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  * This program is free software; you can redistribute it and/or modify
  16  * it under the terms of the GNU General Public License as published by
  17  * the Free Software Foundation; either version 2, or (at your option)
  18  * any later version.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * (for example /usr/src/linux/COPYING); if not, write to the Free
  22  * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23  */
  24
  25 #include <linux/raid/raid1.h>
  26
  27 /*
  28  * Number of guaranteed r1bios in case of extreme VM load:
  29  */
  30 #define NR_RAID1_BIOS 256
  31
  32 static mdk_personality_t raid1_personality;
  33
  34 static void unplug_slaves(mddev_t *mddev);
  35
  36
  37 static void * r1bio_pool_alloc(int gfp_flags, void *data)
  38 {
  39         struct pool_info *pi = data;
  40         r1bio_t *r1_bio;
  41         int size = offsetof(r1bio_t, bios[pi->raid_disks]);
  42
  43         /* allocate a r1bio with room for raid_disks entries in the bios array */
  44         r1_bio = kmalloc(size, gfp_flags);
  45         if (r1_bio)
  46                 memset(r1_bio, 0, size);
  47         else
  48                 unplug_slaves(pi->mddev);
  49
  50         return r1_bio;
  51 }
  52
  53 static void r1bio_pool_free(void *r1_bio, void *data)
  54 {
  55         kfree(r1_bio);
  56 }
  57
  58 #define RESYNC_BLOCK_SIZE (64*1024)
  59 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
  60 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
  61 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
  62 #define RESYNC_WINDOW (2048*1024)
  63
  64 static void * r1buf_pool_alloc(int gfp_flags, void *data)
  65 {
  66         struct pool_info *pi = data;
  67         struct page *page;
  68         r1bio_t *r1_bio;
  69         struct bio *bio;
  70         int i, j;
  71
  72         r1_bio = r1bio_pool_alloc(gfp_flags, pi);
  73         if (!r1_bio) {
  74                 unplug_slaves(pi->mddev);
  75                 return NULL;
  76         }
  77
  78         /*
  79          * Allocate bios : 1 for reading, n-1 for writing
  80          */
  81         for (j = pi->raid_disks ; j-- ; ) {
  82                 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
  83                 if (!bio)
  84                         goto out_free_bio;
  85                 r1_bio->bios[j] = bio;
  86         }
  87         /*
  88          * Allocate RESYNC_PAGES data pages and attach them to
  89          * the first bio;
  90          */
  91         bio = r1_bio->bios[0];
  92         for (i = 0; i < RESYNC_PAGES; i++) {
  93                 page = alloc_page(gfp_flags);
  94                 if (unlikely(!page))
  95                         goto out_free_pages;
  96
  97                 bio->bi_io_vec[i].bv_page = page;
  98         }
  99
 100         r1_bio->master_bio = NULL;
 101
 102         return r1_bio;
 103
 104 out_free_pages:
 105         for ( ; i > 0 ; i--)
 106                 __free_page(bio->bi_io_vec[i-1].bv_page);
 107 out_free_bio:
 108         while ( ++j < pi->raid_disks )
 109                 bio_put(r1_bio->bios[j]);
 110         r1bio_pool_free(r1_bio, data);
 111         return NULL;
 112 }
 113
 114 static void r1buf_pool_free(void *__r1_bio, void *data)
 115 {
 116         struct pool_info *pi = data;
 117         int i;
 118         r1bio_t *r1bio = __r1_bio;
 119         struct bio *bio = r1bio->bios[0];
 120
 121         for (i = 0; i < RESYNC_PAGES; i++) {
 122                 __free_page(bio->bi_io_vec[i].bv_page);
 123                 bio->bi_io_vec[i].bv_page = NULL;
 124         }
 125         for (i=0 ; i < pi->raid_disks; i++)
 126                 bio_put(r1bio->bios[i]);
 127
 128         r1bio_pool_free(r1bio, data);
 129 }
 130
 131 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
 132 {
 133         int i;
 134
 135         for (i = 0; i < conf->raid_disks; i++) {
 136                 struct bio **bio = r1_bio->bios + i;
 137                 if (*bio)
 138                         bio_put(*bio);
 139                 *bio = NULL;
 140         }
 141 }
 142
 143 static inline void free_r1bio(r1bio_t *r1_bio)
 144 {
 145         unsigned long flags;
 146
 147         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 148
 149         /*
 150          * Wake up any possible resync thread that waits for the device
 151          * to go idle.
 152          */
 153         spin_lock_irqsave(&conf->resync_lock, flags);
 154         if (!--conf->nr_pending) {
 155                 wake_up(&conf->wait_idle);
 156                 wake_up(&conf->wait_resume);
 157         }
 158         spin_unlock_irqrestore(&conf->resync_lock, flags);
 159
 160         put_all_bios(conf, r1_bio);
 161         mempool_free(r1_bio, conf->r1bio_pool);
 162 }
 163
 164 static inline void put_buf(r1bio_t *r1_bio)
 165 {
 166         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 167         unsigned long flags;
 168
 169         mempool_free(r1_bio, conf->r1buf_pool);
 170
 171         spin_lock_irqsave(&conf->resync_lock, flags);
 172         if (!conf->barrier)
 173                 BUG();
 174         --conf->barrier;
 175         wake_up(&conf->wait_resume);
 176         wake_up(&conf->wait_idle);
 177
 178         if (!--conf->nr_pending) {
 179                 wake_up(&conf->wait_idle);
 180                 wake_up(&conf->wait_resume);
 181         }
 182         spin_unlock_irqrestore(&conf->resync_lock, flags);
 183 }
 184
 185 static void reschedule_retry(r1bio_t *r1_bio)
 186 {
 187         unsigned long flags;
 188         mddev_t *mddev = r1_bio->mddev;
 189         conf_t *conf = mddev_to_conf(mddev);
 190
 191         spin_lock_irqsave(&conf->device_lock, flags);
 192         list_add(&r1_bio->retry_list, &conf->retry_list);
 193         spin_unlock_irqrestore(&conf->device_lock, flags);
 194
 195         md_wakeup_thread(mddev->thread);
 196 }
 197
 198 /*
 199  * raid_end_bio_io() is called when we have finished servicing a mirrored
 200  * operation and are ready to return a success/failure code to the buffer
 201  * cache layer.
