->raid_disks * 2; primary++)
2296		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
2297		    !r1_bio->bios[primary]->bi_status) {
2298			r1_bio->bios[primary]->bi_end_io = NULL;
2299			rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
2300			break;
2301		}
2302	r1_bio->read_disk = primary;
2303	for (i = 0; i < conf->raid_disks * 2; i++) {
2304		int j = 0;
2305		struct bio *pbio = r1_bio->bios[primary];
2306		struct bio *sbio = r1_bio->bios[i];
2307		blk_status_t status = sbio->bi_status;
2308		struct page **ppages = get_resync_pages(pbio)->pages;
2309		struct page **spages = get_resync_pages(sbio)->pages;
2310		struct bio_vec *bi;
2311		int page_len[RESYNC_PAGES] = { 0 };
2312		struct bvec_iter_all iter_all;
2313
2314		if (sbio->bi_end_io != end_sync_read)
2315			continue;
2316		/* Now we can 'fixup' the error value */
2317		sbio->bi_status = 0;
2318
2319		bio_for_each_segment_all(bi, sbio, iter_all)
2320			page_len[j++] = bi->bv_len;
2321
2322		if (!status) {
2323			for (j = vcnt; j-- ; ) {
2324				if (memcmp(page_address(ppages[j]),
2325					   page_address(spages[j]),
2326					   page_len[j]))
 
 
 
2327					break;
2328			}
2329		} else
2330			j = 0;
2331		if (j >= 0)
2332			atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
2333		if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
2334			      && !status)) {
2335			/* No need to write to this device. */
2336			sbio->bi_end_io = NULL;
2337			rdev_dec_pending(conf->mirrors[i].rdev, mddev);
2338			continue;
2339		}
2340
2341		bio_copy_data(sbio, pbio);
2342	}
2343}
2344
2345static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
2346{
2347	struct r1conf *conf = mddev->private;
2348	int i;
2349	int disks = conf->raid_disks * 2;
2350	struct bio *wbio;
 
 
2351
2352	if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2353		/* ouch - failed to read all of that. */
2354		if (!fix_sync_read_error(r1_bio))
2355			return;
2356
2357	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2358		process_checks(r1_bio);
2359
2360	/*
2361	 * schedule writes
2362	 */
2363	atomic_set(&r1_bio->remaining, 1);
2364	for (i = 0; i < disks ; i++) {
2365		wbio = r1_bio->bios[i];
2366		if (wbio->bi_end_io == NULL ||
2367		    (wbio->bi_end_io == end_sync_read &&
2368		     (i == r1_bio->read_disk ||
2369		      !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
2370			continue;
2371		if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) {
2372			abort_sync_write(mddev, r1_bio);
2373			continue;
2374		}
2375
2376		wbio->bi_opf = REQ_OP_WRITE;
2377		if (test_bit(FailFast, &conf->mirrors[i].rdev->flags))
2378			wbio->bi_opf |= MD_FAILFAST;
2379
2380		wbio->bi_end_io = end_sync_write;
2381		atomic_inc(&r1_bio->remaining);
2382		md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
2383
2384		submit_bio_noacct(wbio);
2385	}
2386
2387	put_sync_write_buf(r1_bio, 1);
 
 
 
 
 
 
 
 
 
 
2388}
2389
2390/*
2391 * This is a kernel thread which:
2392 *
2393 *	1.	Retries failed read operations on working mirrors.
2394 *	2.	Updates the raid superblock when problems encounter.
2395 *	3.	Performs writes following reads for array synchronising.
2396 */
2397
2398static void fix_read_error(struct r1conf *conf, struct r1bio *r1_bio)
 
2399{
2400	sector_t sect = r1_bio->sector;
2401	int sectors = r1_bio->sectors;
2402	int read_disk = r1_bio->read_disk;
2403	struct mddev *mddev = conf->mddev;
2404	struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2405
2406	if (exceed_read_errors(mddev, rdev)) {
2407		r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2408		return;
2409	}
2410
2411	while(sectors) {
2412		int s = sectors;
2413		int d = read_disk;
2414		int success = 0;
2415		int start;
 
2416
2417		if (s > (PAGE_SIZE>>9))
2418			s = PAGE_SIZE >> 9;
2419
2420		do {
2421			rdev = conf->mirrors[d].rdev;
 
 
 
 
2422			if (rdev &&
2423			    (test_bit(In_sync, &rdev->flags) ||
2424			     (!test_bit(Faulty, &rdev->flags) &&
2425			      rdev->recovery_offset >= sect + s)) &&
2426			    rdev_has_badblock(rdev, sect, s) == 0) {
 
2427				atomic_inc(&rdev->nr_pending);
 
2428				if (sync_page_io(rdev, sect, s<<9,
2429					 conf->tmppage, REQ_OP_READ, false))
2430					success = 1;
2431				rdev_dec_pending(rdev, mddev);
2432				if (success)
2433					break;
2434			}
2435
2436			d++;
2437			if (d == conf->raid_disks * 2)
2438				d = 0;
2439		} while (d != read_disk);
2440
2441		if (!success) {
2442			/* Cannot read from anywhere - mark it bad */
2443			struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2444			if (!rdev_set_badblocks(rdev, sect, s, 0))
2445				md_error(mddev, rdev);
2446			break;
2447		}
2448		/* write it back and re-read */
2449		start = d;
2450		while (d != read_disk) {
2451			if (d==0)
2452				d = conf->raid_disks * 2;
2453			d--;
2454			rdev = conf->mirrors[d].rdev;
 
2455			if (rdev &&
2456			    !test_bit(Faulty, &rdev->flags)) {
2457				atomic_inc(&rdev->nr_pending);
 
2458				r1_sync_page_io(rdev, sect, s,
2459						conf->tmppage, REQ_OP_WRITE);
2460				rdev_dec_pending(rdev, mddev);
2461			}
 
2462		}
2463		d = start;
2464		while (d != read_disk) {
 
2465			if (d==0)
2466				d = conf->raid_disks * 2;
2467			d--;
2468			rdev = conf->mirrors[d].rdev;
 
2469			if (rdev &&
2470			    !test_bit(Faulty, &rdev->flags)) {
2471				atomic_inc(&rdev->nr_pending);
 
2472				if (r1_sync_page_io(rdev, sect, s,
2473						conf->tmppage, REQ_OP_READ)) {
2474					atomic_add(s, &rdev->corrected_errors);
2475					pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %pg)\n",
2476						mdname(mddev), s,
2477						(unsigned long long)(sect +
2478								     rdev->data_offset),
2479						rdev->bdev);
2480				}
2481				rdev_dec_pending(rdev, mddev);
2482			}
 
2483		}
2484		sectors -= s;
2485		sect += s;
2486	}
2487}
2488
2489static int narrow_write_error(struct r1bio *r1_bio, int i)
2490{
2491	struct mddev *mddev = r1_bio->mddev;
2492	struct r1conf *conf = mddev->private;
2493	struct md_rdev *rdev = conf->mirrors[i].rdev;
2494
2495	/* bio has the data to be written to device 'i' where
2496	 * we just recently had a write error.
2497	 * We repeatedly clone the bio and trim down to one block,
2498	 * then try the write.  Where the write fails we record
2499	 * a bad block.
2500	 * It is conceivable that the bio doesn't exactly align with
2501	 * blocks.  We must handle this somehow.
2502	 *
2503	 * We currently own a reference on the rdev.
2504	 */
2505
2506	int block_sectors;
2507	sector_t sector;
2508	int sectors;
2509	int sect_to_write = r1_bio->sectors;
2510	int ok = 1;
2511
2512	if (rdev->badblocks.shift < 0)
2513		return 0;
2514
2515	block_sectors = roundup(1 << rdev->badblocks.shift,
2516				bdev_logical_block_size(rdev->bdev) >> 9);
2517	sector = r1_bio->sector;
2518	sectors = ((sector + block_sectors)
2519		   & ~(sector_t)(block_sectors - 1))
2520		- sector;
2521
2522	while (sect_to_write) {
2523		struct bio *wbio;
2524		if (sectors > sect_to_write)
2525			sectors = sect_to_write;
2526		/* Write at 'sector' for 'sectors'*/
2527
2528		if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2529			wbio = bio_alloc_clone(rdev->bdev,
2530					       r1_bio->behind_master_bio,
2531					       GFP_NOIO, &mddev->bio_set);
 
 
 
 
 
 
 
 
 
2532		} else {
2533			wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio,
2534					       GFP_NOIO, &mddev->bio_set);
2535		}
2536
2537		wbio->bi_opf = REQ_OP_WRITE;
2538		wbio->bi_iter.bi_sector = r1_bio->sector;
2539		wbio->bi_iter.bi_size = r1_bio->sectors << 9;
2540
2541		bio_trim(wbio, sector - r1_bio->sector, sectors);
2542		wbio->bi_iter.bi_sector += rdev->data_offset;
 
2543
2544		if (submit_bio_wait(wbio) < 0)
2545			/* failure! */
2546			ok = rdev_set_badblocks(rdev, sector,
2547						sectors, 0)
2548				&& ok;
2549
2550		bio_put(wbio);
2551		sect_to_write -= sectors;
2552		sector += sectors;
2553		sectors = block_sectors;
2554	}
2555	return ok;
2556}
2557
2558static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2559{
2560	int m;
2561	int s = r1_bio->sectors;
2562	for (m = 0; m < conf->raid_disks * 2 ; m++) {
2563		struct md_rdev *rdev = conf->mirrors[m].rdev;
2564		struct bio *bio = r1_bio->bios[m];
2565		if (bio->bi_end_io == NULL)
2566			continue;
2567		if (!bio->bi_status &&
2568		    test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2569			rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
2570		}
2571		if (bio->bi_status &&
2572		    test_bit(R1BIO_WriteError, &r1_bio->state)) {
2573			if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2574				md_error(conf->mddev, rdev);
2575		}
2576	}
2577	put_buf(r1_bio);
2578	md_done_sync(conf->mddev, s, 1);
2579}
2580
2581static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2582{
2583	int m, idx;
2584	bool fail = false;
2585
2586	for (m = 0; m < conf->raid_disks * 2 ; m++)
2587		if (r1_bio->bios[m] == IO_MADE_GOOD) {
2588			struct md_rdev *rdev = conf->mirrors[m].rdev;
2589			rdev_clear_badblocks(rdev,
2590					     r1_bio->sector,
2591					     r1_bio->sectors, 0);
2592			rdev_dec_pending(rdev, conf->mddev);
2593		} else if (r1_bio->bios[m] != NULL) {
2594			/* This drive got a write error.  We need to
2595			 * narrow down and record precise write
2596			 * errors.
2597			 */
2598			fail = true;
2599			if (!narrow_write_error(r1_bio, m))
2600				md_error(conf->mddev,
2601					 conf->mirrors[m].rdev);
2602				/* an I/O failed, we can't clear the bitmap */
 
 
2603			rdev_dec_pending(conf->mirrors[m].rdev,
2604					 conf->mddev);
2605		}
2606	if (fail) {
2607		spin_lock_irq(&conf->device_lock);
2608		list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2609		idx = sector_to_idx(r1_bio->sector);
2610		atomic_inc(&conf->nr_queued[idx]);
2611		spin_unlock_irq(&conf->device_lock);
2612		/*
2613		 * In case freeze_array() is waiting for condition
2614		 * get_unqueued_pending() == extra to be true.
2615		 */
2616		wake_up(&conf->wait_barrier);
2617		md_wakeup_thread(conf->mddev->thread);
2618	} else {
2619		if (test_bit(R1BIO_WriteError, &r1_bio->state))
2620			close_write(r1_bio);
2621		raid_end_bio_io(r1_bio);
2622	}
2623}
2624
2625static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
2626{
 
 
2627	struct mddev *mddev = conf->mddev;
2628	struct bio *bio;
 
2629	struct md_rdev *rdev;
2630	sector_t sector;
 
2631
2632	clear_bit(R1BIO_ReadError, &r1_bio->state);
2633	/* we got a read error. Maybe the drive is bad.  Maybe just
2634	 * the block and we can fix it.
2635	 * We freeze all other IO, and try reading the block from
2636	 * other devices.  When we find one, we re-write
2637	 * and check it that fixes the read error.
2638	 * This is all done synchronously while the array is
2639	 * frozen
2640	 */
2641
2642	bio = r1_bio->bios[r1_bio->read_disk];
 
 
 
2643	bio_put(bio);
2644	r1_bio->bios[r1_bio->read_disk] = NULL;
2645
2646	rdev = conf->mirrors[r1_bio->read_disk].rdev;
2647	if (mddev->ro == 0
2648	    && !test_bit(FailFast, &rdev->flags)) {
2649		freeze_array(conf, 1);
2650		fix_read_error(conf, r1_bio);
 
2651		unfreeze_array(conf);
2652	} else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) {
2653		md_error(mddev, rdev);
2654	} else {
2655		r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2656	}
2657
2658	rdev_dec_pending(rdev, conf->mddev);
2659	sector = r1_bio->sector;
2660	bio = r1_bio->master_bio;
2661
2662	/* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */
2663	r1_bio->state = 0;
2664	raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio);
2665	allow_barrier(conf, sector);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2666}
2667
2668static void raid1d(struct md_thread *thread)
2669{
2670	struct mddev *mddev = thread->mddev;
2671	struct r1bio *r1_bio;
2672	unsigned long flags;
2673	struct r1conf *conf = mddev->private;
2674	struct list_head *head = &conf->retry_list;
2675	struct blk_plug plug;
2676	int idx;
2677
2678	md_check_recovery(mddev);
2679
2680	if (!list_empty_careful(&conf->bio_end_io_list) &&
2681	    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
2682		LIST_HEAD(tmp);
2683		spin_lock_irqsave(&conf->device_lock, flags);
2684		if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
2685			list_splice_init(&conf->bio_end_io_list, &tmp);
 
 
 
 
2686		spin_unlock_irqrestore(&conf->device_lock, flags);
2687		while (!list_empty(&tmp)) {
2688			r1_bio = list_first_entry(&tmp, struct r1bio,
2689						  retry_list);
2690			list_del(&r1_bio->retry_list);
2691			idx = sector_to_idx(r1_bio->sector);
2692			atomic_dec(&conf->nr_queued[idx]);
2693			if (test_bit(R1BIO_WriteError, &r1_bio->state))
2694				close_write(r1_bio);
2695			raid_end_bio_io(r1_bio);
2696		}
2697	}
2698
2699	blk_start_plug(&plug);
2700	for (;;) {
2701
2702		flush_pending_writes(conf);
2703
2704		spin_lock_irqsave(&conf->device_lock, flags);
2705		if (list_empty(head)) {
2706			spin_unlock_irqrestore(&conf->device_lock, flags);
2707			break;
2708		}
2709		r1_bio = list_entry(head->prev, struct r1bio, retry_list);
2710		list_del(head->prev);
2711		idx = sector_to_idx(r1_bio->sector);
2712		atomic_dec(&conf->nr_queued[idx]);
2713		spin_unlock_irqrestore(&conf->device_lock, flags);
2714
2715		mddev = r1_bio->mddev;
2716		conf = mddev->private;
2717		if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
2718			if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2719			    test_bit(R1BIO_WriteError, &r1_bio->state))
2720				handle_sync_write_finished(conf, r1_bio);
2721			else
2722				sync_request_write(mddev, r1_bio);
2723		} else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2724			   test_bit(R1BIO_WriteError, &r1_bio->state))
2725			handle_write_finished(conf, r1_bio);
2726		else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2727			handle_read_error(conf, r1_bio);
2728		else
2729			WARN_ON_ONCE(1);
 
 
 
2730
2731		cond_resched();
2732		if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING))
2733			md_check_recovery(mddev);
2734	}
2735	blk_finish_plug(&plug);
2736}
2737
2738static int init_resync(struct r1conf *conf)
2739{
2740	int buffs;
2741
2742	buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
2743	BUG_ON(mempool_initialized(&conf->r1buf_pool));
2744
2745	return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc,
2746			    r1buf_pool_free, conf->poolinfo);
2747}
2748
2749static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf)
2750{
2751	struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO);
2752	struct resync_pages *rps;
2753	struct bio *bio;
2754	int i;
2755
2756	for (i = conf->poolinfo->raid_disks; i--; ) {
2757		bio = r1bio->bios[i];
2758		rps = bio->bi_private;
2759		bio_reset(bio, NULL, 0);
2760		bio->bi_private = rps;
2761	}
2762	r1bio->master_bio = NULL;
2763	return r1bio;
2764}
2765
2766/*
2767 * perform a "sync" on one "block"
2768 *
2769 * We need to make sure that no normal I/O request - particularly write
2770 * requests - conflict with active sync requests.
2771 *
2772 * This is achieved by tracking pending requests and a 'barrier' concept
2773 * that can be installed to exclude normal IO requests.
2774 */
2775
2776static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2777				   sector_t max_sector, int *skipped)
2778{
2779	struct r1conf *conf = mddev->private;
2780	struct r1bio *r1_bio;
2781	struct bio *bio;
2782	sector_t nr_sectors;
2783	int disk = -1;
2784	int i;
2785	int wonly = -1;
2786	int write_targets = 0, read_targets = 0;
2787	sector_t sync_blocks;
2788	bool still_degraded = false;
2789	int good_sectors = RESYNC_SECTORS;
2790	int min_bad = 0; /* number of sectors that are bad in all devices */
2791	int idx = sector_to_idx(sector_nr);
2792	int page_idx = 0;
2793
2794	if (!mempool_initialized(&conf->r1buf_pool))
2795		if (init_resync(conf))
2796			return 0;
2797
 
2798	if (sector_nr >= max_sector) {
2799		/* If we aborted, we need to abort the
2800		 * sync on the 'current' bitmap chunk (there will
2801		 * only be one in raid1 resync.
2802		 * We can find the current addess in mddev->curr_resync
2803		 */
2804		if (mddev->curr_resync < max_sector) /* aborted */
2805			mddev->bitmap_ops->end_sync(mddev, mddev->curr_resync,
2806						    &sync_blocks);
2807		else /* completed sync */
2808			conf->fullsync = 0;
2809
2810		mddev->bitmap_ops->close_sync(mddev);
2811		close_sync(conf);
2812
2813		if (mddev_is_clustered(mddev)) {
2814			conf->cluster_sync_low = 0;
2815			conf->cluster_sync_high = 0;
2816		}
2817		return 0;
2818	}
2819
2820	if (mddev->bitmap == NULL &&
2821	    mddev->recovery_cp == MaxSector &&
2822	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
2823	    conf->fullsync == 0) {
2824		*skipped = 1;
2825		return max_sector - sector_nr;
2826	}
2827	/* before building a request, check if we can skip these blocks..
2828	 * This call the bitmap_start_sync doesn't actually record anything
2829	 */
2830	if (!mddev->bitmap_ops->start_sync(mddev, sector_nr, &sync_blocks, true) &&
2831	    !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2832		/* We can skip this block, and probably several more */
2833		*skipped = 1;
2834		return sync_blocks;
2835	}
2836
2837	/*
2838	 * If there is non-resync activity waiting for a turn, then let it
2839	 * though before starting on this new sync request.
2840	 */
2841	if (atomic_read(&conf->nr_waiting[idx]))
2842		schedule_timeout_uninterruptible(1);
2843
2844	/* we are incrementing sector_nr below. To be safe, we check against
2845	 * sector_nr + two times RESYNC_SECTORS
2846	 */
2847
2848	mddev->bitmap_ops->cond_end_sync(mddev, sector_nr,
2849		mddev_is_clustered(mddev) &&
2850		(sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
2851
2852	if (raise_barrier(conf, sector_nr))
2853		return 0;
2854
2855	r1_bio = raid1_alloc_init_r1buf(conf);
2856
 
2857	/*
2858	 * If we get a correctably read error during resync or recovery,
2859	 * we might want to read from a different device.  So we
2860	 * flag all drives that could conceivably be read from for READ,
2861	 * and any others (which will be non-In_sync devices) for WRITE.
2862	 * If a read fails, we try reading from something else for which READ
2863	 * is OK.
2864	 */
2865
2866	r1_bio->mddev = mddev;
2867	r1_bio->sector = sector_nr;
2868	r1_bio->state = 0;
2869	set_bit(R1BIO_IsSync, &r1_bio->state);
2870	/* make sure good_sectors won't go across barrier unit boundary */
2871	good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors);
2872
2873	for (i = 0; i < conf->raid_disks * 2; i++) {
2874		struct md_rdev *rdev;
2875		bio = r1_bio->bios[i];
 
