SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/md/raid1.c

   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 "dm-bio-list.h"
  35 #include <linux/raid/raid1.h>
  36 #include <linux/raid/bitmap.h>
  37
  38 #define DEBUG 0
  39 #if DEBUG
  40 #define PRINTK(x...) printk(x)
  41 #else
  42 #define PRINTK(x...)
  43 #endif
  44
  45 /*
  46  * Number of guaranteed r1bios in case of extreme VM load:
  47  */
  48 #define NR_RAID1_BIOS 256
  49
  50 static mdk_personality_t raid1_personality;
  51
  52 static void unplug_slaves(mddev_t *mddev);
  53
  54 static void allow_barrier(conf_t *conf);
  55 static void lower_barrier(conf_t *conf);
  56
  57 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
  58 {
  59         struct pool_info *pi = data;
  60         r1bio_t *r1_bio;
  61         int size = offsetof(r1bio_t, bios[pi->raid_disks]);
  62
  63         /* allocate a r1bio with room for raid_disks entries in the bios array */
  64         r1_bio = kmalloc(size, gfp_flags);
  65         if (r1_bio)
  66                 memset(r1_bio, 0, size);
  67         else
  68                 unplug_slaves(pi->mddev);
  69
  70         return r1_bio;
  71 }
  72
  73 static void r1bio_pool_free(void *r1_bio, void *data)
  74 {
  75         kfree(r1_bio);
  76 }
  77
  78 #define RESYNC_BLOCK_SIZE (64*1024)
  79 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
  80 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
  81 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
  82 #define RESYNC_WINDOW (2048*1024)
  83
  84 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
  85 {
  86         struct pool_info *pi = data;
  87         struct page *page;
  88         r1bio_t *r1_bio;
  89         struct bio *bio;
  90         int i, j;
  91
  92         r1_bio = r1bio_pool_alloc(gfp_flags, pi);
  93         if (!r1_bio) {
  94                 unplug_slaves(pi->mddev);
  95                 return NULL;
  96         }
  97
  98         /*
  99          * Allocate bios : 1 for reading, n-1 for writing
 100          */
 101         for (j = pi->raid_disks ; j-- ; ) {
 102                 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
 103                 if (!bio)
 104                         goto out_free_bio;
 105                 r1_bio->bios[j] = bio;
 106         }
 107         /*
 108          * Allocate RESYNC_PAGES data pages and attach them to
 109          * the first bio;
 110          */
 111         bio = r1_bio->bios[0];
 112         for (i = 0; i < RESYNC_PAGES; i++) {
 113                 page = alloc_page(gfp_flags);
 114                 if (unlikely(!page))
 115                         goto out_free_pages;
 116
 117                 bio->bi_io_vec[i].bv_page = page;
 118         }
 119
 120         r1_bio->master_bio = NULL;
 121
 122         return r1_bio;
 123
 124 out_free_pages:
 125         for ( ; i > 0 ; i--)
 126                 __free_page(bio->bi_io_vec[i-1].bv_page);
 127 out_free_bio:
 128         while ( ++j < pi->raid_disks )
 129                 bio_put(r1_bio->bios[j]);
 130         r1bio_pool_free(r1_bio, data);
 131         return NULL;
 132 }
 133
 134 static void r1buf_pool_free(void *__r1_bio, void *data)
 135 {
 136         struct pool_info *pi = data;
 137         int i;
 138         r1bio_t *r1bio = __r1_bio;
 139         struct bio *bio = r1bio->bios[0];
 140
 141         for (i = 0; i < RESYNC_PAGES; i++) {
 142                 __free_page(bio->bi_io_vec[i].bv_page);
 143                 bio->bi_io_vec[i].bv_page = NULL;
 144         }
 145         for (i=0 ; i < pi->raid_disks; i++)
 146                 bio_put(r1bio->bios[i]);
 147
 148         r1bio_pool_free(r1bio, data);
 149 }
 150
 151 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
 152 {
 153         int i;
 154
 155         for (i = 0; i < conf->raid_disks; i++) {
 156                 struct bio **bio = r1_bio->bios + i;
 157                 if (*bio)
 158                         bio_put(*bio);
 159                 *bio = NULL;
 160         }
 161 }
 162
 163 static inline void free_r1bio(r1bio_t *r1_bio)
 164 {
 165         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 166
 167         /*
 168          * Wake up any possible resync thread that waits for the device
 169          * to go idle.
 170          */
 171         allow_barrier(conf);
 172
 173         put_all_bios(conf, r1_bio);
 174         mempool_free(r1_bio, conf->r1bio_pool);
 175 }
 176
 177 static inline void put_buf(r1bio_t *r1_bio)
 178 {
 179         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 180         int i;
 181
 182         for (i=0; i<conf->raid_disks; i++) {
 183                 struct bio *bio = r1_bio->bios[i];
 184                 if (bio->bi_end_io)
 185                         rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
 186         }
 187
 188         mempool_free(r1_bio, conf->r1buf_pool);
 189
 190         lower_barrier(conf);
 191 }
 192
 193 static void reschedule_retry(r1bio_t *r1_bio)
 194 {
 195         unsigned long flags;
 196         mddev_t *mddev = r1_bio->mddev;
 197         conf_t *conf = mddev_to_conf(mddev);
 198
 199         spin_lock_irqsave(&conf->device_lock, flags);
 200         list_add(&r1_bio->retry_list, &conf->retry_list);
 201         conf->nr_queued ++;
 202         spin_unlock_irqrestore(&conf->device_lock, flags);
 203
 204         wake_up(&conf->wait_barrier);
 205         md_wakeup_thread(mddev->thread);
 206 }
 207
 208 /*
 209  * raid_end_bio_io() is called when we have finished servicing a mirrored
 210  * operation and are ready to return a success/failure code to the buffer
 211  * cache layer.
 212  */
 213 static void raid_end_bio_io(r1bio_t *r1_bio)
 214 {
 215         struct bio *bio = r1_bio->master_bio;
 216
 217         /* if nobody has done the final endio yet, do it now */
 218         if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
 219                 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
 220                         (bio_data_dir(bio) == WRITE) ? "write" : "read",
 221                         (unsigned long long) bio->bi_sector,
 222                         (unsigned long long) bio->bi_sector +
 223                                 (bio->bi_size >> 9) - 1);
 224
 225                 bio_endio(bio, bio->bi_size,
 226                         test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
 227         }
 228         free_r1bio(r1_bio);
 229 }
 230
 231 /*
 232  * Update disk head position estimator based on IRQ completion info.
 233  */
 234 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
 235 {
 236         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 237
 238         conf->mirrors[disk].head_position =
 239                 r1_bio->sector + (r1_bio->sectors);
 240 }
 241
 242 static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
 243 {
 244         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 245         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 246         int mirror;
 247         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 248
 249         if (bio->bi_size)
 250                 return 1;
 251
 252         mirror = r1_bio->read_disk;
 253         /*
 254          * this branch is our 'one mirror IO has finished' event handler:
 255          */
 256         update_head_pos(mirror, r1_bio);
 257
 258         if (uptodate || conf->working_disks <= 1) {
 259                 /*
 260                  * Set R1BIO_Uptodate in our master bio, so that
 261                  * we will return a good error code for to the higher
 262                  * levels even if IO on some other mirrored buffer fails.
 263                  *
 264                  * The 'master' represents the composite IO operation to
 265                  * user-side. So if something waits for IO, then it will
 266                  * wait for the 'master' bio.
 267                  */
 268                 set_bit(R1BIO_Uptodate, &r1_bio->state);
 269
 270                 raid_end_bio_io(r1_bio);
 271         } else {
 272                 /*
 273                  * oops, read error:
 274                  */
 275                 char b[BDEVNAME_SIZE];
 276                 if (printk_ratelimit())
 277                         printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
 278                                bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
 279                 reschedule_retry(r1_bio);
 280         }
 281
 282         rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 283         return 0;
 284 }
 285
 286 static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
 287 {
 288         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 289         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 290         int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
 291         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 292
 293         if (bio->bi_size)
 294                 return 1;
 295
 296         for (mirror = 0; mirror < conf->raid_disks; mirror++)
 297                 if (r1_bio->bios[mirror] == bio)
 298                         break;
 299
 300         if (error == -ENOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
 301                 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
 302                 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
 303                 r1_bio->mddev->barriers_work = 0;
 304         } else {
 305                 /*
 306                  * this branch is our 'one mirror IO has finished' event handler:
 307                  */
 308                 r1_bio->bios[mirror] = NULL;
 309                 if (!uptodate) {
 310                         md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
 311                         /* an I/O failed, we can't clear the bitmap */
 312                         set_bit(R1BIO_Degraded, &r1_bio->state);
 313                 } else
 314                         /*
 315                          * Set R1BIO_Uptodate in our master bio, so that
 316                          * we will return a good error code for to the higher
 317                          * levels even if IO on some other mirrored buffer fails.
