fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/random.h>
  23 #include <linux/iocontext.h>
  24 #include <asm/div64.h>
  25 #include "compat.h"
  26 #include "ctree.h"
  27 #include "extent_map.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "print-tree.h"
  31 #include "volumes.h"
  32 #include "async-thread.h"
  33
  34 struct map_lookup {
  35         u64 type;
  36         int io_align;
  37         int io_width;
  38         int stripe_len;
  39         int sector_size;
  40         int num_stripes;
  41         int sub_stripes;
  42         struct btrfs_bio_stripe stripes[];
  43 };
  44
  45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  46                                 struct btrfs_root *root,
  47                                 struct btrfs_device *device);
  48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  49
  50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  51                             (sizeof(struct btrfs_bio_stripe) * (n)))
  52
  53 static DEFINE_MUTEX(uuid_mutex);
  54 static LIST_HEAD(fs_uuids);
  55
  56 void btrfs_lock_volumes(void)
  57 {
  58         mutex_lock(&uuid_mutex);
  59 }
  60
  61 void btrfs_unlock_volumes(void)
  62 {
  63         mutex_unlock(&uuid_mutex);
  64 }
  65
  66 static void lock_chunks(struct btrfs_root *root)
  67 {
  68         mutex_lock(&root->fs_info->chunk_mutex);
  69 }
  70
  71 static void unlock_chunks(struct btrfs_root *root)
  72 {
  73         mutex_unlock(&root->fs_info->chunk_mutex);
  74 }
  75
  76 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  77 {
  78         struct btrfs_device *device;
  79         WARN_ON(fs_devices->opened);
  80         while (!list_empty(&fs_devices->devices)) {
  81                 device = list_entry(fs_devices->devices.next,
  82                                     struct btrfs_device, dev_list);
  83                 list_del(&device->dev_list);
  84                 kfree(device->name);
  85                 kfree(device);
  86         }
  87         kfree(fs_devices);
  88 }
  89
  90 int btrfs_cleanup_fs_uuids(void)
  91 {
  92         struct btrfs_fs_devices *fs_devices;
  93
  94         while (!list_empty(&fs_uuids)) {
  95                 fs_devices = list_entry(fs_uuids.next,
  96                                         struct btrfs_fs_devices, list);
  97                 list_del(&fs_devices->list);
  98                 free_fs_devices(fs_devices);
  99         }
 100         return 0;
 101 }
 102
 103 static noinline struct btrfs_device *__find_device(struct list_head *head,
 104                                                    u64 devid, u8 *uuid)
 105 {
 106         struct btrfs_device *dev;
 107
 108         list_for_each_entry(dev, head, dev_list) {
 109                 if (dev->devid == devid &&
 110                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 111                         return dev;
 112                 }
 113         }
 114         return NULL;
 115 }
 116
 117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 118 {
 119         struct btrfs_fs_devices *fs_devices;
 120
 121         list_for_each_entry(fs_devices, &fs_uuids, list) {
 122                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 123                         return fs_devices;
 124         }
 125         return NULL;
 126 }
 127
 128 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 129                         struct bio *head, struct bio *tail)
 130 {
 131
 132         struct bio *old_head;
 133
 134         old_head = pending_bios->head;
 135         pending_bios->head = head;
 136         if (pending_bios->tail)
 137                 tail->bi_next = old_head;
 138         else
 139                 pending_bios->tail = tail;
 140 }
 141
 142 /*
 143  * we try to collect pending bios for a device so we don't get a large
 144  * number of procs sending bios down to the same device.  This greatly
 145  * improves the schedulers ability to collect and merge the bios.
 146  *
 147  * But, it also turns into a long list of bios to process and that is sure
 148  * to eventually make the worker thread block.  The solution here is to
 149  * make some progress and then put this work struct back at the end of
 150  * the list if the block device is congested.  This way, multiple devices
 151  * can make progress from a single worker thread.
 152  */
 153 static noinline int run_scheduled_bios(struct btrfs_device *device)
 154 {
 155         struct bio *pending;
 156         struct backing_dev_info *bdi;
 157         struct btrfs_fs_info *fs_info;
 158         struct btrfs_pending_bios *pending_bios;
 159         struct bio *tail;
 160         struct bio *cur;
 161         int again = 0;
 162         unsigned long num_run;
 163         unsigned long num_sync_run;
 164         unsigned long batch_run = 0;
 165         unsigned long limit;
 166         unsigned long last_waited = 0;
 167         int force_reg = 0;
 168
 169         bdi = blk_get_backing_dev_info(device->bdev);
 170         fs_info = device->dev_root->fs_info;
 171         limit = btrfs_async_submit_limit(fs_info);
 172         limit = limit * 2 / 3;
 173
 174         /* we want to make sure that every time we switch from the sync
 175          * list to the normal list, we unplug
 176          */
 177         num_sync_run = 0;
 178
 179 loop:
 180         spin_lock(&device->io_lock);
 181
 182 loop_lock:
 183         num_run = 0;
 184
 185         /* take all the bios off the list at once and process them
 186          * later on (without the lock held).  But, remember the
 187          * tail and other pointers so the bios can be properly reinserted
 188          * into the list if we hit congestion
 189          */
 190         if (!force_reg && device->pending_sync_bios.head) {
 191                 pending_bios = &device->pending_sync_bios;
 192                 force_reg = 1;
 193         } else {
 194                 pending_bios = &device->pending_bios;
 195                 force_reg = 0;
 196         }
 197
 198         pending = pending_bios->head;
 199         tail = pending_bios->tail;
 200         WARN_ON(pending && !tail);
 201
 202         /*
 203          * if pending was null this time around, no bios need processing
 204          * at all and we can stop.  Otherwise it'll loop back up again
 205          * and do an additional check so no bios are missed.
 206          *
 207          * device->running_pending is used to synchronize with the
 208          * schedule_bio code.
 209          */
 210         if (device->pending_sync_bios.head == NULL &&
 211             device->pending_bios.head == NULL) {
 212                 again = 0;
 213                 device->running_pending = 0;
 214         } else {
 215                 again = 1;
 216                 device->running_pending = 1;
 217         }
 218
 219         pending_bios->head = NULL;
 220         pending_bios->tail = NULL;
 221
 222         spin_unlock(&device->io_lock);
 223
 224         /*
 225          * if we're doing the regular priority list, make sure we unplug
 226          * for any high prio bios we've sent down
 227          */
 228         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
 229                 num_sync_run = 0;
 230                 blk_run_backing_dev(bdi, NULL);
 231         }
 232
 233         while (pending) {
 234
 235                 rmb();
 236                 /* we want to work on both lists, but do more bios on the
 237                  * sync list than the regular list
 238                  */
 239                 if ((num_run > 32 &&
 240                     pending_bios != &device->pending_sync_bios &&
 241                     device->pending_sync_bios.head) ||
 242                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 243                     device->pending_bios.head)) {
 244                         spin_lock(&device->io_lock);
 245                         requeue_list(pending_bios, pending, tail);
 246                         goto loop_lock;
 247                 }
 248
 249                 cur = pending;
 250                 pending = pending->bi_next;
 251                 cur->bi_next = NULL;
 252                 atomic_dec(&fs_info->nr_async_bios);
 253
 254                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 255                     waitqueue_active(&fs_info->async_submit_wait))
 256                         wake_up(&fs_info->async_submit_wait);
 257
 258                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 259                 submit_bio(cur->bi_rw, cur);
 260                 num_run++;
 261                 batch_run++;
 262
 263                 if (bio_sync(cur))
 264                         num_sync_run++;
 265
 266                 if (need_resched()) {
 267                         if (num_sync_run) {
 268                                 blk_run_backing_dev(bdi, NULL);
 269                                 num_sync_run = 0;
 270                         }
 271                         cond_resched();
 272                 }
 273
 274                 /*
 275                  * we made progress, there is more work to do and the bdi
 276                  * is now congested.  Back off and let other work structs
 277                  * run instead
 278                  */
 279                 if (pending && bdi_write_congested(bdi) && batch_run > 32 &&
 280                     fs_info->fs_devices->open_devices > 1) {
 281                         struct io_context *ioc;
 282
 283                         ioc = current->io_context;
 284
 285                         /*
 286                          * the main goal here is that we don't want to
 287                          * block if we're going to be able to submit
 288                          * more requests without blocking.
 289                          *
 290                          * This code does two great things, it pokes into
 291                          * the elevator code from a filesystem _and_
 292                          * it makes assumptions about how batching works.
