drivers/md/raid1.c

   1 /*
   2  * raid1.c : Multiple Devices driver for Linux
   3  *
   4  * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
   5  *
   6  * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
   7  *
   8  * RAID-1 management functions.
   9  *
  10  * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
  11  *
  12  * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
  13  * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
  14  *
  15  * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
  16  * bitmapped intelligence in resync:
  17  *
  18  *      - bitmap marked during normal i/o
  19  *      - bitmap used to skip nondirty blocks during sync
  20  *
  21  * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
  22  * - persistent bitmap code
  23  *
  24  * This program is free software; you can redistribute it and/or modify
  25  * it under the terms of the GNU General Public License as published by
  26  * the Free Software Foundation; either version 2, or (at your option)
  27  * any later version.
  28  *
  29  * You should have received a copy of the GNU General Public License
  30  * (for example /usr/src/linux/COPYING); if not, write to the Free
  31  * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  32  */
  33
  34 #include "dm-bio-list.h"
  35 #include <linux/raid/raid1.h>
  36 #include <linux/raid/bitmap.h>
  37
  38 #define DEBUG 0
  39 #if DEBUG
  40 #define PRINTK(x...) printk(x)
  41 #else
  42 #define PRINTK(x...)
  43 #endif
  44
  45 /*
  46  * Number of guaranteed r1bios in case of extreme VM load:
  47  */
  48 #define NR_RAID1_BIOS 256
  49
  50 static mdk_personality_t raid1_personality;
  51
  52 static void unplug_slaves(mddev_t *mddev);
  53
  54
  55 static void * r1bio_pool_alloc(unsigned int __nocast gfp_flags, void *data)
  56 {
  57         struct pool_info *pi = data;
  58         r1bio_t *r1_bio;
  59         int size = offsetof(r1bio_t, bios[pi->raid_disks]);
  60
  61         /* allocate a r1bio with room for raid_disks entries in the bios array */
  62         r1_bio = kmalloc(size, gfp_flags);
  63         if (r1_bio)
  64                 memset(r1_bio, 0, size);
  65         else
  66                 unplug_slaves(pi->mddev);
  67
  68         return r1_bio;
  69 }
  70
  71 static void r1bio_pool_free(void *r1_bio, void *data)
  72 {
  73         kfree(r1_bio);
  74 }
  75
  76 #define RESYNC_BLOCK_SIZE (64*1024)
  77 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
  78 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
  79 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
  80 #define RESYNC_WINDOW (2048*1024)
  81
  82 static void * r1buf_pool_alloc(unsigned int __nocast gfp_flags, void *data)
  83 {
  84         struct pool_info *pi = data;
  85         struct page *page;
  86         r1bio_t *r1_bio;
  87         struct bio *bio;
  88         int i, j;
  89
  90         r1_bio = r1bio_pool_alloc(gfp_flags, pi);
  91         if (!r1_bio) {
  92                 unplug_slaves(pi->mddev);
  93                 return NULL;
  94         }
  95
  96         /*
  97          * Allocate bios : 1 for reading, n-1 for writing
  98          */
  99         for (j = pi->raid_disks ; j-- ; ) {
 100                 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
 101                 if (!bio)
 102                         goto out_free_bio;
 103                 r1_bio->bios[j] = bio;
 104         }
 105         /*
 106          * Allocate RESYNC_PAGES data pages and attach them to
 107          * the first bio;
 108          */
 109         bio = r1_bio->bios[0];
 110         for (i = 0; i < RESYNC_PAGES; i++) {
 111                 page = alloc_page(gfp_flags);
 112                 if (unlikely(!page))
 113                         goto out_free_pages;
 114
 115                 bio->bi_io_vec[i].bv_page = page;
 116         }
 117
 118         r1_bio->master_bio = NULL;
 119
 120         return r1_bio;
 121
 122 out_free_pages:
 123         for ( ; i > 0 ; i--)
 124                 __free_page(bio->bi_io_vec[i-1].bv_page);
 125 out_free_bio:
 126         while ( ++j < pi->raid_disks )
 127                 bio_put(r1_bio->bios[j]);
 128         r1bio_pool_free(r1_bio, data);
 129         return NULL;
 130 }
 131
 132 static void r1buf_pool_free(void *__r1_bio, void *data)
 133 {
 134         struct pool_info *pi = data;
 135         int i;
 136         r1bio_t *r1bio = __r1_bio;
 137         struct bio *bio = r1bio->bios[0];
 138
 139         for (i = 0; i < RESYNC_PAGES; i++) {
 140                 __free_page(bio->bi_io_vec[i].bv_page);
 141                 bio->bi_io_vec[i].bv_page = NULL;
 142         }
 143         for (i=0 ; i < pi->raid_disks; i++)
 144                 bio_put(r1bio->bios[i]);
 145
 146         r1bio_pool_free(r1bio, data);
 147 }
 148
 149 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
 150 {
 151         int i;
 152
 153         for (i = 0; i < conf->raid_disks; i++) {
 154                 struct bio **bio = r1_bio->bios + i;
 155                 if (*bio)
 156                         bio_put(*bio);
 157                 *bio = NULL;
 158         }
 159 }
 160
 161 static inline void free_r1bio(r1bio_t *r1_bio)
 162 {
 163         unsigned long flags;
 164
 165         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 166
 167         /*
 168          * Wake up any possible resync thread that waits for the device
 169          * to go idle.
 170          */
 171         spin_lock_irqsave(&conf->resync_lock, flags);
 172         if (!--conf->nr_pending) {
 173                 wake_up(&conf->wait_idle);
 174                 wake_up(&conf->wait_resume);
 175         }
 176         spin_unlock_irqrestore(&conf->resync_lock, flags);
 177
 178         put_all_bios(conf, r1_bio);
 179         mempool_free(r1_bio, conf->r1bio_pool);
 180 }
 181
 182 static inline void put_buf(r1bio_t *r1_bio)
 183 {
 184         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 185         unsigned long flags;
 186
 187         mempool_free(r1_bio, conf->r1buf_pool);
 188
 189         spin_lock_irqsave(&conf->resync_lock, flags);
 190         if (!conf->barrier)
 191                 BUG();
 192         --conf->barrier;
 193         wake_up(&conf->wait_resume);
 194         wake_up(&conf->wait_idle);
 195
 196         if (!--conf->nr_pending) {
 197                 wake_up(&conf->wait_idle);
 198                 wake_up(&conf->wait_resume);
 199         }
 200         spin_unlock_irqrestore(&conf->resync_lock, flags);
 201 }
 202
 203 static void reschedule_retry(r1bio_t *r1_bio)
 204 {
 205         unsigned long flags;
 206         mddev_t *mddev = r1_bio->mddev;
 207         conf_t *conf = mddev_to_conf(mddev);
 208
 209         spin_lock_irqsave(&conf->device_lock, flags);
 210         list_add(&r1_bio->retry_list, &conf->retry_list);
 211         spin_unlock_irqrestore(&conf->device_lock, flags);
 212
 213         md_wakeup_thread(mddev->thread);
 214 }
 215
 216 /*
 217  * raid_end_bio_io() is called when we have finished servicing a mirrored
 218  * operation and are ready to return a success/failure code to the buffer
 219  * cache layer.