2876
2877		rdev = conf->mirrors[i].rdev;
2878		if (rdev == NULL ||
2879		    test_bit(Faulty, &rdev->flags)) {
2880			if (i < conf->raid_disks)
2881				still_degraded = true;
2882		} else if (!test_bit(In_sync, &rdev->flags)) {
2883			bio->bi_opf = REQ_OP_WRITE;
2884			bio->bi_end_io = end_sync_write;
2885			write_targets ++;
2886		} else {
2887			/* may need to read from here */
2888			sector_t first_bad = MaxSector;
2889			int bad_sectors;
2890
2891			if (is_badblock(rdev, sector_nr, good_sectors,
2892					&first_bad, &bad_sectors)) {
2893				if (first_bad > sector_nr)
2894					good_sectors = first_bad - sector_nr;
2895				else {
2896					bad_sectors -= (sector_nr - first_bad);
2897					if (min_bad == 0 ||
2898					    min_bad > bad_sectors)
2899						min_bad = bad_sectors;
2900				}
2901			}
2902			if (sector_nr < first_bad) {
2903				if (test_bit(WriteMostly, &rdev->flags)) {
2904					if (wonly < 0)
2905						wonly = i;
2906				} else {
2907					if (disk < 0)
2908						disk = i;
2909				}
2910				bio->bi_opf = REQ_OP_READ;
2911				bio->bi_end_io = end_sync_read;
2912				read_targets++;
2913			} else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2914				test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2915				!test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2916				/*
2917				 * The device is suitable for reading (InSync),
2918				 * but has bad block(s) here. Let's try to correct them,
2919				 * if we are doing resync or repair. Otherwise, leave
2920				 * this device alone for this sync request.
2921				 */
2922				bio->bi_opf = REQ_OP_WRITE;
2923				bio->bi_end_io = end_sync_write;
2924				write_targets++;
2925			}
2926		}
2927		if (rdev && bio->bi_end_io) {
2928			atomic_inc(&rdev->nr_pending);
2929			bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
2930			bio_set_dev(bio, rdev->bdev);
 
2931			if (test_bit(FailFast, &rdev->flags))
2932				bio->bi_opf |= MD_FAILFAST;
2933		}
2934	}
 
2935	if (disk < 0)
2936		disk = wonly;
2937	r1_bio->read_disk = disk;
2938
2939	if (read_targets == 0 && min_bad > 0) {
2940		/* These sectors are bad on all InSync devices, so we
2941		 * need to mark them bad on all write targets
2942		 */
2943		int ok = 1;
2944		for (i = 0 ; i < conf->raid_disks * 2 ; i++)
2945			if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
2946				struct md_rdev *rdev = conf->mirrors[i].rdev;
2947				ok = rdev_set_badblocks(rdev, sector_nr,
2948							min_bad, 0
2949					) && ok;
2950			}
2951		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2952		*skipped = 1;
2953		put_buf(r1_bio);
2954
2955		if (!ok) {
2956			/* Cannot record the badblocks, so need to
2957			 * abort the resync.
2958			 * If there are multiple read targets, could just
2959			 * fail the really bad ones ???
2960			 */
2961			conf->recovery_disabled = mddev->recovery_disabled;
2962			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2963			return 0;
2964		} else
2965			return min_bad;
2966
2967	}
2968	if (min_bad > 0 && min_bad < good_sectors) {
2969		/* only resync enough to reach the next bad->good
2970		 * transition */
2971		good_sectors = min_bad;
2972	}
2973
2974	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2975		/* extra read targets are also write targets */
2976		write_targets += read_targets-1;
2977
2978	if (write_targets == 0 || read_targets == 0) {
2979		/* There is nowhere to write, so all non-sync
2980		 * drives must be failed - so we are finished
2981		 */
2982		sector_t rv;
2983		if (min_bad > 0)
2984			max_sector = sector_nr + min_bad;
2985		rv = max_sector - sector_nr;
2986		*skipped = 1;
2987		put_buf(r1_bio);
2988		return rv;
2989	}
2990
2991	if (max_sector > mddev->resync_max)
2992		max_sector = mddev->resync_max; /* Don't do IO beyond here */
2993	if (max_sector > sector_nr + good_sectors)
2994		max_sector = sector_nr + good_sectors;
2995	nr_sectors = 0;
2996	sync_blocks = 0;
2997	do {
2998		struct page *page;
2999		int len = PAGE_SIZE;
3000		if (sector_nr + (len>>9) > max_sector)
3001			len = (max_sector - sector_nr) << 9;
3002		if (len == 0)
3003			break;
3004		if (sync_blocks == 0) {
3005			if (!mddev->bitmap_ops->start_sync(mddev, sector_nr,
3006						&sync_blocks, still_degraded) &&
3007			    !conf->fullsync &&
3008			    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3009				break;
3010			if ((len >> 9) > sync_blocks)
3011				len = sync_blocks<<9;
3012		}
3013
3014		for (i = 0 ; i < conf->raid_disks * 2; i++) {
3015			struct resync_pages *rp;
3016
3017			bio = r1_bio->bios[i];
3018			rp = get_resync_pages(bio);
3019			if (bio->bi_end_io) {
3020				page = resync_fetch_page(rp, page_idx);
3021
3022				/*
3023				 * won't fail because the vec table is big
3024				 * enough to hold all these pages
3025				 */
3026				__bio_add_page(bio, page, len, 0);
 
 
 
 
 
 
 
 
 
3027			}
3028		}
3029		nr_sectors += len>>9;
3030		sector_nr += len>>9;
3031		sync_blocks -= (len>>9);
3032	} while (++page_idx < RESYNC_PAGES);
3033
3034	r1_bio->sectors = nr_sectors;
3035
3036	if (mddev_is_clustered(mddev) &&
3037			conf->cluster_sync_high < sector_nr + nr_sectors) {
3038		conf->cluster_sync_low = mddev->curr_resync_completed;
3039		conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
3040		/* Send resync message */
3041		md_cluster_ops->resync_info_update(mddev,
3042				conf->cluster_sync_low,
3043				conf->cluster_sync_high);
3044	}
3045
3046	/* For a user-requested sync, we read all readable devices and do a
3047	 * compare
3048	 */
3049	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3050		atomic_set(&r1_bio->remaining, read_targets);
3051		for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
3052			bio = r1_bio->bios[i];
3053			if (bio->bi_end_io == end_sync_read) {
3054				read_targets--;
3055				md_sync_acct_bio(bio, nr_sectors);
3056				if (read_targets == 1)
3057					bio->bi_opf &= ~MD_FAILFAST;
3058				submit_bio_noacct(bio);
3059			}
3060		}
3061	} else {
3062		atomic_set(&r1_bio->remaining, 1);
3063		bio = r1_bio->bios[r1_bio->read_disk];
3064		md_sync_acct_bio(bio, nr_sectors);
3065		if (read_targets == 1)
3066			bio->bi_opf &= ~MD_FAILFAST;
3067		submit_bio_noacct(bio);
 
3068	}
3069	return nr_sectors;
3070}
3071
3072static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
3073{
3074	if (sectors)
3075		return sectors;
3076
3077	return mddev->dev_sectors;
3078}
3079
3080static struct r1conf *setup_conf(struct mddev *mddev)
3081{
3082	struct r1conf *conf;
3083	int i;
3084	struct raid1_info *disk;
3085	struct md_rdev *rdev;
3086	int err = -ENOMEM;
3087
3088	conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
3089	if (!conf)
3090		goto abort;
3091
3092	conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR,
3093				   sizeof(atomic_t), GFP_KERNEL);
3094	if (!conf->nr_pending)
3095		goto abort;
3096
3097	conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR,
3098				   sizeof(atomic_t), GFP_KERNEL);
3099	if (!conf->nr_waiting)
3100		goto abort;
3101
3102	conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR,
3103				  sizeof(atomic_t), GFP_KERNEL);
3104	if (!conf->nr_queued)
3105		goto abort;
3106
3107	conf->barrier = kcalloc(BARRIER_BUCKETS_NR,
3108				sizeof(atomic_t), GFP_KERNEL);
3109	if (!conf->barrier)
3110		goto abort;
3111
3112	conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info),
3113					    mddev->raid_disks, 2),
3114				GFP_KERNEL);
3115	if (!conf->mirrors)
3116		goto abort;
3117
3118	conf->tmppage = alloc_page(GFP_KERNEL);
3119	if (!conf->tmppage)
3120		goto abort;
3121
3122	conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
3123	if (!conf->poolinfo)
3124		goto abort;
3125	conf->poolinfo->raid_disks = mddev->raid_disks * 2;
3126	err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc,
3127			   rbio_pool_free, conf->poolinfo);
3128	if (err)
3129		goto abort;
3130
3131	err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0);
3132	if (err)
3133		goto abort;
3134
3135	conf->poolinfo->mddev = mddev;
3136
3137	err = -EINVAL;
3138	spin_lock_init(&conf->device_lock);
3139	conf->raid_disks = mddev->raid_disks;
3140	rdev_for_each(rdev, mddev) {
 
3141		int disk_idx = rdev->raid_disk;
3142
3143		if (disk_idx >= conf->raid_disks || disk_idx < 0)
3144			continue;
 
 
 
 
3145
3146		if (!raid1_add_conf(conf, rdev, disk_idx,
3147				    test_bit(Replacement, &rdev->flags)))
3148			goto abort;
 
 
 
 
 
3149	}
 
3150	conf->mddev = mddev;
3151	INIT_LIST_HEAD(&conf->retry_list);
3152	INIT_LIST_HEAD(&conf->bio_end_io_list);
3153
3154	spin_lock_init(&conf->resync_lock);
3155	init_waitqueue_head(&conf->wait_barrier);
3156
3157	bio_list_init(&conf->pending_bio_list);
 
3158	conf->recovery_disabled = mddev->recovery_disabled - 1;
3159
 
 
 
3160	err = -EIO;
3161	for (i = 0; i < conf->raid_disks * 2; i++) {
3162
3163		disk = conf->mirrors + i;
3164
3165		if (i < conf->raid_disks &&
3166		    disk[conf->raid_disks].rdev) {
3167			/* This slot has a replacement. */
3168			if (!disk->rdev) {
3169				/* No original, just make the replacement
3170				 * a recovering spare
3171				 */
3172				disk->rdev =
3173					disk[conf->raid_disks].rdev;
3174				disk[conf->raid_disks].rdev = NULL;
3175			} else if (!test_bit(In_sync, &disk->rdev->flags))
3176				/* Original is not in_sync - bad */
3177				goto abort;
3178		}
3179
3180		if (!disk->rdev ||
3181		    !test_bit(In_sync, &disk->rdev->flags)) {
3182			disk->head_position = 0;
3183			if (disk->rdev &&
3184			    (disk->rdev->saved_raid_disk < 0))
3185				conf->fullsync = 1;
3186		}
3187	}
3188
3189	err = -ENOMEM;
3190	rcu_assign_pointer(conf->thread,
3191			   md_register_thread(raid1d, mddev, "raid1"));
3192	if (!conf->thread)
3193		goto abort;
3194
3195	return conf;
3196
3197 abort:
3198	if (conf) {
3199		mempool_exit(&conf->r1bio_pool);
3200		kfree(conf->mirrors);
3201		safe_put_page(conf->tmppage);
3202		kfree(conf->poolinfo);
3203		kfree(conf->nr_pending);
3204		kfree(conf->nr_waiting);
3205		kfree(conf->nr_queued);
3206		kfree(conf->barrier);
3207		bioset_exit(&conf->bio_split);
3208		kfree(conf);
3209	}
3210	return ERR_PTR(err);
3211}
3212
3213static int raid1_set_limits(struct mddev *mddev)
3214{
3215	struct queue_limits lim;
3216	int err;
3217
3218	md_init_stacking_limits(&lim);
3219	lim.max_write_zeroes_sectors = 0;
3220	lim.features |= BLK_FEAT_ATOMIC_WRITES_STACKED;
3221	err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
3222	if (err)
3223		return err;
3224	return queue_limits_set(mddev->gendisk->queue, &lim);
3225}
3226
3227static int raid1_run(struct mddev *mddev)
3228{
3229	struct r1conf *conf;
3230	int i;
 
3231	int ret;
 
3232
3233	if (mddev->level != 1) {
3234		pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n",
3235			mdname(mddev), mddev->level);
3236		return -EIO;
3237	}
3238	if (mddev->reshape_position != MaxSector) {
3239		pr_warn("md/raid1:%s: reshape_position set but not supported\n",
3240			mdname(mddev));
3241		return -EIO;
3242	}
3243
3244	/*
3245	 * copy the already verified devices into our private RAID1
3246	 * bookkeeping area. [whatever we allocate in run(),
3247	 * should be freed in raid1_free()]
3248	 */
3249	if (mddev->private == NULL)
3250		conf = setup_conf(mddev);
3251	else
3252		conf = mddev->private;
3253
3254	if (IS_ERR(conf))
3255		return PTR_ERR(conf);
3256
3257	if (!mddev_is_dm(mddev)) {
3258		ret = raid1_set_limits(mddev);
3259		if (ret)
3260			return ret;
 
 
 
 
 
 
3261	}
3262
3263	mddev->degraded = 0;
3264	for (i = 0; i < conf->raid_disks; i++)
3265		if (conf->mirrors[i].rdev == NULL ||
3266		    !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
3267		    test_bit(Faulty, &conf->mirrors[i].rdev->flags))
3268			mddev->degraded++;
3269	/*
3270	 * RAID1 needs at least one disk in active
3271	 */
3272	if (conf->raid_disks - mddev->degraded < 1) {
3273		md_unregister_thread(mddev, &conf->thread);
3274		return -EINVAL;
3275	}
3276
3277	if (conf->raid_disks - mddev->degraded == 1)
3278		mddev->recovery_cp = MaxSector;
3279
3280	if (mddev->recovery_cp != MaxSector)
3281		pr_info("md/raid1:%s: not clean -- starting background reconstruction\n",
3282			mdname(mddev));
3283	pr_info("md/raid1:%s: active with %d out of %d mirrors\n",
3284		mdname(mddev), mddev->raid_disks - mddev->degraded,
3285		mddev->raid_disks);
3286
3287	/*
3288	 * Ok, everything is just fine now
3289	 */
3290	rcu_assign_pointer(mddev->thread, conf->thread);
3291	rcu_assign_pointer(conf->thread, NULL);
3292	mddev->private = conf;
3293	set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
3294
3295	md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
3296
3297	ret = md_integrity_register(mddev);
3298	if (ret)
3299		md_unregister_thread(mddev, &mddev->thread);
 
 
 
 
 
 
 
 
 
 
 
3300	return ret;
3301}
3302
3303static void raid1_free(struct mddev *mddev, void *priv)
3304{
3305	struct r1conf *conf = priv;
3306
3307	mempool_exit(&conf->r1bio_pool);
3308	kfree(conf->mirrors);
3309	safe_put_page(conf->tmppage);
3310	kfree(conf->poolinfo);
3311	kfree(conf->nr_pending);
3312	kfree(conf->nr_waiting);
3313	kfree(conf->nr_queued);
3314	kfree(conf->barrier);
3315	bioset_exit(&conf->bio_split);
3316	kfree(conf);
3317}
3318
3319static int raid1_resize(struct mddev *mddev, sector_t sectors)
3320{
3321	/* no resync is happening, and there is enough space
3322	 * on all devices, so we can resize.
3323	 * We need to make sure resync covers any new space.
3324	 * If the array is shrinking we should possibly wait until
3325	 * any io in the removed space completes, but it hardly seems
3326	 * worth it.
3327	 */
3328	sector_t newsize = raid1_size(mddev, sectors, 0);
3329	int ret;
3330
3331	if (mddev->external_size &&
3332	    mddev->array_sectors > newsize)
3333		return -EINVAL;
3334
3335	ret = mddev->bitmap_ops->resize(mddev, newsize, 0, false);
3336	if (ret)
3337		return ret;
3338
3339	md_set_array_sectors(mddev, newsize);
 
 
3340	if (sectors > mddev->dev_sectors &&
3341	    mddev->recovery_cp > mddev->dev_sectors) {
3342		mddev->recovery_cp = mddev->dev_sectors;
3343		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3344	}
3345	mddev->dev_sectors = sectors;
3346	mddev->resync_max_sectors = sectors;
3347	return 0;
3348}
3349
3350static int raid1_reshape(struct mddev *mddev)
3351{
3352	/* We need to:
3353	 * 1/ resize the r1bio_pool
3354	 * 2/ resize conf->mirrors
3355	 *
3356	 * We allocate a new r1bio_pool if we can.
3357	 * Then raise a device barrier and wait until all IO stops.
3358	 * Then resize conf->mirrors and swap in the new r1bio pool.
3359	 *
3360	 * At the same time, we "pack" the devices so that all the missing
3361	 * devices have the higher raid_disk numbers.
3362	 */
3363	mempool_t newpool, oldpool;
3364	struct pool_info *newpoolinfo;
3365	struct raid1_info *newmirrors;
3366	struct r1conf *conf = mddev->private;
3367	int cnt, raid_disks;
3368	unsigned long flags;
3369	int d, d2;
3370	int ret;
3371
3372	memset(&newpool, 0, sizeof(newpool));
3373	memset(&oldpool, 0, sizeof(oldpool));
3374
3375	/* Cannot change chunk_size, layout, or level */
3376	if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
3377	    mddev->layout != mddev->new_layout ||
3378	    mddev->level != mddev->new_level) {
3379		mddev->new_chunk_sectors = mddev->chunk_sectors;
3380		mddev->new_layout = mddev->layout;
3381		mddev->new_level = mddev->level;
3382		return -EINVAL;
3383	}
3384
3385	if (!mddev_is_clustered(mddev))
3386		md_allow_write(mddev);
 
 
 
3387
3388	raid_disks = mddev->raid_disks + mddev->delta_disks;
3389
3390	if (raid_disks < conf->raid_disks) {
3391		cnt=0;
3392		for (d= 0; d < conf->raid_disks; d++)
3393			if (conf->mirrors[d].rdev)
3394				cnt++;
3395		if (cnt > raid_disks)
3396			return -EBUSY;
3397	}
3398
3399	newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3400	if (!newpoolinfo)
3401		return -ENOMEM;
3402	newpoolinfo->mddev = mddev;
3403	newpoolinfo->raid_disks = raid_disks * 2;
3404
3405	ret = mempool_init(&newpool, NR_RAID_BIOS, r1bio_pool_alloc,
3406			   rbio_pool_free, newpoolinfo);
3407	if (ret) {
3408		kfree(newpoolinfo);
3409		return ret;
3410	}
3411	newmirrors = kzalloc(array3_size(sizeof(struct raid1_info),
3412					 raid_disks, 2),
3413			     GFP_KERNEL);
3414	if (!newmirrors) {
3415		kfree(newpoolinfo);
3416		mempool_exit(&newpool);
3417		return -ENOMEM;
3418	}
3419
3420	freeze_array(conf, 0);
3421
3422	/* ok, everything is stopped */
3423	oldpool = conf->r1bio_pool;
3424	conf->r1bio_pool = newpool;
3425
3426	for (d = d2 = 0; d < conf->raid_disks; d++) {
3427		struct md_rdev *rdev = conf->mirrors[d].rdev;
3428		if (rdev && rdev->raid_disk != d2) {
3429			sysfs_unlink_rdev(mddev, rdev);
3430			rdev->raid_disk = d2;
3431			sysfs_unlink_rdev(mddev, rdev);
3432			if (sysfs_link_rdev(mddev, rdev))
3433				pr_warn("md/raid1:%s: cannot register rd%d\n",
3434					mdname(mddev), rdev->raid_disk);
3435		}
3436		if (rdev)
3437			newmirrors[d2++].rdev = rdev;
3438	}
3439	kfree(conf->mirrors);
3440	conf->mirrors = newmirrors;
3441	kfree(conf->poolinfo);
3442	conf->poolinfo = newpoolinfo;
3443
3444	spin_lock_irqsave(&conf->device_lock, flags);
3445	mddev->degraded += (raid_disks - conf->raid_disks);
3446	spin_unlock_irqrestore(&conf->device_lock, flags);
3447	conf->raid_disks = mddev->raid_disks = raid_disks;
3448	mddev->delta_disks = 0;
3449
3450	unfreeze_array(conf);
3451
3452	set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3453	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3454	md_wakeup_thread(mddev->thread);
3455
3456	mempool_exit(&oldpool);
3457	return 0;
3458}
3459
3460static void raid1_quiesce(struct mddev *mddev, int quiesce)
3461{
3462	struct r1conf *conf = mddev->private;
3463
3464	if (quiesce)
 