 293                          */
 294                         if (ioc && ioc->nr_batch_requests > 0 &&
 295                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 296                             (last_waited == 0 ||
 297                              ioc->last_waited == last_waited)) {
 298                                 /*
 299                                  * we want to go through our batch of
 300                                  * requests and stop.  So, we copy out
 301                                  * the ioc->last_waited time and test
 302                                  * against it before looping
 303                                  */
 304                                 last_waited = ioc->last_waited;
 305                                 if (need_resched()) {
 306                                         if (num_sync_run) {
 307                                                 blk_run_backing_dev(bdi, NULL);
 308                                                 num_sync_run = 0;
 309                                         }
 310                                         cond_resched();
 311                                 }
 312                                 continue;
 313                         }
 314                         spin_lock(&device->io_lock);
 315                         requeue_list(pending_bios, pending, tail);
 316                         device->running_pending = 1;
 317
 318                         spin_unlock(&device->io_lock);
 319                         btrfs_requeue_work(&device->work);
 320                         goto done;
 321                 }
 322         }
 323
 324         if (num_sync_run) {
 325                 num_sync_run = 0;
 326                 blk_run_backing_dev(bdi, NULL);
 327         }
 328
 329         cond_resched();
 330         if (again)
 331                 goto loop;
 332
 333         spin_lock(&device->io_lock);
 334         if (device->pending_bios.head || device->pending_sync_bios.head)
 335                 goto loop_lock;
 336         spin_unlock(&device->io_lock);
 337
 338         /*
 339          * IO has already been through a long path to get here.  Checksumming,
 340          * async helper threads, perhaps compression.  We've done a pretty
 341          * good job of collecting a batch of IO and should just unplug
 342          * the device right away.
 343          *
 344          * This will help anyone who is waiting on the IO, they might have
 345          * already unplugged, but managed to do so before the bio they
 346          * cared about found its way down here.
 347          */
 348         blk_run_backing_dev(bdi, NULL);
 349 done:
 350         return 0;
 351 }
 352
 353 static void pending_bios_fn(struct btrfs_work *work)
 354 {
 355         struct btrfs_device *device;
 356
 357         device = container_of(work, struct btrfs_device, work);
 358         run_scheduled_bios(device);
 359 }
 360
 361 static noinline int device_list_add(const char *path,
 362                            struct btrfs_super_block *disk_super,
 363                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 364 {
 365         struct btrfs_device *device;
 366         struct btrfs_fs_devices *fs_devices;
 367         u64 found_transid = btrfs_super_generation(disk_super);
 368
 369         fs_devices = find_fsid(disk_super->fsid);
 370         if (!fs_devices) {
 371                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 372                 if (!fs_devices)
 373                         return -ENOMEM;
 374                 INIT_LIST_HEAD(&fs_devices->devices);
 375                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 376                 list_add(&fs_devices->list, &fs_uuids);
 377                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 378                 fs_devices->latest_devid = devid;
 379                 fs_devices->latest_trans = found_transid;
 380                 device = NULL;
 381         } else {
 382                 device = __find_device(&fs_devices->devices, devid,
 383                                        disk_super->dev_item.uuid);
 384         }
 385         if (!device) {
 386                 if (fs_devices->opened)
 387                         return -EBUSY;
 388
 389                 device = kzalloc(sizeof(*device), GFP_NOFS);
 390                 if (!device) {
 391                         /* we can safely leave the fs_devices entry around */
 392                         return -ENOMEM;
 393                 }
 394                 device->devid = devid;
 395                 device->work.func = pending_bios_fn;
 396                 memcpy(device->uuid, disk_super->dev_item.uuid,
 397                        BTRFS_UUID_SIZE);
 398                 device->barriers = 1;
 399                 spin_lock_init(&device->io_lock);
 400                 device->name = kstrdup(path, GFP_NOFS);
 401                 if (!device->name) {
 402                         kfree(device);
 403                         return -ENOMEM;
 404                 }
 405                 INIT_LIST_HEAD(&device->dev_alloc_list);
 406                 list_add(&device->dev_list, &fs_devices->devices);
 407                 device->fs_devices = fs_devices;
 408                 fs_devices->num_devices++;
 409         }
 410
 411         if (found_transid > fs_devices->latest_trans) {
 412                 fs_devices->latest_devid = devid;
 413                 fs_devices->latest_trans = found_transid;
 414         }
 415         *fs_devices_ret = fs_devices;
 416         return 0;
 417 }
 418
 419 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 420 {
 421         struct btrfs_fs_devices *fs_devices;
 422         struct btrfs_device *device;
 423         struct btrfs_device *orig_dev;
 424
 425         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 426         if (!fs_devices)
 427                 return ERR_PTR(-ENOMEM);
 428
 429         INIT_LIST_HEAD(&fs_devices->devices);
 430         INIT_LIST_HEAD(&fs_devices->alloc_list);
 431         INIT_LIST_HEAD(&fs_devices->list);
 432         fs_devices->latest_devid = orig->latest_devid;
 433         fs_devices->latest_trans = orig->latest_trans;
 434         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 435
 436         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 437                 device = kzalloc(sizeof(*device), GFP_NOFS);
 438                 if (!device)
 439                         goto error;
 440
 441                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 442                 if (!device->name)
 443                         goto error;
 444
 445                 device->devid = orig_dev->devid;
 446                 device->work.func = pending_bios_fn;
 447                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 448                 device->barriers = 1;
 449                 spin_lock_init(&device->io_lock);
 450                 INIT_LIST_HEAD(&device->dev_list);
 451                 INIT_LIST_HEAD(&device->dev_alloc_list);
 452
 453                 list_add(&device->dev_list, &fs_devices->devices);
 454                 device->fs_devices = fs_devices;
 455                 fs_devices->num_devices++;
 456         }
 457         return fs_devices;
 458 error:
 459         free_fs_devices(fs_devices);
 460         return ERR_PTR(-ENOMEM);
 461 }
 462
 463 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 464 {
 465         struct btrfs_device *device, *next;
 466
 467         mutex_lock(&uuid_mutex);
 468 again:
 469         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 470                 if (device->in_fs_metadata)
 471                         continue;
 472
 473                 if (device->bdev) {
 474                         close_bdev_exclusive(device->bdev, device->mode);
 475                         device->bdev = NULL;
 476                         fs_devices->open_devices--;
 477                 }
 478                 if (device->writeable) {
 479                         list_del_init(&device->dev_alloc_list);
 480                         device->writeable = 0;
 481                         fs_devices->rw_devices--;
 482                 }
 483                 list_del_init(&device->dev_list);
 484                 fs_devices->num_devices--;
 485                 kfree(device->name);
 486                 kfree(device);
 487         }
 488
 489         if (fs_devices->seed) {
 490                 fs_devices = fs_devices->seed;
 491                 goto again;
 492         }
 493
 494         mutex_unlock(&uuid_mutex);
 495         return 0;
 496 }
 497
 498 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 499 {
 500         struct btrfs_device *device;
 501
 502         if (--fs_devices->opened > 0)
 503                 return 0;
 504
 505         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 506                 if (device->bdev) {
 507                         close_bdev_exclusive(device->bdev, device->mode);
 508                         fs_devices->open_devices--;
 509                 }
 510                 if (device->writeable) {
 511                         list_del_init(&device->dev_alloc_list);
 512                         fs_devices->rw_devices--;
 513                 }
 514
 515                 device->bdev = NULL;
 516                 device->writeable = 0;
 517                 device->in_fs_metadata = 0;
 518         }
 519         WARN_ON(fs_devices->open_devices);
 520         WARN_ON(fs_devices->rw_devices);
 521         fs_devices->opened = 0;
 522         fs_devices->seeding = 0;