 318                          *
 319                          * The 'master' represents the composite IO operation to
 320                          * user-side. So if something waits for IO, then it will
 321                          * wait for the 'master' bio.
 322                          */
 323                         set_bit(R1BIO_Uptodate, &r1_bio->state);
 324
 325                 update_head_pos(mirror, r1_bio);
 326
 327                 if (behind) {
 328                         if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
 329                                 atomic_dec(&r1_bio->behind_remaining);
 330
 331                         /* In behind mode, we ACK the master bio once the I/O has safely
 332                          * reached all non-writemostly disks. Setting the Returned bit
 333                          * ensures that this gets done only once -- we don't ever want to
 334                          * return -EIO here, instead we'll wait */
 335
 336                         if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
 337                             test_bit(R1BIO_Uptodate, &r1_bio->state)) {
 338                                 /* Maybe we can return now */
 339                                 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
 340                                         struct bio *mbio = r1_bio->master_bio;
 341                                         PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
 342                                                (unsigned long long) mbio->bi_sector,
 343                                                (unsigned long long) mbio->bi_sector +
 344                                                (mbio->bi_size >> 9) - 1);
 345                                         bio_endio(mbio, mbio->bi_size, 0);
 346                                 }
 347                         }
 348                 }
 349         }
 350         /*
 351          *
 352          * Let's see if all mirrored write operations have finished
 353          * already.
 354          */
 355         if (atomic_dec_and_test(&r1_bio->remaining)) {
 356                 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
 357                         reschedule_retry(r1_bio);
 358                         /* Don't dec_pending yet, we want to hold
 359                          * the reference over the retry
 360                          */
 361                         return 0;
 362                 }
 363                 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
 364                         /* free extra copy of the data pages */
 365                         int i = bio->bi_vcnt;
 366                         while (i--)
 367                                 __free_page(bio->bi_io_vec[i].bv_page);
 368                 }
 369                 /* clear the bitmap if all writes complete successfully */
 370                 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
 371                                 r1_bio->sectors,
 372                                 !test_bit(R1BIO_Degraded, &r1_bio->state),
 373                                 behind);
 374                 md_write_end(r1_bio->mddev);
 375                 raid_end_bio_io(r1_bio);
 376         }
 377
 378         if (r1_bio->bios[mirror]==NULL)
 379                 bio_put(bio);
 380
 381         rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 382         return 0;
 383 }
 384
 385
 386 /*
 387  * This routine returns the disk from which the requested read should
 388  * be done. There is a per-array 'next expected sequential IO' sector
 389  * number - if this matches on the next IO then we use the last disk.
 390  * There is also a per-disk 'last know head position' sector that is
 391  * maintained from IRQ contexts, both the normal and the resync IO
 392  * completion handlers update this position correctly. If there is no
 393  * perfect sequential match then we pick the disk whose head is closest.
 394  *
 395  * If there are 2 mirrors in the same 2 devices, performance degrades
 396  * because position is mirror, not device based.
 397  *
 398  * The rdev for the device selected will have nr_pending incremented.
 399  */
 400 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
 401 {
 402         const unsigned long this_sector = r1_bio->sector;
 403         int new_disk = conf->last_used, disk = new_disk;
 404         int wonly_disk = -1;
 405         const int sectors = r1_bio->sectors;
 406         sector_t new_distance, current_distance;
 407         mdk_rdev_t *rdev;
 408
 409         rcu_read_lock();
 410         /*
 411          * Check if we can balance. We can balance on the whole
 412          * device if no resync is going on, or below the resync window.
 413          * We take the first readable disk when above the resync window.
 414          */
 415  retry:
 416         if (conf->mddev->recovery_cp < MaxSector &&
 417             (this_sector + sectors >= conf->next_resync)) {
 418                 /* Choose the first operation device, for consistancy */
 419                 new_disk = 0;
 420
 421                 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
 422                      !rdev || !test_bit(In_sync, &rdev->flags)
 423                              || test_bit(WriteMostly, &rdev->flags);
 424                      rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
 425
 426                         if (rdev && test_bit(In_sync, &rdev->flags))
 427                                 wonly_disk = new_disk;
 428
 429                         if (new_disk == conf->raid_disks - 1) {
 430                                 new_disk = wonly_disk;
 431                                 break;
 432                         }
 433                 }
 434                 goto rb_out;
 435         }
 436
 437
 438         /* make sure the disk is operational */
 439         for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
 440              !rdev || !test_bit(In_sync, &rdev->flags) ||
 441                      test_bit(WriteMostly, &rdev->flags);
 442              rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
 443
 444                 if (rdev && test_bit(In_sync, &rdev->flags))
 445                         wonly_disk = new_disk;
 446
 447                 if (new_disk <= 0)
 448                         new_disk = conf->raid_disks;
 449                 new_disk--;
 450                 if (new_disk == disk) {
 451                         new_disk = wonly_disk;
 452                         break;
 453                 }
 454         }
 455
 456         if (new_disk < 0)
 457                 goto rb_out;
 458
 459         disk = new_disk;
 460         /* now disk == new_disk == starting point for search */
 461
 462         /*
 463          * Don't change to another disk for sequential reads:
 464          */
 465         if (conf->next_seq_sect == this_sector)
 466                 goto rb_out;
 467         if (this_sector == conf->mirrors[new_disk].head_position)
 468                 goto rb_out;
 469
 470         current_distance = abs(this_sector - conf->mirrors[disk].head_position);
 471
 472         /* Find the disk whose head is closest */
 473
 474         do {
 475                 if (disk <= 0)
 476                         disk = conf->raid_disks;
 477                 disk--;
 478
 479                 rdev = rcu_dereference(conf->mirrors[disk].rdev);
 480
 481                 if (!rdev ||
 482                     !test_bit(In_sync, &rdev->flags) ||
 483                     test_bit(WriteMostly, &rdev->flags))
 484                         continue;
 485
 486                 if (!atomic_read(&rdev->nr_pending)) {
 487                         new_disk = disk;
 488                         break;
 489                 }
 490                 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
 491                 if (new_distance < current_distance) {
 492                         current_distance = new_distance;
 493                         new_disk = disk;
 494                 }
 495         } while (disk != conf->last_used);
 496
 497  rb_out:
 498
 499
 500         if (new_disk >= 0) {
 501                 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
 502                 if (!rdev)
 503                         goto retry;
 504                 atomic_inc(&rdev->nr_pending);
 505                 if (!test_bit(In_sync, &rdev->flags)) {
 506                         /* cannot risk returning a device that failed
 507                          * before we inc'ed nr_pending
 508                          */
 509                         atomic_dec(&rdev->nr_pending);
 510                         goto retry;
 511                 }
 512                 conf->next_seq_sect = this_sector + sectors;
 513                 conf->last_used = new_disk;
 514         }
 515         rcu_read_unlock();
 516
 517         return new_disk;
 518 }
 519
 520 static void unplug_slaves(mddev_t *mddev)
 521 {
 522         conf_t *conf = mddev_to_conf(mddev);
 523         int i;
 524
 525         rcu_read_lock();
 526         for (i=0; i<mddev->raid_disks; i++) {
 527                 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
 528                 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
 529                         request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
 530
 531                         atomic_inc(&rdev->nr_pending);
 532                         rcu_read_unlock();
 533
 534                         if (r_queue->unplug_fn)
 535                                 r_queue->unplug_fn(r_queue);
 536
 537                         rdev_dec_pending(rdev, mddev);
 538                         rcu_read_lock();
 539                 }
 540         }
 541         rcu_read_unlock();
 542 }
 543
 544 static void raid1_unplug(request_queue_t *q)
 545 {
 546         mddev_t *mddev = q->queuedata;
 547
 548         unplug_slaves(mddev);
 549         md_wakeup_thread(mddev->thread);
 550 }
 551
 552 static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
 553                              sector_t *error_sector)
 554 {
 555         mddev_t *mddev = q->queuedata;
 556         conf_t *conf = mddev_to_conf(mddev);
 557         int i, ret = 0;
 558
 559         rcu_read_lock();
 560         for (i=0; i<mddev->raid_disks && ret == 0; i++) {
 561                 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
 562                 if (rdev && !test_bit(Faulty, &rdev->flags)) {
 563                         struct block_device *bdev = rdev->bdev;
 564                         request_queue_t *r_queue = bdev_get_queue(bdev);
 565
 566                         if (!r_queue->issue_flush_fn)
 567                                 ret = -EOPNOTSUPP;
 568                         else {
 569                                 atomic_inc(&rdev->nr_pending);
 570                                 rcu_read_unlock();
 571                                 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
 572                                                               error_sector);
 573                                 rdev_dec_pending(rdev, mddev);
 574                                 rcu_read_lock();
 575                         }
 576                 }
 577         }
 578         rcu_read_unlock();
 579         return ret;
 580 }
 581
 582 /* Barriers....