 
 
 
3465		freeze_array(conf, 0);
3466	else
 
3467		unfreeze_array(conf);
 
 
3468}
3469
3470static void *raid1_takeover(struct mddev *mddev)
3471{
3472	/* raid1 can take over:
3473	 *  raid5 with 2 devices, any layout or chunk size
3474	 */
3475	if (mddev->level == 5 && mddev->raid_disks == 2) {
3476		struct r1conf *conf;
3477		mddev->new_level = 1;
3478		mddev->new_layout = 0;
3479		mddev->new_chunk_sectors = 0;
3480		conf = setup_conf(mddev);
3481		if (!IS_ERR(conf)) {
3482			/* Array must appear to be quiesced */
3483			conf->array_frozen = 1;
3484			mddev_clear_unsupported_flags(mddev,
3485				UNSUPPORTED_MDDEV_FLAGS);
3486		}
3487		return conf;
3488	}
3489	return ERR_PTR(-EINVAL);
3490}
3491
3492static struct md_personality raid1_personality =
3493{
3494	.name		= "raid1",
3495	.level		= 1,
3496	.owner		= THIS_MODULE,
3497	.make_request	= raid1_make_request,
3498	.run		= raid1_run,
3499	.free		= raid1_free,
3500	.status		= raid1_status,
3501	.error_handler	= raid1_error,
3502	.hot_add_disk	= raid1_add_disk,
3503	.hot_remove_disk= raid1_remove_disk,
3504	.spare_active	= raid1_spare_active,
3505	.sync_request	= raid1_sync_request,
3506	.resize		= raid1_resize,
3507	.size		= raid1_size,
3508	.check_reshape	= raid1_reshape,
3509	.quiesce	= raid1_quiesce,
3510	.takeover	= raid1_takeover,
 
3511};
3512
3513static int __init raid_init(void)
3514{
3515	return register_md_personality(&raid1_personality);
3516}
3517
3518static void raid_exit(void)
3519{
3520	unregister_md_personality(&raid1_personality);
3521}
3522
3523module_init(raid_init);
3524module_exit(raid_exit);
3525MODULE_LICENSE("GPL");
3526MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3527MODULE_ALIAS("md-personality-3"); /* RAID1 */
3528MODULE_ALIAS("md-raid1");
3529MODULE_ALIAS("md-level-1");
v4.10.11
 
   1/*
   2 * raid1.c : Multiple Devices driver for Linux
   3 *
   4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
   5 *
   6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
   7 *
   8 * RAID-1 management functions.
   9 *
  10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
  11 *
  12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
  13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
  14 *
  15 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
  16 * bitmapped intelligence in resync:
  17 *
  18 *      - bitmap marked during normal i/o
  19 *      - bitmap used to skip nondirty blocks during sync
  20 *
  21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
  22 * - persistent bitmap code
  23 *
  24 * This program is free software; you can redistribute it and/or modify
  25 * it under the terms of the GNU General Public License as published by
  26 * the Free Software Foundation; either version 2, or (at your option)
  27 * any later version.
  28 *
  29 * You should have received a copy of the GNU General Public License
  30 * (for example /usr/src/linux/COPYING); if not, write to the Free
  31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  32 */
  33
  34#include <linux/slab.h>
  35#include <linux/delay.h>
  36#include <linux/blkdev.h>
  37#include <linux/module.h>
  38#include <linux/seq_file.h>
  39#include <linux/ratelimit.h>
 
 
  40#include <trace/events/block.h>
 
  41#include "md.h"
  42#include "raid1.h"
  43#include "bitmap.h"
  44
  45#define UNSUPPORTED_MDDEV_FLAGS		\
  46	((1L << MD_HAS_JOURNAL) |	\
  47	 (1L << MD_JOURNAL_CLEAN))
 
 
 
 
 
 
 
 
 
 
 
 
 
  48
  49/*
  50 * Number of guaranteed r1bios in case of extreme VM load:
  51 */
  52#define	NR_RAID1_BIOS 256
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  53
  54/* when we get a read error on a read-only array, we redirect to another
  55 * device without failing the first device, or trying to over-write to
  56 * correct the read error.  To keep track of bad blocks on a per-bio
  57 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
  58 */
  59#define IO_BLOCKED ((struct bio *)1)
  60/* When we successfully write to a known bad-block, we need to remove the
  61 * bad-block marking which must be done from process context.  So we record
  62 * the success by setting devs[n].bio to IO_MADE_GOOD
  63 */
  64#define IO_MADE_GOOD ((struct bio *)2)
  65
  66#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
 
 
 
 
 
  67
  68/* When there are this many requests queue to be written by
  69 * the raid1 thread, we become 'congested' to provide back-pressure
  70 * for writeback.
  71 */
  72static int max_queued_requests = 1024;
 
  73
  74static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
  75			  sector_t bi_sector);
  76static void lower_barrier(struct r1conf *conf);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  77
  78#define raid1_log(md, fmt, args...)				\
  79	do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0)
 
 
 
 
 
 
  80
  81static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
  82{
  83	struct pool_info *pi = data;
  84	int size = offsetof(struct r1bio, bios[pi->raid_disks]);
  85
  86	/* allocate a r1bio with room for raid_disks entries in the bios array */
  87	return kzalloc(size, gfp_flags);
  88}
  89
  90static void r1bio_pool_free(void *r1_bio, void *data)
  91{
  92	kfree(r1_bio);
  93}
  94
  95#define RESYNC_BLOCK_SIZE (64*1024)
  96#define RESYNC_DEPTH 32
  97#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
  98#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
  99#define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
 100#define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
 101#define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
 102#define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
 103#define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
 104
 105static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
 106{
 107	struct pool_info *pi = data;
 108	struct r1bio *r1_bio;
 109	struct bio *bio;
 110	int need_pages;
 111	int i, j;
 
 112
 113	r1_bio = r1bio_pool_alloc(gfp_flags, pi);
 114	if (!r1_bio)
 115		return NULL;
 116
 
 
 
 
 
 117	/*
 118	 * Allocate bios : 1 for reading, n-1 for writing
 119	 */
 120	for (j = pi->raid_disks ; j-- ; ) {
 121		bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
 122		if (!bio)
 123			goto out_free_bio;
 
 124		r1_bio->bios[j] = bio;
 125	}
 126	/*
 127	 * Allocate RESYNC_PAGES data pages and attach them to
 128	 * the first bio.
 129	 * If this is a user-requested check/repair, allocate
 130	 * RESYNC_PAGES for each bio.
 131	 */
 132	if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
 133		need_pages = pi->raid_disks;
 134	else
 135		need_pages = 1;
 136	for (j = 0; j < need_pages; j++) {
 
 
 137		bio = r1_bio->bios[j];
 138		bio->bi_vcnt = RESYNC_PAGES;
 139
 140		if (bio_alloc_pages(bio, gfp_flags))
 141			goto out_free_pages;
 142	}
 143	/* If not user-requests, copy the page pointers to all bios */
 144	if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
 145		for (i=0; i<RESYNC_PAGES ; i++)
 146			for (j=1; j<pi->raid_disks; j++)
 147				r1_bio->bios[j]->bi_io_vec[i].bv_page =
 148					r1_bio->bios[0]->bi_io_vec[i].bv_page;
 
 149	}
 150
 151	r1_bio->master_bio = NULL;
 152
 153	return r1_bio;
 154
 155out_free_pages:
 156	while (--j >= 0)
 157		bio_free_pages(r1_bio->bios[j]);
 158
 159out_free_bio:
 160	while (++j < pi->raid_disks)
 161		bio_put(r1_bio->bios[j]);
 162	r1bio_pool_free(r1_bio, data);
 
 
 
 
 
 163	return NULL;
 164}
 165
 166static void r1buf_pool_free(void *__r1_bio, void *data)
 167{
 168	struct pool_info *pi = data;
 169	int i,j;
 170	struct r1bio *r1bio = __r1_bio;
 
 171
 172	for (i = 0; i < RESYNC_PAGES; i++)
 173		for (j = pi->raid_disks; j-- ;) {
 174			if (j == 0 ||
 175			    r1bio->bios[j]->bi_io_vec[i].bv_page !=
 176			    r1bio->bios[0]->bi_io_vec[i].bv_page)
 177				safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
 178		}
 179	for (i=0 ; i < pi->raid_disks; i++)
 180		bio_put(r1bio->bios[i]);
 181
 182	r1bio_pool_free(r1bio, data);
 183}
 184
 185static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
 186{
 187	int i;
 188
 189	for (i = 0; i < conf->raid_disks * 2; i++) {
 190		struct bio **bio = r1_bio->bios + i;
 191		if (!BIO_SPECIAL(*bio))
 192			bio_put(*bio);
 193		*bio = NULL;
 194	}
 195}
 196
 197static void free_r1bio(struct r1bio *r1_bio)
 198{
 199	struct r1conf *conf = r1_bio->mddev->private;
 200
 201	put_all_bios(conf, r1_bio);
 202	mempool_free(r1_bio, conf->r1bio_pool);
 203}
 204
 205static void put_buf(struct r1bio *r1_bio)
 206{
 207	struct r1conf *conf = r1_bio->mddev->private;
 
 208	int i;
 209
 210	for (i = 0; i < conf->raid_disks * 2; i++) {
 211		struct bio *bio = r1_bio->bios[i];
 212		if (bio->bi_end_io)
 213			rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
 214	}
 215
 216	mempool_free(r1_bio, conf->r1buf_pool);
 217
 218	lower_barrier(conf);
 219}
 220
 221static void reschedule_retry(struct r1bio *r1_bio)
 222{
 223	unsigned long flags;
 224	struct mddev *mddev = r1_bio->mddev;
 225	struct r1conf *conf = mddev->private;
 
 226
 
 227	spin_lock_irqsave(&conf->device_lock, flags);
 228	list_add(&r1_bio->retry_list, &conf->retry_list);
 229	conf->nr_queued ++;
 230	spin_unlock_irqrestore(&conf->device_lock, flags);
 231
 232	wake_up(&conf->wait_barrier);
 233	md_wakeup_thread(mddev->thread);
 234}
 235
 236/*
 237 * raid_end_bio_io() is called when we have finished servicing a mirrored
 238 * operation and are ready to return a success/failure code to the buffer
 239 * cache layer.
 240 */
 241static void call_bio_endio(struct r1bio *r1_bio)
 242{
 243	struct bio *bio = r1_bio->master_bio;
 244	int done;
 245	struct r1conf *conf = r1_bio->mddev->private;
 246	sector_t start_next_window = r1_bio->start_next_window;
 247	sector_t bi_sector = bio->bi_iter.bi_sector;
 248
 249	if (bio->bi_phys_segments) {
 250		unsigned long flags;
 251		spin_lock_irqsave(&conf->device_lock, flags);
 252		bio->bi_phys_segments--;
 253		done = (bio->bi_phys_segments == 0);
 254		spin_unlock_irqrestore(&conf->device_lock, flags);
 255		/*
 256		 * make_request() might be waiting for
 257		 * bi_phys_segments to decrease
 258		 */
 259		wake_up(&conf->wait_barrier);
 260	} else
 261		done = 1;
 262
 263	if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
 264		bio->bi_error = -EIO;
 265
 266	if (done) {
 267		bio_endio(bio);
 268		/*
 269		 * Wake up any possible resync thread that waits for the device
 270		 * to go idle.
 271		 */
 272		allow_barrier(conf, start_next_window, bi_sector);
 273	}
 274}
 275
 276static void raid_end_bio_io(struct r1bio *r1_bio)
 277{
 278	struct bio *bio = r1_bio->master_bio;
 
 
 279
 280	/* if nobody has done the final endio yet, do it now */
 281	if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
 282		pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
 283			 (bio_data_dir(bio) == WRITE) ? "write" : "read",
 284			 (unsigned long long) bio->bi_iter.bi_sector,
 285			 (unsigned long long) bio_end_sector(bio) - 1);
 286
 287		call_bio_endio(r1_bio);
 288	}
 
 289	free_r1bio(r1_bio);
 
 
 
 
 
 290}
 291
 292/*
 293 * Update disk head position estimator based on IRQ completion info.
 294 */
 295static inline void update_head_pos(int disk, struct r1bio *r1_bio)
 296{
 297	struct r1conf *conf = r1_bio->mddev->private;
 298
 299	conf->mirrors[disk].head_position =
 300		r1_bio->sector + (r1_bio->sectors);
 301}
 302
 303/*
 304 * Find the disk number which triggered given bio
 305 */
 306static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
 307{
 308	int mirror;
 309	struct r1conf *conf = r1_bio->mddev->private;
 310	int raid_disks = conf->raid_disks;
 311
 312	for (mirror = 0; mirror < raid_disks * 2; mirror++)
 313		if (r1_bio->bios[mirror] == bio)
 314			break;
 315
 316	BUG_ON(mirror == raid_disks * 2);
 317	update_head_pos(mirror, r1_bio);
 318
 319	return mirror;
 320}
 321
 322static void raid1_end_read_request(struct bio *bio)
 323{
 324	int uptodate = !bio->bi_error;
 325	struct r1bio *r1_bio = bio->bi_private;
 326	struct r1conf *conf = r1_bio->mddev->private;
 327	struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;
 328
 329	/*
 330	 * this branch is our 'one mirror IO has finished' event handler:
 331	 */
 332	update_head_pos(r1_bio->read_disk, r1_bio);
 333
 334	if (uptodate)
 335		set_bit(R1BIO_Uptodate, &r1_bio->state);
 336	else if (test_bit(FailFast, &rdev->flags) &&
 337		 test_bit(R1BIO_FailFast, &r1_bio->state))
 338		/* This was a fail-fast read so we definitely
 339		 * want to retry */
 340		;
 341	else {
 342		/* If all other devices have failed, we want to return
 343		 * the error upwards rather than fail the last device.
 344		 * Here we redefine "uptodate" to mean "Don't want to retry"
 345		 */
 346		unsigned long flags;
 347		spin_lock_irqsave(&conf->device_lock, flags);
 348		if (r1_bio->mddev->degraded == conf->raid_disks ||
 349		    (r1_bio->mddev->degraded == conf->raid_disks-1 &&
 350		     test_bit(In_sync, &rdev->flags)))
 351			uptodate = 1;
 352		spin_unlock_irqrestore(&conf->device_lock, flags);
 353	}
 354
 355	if (uptodate) {
 356		raid_end_bio_io(r1_bio);
 357		rdev_dec_pending(rdev, conf->mddev);
 358	} else {
 359		/*
 360		 * oops, read error:
 361		 */
 362		char b[BDEVNAME_SIZE];
 363		pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n",
 364				   mdname(conf->mddev),
 365				   bdevname(rdev->bdev, b),
 366				   (unsigned long long)r1_bio->sector);
 367		set_bit(R1BIO_ReadError, &r1_bio->state);
 368		reschedule_retry(r1_bio);
 369		/* don't drop the reference on read_disk yet */
 370	}
 371}
 372
 373static void close_write(struct r1bio *r1_bio)
 374{
 
 
 375	/* it really is the end of this request */
 376	if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
 377		/* free extra copy of the data pages */
 378		int i = r1_bio->behind_page_count;
 379		while (i--)
 380			safe_put_page(r1_bio->behind_bvecs[i].bv_page);
 381		kfree(r1_bio->behind_bvecs);
 382		r1_bio->behind_bvecs = NULL;
 383	}
 384	/* clear the bitmap if all writes complete successfully */
 385	bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
 386			r1_bio->sectors,
 387			!test_bit(R1BIO_Degraded, &r1_bio->state),
 388			test_bit(R1BIO_BehindIO, &r1_bio->state));
 389	md_write_end(r1_bio->mddev);
 390}
 391
 392static void r1_bio_write_done(struct r1bio *r1_bio)
 393{
 394	if (!atomic_dec_and_test(&r1_bio->remaining))
 395		return;
 396
 397	if (test_bit(R1BIO_WriteError, &r1_bio->state))
 398		reschedule_retry(r1_bio);
 399	else {
 400		close_write(r1_bio);
 401		if (test_bit(R1BIO_MadeGood, &r1_bio->state))
 402			reschedule_retry(r1_bio);
 403		else
 404			raid_end_bio_io(r1_bio);
 405	}
 406}
 407
 408static void raid1_end_write_request(struct bio *bio)
 409{
 410	struct r1bio *r1_bio = bio->bi_private;
 411	int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
 412	struct r1conf *conf = r1_bio->mddev->private;
 413	struct bio *to_put = NULL;
 414	int mirror = find_bio_disk(r1_bio, bio);
 415	struct md_rdev *rdev = conf->mirrors[mirror].rdev;
 416	bool discard_error;
 
 
 417
 418	discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD;
 419
 420	/*
 421	 * 'one mirror IO has finished' event handler:
 422	 */
 423	if (bio->bi_error && !discard_error) {
 424		set_bit(WriteErrorSeen,	&rdev->flags);
 425		if (!test_and_set_bit(WantReplacement, &rdev->flags))
 426			set_bit(MD_RECOVERY_NEEDED, &
 427				conf->mddev->recovery);
 428
 429		if (test_bit(FailFast, &rdev->flags) &&
 430		    (bio->bi_opf & MD_FAILFAST) &&
 431		    /* We never try FailFast to WriteMostly devices */
 432		    !test_bit(WriteMostly, &rdev->flags)) {
 433			md_error(r1_bio->mddev, rdev);
 434			if (!test_bit(Faulty, &rdev->flags))
 435				/* This is the only remaining device,
 436				 * We need to retry the write without
 437				 * FailFast
 438				 */
 439				set_bit(R1BIO_WriteError, &r1_bio->state);
 440			else {
 441				/* Finished with this branch */
 442				r1_bio->bios[mirror] = NULL;
 443				to_put = bio;
 444			}
 445		} else
 446			set_bit(R1BIO_WriteError, &r1_bio->state);
 
 
 
 
 
 447	} else {
 448		/*
 449		 * Set R1BIO_Uptodate in our master bio, so that we
 450		 * will return a good error code for to the higher
 451		 * levels even if IO on some other mirrored buffer
 452		 * fails.
 453		 *
 454		 * The 'master' represents the composite IO operation
 455		 * to user-side. So if something waits for IO, then it
 456		 * will wait for the 'master' bio.
 457		 */
 458		sector_t first_bad;
 459		int bad_sectors;
 460
 461		r1_bio->bios[mirror] = NULL;
 462		to_put = bio;
 463		/*
 464		 * Do not set R1BIO_Uptodate if the current device is
 465		 * rebuilding or Faulty. This is because we cannot use
 466		 * such device for properly reading the data back (we could
 467		 * potentially use it, if the current write would have felt
 468		 * before rdev->recovery_offset, but for simplicity we don't
 469		 * check this here.
 470		 */
 471		if (test_bit(In_sync, &rdev->flags) &&
 472		    !test_bit(Faulty, &rdev->flags))
 473			set_bit(R1BIO_Uptodate, &r1_bio->state);
 474
 475		/* Maybe we can clear some bad blocks. */
 476		if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
 477				&first_bad, &bad_sectors) && !discard_error) {
 478			r1_bio->bios[mirror] = IO_MADE_GOOD;
 479			set_bit(R1BIO_MadeGood, &r1_bio->state);
 480		}
 481	}
 482
 483	if (behind) {
 