 523
 524         return 0;
 525 }
 526
 527 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 528 {
 529         struct btrfs_fs_devices *seed_devices = NULL;
 530         int ret;
 531
 532         mutex_lock(&uuid_mutex);
 533         ret = __btrfs_close_devices(fs_devices);
 534         if (!fs_devices->opened) {
 535                 seed_devices = fs_devices->seed;
 536                 fs_devices->seed = NULL;
 537         }
 538         mutex_unlock(&uuid_mutex);
 539
 540         while (seed_devices) {
 541                 fs_devices = seed_devices;
 542                 seed_devices = fs_devices->seed;
 543                 __btrfs_close_devices(fs_devices);
 544                 free_fs_devices(fs_devices);
 545         }
 546         return ret;
 547 }
 548
 549 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 550                                 fmode_t flags, void *holder)
 551 {
 552         struct block_device *bdev;
 553         struct list_head *head = &fs_devices->devices;
 554         struct btrfs_device *device;
 555         struct block_device *latest_bdev = NULL;
 556         struct buffer_head *bh;
 557         struct btrfs_super_block *disk_super;
 558         u64 latest_devid = 0;
 559         u64 latest_transid = 0;
 560         u64 devid;
 561         int seeding = 1;
 562         int ret = 0;
 563
 564         list_for_each_entry(device, head, dev_list) {
 565                 if (device->bdev)
 566                         continue;
 567                 if (!device->name)
 568                         continue;
 569
 570                 bdev = open_bdev_exclusive(device->name, flags, holder);
 571                 if (IS_ERR(bdev)) {
 572                         printk(KERN_INFO "open %s failed\n", device->name);
 573                         goto error;
 574                 }
 575                 set_blocksize(bdev, 4096);
 576
 577                 bh = btrfs_read_dev_super(bdev);
 578                 if (!bh)
 579                         goto error_close;
 580
 581                 disk_super = (struct btrfs_super_block *)bh->b_data;
 582                 devid = le64_to_cpu(disk_super->dev_item.devid);
 583                 if (devid != device->devid)
 584                         goto error_brelse;
 585
 586                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 587                            BTRFS_UUID_SIZE))
 588                         goto error_brelse;
 589
 590                 device->generation = btrfs_super_generation(disk_super);
 591                 if (!latest_transid || device->generation > latest_transid) {
 592                         latest_devid = devid;
 593                         latest_transid = device->generation;
 594                         latest_bdev = bdev;
 595                 }
 596
 597                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 598                         device->writeable = 0;
 599                 } else {
 600                         device->writeable = !bdev_read_only(bdev);
 601                         seeding = 0;
 602                 }
 603
 604                 device->bdev = bdev;
 605                 device->in_fs_metadata = 0;
 606                 device->mode = flags;
 607
 608                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 609                         fs_devices->rotating = 1;
 610
 611                 fs_devices->open_devices++;
 612                 if (device->writeable) {
 613                         fs_devices->rw_devices++;
 614                         list_add(&device->dev_alloc_list,
 615                                  &fs_devices->alloc_list);
 616                 }
 617                 continue;
 618
 619 error_brelse:
 620                 brelse(bh);
 621 error_close:
 622                 close_bdev_exclusive(bdev, FMODE_READ);
 623 error:
 624                 continue;
 625         }
 626         if (fs_devices->open_devices == 0) {
 627                 ret = -EIO;
 628                 goto out;
 629         }
 630         fs_devices->seeding = seeding;
 631         fs_devices->opened = 1;
 632         fs_devices->latest_bdev = latest_bdev;
 633         fs_devices->latest_devid = latest_devid;
 634         fs_devices->latest_trans = latest_transid;
 635         fs_devices->total_rw_bytes = 0;
 636 out:
 637         return ret;
 638 }
 639
 640 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 641                        fmode_t flags, void *holder)
 642 {
 643         int ret;
 644
 645         mutex_lock(&uuid_mutex);
 646         if (fs_devices->opened) {
 647                 fs_devices->opened++;
 648                 ret = 0;
 649         } else {
 650                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 651         }
 652         mutex_unlock(&uuid_mutex);
 653         return ret;
 654 }
 655
 656 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 657                           struct btrfs_fs_devices **fs_devices_ret)
 658 {
 659         struct btrfs_super_block *disk_super;
 660         struct block_device *bdev;
 661         struct buffer_head *bh;
 662         int ret;
 663         u64 devid;
 664         u64 transid;
 665
 666         mutex_lock(&uuid_mutex);
 667
 668         bdev = open_bdev_exclusive(path, flags, holder);
 669
 670         if (IS_ERR(bdev)) {
 671                 ret = PTR_ERR(bdev);
 672                 goto error;
 673         }
 674
 675         ret = set_blocksize(bdev, 4096);
 676         if (ret)
 677                 goto error_close;
 678         bh = btrfs_read_dev_super(bdev);
 679         if (!bh) {
 680                 ret = -EIO;
 681                 goto error_close;
 682         }
 683         disk_super = (struct btrfs_super_block *)bh->b_data;
 684         devid = le64_to_cpu(disk_super->dev_item.devid);
 685         transid = btrfs_super_generation(disk_super);
 686         if (disk_super->label[0])
 687                 printk(KERN_INFO "device label %s ", disk_super->label);
 688         else {
 689                 /* FIXME, make a readl uuid parser */
 690                 printk(KERN_INFO "device fsid %llx-%llx ",
 691                        *(unsigned long long *)disk_super->fsid,
 692                        *(unsigned long long *)(disk_super->fsid + 8));
 693         }
 694         printk(KERN_CONT "devid %llu transid %llu %s\n",
 695                (unsigned long long)devid, (unsigned long long)transid, path);
 696         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 697
 698         brelse(bh);
 699 error_close:
 700         close_bdev_exclusive(bdev, flags);
 701 error:
 702         mutex_unlock(&uuid_mutex);
 703         return ret;
 704 }
 705
 706 /*
 707  * this uses a pretty simple search, the expectation is that it is
 708  * called very infrequently and that a given device has a small number
 709  * of extents
 710  */
 711 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
 712                                          struct btrfs_device *device,
 713                                          u64 num_bytes, u64 *start)
 714 {
 715         struct btrfs_key key;
 716         struct btrfs_root *root = device->dev_root;
 717         struct btrfs_dev_extent *dev_extent = NULL;
 718         struct btrfs_path *path;
 719         u64 hole_size = 0;
 720         u64 last_byte = 0;
 721         u64 search_start = 0;
 722         u64 search_end = device->total_bytes;
 723         int ret;
 724         int slot = 0;
 725         int start_found;
 726         struct extent_buffer *l;
 727
 728         path = btrfs_alloc_path();
 729         if (!path)
 730                 return -ENOMEM;
 731         path->reada = 2;
 732         start_found = 0;
 733
 734         /* FIXME use last free of some kind */
 735
 736         /* we don't want to overwrite the superblock on the drive,
 737          * so we make sure to start at an offset of at least 1MB
 738          */
 739         search_start = max((u64)1024 * 1024, search_start);
 740
 741         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 742                 search_start = max(root->fs_info->alloc_start, search_start);
 743
 744         key.objectid = device->devid;
 745         key.offset = search_start;
 746         key.type = BTRFS_DEV_EXTENT_KEY;
 747         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 748         if (ret < 0)
 749                 goto error;
 750         ret = btrfs_previous_item(root, path, 0, key.type);
 751         if (ret < 0)
 752                 goto error;
 753         l = path->nodes[0];
 754         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 755         while (1) {
 756                 l = path->nodes[0];
 757                 slot = path->slots[0];
 758                 if (slot >= btrfs_header_nritems(l)) {
 759                         ret = btrfs_next_leaf(root, path);
 760                         if (ret == 0)
 761                                 continue;
 762                         if (ret < 0)
 763                                 goto error;
 764 no_more_items:
 765                         if (!start_found) {
 766                                 if (search_start >= search_end) {
 767                                         ret = -ENOSPC;
 768                                         goto error;
 769                                 }
 770                                 *start = search_start;
 771                                 start_found = 1;
 772                                 goto check_pending;
 773                         }
 774                         *start = last_byte > search_start ?