 583  * Sometimes we need to suspend IO while we do something else,
 584  * either some resync/recovery, or reconfigure the array.
 585  * To do this we raise a 'barrier'.
 586  * The 'barrier' is a counter that can be raised multiple times
 587  * to count how many activities are happening which preclude
 588  * normal IO.
 589  * We can only raise the barrier if there is no pending IO.
 590  * i.e. if nr_pending == 0.
 591  * We choose only to raise the barrier if no-one is waiting for the
 592  * barrier to go down.  This means that as soon as an IO request
 593  * is ready, no other operations which require a barrier will start
 594  * until the IO request has had a chance.
 595  *
 596  * So: regular IO calls 'wait_barrier'.  When that returns there
 597  *    is no backgroup IO happening,  It must arrange to call
 598  *    allow_barrier when it has finished its IO.
 599  * backgroup IO calls must call raise_barrier.  Once that returns
 600  *    there is no normal IO happeing.  It must arrange to call
 601  *    lower_barrier when the particular background IO completes.
 602  */
 603 #define RESYNC_DEPTH 32
 604
 605 static void raise_barrier(conf_t *conf)
 606 {
 607         spin_lock_irq(&conf->resync_lock);
 608
 609         /* Wait until no block IO is waiting */
 610         wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
 611                             conf->resync_lock,
 612                             raid1_unplug(conf->mddev->queue));
 613
 614         /* block any new IO from starting */
 615         conf->barrier++;
 616
 617         /* No wait for all pending IO to complete */
 618         wait_event_lock_irq(conf->wait_barrier,
 619                             !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
 620                             conf->resync_lock,
 621                             raid1_unplug(conf->mddev->queue));
 622
 623         spin_unlock_irq(&conf->resync_lock);
 624 }
 625
 626 static void lower_barrier(conf_t *conf)
 627 {
 628         unsigned long flags;
 629         spin_lock_irqsave(&conf->resync_lock, flags);
 630         conf->barrier--;
 631         spin_unlock_irqrestore(&conf->resync_lock, flags);
 632         wake_up(&conf->wait_barrier);
 633 }
 634
 635 static void wait_barrier(conf_t *conf)
 636 {
 637         spin_lock_irq(&conf->resync_lock);
 638         if (conf->barrier) {
 639                 conf->nr_waiting++;
 640                 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
 641                                     conf->resync_lock,
 642                                     raid1_unplug(conf->mddev->queue));
 643                 conf->nr_waiting--;
 644         }
 645         conf->nr_pending++;
 646         spin_unlock_irq(&conf->resync_lock);
 647 }
 648
 649 static void allow_barrier(conf_t *conf)
 650 {
 651         unsigned long flags;
 652         spin_lock_irqsave(&conf->resync_lock, flags);
 653         conf->nr_pending--;
 654         spin_unlock_irqrestore(&conf->resync_lock, flags);
 655         wake_up(&conf->wait_barrier);
 656 }
 657
 658 static void freeze_array(conf_t *conf)
 659 {
 660         /* stop syncio and normal IO and wait for everything to
 661          * go quite.
 662          * We increment barrier and nr_waiting, and then
 663          * wait until barrier+nr_pending match nr_queued+2
 664          */
 665         spin_lock_irq(&conf->resync_lock);
 666         conf->barrier++;
 667         conf->nr_waiting++;
 668         wait_event_lock_irq(conf->wait_barrier,
 669                             conf->barrier+conf->nr_pending == conf->nr_queued+2,
 670                             conf->resync_lock,
 671                             raid1_unplug(conf->mddev->queue));
 672         spin_unlock_irq(&conf->resync_lock);
 673 }
 674 static void unfreeze_array(conf_t *conf)
 675 {
 676         /* reverse the effect of the freeze */
 677         spin_lock_irq(&conf->resync_lock);
 678         conf->barrier--;
 679         conf->nr_waiting--;
 680         wake_up(&conf->wait_barrier);
 681         spin_unlock_irq(&conf->resync_lock);
 682 }
 683
 684
 685 /* duplicate the data pages for behind I/O */
 686 static struct page **alloc_behind_pages(struct bio *bio)
 687 {
 688         int i;
 689         struct bio_vec *bvec;
 690         struct page **pages = kmalloc(bio->bi_vcnt * sizeof(struct page *),
 691                                         GFP_NOIO);
 692         if (unlikely(!pages))
 693                 goto do_sync_io;
 694
 695         memset(pages, 0, bio->bi_vcnt * sizeof(struct page *));
 696
 697         bio_for_each_segment(bvec, bio, i) {
 698                 pages[i] = alloc_page(GFP_NOIO);
 699                 if (unlikely(!pages[i]))
 700                         goto do_sync_io;
 701                 memcpy(kmap(pages[i]) + bvec->bv_offset,
 702                         kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
 703                 kunmap(pages[i]);
 704                 kunmap(bvec->bv_page);
 705         }
 706
 707         return pages;
 708
 709 do_sync_io:
 710         if (pages)
 711                 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
 712                         __free_page(pages[i]);
 713         kfree(pages);
 714         PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
 715         return NULL;
 716 }
 717
 718 static int make_request(request_queue_t *q, struct bio * bio)
 719 {
 720         mddev_t *mddev = q->queuedata;
 721         conf_t *conf = mddev_to_conf(mddev);
 722         mirror_info_t *mirror;
 723         r1bio_t *r1_bio;
 724         struct bio *read_bio;
 725         int i, targets = 0, disks;
 726         mdk_rdev_t *rdev;
 727         struct bitmap *bitmap = mddev->bitmap;
 728         unsigned long flags;
 729         struct bio_list bl;
 730         struct page **behind_pages = NULL;
 731         const int rw = bio_data_dir(bio);
 732         int do_barriers;
 733
 734         if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
 735                 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
 736                 return 0;
 737         }
 738
 739         /*
 740          * Register the new request and wait if the reconstruction
 741          * thread has put up a bar for new requests.
 742          * Continue immediately if no resync is active currently.
 743          */
 744         md_write_start(mddev, bio); /* wait on superblock update early */
 745
 746         wait_barrier(conf);
 747
 748         disk_stat_inc(mddev->gendisk, ios[rw]);
 749         disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
 750
 751         /*
 752          * make_request() can abort the operation when READA is being
 753          * used and no empty request is available.