 202  */
 203 static void raid_end_bio_io(r1bio_t *r1_bio)
 204 {
 205         struct bio *bio = r1_bio->master_bio;
 206
 207         bio_endio(bio, bio->bi_size,
 208                 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
 209         free_r1bio(r1_bio);
 210 }
 211
 212 /*
 213  * Update disk head position estimator based on IRQ completion info.
 214  */
 215 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
 216 {
 217         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 218
 219         conf->mirrors[disk].head_position =
 220                 r1_bio->sector + (r1_bio->sectors);
 221 }
 222
 223 static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
 224 {
 225         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 226         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 227         int mirror;
 228         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 229
 230         if (bio->bi_size)
 231                 return 1;
 232
 233         mirror = r1_bio->read_disk;
 234         /*
 235          * this branch is our 'one mirror IO has finished' event handler:
 236          */
 237         if (!uptodate)
 238                 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
 239         else
 240                 /*
 241                  * Set R1BIO_Uptodate in our master bio, so that
 242                  * we will return a good error code for to the higher
 243                  * levels even if IO on some other mirrored buffer fails.
 244                  *
 245                  * The 'master' represents the composite IO operation to
 246                  * user-side. So if something waits for IO, then it will
 247                  * wait for the 'master' bio.
 248                  */
 249                 set_bit(R1BIO_Uptodate, &r1_bio->state);
 250
 251         update_head_pos(mirror, r1_bio);
 252
 253         /*
 254          * we have only one bio on the read side
 255          */
 256         if (uptodate)
 257                 raid_end_bio_io(r1_bio);
 258         else {
 259                 /*
 260                  * oops, read error:
 261                  */
 262                 char b[BDEVNAME_SIZE];
 263                 if (printk_ratelimit())
 264                         printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
 265                                bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
 266                 reschedule_retry(r1_bio);
 267         }
 268
 269         rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 270         return 0;
 271 }
 272
 273 static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
 274 {
 275         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 276         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 277         int mirror;
 278         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 279
 280         if (bio->bi_size)
 281                 return 1;
 282
 283         for (mirror = 0; mirror < conf->raid_disks; mirror++)
 284                 if (r1_bio->bios[mirror] == bio)
 285                         break;
 286
 287         /*
 288          * this branch is our 'one mirror IO has finished' event handler:
 289          */
 290         if (!uptodate)
 291                 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
 292         else
 293                 /*
 294                  * Set R1BIO_Uptodate in our master bio, so that
 295                  * we will return a good error code for to the higher
 296                  * levels even if IO on some other mirrored buffer fails.
 297                  *
 298                  * The 'master' represents the composite IO operation to
 299                  * user-side. So if something waits for IO, then it will
 300                  * wait for the 'master' bio.
 301                  */
 302                 set_bit(R1BIO_Uptodate, &r1_bio->state);
 303
 304         update_head_pos(mirror, r1_bio);
 305
 306         /*
 307          *
 308          * Let's see if all mirrored write operations have finished
 309          * already.
 310          */
 311         if (atomic_dec_and_test(&r1_bio->remaining)) {
 312                 md_write_end(r1_bio->mddev);
 313                 raid_end_bio_io(r1_bio);
 314         }
 315
 316         rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 317         return 0;
 318 }
 319
 320
 321 /*
 322  * This routine returns the disk from which the requested read should
 323  * be done. There is a per-array 'next expected sequential IO' sector
 324  * number - if this matches on the next IO then we use the last disk.
 325  * There is also a per-disk 'last know head position' sector that is
 326  * maintained from IRQ contexts, both the normal and the resync IO
 327  * completion handlers update this position correctly. If there is no
 328  * perfect sequential match then we pick the disk whose head is closest.
 329  *
 330  * If there are 2 mirrors in the same 2 devices, performance degrades
 331  * because position is mirror, not device based.
 332  *
 333  * The rdev for the device selected will have nr_pending incremented.
 334  */
 335 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
 336 {
 337         const unsigned long this_sector = r1_bio->sector;
 338         int new_disk = conf->last_used, disk = new_disk;
 339         const int sectors = r1_bio->sectors;
 340         sector_t new_distance, current_distance;
 341
 342         rcu_read_lock();
 343         /*
 344          * Check if it if we can balance. We can balance on the whole
 345          * device if no resync is going on, or below the resync window.
 346          * We take the first readable disk when above the resync window.