 
 484		if (test_bit(WriteMostly, &rdev->flags))
 485			atomic_dec(&r1_bio->behind_remaining);
 486
 487		/*
 488		 * In behind mode, we ACK the master bio once the I/O
 489		 * has safely reached all non-writemostly
 490		 * disks. Setting the Returned bit ensures that this
 491		 * gets done only once -- we don't ever want to return
 492		 * -EIO here, instead we'll wait
 493		 */
 494		if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
 495		    test_bit(R1BIO_Uptodate, &r1_bio->state)) {
 496			/* Maybe we can return now */
 497			if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
 498				struct bio *mbio = r1_bio->master_bio;
 499				pr_debug("raid1: behind end write sectors"
 500					 " %llu-%llu\n",
 501					 (unsigned long long) mbio->bi_iter.bi_sector,
 502					 (unsigned long long) bio_end_sector(mbio) - 1);
 503				call_bio_endio(r1_bio);
 504			}
 505		}
 506	}
 
 507	if (r1_bio->bios[mirror] == NULL)
 508		rdev_dec_pending(rdev, conf->mddev);
 509
 510	/*
 511	 * Let's see if all mirrored write operations have finished
 512	 * already.
 513	 */
 514	r1_bio_write_done(r1_bio);
 515
 516	if (to_put)
 517		bio_put(to_put);
 518}
 519
 520/*
 521 * This routine returns the disk from which the requested read should
 522 * be done. There is a per-array 'next expected sequential IO' sector
 523 * number - if this matches on the next IO then we use the last disk.
 524 * There is also a per-disk 'last know head position' sector that is
 525 * maintained from IRQ contexts, both the normal and the resync IO
 526 * completion handlers update this position correctly. If there is no
 527 * perfect sequential match then we pick the disk whose head is closest.
 528 *
 529 * If there are 2 mirrors in the same 2 devices, performance degrades
 530 * because position is mirror, not device based.
 531 *
 532 * The rdev for the device selected will have nr_pending incremented.
 533 */
 534static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 535{
 536	const sector_t this_sector = r1_bio->sector;
 537	int sectors;
 538	int best_good_sectors;
 539	int best_disk, best_dist_disk, best_pending_disk;
 540	int has_nonrot_disk;
 541	int disk;
 542	sector_t best_dist;
 543	unsigned int min_pending;
 544	struct md_rdev *rdev;
 545	int choose_first;
 546	int choose_next_idle;
 547
 548	rcu_read_lock();
 549	/*
 550	 * Check if we can balance. We can balance on the whole
 551	 * device if no resync is going on, or below the resync window.
 552	 * We take the first readable disk when above the resync window.
 553	 */
 554 retry:
 555	sectors = r1_bio->sectors;
 556	best_disk = -1;
 557	best_dist_disk = -1;
 558	best_dist = MaxSector;
 559	best_pending_disk = -1;
 560	min_pending = UINT_MAX;
 561	best_good_sectors = 0;
 562	has_nonrot_disk = 0;
 563	choose_next_idle = 0;
 564	clear_bit(R1BIO_FailFast, &r1_bio->state);
 
 
 
 
 
 
 
 
 
 
 
 565
 566	if ((conf->mddev->recovery_cp < this_sector + sectors) ||
 567	    (mddev_is_clustered(conf->mddev) &&
 568	    md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
 569		    this_sector + sectors)))
 570		choose_first = 1;
 571	else
 572		choose_first = 0;
 573
 574	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
 575		sector_t dist;
 576		sector_t first_bad;
 577		int bad_sectors;
 578		unsigned int pending;
 579		bool nonrot;
 580
 581		rdev = rcu_dereference(conf->mirrors[disk].rdev);
 582		if (r1_bio->bios[disk] == IO_BLOCKED
 583		    || rdev == NULL
 584		    || test_bit(Faulty, &rdev->flags))
 585			continue;
 586		if (!test_bit(In_sync, &rdev->flags) &&
 587		    rdev->recovery_offset < this_sector + sectors)
 
 
 
 588			continue;
 589		if (test_bit(WriteMostly, &rdev->flags)) {
 590			/* Don't balance among write-mostly, just
 591			 * use the first as a last resort */
 592			if (best_dist_disk < 0) {
 593				if (is_badblock(rdev, this_sector, sectors,
 594						&first_bad, &bad_sectors)) {
 595					if (first_bad <= this_sector)
 596						/* Cannot use this */
 597						continue;
 598					best_good_sectors = first_bad - this_sector;
 599				} else
 600					best_good_sectors = sectors;
 601				best_dist_disk = disk;
 602				best_pending_disk = disk;
 603			}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 604			continue;
 
 
 
 
 
 
 
 
 605		}
 606		/* This is a reasonable device to use.  It might
 607		 * even be best.
 
 
 608		 */
 609		if (is_badblock(rdev, this_sector, sectors,
 610				&first_bad, &bad_sectors)) {
 611			if (best_dist < MaxSector)
 612				/* already have a better device */
 613				continue;
 614			if (first_bad <= this_sector) {
 615				/* cannot read here. If this is the 'primary'
 616				 * device, then we must not read beyond
 617				 * bad_sectors from another device..
 618				 */
 619				bad_sectors -= (this_sector - first_bad);
 620				if (choose_first && sectors > bad_sectors)
 621					sectors = bad_sectors;
 622				if (best_good_sectors > sectors)
 623					best_good_sectors = sectors;
 624
 625			} else {
 626				sector_t good_sectors = first_bad - this_sector;
 627				if (good_sectors > best_good_sectors) {
 628					best_good_sectors = good_sectors;
 629					best_disk = disk;
 630				}
 631				if (choose_first)
 632					break;
 633			}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 634			continue;
 635		} else
 636			best_good_sectors = sectors;
 637
 638		if (best_disk >= 0)
 639			/* At least two disks to choose from so failfast is OK */
 640			set_bit(R1BIO_FailFast, &r1_bio->state);
 641
 642		nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
 643		has_nonrot_disk |= nonrot;
 644		pending = atomic_read(&rdev->nr_pending);
 645		dist = abs(this_sector - conf->mirrors[disk].head_position);
 646		if (choose_first) {
 647			best_disk = disk;
 648			break;
 649		}
 650		/* Don't change to another disk for sequential reads */
 651		if (conf->mirrors[disk].next_seq_sect == this_sector
 652		    || dist == 0) {
 653			int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
 654			struct raid1_info *mirror = &conf->mirrors[disk];
 655
 656			best_disk = disk;
 657			/*
 658			 * If buffered sequential IO size exceeds optimal
 659			 * iosize, check if there is idle disk. If yes, choose
 660			 * the idle disk. read_balance could already choose an
 661			 * idle disk before noticing it's a sequential IO in
 662			 * this disk. This doesn't matter because this disk
 663			 * will idle, next time it will be utilized after the
 664			 * first disk has IO size exceeds optimal iosize. In
 665			 * this way, iosize of the first disk will be optimal
 666			 * iosize at least. iosize of the second disk might be
 667			 * small, but not a big deal since when the second disk
 668			 * starts IO, the first disk is likely still busy.
 669			 */
 670			if (nonrot && opt_iosize > 0 &&
 671			    mirror->seq_start != MaxSector &&
 672			    mirror->next_seq_sect > opt_iosize &&
 673			    mirror->next_seq_sect - opt_iosize >=
 674			    mirror->seq_start) {
 675				choose_next_idle = 1;
 676				continue;
 677			}
 678			break;
 679		}
 680
 681		if (choose_next_idle)
 682			continue;
 683
 684		if (min_pending > pending) {
 685			min_pending = pending;
 686			best_pending_disk = disk;
 687		}
 688
 689		if (dist < best_dist) {
 690			best_dist = dist;
 691			best_dist_disk = disk;
 692		}
 693	}
 694
 695	/*
 
 
 
 
 
 
 
 696	 * If all disks are rotational, choose the closest disk. If any disk is
 697	 * non-rotational, choose the disk with less pending request even the
 698	 * disk is rotational, which might/might not be optimal for raids with
 699	 * mixed ratation/non-rotational disks depending on workload.
 700	 */
 701	if (best_disk == -1) {
 702		if (has_nonrot_disk || min_pending == 0)
 703			best_disk = best_pending_disk;
 704		else
 705			best_disk = best_dist_disk;
 706	}
 707
 708	if (best_disk >= 0) {
 709		rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
 710		if (!rdev)
 711			goto retry;
 712		atomic_inc(&rdev->nr_pending);
 713		sectors = best_good_sectors;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 714
 715		if (conf->mirrors[best_disk].next_seq_sect != this_sector)
 716			conf->mirrors[best_disk].seq_start = this_sector;
 717
 718		conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
 
 
 
 
 
 
 
 
 
 719	}
 720	rcu_read_unlock();
 721	*max_sectors = sectors;
 722
 723	return best_disk;
 
 
 
 
 
 
 
 
 
 724}
 725
 726static int raid1_congested(struct mddev *mddev, int bits)
 727{
 728	struct r1conf *conf = mddev->private;
 729	int i, ret = 0;
 
 730
 731	if ((bits & (1 << WB_async_congested)) &&
 732	    conf->pending_count >= max_queued_requests)
 733		return 1;
 
 
 734
 735	rcu_read_lock();
 736	for (i = 0; i < conf->raid_disks * 2; i++) {
 737		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
 738		if (rdev && !test_bit(Faulty, &rdev->flags)) {
 739			struct request_queue *q = bdev_get_queue(rdev->bdev);
 740
 741			BUG_ON(!q);
 742
 743			/* Note the '|| 1' - when read_balance prefers
 744			 * non-congested targets, it can be removed
 745			 */
 746			if ((bits & (1 << WB_async_congested)) || 1)
 747				ret |= bdi_congested(&q->backing_dev_info, bits);
 748			else
 749				ret &= bdi_congested(&q->backing_dev_info, bits);
 750		}
 751	}
 752	rcu_read_unlock();
 753	return ret;
 754}
 755
 756static void flush_pending_writes(struct r1conf *conf)
 757{
 758	/* Any writes that have been queued but are awaiting
 759	 * bitmap updates get flushed here.
 760	 */
 761	spin_lock_irq(&conf->device_lock);
 762
 763	if (conf->pending_bio_list.head) {
 
 764		struct bio *bio;
 
 765		bio = bio_list_get(&conf->pending_bio_list);
 766		conf->pending_count = 0;
 767		spin_unlock_irq(&conf->device_lock);
 768		/* flush any pending bitmap writes to
 769		 * disk before proceeding w/ I/O */
 770		bitmap_unplug(conf->mddev->bitmap);
 771		wake_up(&conf->wait_barrier);
 772
 773		while (bio) { /* submit pending writes */
 774			struct bio *next = bio->bi_next;
 775			struct md_rdev *rdev = (void*)bio->bi_bdev;
 776			bio->bi_next = NULL;
 777			bio->bi_bdev = rdev->bdev;
 778			if (test_bit(Faulty, &rdev->flags)) {
 779				bio->bi_error = -EIO;
 780				bio_endio(bio);
 781			} else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
 782					    !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
 783				/* Just ignore it */
 784				bio_endio(bio);
 785			else
 786				generic_make_request(bio);
 787			bio = next;
 788		}
 789	} else
 790		spin_unlock_irq(&conf->device_lock);
 791}
 792
 793/* Barriers....
 794 * Sometimes we need to suspend IO while we do something else,
 795 * either some resync/recovery, or reconfigure the array.
 796 * To do this we raise a 'barrier'.
 797 * The 'barrier' is a counter that can be raised multiple times
 798 * to count how many activities are happening which preclude
 799 * normal IO.
 800 * We can only raise the barrier if there is no pending IO.
 801 * i.e. if nr_pending == 0.
 802 * We choose only to raise the barrier if no-one is waiting for the
 803 * barrier to go down.  This means that as soon as an IO request
 804 * is ready, no other operations which require a barrier will start
 805 * until the IO request has had a chance.
 806 *
 807 * So: regular IO calls 'wait_barrier'.  When that returns there
 808 *    is no backgroup IO happening,  It must arrange to call
 809 *    allow_barrier when it has finished its IO.
 810 * backgroup IO calls must call raise_barrier.  Once that returns
 811 *    there is no normal IO happeing.  It must arrange to call
 812 *    lower_barrier when the particular background IO completes.
 
 
 
 813 */
 814static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
 815{
 
 
 816	spin_lock_irq(&conf->resync_lock);
 817
 818	/* Wait until no block IO is waiting */
 819	wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
 
 820			    conf->resync_lock);
 821
 822	/* block any new IO from starting */
 823	conf->barrier++;
 824	conf->next_resync = sector_nr;
 
 
 
 
 
 
 
 
 825
 826	/* For these conditions we must wait:
 827	 * A: while the array is in frozen state
 828	 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
 829	 *    the max count which allowed.
 830	 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
 831	 *    next resync will reach to the window which normal bios are
 832	 *    handling.
 833	 * D: while there are any active requests in the current window.
 834	 */
 835	wait_event_lock_irq(conf->wait_barrier,
 836			    !conf->array_frozen &&
 837			    conf->barrier < RESYNC_DEPTH &&
 838			    conf->current_window_requests == 0 &&
 839			    (conf->start_next_window >=
 840			     conf->next_resync + RESYNC_SECTORS),
 841			    conf->resync_lock);
 842
 843	conf->nr_pending++;
 
 
 
 
 
 
 
 844	spin_unlock_irq(&conf->resync_lock);
 
 
 845}
 846
 847static void lower_barrier(struct r1conf *conf)
 848{
 849	unsigned long flags;
 850	BUG_ON(conf->barrier <= 0);
 851	spin_lock_irqsave(&conf->resync_lock, flags);
 852	conf->barrier--;
 853	conf->nr_pending--;
 854	spin_unlock_irqrestore(&conf->resync_lock, flags);
 855	wake_up(&conf->wait_barrier);
 856}
 857
 858static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
 859{
 860	bool wait = false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 861
 862	if (conf->array_frozen || !bio)
 863		wait = true;
 864	else if (conf->barrier && bio_data_dir(bio) == WRITE) {
 865		if ((conf->mddev->curr_resync_completed
 866		     >= bio_end_sector(bio)) ||
 867		    (conf->start_next_window + NEXT_NORMALIO_DISTANCE
 868		     <= bio->bi_iter.bi_sector))
 869			wait = false;
 870		else
 871			wait = true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 872	}
 873
 874	return wait;
 
 
 875}
 876
 877static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
 878{
 879	sector_t sector = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 880
 881	spin_lock_irq(&conf->resync_lock);
 882	if (need_to_wait_for_sync(conf, bio)) {
 883		conf->nr_waiting++;
 884		/* Wait for the barrier to drop.
 885		 * However if there are already pending
 886		 * requests (preventing the barrier from
 887		 * rising completely), and the
 888		 * per-process bio queue isn't empty,
 889		 * then don't wait, as we need to empty
 890		 * that queue to allow conf->start_next_window
 891		 * to increase.
 892		 */
 893		raid1_log(conf->mddev, "wait barrier");
 
 
 894		wait_event_lock_irq(conf->wait_barrier,
 895				    !conf->array_frozen &&
 896				    (!conf->barrier ||
 897				     ((conf->start_next_window <
 898				       conf->next_resync + RESYNC_SECTORS) &&
 899				      current->bio_list &&
 900				      !bio_list_empty(current->bio_list))),
 901				    conf->resync_lock);
 902		conf->nr_waiting--;
 903	}
 904
 905	if (bio && bio_data_dir(bio) == WRITE) {
 906		if (bio->bi_iter.bi_sector >= conf->next_resync) {
 907			if (conf->start_next_window == MaxSector)
 908				conf->start_next_window =
 909					conf->next_resync +
 910					NEXT_NORMALIO_DISTANCE;
 911
 912			if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
 913			    <= bio->bi_iter.bi_sector)
 914				conf->next_window_requests++;
 915			else
 916				conf->current_window_requests++;
 917			sector = conf->start_next_window;
 918		}
 919	}
 920
 921	conf->nr_pending++;
 922	spin_unlock_irq(&conf->resync_lock);
 923	return sector;
 924}
 925
 926static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
 927			  sector_t bi_sector)
 928{
 929	unsigned long flags;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 930
 931	spin_lock_irqsave(&conf->resync_lock, flags);
 932	conf->nr_pending--;
 933	if (start_next_window) {
 934		if (start_next_window == conf->start_next_window) {
 935			if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
 936			    <= bi_sector)
 937				conf->next_window_requests--;
 938			else
 939				conf->current_window_requests--;
 940		} else
 941			conf->current_window_requests--;
 942
 943		if (!conf->current_window_requests) {
 944			if (conf->next_window_requests) {
 945				conf->current_window_requests =
 946					conf->next_window_requests;
 947				conf->next_window_requests = 0;
 948				conf->start_next_window +=
 949					NEXT_NORMALIO_DISTANCE;
 950			} else
 951				conf->start_next_window = MaxSector;
 952		}
 953	}
 954	spin_unlock_irqrestore(&conf->resync_lock, flags);
 955	wake_up(&conf->wait_barrier);
 956}
 957
 958static void freeze_array(struct r1conf *conf, int extra)
 959{
 960	/* stop syncio and normal IO and wait for everything to
 961	 * go quite.
 962	 * We wait until nr_pending match nr_queued+extra
 963	 * This is called in the context of one normal IO request
 964	 * that has failed. Thus any sync request that might be pending
 965	 * will be blocked by nr_pending, and we need to wait for
 966	 * pending IO requests to complete or be queued for re-try.
 967	 * Thus the number queued (nr_queued) plus this request (extra)
 968	 * must match the number of pending IOs (nr_pending) before
 969	 * we continue.
 