 754          *
 755          */
 756         r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
 757
 758         r1_bio->master_bio = bio;
 759         r1_bio->sectors = bio->bi_size >> 9;
 760         r1_bio->state = 0;
 761         r1_bio->mddev = mddev;
 762         r1_bio->sector = bio->bi_sector;
 763
 764         if (rw == READ) {
 765                 /*
 766                  * read balancing logic:
 767                  */
 768                 int rdisk = read_balance(conf, r1_bio);
 769
 770                 if (rdisk < 0) {
 771                         /* couldn't find anywhere to read from */
 772                         raid_end_bio_io(r1_bio);
 773                         return 0;
 774                 }
 775                 mirror = conf->mirrors + rdisk;
 776
 777                 r1_bio->read_disk = rdisk;
 778
 779                 read_bio = bio_clone(bio, GFP_NOIO);
 780
 781                 r1_bio->bios[rdisk] = read_bio;
 782
 783                 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
 784                 read_bio->bi_bdev = mirror->rdev->bdev;
 785                 read_bio->bi_end_io = raid1_end_read_request;
 786                 read_bio->bi_rw = READ;
 787                 read_bio->bi_private = r1_bio;
 788
 789                 generic_make_request(read_bio);
 790                 return 0;
 791         }
 792
 793         /*
 794          * WRITE:
 795          */
 796         /* first select target devices under spinlock and
 797          * inc refcount on their rdev.  Record them by setting
 798          * bios[x] to bio
 799          */
 800         disks = conf->raid_disks;
 801 #if 0
 802         { static int first=1;
 803         if (first) printk("First Write sector %llu disks %d\n",
 804                           (unsigned long long)r1_bio->sector, disks);
 805         first = 0;
 806         }
 807 #endif
 808         rcu_read_lock();
 809         for (i = 0;  i < disks; i++) {
 810                 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
 811                     !test_bit(Faulty, &rdev->flags)) {
 812                         atomic_inc(&rdev->nr_pending);
 813                         if (test_bit(Faulty, &rdev->flags)) {
 814                                 atomic_dec(&rdev->nr_pending);
 815                                 r1_bio->bios[i] = NULL;
 816                         } else
 817                                 r1_bio->bios[i] = bio;
 818                         targets++;
 819                 } else
 820                         r1_bio->bios[i] = NULL;
 821         }
 822         rcu_read_unlock();
 823
 824         BUG_ON(targets == 0); /* we never fail the last device */
 825
 826         if (targets < conf->raid_disks) {
 827                 /* array is degraded, we will not clear the bitmap
 828                  * on I/O completion (see raid1_end_write_request) */
 829                 set_bit(R1BIO_Degraded, &r1_bio->state);
 830         }
 831
 832         /* do behind I/O ? */
 833         if (bitmap &&
 834             atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
 835             (behind_pages = alloc_behind_pages(bio)) != NULL)
 836                 set_bit(R1BIO_BehindIO, &r1_bio->state);
 837
 838         atomic_set(&r1_bio->remaining, 0);
 839         atomic_set(&r1_bio->behind_remaining, 0);
 840
 841         do_barriers = bio->bi_rw & BIO_RW_BARRIER;
 842         if (do_barriers)
 843                 set_bit(R1BIO_Barrier, &r1_bio->state);
 844
 845         bio_list_init(&bl);
 846         for (i = 0; i < disks; i++) {
 847                 struct bio *mbio;
 848                 if (!r1_bio->bios[i])
 849                         continue;
 850
 851                 mbio = bio_clone(bio, GFP_NOIO);
 852                 r1_bio->bios[i] = mbio;
 853
 854                 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
 855                 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
 856                 mbio->bi_end_io = raid1_end_write_request;
 857                 mbio->bi_rw = WRITE | do_barriers;
 858                 mbio->bi_private = r1_bio;
 859
 860                 if (behind_pages) {
 861                         struct bio_vec *bvec;
 862                         int j;
 863
 864                         /* Yes, I really want the '__' version so that
 865                          * we clear any unused pointer in the io_vec, rather
 866                          * than leave them unchanged.  This is important
 867                          * because when we come to free the pages, we won't
 868                          * know the originial bi_idx, so we just free
 869                          * them all
 870                          */
 871                         __bio_for_each_segment(bvec, mbio, j, 0)
 872                                 bvec->bv_page = behind_pages[j];
 873                         if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
 874                                 atomic_inc(&r1_bio->behind_remaining);
 875                 }
 876
 877                 atomic_inc(&r1_bio->remaining);
 878
 879                 bio_list_add(&bl, mbio);
 880         }
 881         kfree(behind_pages); /* the behind pages are attached to the bios now */
 882
 883         bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
 884                                 test_bit(R1BIO_BehindIO, &r1_bio->state));
 885         spin_lock_irqsave(&conf->device_lock, flags);
 886         bio_list_merge(&conf->pending_bio_list, &bl);
 887         bio_list_init(&bl);
 888
 889         blk_plug_device(mddev->queue);
 890         spin_unlock_irqrestore(&conf->device_lock, flags);
 891
 892 #if 0
 893         while ((bio = bio_list_pop(&bl)) != NULL)
 894                 generic_make_request(bio);
 895 #endif
 896
 897         return 0;
 898 }
 899
 900 static void status(struct seq_file *seq, mddev_t *mddev)
 901 {
 902         conf_t *conf = mddev_to_conf(mddev);
 903         int i;
 904
 905         seq_printf(seq, " [%d/%d] [", conf->raid_disks,
 906                                                 conf->working_disks);
 907         for (i = 0; i < conf->raid_disks; i++)
 908                 seq_printf(seq, "%s",
 909                               conf->mirrors[i].rdev &&
 910                               test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
 911         seq_printf(seq, "]");
 912 }
 913
 914
 915 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
 916 {
 917         char b[BDEVNAME_SIZE];
 918         conf_t *conf = mddev_to_conf(mddev);
 919
 920         /*
 921          * If it is not operational, then we have already marked it as dead
 922          * else if it is the last working disks, ignore the error, let the
 923          * next level up know.
 924          * else mark the drive as failed
 925          */
 926         if (test_bit(In_sync, &rdev->flags)
 927             && conf->working_disks == 1)
 928                 /*
 929                  * Don't fail the drive, act as though we were just a
 930                  * normal single drive
 931                  */
 932                 return;
 933         if (test_bit(In_sync, &rdev->flags)) {
 934                 mddev->degraded++;
 935                 conf->working_disks--;
 936                 /*
 937                  * if recovery is running, make sure it aborts.
 938                  */
 939                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
 940         }
 941         clear_bit(In_sync, &rdev->flags);
 942         set_bit(Faulty, &rdev->flags);
 943         mddev->sb_dirty = 1;
 944         printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
 945                 "       Operation continuing on %d devices\n",
 946                 bdevname(rdev->bdev,b), conf->working_disks);
 947 }
 948
 949 static void print_conf(conf_t *conf)
 950 {
 951         int i;
 952         mirror_info_t *tmp;
 953
 954         printk("RAID1 conf printout:\n");
 955         if (!conf) {
 956                 printk("(!conf)\n");
 957                 return;
 958         }
 959         printk(" --- wd:%d rd:%d\n", conf->working_disks,
 960                 conf->raid_disks);
 961
 962         for (i = 0; i < conf->raid_disks; i++) {
 963                 char b[BDEVNAME_SIZE];
 964                 tmp = conf->mirrors + i;
 965                 if (tmp->rdev)
 966                         printk(" disk %d, wo:%d, o:%d, dev:%s\n",
 967                                 i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags),
 968                                 bdevname(tmp->rdev->bdev,b));
 969         }
 970 }
 971
 972 static void close_sync(conf_t *conf)
 973 {
 974         wait_barrier(conf);
 975         allow_barrier(conf);
 976
 977         mempool_destroy(conf->r1buf_pool);
 978         conf->r1buf_pool = NULL;
 979 }
 980
 981 static int raid1_spare_active(mddev_t *mddev)
 982 {
 983         int i;
 984         conf_t *conf = mddev->private;
 985         mirror_info_t *tmp;
 986
 987         /*
 988          * Find all failed disks within the RAID1 configuration
 989          * and mark them readable
 990          */
 991         for (i = 0; i < conf->raid_disks; i++) {
 992                 tmp = conf->mirrors + i;
 993                 if (tmp->rdev
 994                     && !test_bit(Faulty, &tmp->rdev->flags)
 995                     && !test_bit(In_sync, &tmp->rdev->flags)) {
 996                         conf->working_disks++;
 997                         mddev->degraded--;
 998                         set_bit(In_sync, &tmp->rdev->flags);
 999                 }
1000         }
1001
1002         print_conf(conf);
1003         return 0;
1004 }
1005
1006
1007 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1008 {
1009         conf_t *conf = mddev->private;
1010         int found = 0;
1011         int mirror = 0;
1012         mirror_info_t *p;
1013
1014         for (mirror=0; mirror < mddev->raid_disks; mirror++)
1015                 if ( !(p=conf->mirrors+mirror)->rdev) {
1016
1017                         blk_queue_stack_limits(mddev->queue,
1018                                                rdev->bdev->bd_disk->queue);
1019                         /* as we don't honour merge_bvec_fn, we must never risk
1020                          * violating it, so limit ->max_sector to one PAGE, as
1021                          * a one page request is never in violation.
