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