 220  */
 221 static void raid_end_bio_io(r1bio_t *r1_bio)
 222 {
 223         struct bio *bio = r1_bio->master_bio;
 224
 225         bio_endio(bio, bio->bi_size,
 226                 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
 227         free_r1bio(r1_bio);
 228 }
 229
 230 /*
 231  * Update disk head position estimator based on IRQ completion info.
 232  */
 233 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
 234 {
 235         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 236
 237         conf->mirrors[disk].head_position =
 238                 r1_bio->sector + (r1_bio->sectors);
 239 }
 240
 241 static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
 242 {
 243         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 244         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 245         int mirror;
 246         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 247
 248         if (bio->bi_size)
 249                 return 1;
 250
 251         mirror = r1_bio->read_disk;
 252         /*
 253          * this branch is our 'one mirror IO has finished' event handler:
 254          */
 255         if (!uptodate)
 256                 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
 257         else
 258                 /*
 259                  * Set R1BIO_Uptodate in our master bio, so that
 260                  * we will return a good error code for to the higher
 261                  * levels even if IO on some other mirrored buffer fails.
 262                  *
 263                  * The 'master' represents the composite IO operation to
 264                  * user-side. So if something waits for IO, then it will
 265                  * wait for the 'master' bio.
 266                  */
 267                 set_bit(R1BIO_Uptodate, &r1_bio->state);
 268
 269         update_head_pos(mirror, r1_bio);
 270
 271         /*
 272          * we have only one bio on the read side
 273          */
 274         if (uptodate)
 275                 raid_end_bio_io(r1_bio);
 276         else {
 277                 /*
 278                  * oops, read error:
 279                  */
 280                 char b[BDEVNAME_SIZE];
 281                 if (printk_ratelimit())
 282                         printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
 283                                bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
 284                 reschedule_retry(r1_bio);
 285         }
 286
 287         rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 288         return 0;
 289 }
 290
 291 static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
 292 {
 293         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 294         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 295         int mirror;
 296         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 297
 298         if (bio->bi_size)
 299                 return 1;
 300
 301         for (mirror = 0; mirror < conf->raid_disks; mirror++)
 302                 if (r1_bio->bios[mirror] == bio)
 303                         break;
 304
 305         /*
 306          * this branch is our 'one mirror IO has finished' event handler:
 307          */
 308         if (!uptodate) {
 309                 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
 310                 /* an I/O failed, we can't clear the bitmap */
 311                 set_bit(R1BIO_Degraded, &r1_bio->state);
 312         } else
 313                 /*
 314                  * Set R1BIO_Uptodate in our master bio, so that
 315                  * we will return a good error code for to the higher
 316                  * levels even if IO on some other mirrored buffer fails.
 317                  *
 318                  * The 'master' represents the composite IO operation to
 319                  * user-side. So if something waits for IO, then it will
 320                  * wait for the 'master' bio.
 321                  */
 322                 set_bit(R1BIO_Uptodate, &r1_bio->state);
 323
 324         update_head_pos(mirror, r1_bio);
 325
 326         /*
 327          *
 328          * Let's see if all mirrored write operations have finished
 329          * already.
 330          */
 331         if (atomic_dec_and_test(&r1_bio->remaining)) {
 332                 /* clear the bitmap if all writes complete successfully */
 333                 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
 334                                 r1_bio->sectors,
 335                                 !test_bit(R1BIO_Degraded, &r1_bio->state));
 336                 md_write_end(r1_bio->mddev);
 337                 raid_end_bio_io(r1_bio);
 338         }
 339
 340         rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 341         return 0;
 342 }
 343
 344
 345 /*
 346  * This routine returns the disk from which the requested read should
 347  * be done. There is a per-array 'next expected sequential IO' sector
 348  * number - if this matches on the next IO then we use the last disk.
 349  * There is also a per-disk 'last know head position' sector that is
 350  * maintained from IRQ contexts, both the normal and the resync IO
 351  * completion handlers update this position correctly. If there is no
 352  * perfect sequential match then we pick the disk whose head is closest.
 353  *
 354  * If there are 2 mirrors in the same 2 devices, performance degrades
 355  * because position is mirror, not device based.
 356  *
 357  * The rdev for the device selected will have nr_pending incremented.
 358  */
 359 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
 360 {
 361         const unsigned long this_sector = r1_bio->sector;
 362         int new_disk = conf->last_used, disk = new_disk;
 363         const int sectors = r1_bio->sectors;
 364         sector_t new_distance, current_distance;
 365         mdk_rdev_t *new_rdev, *rdev;
 366
 367         rcu_read_lock();
 368         /*
 369          * Check if it if we can balance. We can balance on the whole
 370          * device if no resync is going on, or below the resync window.
 371          * We take the first readable disk when above the resync window.