1022                          */
1023                         if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1024                             mddev->queue->max_sectors > (PAGE_SIZE>>9))
1025                                 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1026
1027                         p->head_position = 0;
1028                         rdev->raid_disk = mirror;
1029                         found = 1;
1030                         /* As all devices are equivalent, we don't need a full recovery
1031                          * if this was recently any drive of the array
1032                          */
1033                         if (rdev->saved_raid_disk < 0)
1034                                 conf->fullsync = 1;
1035                         rcu_assign_pointer(p->rdev, rdev);
1036                         break;
1037                 }
1038
1039         print_conf(conf);
1040         return found;
1041 }
1042
1043 static int raid1_remove_disk(mddev_t *mddev, int number)
1044 {
1045         conf_t *conf = mddev->private;
1046         int err = 0;
1047         mdk_rdev_t *rdev;
1048         mirror_info_t *p = conf->mirrors+ number;
1049
1050         print_conf(conf);
1051         rdev = p->rdev;
1052         if (rdev) {
1053                 if (test_bit(In_sync, &rdev->flags) ||
1054                     atomic_read(&rdev->nr_pending)) {
1055                         err = -EBUSY;
1056                         goto abort;
1057                 }
1058                 p->rdev = NULL;
1059                 synchronize_rcu();
1060                 if (atomic_read(&rdev->nr_pending)) {
1061                         /* lost the race, try later */
1062                         err = -EBUSY;
1063                         p->rdev = rdev;
1064                 }
1065         }
1066 abort:
1067
1068         print_conf(conf);
1069         return err;
1070 }
1071
1072
1073 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
1074 {
1075         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1076
1077         if (bio->bi_size)
1078                 return 1;
1079
1080         if (r1_bio->bios[r1_bio->read_disk] != bio)
1081                 BUG();
1082         update_head_pos(r1_bio->read_disk, r1_bio);
1083         /*
1084          * we have read a block, now it needs to be re-written,
1085          * or re-read if the read failed.
1086          * We don't do much here, just schedule handling by raid1d
1087          */
1088         if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1089                 set_bit(R1BIO_Uptodate, &r1_bio->state);
1090         reschedule_retry(r1_bio);
1091         return 0;
1092 }
1093
1094 static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
1095 {
1096         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1097         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1098         mddev_t *mddev = r1_bio->mddev;
1099         conf_t *conf = mddev_to_conf(mddev);
1100         int i;
1101         int mirror=0;
1102
1103         if (bio->bi_size)
1104                 return 1;
1105
1106         for (i = 0; i < conf->raid_disks; i++)
1107                 if (r1_bio->bios[i] == bio) {
1108                         mirror = i;
1109                         break;
1110                 }
1111         if (!uptodate)
1112                 md_error(mddev, conf->mirrors[mirror].rdev);
1113
1114         update_head_pos(mirror, r1_bio);
1115
1116         if (atomic_dec_and_test(&r1_bio->remaining)) {
1117                 md_done_sync(mddev, r1_bio->sectors, uptodate);
1118                 put_buf(r1_bio);
1119         }
1120         return 0;
1121 }
1122
1123 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1124 {
1125         conf_t *conf = mddev_to_conf(mddev);
1126         int i;
1127         int disks = conf->raid_disks;
1128         struct bio *bio, *wbio;
1129
1130         bio = r1_bio->bios[r1_bio->read_disk];
1131
1132
1133         /*
1134          * schedule writes
1135          */
1136         if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1137                 /* ouch - failed to read all of that.
1138                  * Try some synchronous reads of other devices to get
1139                  * good data, much like with normal read errors.  Only
1140                  * read into the pages we already have so they we don't
1141                  * need to re-issue the read request.
1142                  * We don't need to freeze the array, because being in an
1143                  * active sync request, there is no normal IO, and
1144                  * no overlapping syncs.
1145                  */
1146                 sector_t sect = r1_bio->sector;
1147                 int sectors = r1_bio->sectors;
1148                 int idx = 0;
1149
1150                 while(sectors) {
1151                         int s = sectors;
1152                         int d = r1_bio->read_disk;
1153                         int success = 0;
1154                         mdk_rdev_t *rdev;
1155
1156                         if (s > (PAGE_SIZE>>9))
1157                                 s = PAGE_SIZE >> 9;
1158                         do {
1159                                 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1160                                         rdev = conf->mirrors[d].rdev;
1161                                         if (sync_page_io(rdev->bdev,
1162                                                          sect + rdev->data_offset,
1163                                                          s<<9,
1164                                                          bio->bi_io_vec[idx].bv_page,
1165                                                          READ)) {
1166                                                 success = 1;
1167                                                 break;
1168                                         }
1169                                 }
1170                                 d++;
1171                                 if (d == conf->raid_disks)
1172                                         d = 0;
1173                         } while (!success && d != r1_bio->read_disk);
1174
1175                         if (success) {
1176                                 /* write it back and re-read */
1177                                 set_bit(R1BIO_Uptodate, &r1_bio->state);
1178                                 while (d != r1_bio->read_disk) {
1179                                         if (d == 0)
1180                                                 d = conf->raid_disks;
1181                                         d--;
1182                                         if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1183                                                 continue;
1184                                         rdev = conf->mirrors[d].rdev;
1185                                         if (sync_page_io(rdev->bdev,
1186                                                          sect + rdev->data_offset,
1187                                                          s<<9,
1188                                                          bio->bi_io_vec[idx].bv_page,
1189                                                          WRITE) == 0 ||
1190                                             sync_page_io(rdev->bdev,
1191                                                          sect + rdev->data_offset,
1192                                                          s<<9,
1193                                                          bio->bi_io_vec[idx].bv_page,
1194                                                          READ) == 0) {
1195                                                 md_error(mddev, rdev);
1196                                         }
1197                                 }
1198                         } else {
1199                                 char b[BDEVNAME_SIZE];
1200                                 /* Cannot read from anywhere, array is toast */
1201                                 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1202                                 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1203                                        " for block %llu\n",
1204                                        bdevname(bio->bi_bdev,b),
1205                                        (unsigned long long)r1_bio->sector);
1206                                 md_done_sync(mddev, r1_bio->sectors, 0);
1207                                 put_buf(r1_bio);
1208                                 return;
1209                         }
1210                         sectors -= s;
1211                         sect += s;
1212                         idx ++;
1213                 }
1214         }
1215         atomic_set(&r1_bio->remaining, 1);
1216         for (i = 0; i < disks ; i++) {
1217                 wbio = r1_bio->bios[i];
1218                 if (wbio->bi_end_io == NULL ||
1219                     (wbio->bi_end_io == end_sync_read &&
1220                      (i == r1_bio->read_disk ||
1221                       !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1222                         continue;
1223
1224                 wbio->bi_rw = WRITE;
1225                 wbio->bi_end_io = end_sync_write;
1226                 atomic_inc(&r1_bio->remaining);
1227                 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
1228
1229                 generic_make_request(wbio);
1230         }
1231
1232         if (atomic_dec_and_test(&r1_bio->remaining)) {
1233                 /* if we're here, all write(s) have completed, so clean up */
1234                 md_done_sync(mddev, r1_bio->sectors, 1);
1235                 put_buf(r1_bio);
1236         }
1237 }
1238
1239 /*
1240  * This is a kernel thread which:
1241  *
1242  *      1.      Retries failed read operations on working mirrors.
1243  *      2.      Updates the raid superblock when problems encounter.
1244  *      3.      Performs writes following reads for array syncronising.