 775                                 last_byte : search_start;
 776                         if (search_end <= *start) {
 777                                 ret = -ENOSPC;
 778                                 goto error;
 779                         }
 780                         goto check_pending;
 781                 }
 782                 btrfs_item_key_to_cpu(l, &key, slot);
 783
 784                 if (key.objectid < device->devid)
 785                         goto next;
 786
 787                 if (key.objectid > device->devid)
 788                         goto no_more_items;
 789
 790                 if (key.offset >= search_start && key.offset > last_byte &&
 791                     start_found) {
 792                         if (last_byte < search_start)
 793                                 last_byte = search_start;
 794                         hole_size = key.offset - last_byte;
 795                         if (key.offset > last_byte &&
 796                             hole_size >= num_bytes) {
 797                                 *start = last_byte;
 798                                 goto check_pending;
 799                         }
 800                 }
 801                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 802                         goto next;
 803
 804                 start_found = 1;
 805                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 806                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 807 next:
 808                 path->slots[0]++;
 809                 cond_resched();
 810         }
 811 check_pending:
 812         /* we have to make sure we didn't find an extent that has already
 813          * been allocated by the map tree or the original allocation
 814          */
 815         BUG_ON(*start < search_start);
 816
 817         if (*start + num_bytes > search_end) {
 818                 ret = -ENOSPC;
 819                 goto error;
 820         }
 821         /* check for pending inserts here */
 822         ret = 0;
 823
 824 error:
 825         btrfs_free_path(path);
 826         return ret;
 827 }
 828
 829 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 830                           struct btrfs_device *device,
 831                           u64 start)
 832 {
 833         int ret;
 834         struct btrfs_path *path;
 835         struct btrfs_root *root = device->dev_root;
 836         struct btrfs_key key;
 837         struct btrfs_key found_key;
 838         struct extent_buffer *leaf = NULL;
 839         struct btrfs_dev_extent *extent = NULL;
 840
 841         path = btrfs_alloc_path();
 842         if (!path)
 843                 return -ENOMEM;
 844
 845         key.objectid = device->devid;
 846         key.offset = start;
 847         key.type = BTRFS_DEV_EXTENT_KEY;
 848
 849         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 850         if (ret > 0) {
 851                 ret = btrfs_previous_item(root, path, key.objectid,
 852                                           BTRFS_DEV_EXTENT_KEY);
 853                 BUG_ON(ret);
 854                 leaf = path->nodes[0];
 855                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 856                 extent = btrfs_item_ptr(leaf, path->slots[0],
 857                                         struct btrfs_dev_extent);
 858                 BUG_ON(found_key.offset > start || found_key.offset +
 859                        btrfs_dev_extent_length(leaf, extent) < start);
 860                 ret = 0;
 861         } else if (ret == 0) {
 862                 leaf = path->nodes[0];
 863                 extent = btrfs_item_ptr(leaf, path->slots[0],
 864                                         struct btrfs_dev_extent);
 865         }
 866         BUG_ON(ret);
 867
 868         if (device->bytes_used > 0)
 869                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 870         ret = btrfs_del_item(trans, root, path);
 871         BUG_ON(ret);
 872
 873         btrfs_free_path(path);
 874         return ret;
 875 }
 876
 877 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 878                            struct btrfs_device *device,
 879                            u64 chunk_tree, u64 chunk_objectid,
 880                            u64 chunk_offset, u64 start, u64 num_bytes)
 881 {
 882         int ret;
 883         struct btrfs_path *path;
 884         struct btrfs_root *root = device->dev_root;
 885         struct btrfs_dev_extent *extent;
 886         struct extent_buffer *leaf;
 887         struct btrfs_key key;
 888
 889         WARN_ON(!device->in_fs_metadata);
 890         path = btrfs_alloc_path();
 891         if (!path)
 892                 return -ENOMEM;
 893
 894         key.objectid = device->devid;
 895         key.offset = start;
 896         key.type = BTRFS_DEV_EXTENT_KEY;
 897         ret = btrfs_insert_empty_item(trans, root, path, &key,
 898                                       sizeof(*extent));
 899         BUG_ON(ret);
 900
 901         leaf = path->nodes[0];
 902         extent = btrfs_item_ptr(leaf, path->slots[0],
 903                                 struct btrfs_dev_extent);
 904         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 905         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 906         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 907
 908         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 909                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 910                     BTRFS_UUID_SIZE);
 911
 912         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 913         btrfs_mark_buffer_dirty(leaf);
 914         btrfs_free_path(path);
 915         return ret;
 916 }
 917
 918 static noinline int find_next_chunk(struct btrfs_root *root,
 919                                     u64 objectid, u64 *offset)
 920 {
 921         struct btrfs_path *path;
 922         int ret;
 923         struct btrfs_key key;
 924         struct btrfs_chunk *chunk;
 925         struct btrfs_key found_key;
 926
 927         path = btrfs_alloc_path();
 928         BUG_ON(!path);
 929
 930         key.objectid = objectid;
 931         key.offset = (u64)-1;
 932         key.type = BTRFS_CHUNK_ITEM_KEY;
 933
 934         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 935         if (ret < 0)
 936                 goto error;
 937
 938         BUG_ON(ret == 0);
 939
 940         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 941         if (ret) {
 942                 *offset = 0;
 943         } else {
 944                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 945                                       path->slots[0]);
 946                 if (found_key.objectid != objectid)
 947                         *offset = 0;
 948                 else {
 949                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 950                                                struct btrfs_chunk);
 951                         *offset = found_key.offset +
 952                                 btrfs_chunk_length(path->nodes[0], chunk);
 953                 }
 954         }
 955         ret = 0;
 956 error:
 957         btrfs_free_path(path);
 958         return ret;
 959 }
 960
 961 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
 962 {
 963         int ret;
 964         struct btrfs_key key;
 965         struct btrfs_key found_key;
 966         struct btrfs_path *path;
 967
 968         root = root->fs_info->chunk_root;
 969
 970         path = btrfs_alloc_path();
 971         if (!path)
 972                 return -ENOMEM;
 973
 974         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 975         key.type = BTRFS_DEV_ITEM_KEY;
 976         key.offset = (u64)-1;
 977
 978         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 979         if (ret < 0)
 980                 goto error;
 981
 982         BUG_ON(ret == 0);
 983
 984         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 985                                   BTRFS_DEV_ITEM_KEY);
 986         if (ret) {
 987                 *objectid = 1;
 988         } else {
 989                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 990                                       path->slots[0]);
 991                 *objectid = found_key.offset + 1;
 992         }
 993         ret = 0;
 994 error:
 995         btrfs_free_path(path);
 996         return ret;
 997 }
 998
 999 /*
1000  * the device information is stored in the chunk root
1001  * the btrfs_device struct should be fully filled in
1002  */
1003 int btrfs_add_device(struct btrfs_trans_handle *trans,
1004                      struct btrfs_root *root,
1005                      struct btrfs_device *device)
1006 {
1007         int ret;
1008         struct btrfs_path *path;
1009         struct btrfs_dev_item *dev_item;
1010         struct extent_buffer *leaf;
1011         struct btrfs_key key;
1012         unsigned long ptr;
1013
1014         root = root->fs_info->chunk_root;
1015
1016         path = btrfs_alloc_path();
1017         if (!path)
1018                 return -ENOMEM;
1019
1020         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1021         key.type = BTRFS_DEV_ITEM_KEY;
1022         key.offset = device->devid;
1023
1024         ret = btrfs_insert_empty_item(trans, root, path, &key,
1025                                       sizeof(*dev_item));
1026         if (ret)
1027                 goto out;
1028
1029         leaf = path->nodes[0];
1030         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1031
1032         btrfs_set_device_id(leaf, dev_item, device->devid);
1033         btrfs_set_device_generation(leaf, dev_item, 0);
1034         btrfs_set_device_type(leaf, dev_item, device->type);
1035         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1036         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1037         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1038         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1039         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1040         btrfs_set_device_group(leaf, dev_item, 0);
1041         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1042         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1043         btrfs_set_device_start_offset(leaf, dev_item, 0);
1044