 372          */
 373  retry:
 374         if (conf->mddev->recovery_cp < MaxSector &&
 375             (this_sector + sectors >= conf->next_resync)) {
 376                 /* Choose the first operation device, for consistancy */
 377                 new_disk = 0;
 378
 379                 while ((new_rdev=conf->mirrors[new_disk].rdev) == NULL ||
 380                        !new_rdev->in_sync) {
 381                         new_disk++;
 382                         if (new_disk == conf->raid_disks) {
 383                                 new_disk = -1;
 384                                 break;
 385                         }
 386                 }
 387                 goto rb_out;
 388         }
 389
 390
 391         /* make sure the disk is operational */
 392         while ((new_rdev=conf->mirrors[new_disk].rdev) == NULL ||
 393                !new_rdev->in_sync) {
 394                 if (new_disk <= 0)
 395                         new_disk = conf->raid_disks;
 396                 new_disk--;
 397                 if (new_disk == disk) {
 398                         new_disk = -1;
 399                         goto rb_out;
 400                 }
 401         }
 402         disk = new_disk;
 403         /* now disk == new_disk == starting point for search */
 404
 405         /*
 406          * Don't change to another disk for sequential reads:
 407          */
 408         if (conf->next_seq_sect == this_sector)
 409                 goto rb_out;
 410         if (this_sector == conf->mirrors[new_disk].head_position)
 411                 goto rb_out;
 412
 413         current_distance = abs(this_sector - conf->mirrors[disk].head_position);
 414
 415         /* Find the disk whose head is closest */
 416
 417         do {
 418                 if (disk <= 0)
 419                         disk = conf->raid_disks;
 420                 disk--;
 421
 422                 if ((rdev=conf->mirrors[disk].rdev) == NULL ||
 423                     !rdev->in_sync)
 424                         continue;
 425
 426                 if (!atomic_read(&rdev->nr_pending)) {
 427                         new_disk = disk;
 428                         new_rdev = rdev;
 429                         break;
 430                 }
 431                 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
 432                 if (new_distance < current_distance) {
 433                         current_distance = new_distance;
 434                         new_disk = disk;
 435                         new_rdev = rdev;
 436                 }
 437         } while (disk != conf->last_used);
 438
 439 rb_out:
 440
 441
 442         if (new_disk >= 0) {
 443                 conf->next_seq_sect = this_sector + sectors;
 444                 conf->last_used = new_disk;
 445                 atomic_inc(&new_rdev->nr_pending);
 446                 if (!new_rdev->in_sync) {
 447                         /* cannot risk returning a device that failed
 448                          * before we inc'ed nr_pending
 449                          */
 450                         atomic_dec(&new_rdev->nr_pending);
 451                         goto retry;
 452                 }
 453         }
 454         rcu_read_unlock();
 455
 456         return new_disk;
 457 }
 458
 459 static void unplug_slaves(mddev_t *mddev)
 460 {
 461         conf_t *conf = mddev_to_conf(mddev);
 462         int i;
 463
 464         rcu_read_lock();
 465         for (i=0; i<mddev->raid_disks; i++) {
 466                 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
 467                 if (rdev && !rdev->faulty && atomic_read(&rdev->nr_pending)) {
 468                         request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
 469
 470                         atomic_inc(&rdev->nr_pending);
 471                         rcu_read_unlock();
 472
 473                         if (r_queue->unplug_fn)
 474                                 r_queue->unplug_fn(r_queue);
 475
 476                         rdev_dec_pending(rdev, mddev);
 477                         rcu_read_lock();
 478                 }
 479         }
 480         rcu_read_unlock();
 481 }
 482
 483 static void raid1_unplug(request_queue_t *q)
 484 {
 485         mddev_t *mddev = q->queuedata;
 486
 487         unplug_slaves(mddev);
 488         md_wakeup_thread(mddev->thread);
 489 }
 490
 491 static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
 492                              sector_t *error_sector)
 493 {
 494         mddev_t *mddev = q->queuedata;
 495         conf_t *conf = mddev_to_conf(mddev);
 496         int i, ret = 0;
 497
 498         rcu_read_lock();
 499         for (i=0; i<mddev->raid_disks && ret == 0; i++) {
 500                 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
 501                 if (rdev && !rdev->faulty) {
 502                         struct block_device *bdev = rdev->bdev;
 503                         request_queue_t *r_queue = bdev_get_queue(bdev);
 504
 505                         if (!r_queue->issue_flush_fn)
 506                                 ret = -EOPNOTSUPP;
 507                         else {
 508                                 atomic_inc(&rdev->nr_pending);
 509                                 rcu_read_unlock();
 510                                 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
 511                                                               error_sector);
 512                                 rdev_dec_pending(rdev, mddev);
 513                                 rcu_read_lock();
 514                         }
 515                 }
 516         }
 517         rcu_read_unlock();
 518         return ret;
 519 }
 520
 521 /*
 522  * Throttle resync depth, so that we can both get proper overlapping of
 523  * requests, but are still able to handle normal requests quickly.
 524  */
 525 #define RESYNC_DEPTH 32
 526
 527 static void device_barrier(conf_t *conf, sector_t sect)
 528 {
 529         spin_lock_irq(&conf->resync_lock);
 530         wait_event_lock_irq(conf->wait_idle, !waitqueue_active(&conf->wait_resume),
 531                             conf->resync_lock, raid1_unplug(conf->mddev->queue));
 532
 533         if (!conf->barrier++) {
 534                 wait_event_lock_irq(conf->wait_idle, !conf->nr_pending,
 535                                     conf->resync_lock, raid1_unplug(conf->mddev->queue));
 536                 if (conf->nr_pending)
 537                         BUG();
 538         }
 539         wait_event_lock_irq(conf->wait_resume, conf->barrier < RESYNC_DEPTH,
 540                             conf->resync_lock, raid1_unplug(conf->mddev->queue));
 541         conf->next_resync = sect;
 542         spin_unlock_irq(&conf->resync_lock);
 543 }
 544
 545 static int make_request(request_queue_t *q, struct bio * bio)
 546 {
 547         mddev_t *mddev = q->queuedata;
 548         conf_t *conf = mddev_to_conf(mddev);
 549         mirror_info_t *mirror;
 550         r1bio_t *r1_bio;
 551         struct bio *read_bio;
 552         int i, targets = 0, disks;
 553         mdk_rdev_t *rdev;
 554         struct bitmap *bitmap = mddev->bitmap;
 555         unsigned long flags;
 556         struct bio_list bl;
 557
 558
 559         /*
 560          * Register the new request and wait if the reconstruction
 561          * thread has put up a bar for new requests.
 562          * Continue immediately if no resync is active currently.
 563          */
 564         md_write_start(mddev, bio); /* wait on superblock update early */
 565
 566         spin_lock_irq(&conf->resync_lock);
 567         wait_event_lock_irq(conf->wait_resume, !conf->barrier, conf->resync_lock, );
 568         conf->nr_pending++;
 569         spin_unlock_irq(&conf->resync_lock);
 570
 571         if (bio_data_dir(bio)==WRITE) {
 572                 disk_stat_inc(mddev->gendisk, writes);
 573                 disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio));
 574         } else {
 575                 disk_stat_inc(mddev->gendisk, reads);
 576                 disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio));
 577         }
 578
 579         /*
 580          * make_request() can abort the operation when READA is being
 581          * used and no empty request is available.