1245  */
1246
1247 static void raid1d(mddev_t *mddev)
1248 {
1249         r1bio_t *r1_bio;
1250         struct bio *bio;
1251         unsigned long flags;
1252         conf_t *conf = mddev_to_conf(mddev);
1253         struct list_head *head = &conf->retry_list;
1254         int unplug=0;
1255         mdk_rdev_t *rdev;
1256
1257         md_check_recovery(mddev);
1258
1259         for (;;) {
1260                 char b[BDEVNAME_SIZE];
1261                 spin_lock_irqsave(&conf->device_lock, flags);
1262
1263                 if (conf->pending_bio_list.head) {
1264                         bio = bio_list_get(&conf->pending_bio_list);
1265                         blk_remove_plug(mddev->queue);
1266                         spin_unlock_irqrestore(&conf->device_lock, flags);
1267                         /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1268                         if (bitmap_unplug(mddev->bitmap) != 0)
1269                                 printk("%s: bitmap file write failed!\n", mdname(mddev));
1270
1271                         while (bio) { /* submit pending writes */
1272                                 struct bio *next = bio->bi_next;
1273                                 bio->bi_next = NULL;
1274                                 generic_make_request(bio);
1275                                 bio = next;
1276                         }
1277                         unplug = 1;
1278
1279                         continue;
1280                 }
1281
1282                 if (list_empty(head))
1283                         break;
1284                 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1285                 list_del(head->prev);
1286                 conf->nr_queued--;
1287                 spin_unlock_irqrestore(&conf->device_lock, flags);
1288
1289                 mddev = r1_bio->mddev;
1290                 conf = mddev_to_conf(mddev);
1291                 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1292                         sync_request_write(mddev, r1_bio);
1293                         unplug = 1;
1294                 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1295                         /* some requests in the r1bio were BIO_RW_BARRIER
1296                          * requests which failed with -ENOTSUPP.  Hohumm..
1297                          * Better resubmit without the barrier.
1298                          * We know which devices to resubmit for, because
1299                          * all others have had their bios[] entry cleared.
1300                          */
1301                         int i;
1302                         clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1303                         clear_bit(R1BIO_Barrier, &r1_bio->state);
1304                         for (i=0; i < conf->raid_disks; i++)
1305                                 if (r1_bio->bios[i]) {
1306                                         struct bio_vec *bvec;
1307                                         int j;
1308
1309                                         bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1310                                         /* copy pages from the failed bio, as
1311                                          * this might be a write-behind device */
1312                                         __bio_for_each_segment(bvec, bio, j, 0)
1313                                                 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1314                                         bio_put(r1_bio->bios[i]);
1315                                         bio->bi_sector = r1_bio->sector +
1316                                                 conf->mirrors[i].rdev->data_offset;
1317                                         bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1318                                         bio->bi_end_io = raid1_end_write_request;
1319                                         bio->bi_rw = WRITE;
1320                                         bio->bi_private = r1_bio;
1321                                         r1_bio->bios[i] = bio;
1322                                         generic_make_request(bio);
1323                                 }
1324                 } else {
1325                         int disk;
1326
1327                         /* we got a read error. Maybe the drive is bad.  Maybe just
1328                          * the block and we can fix it.
1329                          * We freeze all other IO, and try reading the block from
1330                          * other devices.  When we find one, we re-write
1331                          * and check it that fixes the read error.
1332                          * This is all done synchronously while the array is
1333                          * frozen
1334                          */
1335                         sector_t sect = r1_bio->sector;
1336                         int sectors = r1_bio->sectors;
1337                         freeze_array(conf);
1338                         while(sectors) {
1339                                 int s = sectors;
1340                                 int d = r1_bio->read_disk;
1341                                 int success = 0;
1342
1343                                 if (s > (PAGE_SIZE>>9))
1344                                         s = PAGE_SIZE >> 9;
1345
1346                                 do {
1347                                         rdev = conf->mirrors[d].rdev;
1348                                         if (rdev &&
1349                                             test_bit(In_sync, &rdev->flags) &&
1350                                             sync_page_io(rdev->bdev,
1351                                                          sect + rdev->data_offset,
1352                                                          s<<9,
1353                                                          conf->tmppage, READ))
1354                                                 success = 1;
1355                                         else {
1356                                                 d++;
1357                                                 if (d == conf->raid_disks)
1358                                                         d = 0;
1359                                         }
1360                                 } while (!success && d != r1_bio->read_disk);
1361
1362                                 if (success) {
1363                                         /* write it back and re-read */
1364                                         while (d != r1_bio->read_disk) {
1365                                                 if (d==0)
1366                                                         d = conf->raid_disks;
1367                                                 d--;
1368                                                 rdev = conf->mirrors[d].rdev;
1369                                                 if (rdev &&
1370                                                     test_bit(In_sync, &rdev->flags)) {
1371                                                         if (sync_page_io(rdev->bdev,
1372                                                                          sect + rdev->data_offset,
1373                                                                          s<<9, conf->tmppage, WRITE) == 0 ||
1374                                                             sync_page_io(rdev->bdev,
1375                                                                          sect + rdev->data_offset,
1376                                                                          s<<9, conf->tmppage, READ) == 0) {
1377                                                                 /* Well, this device is dead */
1378                                                                 md_error(mddev, rdev);
1379                                                         }
1380                                                 }
1381                                         }
1382                                 } else {
1383                                         /* Cannot read from anywhere -- bye bye array */
1384                                         md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1385                                         break;
1386                                 }
1387                                 sectors -= s;
1388                                 sect += s;
1389                         }
1390
1391
1392                         unfreeze_array(conf);
1393
1394                         bio = r1_bio->bios[r1_bio->read_disk];
1395                         if ((disk=read_balance(conf, r1_bio)) == -1) {
1396                                 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1397                                        " read error for block %llu\n",
1398                                        bdevname(bio->bi_bdev,b),
1399                                        (unsigned long long)r1_bio->sector);
1400                                 raid_end_bio_io(r1_bio);
1401                         } else {
1402                                 r1_bio->bios[r1_bio->read_disk] = NULL;
1403                                 r1_bio->read_disk = disk;
1404                                 bio_put(bio);
1405                                 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1406                                 r1_bio->bios[r1_bio->read_disk] = bio;
1407                                 rdev = conf->mirrors[disk].rdev;
1408                                 if (printk_ratelimit())
1409                                         printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1410                                                " another mirror\n",
1411                                                bdevname(rdev->bdev,b),
1412                                                (unsigned long long)r1_bio->sector);
1413                                 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1414                                 bio->bi_bdev = rdev->bdev;
1415                                 bio->bi_end_io = raid1_end_read_request;
1416                                 bio->bi_rw = READ;
1417                                 bio->bi_private = r1_bio;
1418                                 unplug = 1;
1419                                 generic_make_request(bio);
1420                         }
1421                 }
1422         }
1423         spin_unlock_irqrestore(&conf->device_lock, flags);
1424         if (unplug)
1425                 unplug_slaves(mddev);
1426 }
1427
1428
1429 static int init_resync(conf_t *conf)
1430 {
1431         int buffs;
1432
1433         buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1434         if (conf->r1buf_pool)
1435                 BUG();
1436         conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1437                                           conf->poolinfo);
1438         if (!conf->r1buf_pool)
1439                 return -ENOMEM;
1440         conf->next_resync = 0;
1441         return 0;
1442 }
1443
1444 /*
1445  * perform a "sync" on one "block"
1446  *
1447  * We need to make sure that no normal I/O request - particularly write
1448  * requests - conflict with active sync requests.
1449  *
1450  * This is achieved by tracking pending requests and a 'barrier' concept
1451  * that can be installed to exclude normal IO requests.
1452  */
1453
1454 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1455 {
1456         conf_t *conf = mddev_to_conf(mddev);
1457         r1bio_t *r1_bio;
1458         struct bio *bio;
1459         sector_t max_sector, nr_sectors;
1460         int disk = -1;
1461         int i;
1462         int wonly = -1;
1463         int write_targets = 0, read_targets = 0;
1464         int sync_blocks;
1465         int still_degraded = 0;
1466
1467         if (!conf->r1buf_pool)
1468         {
1469 /*
1470                 printk("sync start - bitmap %p\n", mddev->bitmap);
1471 */
1472                 if (init_resync(conf))
1473                         return 0;
1474         }
1475
1476         max_sector = mddev->size << 1;
1477         if (sector_nr >= max_sector) {
1478                 /* If we aborted, we need to abort the
1479                  * sync on the 'current' bitmap chunk (there will
1480                  * only be one in raid1 resync.
1481                  * We can find the current addess in mddev->curr_resync
1482                  */
1483                 if (mddev->curr_resync < max_sector) /* aborted */
1484                         bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1485                                                 &sync_blocks, 1);
1486                 else /* completed sync */
1487                         conf->fullsync = 0;
1488
1489                 bitmap_close_sync(mddev->bitmap);
1490                 close_sync(conf);
1491                 return 0;
1492         }
1493
1494         /* before building a request, check if we can skip these blocks..