1045         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1046         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1047         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1048         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1049         btrfs_mark_buffer_dirty(leaf);
1050
1051         ret = 0;
1052 out:
1053         btrfs_free_path(path);
1054         return ret;
1055 }
1056
1057 static int btrfs_rm_dev_item(struct btrfs_root *root,
1058                              struct btrfs_device *device)
1059 {
1060         int ret;
1061         struct btrfs_path *path;
1062         struct btrfs_key key;
1063         struct btrfs_trans_handle *trans;
1064
1065         root = root->fs_info->chunk_root;
1066
1067         path = btrfs_alloc_path();
1068         if (!path)
1069                 return -ENOMEM;
1070
1071         trans = btrfs_start_transaction(root, 1);
1072         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1073         key.type = BTRFS_DEV_ITEM_KEY;
1074         key.offset = device->devid;
1075         lock_chunks(root);
1076
1077         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1078         if (ret < 0)
1079                 goto out;
1080
1081         if (ret > 0) {
1082                 ret = -ENOENT;
1083                 goto out;
1084         }
1085
1086         ret = btrfs_del_item(trans, root, path);
1087         if (ret)
1088                 goto out;
1089 out:
1090         btrfs_free_path(path);
1091         unlock_chunks(root);
1092         btrfs_commit_transaction(trans, root);
1093         return ret;
1094 }
1095
1096 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1097 {
1098         struct btrfs_device *device;
1099         struct btrfs_device *next_device;
1100         struct block_device *bdev;
1101         struct buffer_head *bh = NULL;
1102         struct btrfs_super_block *disk_super;
1103         u64 all_avail;
1104         u64 devid;
1105         u64 num_devices;
1106         u8 *dev_uuid;
1107         int ret = 0;
1108
1109         mutex_lock(&uuid_mutex);
1110         mutex_lock(&root->fs_info->volume_mutex);
1111
1112         all_avail = root->fs_info->avail_data_alloc_bits |
1113                 root->fs_info->avail_system_alloc_bits |
1114                 root->fs_info->avail_metadata_alloc_bits;
1115
1116         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1117             root->fs_info->fs_devices->rw_devices <= 4) {
1118                 printk(KERN_ERR "btrfs: unable to go below four devices "
1119                        "on raid10\n");
1120                 ret = -EINVAL;
1121                 goto out;
1122         }
1123
1124         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1125             root->fs_info->fs_devices->rw_devices <= 2) {
1126                 printk(KERN_ERR "btrfs: unable to go below two "
1127                        "devices on raid1\n");
1128                 ret = -EINVAL;
1129                 goto out;
1130         }
1131
1132         if (strcmp(device_path, "missing") == 0) {
1133                 struct list_head *devices;
1134                 struct btrfs_device *tmp;
1135
1136                 device = NULL;
1137                 devices = &root->fs_info->fs_devices->devices;
1138                 list_for_each_entry(tmp, devices, dev_list) {
1139                         if (tmp->in_fs_metadata && !tmp->bdev) {
1140                                 device = tmp;
1141                                 break;
1142                         }
1143                 }
1144                 bdev = NULL;
1145                 bh = NULL;
1146                 disk_super = NULL;
1147                 if (!device) {
1148                         printk(KERN_ERR "btrfs: no missing devices found to "
1149                                "remove\n");
1150                         goto out;
1151                 }
1152         } else {
1153                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1154                                       root->fs_info->bdev_holder);
1155                 if (IS_ERR(bdev)) {
1156                         ret = PTR_ERR(bdev);
1157                         goto out;
1158                 }
1159
1160                 set_blocksize(bdev, 4096);
1161                 bh = btrfs_read_dev_super(bdev);
1162                 if (!bh) {
1163                         ret = -EIO;
1164                         goto error_close;
1165                 }
1166                 disk_super = (struct btrfs_super_block *)bh->b_data;
1167                 devid = le64_to_cpu(disk_super->dev_item.devid);
1168                 dev_uuid = disk_super->dev_item.uuid;
1169                 device = btrfs_find_device(root, devid, dev_uuid,
1170                                            disk_super->fsid);
1171                 if (!device) {
1172                         ret = -ENOENT;
1173                         goto error_brelse;
1174                 }
1175         }
1176
1177         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1178                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1179                        "device\n");
1180                 ret = -EINVAL;
1181                 goto error_brelse;
1182         }
1183
1184         if (device->writeable) {
1185                 list_del_init(&device->dev_alloc_list);
1186                 root->fs_info->fs_devices->rw_devices--;
1187         }
1188
1189         ret = btrfs_shrink_device(device, 0);
1190         if (ret)
1191                 goto error_brelse;
1192
1193         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1194         if (ret)
1195                 goto error_brelse;
1196
1197         device->in_fs_metadata = 0;
1198         list_del_init(&device->dev_list);
1199         device->fs_devices->num_devices--;
1200
1201         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1202                                  struct btrfs_device, dev_list);
1203         if (device->bdev == root->fs_info->sb->s_bdev)
1204                 root->fs_info->sb->s_bdev = next_device->bdev;
1205         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1206                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1207
1208         if (device->bdev) {
1209                 close_bdev_exclusive(device->bdev, device->mode);
1210                 device->bdev = NULL;
1211                 device->fs_devices->open_devices--;
1212         }
1213
1214         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1215         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1216
1217         if (device->fs_devices->open_devices == 0) {
1218                 struct btrfs_fs_devices *fs_devices;
1219                 fs_devices = root->fs_info->fs_devices;
1220                 while (fs_devices) {
1221                         if (fs_devices->seed == device->fs_devices)
1222                                 break;
1223                         fs_devices = fs_devices->seed;
1224                 }
1225                 fs_devices->seed = device->fs_devices->seed;
1226                 device->fs_devices->seed = NULL;
1227                 __btrfs_close_devices(device->fs_devices);
1228                 free_fs_devices(device->fs_devices);
1229         }
1230
1231         /*
1232          * at this point, the device is zero sized.  We want to
1233          * remove it from the devices list and zero out the old super
1234          */
1235         if (device->writeable) {
1236                 /* make sure this device isn't detected as part of
1237                  * the FS anymore
1238                  */
1239                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1240                 set_buffer_dirty(bh);
1241                 sync_dirty_buffer(bh);
1242         }
1243
1244         kfree(device->name);
1245         kfree(device);
1246         ret = 0;
1247
1248 error_brelse:
1249         brelse(bh);
1250 error_close:
1251         if (bdev)
1252                 close_bdev_exclusive(bdev, FMODE_READ);
1253 out:
1254         mutex_unlock(&root->fs_info->volume_mutex);
1255         mutex_unlock(&uuid_mutex);
1256         return ret;
1257 }
1258
1259 /*
1260  * does all the dirty work required for changing file system's UUID.
1261  */
1262 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1263                                 struct btrfs_root *root)
1264 {
1265         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1266         struct btrfs_fs_devices *old_devices;
1267         struct btrfs_fs_devices *seed_devices;
1268         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1269         struct btrfs_device *device;
1270         u64 super_flags;
1271
1272         BUG_ON(!mutex_is_locked(&uuid_mutex));
1273         if (!fs_devices->seeding)
1274                 return -EINVAL;
1275
1276         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1277         if (!seed_devices)
1278                 return -ENOMEM;
1279
1280         old_devices = clone_fs_devices(fs_devices);
1281         if (IS_ERR(old_devices)) {
1282                 kfree(seed_devices);
1283                 return PTR_ERR(old_devices);
1284         }
1285
1286         list_add(&old_devices->list, &fs_uuids);
1287
1288         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1289         seed_devices->opened = 1;
1290         INIT_LIST_HEAD(&seed_devices->devices);
1291         INIT_LIST_HEAD(&seed_devices->alloc_list);
1292         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1293         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1294         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1295                 device->fs_devices = seed_devices;
1296         }
1297
1298         fs_devices->seeding = 0;
1299         fs_devices->num_devices = 0;
1300         fs_devices->open_devices = 0;
1301         fs_devices->seed = seed_devices;
1302
1303         generate_random_uuid(fs_devices->fsid);
1304         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1305         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1306         super_flags = btrfs_super_flags(disk_super) &
1307                       ~BTRFS_SUPER_FLAG_SEEDING;
1308         btrfs_set_super_flags(disk_super, super_flags);
1309
1310         return 0;
1311 }
1312
1313 /*
1314  * strore the expected generation for seed devices in device items.