 582          *
 583          */
 584         r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
 585
 586         r1_bio->master_bio = bio;
 587         r1_bio->sectors = bio->bi_size >> 9;
 588         r1_bio->state = 0;
 589         r1_bio->mddev = mddev;
 590         r1_bio->sector = bio->bi_sector;
 591
 592         r1_bio->state = 0;
 593
 594         if (bio_data_dir(bio) == READ) {
 595                 /*
 596                  * read balancing logic:
 597                  */
 598                 int rdisk = read_balance(conf, r1_bio);
 599
 600                 if (rdisk < 0) {
 601                         /* couldn't find anywhere to read from */
 602                         raid_end_bio_io(r1_bio);
 603                         return 0;
 604                 }
 605                 mirror = conf->mirrors + rdisk;
 606
 607                 r1_bio->read_disk = rdisk;
 608
 609                 read_bio = bio_clone(bio, GFP_NOIO);
 610
 611                 r1_bio->bios[rdisk] = read_bio;
 612
 613                 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
 614                 read_bio->bi_bdev = mirror->rdev->bdev;
 615                 read_bio->bi_end_io = raid1_end_read_request;
 616                 read_bio->bi_rw = READ;
 617                 read_bio->bi_private = r1_bio;
 618
 619                 generic_make_request(read_bio);
 620                 return 0;
 621         }
 622
 623         /*
 624          * WRITE:
 625          */
 626         /* first select target devices under spinlock and
 627          * inc refcount on their rdev.  Record them by setting
 628          * bios[x] to bio
 629          */
 630         disks = conf->raid_disks;
 631 #if 0
 632         { static int first=1;
 633         if (first) printk("First Write sector %llu disks %d\n",
 634                           (unsigned long long)r1_bio->sector, disks);
 635         first = 0;
 636         }
 637 #endif
 638         rcu_read_lock();
 639         for (i = 0;  i < disks; i++) {
 640                 if ((rdev=conf->mirrors[i].rdev) != NULL &&
 641                     !rdev->faulty) {
 642                         atomic_inc(&rdev->nr_pending);
 643                         if (rdev->faulty) {
 644                                 atomic_dec(&rdev->nr_pending);
 645                                 r1_bio->bios[i] = NULL;
 646                         } else
 647                                 r1_bio->bios[i] = bio;
 648                         targets++;
 649                 } else
 650                         r1_bio->bios[i] = NULL;
 651         }
 652         rcu_read_unlock();
 653
 654         if (targets < conf->raid_disks) {
 655                 /* array is degraded, we will not clear the bitmap
 656                  * on I/O completion (see raid1_end_write_request) */
 657                 set_bit(R1BIO_Degraded, &r1_bio->state);
 658         }
 659
 660         atomic_set(&r1_bio->remaining, 0);
 661
 662         bio_list_init(&bl);
 663         for (i = 0; i < disks; i++) {
 664                 struct bio *mbio;
 665                 if (!r1_bio->bios[i])
 666                         continue;
 667
 668                 mbio = bio_clone(bio, GFP_NOIO);
 669                 r1_bio->bios[i] = mbio;
 670
 671                 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
 672                 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
 673                 mbio->bi_end_io = raid1_end_write_request;
 674                 mbio->bi_rw = WRITE;
 675                 mbio->bi_private = r1_bio;
 676
 677                 atomic_inc(&r1_bio->remaining);
 678
 679                 bio_list_add(&bl, mbio);
 680         }
 681
 682         bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors);
 683         spin_lock_irqsave(&conf->device_lock, flags);
 684         bio_list_merge(&conf->pending_bio_list, &bl);
 685         bio_list_init(&bl);
 686
 687         blk_plug_device(mddev->queue);
 688         spin_unlock_irqrestore(&conf->device_lock, flags);
 689
 690 #if 0
 691         while ((bio = bio_list_pop(&bl)) != NULL)
 692                 generic_make_request(bio);
 693 #endif
 694
 695         return 0;
 696 }
 697
 698 static void status(struct seq_file *seq, mddev_t *mddev)
 699 {
 700         conf_t *conf = mddev_to_conf(mddev);
 701         int i;
 702
 703         seq_printf(seq, " [%d/%d] [", conf->raid_disks,
 704                                                 conf->working_disks);
 705         for (i = 0; i < conf->raid_disks; i++)
 706                 seq_printf(seq, "%s",
 707                               conf->mirrors[i].rdev &&
 708                               conf->mirrors[i].rdev->in_sync ? "U" : "_");
 709         seq_printf(seq, "]");
 710 }
 711
 712
 713 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
 714 {
 715         char b[BDEVNAME_SIZE];
 716         conf_t *conf = mddev_to_conf(mddev);
 717
 718         /*
 719          * If it is not operational, then we have already marked it as dead
 720          * else if it is the last working disks, ignore the error, let the
 721          * next level up know.
 722          * else mark the drive as failed
 723          */
 724         if (rdev->in_sync
 725             && conf->working_disks == 1)
 726                 /*
 727                  * Don't fail the drive, act as though we were just a
 728                  * normal single drive
 729                  */
 730                 return;
 731         if (rdev->in_sync) {
 732                 mddev->degraded++;
 733                 conf->working_disks--;
 734                 /*
 735                  * if recovery is running, make sure it aborts.
 736                  */
 737                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
 738         }
 739         rdev->in_sync = 0;
 740         rdev->faulty = 1;
 741         mddev->sb_dirty = 1;
 742         printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
 743                 "       Operation continuing on %d devices\n",
 744                 bdevname(rdev->bdev,b), conf->working_disks);
 745 }
 746
 747 static void print_conf(conf_t *conf)
 748 {
 749         int i;
 750         mirror_info_t *tmp;
 751
 752         printk("RAID1 conf printout:\n");
 753         if (!conf) {
 754                 printk("(!conf)\n");
 755                 return;
 756         }
 757         printk(" --- wd:%d rd:%d\n", conf->working_disks,
 758                 conf->raid_disks);
 759
 760         for (i = 0; i < conf->raid_disks; i++) {
 761                 char b[BDEVNAME_SIZE];
 762                 tmp = conf->mirrors + i;
 763                 if (tmp->rdev)
 764                         printk(" disk %d, wo:%d, o:%d, dev:%s\n",
 765                                 i, !tmp->rdev->in_sync, !tmp->rdev->faulty,
 766                                 bdevname(tmp->rdev->bdev,b));
 767         }
 768 }
 769
 770 static void close_sync(conf_t *conf)
 771 {
 772         spin_lock_irq(&conf->resync_lock);
 773         wait_event_lock_irq(conf->wait_resume, !conf->barrier,
 774                             conf->resync_lock,  raid1_unplug(conf->mddev->queue));
 775         spin_unlock_irq(&conf->resync_lock);
 776
 777         if (conf->barrier) BUG();
 778         if (waitqueue_active(&conf->wait_idle)) BUG();
 779
 780         mempool_destroy(conf->r1buf_pool);
 781         conf->r1buf_pool = NULL;
 782 }
 783
 784 static int raid1_spare_active(mddev_t *mddev)
 785 {
 786         int i;
 787         conf_t *conf = mddev->private;
 788         mirror_info_t *tmp;
 789
 790         /*
 791          * Find all failed disks within the RAID1 configuration
 792          * and mark them readable
 793          */
 794         for (i = 0; i < conf->raid_disks; i++) {
 795                 tmp = conf->mirrors + i;
 796                 if (tmp->rdev
 797                     && !tmp->rdev->faulty
 798                     && !tmp->rdev->in_sync) {
 799                         conf->working_disks++;
 800                         mddev->degraded--;
 801                         tmp->rdev->in_sync = 1;
 802                 }
 803         }
 804
 805         print_conf(conf);
 806         return 0;
 807 }
 808
 809
 810 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
 811 {
 812         conf_t *conf = mddev->private;
 813         int found = 0;
 814         int mirror = 0;
 815         mirror_info_t *p;
 816
 817         if (rdev->saved_raid_disk >= 0 &&
 818             conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
 819                 mirror = rdev->saved_raid_disk;
 820         for (mirror=0; mirror < mddev->raid_disks; mirror++)
 821                 if ( !(p=conf->mirrors+mirror)->rdev) {
 822
 823                         blk_queue_stack_limits(mddev->queue,
 824                                                rdev->bdev->bd_disk->queue);
 825                         /* as we don't honour merge_bvec_fn, we must never risk
 826                          * violating it, so limit ->max_sector to one PAGE, as
 827                          * a one page request is never in violation.