 347          */
 348         if (conf->mddev->recovery_cp < MaxSector &&
 349             (this_sector + sectors >= conf->next_resync)) {
 350                 /* Choose the first operation device, for consistancy */
 351                 new_disk = 0;
 352
 353                 while (!conf->mirrors[new_disk].rdev ||
 354                        !conf->mirrors[new_disk].rdev->in_sync) {
 355                         new_disk++;
 356                         if (new_disk == conf->raid_disks) {
 357                                 new_disk = -1;
 358                                 break;
 359                         }
 360                 }
 361                 goto rb_out;
 362         }
 363
 364
 365         /* make sure the disk is operational */
 366         while (!conf->mirrors[new_disk].rdev ||
 367                !conf->mirrors[new_disk].rdev->in_sync) {
 368                 if (new_disk <= 0)
 369                         new_disk = conf->raid_disks;
 370                 new_disk--;
 371                 if (new_disk == disk) {
 372                         new_disk = -1;
 373                         goto rb_out;
 374                 }
 375         }
 376         disk = new_disk;
 377         /* now disk == new_disk == starting point for search */
 378
 379         /*
 380          * Don't change to another disk for sequential reads:
 381          */
 382         if (conf->next_seq_sect == this_sector)
 383                 goto rb_out;
 384         if (this_sector == conf->mirrors[new_disk].head_position)
 385                 goto rb_out;
 386
 387         current_distance = abs(this_sector - conf->mirrors[disk].head_position);
 388
 389         /* Find the disk whose head is closest */
 390
 391         do {
 392                 if (disk <= 0)
 393                         disk = conf->raid_disks;
 394                 disk--;
 395
 396                 if (!conf->mirrors[disk].rdev ||
 397                     !conf->mirrors[disk].rdev->in_sync)
 398                         continue;
 399
 400                 if (!atomic_read(&conf->mirrors[disk].rdev->nr_pending)) {
 401                         new_disk = disk;
 402                         break;
 403                 }
 404                 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
 405                 if (new_distance < current_distance) {
 406                         current_distance = new_distance;
 407                         new_disk = disk;
 408                 }
 409         } while (disk != conf->last_used);
 410
 411 rb_out:
 412
 413
 414         if (new_disk >= 0) {
 415                 conf->next_seq_sect = this_sector + sectors;
 416                 conf->last_used = new_disk;
 417                 atomic_inc(&conf->mirrors[new_disk].rdev->nr_pending);
 418         }
 419         rcu_read_unlock();
 420
 421         return new_disk;
 422 }
 423
 424 static void unplug_slaves(mddev_t *mddev)
 425 {
 426         conf_t *conf = mddev_to_conf(mddev);
 427         int i;
 428
 429         rcu_read_lock();
 430         for (i=0; i<mddev->raid_disks; i++) {
 431                 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
 432                 if (rdev && !rdev->faulty && atomic_read(&rdev->nr_pending)) {
 433                         request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
 434
 435                         atomic_inc(&rdev->nr_pending);
 436                         rcu_read_unlock();
 437
 438                         if (r_queue->unplug_fn)
 439                                 r_queue->unplug_fn(r_queue);
 440
 441                         rdev_dec_pending(rdev, mddev);
 442                         rcu_read_lock();
 443                 }
 444         }
 445         rcu_read_unlock();
 446 }
 447
 448 static void raid1_unplug(request_queue_t *q)
 449 {
 450         unplug_slaves(q->queuedata);
 451 }
 452
 453 static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
 454                              sector_t *error_sector)
 455 {
 456         mddev_t *mddev = q->queuedata;
 457         conf_t *conf = mddev_to_conf(mddev);
 458         int i, ret = 0;
 459
 460         rcu_read_lock();
 461         for (i=0; i<mddev->raid_disks && ret == 0; i++) {
 462                 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
 463                 if (rdev && !rdev->faulty) {
 464                         struct block_device *bdev = rdev->bdev;
 465                         request_queue_t *r_queue = bdev_get_queue(bdev);
 466
 467                         if (!r_queue->issue_flush_fn)
 468                                 ret = -EOPNOTSUPP;
 469                         else {
 470                                 atomic_inc(&rdev->nr_pending);
 471                                 rcu_read_unlock();
 472                                 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
 473                                                               error_sector);
 474                                 rdev_dec_pending(rdev, mddev);
 475                                 rcu_read_lock();
 476                         }
 477                 }
 478         }
 479         rcu_read_unlock();
 480         return ret;
 481 }
 482
 483 /*
 484  * Throttle resync depth, so that we can both get proper overlapping of
 485  * requests, but are still able to handle normal requests quickly.
 486  */
 487 #define RESYNC_DEPTH 32
 488
 489 static void device_barrier(conf_t *conf, sector_t sect)
 490 {
 491         spin_lock_irq(&conf->resync_lock);
 492         wait_event_lock_irq(conf->wait_idle, !waitqueue_active(&conf->wait_resume),
 493                             conf->resync_lock, unplug_slaves(conf->mddev));
 494
 495         if (!conf->barrier++) {
 496                 wait_event_lock_irq(conf->wait_idle, !conf->nr_pending,
 497                                     conf->resync_lock, unplug_slaves(conf->mddev));
 498                 if (conf->nr_pending)
 499                         BUG();
 500         }
 501         wait_event_lock_irq(conf->wait_resume, conf->barrier < RESYNC_DEPTH,
 502                             conf->resync_lock, unplug_slaves(conf->mddev));
 503         conf->next_resync = sect;
 504         spin_unlock_irq(&conf->resync_lock);
 505 }
 506
 507 static int make_request(request_queue_t *q, struct bio * bio)
 508 {
 509         mddev_t *mddev = q->queuedata;
 510         conf_t *conf = mddev_to_conf(mddev);
 511         mirror_info_t *mirror;
 512         r1bio_t *r1_bio;
 513         struct bio *read_bio;
 514         int i, disks;
 515
 516         /*
 517          * Register the new request and wait if the reconstruction
 518          * thread has put up a bar for new requests.
 519          * Continue immediately if no resync is active currently.
 520          */
 521         spin_lock_irq(&conf->resync_lock);
 522         wait_event_lock_irq(conf->wait_resume, !conf->barrier, conf->resync_lock, );
 523         conf->nr_pending++;
 524         spin_unlock_irq(&conf->resync_lock);
 525
 526         if (bio_data_dir(bio)==WRITE) {
 527                 disk_stat_inc(mddev->gendisk, writes);
 528                 disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio));
 529         } else {
 530                 disk_stat_inc(mddev->gendisk, reads);
 531                 disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio));
 532         }
 533
 534         /*
 535          * make_request() can abort the operation when READA is being
 536          * used and no empty request is available.