1495          * This call the bitmap_start_sync doesn't actually record anything
1496          */
1497         if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1498             !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1499                 /* We can skip this block, and probably several more */
1500                 *skipped = 1;
1501                 return sync_blocks;
1502         }
1503         /*
1504          * If there is non-resync activity waiting for a turn,
1505          * and resync is going fast enough,
1506          * then let it though before starting on this new sync request.
1507          */
1508         if (!go_faster && conf->nr_waiting)
1509                 msleep_interruptible(1000);
1510
1511         raise_barrier(conf);
1512
1513         conf->next_resync = sector_nr;
1514
1515         r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1516         rcu_read_lock();
1517         /*
1518          * If we get a correctably read error during resync or recovery,
1519          * we might want to read from a different device.  So we
1520          * flag all drives that could conceivably be read from for READ,
1521          * and any others (which will be non-In_sync devices) for WRITE.
1522          * If a read fails, we try reading from something else for which READ
1523          * is OK.
1524          */
1525
1526         r1_bio->mddev = mddev;
1527         r1_bio->sector = sector_nr;
1528         r1_bio->state = 0;
1529         set_bit(R1BIO_IsSync, &r1_bio->state);
1530
1531         for (i=0; i < conf->raid_disks; i++) {
1532                 mdk_rdev_t *rdev;
1533                 bio = r1_bio->bios[i];
1534
1535                 /* take from bio_init */
1536                 bio->bi_next = NULL;
1537                 bio->bi_flags |= 1 << BIO_UPTODATE;
1538                 bio->bi_rw = 0;
1539                 bio->bi_vcnt = 0;
1540                 bio->bi_idx = 0;
1541                 bio->bi_phys_segments = 0;
1542                 bio->bi_hw_segments = 0;
1543                 bio->bi_size = 0;
1544                 bio->bi_end_io = NULL;
1545                 bio->bi_private = NULL;
1546
1547                 rdev = rcu_dereference(conf->mirrors[i].rdev);
1548                 if (rdev == NULL ||
1549                            test_bit(Faulty, &rdev->flags)) {
1550                         still_degraded = 1;
1551                         continue;
1552                 } else if (!test_bit(In_sync, &rdev->flags)) {
1553                         bio->bi_rw = WRITE;
1554                         bio->bi_end_io = end_sync_write;
1555                         write_targets ++;
1556                 } else {
1557                         /* may need to read from here */
1558                         bio->bi_rw = READ;
1559                         bio->bi_end_io = end_sync_read;
1560                         if (test_bit(WriteMostly, &rdev->flags)) {
1561                                 if (wonly < 0)
1562                                         wonly = i;
1563                         } else {
1564                                 if (disk < 0)
1565                                         disk = i;
1566                         }
1567                         read_targets++;
1568                 }
1569                 atomic_inc(&rdev->nr_pending);
1570                 bio->bi_sector = sector_nr + rdev->data_offset;
1571                 bio->bi_bdev = rdev->bdev;
1572                 bio->bi_private = r1_bio;
1573         }
1574         rcu_read_unlock();
1575         if (disk < 0)
1576                 disk = wonly;
1577         r1_bio->read_disk = disk;
1578
1579         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1580                 /* extra read targets are also write targets */
1581                 write_targets += read_targets-1;
1582
1583         if (write_targets == 0 || read_targets == 0) {
1584                 /* There is nowhere to write, so all non-sync
1585                  * drives must be failed - so we are finished
1586                  */
1587                 sector_t rv = max_sector - sector_nr;
1588                 *skipped = 1;
1589                 put_buf(r1_bio);
1590                 return rv;
1591         }
1592
1593         nr_sectors = 0;
1594         sync_blocks = 0;
1595         do {
1596                 struct page *page;
1597                 int len = PAGE_SIZE;
1598                 if (sector_nr + (len>>9) > max_sector)
1599                         len = (max_sector - sector_nr) << 9;
1600                 if (len == 0)
1601                         break;
1602                 if (sync_blocks == 0) {
1603                         if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1604                                                &sync_blocks, still_degraded) &&
1605                             !conf->fullsync &&
1606                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1607                                 break;
1608                         if (sync_blocks < (PAGE_SIZE>>9))
1609                                 BUG();
1610                         if (len > (sync_blocks<<9))
1611                                 len = sync_blocks<<9;
1612                 }
1613
1614                 for (i=0 ; i < conf->raid_disks; i++) {
1615                         bio = r1_bio->bios[i];
1616                         if (bio->bi_end_io) {
1617                                 page = r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page;
1618                                 if (bio_add_page(bio, page, len, 0) == 0) {
1619                                         /* stop here */
1620                                         r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page = page;
1621                                         while (i > 0) {
1622                                                 i--;
1623                                                 bio = r1_bio->bios[i];
1624                                                 if (bio->bi_end_io==NULL)
1625                                                         continue;
1626                                                 /* remove last page from this bio */
1627                                                 bio->bi_vcnt--;
1628                                                 bio->bi_size -= len;
1629                                                 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1630                                         }
1631                                         goto bio_full;
1632                                 }
1633                         }
1634                 }
1635                 nr_sectors += len>>9;
1636                 sector_nr += len>>9;
1637                 sync_blocks -= (len>>9);
1638         } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1639  bio_full:
1640         bio = r1_bio->bios[r1_bio->read_disk];
1641         r1_bio->sectors = nr_sectors;
1642
1643         md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev, nr_sectors);
1644
1645         generic_make_request(bio);
1646
1647         return nr_sectors;
1648 }
1649
1650 static int run(mddev_t *mddev)
1651 {
1652         conf_t *conf;
1653         int i, j, disk_idx;
1654         mirror_info_t *disk;
1655         mdk_rdev_t *rdev;
1656         struct list_head *tmp;
1657
1658         if (mddev->level != 1) {
1659                 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1660                        mdname(mddev), mddev->level);
1661                 goto out;
1662         }
1663         /*
1664          * copy the already verified devices into our private RAID1
1665          * bookkeeping area. [whatever we allocate in run(),
1666          * should be freed in stop()]
1667          */
1668         conf = kmalloc(sizeof(conf_t), GFP_KERNEL);
1669         mddev->private = conf;
1670         if (!conf)
1671                 goto out_no_mem;
1672
1673         memset(conf, 0, sizeof(*conf));
1674         conf->mirrors = kmalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1675                                  GFP_KERNEL);
1676         if (!conf->mirrors)
1677                 goto out_no_mem;
1678
1679         memset(conf->mirrors, 0, sizeof(struct mirror_info)*mddev->raid_disks);
1680
1681         conf->tmppage = alloc_page(GFP_KERNEL);
1682         if (!conf->tmppage)
1683                 goto out_no_mem;
1684
1685         conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1686         if (!conf->poolinfo)
1687                 goto out_no_mem;
1688         conf->poolinfo->mddev = mddev;
1689         conf->poolinfo->raid_disks = mddev->raid_disks;
1690         conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1691                                           r1bio_pool_free,
1692                                           conf->poolinfo);
1693         if (!conf->r1bio_pool)
1694                 goto out_no_mem;
1695
1696         ITERATE_RDEV(mddev, rdev, tmp) {
1697                 disk_idx = rdev->raid_disk;
1698                 if (disk_idx >= mddev->raid_disks
1699                     || disk_idx < 0)
1700                         continue;
1701                 disk = conf->mirrors + disk_idx;
1702
1703                 disk->rdev = rdev;
1704
1705                 blk_queue_stack_limits(mddev->queue,
1706                                        rdev->bdev->bd_disk->queue);
1707                 /* as we don't honour merge_bvec_fn, we must never risk
1708                  * violating it, so limit ->max_sector to one PAGE, as
1709                  * a one page request is never in violation.