 828                          */
 829                         if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 830                             mddev->queue->max_sectors > (PAGE_SIZE>>9))
 831                                 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 832
 833                         p->head_position = 0;
 834                         rdev->raid_disk = mirror;
 835                         found = 1;
 836                         if (rdev->saved_raid_disk != mirror)
 837                                 conf->fullsync = 1;
 838                         p->rdev = rdev;
 839                         break;
 840                 }
 841
 842         print_conf(conf);
 843         return found;
 844 }
 845
 846 static int raid1_remove_disk(mddev_t *mddev, int number)
 847 {
 848         conf_t *conf = mddev->private;
 849         int err = 0;
 850         mdk_rdev_t *rdev;
 851         mirror_info_t *p = conf->mirrors+ number;
 852
 853         print_conf(conf);
 854         rdev = p->rdev;
 855         if (rdev) {
 856                 if (rdev->in_sync ||
 857                     atomic_read(&rdev->nr_pending)) {
 858                         err = -EBUSY;
 859                         goto abort;
 860                 }
 861                 p->rdev = NULL;
 862                 synchronize_rcu();
 863                 if (atomic_read(&rdev->nr_pending)) {
 864                         /* lost the race, try later */
 865                         err = -EBUSY;
 866                         p->rdev = rdev;
 867                 }
 868         }
 869 abort:
 870
 871         print_conf(conf);
 872         return err;
 873 }
 874
 875
 876 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
 877 {
 878         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 879         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 880         conf_t *conf = mddev_to_conf(r1_bio->mddev);
 881
 882         if (bio->bi_size)
 883                 return 1;
 884
 885         if (r1_bio->bios[r1_bio->read_disk] != bio)
 886                 BUG();
 887         update_head_pos(r1_bio->read_disk, r1_bio);
 888         /*
 889          * we have read a block, now it needs to be re-written,
 890          * or re-read if the read failed.
 891          * We don't do much here, just schedule handling by raid1d
 892          */
 893         if (!uptodate) {
 894                 md_error(r1_bio->mddev,
 895                          conf->mirrors[r1_bio->read_disk].rdev);
 896         } else
 897                 set_bit(R1BIO_Uptodate, &r1_bio->state);
 898         rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
 899         reschedule_retry(r1_bio);
 900         return 0;
 901 }
 902
 903 static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
 904 {
 905         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 906         r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
 907         mddev_t *mddev = r1_bio->mddev;
 908         conf_t *conf = mddev_to_conf(mddev);
 909         int i;
 910         int mirror=0;
 911
 912         if (bio->bi_size)
 913                 return 1;
 914
 915         for (i = 0; i < conf->raid_disks; i++)
 916                 if (r1_bio->bios[i] == bio) {
 917                         mirror = i;
 918                         break;
 919                 }
 920         if (!uptodate)
 921                 md_error(mddev, conf->mirrors[mirror].rdev);
 922
 923         update_head_pos(mirror, r1_bio);
 924
 925         if (atomic_dec_and_test(&r1_bio->remaining)) {
 926                 md_done_sync(mddev, r1_bio->sectors, uptodate);
 927                 put_buf(r1_bio);
 928         }
 929         rdev_dec_pending(conf->mirrors[mirror].rdev, mddev);
 930         return 0;
 931 }
 932
 933 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
 934 {
 935         conf_t *conf = mddev_to_conf(mddev);
 936         int i;
 937         int disks = conf->raid_disks;
 938         struct bio *bio, *wbio;
 939
 940         bio = r1_bio->bios[r1_bio->read_disk];
 941
 942 /*
 943         if (r1_bio->sector == 0) printk("First sync write startss\n");
 944 */
 945         /*
 946          * schedule writes
 947          */
 948         if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
 949                 /*
 950                  * There is no point trying a read-for-reconstruct as
 951                  * reconstruct is about to be aborted
 952                  */
 953                 char b[BDEVNAME_SIZE];
 954                 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
 955                         " for block %llu\n",
 956                         bdevname(bio->bi_bdev,b),
 957                         (unsigned long long)r1_bio->sector);
 958                 md_done_sync(mddev, r1_bio->sectors, 0);
 959                 put_buf(r1_bio);
 960                 return;
 961         }
 962
 963         atomic_set(&r1_bio->remaining, 1);
 964         for (i = 0; i < disks ; i++) {
 965                 wbio = r1_bio->bios[i];
 966                 if (wbio->bi_end_io != end_sync_write)
 967                         continue;
 968
 969                 atomic_inc(&conf->mirrors[i].rdev->nr_pending);
 970                 atomic_inc(&r1_bio->remaining);
 971                 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
 972
 973                 generic_make_request(wbio);
 974         }
 975
 976         if (atomic_dec_and_test(&r1_bio->remaining)) {
 977                 /* if we're here, all write(s) have completed, so clean up */
 978                 md_done_sync(mddev, r1_bio->sectors, 1);
 979                 put_buf(r1_bio);
 980         }
 981 }
 982
 983 /*
 984  * This is a kernel thread which:
 985  *
 986  *      1.      Retries failed read operations on working mirrors.
 987  *      2.      Updates the raid superblock when problems encounter.
 988  *      3.      Performs writes following reads for array syncronising.