 
 
 
 
 
 
 
 
 
 
 
 970	 */
 971	spin_lock_irq(&conf->resync_lock);
 972	conf->array_frozen = 1;
 973	raid1_log(conf->mddev, "wait freeze");
 974	wait_event_lock_irq_cmd(conf->wait_barrier,
 975				conf->nr_pending == conf->nr_queued+extra,
 976				conf->resync_lock,
 977				flush_pending_writes(conf));
 
 978	spin_unlock_irq(&conf->resync_lock);
 979}
 980static void unfreeze_array(struct r1conf *conf)
 981{
 982	/* reverse the effect of the freeze */
 983	spin_lock_irq(&conf->resync_lock);
 984	conf->array_frozen = 0;
 
 985	wake_up(&conf->wait_barrier);
 986	spin_unlock_irq(&conf->resync_lock);
 987}
 988
 989/* duplicate the data pages for behind I/O
 990 */
 991static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
 992{
 993	int i;
 994	struct bio_vec *bvec;
 995	struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
 996					GFP_NOIO);
 997	if (unlikely(!bvecs))
 998		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 999
1000	bio_for_each_segment_all(bvec, bio, i) {
1001		bvecs[i] = *bvec;
1002		bvecs[i].bv_page = alloc_page(GFP_NOIO);
1003		if (unlikely(!bvecs[i].bv_page))
1004			goto do_sync_io;
1005		memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
1006		       kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
1007		kunmap(bvecs[i].bv_page);
1008		kunmap(bvec->bv_page);
1009	}
1010	r1_bio->behind_bvecs = bvecs;
1011	r1_bio->behind_page_count = bio->bi_vcnt;
 
 
1012	set_bit(R1BIO_BehindIO, &r1_bio->state);
 
1013	return;
1014
1015do_sync_io:
1016	for (i = 0; i < bio->bi_vcnt; i++)
1017		if (bvecs[i].bv_page)
1018			put_page(bvecs[i].bv_page);
1019	kfree(bvecs);
1020	pr_debug("%dB behind alloc failed, doing sync I/O\n",
1021		 bio->bi_iter.bi_size);
 
 
1022}
1023
1024struct raid1_plug_cb {
1025	struct blk_plug_cb	cb;
1026	struct bio_list		pending;
1027	int			pending_cnt;
1028};
1029
1030static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
1031{
1032	struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1033						  cb);
1034	struct mddev *mddev = plug->cb.data;
1035	struct r1conf *conf = mddev->private;
1036	struct bio *bio;
1037
1038	if (from_schedule || current->bio_list) {
1039		spin_lock_irq(&conf->device_lock);
1040		bio_list_merge(&conf->pending_bio_list, &plug->pending);
1041		conf->pending_count += plug->pending_cnt;
1042		spin_unlock_irq(&conf->device_lock);
1043		wake_up(&conf->wait_barrier);
1044		md_wakeup_thread(mddev->thread);
1045		kfree(plug);
1046		return;
1047	}
1048
1049	/* we aren't scheduling, so we can do the write-out directly. */
1050	bio = bio_list_get(&plug->pending);
1051	bitmap_unplug(mddev->bitmap);
1052	wake_up(&conf->wait_barrier);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1053
1054	while (bio) { /* submit pending writes */
1055		struct bio *next = bio->bi_next;
1056		struct md_rdev *rdev = (void*)bio->bi_bdev;
1057		bio->bi_next = NULL;
1058		bio->bi_bdev = rdev->bdev;
1059		if (test_bit(Faulty, &rdev->flags)) {
1060			bio->bi_error = -EIO;
1061			bio_endio(bio);
1062		} else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
1063				    !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1064			/* Just ignore it */
1065			bio_endio(bio);
1066		else
1067			generic_make_request(bio);
1068		bio = next;
1069	}
1070	kfree(plug);
1071}
1072
1073static void raid1_read_request(struct mddev *mddev, struct bio *bio,
1074				 struct r1bio *r1_bio)
1075{
1076	struct r1conf *conf = mddev->private;
1077	struct raid1_info *mirror;
1078	struct bio *read_bio;
1079	struct bitmap *bitmap = mddev->bitmap;
1080	const int op = bio_op(bio);
1081	const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1082	int sectors_handled;
1083	int max_sectors;
1084	int rdisk;
 
1085
1086	wait_barrier(conf, bio);
 
 
 
 
 
1087
1088read_again:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1089	rdisk = read_balance(conf, r1_bio, &max_sectors);
1090
1091	if (rdisk < 0) {
1092		/* couldn't find anywhere to read from */
 
 
 
 
 
1093		raid_end_bio_io(r1_bio);
1094		return;
1095	}
1096	mirror = conf->mirrors + rdisk;
1097
1098	if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1099	    bitmap) {
 
 
 
 
 
1100		/*
1101		 * Reading from a write-mostly device must take care not to
1102		 * over-take any writes that are 'behind'
1103		 */
1104		raid1_log(mddev, "wait behind writes");
1105		wait_event(bitmap->behind_wait,
1106			   atomic_read(&bitmap->behind_writes) == 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1107	}
 
1108	r1_bio->read_disk = rdisk;
1109	r1_bio->start_next_window = 0;
1110
1111	read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1112	bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
1113		 max_sectors);
 
1114
1115	r1_bio->bios[rdisk] = read_bio;
1116
1117	read_bio->bi_iter.bi_sector = r1_bio->sector +
1118		mirror->rdev->data_offset;
1119	read_bio->bi_bdev = mirror->rdev->bdev;
1120	read_bio->bi_end_io = raid1_end_read_request;
1121	bio_set_op_attrs(read_bio, op, do_sync);
1122	if (test_bit(FailFast, &mirror->rdev->flags) &&
1123	    test_bit(R1BIO_FailFast, &r1_bio->state))
1124	        read_bio->bi_opf |= MD_FAILFAST;
1125	read_bio->bi_private = r1_bio;
 
 
 
1126
1127	if (mddev->gendisk)
1128	        trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev),
1129	                              read_bio, disk_devt(mddev->gendisk),
1130	                              r1_bio->sector);
 
 
1131
1132	if (max_sectors < r1_bio->sectors) {
1133		/*
1134		 * could not read all from this device, so we will need another
1135		 * r1_bio.
1136		 */
1137		sectors_handled = (r1_bio->sector + max_sectors
1138				   - bio->bi_iter.bi_sector);
1139		r1_bio->sectors = max_sectors;
1140		spin_lock_irq(&conf->device_lock);
1141		if (bio->bi_phys_segments == 0)
1142			bio->bi_phys_segments = 2;
1143		else
1144			bio->bi_phys_segments++;
1145		spin_unlock_irq(&conf->device_lock);
1146
1147		/*
1148		 * Cannot call generic_make_request directly as that will be
1149		 * queued in __make_request and subsequent mempool_alloc might
1150		 * block waiting for it.  So hand bio over to raid1d.
1151		 */
1152		reschedule_retry(r1_bio);
 
 
 
1153
1154		r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
 
 
 
 
 
 
1155
1156		r1_bio->master_bio = bio;
1157		r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1158		r1_bio->state = 0;
1159		r1_bio->mddev = mddev;
1160		r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
1161		goto read_again;
1162	} else
1163		generic_make_request(read_bio);
1164}
1165
1166static void raid1_write_request(struct mddev *mddev, struct bio *bio,
1167				struct r1bio *r1_bio)
1168{
1169	struct r1conf *conf = mddev->private;
1170	int i, disks;
1171	struct bitmap *bitmap = mddev->bitmap;
1172	unsigned long flags;
1173	const int op = bio_op(bio);
1174	const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1175	const unsigned long do_flush_fua = (bio->bi_opf &
1176						(REQ_PREFLUSH | REQ_FUA));
1177	struct md_rdev *blocked_rdev;
1178	struct blk_plug_cb *cb;
1179	struct raid1_plug_cb *plug = NULL;
1180	int first_clone;
1181	int sectors_handled;
1182	int max_sectors;
1183	sector_t start_next_window;
 
1184
1185	/*
1186	 * Register the new request and wait if the reconstruction
1187	 * thread has put up a bar for new requests.
1188	 * Continue immediately if no resync is active currently.
1189	 */
1190
1191	md_write_start(mddev, bio); /* wait on superblock update early */
1192
1193	if ((bio_end_sector(bio) > mddev->suspend_lo &&
1194	    bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1195	    (mddev_is_clustered(mddev) &&
1196	     md_cluster_ops->area_resyncing(mddev, WRITE,
1197		     bio->bi_iter.bi_sector, bio_end_sector(bio)))) {
1198
1199		/*
1200		 * As the suspend_* range is controlled by userspace, we want
1201		 * an interruptible wait.
1202		 */
1203		DEFINE_WAIT(w);
 
 
 
 
1204		for (;;) {
1205			flush_signals(current);
1206			prepare_to_wait(&conf->wait_barrier,
1207					&w, TASK_INTERRUPTIBLE);
1208			if (bio_end_sector(bio) <= mddev->suspend_lo ||
1209			    bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1210			    (mddev_is_clustered(mddev) &&
1211			     !md_cluster_ops->area_resyncing(mddev, WRITE,
1212				     bio->bi_iter.bi_sector,
1213				     bio_end_sector(bio))))
1214				break;
1215			schedule();
1216		}
1217		finish_wait(&conf->wait_barrier, &w);
1218	}
1219	start_next_window = wait_barrier(conf, bio);
1220
1221	if (conf->pending_count >= max_queued_requests) {
1222		md_wakeup_thread(mddev->thread);
1223		raid1_log(mddev, "wait queued");
1224		wait_event(conf->wait_barrier,
1225			   conf->pending_count < max_queued_requests);
 
 
 
 
 
 
 
 
 
1226	}
 
 
 
 
1227	/* first select target devices under rcu_lock and
1228	 * inc refcount on their rdev.  Record them by setting
1229	 * bios[x] to bio
1230	 * If there are known/acknowledged bad blocks on any device on
1231	 * which we have seen a write error, we want to avoid writing those
1232	 * blocks.
1233	 * This potentially requires several writes to write around
1234	 * the bad blocks.  Each set of writes gets it's own r1bio
1235	 * with a set of bios attached.
1236	 */
1237
1238	disks = conf->raid_disks * 2;
1239 retry_write:
1240	r1_bio->start_next_window = start_next_window;
1241	blocked_rdev = NULL;
1242	rcu_read_lock();
1243	max_sectors = r1_bio->sectors;
1244	for (i = 0;  i < disks; i++) {
1245		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1246		if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1247			atomic_inc(&rdev->nr_pending);
1248			blocked_rdev = rdev;
1249			break;
1250		}
 
 
 
 
1251		r1_bio->bios[i] = NULL;
1252		if (!rdev || test_bit(Faulty, &rdev->flags)) {
1253			if (i < conf->raid_disks)
1254				set_bit(R1BIO_Degraded, &r1_bio->state);
1255			continue;
1256		}
1257
1258		atomic_inc(&rdev->nr_pending);
1259		if (test_bit(WriteErrorSeen, &rdev->flags)) {
1260			sector_t first_bad;
1261			int bad_sectors;
1262			int is_bad;
1263
1264			is_bad = is_badblock(rdev, r1_bio->sector, max_sectors,
1265					     &first_bad, &bad_sectors);
1266			if (is_bad < 0) {
1267				/* mustn't write here until the bad block is
1268				 * acknowledged*/
1269				set_bit(BlockedBadBlocks, &rdev->flags);
1270				blocked_rdev = rdev;
1271				break;
1272			}
1273			if (is_bad && first_bad <= r1_bio->sector) {
1274				/* Cannot write here at all */
1275				bad_sectors -= (r1_bio->sector - first_bad);
1276				if (bad_sectors < max_sectors)
1277					/* mustn't write more than bad_sectors
1278					 * to other devices yet
1279					 */
1280					max_sectors = bad_sectors;
1281				rdev_dec_pending(rdev, mddev);
1282				/* We don't set R1BIO_Degraded as that
1283				 * only applies if the disk is
1284				 * missing, so it might be re-added,
1285				 * and we want to know to recover this
1286				 * chunk.
1287				 * In this case the device is here,
1288				 * and the fact that this chunk is not
1289				 * in-sync is recorded in the bad
1290				 * block log
1291				 */
1292				continue;
1293			}
1294			if (is_bad) {
1295				int good_sectors = first_bad - r1_bio->sector;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1296				if (good_sectors < max_sectors)
1297					max_sectors = good_sectors;
1298			}
1299		}
1300		r1_bio->bios[i] = bio;
1301	}
1302	rcu_read_unlock();
1303
1304	if (unlikely(blocked_rdev)) {
1305		/* Wait for this device to become unblocked */
1306		int j;
1307		sector_t old = start_next_window;
1308
1309		for (j = 0; j < i; j++)
1310			if (r1_bio->bios[j])
1311				rdev_dec_pending(conf->mirrors[j].rdev, mddev);
1312		r1_bio->state = 0;
1313		allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
1314		raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk);
1315		md_wait_for_blocked_rdev(blocked_rdev, mddev);
1316		start_next_window = wait_barrier(conf, bio);
1317		/*
1318		 * We must make sure the multi r1bios of bio have
1319		 * the same value of bi_phys_segments
1320		 */
1321		if (bio->bi_phys_segments && old &&
1322		    old != start_next_window)
1323			/* Wait for the former r1bio(s) to complete */
1324			wait_event(conf->wait_barrier,
1325				   bio->bi_phys_segments == 1);
1326		goto retry_write;
1327	}
1328
1329	if (max_sectors < r1_bio->sectors) {
1330		/* We are splitting this write into multiple parts, so
1331		 * we need to prepare for allocating another r1_bio.
1332		 */
1333		r1_bio->sectors = max_sectors;
1334		spin_lock_irq(&conf->device_lock);
1335		if (bio->bi_phys_segments == 0)
1336			bio->bi_phys_segments = 2;
1337		else
1338			bio->bi_phys_segments++;
1339		spin_unlock_irq(&conf->device_lock);
1340	}
1341	sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
1342
 
 
1343	atomic_set(&r1_bio->remaining, 1);
1344	atomic_set(&r1_bio->behind_remaining, 0);
1345
1346	first_clone = 1;
 
1347	for (i = 0; i < disks; i++) {
1348		struct bio *mbio;
 
1349		if (!r1_bio->bios[i])
1350			continue;
1351
1352		mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1353		bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector,
1354			 max_sectors);
 
 
1355
1356		if (first_clone) {
1357			/* do behind I/O ?
1358			 * Not if there are too many, or cannot
1359			 * allocate memory, or a reader on WriteMostly
1360			 * is waiting for behind writes to flush */
1361			if (bitmap &&
1362			    (atomic_read(&bitmap->behind_writes)
1363			     < mddev->bitmap_info.max_write_behind) &&
1364			    !waitqueue_active(&bitmap->behind_wait))
1365				alloc_behind_pages(mbio, r1_bio);
1366
1367			bitmap_startwrite(bitmap, r1_bio->sector,
1368					  r1_bio->sectors,
1369					  test_bit(R1BIO_BehindIO,
1370						   &r1_bio->state));
1371			first_clone = 0;
1372		}
1373		if (r1_bio->behind_bvecs) {
1374			struct bio_vec *bvec;
1375			int j;
1376
1377			/*
1378			 * We trimmed the bio, so _all is legit
1379			 */
1380			bio_for_each_segment_all(bvec, mbio, j)
1381				bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
1382			if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
 
1383				atomic_inc(&r1_bio->behind_remaining);
 
 
 
 
 
 
1384		}
1385
1386		r1_bio->bios[i] = mbio;
1387
1388		mbio->bi_iter.bi_sector	= (r1_bio->sector +
1389				   conf->mirrors[i].rdev->data_offset);
1390		mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1391		mbio->bi_end_io	= raid1_end_write_request;
1392		bio_set_op_attrs(mbio, op, do_flush_fua | do_sync);
1393		if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) &&
1394		    !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) &&
 
1395		    conf->raid_disks - mddev->degraded > 1)
1396			mbio->bi_opf |= MD_FAILFAST;
1397		mbio->bi_private = r1_bio;
1398
1399		atomic_inc(&r1_bio->remaining);
1400
1401		if (mddev->gendisk)
1402			trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev),
1403					      mbio, disk_devt(mddev->gendisk),
1404					      r1_bio->sector);
1405		/* flush_pending_writes() needs access to the rdev so...*/
1406		mbio->bi_bdev = (void*)conf->mirrors[i].rdev;
1407
1408		cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1409		if (cb)
1410			plug = container_of(cb, struct raid1_plug_cb, cb);
1411		else
1412			plug = NULL;
1413		spin_lock_irqsave(&conf->device_lock, flags);
1414		if (plug) {
1415			bio_list_add(&plug->pending, mbio);
1416			plug->pending_cnt++;
1417		} else {
1418			bio_list_add(&conf->pending_bio_list, mbio);
1419			conf->pending_count++;
 
1420		}
1421		spin_unlock_irqrestore(&conf->device_lock, flags);
1422		if (!plug)
1423			md_wakeup_thread(mddev->thread);
1424	}
1425	/* Mustn't call r1_bio_write_done before this next test,
1426	 * as it could result in the bio being freed.
1427	 */
1428	if (sectors_handled < bio_sectors(bio)) {
1429		r1_bio_write_done(r1_bio);
1430		/* We need another r1_bio.  It has already been counted
1431		 * in bio->bi_phys_segments
1432		 */
1433		r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1434		r1_bio->master_bio = bio;
1435		r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1436		r1_bio->state = 0;
1437		r1_bio->mddev = mddev;
1438		r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
1439		goto retry_write;
1440	}
1441
1442	r1_bio_write_done(r1_bio);
1443
1444	/* In case raid1d snuck in to freeze_array */
1445	wake_up(&conf->wait_barrier);
 
 
 
 
 
 
 
 
 
 
 
 
1446}
1447
1448static void raid1_make_request(struct mddev *mddev, struct bio *bio)
1449{
1450	struct r1conf *conf = mddev->private;
1451	struct r1bio *r1_bio;
1452
1453	/*
1454	 * make_request() can abort the operation when read-ahead is being
1455	 * used and no empty request is available.
1456	 *
1457	 */
1458	r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1459
1460	r1_bio->master_bio = bio;
1461	r1_bio->sectors = bio_sectors(bio);
1462	r1_bio->state = 0;
1463	r1_bio->mddev = mddev;
1464	r1_bio->sector = bio->bi_iter.bi_sector;
1465
1466	/*
1467	 * We might need to issue multiple reads to different devices if there
1468	 * are bad blocks around, so we keep track of the number of reads in
1469	 * bio->bi_phys_segments.  If this is 0, there is only one r1_bio and
1470	 * no locking will be needed when requests complete.  If it is
1471	 * non-zero, then it is the number of not-completed requests.
1472	 */
1473	bio->bi_phys_segments = 0;
1474	bio_clear_flag(bio, BIO_SEG_VALID);
1475
1476	if (bio_data_dir(bio) == READ)
1477		raid1_read_request(mddev, bio, r1_bio);
1478	else
1479		raid1_write_request(mddev, bio, r1_bio);
 
 
 
1480}
1481
1482static void raid1_status(struct seq_file *seq, struct mddev *mddev)
1483{
1484	struct r1conf *conf = mddev->private;
1485	int i;
1486
 
 
1487	seq_printf(seq, " [%d/%d] [", conf->raid_disks,
1488		   conf->raid_disks - mddev->degraded);
1489	rcu_read_lock();
1490	for (i = 0; i < conf->raid_disks; i++) {
1491		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
 
1492		seq_printf(seq, "%s",
1493			   rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1494	}
1495	rcu_read_unlock();
1496	seq_printf(seq, "]");
1497}
1498
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1499static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
1500{
1501	char b[BDEVNAME_SIZE];
1502	struct r1conf *conf = mddev->private;
1503	unsigned long flags;
1504
1505	/*
1506	 * If it is not operational, then we have already marked it as dead
1507	 * else if it is the last working disks, ignore the error, let the
1508	 * next level up know.
1509	 * else mark the drive as failed
1510	 */
1511	spin_lock_irqsave(&conf->device_lock, flags);
1512	if (test_bit(In_sync, &rdev->flags)
1513	    && (conf->raid_disks - mddev->degraded) == 1) {
1514		/*
1515		 * Don't fail the drive, act as though we were just a
1516		 * normal single drive.
1517		 * However don't try a recovery from this drive as
1518		 * it is very likely to fail.
1519		 */
1520		conf->recovery_disabled = mddev->recovery_disabled;
1521		spin_unlock_irqrestore(&conf->device_lock, flags);
1522		return;
1523	}
1524	set_bit(Blocked, &rdev->flags);
1525	if (test_and_clear_bit(In_sync, &rdev->flags)) {
1526		mddev->degraded++;
1527		set_bit(Faulty, &rdev->flags);
1528	} else
1529		set_bit(Faulty, &rdev->flags);
1530	spin_unlock_irqrestore(&conf->device_lock, flags);
1531	/*
1532	 * if recovery is running, make sure it aborts.
1533	 */
1534	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1535	set_mask_bits(&mddev->sb_flags, 0,
1536		      BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
1537	pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n"
1538		"md/raid1:%s: Operation continuing on %d devices.\n",
1539		mdname(mddev), bdevname(rdev->bdev, b),
1540		mdname(mddev), conf->raid_disks - mddev->degraded);
1541}
1542
1543static void print_conf(struct r1conf *conf)
1544{
1545	int i;
1546
1547	pr_debug("RAID1 conf printout:\n");
1548	if (!conf) {
1549		pr_debug("(!conf)\n");
1550		return;
1551	}
1552	pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1553		 conf->raid_disks);
1554
1555	rcu_read_lock();
1556	for (i = 0; i < conf->raid_disks; i++) {
1557		char b[BDEVNAME_SIZE];
1558		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1559		if (rdev)
1560			pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n",
1561				 i, !test_bit(In_sync, &rdev->flags),
1562				 !test_bit(Faulty, &rdev->flags),
1563				 bdevname(rdev->bdev,b));
1564	}
1565	rcu_read_unlock();
1566}
1567
1568static void close_sync(struct r1conf *conf)
1569{
1570	wait_barrier(conf, NULL);
1571	allow_barrier(conf, 0, 0);
1572
1573	mempool_destroy(conf->r1buf_pool);
1574	conf->r1buf_pool = NULL;
 