1315  */
1316 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1317                                struct btrfs_root *root)
1318 {
1319         struct btrfs_path *path;
1320         struct extent_buffer *leaf;
1321         struct btrfs_dev_item *dev_item;
1322         struct btrfs_device *device;
1323         struct btrfs_key key;
1324         u8 fs_uuid[BTRFS_UUID_SIZE];
1325         u8 dev_uuid[BTRFS_UUID_SIZE];
1326         u64 devid;
1327         int ret;
1328
1329         path = btrfs_alloc_path();
1330         if (!path)
1331                 return -ENOMEM;
1332
1333         root = root->fs_info->chunk_root;
1334         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1335         key.offset = 0;
1336         key.type = BTRFS_DEV_ITEM_KEY;
1337
1338         while (1) {
1339                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1340                 if (ret < 0)
1341                         goto error;
1342
1343                 leaf = path->nodes[0];
1344 next_slot:
1345                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1346                         ret = btrfs_next_leaf(root, path);
1347                         if (ret > 0)
1348                                 break;
1349                         if (ret < 0)
1350                                 goto error;
1351                         leaf = path->nodes[0];
1352                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1353                         btrfs_release_path(root, path);
1354                         continue;
1355                 }
1356
1357                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1358                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1359                     key.type != BTRFS_DEV_ITEM_KEY)
1360                         break;
1361
1362                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1363                                           struct btrfs_dev_item);
1364                 devid = btrfs_device_id(leaf, dev_item);
1365                 read_extent_buffer(leaf, dev_uuid,
1366                                    (unsigned long)btrfs_device_uuid(dev_item),
1367                                    BTRFS_UUID_SIZE);
1368                 read_extent_buffer(leaf, fs_uuid,
1369                                    (unsigned long)btrfs_device_fsid(dev_item),
1370                                    BTRFS_UUID_SIZE);
1371                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1372                 BUG_ON(!device);
1373
1374                 if (device->fs_devices->seeding) {
1375                         btrfs_set_device_generation(leaf, dev_item,
1376                                                     device->generation);
1377                         btrfs_mark_buffer_dirty(leaf);
1378                 }
1379
1380                 path->slots[0]++;
1381                 goto next_slot;
1382         }
1383         ret = 0;
1384 error:
1385         btrfs_free_path(path);
1386         return ret;
1387 }
1388
1389 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1390 {
1391         struct btrfs_trans_handle *trans;
1392         struct btrfs_device *device;
1393         struct block_device *bdev;
1394         struct list_head *devices;
1395         struct super_block *sb = root->fs_info->sb;
1396         u64 total_bytes;
1397         int seeding_dev = 0;
1398         int ret = 0;
1399
1400         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1401                 return -EINVAL;
1402
1403         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1404         if (!bdev)
1405                 return -EIO;
1406
1407         if (root->fs_info->fs_devices->seeding) {
1408                 seeding_dev = 1;
1409                 down_write(&sb->s_umount);
1410                 mutex_lock(&uuid_mutex);
1411         }
1412
1413         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1414         mutex_lock(&root->fs_info->volume_mutex);
1415
1416         devices = &root->fs_info->fs_devices->devices;
1417         list_for_each_entry(device, devices, dev_list) {
1418                 if (device->bdev == bdev) {
1419                         ret = -EEXIST;
1420                         goto error;
1421                 }
1422         }
1423
1424         device = kzalloc(sizeof(*device), GFP_NOFS);
1425         if (!device) {
1426                 /* we can safely leave the fs_devices entry around */
1427                 ret = -ENOMEM;
1428                 goto error;
1429         }
1430
1431         device->name = kstrdup(device_path, GFP_NOFS);
1432         if (!device->name) {
1433                 kfree(device);
1434                 ret = -ENOMEM;
1435                 goto error;
1436         }
1437
1438         ret = find_next_devid(root, &device->devid);
1439         if (ret) {
1440                 kfree(device);
1441                 goto error;
1442         }
1443
1444         trans = btrfs_start_transaction(root, 1);
1445         lock_chunks(root);
1446
1447         device->barriers = 1;
1448         device->writeable = 1;
1449         device->work.func = pending_bios_fn;
1450         generate_random_uuid(device->uuid);
1451         spin_lock_init(&device->io_lock);
1452         device->generation = trans->transid;
1453         device->io_width = root->sectorsize;
1454         device->io_align = root->sectorsize;
1455         device->sector_size = root->sectorsize;
1456         device->total_bytes = i_size_read(bdev->bd_inode);
1457         device->disk_total_bytes = device->total_bytes;
1458         device->dev_root = root->fs_info->dev_root;
1459         device->bdev = bdev;
1460         device->in_fs_metadata = 1;
1461         device->mode = 0;
1462         set_blocksize(device->bdev, 4096);
1463
1464         if (seeding_dev) {
1465                 sb->s_flags &= ~MS_RDONLY;
1466                 ret = btrfs_prepare_sprout(trans, root);
1467                 BUG_ON(ret);
1468         }
1469
1470         device->fs_devices = root->fs_info->fs_devices;
1471         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1472         list_add(&device->dev_alloc_list,
1473                  &root->fs_info->fs_devices->alloc_list);
1474         root->fs_info->fs_devices->num_devices++;
1475         root->fs_info->fs_devices->open_devices++;
1476         root->fs_info->fs_devices->rw_devices++;
1477         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1478
1479         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1480                 root->fs_info->fs_devices->rotating = 1;
1481
1482         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1483         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1484                                     total_bytes + device->total_bytes);
1485
1486         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1487         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1488                                     total_bytes + 1);
1489
1490         if (seeding_dev) {
1491                 ret = init_first_rw_device(trans, root, device);
1492                 BUG_ON(ret);
1493                 ret = btrfs_finish_sprout(trans, root);
1494                 BUG_ON(ret);
1495         } else {
1496                 ret = btrfs_add_device(trans, root, device);
1497         }
1498
1499         /*
1500          * we've got more storage, clear any full flags on the space
1501          * infos
1502          */
1503         btrfs_clear_space_info_full(root->fs_info);
1504
1505         unlock_chunks(root);
1506         btrfs_commit_transaction(trans, root);
1507
1508         if (seeding_dev) {
1509                 mutex_unlock(&uuid_mutex);
1510                 up_write(&sb->s_umount);
1511
1512                 ret = btrfs_relocate_sys_chunks(root);
1513                 BUG_ON(ret);
1514         }
1515 out:
1516         mutex_unlock(&root->fs_info->volume_mutex);
1517         return ret;
1518 error:
1519         close_bdev_exclusive(bdev, 0);
1520         if (seeding_dev) {
1521                 mutex_unlock(&uuid_mutex);
1522                 up_write(&sb->s_umount);
1523         }
1524         goto out;
1525 }
1526
1527 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1528                                         struct btrfs_device *device)
1529 {
1530         int ret;
1531         struct btrfs_path *path;
1532         struct btrfs_root *root;
1533         struct btrfs_dev_item *dev_item;
1534         struct extent_buffer *leaf;
1535         struct btrfs_key key;
1536
1537         root = device->dev_root->fs_info->chunk_root;
1538
1539         path = btrfs_alloc_path();
1540         if (!path)
1541                 return -ENOMEM;
1542
1543         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1544         key.type = BTRFS_DEV_ITEM_KEY;
1545         key.offset = device->devid;
1546
1547         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1548         if (ret < 0)
1549                 goto out;
1550
1551         if (ret > 0) {
1552                 ret = -ENOENT;
1553                 goto out;
1554         }
1555
1556         leaf = path->nodes[0];
1557         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1558
1559         btrfs_set_device_id(leaf, dev_item, device->devid);
1560         btrfs_set_device_type(leaf, dev_item, device->type);
1561         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1562         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1563         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1564         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1565         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1566         btrfs_mark_buffer_dirty(leaf);
1567
1568 out:
1569         btrfs_free_path(path);
1570         return ret;
1571 }
1572
1573 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1574                       struct btrfs_device *device, u64 new_size)
1575 {
1576         struct btrfs_super_block *super_copy =
1577                 &device->dev_root->fs_info->super_copy;
1578         u64 old_total = btrfs_super_total_bytes(super_copy);
1579         u64 diff = new_size - device->total_bytes;
1580
1581         if (!device->writeable)
1582                 return -EACCES;
1583         if (new_size <= device->total_bytes)
1584                 return -EINVAL;
1585
1586         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1587         device->fs_devices->total_rw_bytes += diff;
1588
1589         device->total_bytes = new_size;
1590         btrfs_clear_space_info_full(device->dev_root->fs_info);
1591
1592         return btrfs_update_device(trans, device);
1593 }
1594
1595 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1596                       struct btrfs_device *device, u64 new_size)
1597 {
1598         int ret;
1599         lock_chunks(device->dev_root);
1600         ret = __btrfs_grow_device(trans, device, new_size);
1601         unlock_chunks(device->dev_root);
1602         return ret;
1603 }
1604
1605 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1606                             struct btrfs_root *root,
1607                             u64 chunk_tree, u64 chunk_objectid,
1608                             u64 chunk_offset)
1609 {
1610         int ret;
1611         struct btrfs_path *path;