 537          *
 538          */
 539         r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
 540
 541         r1_bio->master_bio = bio;
 542         r1_bio->sectors = bio->bi_size >> 9;
 543
 544         r1_bio->mddev = mddev;
 545         r1_bio->sector = bio->bi_sector;
 546
 547         r1_bio->state = 0;
 548
 549         if (bio_data_dir(bio) == READ) {
 550                 /*
 551                  * read balancing logic:
 552                  */
 553                 int rdisk = read_balance(conf, r1_bio);
 554
 555                 if (rdisk < 0) {
 556                         /* couldn't find anywhere to read from */
 557                         raid_end_bio_io(r1_bio);
 558                         return 0;
 559                 }
 560                 mirror = conf->mirrors + rdisk;
 561
 562                 r1_bio->read_disk = rdisk;
 563
 564                 read_bio = bio_clone(bio, GFP_NOIO);
 565
 566                 r1_bio->bios[rdisk] = read_bio;
 567
 568                 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
 569                 read_bio->bi_bdev = mirror->rdev->bdev;
 570                 read_bio->bi_end_io = raid1_end_read_request;
 571                 read_bio->bi_rw = READ;
 572                 read_bio->bi_private = r1_bio;
 573
 574                 generic_make_request(read_bio);
 575                 return 0;
 576         }
 577
 578         /*
 579          * WRITE:
 580          */
 581         /* first select target devices under spinlock and
 582          * inc refcount on their rdev.  Record them by setting
 583          * bios[x] to bio
 584          */
 585         disks = conf->raid_disks;
 586         rcu_read_lock();
 587         for (i = 0;  i < disks; i++) {
 588                 if (conf->mirrors[i].rdev &&
 589                     !conf->mirrors[i].rdev->faulty) {
 590                         atomic_inc(&conf->mirrors[i].rdev->nr_pending);
 591                         r1_bio->bios[i] = bio;
 592                 } else
 593                         r1_bio->bios[i] = NULL;
 594         }
 595         rcu_read_unlock();
 596
 597         atomic_set(&r1_bio->remaining, 1);
 598         md_write_start(mddev);
 599         for (i = 0; i < disks; i++) {
 600                 struct bio *mbio;
 601                 if (!r1_bio->bios[i])
 602                         continue;
 603
 604                 mbio = bio_clone(bio, GFP_NOIO);
 605                 r1_bio->bios[i] = mbio;
 606
 607                 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
 608                 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
 609                 mbio->bi_end_io = raid1_end_write_request;
 610                 mbio->bi_rw = WRITE;
 611                 mbio->bi_private = r1_bio;
 612
 613                 atomic_inc(&r1_bio->remaining);
 614                 generic_make_request(mbio);
 615         }
 616
 617         if (atomic_dec_and_test(&r1_bio->remaining)) {
 618                 md_write_end(mddev);
 619                 raid_end_bio_io(r1_bio);
 620         }
 621
 622         return 0;
 623 }
 624
 625 static void status(struct seq_file *seq, mddev_t *mddev)
 626 {
 627         conf_t *conf = mddev_to_conf(mddev);
 628         int i;
 629
 630         seq_printf(seq, " [%d/%d] [", conf->raid_disks,
 631                                                 conf->working_disks);
 632         for (i = 0; i < conf->raid_disks; i++)
 633                 seq_printf(seq, "%s",
 634                               conf->mirrors[i].rdev &&
 635                               conf->mirrors[i].rdev->in_sync ? "U" : "_");
 636         seq_printf(seq, "]");
 637 }
 638
 639
 640 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
 641 {
 642         char b[BDEVNAME_SIZE];
 643         conf_t *conf = mddev_to_conf(mddev);
 644
 645         /*
 646          * If it is not operational, then we have already marked it as dead
 647          * else if it is the last working disks, ignore the error, let the
 648          * next level up know.
 649          * else mark the drive as failed
 650          */
 651         if (rdev->in_sync
 652             && conf->working_disks == 1)
 653                 /*
 654                  * Don't fail the drive, act as though we were just a
 655                  * normal single drive
 656                  */
 657                 return;
 658         if (rdev->in_sync) {
 659                 mddev->degraded++;
 660                 conf->working_disks--;
 661                 /*
 662                  * if recovery is running, make sure it aborts.
 663                  */
 664                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
 665         }
 666         rdev->in_sync = 0;
 667         rdev->faulty = 1;
 668         mddev->sb_dirty = 1;
 669         printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
 670                 "       Operation continuing on %d devices\n",
 671                 bdevname(rdev->bdev,b), conf->working_disks);
 672 }
 673
 674 static void print_conf(conf_t *conf)
 675 {
 676         int i;
 677         mirror_info_t *tmp;
 678
 679         printk("RAID1 conf printout:\n");
 680         if (!conf) {
 681                 printk("(!conf)\n");
 682                 return;
 683         }
 684         printk(" --- wd:%d rd:%d\n", conf->working_disks,
 685                 conf->raid_disks);
 686
 687         for (i = 0; i < conf->raid_disks; i++) {
 688                 char b[BDEVNAME_SIZE];
 689                 tmp = conf->mirrors + i;
 690                 if (tmp->rdev)
 691                         printk(" disk %d, wo:%d, o:%d, dev:%s\n",
 692                                 i, !tmp->rdev->in_sync, !tmp->rdev->faulty,
 693                                 bdevname(tmp->rdev->bdev,b));
 694         }
 695 }
 696
 697 static void close_sync(conf_t *conf)
 698 {
 699         spin_lock_irq(&conf->resync_lock);
 700         wait_event_lock_irq(conf->wait_resume, !conf->barrier,
 701                             conf->resync_lock,  unplug_slaves(conf->mddev));
 702         spin_unlock_irq(&conf->resync_lock);
 703
 704         if (conf->barrier) BUG();
 705         if (waitqueue_active(&conf->wait_idle)) BUG();
 706
 707         mempool_destroy(conf->r1buf_pool);
 708         conf->r1buf_pool = NULL;
 709 }
 710
 711 static int raid1_spare_active(mddev_t *mddev)
 712 {
 713         int i;
 714         conf_t *conf = mddev->private;
 715         mirror_info_t *tmp;
 716
 717         /*
 718          * Find all failed disks within the RAID1 configuration
 719          * and mark them readable
 720          */
 721         for (i = 0; i < conf->raid_disks; i++) {
 722                 tmp = conf->mirrors + i;
 723                 if (tmp->rdev
 724                     && !tmp->rdev->faulty
 725                     && !tmp->rdev->in_sync) {
 726                         conf->working_disks++;
 727                         mddev->degraded--;
 728                         tmp->rdev->in_sync = 1;
 729                 }
 730         }
 731
 732         print_conf(conf);
 733         return 0;
 734 }
 735
 736
 737 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
 738 {
 739         conf_t *conf = mddev->private;
 740         int found = 0;
 741         int mirror;
 742         mirror_info_t *p;
 743
 744         for (mirror=0; mirror < mddev->raid_disks; mirror++)
 745                 if ( !(p=conf->mirrors+mirror)->rdev) {
 746
 747                         blk_queue_stack_limits(mddev->queue,
 748                                                rdev->bdev->bd_disk->queue);
 749                         /* as we don't honour merge_bvec_fn, we must never risk
 750                          * violating it, so limit ->max_sector to one PAGE, as
 751                          * a one page request is never in violation.