 
1575
1576	spin_lock_irq(&conf->resync_lock);
1577	conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
1578	conf->start_next_window = MaxSector;
1579	conf->current_window_requests +=
1580		conf->next_window_requests;
1581	conf->next_window_requests = 0;
1582	spin_unlock_irq(&conf->resync_lock);
1583}
1584
1585static int raid1_spare_active(struct mddev *mddev)
1586{
1587	int i;
1588	struct r1conf *conf = mddev->private;
1589	int count = 0;
1590	unsigned long flags;
1591
1592	/*
1593	 * Find all failed disks within the RAID1 configuration
1594	 * and mark them readable.
1595	 * Called under mddev lock, so rcu protection not needed.
1596	 * device_lock used to avoid races with raid1_end_read_request
1597	 * which expects 'In_sync' flags and ->degraded to be consistent.
1598	 */
1599	spin_lock_irqsave(&conf->device_lock, flags);
1600	for (i = 0; i < conf->raid_disks; i++) {
1601		struct md_rdev *rdev = conf->mirrors[i].rdev;
1602		struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1603		if (repl
1604		    && !test_bit(Candidate, &repl->flags)
1605		    && repl->recovery_offset == MaxSector
1606		    && !test_bit(Faulty, &repl->flags)
1607		    && !test_and_set_bit(In_sync, &repl->flags)) {
1608			/* replacement has just become active */
1609			if (!rdev ||
1610			    !test_and_clear_bit(In_sync, &rdev->flags))
1611				count++;
1612			if (rdev) {
1613				/* Replaced device not technically
1614				 * faulty, but we need to be sure
1615				 * it gets removed and never re-added
1616				 */
1617				set_bit(Faulty, &rdev->flags);
1618				sysfs_notify_dirent_safe(
1619					rdev->sysfs_state);
1620			}
1621		}
1622		if (rdev
1623		    && rdev->recovery_offset == MaxSector
1624		    && !test_bit(Faulty, &rdev->flags)
1625		    && !test_and_set_bit(In_sync, &rdev->flags)) {
1626			count++;
1627			sysfs_notify_dirent_safe(rdev->sysfs_state);
1628		}
1629	}
1630	mddev->degraded -= count;
1631	spin_unlock_irqrestore(&conf->device_lock, flags);
1632
1633	print_conf(conf);
1634	return count;
1635}
1636
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1637static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1638{
1639	struct r1conf *conf = mddev->private;
1640	int err = -EEXIST;
1641	int mirror = 0;
1642	struct raid1_info *p;
1643	int first = 0;
1644	int last = conf->raid_disks - 1;
1645
1646	if (mddev->recovery_disabled == conf->recovery_disabled)
1647		return -EBUSY;
1648
1649	if (md_integrity_add_rdev(rdev, mddev))
1650		return -ENXIO;
1651
1652	if (rdev->raid_disk >= 0)
1653		first = last = rdev->raid_disk;
1654
1655	/*
1656	 * find the disk ... but prefer rdev->saved_raid_disk
1657	 * if possible.
1658	 */
1659	if (rdev->saved_raid_disk >= 0 &&
1660	    rdev->saved_raid_disk >= first &&
 
1661	    conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1662		first = last = rdev->saved_raid_disk;
1663
1664	for (mirror = first; mirror <= last; mirror++) {
1665		p = conf->mirrors+mirror;
1666		if (!p->rdev) {
 
 
 
1667
1668			if (mddev->gendisk)
1669				disk_stack_limits(mddev->gendisk, rdev->bdev,
1670						  rdev->data_offset << 9);
1671
1672			p->head_position = 0;
1673			rdev->raid_disk = mirror;
1674			err = 0;
1675			/* As all devices are equivalent, we don't need a full recovery
1676			 * if this was recently any drive of the array
1677			 */
1678			if (rdev->saved_raid_disk < 0)
1679				conf->fullsync = 1;
1680			rcu_assign_pointer(p->rdev, rdev);
1681			break;
1682		}
1683		if (test_bit(WantReplacement, &p->rdev->flags) &&
1684		    p[conf->raid_disks].rdev == NULL) {
1685			/* Add this device as a replacement */
1686			clear_bit(In_sync, &rdev->flags);
1687			set_bit(Replacement, &rdev->flags);
1688			rdev->raid_disk = mirror;
1689			err = 0;
1690			conf->fullsync = 1;
1691			rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1692			break;
1693		}
 
1694	}
1695	if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
1696		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1697	print_conf(conf);
1698	return err;
1699}
1700
1701static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1702{
1703	struct r1conf *conf = mddev->private;
1704	int err = 0;
1705	int number = rdev->raid_disk;
1706	struct raid1_info *p = conf->mirrors + number;
1707
1708	if (rdev != p->rdev)
1709		p = conf->mirrors + conf->raid_disks + number;
 
 
 
 
 
1710
1711	print_conf(conf);
1712	if (rdev == p->rdev) {
1713		if (test_bit(In_sync, &rdev->flags) ||
1714		    atomic_read(&rdev->nr_pending)) {
1715			err = -EBUSY;
1716			goto abort;
1717		}
1718		/* Only remove non-faulty devices if recovery
1719		 * is not possible.
1720		 */
1721		if (!test_bit(Faulty, &rdev->flags) &&
1722		    mddev->recovery_disabled != conf->recovery_disabled &&
1723		    mddev->degraded < conf->raid_disks) {
1724			err = -EBUSY;
1725			goto abort;
1726		}
1727		p->rdev = NULL;
1728		if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1729			synchronize_rcu();
1730			if (atomic_read(&rdev->nr_pending)) {
1731				/* lost the race, try later */
1732				err = -EBUSY;
1733				p->rdev = rdev;
1734				goto abort;
1735			}
1736		}
1737		if (conf->mirrors[conf->raid_disks + number].rdev) {
1738			/* We just removed a device that is being replaced.
1739			 * Move down the replacement.  We drain all IO before
1740			 * doing this to avoid confusion.
1741			 */
1742			struct md_rdev *repl =
1743				conf->mirrors[conf->raid_disks + number].rdev;
1744			freeze_array(conf, 0);
 
 
 
 
 
 
 
 
 
 
 
1745			clear_bit(Replacement, &repl->flags);
1746			p->rdev = repl;
1747			conf->mirrors[conf->raid_disks + number].rdev = NULL;
1748			unfreeze_array(conf);
1749			clear_bit(WantReplacement, &rdev->flags);
1750		} else
1751			clear_bit(WantReplacement, &rdev->flags);
1752		err = md_integrity_register(mddev);
1753	}
1754abort:
1755
1756	print_conf(conf);
1757	return err;
1758}
1759
1760static void end_sync_read(struct bio *bio)
1761{
1762	struct r1bio *r1_bio = bio->bi_private;
1763
1764	update_head_pos(r1_bio->read_disk, r1_bio);
1765
1766	/*
1767	 * we have read a block, now it needs to be re-written,
1768	 * or re-read if the read failed.
1769	 * We don't do much here, just schedule handling by raid1d
1770	 */
1771	if (!bio->bi_error)
1772		set_bit(R1BIO_Uptodate, &r1_bio->state);
1773
1774	if (atomic_dec_and_test(&r1_bio->remaining))
1775		reschedule_retry(r1_bio);
1776}
1777
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1778static void end_sync_write(struct bio *bio)
1779{
1780	int uptodate = !bio->bi_error;
1781	struct r1bio *r1_bio = bio->bi_private;
1782	struct mddev *mddev = r1_bio->mddev;
1783	struct r1conf *conf = mddev->private;
1784	sector_t first_bad;
1785	int bad_sectors;
1786	struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
1787
1788	if (!uptodate) {
1789		sector_t sync_blocks = 0;
1790		sector_t s = r1_bio->sector;
1791		long sectors_to_go = r1_bio->sectors;
1792		/* make sure these bits doesn't get cleared. */
1793		do {
1794			bitmap_end_sync(mddev->bitmap, s,
1795					&sync_blocks, 1);
1796			s += sync_blocks;
1797			sectors_to_go -= sync_blocks;
1798		} while (sectors_to_go > 0);
1799		set_bit(WriteErrorSeen, &rdev->flags);
1800		if (!test_and_set_bit(WantReplacement, &rdev->flags))
1801			set_bit(MD_RECOVERY_NEEDED, &
1802				mddev->recovery);
1803		set_bit(R1BIO_WriteError, &r1_bio->state);
1804	} else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
1805			       &first_bad, &bad_sectors) &&
1806		   !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1807				r1_bio->sector,
1808				r1_bio->sectors,
1809				&first_bad, &bad_sectors)
1810		)
1811		set_bit(R1BIO_MadeGood, &r1_bio->state);
 
1812
1813	if (atomic_dec_and_test(&r1_bio->remaining)) {
1814		int s = r1_bio->sectors;
1815		if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1816		    test_bit(R1BIO_WriteError, &r1_bio->state))
1817			reschedule_retry(r1_bio);
1818		else {
1819			put_buf(r1_bio);
1820			md_done_sync(mddev, s, uptodate);
1821		}
1822	}
1823}
1824
1825static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
1826			    int sectors, struct page *page, int rw)
1827{
1828	if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
1829		/* success */
1830		return 1;
1831	if (rw == WRITE) {
1832		set_bit(WriteErrorSeen, &rdev->flags);
1833		if (!test_and_set_bit(WantReplacement,
1834				      &rdev->flags))
1835			set_bit(MD_RECOVERY_NEEDED, &
1836				rdev->mddev->recovery);
1837	}
1838	/* need to record an error - either for the block or the device */
1839	if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1840		md_error(rdev->mddev, rdev);
1841	return 0;
1842}
1843
1844static int fix_sync_read_error(struct r1bio *r1_bio)
1845{
1846	/* Try some synchronous reads of other devices to get
1847	 * good data, much like with normal read errors.  Only
1848	 * read into the pages we already have so we don't
1849	 * need to re-issue the read request.
1850	 * We don't need to freeze the array, because being in an
1851	 * active sync request, there is no normal IO, and
1852	 * no overlapping syncs.
1853	 * We don't need to check is_badblock() again as we
1854	 * made sure that anything with a bad block in range
1855	 * will have bi_end_io clear.
1856	 */
1857	struct mddev *mddev = r1_bio->mddev;
1858	struct r1conf *conf = mddev->private;
1859	struct bio *bio = r1_bio->bios[r1_bio->read_disk];
 
1860	sector_t sect = r1_bio->sector;
1861	int sectors = r1_bio->sectors;
1862	int idx = 0;
1863	struct md_rdev *rdev;
1864
1865	rdev = conf->mirrors[r1_bio->read_disk].rdev;
1866	if (test_bit(FailFast, &rdev->flags)) {
1867		/* Don't try recovering from here - just fail it
1868		 * ... unless it is the last working device of course */
1869		md_error(mddev, rdev);
1870		if (test_bit(Faulty, &rdev->flags))
1871			/* Don't try to read from here, but make sure
1872			 * put_buf does it's thing
1873			 */
1874			bio->bi_end_io = end_sync_write;
1875	}
1876
1877	while(sectors) {
1878		int s = sectors;
1879		int d = r1_bio->read_disk;
1880		int success = 0;
1881		int start;
1882
1883		if (s > (PAGE_SIZE>>9))
1884			s = PAGE_SIZE >> 9;
1885		do {
1886			if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1887				/* No rcu protection needed here devices
1888				 * can only be removed when no resync is
1889				 * active, and resync is currently active
1890				 */
1891				rdev = conf->mirrors[d].rdev;
1892				if (sync_page_io(rdev, sect, s<<9,
1893						 bio->bi_io_vec[idx].bv_page,
1894						 REQ_OP_READ, 0, false)) {
1895					success = 1;
1896					break;
1897				}
1898			}
1899			d++;
1900			if (d == conf->raid_disks * 2)
1901				d = 0;
1902		} while (!success && d != r1_bio->read_disk);
1903
1904		if (!success) {
1905			char b[BDEVNAME_SIZE];
1906			int abort = 0;
1907			/* Cannot read from anywhere, this block is lost.
1908			 * Record a bad block on each device.  If that doesn't
1909			 * work just disable and interrupt the recovery.
1910			 * Don't fail devices as that won't really help.
1911			 */
1912			pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
1913					    mdname(mddev),
1914					    bdevname(bio->bi_bdev, b),
1915					    (unsigned long long)r1_bio->sector);
1916			for (d = 0; d < conf->raid_disks * 2; d++) {
1917				rdev = conf->mirrors[d].rdev;
1918				if (!rdev || test_bit(Faulty, &rdev->flags))
1919					continue;
1920				if (!rdev_set_badblocks(rdev, sect, s, 0))
1921					abort = 1;
1922			}
1923			if (abort) {
1924				conf->recovery_disabled =
1925					mddev->recovery_disabled;
1926				set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1927				md_done_sync(mddev, r1_bio->sectors, 0);
1928				put_buf(r1_bio);
1929				return 0;
1930			}
1931			/* Try next page */
1932			sectors -= s;
1933			sect += s;
1934			idx++;
1935			continue;
1936		}
1937
1938		start = d;
1939		/* write it back and re-read */
1940		while (d != r1_bio->read_disk) {
1941			if (d == 0)
1942				d = conf->raid_disks * 2;
1943			d--;
1944			if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1945				continue;
1946			rdev = conf->mirrors[d].rdev;
1947			if (r1_sync_page_io(rdev, sect, s,
1948					    bio->bi_io_vec[idx].bv_page,
1949					    WRITE) == 0) {
1950				r1_bio->bios[d]->bi_end_io = NULL;
1951				rdev_dec_pending(rdev, mddev);
1952			}
1953		}
1954		d = start;
1955		while (d != r1_bio->read_disk) {
1956			if (d == 0)
1957				d = conf->raid_disks * 2;
1958			d--;
1959			if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1960				continue;
1961			rdev = conf->mirrors[d].rdev;
1962			if (r1_sync_page_io(rdev, sect, s,
1963					    bio->bi_io_vec[idx].bv_page,
1964					    READ) != 0)
1965				atomic_add(s, &rdev->corrected_errors);
1966		}
1967		sectors -= s;
1968		sect += s;
1969		idx ++;
1970	}
1971	set_bit(R1BIO_Uptodate, &r1_bio->state);
1972	bio->bi_error = 0;
1973	return 1;
1974}
1975
1976static void process_checks(struct r1bio *r1_bio)
1977{
1978	/* We have read all readable devices.  If we haven't
1979	 * got the block, then there is no hope left.
1980	 * If we have, then we want to do a comparison
1981	 * and skip the write if everything is the same.
1982	 * If any blocks failed to read, then we need to
1983	 * attempt an over-write
1984	 */
1985	struct mddev *mddev = r1_bio->mddev;
1986	struct r1conf *conf = mddev->private;
1987	int primary;
1988	int i;
1989	int vcnt;
1990
1991	/* Fix variable parts of all bios */
1992	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1993	for (i = 0; i < conf->raid_disks * 2; i++) {
1994		int j;
1995		int size;
1996		int error;
1997		struct bio *b = r1_bio->bios[i];
 
1998		if (b->bi_end_io != end_sync_read)
1999			continue;
2000		/* fixup the bio for reuse, but preserve errno */
2001		error = b->bi_error;
2002		bio_reset(b);
2003		b->bi_error = error;
2004		b->bi_vcnt = vcnt;
2005		b->bi_iter.bi_size = r1_bio->sectors << 9;
2006		b->bi_iter.bi_sector = r1_bio->sector +
2007			conf->mirrors[i].rdev->data_offset;
2008		b->bi_bdev = conf->mirrors[i].rdev->bdev;
2009		b->bi_end_io = end_sync_read;
2010		b->bi_private = r1_bio;
 
2011
2012		size = b->bi_iter.bi_size;
2013		for (j = 0; j < vcnt ; j++) {
2014			struct bio_vec *bi;
2015			bi = &b->bi_io_vec[j];
2016			bi->bv_offset = 0;
2017			if (size > PAGE_SIZE)
2018				bi->bv_len = PAGE_SIZE;
2019			else
2020				bi->bv_len = size;
2021			size -= PAGE_SIZE;
2022		}
2023	}
2024	for (primary = 0; primary < conf->raid_disks * 2; primary++)
2025		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
2026		    !r1_bio->bios[primary]->bi_error) {
2027			r1_bio->bios[primary]->bi_end_io = NULL;
2028			rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
2029			break;
2030		}
2031	r1_bio->read_disk = primary;
2032	for (i = 0; i < conf->raid_disks * 2; i++) {
2033		int j;
2034		struct bio *pbio = r1_bio->bios[primary];
2035		struct bio *sbio = r1_bio->bios[i];
2036		int error = sbio->bi_error;
 
 
 
 
 
2037
2038		if (sbio->bi_end_io != end_sync_read)
2039			continue;
2040		/* Now we can 'fixup' the error value */
2041		sbio->bi_error = 0;
 
 
 
2042
2043		if (!error) {
2044			for (j = vcnt; j-- ; ) {
2045				struct page *p, *s;
2046				p = pbio->bi_io_vec[j].bv_page;
2047				s = sbio->bi_io_vec[j].bv_page;
2048				if (memcmp(page_address(p),
2049					   page_address(s),
2050					   sbio->bi_io_vec[j].bv_len))
2051					break;
2052			}
2053		} else
2054			j = 0;
2055		if (j >= 0)
2056			atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
2057		if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
2058			      && !error)) {
2059			/* No need to write to this device. */
2060			sbio->bi_end_io = NULL;
2061			rdev_dec_pending(conf->mirrors[i].rdev, mddev);
2062			continue;
2063		}
2064
2065		bio_copy_data(sbio, pbio);
2066	}
2067}
2068
2069static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
2070{
2071	struct r1conf *conf = mddev->private;
2072	int i;
2073	int disks = conf->raid_disks * 2;
2074	struct bio *bio, *wbio;
2075
2076	bio = r1_bio->bios[r1_bio->read_disk];
2077
2078	if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2079		/* ouch - failed to read all of that. */
2080		if (!fix_sync_read_error(r1_bio))
2081			return;
2082
2083	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2084		process_checks(r1_bio);
2085
2086	/*
2087	 * schedule writes
2088	 */
2089	atomic_set(&r1_bio->remaining, 1);
2090	for (i = 0; i < disks ; i++) {
2091		wbio = r1_bio->bios[i];
2092		if (wbio->bi_end_io == NULL ||
2093		    (wbio->bi_end_io == end_sync_read &&
2094		     (i == r1_bio->read_disk ||
2095		      !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
2096			continue;
 
 
 
 
2097
2098		bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2099		if (test_bit(FailFast, &conf->mirrors[i].rdev->flags))
2100			wbio->bi_opf |= MD_FAILFAST;
2101
2102		wbio->bi_end_io = end_sync_write;
2103		atomic_inc(&r1_bio->remaining);
2104		md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
2105
2106		generic_make_request(wbio);
2107	}
2108
2109	if (atomic_dec_and_test(&r1_bio->remaining)) {
2110		/* if we're here, all write(s) have completed, so clean up */
2111		int s = r1_bio->sectors;
2112		if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2113		    test_bit(R1BIO_WriteError, &r1_bio->state))
2114			reschedule_retry(r1_bio);
2115		else {
2116			put_buf(r1_bio);
2117			md_done_sync(mddev, s, 1);
2118		}
2119	}
2120}
2121
2122/*
2123 * This is a kernel thread which:
2124 *
2125 *	1.	Retries failed read operations on working mirrors.
2126 *	2.	Updates the raid superblock when problems encounter.
2127 *	3.	Performs writes following reads for array synchronising.
2128 */
2129
2130static void fix_read_error(struct r1conf *conf, int read_disk,
2131			   sector_t sect, int sectors)
2132{
 
 
 
2133	struct mddev *mddev = conf->mddev;
 
 
 
 
 
 
 