 989  */
 990
 991 static void raid1d(mddev_t *mddev)
 992 {
 993         r1bio_t *r1_bio;
 994         struct bio *bio;
 995         unsigned long flags;
 996         conf_t *conf = mddev_to_conf(mddev);
 997         struct list_head *head = &conf->retry_list;
 998         int unplug=0;
 999         mdk_rdev_t *rdev;
1000
1001         md_check_recovery(mddev);
1002
1003         for (;;) {
1004                 char b[BDEVNAME_SIZE];
1005                 spin_lock_irqsave(&conf->device_lock, flags);
1006
1007                 if (conf->pending_bio_list.head) {
1008                         bio = bio_list_get(&conf->pending_bio_list);
1009                         blk_remove_plug(mddev->queue);
1010                         spin_unlock_irqrestore(&conf->device_lock, flags);
1011                         /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1012                         if (bitmap_unplug(mddev->bitmap) != 0)
1013                                 printk("%s: bitmap file write failed!\n", mdname(mddev));
1014
1015                         while (bio) { /* submit pending writes */
1016                                 struct bio *next = bio->bi_next;
1017                                 bio->bi_next = NULL;
1018                                 generic_make_request(bio);
1019                                 bio = next;
1020                         }
1021                         unplug = 1;
1022
1023                         continue;
1024                 }
1025
1026                 if (list_empty(head))
1027                         break;
1028                 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1029                 list_del(head->prev);
1030                 spin_unlock_irqrestore(&conf->device_lock, flags);
1031
1032                 mddev = r1_bio->mddev;
1033                 conf = mddev_to_conf(mddev);
1034                 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1035                         sync_request_write(mddev, r1_bio);
1036                         unplug = 1;
1037                 } else {
1038                         int disk;
1039                         bio = r1_bio->bios[r1_bio->read_disk];
1040                         if ((disk=read_balance(conf, r1_bio)) == -1) {
1041                                 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1042                                        " read error for block %llu\n",
1043                                        bdevname(bio->bi_bdev,b),
1044                                        (unsigned long long)r1_bio->sector);
1045                                 raid_end_bio_io(r1_bio);
1046                         } else {
1047                                 r1_bio->bios[r1_bio->read_disk] = NULL;
1048                                 r1_bio->read_disk = disk;
1049                                 bio_put(bio);
1050                                 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1051                                 r1_bio->bios[r1_bio->read_disk] = bio;
1052                                 rdev = conf->mirrors[disk].rdev;
1053                                 if (printk_ratelimit())
1054                                         printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1055                                                " another mirror\n",
1056                                                bdevname(rdev->bdev,b),
1057                                                (unsigned long long)r1_bio->sector);
1058                                 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1059                                 bio->bi_bdev = rdev->bdev;
1060                                 bio->bi_end_io = raid1_end_read_request;
1061                                 bio->bi_rw = READ;
1062                                 bio->bi_private = r1_bio;
1063                                 unplug = 1;
1064                                 generic_make_request(bio);
1065                         }
1066                 }
1067         }
1068         spin_unlock_irqrestore(&conf->device_lock, flags);
1069         if (unplug)
1070                 unplug_slaves(mddev);
1071 }
1072
1073
1074 static int init_resync(conf_t *conf)
1075 {
1076         int buffs;
1077
1078         buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1079         if (conf->r1buf_pool)
1080                 BUG();
1081         conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1082                                           conf->poolinfo);
1083         if (!conf->r1buf_pool)
1084                 return -ENOMEM;
1085         conf->next_resync = 0;
1086         return 0;
1087 }
1088
1089 /*
1090  * perform a "sync" on one "block"
1091  *
1092  * We need to make sure that no normal I/O request - particularly write
1093  * requests - conflict with active sync requests.
1094  *
1095  * This is achieved by tracking pending requests and a 'barrier' concept
1096  * that can be installed to exclude normal IO requests.
1097  */
1098
1099 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1100 {
1101         conf_t *conf = mddev_to_conf(mddev);
1102         mirror_info_t *mirror;
1103         r1bio_t *r1_bio;
1104         struct bio *bio;
1105         sector_t max_sector, nr_sectors;
1106         int disk;
1107         int i;
1108         int write_targets = 0;
1109         int sync_blocks;
1110         int still_degraded = 0;
1111
1112         if (!conf->r1buf_pool)
1113         {
1114 /*
1115                 printk("sync start - bitmap %p\n", mddev->bitmap);
1116 */
1117                 if (init_resync(conf))
1118                         return 0;
1119         }
1120
1121         max_sector = mddev->size << 1;
1122         if (sector_nr >= max_sector) {
1123                 /* If we aborted, we need to abort the
1124                  * sync on the 'current' bitmap chunk (there will
1125                  * only be one in raid1 resync.
1126                  * We can find the current addess in mddev->curr_resync
1127                  */
1128                 if (mddev->curr_resync < max_sector) /* aborted */
1129                         bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1130                                                 &sync_blocks, 1);
1131                 else /* completed sync */
1132                         conf->fullsync = 0;
1133
1134                 bitmap_close_sync(mddev->bitmap);
1135                 close_sync(conf);
1136                 return 0;
1137         }
1138
1139         /* before building a request, check if we can skip these blocks..
1140          * This call the bitmap_start_sync doesn't actually record anything
1141          */
1142         if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1143             !conf->fullsync) {
1144                 /* We can skip this block, and probably several more */
1145                 *skipped = 1;
1146                 return sync_blocks;
1147         }
1148         /*
1149          * If there is non-resync activity waiting for us then
1150          * put in a delay to throttle resync.
1151          */
1152         if (!go_faster && waitqueue_active(&conf->wait_resume))
1153                 msleep_interruptible(1000);
1154         device_barrier(conf, sector_nr + RESYNC_SECTORS);
1155
1156         /*
1157          * If reconstructing, and >1 working disc,
1158          * could dedicate one to rebuild and others to
1159          * service read requests ..