 752                          */
 753                         if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 754                             mddev->queue->max_sectors > (PAGE_SIZE>>9))
 755                                 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 756
 757                         p->head_position = 0;
 758                         rdev->raid_disk = mirror;
 759                         found = 1;
 760                         p->rdev = rdev;
 761                         break;
 762                 }
 763
 764         print_conf(conf);
 765         return found;
 766 }
 767
 768 static int raid1_remove_disk(mddev_t *mddev, int number)
 769 {
 770         conf_t *conf = mddev->private;
 771         int err = 0;
 772         mdk_rdev_t *rdev;
 773         mirror_info_t *p = conf->mirrors+ number;
 774
 775         print_conf(conf);
 776         rdev = p->rdev;
 777         if (rdev) {
 778                 if (rdev->in_sync ||
 779                     atomic_read(&rdev->nr_pending)) {
 780                         err = -EBUSY;
 781                         goto abort;
 782                 }
 783                 p->rdev = NULL;
 784                 synchronize_kernel();
 785                 if (atomic_read(&rdev->nr_pending)) {
 786                         /* lost the race, try later */
 787                         err = -EBUSY;
 788                         p->rdev = rdev;
 789                 }
 790         }
 791 abort:
 792
 793         print_conf(conf);
 794         return err;
 795 }
 796
 797
 798 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
 799 {
 800         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 801         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 802         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 803
 804         if (bio->bi_size)
 805                 return 1;
 806
 807         if (r1_bio->bios[r1_bio->read_disk] != bio)
 808                 BUG();
 809         update_head_pos(r1_bio->read_disk, r1_bio);
 810         /*
 811          * we have read a block, now it needs to be re-written,
 812          * or re-read if the read failed.
 813          * We don't do much here, just schedule handling by raid1d
 814          */
 815         if (!uptodate)
 816                 md_error(r1_bio->mddev,
 817                          conf->mirrors[r1_bio->read_disk].rdev);
 818         else
 819                 set_bit(R1BIO_Uptodate, &r1_bio->state);
 820         rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
 821         reschedule_retry(r1_bio);
 822         return 0;
 823 }
 824
 825 static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
 826 {
 827         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 828         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 829         mddev_t *mddev = r1_bio->mddev;
 830         conf_t *conf = mddev_to_conf(mddev);
 831         int i;
 832         int mirror=0;
 833
 834         if (bio->bi_size)
 835                 return 1;
 836
 837         for (i = 0; i < conf->raid_disks; i++)
 838                 if (r1_bio->bios[i] == bio) {
 839                         mirror = i;
 840                         break;
 841                 }
 842         if (!uptodate)
 843                 md_error(mddev, conf->mirrors[mirror].rdev);
 844         update_head_pos(mirror, r1_bio);
 845
 846         if (atomic_dec_and_test(&r1_bio->remaining)) {
 847                 md_done_sync(mddev, r1_bio->sectors, uptodate);
 848                 put_buf(r1_bio);
 849         }
 850         rdev_dec_pending(conf->mirrors[mirror].rdev, mddev);
 851         return 0;
 852 }
 853
 854 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
 855 {
 856         conf_t *conf = mddev_to_conf(mddev);
 857         int i;
 858         int disks = conf->raid_disks;
 859         struct bio *bio, *wbio;
 860
 861         bio = r1_bio->bios[r1_bio->read_disk];
 862
 863         /*
 864          * schedule writes
 865          */
 866         if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
 867                 /*
 868                  * There is no point trying a read-for-reconstruct as
 869                  * reconstruct is about to be aborted
 870                  */
 871                 char b[BDEVNAME_SIZE];
 872                 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
 873                         " for block %llu\n",
 874                         bdevname(bio->bi_bdev,b),
 875                         (unsigned long long)r1_bio->sector);
 876                 md_done_sync(mddev, r1_bio->sectors, 0);
 877                 put_buf(r1_bio);
 878                 return;
 879         }
 880
 881         atomic_set(&r1_bio->remaining, 1);
 882         for (i = 0; i < disks ; i++) {
 883                 wbio = r1_bio->bios[i];
 884                 if (wbio->bi_end_io != end_sync_write)
 885                         continue;
 886
 887                 atomic_inc(&conf->mirrors[i].rdev->nr_pending);
 888                 atomic_inc(&r1_bio->remaining);
 889                 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
 890                 generic_make_request(wbio);
 891         }
 892
 893         if (atomic_dec_and_test(&r1_bio->remaining)) {
 894                 md_done_sync(mddev, r1_bio->sectors, 1);
 895                 put_buf(r1_bio);
 896         }
 897 }
 898
 899 /*
 900  * This is a kernel thread which:
 901  *
 902  *      1.      Retries failed read operations on working mirrors.
 903  *      2.      Updates the raid superblock when problems encounter.
 904  *      3.      Performs writes following reads for array syncronising.