2134	while(sectors) {
2135		int s = sectors;
2136		int d = read_disk;
2137		int success = 0;
2138		int start;
2139		struct md_rdev *rdev;
2140
2141		if (s > (PAGE_SIZE>>9))
2142			s = PAGE_SIZE >> 9;
2143
2144		do {
2145			sector_t first_bad;
2146			int bad_sectors;
2147
2148			rcu_read_lock();
2149			rdev = rcu_dereference(conf->mirrors[d].rdev);
2150			if (rdev &&
2151			    (test_bit(In_sync, &rdev->flags) ||
2152			     (!test_bit(Faulty, &rdev->flags) &&
2153			      rdev->recovery_offset >= sect + s)) &&
2154			    is_badblock(rdev, sect, s,
2155					&first_bad, &bad_sectors) == 0) {
2156				atomic_inc(&rdev->nr_pending);
2157				rcu_read_unlock();
2158				if (sync_page_io(rdev, sect, s<<9,
2159					 conf->tmppage, REQ_OP_READ, 0, false))
2160					success = 1;
2161				rdev_dec_pending(rdev, mddev);
2162				if (success)
2163					break;
2164			} else
2165				rcu_read_unlock();
2166			d++;
2167			if (d == conf->raid_disks * 2)
2168				d = 0;
2169		} while (!success && d != read_disk);
2170
2171		if (!success) {
2172			/* Cannot read from anywhere - mark it bad */
2173			struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2174			if (!rdev_set_badblocks(rdev, sect, s, 0))
2175				md_error(mddev, rdev);
2176			break;
2177		}
2178		/* write it back and re-read */
2179		start = d;
2180		while (d != read_disk) {
2181			if (d==0)
2182				d = conf->raid_disks * 2;
2183			d--;
2184			rcu_read_lock();
2185			rdev = rcu_dereference(conf->mirrors[d].rdev);
2186			if (rdev &&
2187			    !test_bit(Faulty, &rdev->flags)) {
2188				atomic_inc(&rdev->nr_pending);
2189				rcu_read_unlock();
2190				r1_sync_page_io(rdev, sect, s,
2191						conf->tmppage, WRITE);
2192				rdev_dec_pending(rdev, mddev);
2193			} else
2194				rcu_read_unlock();
2195		}
2196		d = start;
2197		while (d != read_disk) {
2198			char b[BDEVNAME_SIZE];
2199			if (d==0)
2200				d = conf->raid_disks * 2;
2201			d--;
2202			rcu_read_lock();
2203			rdev = rcu_dereference(conf->mirrors[d].rdev);
2204			if (rdev &&
2205			    !test_bit(Faulty, &rdev->flags)) {
2206				atomic_inc(&rdev->nr_pending);
2207				rcu_read_unlock();
2208				if (r1_sync_page_io(rdev, sect, s,
2209						    conf->tmppage, READ)) {
2210					atomic_add(s, &rdev->corrected_errors);
2211					pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n",
2212						mdname(mddev), s,
2213						(unsigned long long)(sect +
2214								     rdev->data_offset),
2215						bdevname(rdev->bdev, b));
2216				}
2217				rdev_dec_pending(rdev, mddev);
2218			} else
2219				rcu_read_unlock();
2220		}
2221		sectors -= s;
2222		sect += s;
2223	}
2224}
2225
2226static int narrow_write_error(struct r1bio *r1_bio, int i)
2227{
2228	struct mddev *mddev = r1_bio->mddev;
2229	struct r1conf *conf = mddev->private;
2230	struct md_rdev *rdev = conf->mirrors[i].rdev;
2231
2232	/* bio has the data to be written to device 'i' where
2233	 * we just recently had a write error.
2234	 * We repeatedly clone the bio and trim down to one block,
2235	 * then try the write.  Where the write fails we record
2236	 * a bad block.
2237	 * It is conceivable that the bio doesn't exactly align with
2238	 * blocks.  We must handle this somehow.
2239	 *
2240	 * We currently own a reference on the rdev.
2241	 */
2242
2243	int block_sectors;
2244	sector_t sector;
2245	int sectors;
2246	int sect_to_write = r1_bio->sectors;
2247	int ok = 1;
2248
2249	if (rdev->badblocks.shift < 0)
2250		return 0;
2251
2252	block_sectors = roundup(1 << rdev->badblocks.shift,
2253				bdev_logical_block_size(rdev->bdev) >> 9);
2254	sector = r1_bio->sector;
2255	sectors = ((sector + block_sectors)
2256		   & ~(sector_t)(block_sectors - 1))
2257		- sector;
2258
2259	while (sect_to_write) {
2260		struct bio *wbio;
2261		if (sectors > sect_to_write)
2262			sectors = sect_to_write;
2263		/* Write at 'sector' for 'sectors'*/
2264
2265		if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2266			unsigned vcnt = r1_bio->behind_page_count;
2267			struct bio_vec *vec = r1_bio->behind_bvecs;
2268
2269			while (!vec->bv_page) {
2270				vec++;
2271				vcnt--;
2272			}
2273
2274			wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2275			memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2276
2277			wbio->bi_vcnt = vcnt;
2278		} else {
2279			wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
 
2280		}
2281
2282		bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2283		wbio->bi_iter.bi_sector = r1_bio->sector;
2284		wbio->bi_iter.bi_size = r1_bio->sectors << 9;
2285
2286		bio_trim(wbio, sector - r1_bio->sector, sectors);
2287		wbio->bi_iter.bi_sector += rdev->data_offset;
2288		wbio->bi_bdev = rdev->bdev;
2289
2290		if (submit_bio_wait(wbio) < 0)
2291			/* failure! */
2292			ok = rdev_set_badblocks(rdev, sector,
2293						sectors, 0)
2294				&& ok;
2295
2296		bio_put(wbio);
2297		sect_to_write -= sectors;
2298		sector += sectors;
2299		sectors = block_sectors;
2300	}
2301	return ok;
2302}
2303
2304static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2305{
2306	int m;
2307	int s = r1_bio->sectors;
2308	for (m = 0; m < conf->raid_disks * 2 ; m++) {
2309		struct md_rdev *rdev = conf->mirrors[m].rdev;
2310		struct bio *bio = r1_bio->bios[m];
2311		if (bio->bi_end_io == NULL)
2312			continue;
2313		if (!bio->bi_error &&
2314		    test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2315			rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
2316		}
2317		if (bio->bi_error &&
2318		    test_bit(R1BIO_WriteError, &r1_bio->state)) {
2319			if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2320				md_error(conf->mddev, rdev);
2321		}
2322	}
2323	put_buf(r1_bio);
2324	md_done_sync(conf->mddev, s, 1);
2325}
2326
2327static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2328{
2329	int m;
2330	bool fail = false;
 
2331	for (m = 0; m < conf->raid_disks * 2 ; m++)
2332		if (r1_bio->bios[m] == IO_MADE_GOOD) {
2333			struct md_rdev *rdev = conf->mirrors[m].rdev;
2334			rdev_clear_badblocks(rdev,
2335					     r1_bio->sector,
2336					     r1_bio->sectors, 0);
2337			rdev_dec_pending(rdev, conf->mddev);
2338		} else if (r1_bio->bios[m] != NULL) {
2339			/* This drive got a write error.  We need to
2340			 * narrow down and record precise write
2341			 * errors.
2342			 */
2343			fail = true;
2344			if (!narrow_write_error(r1_bio, m)) {
2345				md_error(conf->mddev,
2346					 conf->mirrors[m].rdev);
2347				/* an I/O failed, we can't clear the bitmap */
2348				set_bit(R1BIO_Degraded, &r1_bio->state);
2349			}
2350			rdev_dec_pending(conf->mirrors[m].rdev,
2351					 conf->mddev);
2352		}
2353	if (fail) {
2354		spin_lock_irq(&conf->device_lock);
2355		list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2356		conf->nr_queued++;
 
2357		spin_unlock_irq(&conf->device_lock);
 
 
 
 
 
2358		md_wakeup_thread(conf->mddev->thread);
2359	} else {
2360		if (test_bit(R1BIO_WriteError, &r1_bio->state))
2361			close_write(r1_bio);
2362		raid_end_bio_io(r1_bio);
2363	}
2364}
2365
2366static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
2367{
2368	int disk;
2369	int max_sectors;
2370	struct mddev *mddev = conf->mddev;
2371	struct bio *bio;
2372	char b[BDEVNAME_SIZE];
2373	struct md_rdev *rdev;
2374	dev_t bio_dev;
2375	sector_t bio_sector;
2376
2377	clear_bit(R1BIO_ReadError, &r1_bio->state);
2378	/* we got a read error. Maybe the drive is bad.  Maybe just
2379	 * the block and we can fix it.
2380	 * We freeze all other IO, and try reading the block from
2381	 * other devices.  When we find one, we re-write
2382	 * and check it that fixes the read error.
2383	 * This is all done synchronously while the array is
2384	 * frozen
2385	 */
2386
2387	bio = r1_bio->bios[r1_bio->read_disk];
2388	bdevname(bio->bi_bdev, b);
2389	bio_dev = bio->bi_bdev->bd_dev;
2390	bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector;
2391	bio_put(bio);
2392	r1_bio->bios[r1_bio->read_disk] = NULL;
2393
2394	rdev = conf->mirrors[r1_bio->read_disk].rdev;
2395	if (mddev->ro == 0
2396	    && !test_bit(FailFast, &rdev->flags)) {
2397		freeze_array(conf, 1);
2398		fix_read_error(conf, r1_bio->read_disk,
2399			       r1_bio->sector, r1_bio->sectors);
2400		unfreeze_array(conf);
 
 
2401	} else {
2402		r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2403	}
2404
2405	rdev_dec_pending(rdev, conf->mddev);
 
 
2406
2407read_more:
2408	disk = read_balance(conf, r1_bio, &max_sectors);
2409	if (disk == -1) {
2410		pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
2411				    mdname(mddev), b, (unsigned long long)r1_bio->sector);
2412		raid_end_bio_io(r1_bio);
2413	} else {
2414		const unsigned long do_sync
2415			= r1_bio->master_bio->bi_opf & REQ_SYNC;
2416		r1_bio->read_disk = disk;
2417		bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2418		bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2419			 max_sectors);
2420		r1_bio->bios[r1_bio->read_disk] = bio;
2421		rdev = conf->mirrors[disk].rdev;
2422		pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n",
2423				    mdname(mddev),
2424				    (unsigned long long)r1_bio->sector,
2425				    bdevname(rdev->bdev, b));
2426		bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
2427		bio->bi_bdev = rdev->bdev;
2428		bio->bi_end_io = raid1_end_read_request;
2429		bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
2430		if (test_bit(FailFast, &rdev->flags) &&
2431		    test_bit(R1BIO_FailFast, &r1_bio->state))
2432			bio->bi_opf |= MD_FAILFAST;
2433		bio->bi_private = r1_bio;
2434		if (max_sectors < r1_bio->sectors) {
2435			/* Drat - have to split this up more */
2436			struct bio *mbio = r1_bio->master_bio;
2437			int sectors_handled = (r1_bio->sector + max_sectors
2438					       - mbio->bi_iter.bi_sector);
2439			r1_bio->sectors = max_sectors;
2440			spin_lock_irq(&conf->device_lock);
2441			if (mbio->bi_phys_segments == 0)
2442				mbio->bi_phys_segments = 2;
2443			else
2444				mbio->bi_phys_segments++;
2445			spin_unlock_irq(&conf->device_lock);
2446			trace_block_bio_remap(bdev_get_queue(bio->bi_bdev),
2447					      bio, bio_dev, bio_sector);
2448			generic_make_request(bio);
2449			bio = NULL;
2450
2451			r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2452
2453			r1_bio->master_bio = mbio;
2454			r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
2455			r1_bio->state = 0;
2456			set_bit(R1BIO_ReadError, &r1_bio->state);
2457			r1_bio->mddev = mddev;
2458			r1_bio->sector = mbio->bi_iter.bi_sector +
2459				sectors_handled;
2460
2461			goto read_more;
2462		} else {
2463			trace_block_bio_remap(bdev_get_queue(bio->bi_bdev),
2464					      bio, bio_dev, bio_sector);
2465			generic_make_request(bio);
2466		}
2467	}
2468}
2469
2470static void raid1d(struct md_thread *thread)
2471{
2472	struct mddev *mddev = thread->mddev;
2473	struct r1bio *r1_bio;
2474	unsigned long flags;
2475	struct r1conf *conf = mddev->private;
2476	struct list_head *head = &conf->retry_list;
2477	struct blk_plug plug;
 
2478
2479	md_check_recovery(mddev);
2480
2481	if (!list_empty_careful(&conf->bio_end_io_list) &&
2482	    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
2483		LIST_HEAD(tmp);
2484		spin_lock_irqsave(&conf->device_lock, flags);
2485		if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
2486			while (!list_empty(&conf->bio_end_io_list)) {
2487				list_move(conf->bio_end_io_list.prev, &tmp);
2488				conf->nr_queued--;
2489			}
2490		}
2491		spin_unlock_irqrestore(&conf->device_lock, flags);
2492		while (!list_empty(&tmp)) {
2493			r1_bio = list_first_entry(&tmp, struct r1bio,
2494						  retry_list);
2495			list_del(&r1_bio->retry_list);
2496			if (mddev->degraded)
2497				set_bit(R1BIO_Degraded, &r1_bio->state);
2498			if (test_bit(R1BIO_WriteError, &r1_bio->state))
2499				close_write(r1_bio);
2500			raid_end_bio_io(r1_bio);
2501		}
2502	}
2503
2504	blk_start_plug(&plug);
2505	for (;;) {
2506
2507		flush_pending_writes(conf);
2508
2509		spin_lock_irqsave(&conf->device_lock, flags);
2510		if (list_empty(head)) {
2511			spin_unlock_irqrestore(&conf->device_lock, flags);
2512			break;
2513		}
2514		r1_bio = list_entry(head->prev, struct r1bio, retry_list);
2515		list_del(head->prev);
2516		conf->nr_queued--;
 
2517		spin_unlock_irqrestore(&conf->device_lock, flags);
2518
2519		mddev = r1_bio->mddev;
2520		conf = mddev->private;
2521		if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
2522			if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2523			    test_bit(R1BIO_WriteError, &r1_bio->state))
2524				handle_sync_write_finished(conf, r1_bio);
2525			else
2526				sync_request_write(mddev, r1_bio);
2527		} else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2528			   test_bit(R1BIO_WriteError, &r1_bio->state))
2529			handle_write_finished(conf, r1_bio);
2530		else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2531			handle_read_error(conf, r1_bio);
2532		else
2533			/* just a partial read to be scheduled from separate
2534			 * context
2535			 */
2536			generic_make_request(r1_bio->bios[r1_bio->read_disk]);
2537
2538		cond_resched();
2539		if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING))
2540			md_check_recovery(mddev);
2541	}
2542	blk_finish_plug(&plug);
2543}
2544
2545static int init_resync(struct r1conf *conf)
2546{
2547	int buffs;
2548
2549	buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
2550	BUG_ON(conf->r1buf_pool);
2551	conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2552					  conf->poolinfo);
2553	if (!conf->r1buf_pool)
2554		return -ENOMEM;
2555	conf->next_resync = 0;
2556	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2557}
2558
2559/*
2560 * perform a "sync" on one "block"
2561 *
2562 * We need to make sure that no normal I/O request - particularly write
2563 * requests - conflict with active sync requests.
2564 *
2565 * This is achieved by tracking pending requests and a 'barrier' concept
2566 * that can be installed to exclude normal IO requests.
2567 */
2568
2569static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2570				   int *skipped)
2571{
2572	struct r1conf *conf = mddev->private;
2573	struct r1bio *r1_bio;
2574	struct bio *bio;
2575	sector_t max_sector, nr_sectors;
2576	int disk = -1;
2577	int i;
2578	int wonly = -1;
2579	int write_targets = 0, read_targets = 0;
2580	sector_t sync_blocks;
2581	int still_degraded = 0;
2582	int good_sectors = RESYNC_SECTORS;
2583	int min_bad = 0; /* number of sectors that are bad in all devices */
 
 
2584
2585	if (!conf->r1buf_pool)
2586		if (init_resync(conf))
2587			return 0;
2588
2589	max_sector = mddev->dev_sectors;
2590	if (sector_nr >= max_sector) {
2591		/* If we aborted, we need to abort the
2592		 * sync on the 'current' bitmap chunk (there will
2593		 * only be one in raid1 resync.
2594		 * We can find the current addess in mddev->curr_resync
2595		 */
2596		if (mddev->curr_resync < max_sector) /* aborted */
2597			bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2598						&sync_blocks, 1);
2599		else /* completed sync */
2600			conf->fullsync = 0;
2601
2602		bitmap_close_sync(mddev->bitmap);
2603		close_sync(conf);
2604
2605		if (mddev_is_clustered(mddev)) {
2606			conf->cluster_sync_low = 0;
2607			conf->cluster_sync_high = 0;
2608		}
2609		return 0;
2610	}
2611
2612	if (mddev->bitmap == NULL &&
2613	    mddev->recovery_cp == MaxSector &&
2614	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
2615	    conf->fullsync == 0) {
2616		*skipped = 1;
2617		return max_sector - sector_nr;
2618	}
2619	/* before building a request, check if we can skip these blocks..
2620	 * This call the bitmap_start_sync doesn't actually record anything
2621	 */
2622	if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
2623	    !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2624		/* We can skip this block, and probably several more */
2625		*skipped = 1;
2626		return sync_blocks;
2627	}
2628
2629	/*
2630	 * If there is non-resync activity waiting for a turn, then let it
2631	 * though before starting on this new sync request.
2632	 */
2633	if (conf->nr_waiting)
2634		schedule_timeout_uninterruptible(1);
2635
2636	/* we are incrementing sector_nr below. To be safe, we check against
2637	 * sector_nr + two times RESYNC_SECTORS
2638	 */
2639
2640	bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2641		mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
2642	r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
 
 
 
2643
2644	raise_barrier(conf, sector_nr);
2645
2646	rcu_read_lock();
2647	/*
2648	 * If we get a correctably read error during resync or recovery,
2649	 * we might want to read from a different device.  So we
2650	 * flag all drives that could conceivably be read from for READ,
2651	 * and any others (which will be non-In_sync devices) for WRITE.
2652	 * If a read fails, we try reading from something else for which READ
2653	 * is OK.
2654	 */
2655
2656	r1_bio->mddev = mddev;
2657	r1_bio->sector = sector_nr;
2658	r1_bio->state = 0;
2659	set_bit(R1BIO_IsSync, &r1_bio->state);
 