1710                  */
1711                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1712                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
1713                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1714
1715                 disk->head_position = 0;
1716                 if (!test_bit(Faulty, &rdev->flags) && test_bit(In_sync, &rdev->flags))
1717                         conf->working_disks++;
1718         }
1719         conf->raid_disks = mddev->raid_disks;
1720         conf->mddev = mddev;
1721         spin_lock_init(&conf->device_lock);
1722         INIT_LIST_HEAD(&conf->retry_list);
1723         if (conf->working_disks == 1)
1724                 mddev->recovery_cp = MaxSector;
1725
1726         spin_lock_init(&conf->resync_lock);
1727         init_waitqueue_head(&conf->wait_barrier);
1728
1729         bio_list_init(&conf->pending_bio_list);
1730         bio_list_init(&conf->flushing_bio_list);
1731
1732         if (!conf->working_disks) {
1733                 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1734                         mdname(mddev));
1735                 goto out_free_conf;
1736         }
1737
1738         mddev->degraded = 0;
1739         for (i = 0; i < conf->raid_disks; i++) {
1740
1741                 disk = conf->mirrors + i;
1742
1743                 if (!disk->rdev) {
1744                         disk->head_position = 0;
1745                         mddev->degraded++;
1746                 }
1747         }
1748
1749         /*
1750          * find the first working one and use it as a starting point
1751          * to read balancing.
1752          */
1753         for (j = 0; j < conf->raid_disks &&
1754                      (!conf->mirrors[j].rdev ||
1755                       !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
1756                 /* nothing */;
1757         conf->last_used = j;
1758
1759
1760         mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1761         if (!mddev->thread) {
1762                 printk(KERN_ERR
1763                        "raid1: couldn't allocate thread for %s\n",
1764                        mdname(mddev));
1765                 goto out_free_conf;
1766         }
1767
1768         printk(KERN_INFO
1769                 "raid1: raid set %s active with %d out of %d mirrors\n",
1770                 mdname(mddev), mddev->raid_disks - mddev->degraded,
1771                 mddev->raid_disks);
1772         /*
1773          * Ok, everything is just fine now
1774          */
1775         mddev->array_size = mddev->size;
1776
1777         mddev->queue->unplug_fn = raid1_unplug;
1778         mddev->queue->issue_flush_fn = raid1_issue_flush;
1779
1780         return 0;
1781
1782 out_no_mem:
1783         printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1784                mdname(mddev));
1785
1786 out_free_conf:
1787         if (conf) {
1788                 if (conf->r1bio_pool)
1789                         mempool_destroy(conf->r1bio_pool);
1790                 kfree(conf->mirrors);
1791                 __free_page(conf->tmppage);
1792                 kfree(conf->poolinfo);
1793                 kfree(conf);
1794                 mddev->private = NULL;
1795         }
1796 out:
1797         return -EIO;
1798 }
1799
1800 static int stop(mddev_t *mddev)
1801 {
1802         conf_t *conf = mddev_to_conf(mddev);
1803         struct bitmap *bitmap = mddev->bitmap;
1804         int behind_wait = 0;
1805
1806         /* wait for behind writes to complete */
1807         while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1808                 behind_wait++;
1809                 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
1810                 set_current_state(TASK_UNINTERRUPTIBLE);
1811                 schedule_timeout(HZ); /* wait a second */
1812                 /* need to kick something here to make sure I/O goes? */
1813         }
1814
1815         md_unregister_thread(mddev->thread);
1816         mddev->thread = NULL;
1817         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
1818         if (conf->r1bio_pool)
1819                 mempool_destroy(conf->r1bio_pool);
1820         kfree(conf->mirrors);
1821         kfree(conf->poolinfo);
1822         kfree(conf);
1823         mddev->private = NULL;
1824         return 0;
1825 }
1826
1827 static int raid1_resize(mddev_t *mddev, sector_t sectors)
1828 {
1829         /* no resync is happening, and there is enough space
1830          * on all devices, so we can resize.
1831          * We need to make sure resync covers any new space.
1832          * If the array is shrinking we should possibly wait until
1833          * any io in the removed space completes, but it hardly seems
1834          * worth it.
1835          */
1836         mddev->array_size = sectors>>1;
1837         set_capacity(mddev->gendisk, mddev->array_size << 1);
1838         mddev->changed = 1;
1839         if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
1840                 mddev->recovery_cp = mddev->size << 1;
1841                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1842         }
1843         mddev->size = mddev->array_size;
1844         mddev->resync_max_sectors = sectors;
1845         return 0;
1846 }
1847
1848 static int raid1_reshape(mddev_t *mddev, int raid_disks)
1849 {
1850         /* We need to:
1851          * 1/ resize the r1bio_pool
1852          * 2/ resize conf->mirrors
1853          *
1854          * We allocate a new r1bio_pool if we can.
1855          * Then raise a device barrier and wait until all IO stops.
1856          * Then resize conf->mirrors and swap in the new r1bio pool.
1857          *
1858          * At the same time, we "pack" the devices so that all the missing
1859          * devices have the higher raid_disk numbers.
1860          */
1861         mempool_t *newpool, *oldpool;
1862         struct pool_info *newpoolinfo;
1863         mirror_info_t *newmirrors;
1864         conf_t *conf = mddev_to_conf(mddev);
1865         int cnt;
1866
1867         int d, d2;
1868
1869         if (raid_disks < conf->raid_disks) {
1870                 cnt=0;
1871                 for (d= 0; d < conf->raid_disks; d++)
1872                         if (conf->mirrors[d].rdev)
1873                                 cnt++;
1874                 if (cnt > raid_disks)
1875                         return -EBUSY;
1876         }
1877
1878         newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
1879         if (!newpoolinfo)
1880                 return -ENOMEM;
1881         newpoolinfo->mddev = mddev;
1882         newpoolinfo->raid_disks = raid_disks;
1883
1884         newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1885                                  r1bio_pool_free, newpoolinfo);
1886         if (!newpool) {
1887                 kfree(newpoolinfo);
1888                 return -ENOMEM;
1889         }
1890         newmirrors = kmalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
1891         if (!newmirrors) {
1892                 kfree(newpoolinfo);
1893                 mempool_destroy(newpool);
1894                 return -ENOMEM;
1895         }
1896         memset(newmirrors, 0, sizeof(struct mirror_info)*raid_disks);
1897
1898         raise_barrier(conf);
1899
1900         /* ok, everything is stopped */
1901         oldpool = conf->r1bio_pool;
1902         conf->r1bio_pool = newpool;
1903
1904         for (d=d2=0; d < conf->raid_disks; d++)
1905                 if (conf->mirrors[d].rdev) {
1906                         conf->mirrors[d].rdev->raid_disk = d2;
1907                         newmirrors[d2++].rdev = conf->mirrors[d].rdev;
1908                 }
1909         kfree(conf->mirrors);
1910         conf->mirrors = newmirrors;
1911         kfree(conf->poolinfo);
1912         conf->poolinfo = newpoolinfo;
1913
1914         mddev->degraded += (raid_disks - conf->raid_disks);
1915         conf->raid_disks = mddev->raid_disks = raid_disks;
1916
1917         conf->last_used = 0; /* just make sure it is in-range */
1918         lower_barrier(conf);
1919
1920         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1921         md_wakeup_thread(mddev->thread);
1922
1923         mempool_destroy(oldpool);
1924         return 0;
1925 }
1926
1927 static void raid1_quiesce(mddev_t *mddev, int state)
1928 {
1929         conf_t *conf = mddev_to_conf(mddev);
1930
1931         switch(state) {
1932         case 1:
1933                 raise_barrier(conf);
1934                 break;
1935         case 0:
1936                 lower_barrier(conf);
1937                 break;
1938         }
1939 }
1940
1941
1942 static mdk_personality_t raid1_personality =
1943 {
1944         .name           = "raid1",
1945         .owner          = THIS_MODULE,
1946         .make_request   = make_request,
1947         .run            = run,
1948         .stop           = stop,
1949         .status         = status,
1950         .error_handler  = error,
1951         .hot_add_disk   = raid1_add_disk,
1952         .hot_remove_disk= raid1_remove_disk,
1953         .spare_active   = raid1_spare_active,
1954         .sync_request   = sync_request,
1955         .resize         = raid1_resize,
1956         .reshape        = raid1_reshape,
1957         .quiesce        = raid1_quiesce,
1958 };
1959
1960 static int __init raid_init(void)
1961 {
1962         return register_md_personality(RAID1, &raid1_personality);
1963 }
1964
1965 static void raid_exit(void)
1966 {
1967         unregister_md_personality(RAID1);
1968 }
1969
1970 module_init(raid_init);
1971 module_exit(raid_exit);
1972 MODULE_LICENSE("GPL");
1973 MODULE_ALIAS("md-personality-3"); /* RAID1 */