 905  */
 906
 907 static void raid1d(mddev_t *mddev)
 908 {
 909         r1bio_t *r1_bio;
 910         struct bio *bio;
 911         unsigned long flags;
 912         conf_t *conf = mddev_to_conf(mddev);
 913         struct list_head *head = &conf->retry_list;
 914         int unplug=0;
 915         mdk_rdev_t *rdev;
 916
 917         md_check_recovery(mddev);
 918         md_handle_safemode(mddev);
 919
 920         for (;;) {
 921                 char b[BDEVNAME_SIZE];
 922                 spin_lock_irqsave(&conf->device_lock, flags);
 923                 if (list_empty(head))
 924                         break;
 925                 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
 926                 list_del(head->prev);
 927                 spin_unlock_irqrestore(&conf->device_lock, flags);
 928
 929                 mddev = r1_bio->mddev;
 930                 conf = mddev_to_conf(mddev);
 931                 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
 932                         sync_request_write(mddev, r1_bio);
 933                         unplug = 1;
 934                 } else {
 935                         int disk;
 936                         bio = r1_bio->bios[r1_bio->read_disk];
 937                         if ((disk=read_balance(conf, r1_bio)) == -1) {
 938                                 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
 939                                        " read error for block %llu\n",
 940                                        bdevname(bio->bi_bdev,b),
 941                                        (unsigned long long)r1_bio->sector);
 942                                 raid_end_bio_io(r1_bio);
 943                         } else {
 944                                 r1_bio->bios[r1_bio->read_disk] = NULL;
 945                                 r1_bio->read_disk = disk;
 946                                 bio_put(bio);
 947                                 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
 948                                 r1_bio->bios[r1_bio->read_disk] = bio;
 949                                 rdev = conf->mirrors[disk].rdev;
 950                                 if (printk_ratelimit())
 951                                         printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
 952                                                " another mirror\n",
 953                                                bdevname(rdev->bdev,b),
 954                                                (unsigned long long)r1_bio->sector);
 955                                 bio->bi_sector = r1_bio->sector + rdev->data_offset;
 956                                 bio->bi_bdev = rdev->bdev;
 957                                 bio->bi_end_io = raid1_end_read_request;
 958                                 bio->bi_rw = READ;
 959                                 bio->bi_private = r1_bio;
 960                                 unplug = 1;
 961                                 generic_make_request(bio);
 962                         }
 963                 }
 964         }
 965         spin_unlock_irqrestore(&conf->device_lock, flags);
 966         if (unplug)
 967                 unplug_slaves(mddev);
 968 }
 969
 970
 971 static int init_resync(conf_t *conf)
 972 {
 973         int buffs;
 974
 975         buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
 976         if (conf->r1buf_pool)
 977                 BUG();
 978         conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
 979                                           conf->poolinfo);
 980         if (!conf->r1buf_pool)
 981                 return -ENOMEM;
 982         conf->next_resync = 0;
 983         return 0;
 984 }
 985
 986 /*
 987  * perform a "sync" on one "block"
 988  *
 989  * We need to make sure that no normal I/O request - particularly write
 990  * requests - conflict with active sync requests.
 991  *
 992  * This is achieved by tracking pending requests and a 'barrier' concept
 993  * that can be installed to exclude normal IO requests.
 994  */
 995
 996 static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster)
 997 {
 998         conf_t *conf = mddev_to_conf(mddev);
 999         mirror_info_t *mirror;
1000         r1bio_t *r1_bio;
1001         struct bio *bio;
1002         sector_t max_sector, nr_sectors;
1003         int disk;
1004         int i;
1005         int write_targets = 0;
1006
1007         if (!conf->r1buf_pool)
1008                 if (init_resync(conf))
1009                         return -ENOMEM;
1010
1011         max_sector = mddev->size << 1;
1012         if (sector_nr >= max_sector) {
1013                 close_sync(conf);
1014                 return 0;
1015         }
1016
1017         /*
1018          * If there is non-resync activity waiting for us then
1019          * put in a delay to throttle resync.
1020          */
1021         if (!go_faster && waitqueue_active(&conf->wait_resume))
1022                 msleep_interruptible(1000);
1023         device_barrier(conf, sector_nr + RESYNC_SECTORS);
1024
1025         /*
1026          * If reconstructing, and >1 working disc,
1027          * could dedicate one to rebuild and others to
1028          * service read requests ..
1029          */
1030         disk = conf->last_used;
1031         /* make sure disk is operational */
1032
1033         while (conf->mirrors[disk].rdev == NULL ||
1034                !conf->mirrors[disk].rdev->in_sync) {
1035                 if (disk <= 0)
1036                         disk = conf->raid_disks;
1037                 disk--;
1038                 if (disk == conf->last_used)
1039                         break;
1040         }
1041         conf->last_used = disk;
1042         atomic_inc(&conf->mirrors[disk].rdev->nr_pending);
1043
1044
1045         mirror = conf->mirrors + disk;
1046
1047         r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1048
1049         spin_lock_irq(&conf->resync_lock);
1050         conf->nr_pending++;
1051         spin_unlock_irq(&conf->resync_lock);
1052
1053         r1_bio->mddev = mddev;
1054         r1_bio->sector = sector_nr;
1055         set_bit(R1BIO_IsSync, &r1_bio->state);
1056         r1_bio->read_disk = disk;
1057
1058         for (i=0; i < conf->raid_disks; i++) {
1059                 bio = r1_bio->bios[i];
1060
1061                 /* take from bio_init */
1062                 bio->bi_next = NULL;
1063                 bio->bi_flags |= 1 << BIO_UPTODATE;
1064                 bio->bi_rw = 0;
1065                 bio->bi_vcnt = 0;
1066                 bio->bi_idx = 0;
1067                 bio->bi_phys_segments = 0;
1068                 bio->bi_hw_segments = 0;
1069                 bio->bi_size = 0;
1070                 bio->bi_end_io = NULL;
1071                 bio->bi_private = NULL;
1072
1073                 if (i == disk) {
1074                         bio->bi_rw = READ;
1075                         bio->bi_end_io = end_sync_read;
1076                 } else if (conf->mirrors[i].rdev &&
1077                            !conf->mirrors[i].rdev->faulty &&
1078                            (!conf->mirrors[i].rdev->in_sync ||