1160          */
1161         disk = conf->last_used;
1162         /* make sure disk is operational */
1163
1164         while (conf->mirrors[disk].rdev == NULL ||
1165                !conf->mirrors[disk].rdev->in_sync) {
1166                 if (disk <= 0)
1167                         disk = conf->raid_disks;
1168                 disk--;
1169                 if (disk == conf->last_used)
1170                         break;
1171         }
1172         conf->last_used = disk;
1173         atomic_inc(&conf->mirrors[disk].rdev->nr_pending);
1174
1175
1176         mirror = conf->mirrors + disk;
1177
1178         r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1179
1180         spin_lock_irq(&conf->resync_lock);
1181         conf->nr_pending++;
1182         spin_unlock_irq(&conf->resync_lock);
1183
1184         r1_bio->mddev = mddev;
1185         r1_bio->sector = sector_nr;
1186         r1_bio->state = 0;
1187         set_bit(R1BIO_IsSync, &r1_bio->state);
1188         r1_bio->read_disk = disk;
1189
1190         for (i=0; i < conf->raid_disks; i++) {
1191                 bio = r1_bio->bios[i];
1192
1193                 /* take from bio_init */
1194                 bio->bi_next = NULL;
1195                 bio->bi_flags |= 1 << BIO_UPTODATE;
1196                 bio->bi_rw = 0;
1197                 bio->bi_vcnt = 0;
1198                 bio->bi_idx = 0;
1199                 bio->bi_phys_segments = 0;
1200                 bio->bi_hw_segments = 0;
1201                 bio->bi_size = 0;
1202                 bio->bi_end_io = NULL;
1203                 bio->bi_private = NULL;
1204
1205                 if (i == disk) {
1206                         bio->bi_rw = READ;
1207                         bio->bi_end_io = end_sync_read;
1208                 } else if (conf->mirrors[i].rdev == NULL ||
1209                            conf->mirrors[i].rdev->faulty) {
1210                         still_degraded = 1;
1211                         continue;
1212                 } else if (!conf->mirrors[i].rdev->in_sync ||
1213                            sector_nr + RESYNC_SECTORS > mddev->recovery_cp) {
1214                         bio->bi_rw = WRITE;
1215                         bio->bi_end_io = end_sync_write;
1216                         write_targets ++;
1217                 } else
1218                         /* no need to read or write here */
1219                         continue;
1220                 bio->bi_sector = sector_nr + conf->mirrors[i].rdev->data_offset;
1221                 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1222                 bio->bi_private = r1_bio;
1223         }
1224
1225         if (write_targets == 0) {
1226                 /* There is nowhere to write, so all non-sync
1227                  * drives must be failed - so we are finished
1228                  */
1229                 sector_t rv = max_sector - sector_nr;
1230                 *skipped = 1;
1231                 put_buf(r1_bio);
1232                 rdev_dec_pending(conf->mirrors[disk].rdev, mddev);
1233                 return rv;
1234         }
1235
1236         nr_sectors = 0;
1237         sync_blocks = 0;
1238         do {
1239                 struct page *page;
1240                 int len = PAGE_SIZE;
1241                 if (sector_nr + (len>>9) > max_sector)
1242                         len = (max_sector - sector_nr) << 9;
1243                 if (len == 0)
1244                         break;
1245                 if (sync_blocks == 0) {
1246                         if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1247                                         &sync_blocks, still_degraded) &&
1248                                         !conf->fullsync)
1249                                 break;
1250                         if (sync_blocks < (PAGE_SIZE>>9))
1251                                 BUG();
1252                         if (len > (sync_blocks<<9))
1253                                 len = sync_blocks<<9;
1254                 }
1255
1256                 for (i=0 ; i < conf->raid_disks; i++) {
1257                         bio = r1_bio->bios[i];
1258                         if (bio->bi_end_io) {
1259                                 page = r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page;
1260                                 if (bio_add_page(bio, page, len, 0) == 0) {
1261                                         /* stop here */
1262                                         r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page = page;
1263                                         while (i > 0) {
1264                                                 i--;
1265                                                 bio = r1_bio->bios[i];
1266                                                 if (bio->bi_end_io==NULL)
1267                                                         continue;
1268                                                 /* remove last page from this bio */
1269                                                 bio->bi_vcnt--;
1270                                                 bio->bi_size -= len;
1271                                                 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1272                                         }
1273                                         goto bio_full;
1274                                 }
1275                         }
1276                 }
1277                 nr_sectors += len>>9;
1278                 sector_nr += len>>9;
1279                 sync_blocks -= (len>>9);
1280         } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1281  bio_full:
1282         bio = r1_bio->bios[disk];
1283         r1_bio->sectors = nr_sectors;
1284
1285         md_sync_acct(mirror->rdev->bdev, nr_sectors);
1286
1287         generic_make_request(bio);
1288
1289         return nr_sectors;
1290 }
1291
1292 static int run(mddev_t *mddev)
1293 {
1294         conf_t *conf;
1295         int i, j, disk_idx;
1296         mirror_info_t *disk;
1297         mdk_rdev_t *rdev;
1298         struct list_head *tmp;
1299
1300         if (mddev->level != 1) {
1301                 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1302                        mdname(mddev), mddev->level);
1303                 goto out;
1304         }
1305         /*
1306          * copy the already verified devices into our private RAID1
1307          * bookkeeping area. [whatever we allocate in run(),
1308          * should be freed in stop()]
1309          */
1310         conf = kmalloc(sizeof(conf_t), GFP_KERNEL);
1311         mddev->private = conf;
1312         if (!conf)
1313                 goto out_no_mem;
1314
1315         memset(conf, 0, sizeof(*conf));
1316         conf->mirrors = kmalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1317                                  GFP_KERNEL);
1318         if (!conf->mirrors)
1319                 goto out_no_mem;
1320
1321         memset(conf->mirrors, 0, sizeof(struct mirror_info)*mddev->raid_disks);
1322
1323         conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1324         if (!conf->poolinfo)
1325                 goto out_no_mem;
1326         conf->poolinfo->mddev = mddev;
1327         conf->poolinfo->raid_disks = mddev->raid_disks;
1328         conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1329                                           r1bio_pool_free,
1330                                           conf->poolinfo);
1331         if (!conf->r1bio_pool)
1332                 goto out_no_mem;
1333
1334         ITERATE_RDEV(mddev, rdev, tmp) {
1335                 disk_idx = rdev->raid_disk;
1336                 if (disk_idx >= mddev->raid_disks
1337                     || disk_idx < 0)
1338                         continue;
1339                 disk = conf->mirrors + disk_idx;
1340
1341                 disk->rdev = rdev;
1342
1343                 blk_queue_stack_limits(mddev->queue,
1344                                        rdev->bdev->bd_disk->queue);
1345                 /* as we don't honour merge_bvec_fn, we must never risk
1346                  * violating it, so limit ->max_sector to one PAGE, as
1347                  * a one page request is never in violation.
1348                  */
1349                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1350                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
1351                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1352
1353                 disk->head_position = 0;
1354                 if (!rdev->faulty && rdev->in_sync)
1355                         conf->working_disks++;
1356         }
1357         conf->raid_disks = mddev->raid_disks;
1358         conf->mddev = mddev;
1359         spin_lock_init(&conf->device_lock);
1360         INIT_LIST_HEAD(&conf->retry_list);
1361         if (conf->working_disks == 1)
1362                 mddev->recovery_cp = MaxSector;
1363
1364         spin_lock_init(&conf->resync_lock);
1365         init_waitqueue_head(&conf->wait_idle);
1366         init_waitqueue_head(&conf->wait_resume);
1367
1368         bio_list_init(&conf->pending_bio_list);
1369         bio_list_init(&conf->flushing_bio_list);
1370
1371         if (!conf->working_disks) {
1372                 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1373                         mdname(mddev));
1374                 goto out_free_conf;
1375         }
1376
1377         mddev->degraded = 0;
1378         for (i = 0; i < conf->raid_disks; i++) {
1379
1380                 disk = conf->mirrors + i;
1381
1382                 if (!disk->rdev) {
1383                         disk->head_position = 0;
1384                         mddev->degraded++;
1385                 }
1386         }
1387
1388         /*
1389          * find the first working one and use it as a starting point
1390          * to read balancing.