1612         struct btrfs_key key;
1613
1614         root = root->fs_info->chunk_root;
1615         path = btrfs_alloc_path();
1616         if (!path)
1617                 return -ENOMEM;
1618
1619         key.objectid = chunk_objectid;
1620         key.offset = chunk_offset;
1621         key.type = BTRFS_CHUNK_ITEM_KEY;
1622
1623         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1624         BUG_ON(ret);
1625
1626         ret = btrfs_del_item(trans, root, path);
1627         BUG_ON(ret);
1628
1629         btrfs_free_path(path);
1630         return 0;
1631 }
1632
1633 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1634                         chunk_offset)
1635 {
1636         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1637         struct btrfs_disk_key *disk_key;
1638         struct btrfs_chunk *chunk;
1639         u8 *ptr;
1640         int ret = 0;
1641         u32 num_stripes;
1642         u32 array_size;
1643         u32 len = 0;
1644         u32 cur;
1645         struct btrfs_key key;
1646
1647         array_size = btrfs_super_sys_array_size(super_copy);
1648
1649         ptr = super_copy->sys_chunk_array;
1650         cur = 0;
1651
1652         while (cur < array_size) {
1653                 disk_key = (struct btrfs_disk_key *)ptr;
1654                 btrfs_disk_key_to_cpu(&key, disk_key);
1655
1656                 len = sizeof(*disk_key);
1657
1658                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1659                         chunk = (struct btrfs_chunk *)(ptr + len);
1660                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1661                         len += btrfs_chunk_item_size(num_stripes);
1662                 } else {
1663                         ret = -EIO;
1664                         break;
1665                 }
1666                 if (key.objectid == chunk_objectid &&
1667                     key.offset == chunk_offset) {
1668                         memmove(ptr, ptr + len, array_size - (cur + len));
1669                         array_size -= len;
1670                         btrfs_set_super_sys_array_size(super_copy, array_size);
1671                 } else {
1672                         ptr += len;
1673                         cur += len;
1674                 }
1675         }
1676         return ret;
1677 }
1678
1679 static int btrfs_relocate_chunk(struct btrfs_root *root,
1680                          u64 chunk_tree, u64 chunk_objectid,
1681                          u64 chunk_offset)
1682 {
1683         struct extent_map_tree *em_tree;
1684         struct btrfs_root *extent_root;
1685         struct btrfs_trans_handle *trans;
1686         struct extent_map *em;
1687         struct map_lookup *map;
1688         int ret;
1689         int i;
1690
1691         root = root->fs_info->chunk_root;
1692         extent_root = root->fs_info->extent_root;
1693         em_tree = &root->fs_info->mapping_tree.map_tree;
1694
1695         /* step one, relocate all the extents inside this chunk */
1696         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1697         BUG_ON(ret);
1698
1699         trans = btrfs_start_transaction(root, 1);
1700         BUG_ON(!trans);
1701
1702         lock_chunks(root);
1703
1704         /*
1705          * step two, delete the device extents and the
1706          * chunk tree entries
1707          */
1708         spin_lock(&em_tree->lock);
1709         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1710         spin_unlock(&em_tree->lock);
1711
1712         BUG_ON(em->start > chunk_offset ||
1713                em->start + em->len < chunk_offset);
1714         map = (struct map_lookup *)em->bdev;
1715
1716         for (i = 0; i < map->num_stripes; i++) {
1717                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1718                                             map->stripes[i].physical);
1719                 BUG_ON(ret);
1720
1721                 if (map->stripes[i].dev) {
1722                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1723                         BUG_ON(ret);
1724                 }
1725         }
1726         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1727                                chunk_offset);
1728
1729         BUG_ON(ret);
1730
1731         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1732                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1733                 BUG_ON(ret);
1734         }
1735
1736         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1737         BUG_ON(ret);
1738
1739         spin_lock(&em_tree->lock);
1740         remove_extent_mapping(em_tree, em);
1741         spin_unlock(&em_tree->lock);
1742
1743         kfree(map);
1744         em->bdev = NULL;
1745
1746         /* once for the tree */
1747         free_extent_map(em);
1748         /* once for us */
1749         free_extent_map(em);
1750
1751         unlock_chunks(root);
1752         btrfs_end_transaction(trans, root);
1753         return 0;
1754 }
1755
1756 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1757 {
1758         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1759         struct btrfs_path *path;
1760         struct extent_buffer *leaf;
1761         struct btrfs_chunk *chunk;
1762         struct btrfs_key key;
1763         struct btrfs_key found_key;
1764         u64 chunk_tree = chunk_root->root_key.objectid;
1765         u64 chunk_type;
1766         int ret;
1767
1768         path = btrfs_alloc_path();
1769         if (!path)
1770                 return -ENOMEM;
1771
1772         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1773         key.offset = (u64)-1;
1774         key.type = BTRFS_CHUNK_ITEM_KEY;
1775
1776         while (1) {
1777                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1778                 if (ret < 0)
1779                         goto error;
1780                 BUG_ON(ret == 0);
1781
1782                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1783                                           key.type);
1784                 if (ret < 0)
1785                         goto error;
1786                 if (ret > 0)
1787                         break;
1788
1789                 leaf = path->nodes[0];
1790                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1791
1792                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1793                                        struct btrfs_chunk);
1794                 chunk_type = btrfs_chunk_type(leaf, chunk);
1795                 btrfs_release_path(chunk_root, path);
1796
1797                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1798                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1799                                                    found_key.objectid,
1800                                                    found_key.offset);
1801                         BUG_ON(ret);
1802                 }
1803
1804                 if (found_key.offset == 0)
1805                         break;
1806                 key.offset = found_key.offset - 1;
1807         }
1808         ret = 0;
1809 error:
1810         btrfs_free_path(path);
1811         return ret;
1812 }
1813
1814 static u64 div_factor(u64 num, int factor)
1815 {
1816         if (factor == 10)
1817                 return num;
1818         num *= factor;
1819         do_div(num, 10);
1820         return num;
1821 }
1822
1823 int btrfs_balance(struct btrfs_root *dev_root)
1824 {
1825         int ret;
1826         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1827         struct btrfs_device *device;
1828         u64 old_size;
1829         u64 size_to_free;
1830         struct btrfs_path *path;
1831         struct btrfs_key key;
1832         struct btrfs_chunk *chunk;
1833         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1834         struct btrfs_trans_handle *trans;
1835         struct btrfs_key found_key;
1836
1837         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1838                 return -EROFS;
1839
1840         mutex_lock(&dev_root->fs_info->volume_mutex);
1841         dev_root = dev_root->fs_info->dev_root;
1842
1843         /* step one make some room on all the devices */
1844         list_for_each_entry(device, devices, dev_list) {
1845                 old_size = device->total_bytes;
1846                 size_to_free = div_factor(old_size, 1);
1847                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1848                 if (!device->writeable ||
1849                     device->total_bytes - device->bytes_used > size_to_free)
1850                         continue;
1851
1852                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1853                 BUG_ON(ret);
1854
1855                 trans = btrfs_start_transaction(dev_root, 1);
1856                 BUG_ON(!trans);
1857
1858                 ret = btrfs_grow_device(trans, device, old_size);
1859                 BUG_ON(ret);
1860
1861                 btrfs_end_transaction(trans, dev_root);
1862         }
1863
1864         /* step two, relocate all the chunks */
1865         path = btrfs_alloc_path();
1866         BUG_ON(!path);
1867
1868         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1869         key.offset = (u64)-1;
1870         key.type = BTRFS_CHUNK_ITEM_KEY;
1871
1872         while (1) {
1873                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1874                 if (ret < 0)
1875                         goto error;
1876
1877                 /*
1878                  * this shouldn't happen, it means the last relocate
1879                  * failed
1880                  */
1881                 if (ret == 0)
1882                         break;
1883
1884                 ret = btrfs_previous_item(chunk_root, path, 0,
1885                                           BTRFS_CHUNK_ITEM_KEY);
1886                 if (ret)
1887                         break;
1888
1889                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1890                                       path->slots[0]);
1891                 if (found_key.objectid != key.objectid)
1892                         break;
1893
1894                 chunk = btrfs_item_ptr(path->nodes[0],
1895                                        path->slots[0],
1896                                        struct btrfs_chunk);
1897                 key.offset = found_key.offset;
1898                 /* chunk zero is special */
1899                 if (key.offset == 0)
1900                         break;
1901
1902                 btrfs_release_path(chunk_root, path);
1903                 ret = btrfs_relocate_chunk(chunk_root,
1904                                            chunk_root->root_key.objectid,
1905                                            found_key.objectid,
1906                                            found_key.offset);
1907                 BUG_ON(ret);
1908         }
1909         ret = 0;
1910 error:
1911         btrfs_free_path(path);
1912         mutex_unlock(&dev_root->fs_info->volume_mutex);
1913         return ret;
1914 }
1915
1916 /*
1917  * shrinking a device means finding all of the device extents past
1918  * the new size, and then following the back refs to the chunks.