1079                             sector_nr + RESYNC_SECTORS > mddev->recovery_cp)) {
1080                         bio->bi_rw = WRITE;
1081                         bio->bi_end_io = end_sync_write;
1082                         write_targets ++;
1083                 } else
1084                         continue;
1085                 bio->bi_sector = sector_nr + conf->mirrors[i].rdev->data_offset;
1086                 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1087                 bio->bi_private = r1_bio;
1088         }
1089         if (write_targets == 0) {
1090                 /* There is nowhere to write, so all non-sync
1091                  * drives must be failed - so we are finished
1092                  */
1093                 int rv = max_sector - sector_nr;
1094                 md_done_sync(mddev, rv, 1);
1095                 put_buf(r1_bio);
1096                 rdev_dec_pending(conf->mirrors[disk].rdev, mddev);
1097                 return rv;
1098         }
1099
1100         nr_sectors = 0;
1101         do {
1102                 struct page *page;
1103                 int len = PAGE_SIZE;
1104                 if (sector_nr + (len>>9) > max_sector)
1105                         len = (max_sector - sector_nr) << 9;
1106                 if (len == 0)
1107                         break;
1108                 for (i=0 ; i < conf->raid_disks; i++) {
1109                         bio = r1_bio->bios[i];
1110                         if (bio->bi_end_io) {
1111                                 page = r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page;
1112                                 if (bio_add_page(bio, page, len, 0) == 0) {
1113                                         /* stop here */
1114                                         r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page = page;
1115                                         while (i > 0) {
1116                                                 i--;
1117                                                 bio = r1_bio->bios[i];
1118                                                 if (bio->bi_end_io==NULL) continue;
1119                                                 /* remove last page from this bio */
1120                                                 bio->bi_vcnt--;
1121                                                 bio->bi_size -= len;
1122                                                 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1123                                         }
1124                                         goto bio_full;
1125                                 }
1126                         }
1127                 }
1128                 nr_sectors += len>>9;
1129                 sector_nr += len>>9;
1130         } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1131  bio_full:
1132         bio = r1_bio->bios[disk];
1133         r1_bio->sectors = nr_sectors;
1134
1135         md_sync_acct(mirror->rdev->bdev, nr_sectors);
1136
1137         generic_make_request(bio);
1138
1139         return nr_sectors;
1140 }
1141
1142 static int run(mddev_t *mddev)
1143 {
1144         conf_t *conf;
1145         int i, j, disk_idx;
1146         mirror_info_t *disk;
1147         mdk_rdev_t *rdev;
1148         struct list_head *tmp;
1149
1150         if (mddev->level != 1) {
1151                 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1152                        mdname(mddev), mddev->level);
1153                 goto out;
1154         }
1155         /*
1156          * copy the already verified devices into our private RAID1
1157          * bookkeeping area. [whatever we allocate in run(),
1158          * should be freed in stop()]
1159          */
1160         conf = kmalloc(sizeof(conf_t), GFP_KERNEL);
1161         mddev->private = conf;
1162         if (!conf)
1163                 goto out_no_mem;
1164
1165         memset(conf, 0, sizeof(*conf));
1166         conf->mirrors = kmalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1167                                  GFP_KERNEL);
1168         if (!conf->mirrors)
1169                 goto out_no_mem;
1170
1171         memset(conf->mirrors, 0, sizeof(struct mirror_info)*mddev->raid_disks);
1172
1173         conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1174         if (!conf->poolinfo)
1175                 goto out_no_mem;
1176         conf->poolinfo->mddev = mddev;
1177         conf->poolinfo->raid_disks = mddev->raid_disks;
1178         conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1179                                           r1bio_pool_free,
1180                                           conf->poolinfo);
1181         if (!conf->r1bio_pool)
1182                 goto out_no_mem;
1183
1184         mddev->queue->unplug_fn = raid1_unplug;
1185
1186         mddev->queue->issue_flush_fn = raid1_issue_flush;
1187
1188         ITERATE_RDEV(mddev, rdev, tmp) {
1189                 disk_idx = rdev->raid_disk;
1190                 if (disk_idx >= mddev->raid_disks
1191                     || disk_idx < 0)
1192                         continue;
1193                 disk = conf->mirrors + disk_idx;
1194
1195                 disk->rdev = rdev;
1196
1197                 blk_queue_stack_limits(mddev->queue,
1198                                        rdev->bdev->bd_disk->queue);
1199                 /* as we don't honour merge_bvec_fn, we must never risk
1200                  * violating it, so limit ->max_sector to one PAGE, as
1201                  * a one page request is never in violation.
1202                  */
1203                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1204                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
1205                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1206
1207                 disk->head_position = 0;
1208                 if (!rdev->faulty && rdev->in_sync)
1209                         conf->working_disks++;
1210         }
1211         conf->raid_disks = mddev->raid_disks;
1212         conf->mddev = mddev;
1213         spin_lock_init(&conf->device_lock);
1214         INIT_LIST_HEAD(&conf->retry_list);
1215         if (conf->working_disks == 1)
1216                 mddev->recovery_cp = MaxSector;
1217
1218         spin_lock_init(&conf->resync_lock);
1219         init_waitqueue_head(&conf->wait_idle);
1220         init_waitqueue_head(&conf->wait_resume);
1221
1222         if (!conf->working_disks) {
1223                 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1224                         mdname(mddev));
1225                 goto out_free_conf;
1226         }
1227
1228         mddev->degraded = 0;
1229         for (i = 0; i < conf->raid_disks; i++) {
1230
1231                 disk = conf->mirrors + i;
1232
1233                 if (!disk->rdev) {
1234                         disk->head_position = 0;
1235                         mddev->degraded++;
1236                 }
1237         }
1238
1239         /*
1240          * find the first working one and use it as a starting point
1241          * to read balancing.