1391          */
1392         for (j = 0; j < conf->raid_disks &&
1393                      (!conf->mirrors[j].rdev ||
1394                       !conf->mirrors[j].rdev->in_sync) ; j++)
1395                 /* nothing */;
1396         conf->last_used = j;
1397
1398
1399         mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1400         if (!mddev->thread) {
1401                 printk(KERN_ERR
1402                        "raid1: couldn't allocate thread for %s\n",
1403                        mdname(mddev));
1404                 goto out_free_conf;
1405         }
1406         if (mddev->bitmap) mddev->thread->timeout = mddev->bitmap->daemon_sleep * HZ;
1407
1408         printk(KERN_INFO
1409                 "raid1: raid set %s active with %d out of %d mirrors\n",
1410                 mdname(mddev), mddev->raid_disks - mddev->degraded,
1411                 mddev->raid_disks);
1412         /*
1413          * Ok, everything is just fine now
1414          */
1415         mddev->array_size = mddev->size;
1416
1417         mddev->queue->unplug_fn = raid1_unplug;
1418         mddev->queue->issue_flush_fn = raid1_issue_flush;
1419
1420         return 0;
1421
1422 out_no_mem:
1423         printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1424                mdname(mddev));
1425
1426 out_free_conf:
1427         if (conf) {
1428                 if (conf->r1bio_pool)
1429                         mempool_destroy(conf->r1bio_pool);
1430                 kfree(conf->mirrors);
1431                 kfree(conf->poolinfo);
1432                 kfree(conf);
1433                 mddev->private = NULL;
1434         }
1435 out:
1436         return -EIO;
1437 }
1438
1439 static int stop(mddev_t *mddev)
1440 {
1441         conf_t *conf = mddev_to_conf(mddev);
1442
1443         md_unregister_thread(mddev->thread);
1444         mddev->thread = NULL;
1445         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
1446         if (conf->r1bio_pool)
1447                 mempool_destroy(conf->r1bio_pool);
1448         kfree(conf->mirrors);
1449         kfree(conf->poolinfo);
1450         kfree(conf);
1451         mddev->private = NULL;
1452         return 0;
1453 }
1454
1455 static int raid1_resize(mddev_t *mddev, sector_t sectors)
1456 {
1457         /* no resync is happening, and there is enough space
1458          * on all devices, so we can resize.
1459          * We need to make sure resync covers any new space.
1460          * If the array is shrinking we should possibly wait until
1461          * any io in the removed space completes, but it hardly seems
1462          * worth it.
1463          */
1464         mddev->array_size = sectors>>1;
1465         set_capacity(mddev->gendisk, mddev->array_size << 1);
1466         mddev->changed = 1;
1467         if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
1468                 mddev->recovery_cp = mddev->size << 1;
1469                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1470         }
1471         mddev->size = mddev->array_size;
1472         mddev->resync_max_sectors = sectors;
1473         return 0;
1474 }
1475
1476 static int raid1_reshape(mddev_t *mddev, int raid_disks)
1477 {
1478         /* We need to:
1479          * 1/ resize the r1bio_pool
1480          * 2/ resize conf->mirrors
1481          *
1482          * We allocate a new r1bio_pool if we can.
1483          * Then raise a device barrier and wait until all IO stops.
1484          * Then resize conf->mirrors and swap in the new r1bio pool.
1485          *
1486          * At the same time, we "pack" the devices so that all the missing
1487          * devices have the higher raid_disk numbers.
1488          */
1489         mempool_t *newpool, *oldpool;
1490         struct pool_info *newpoolinfo;
1491         mirror_info_t *newmirrors;
1492         conf_t *conf = mddev_to_conf(mddev);
1493         int cnt;
1494
1495         int d, d2;
1496
1497         if (raid_disks < conf->raid_disks) {
1498                 cnt=0;
1499                 for (d= 0; d < conf->raid_disks; d++)
1500                         if (conf->mirrors[d].rdev)
1501                                 cnt++;
1502                 if (cnt > raid_disks)
1503                         return -EBUSY;
1504         }
1505
1506         newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
1507         if (!newpoolinfo)
1508                 return -ENOMEM;
1509         newpoolinfo->mddev = mddev;
1510         newpoolinfo->raid_disks = raid_disks;
1511
1512         newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1513                                  r1bio_pool_free, newpoolinfo);
1514         if (!newpool) {
1515                 kfree(newpoolinfo);
1516                 return -ENOMEM;
1517         }
1518         newmirrors = kmalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
1519         if (!newmirrors) {
1520                 kfree(newpoolinfo);
1521                 mempool_destroy(newpool);
1522                 return -ENOMEM;
1523         }
1524         memset(newmirrors, 0, sizeof(struct mirror_info)*raid_disks);
1525
1526         spin_lock_irq(&conf->resync_lock);
1527         conf->barrier++;
1528         wait_event_lock_irq(conf->wait_idle, !conf->nr_pending,
1529                             conf->resync_lock, raid1_unplug(mddev->queue));
1530         spin_unlock_irq(&conf->resync_lock);
1531
1532         /* ok, everything is stopped */
1533         oldpool = conf->r1bio_pool;
1534         conf->r1bio_pool = newpool;
1535
1536         for (d=d2=0; d < conf->raid_disks; d++)
1537                 if (conf->mirrors[d].rdev) {
1538                         conf->mirrors[d].rdev->raid_disk = d2;
1539                         newmirrors[d2++].rdev = conf->mirrors[d].rdev;
1540                 }
1541         kfree(conf->mirrors);
1542         conf->mirrors = newmirrors;
1543         kfree(conf->poolinfo);
1544         conf->poolinfo = newpoolinfo;
1545
1546         mddev->degraded += (raid_disks - conf->raid_disks);
1547         conf->raid_disks = mddev->raid_disks = raid_disks;
1548
1549         conf->last_used = 0; /* just make sure it is in-range */
1550         spin_lock_irq(&conf->resync_lock);
1551         conf->barrier--;
1552         spin_unlock_irq(&conf->resync_lock);
1553         wake_up(&conf->wait_resume);
1554         wake_up(&conf->wait_idle);
1555
1556
1557         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1558         md_wakeup_thread(mddev->thread);
1559
1560         mempool_destroy(oldpool);
1561         return 0;
1562 }
1563
1564
1565 static mdk_personality_t raid1_personality =
1566 {
1567         .name           = "raid1",
1568         .owner          = THIS_MODULE,
1569         .make_request   = make_request,
1570         .run            = run,
1571         .stop           = stop,
1572         .status         = status,
1573         .error_handler  = error,
1574         .hot_add_disk   = raid1_add_disk,
1575         .hot_remove_disk= raid1_remove_disk,
1576         .spare_active   = raid1_spare_active,
1577         .sync_request   = sync_request,
1578         .resize         = raid1_resize,
1579         .reshape        = raid1_reshape,
1580 };
1581
1582 static int __init raid_init(void)
1583 {
1584         return register_md_personality(RAID1, &raid1_personality);
1585 }
1586
1587 static void raid_exit(void)
1588 {
1589         unregister_md_personality(RAID1);
1590 }
1591
1592 module_init(raid_init);
1593 module_exit(raid_exit);
1594 MODULE_LICENSE("GPL");
1595 MODULE_ALIAS("md-personality-3"); /* RAID1 */