 
2660
2661	for (i = 0; i < conf->raid_disks * 2; i++) {
2662		struct md_rdev *rdev;
2663		bio = r1_bio->bios[i];
2664		bio_reset(bio);
2665
2666		rdev = rcu_dereference(conf->mirrors[i].rdev);
2667		if (rdev == NULL ||
2668		    test_bit(Faulty, &rdev->flags)) {
2669			if (i < conf->raid_disks)
2670				still_degraded = 1;
2671		} else if (!test_bit(In_sync, &rdev->flags)) {
2672			bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2673			bio->bi_end_io = end_sync_write;
2674			write_targets ++;
2675		} else {
2676			/* may need to read from here */
2677			sector_t first_bad = MaxSector;
2678			int bad_sectors;
2679
2680			if (is_badblock(rdev, sector_nr, good_sectors,
2681					&first_bad, &bad_sectors)) {
2682				if (first_bad > sector_nr)
2683					good_sectors = first_bad - sector_nr;
2684				else {
2685					bad_sectors -= (sector_nr - first_bad);
2686					if (min_bad == 0 ||
2687					    min_bad > bad_sectors)
2688						min_bad = bad_sectors;
2689				}
2690			}
2691			if (sector_nr < first_bad) {
2692				if (test_bit(WriteMostly, &rdev->flags)) {
2693					if (wonly < 0)
2694						wonly = i;
2695				} else {
2696					if (disk < 0)
2697						disk = i;
2698				}
2699				bio_set_op_attrs(bio, REQ_OP_READ, 0);
2700				bio->bi_end_io = end_sync_read;
2701				read_targets++;
2702			} else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2703				test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2704				!test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2705				/*
2706				 * The device is suitable for reading (InSync),
2707				 * but has bad block(s) here. Let's try to correct them,
2708				 * if we are doing resync or repair. Otherwise, leave
2709				 * this device alone for this sync request.
2710				 */
2711				bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2712				bio->bi_end_io = end_sync_write;
2713				write_targets++;
2714			}
2715		}
2716		if (bio->bi_end_io) {
2717			atomic_inc(&rdev->nr_pending);
2718			bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
2719			bio->bi_bdev = rdev->bdev;
2720			bio->bi_private = r1_bio;
2721			if (test_bit(FailFast, &rdev->flags))
2722				bio->bi_opf |= MD_FAILFAST;
2723		}
2724	}
2725	rcu_read_unlock();
2726	if (disk < 0)
2727		disk = wonly;
2728	r1_bio->read_disk = disk;
2729
2730	if (read_targets == 0 && min_bad > 0) {
2731		/* These sectors are bad on all InSync devices, so we
2732		 * need to mark them bad on all write targets
2733		 */
2734		int ok = 1;
2735		for (i = 0 ; i < conf->raid_disks * 2 ; i++)
2736			if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
2737				struct md_rdev *rdev = conf->mirrors[i].rdev;
2738				ok = rdev_set_badblocks(rdev, sector_nr,
2739							min_bad, 0
2740					) && ok;
2741			}
2742		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2743		*skipped = 1;
2744		put_buf(r1_bio);
2745
2746		if (!ok) {
2747			/* Cannot record the badblocks, so need to
2748			 * abort the resync.
2749			 * If there are multiple read targets, could just
2750			 * fail the really bad ones ???
2751			 */
2752			conf->recovery_disabled = mddev->recovery_disabled;
2753			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2754			return 0;
2755		} else
2756			return min_bad;
2757
2758	}
2759	if (min_bad > 0 && min_bad < good_sectors) {
2760		/* only resync enough to reach the next bad->good
2761		 * transition */
2762		good_sectors = min_bad;
2763	}
2764
2765	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2766		/* extra read targets are also write targets */
2767		write_targets += read_targets-1;
2768
2769	if (write_targets == 0 || read_targets == 0) {
2770		/* There is nowhere to write, so all non-sync
2771		 * drives must be failed - so we are finished
2772		 */
2773		sector_t rv;
2774		if (min_bad > 0)
2775			max_sector = sector_nr + min_bad;
2776		rv = max_sector - sector_nr;
2777		*skipped = 1;
2778		put_buf(r1_bio);
2779		return rv;
2780	}
2781
2782	if (max_sector > mddev->resync_max)
2783		max_sector = mddev->resync_max; /* Don't do IO beyond here */
2784	if (max_sector > sector_nr + good_sectors)
2785		max_sector = sector_nr + good_sectors;
2786	nr_sectors = 0;
2787	sync_blocks = 0;
2788	do {
2789		struct page *page;
2790		int len = PAGE_SIZE;
2791		if (sector_nr + (len>>9) > max_sector)
2792			len = (max_sector - sector_nr) << 9;
2793		if (len == 0)
2794			break;
2795		if (sync_blocks == 0) {
2796			if (!bitmap_start_sync(mddev->bitmap, sector_nr,
2797					       &sync_blocks, still_degraded) &&
2798			    !conf->fullsync &&
2799			    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2800				break;
2801			if ((len >> 9) > sync_blocks)
2802				len = sync_blocks<<9;
2803		}
2804
2805		for (i = 0 ; i < conf->raid_disks * 2; i++) {
 
 
2806			bio = r1_bio->bios[i];
 
2807			if (bio->bi_end_io) {
2808				page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
2809				if (bio_add_page(bio, page, len, 0) == 0) {
2810					/* stop here */
2811					bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
2812					while (i > 0) {
2813						i--;
2814						bio = r1_bio->bios[i];
2815						if (bio->bi_end_io==NULL)
2816							continue;
2817						/* remove last page from this bio */
2818						bio->bi_vcnt--;
2819						bio->bi_iter.bi_size -= len;
2820						bio_clear_flag(bio, BIO_SEG_VALID);
2821					}
2822					goto bio_full;
2823				}
2824			}
2825		}
2826		nr_sectors += len>>9;
2827		sector_nr += len>>9;
2828		sync_blocks -= (len>>9);
2829	} while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2830 bio_full:
2831	r1_bio->sectors = nr_sectors;
2832
2833	if (mddev_is_clustered(mddev) &&
2834			conf->cluster_sync_high < sector_nr + nr_sectors) {
2835		conf->cluster_sync_low = mddev->curr_resync_completed;
2836		conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2837		/* Send resync message */
2838		md_cluster_ops->resync_info_update(mddev,
2839				conf->cluster_sync_low,
2840				conf->cluster_sync_high);
2841	}
2842
2843	/* For a user-requested sync, we read all readable devices and do a
2844	 * compare
2845	 */
2846	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2847		atomic_set(&r1_bio->remaining, read_targets);
2848		for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
2849			bio = r1_bio->bios[i];
2850			if (bio->bi_end_io == end_sync_read) {
2851				read_targets--;
2852				md_sync_acct(bio->bi_bdev, nr_sectors);
2853				if (read_targets == 1)
2854					bio->bi_opf &= ~MD_FAILFAST;
2855				generic_make_request(bio);
2856			}
2857		}
2858	} else {
2859		atomic_set(&r1_bio->remaining, 1);
2860		bio = r1_bio->bios[r1_bio->read_disk];
2861		md_sync_acct(bio->bi_bdev, nr_sectors);
2862		if (read_targets == 1)
2863			bio->bi_opf &= ~MD_FAILFAST;
2864		generic_make_request(bio);
2865
2866	}
2867	return nr_sectors;
2868}
2869
2870static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
2871{
2872	if (sectors)
2873		return sectors;
2874
2875	return mddev->dev_sectors;
2876}
2877
2878static struct r1conf *setup_conf(struct mddev *mddev)
2879{
2880	struct r1conf *conf;
2881	int i;
2882	struct raid1_info *disk;
2883	struct md_rdev *rdev;
2884	int err = -ENOMEM;
2885
2886	conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
2887	if (!conf)
2888		goto abort;
2889
2890	conf->mirrors = kzalloc(sizeof(struct raid1_info)
2891				* mddev->raid_disks * 2,
2892				 GFP_KERNEL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2893	if (!conf->mirrors)
2894		goto abort;
2895
2896	conf->tmppage = alloc_page(GFP_KERNEL);
2897	if (!conf->tmppage)
2898		goto abort;
2899
2900	conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
2901	if (!conf->poolinfo)
2902		goto abort;
2903	conf->poolinfo->raid_disks = mddev->raid_disks * 2;
2904	conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2905					  r1bio_pool_free,
2906					  conf->poolinfo);
2907	if (!conf->r1bio_pool)
 
 
 
2908		goto abort;
2909
2910	conf->poolinfo->mddev = mddev;
2911
2912	err = -EINVAL;
2913	spin_lock_init(&conf->device_lock);
 
2914	rdev_for_each(rdev, mddev) {
2915		struct request_queue *q;
2916		int disk_idx = rdev->raid_disk;
2917		if (disk_idx >= mddev->raid_disks
2918		    || disk_idx < 0)
2919			continue;
2920		if (test_bit(Replacement, &rdev->flags))
2921			disk = conf->mirrors + mddev->raid_disks + disk_idx;
2922		else
2923			disk = conf->mirrors + disk_idx;
2924
2925		if (disk->rdev)
 
2926			goto abort;
2927		disk->rdev = rdev;
2928		q = bdev_get_queue(rdev->bdev);
2929
2930		disk->head_position = 0;
2931		disk->seq_start = MaxSector;
2932	}
2933	conf->raid_disks = mddev->raid_disks;
2934	conf->mddev = mddev;
2935	INIT_LIST_HEAD(&conf->retry_list);
2936	INIT_LIST_HEAD(&conf->bio_end_io_list);
2937
2938	spin_lock_init(&conf->resync_lock);
2939	init_waitqueue_head(&conf->wait_barrier);
2940
2941	bio_list_init(&conf->pending_bio_list);
2942	conf->pending_count = 0;
2943	conf->recovery_disabled = mddev->recovery_disabled - 1;
2944
2945	conf->start_next_window = MaxSector;
2946	conf->current_window_requests = conf->next_window_requests = 0;
2947
2948	err = -EIO;
2949	for (i = 0; i < conf->raid_disks * 2; i++) {
2950
2951		disk = conf->mirrors + i;
2952
2953		if (i < conf->raid_disks &&
2954		    disk[conf->raid_disks].rdev) {
2955			/* This slot has a replacement. */
2956			if (!disk->rdev) {
2957				/* No original, just make the replacement
2958				 * a recovering spare
2959				 */
2960				disk->rdev =
2961					disk[conf->raid_disks].rdev;
2962				disk[conf->raid_disks].rdev = NULL;
2963			} else if (!test_bit(In_sync, &disk->rdev->flags))
2964				/* Original is not in_sync - bad */
2965				goto abort;
2966		}
2967
2968		if (!disk->rdev ||
2969		    !test_bit(In_sync, &disk->rdev->flags)) {
2970			disk->head_position = 0;
2971			if (disk->rdev &&
2972			    (disk->rdev->saved_raid_disk < 0))
2973				conf->fullsync = 1;
2974		}
2975	}
2976
2977	err = -ENOMEM;
2978	conf->thread = md_register_thread(raid1d, mddev, "raid1");
 
2979	if (!conf->thread)
2980		goto abort;
2981
2982	return conf;
2983
2984 abort:
2985	if (conf) {
2986		mempool_destroy(conf->r1bio_pool);
2987		kfree(conf->mirrors);
2988		safe_put_page(conf->tmppage);
2989		kfree(conf->poolinfo);
 
 
 
 
 
2990		kfree(conf);
2991	}
2992	return ERR_PTR(err);
2993}
2994
2995static void raid1_free(struct mddev *mddev, void *priv);
 
 
 
 
 
 
 
 
 
 
 
 
 
2996static int raid1_run(struct mddev *mddev)
2997{
2998	struct r1conf *conf;
2999	int i;
3000	struct md_rdev *rdev;
3001	int ret;
3002	bool discard_supported = false;
3003
3004	if (mddev->level != 1) {
3005		pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n",
3006			mdname(mddev), mddev->level);
3007		return -EIO;
3008	}
3009	if (mddev->reshape_position != MaxSector) {
3010		pr_warn("md/raid1:%s: reshape_position set but not supported\n",
3011			mdname(mddev));
3012		return -EIO;
3013	}
 
3014	/*
3015	 * copy the already verified devices into our private RAID1
3016	 * bookkeeping area. [whatever we allocate in run(),
3017	 * should be freed in raid1_free()]
3018	 */
3019	if (mddev->private == NULL)
3020		conf = setup_conf(mddev);
3021	else
3022		conf = mddev->private;
3023
3024	if (IS_ERR(conf))
3025		return PTR_ERR(conf);
3026
3027	if (mddev->queue)
3028		blk_queue_max_write_same_sectors(mddev->queue, 0);
3029
3030	rdev_for_each(rdev, mddev) {
3031		if (!mddev->gendisk)
3032			continue;
3033		disk_stack_limits(mddev->gendisk, rdev->bdev,
3034				  rdev->data_offset << 9);
3035		if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
3036			discard_supported = true;
3037	}
3038
3039	mddev->degraded = 0;
3040	for (i=0; i < conf->raid_disks; i++)
3041		if (conf->mirrors[i].rdev == NULL ||
3042		    !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
3043		    test_bit(Faulty, &conf->mirrors[i].rdev->flags))
3044			mddev->degraded++;
 
 
 
 
 
 
 
3045
3046	if (conf->raid_disks - mddev->degraded == 1)
3047		mddev->recovery_cp = MaxSector;
3048
3049	if (mddev->recovery_cp != MaxSector)
3050		pr_info("md/raid1:%s: not clean -- starting background reconstruction\n",
3051			mdname(mddev));
3052	pr_info("md/raid1:%s: active with %d out of %d mirrors\n",
3053		mdname(mddev), mddev->raid_disks - mddev->degraded,
3054		mddev->raid_disks);
3055
3056	/*
3057	 * Ok, everything is just fine now
3058	 */
3059	mddev->thread = conf->thread;
3060	conf->thread = NULL;
3061	mddev->private = conf;
3062	set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
3063
3064	md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
3065
3066	if (mddev->queue) {
3067		if (discard_supported)
3068			queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
3069						mddev->queue);
3070		else
3071			queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
3072						  mddev->queue);
3073	}
3074
3075	ret =  md_integrity_register(mddev);
3076	if (ret) {
3077		md_unregister_thread(&mddev->thread);
3078		raid1_free(mddev, conf);
3079	}
3080	return ret;
3081}
3082
3083static void raid1_free(struct mddev *mddev, void *priv)
3084{
3085	struct r1conf *conf = priv;
3086
3087	mempool_destroy(conf->r1bio_pool);
3088	kfree(conf->mirrors);
3089	safe_put_page(conf->tmppage);
3090	kfree(conf->poolinfo);
 
 
 
 
 
3091	kfree(conf);
3092}
3093
3094static int raid1_resize(struct mddev *mddev, sector_t sectors)
3095{
3096	/* no resync is happening, and there is enough space
3097	 * on all devices, so we can resize.
3098	 * We need to make sure resync covers any new space.
3099	 * If the array is shrinking we should possibly wait until
3100	 * any io in the removed space completes, but it hardly seems
3101	 * worth it.
3102	 */
3103	sector_t newsize = raid1_size(mddev, sectors, 0);
 
 
3104	if (mddev->external_size &&
3105	    mddev->array_sectors > newsize)
3106		return -EINVAL;
3107	if (mddev->bitmap) {
3108		int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
3109		if (ret)
3110			return ret;
3111	}
3112	md_set_array_sectors(mddev, newsize);
3113	set_capacity(mddev->gendisk, mddev->array_sectors);
3114	revalidate_disk(mddev->gendisk);
3115	if (sectors > mddev->dev_sectors &&
3116	    mddev->recovery_cp > mddev->dev_sectors) {
3117		mddev->recovery_cp = mddev->dev_sectors;
3118		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3119	}
3120	mddev->dev_sectors = sectors;
3121	mddev->resync_max_sectors = sectors;
3122	return 0;
3123}
3124
3125static int raid1_reshape(struct mddev *mddev)
3126{
3127	/* We need to:
3128	 * 1/ resize the r1bio_pool
3129	 * 2/ resize conf->mirrors
3130	 *
3131	 * We allocate a new r1bio_pool if we can.
3132	 * Then raise a device barrier and wait until all IO stops.
3133	 * Then resize conf->mirrors and swap in the new r1bio pool.
3134	 *
3135	 * At the same time, we "pack" the devices so that all the missing
3136	 * devices have the higher raid_disk numbers.
3137	 */
3138	mempool_t *newpool, *oldpool;
3139	struct pool_info *newpoolinfo;
3140	struct raid1_info *newmirrors;
3141	struct r1conf *conf = mddev->private;
3142	int cnt, raid_disks;
3143	unsigned long flags;
3144	int d, d2, err;
 
 
 
 
3145
3146	/* Cannot change chunk_size, layout, or level */
3147	if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
3148	    mddev->layout != mddev->new_layout ||
3149	    mddev->level != mddev->new_level) {
3150		mddev->new_chunk_sectors = mddev->chunk_sectors;
3151		mddev->new_layout = mddev->layout;
3152		mddev->new_level = mddev->level;
3153		return -EINVAL;
3154	}
3155
3156	if (!mddev_is_clustered(mddev)) {
3157		err = md_allow_write(mddev);
3158		if (err)
3159			return err;
3160	}
3161
3162	raid_disks = mddev->raid_disks + mddev->delta_disks;
3163
3164	if (raid_disks < conf->raid_disks) {
3165		cnt=0;
3166		for (d= 0; d < conf->raid_disks; d++)
3167			if (conf->mirrors[d].rdev)
3168				cnt++;
3169		if (cnt > raid_disks)
3170			return -EBUSY;
3171	}
3172
3173	newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3174	if (!newpoolinfo)
3175		return -ENOMEM;
3176	newpoolinfo->mddev = mddev;
3177	newpoolinfo->raid_disks = raid_disks * 2;
3178
3179	newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3180				 r1bio_pool_free, newpoolinfo);
3181	if (!newpool) {
3182		kfree(newpoolinfo);
3183		return -ENOMEM;
3184	}
3185	newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
 
3186			     GFP_KERNEL);
3187	if (!newmirrors) {
3188		kfree(newpoolinfo);
3189		mempool_destroy(newpool);
3190		return -ENOMEM;
3191	}
3192
3193	freeze_array(conf, 0);
3194
3195	/* ok, everything is stopped */
3196	oldpool = conf->r1bio_pool;
3197	conf->r1bio_pool = newpool;
3198
3199	for (d = d2 = 0; d < conf->raid_disks; d++) {
3200		struct md_rdev *rdev = conf->mirrors[d].rdev;
3201		if (rdev && rdev->raid_disk != d2) {
3202			sysfs_unlink_rdev(mddev, rdev);
3203			rdev->raid_disk = d2;
3204			sysfs_unlink_rdev(mddev, rdev);
3205			if (sysfs_link_rdev(mddev, rdev))
3206				pr_warn("md/raid1:%s: cannot register rd%d\n",
3207					mdname(mddev), rdev->raid_disk);
3208		}
3209		if (rdev)
3210			newmirrors[d2++].rdev = rdev;
3211	}
3212	kfree(conf->mirrors);
3213	conf->mirrors = newmirrors;
3214	kfree(conf->poolinfo);
3215	conf->poolinfo = newpoolinfo;
3216
3217	spin_lock_irqsave(&conf->device_lock, flags);
3218	mddev->degraded += (raid_disks - conf->raid_disks);
3219	spin_unlock_irqrestore(&conf->device_lock, flags);
3220	conf->raid_disks = mddev->raid_disks = raid_disks;
3221	mddev->delta_disks = 0;
3222
3223	unfreeze_array(conf);
3224
3225	set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3226	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3227	md_wakeup_thread(mddev->thread);
3228
3229	mempool_destroy(oldpool);
3230	return 0;
3231}
3232
3233static void raid1_quiesce(struct mddev *mddev, int state)
3234{
3235	struct r1conf *conf = mddev->private;
3236
3237	switch(state) {
3238	case 2: /* wake for suspend */
3239		wake_up(&conf->wait_barrier);
3240		break;
3241	case 1:
3242		freeze_array(conf, 0);
3243		break;
3244	case 0:
3245		unfreeze_array(conf);
3246		break;
3247	}
3248}
3249
3250static void *raid1_takeover(struct mddev *mddev)
3251{
3252	/* raid1 can take over:
3253	 *  raid5 with 2 devices, any layout or chunk size
3254	 */
3255	if (mddev->level == 5 && mddev->raid_disks == 2) {
3256		struct r1conf *conf;
3257		mddev->new_level = 1;
3258		mddev->new_layout = 0;
3259		mddev->new_chunk_sectors = 0;
3260		conf = setup_conf(mddev);
3261		if (!IS_ERR(conf)) {
3262			/* Array must appear to be quiesced */
3263			conf->array_frozen = 1;
3264			mddev_clear_unsupported_flags(mddev,
3265				UNSUPPORTED_MDDEV_FLAGS);
3266		}
3267		return conf;
3268	}
3269	return ERR_PTR(-EINVAL);
3270}
3271
3272static struct md_personality raid1_personality =
3273{
3274	.name		= "raid1",
3275	.level		= 1,
3276	.owner		= THIS_MODULE,
3277	.make_request	= raid1_make_request,
3278	.run		= raid1_run,
3279	.free		= raid1_free,
3280	.status		= raid1_status,
3281	.error_handler	= raid1_error,
3282	.hot_add_disk	= raid1_add_disk,
3283	.hot_remove_disk= raid1_remove_disk,
3284	.spare_active	= raid1_spare_active,
3285	.sync_request	= raid1_sync_request,
3286	.resize		= raid1_resize,
3287	.size		= raid1_size,
3288	.check_reshape	= raid1_reshape,
3289	.quiesce	= raid1_quiesce,
3290	.takeover	= raid1_takeover,
3291	.congested	= raid1_congested,
3292};
3293
3294static int __init raid_init(void)
3295{
3296	return register_md_personality(&raid1_personality);
3297}
3298
3299static void raid_exit(void)
3300{
3301	unregister_md_personality(&raid1_personality);
3302}
3303
3304module_init(raid_init);
3305module_exit(raid_exit);
3306MODULE_LICENSE("GPL");
3307MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3308MODULE_ALIAS("md-personality-3"); /* RAID1 */
3309MODULE_ALIAS("md-raid1");
3310MODULE_ALIAS("md-level-1");
3311
3312module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);