1242          */
1243         for (j = 0; j < conf->raid_disks &&
1244                      (!conf->mirrors[j].rdev ||
1245                       !conf->mirrors[j].rdev->in_sync) ; j++)
1246                 /* nothing */;
1247         conf->last_used = j;
1248
1249
1250
1251         {
1252                 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1253                 if (!mddev->thread) {
1254                         printk(KERN_ERR
1255                                 "raid1: couldn't allocate thread for %s\n",
1256                                 mdname(mddev));
1257                         goto out_free_conf;
1258                 }
1259         }
1260         printk(KERN_INFO
1261                 "raid1: raid set %s active with %d out of %d mirrors\n",
1262                 mdname(mddev), mddev->raid_disks - mddev->degraded,
1263                 mddev->raid_disks);
1264         /*
1265          * Ok, everything is just fine now
1266          */
1267         mddev->array_size = mddev->size;
1268
1269         return 0;
1270
1271 out_no_mem:
1272         printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1273                mdname(mddev));
1274
1275 out_free_conf:
1276         if (conf) {
1277                 if (conf->r1bio_pool)
1278                         mempool_destroy(conf->r1bio_pool);
1279                 if (conf->mirrors)
1280                         kfree(conf->mirrors);
1281                 if (conf->poolinfo)
1282                         kfree(conf->poolinfo);
1283                 kfree(conf);
1284                 mddev->private = NULL;
1285         }
1286 out:
1287         return -EIO;
1288 }
1289
1290 static int stop(mddev_t *mddev)
1291 {
1292         conf_t *conf = mddev_to_conf(mddev);
1293
1294         md_unregister_thread(mddev->thread);
1295         mddev->thread = NULL;
1296         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
1297         if (conf->r1bio_pool)
1298                 mempool_destroy(conf->r1bio_pool);
1299         if (conf->mirrors)
1300                 kfree(conf->mirrors);
1301         if (conf->poolinfo)
1302                 kfree(conf->poolinfo);
1303         kfree(conf);
1304         mddev->private = NULL;
1305         return 0;
1306 }
1307
1308 static int raid1_resize(mddev_t *mddev, sector_t sectors)
1309 {
1310         /* no resync is happening, and there is enough space
1311          * on all devices, so we can resize.
1312          * We need to make sure resync covers any new space.
1313          * If the array is shrinking we should possibly wait until
1314          * any io in the removed space completes, but it hardly seems
1315          * worth it.
1316          */
1317         mddev->array_size = sectors>>1;
1318         set_capacity(mddev->gendisk, mddev->array_size << 1);
1319         mddev->changed = 1;
1320         if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
1321                 mddev->recovery_cp = mddev->size << 1;
1322                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1323         }
1324         mddev->size = mddev->array_size;
1325         return 0;
1326 }
1327
1328 static int raid1_reshape(mddev_t *mddev, int raid_disks)
1329 {
1330         /* We need to:
1331          * 1/ resize the r1bio_pool
1332          * 2/ resize conf->mirrors
1333          *
1334          * We allocate a new r1bio_pool if we can.
1335          * Then raise a device barrier and wait until all IO stops.
1336          * Then resize conf->mirrors and swap in the new r1bio pool.
1337          */
1338         mempool_t *newpool, *oldpool;
1339         struct pool_info *newpoolinfo;
1340         mirror_info_t *newmirrors;
1341         conf_t *conf = mddev_to_conf(mddev);
1342
1343         int d;
1344
1345         for (d= raid_disks; d < conf->raid_disks; d++)
1346                 if (conf->mirrors[d].rdev)
1347                         return -EBUSY;
1348
1349         newpoolinfo = kmalloc(sizeof(newpoolinfo), GFP_KERNEL);
1350         if (!newpoolinfo)
1351                 return -ENOMEM;
1352         newpoolinfo->mddev = mddev;
1353         newpoolinfo->raid_disks = raid_disks;
1354
1355         newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1356                                  r1bio_pool_free, newpoolinfo);
1357         if (!newpool) {
1358                 kfree(newpoolinfo);
1359                 return -ENOMEM;
1360         }
1361         newmirrors = kmalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
1362         if (!newmirrors) {
1363                 kfree(newpoolinfo);
1364                 mempool_destroy(newpool);
1365                 return -ENOMEM;
1366         }
1367         memset(newmirrors, 0, sizeof(struct mirror_info)*raid_disks);
1368
1369         spin_lock_irq(&conf->resync_lock);
1370         conf->barrier++;
1371         wait_event_lock_irq(conf->wait_idle, !conf->nr_pending,
1372                             conf->resync_lock, unplug_slaves(mddev));
1373         spin_unlock_irq(&conf->resync_lock);
1374
1375         /* ok, everything is stopped */
1376         oldpool = conf->r1bio_pool;
1377         conf->r1bio_pool = newpool;
1378         for (d=0; d < raid_disks && d < conf->raid_disks; d++)
1379                 newmirrors[d] = conf->mirrors[d];
1380         kfree(conf->mirrors);
1381         conf->mirrors = newmirrors;
1382         kfree(conf->poolinfo);
1383         conf->poolinfo = newpoolinfo;
1384
1385         mddev->degraded += (raid_disks - conf->raid_disks);
1386         conf->raid_disks = mddev->raid_disks = raid_disks;
1387
1388         spin_lock_irq(&conf->resync_lock);
1389         conf->barrier--;
1390         spin_unlock_irq(&conf->resync_lock);
1391         wake_up(&conf->wait_resume);
1392         wake_up(&conf->wait_idle);
1393
1394
1395         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1396         md_wakeup_thread(mddev->thread);
1397
1398         mempool_destroy(oldpool);
1399         return 0;
1400 }
1401
1402
1403 static mdk_personality_t raid1_personality =
1404 {
1405         .name           = "raid1",
1406         .owner          = THIS_MODULE,
1407         .make_request   = make_request,
1408         .run            = run,
1409         .stop           = stop,
1410         .status         = status,
1411         .error_handler  = error,
1412         .hot_add_disk   = raid1_add_disk,
1413         .hot_remove_disk= raid1_remove_disk,
1414         .spare_active   = raid1_spare_active,
1415         .sync_request   = sync_request,
1416         .resize         = raid1_resize,
1417         .reshape        = raid1_reshape,
1418 };
1419
1420 static int __init raid_init(void)
1421 {
1422         return register_md_personality(RAID1, &raid1_personality);
1423 }
1424
1425 static void raid_exit(void)
1426 {
1427         unregister_md_personality(RAID1);
1428 }
1429
1430 module_init(raid_init);
1431 module_exit(raid_exit);
1432 MODULE_LICENSE("GPL");
1433 MODULE_ALIAS("md-personality-3"); /* RAID1 */