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