1919  * The chunk relocation code actually frees the device extent
1920  */
1921 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1922 {
1923         struct btrfs_trans_handle *trans;
1924         struct btrfs_root *root = device->dev_root;
1925         struct btrfs_dev_extent *dev_extent = NULL;
1926         struct btrfs_path *path;
1927         u64 length;
1928         u64 chunk_tree;
1929         u64 chunk_objectid;
1930         u64 chunk_offset;
1931         int ret;
1932         int slot;
1933         struct extent_buffer *l;
1934         struct btrfs_key key;
1935         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1936         u64 old_total = btrfs_super_total_bytes(super_copy);
1937         u64 diff = device->total_bytes - new_size;
1938
1939         if (new_size >= device->total_bytes)
1940                 return -EINVAL;
1941
1942         path = btrfs_alloc_path();
1943         if (!path)
1944                 return -ENOMEM;
1945
1946         trans = btrfs_start_transaction(root, 1);
1947         if (!trans) {
1948                 ret = -ENOMEM;
1949                 goto done;
1950         }
1951
1952         path->reada = 2;
1953
1954         lock_chunks(root);
1955
1956         device->total_bytes = new_size;
1957         if (device->writeable)
1958                 device->fs_devices->total_rw_bytes -= diff;
1959         unlock_chunks(root);
1960         btrfs_end_transaction(trans, root);
1961
1962         key.objectid = device->devid;
1963         key.offset = (u64)-1;
1964         key.type = BTRFS_DEV_EXTENT_KEY;
1965
1966         while (1) {
1967                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1968                 if (ret < 0)
1969                         goto done;
1970
1971                 ret = btrfs_previous_item(root, path, 0, key.type);
1972                 if (ret < 0)
1973                         goto done;
1974                 if (ret) {
1975                         ret = 0;
1976                         goto done;
1977                 }
1978
1979                 l = path->nodes[0];
1980                 slot = path->slots[0];
1981                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1982
1983                 if (key.objectid != device->devid)
1984                         goto done;
1985
1986                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1987                 length = btrfs_dev_extent_length(l, dev_extent);
1988
1989                 if (key.offset + length <= new_size)
1990                         break;
1991
1992                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1993                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1994                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1995                 btrfs_release_path(root, path);
1996
1997                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1998                                            chunk_offset);
1999                 if (ret)
2000                         goto done;
2001         }
2002
2003         /* Shrinking succeeded, else we would be at "done". */
2004         trans = btrfs_start_transaction(root, 1);
2005         if (!trans) {
2006                 ret = -ENOMEM;
2007                 goto done;
2008         }
2009         lock_chunks(root);
2010
2011         device->disk_total_bytes = new_size;
2012         /* Now btrfs_update_device() will change the on-disk size. */
2013         ret = btrfs_update_device(trans, device);
2014         if (ret) {
2015                 unlock_chunks(root);
2016                 btrfs_end_transaction(trans, root);
2017                 goto done;
2018         }
2019         WARN_ON(diff > old_total);
2020         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2021         unlock_chunks(root);
2022         btrfs_end_transaction(trans, root);
2023 done:
2024         btrfs_free_path(path);
2025         return ret;
2026 }
2027
2028 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2029                            struct btrfs_root *root,
2030                            struct btrfs_key *key,
2031                            struct btrfs_chunk *chunk, int item_size)
2032 {
2033         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2034         struct btrfs_disk_key disk_key;
2035         u32 array_size;
2036         u8 *ptr;
2037
2038         array_size = btrfs_super_sys_array_size(super_copy);
2039         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2040                 return -EFBIG;
2041
2042         ptr = super_copy->sys_chunk_array + array_size;
2043         btrfs_cpu_key_to_disk(&disk_key, key);
2044         memcpy(ptr, &disk_key, sizeof(disk_key));
2045         ptr += sizeof(disk_key);
2046         memcpy(ptr, chunk, item_size);
2047         item_size += sizeof(disk_key);
2048         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2049         return 0;
2050 }
2051
2052 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2053                                         int num_stripes, int sub_stripes)
2054 {
2055         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2056                 return calc_size;
2057         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2058                 return calc_size * (num_stripes / sub_stripes);
2059         else
2060                 return calc_size * num_stripes;
2061 }
2062
2063 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2064                                struct btrfs_root *extent_root,
2065                                struct map_lookup **map_ret,
2066                                u64 *num_bytes, u64 *stripe_size,
2067                                u64 start, u64 type)
2068 {
2069         struct btrfs_fs_info *info = extent_root->fs_info;
2070         struct btrfs_device *device = NULL;
2071         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2072         struct list_head *cur;
2073         struct map_lookup *map = NULL;
2074         struct extent_map_tree *em_tree;
2075         struct extent_map *em;
2076         struct list_head private_devs;
2077         int min_stripe_size = 1 * 1024 * 1024;
2078         u64 calc_size = 1024 * 1024 * 1024;
2079         u64 max_chunk_size = calc_size;
2080         u64 min_free;
2081         u64 avail;
2082         u64 max_avail = 0;
2083         u64 dev_offset;
2084         int num_stripes = 1;
2085         int min_stripes = 1;
2086         int sub_stripes = 0;
2087         int looped = 0;
2088         int ret;
2089         int index;
2090         int stripe_len = 64 * 1024;
2091
2092         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2093             (type & BTRFS_BLOCK_GROUP_DUP)) {
2094                 WARN_ON(1);
2095                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2096         }
2097         if (list_empty(&fs_devices->alloc_list))
2098                 return -ENOSPC;
2099
2100         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2101                 num_stripes = fs_devices->rw_devices;
2102                 min_stripes = 2;
2103         }
2104         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2105                 num_stripes = 2;
2106                 min_stripes = 2;
2107         }
2108         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2109                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2110                 if (num_stripes < 2)
2111                         return -ENOSPC;
2112                 min_stripes = 2;
2113         }
2114         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2115                 num_stripes = fs_devices->rw_devices;
2116                 if (num_stripes < 4)
2117                         return -ENOSPC;
2118                 num_stripes &= ~(u32)1;
2119                 sub_stripes = 2;
2120                 min_stripes = 4;
2121         }
2122
2123         if (type & BTRFS_BLOCK_GROUP_DATA) {
2124                 max_chunk_size = 10 * calc_size;
2125                 min_stripe_size = 64 * 1024 * 1024;
2126         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2127                 max_chunk_size = 4 * calc_size;
2128                 min_stripe_size = 32 * 1024 * 1024;
2129         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2130                 calc_size = 8 * 1024 * 1024;
2131                 max_chunk_size = calc_size * 2;
2132                 min_stripe_size = 1 * 1024 * 1024;
2133         }
2134
2135         /* we don't want a chunk larger than 10% of writeable space */
2136         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2137                              max_chunk_size);
2138
2139 again:
2140         if (!map || map->num_stripes != num_stripes) {
2141                 kfree(map);
2142                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2143                 if (!map)
2144                         return -ENOMEM;
2145                 map->num_stripes = num_stripes;
2146         }
2147
2148         if (calc_size * num_stripes > max_chunk_size) {
2149                 calc_size = max_chunk_size;
2150                 do_div(calc_size, num_stripes);
2151                 do_div(calc_size, stripe_len);
2152                 calc_size *= stripe_len;
2153         }
2154         /* we don't want tiny stripes */
2155         calc_size = max_t(u64, min_stripe_size, calc_size);
2156
2157         do_div(calc_size, stripe_len);
2158         calc_size *= stripe_len;
2159
2160         cur = fs_devices->alloc_list.next;
2161         index = 0;
2162
2163         if (type & BTRFS_BLOCK_GROUP_DUP)
2164                 min_free = calc_size * 2;
2165         else
2166                 min_free = calc_size;
2167
2168         /*
2169          * we add 1MB because we never use the first 1MB of the device, unless
2170          * we've looped, then we are likely allocating the maximum amount of
2171          * space left already
2172          */
2173         if (!looped)
2174                 min_free += 1024 * 1024;
2175
2176         INIT_LIST_HEAD(&private_devs);
2177         while (index < num_stripes) {
2178                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2179                 BUG_ON(!device->writeable);
2180                 if (device->total_bytes > device->bytes_used)
2181                         avail = device->total_bytes - device->bytes_used;
2182                 else
2183                         avail = 0;
2184                 cur = cur->next;
2185
2186                 if (device->in_fs_metadata && avail >= min_free) {
2187                         ret = find_free_dev_extent(trans, device,
2188                                                    min_free, &dev_offset);
2189                         if (ret == 0) {
2190                                 list_move_tail(&device->dev_alloc_list,
2191                                                &private_devs);
2192                                 map->stripes[index].dev = device;
2193                                 map->stripes[index].physical = dev_offset;
2194                                 index++;
2195                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2196                                         map->stripes[index].dev = device;
2197                                         map->stripes[index].physical =
2198                                                 dev_offset + calc_size;
2199                                         index++;
2200                                 }
2201                         }
2202                 } else if (device->in_fs_metadata && avail > max_avail)
2203                         max_avail = avail;
2204                 if (cur == &fs_devices->alloc_list)
2205                         break;
2206         }
2207         list_splice(&private_devs, &fs_devices->alloc_list);
2208         if (index < num_stripes) {
2209                 if (index >= min_stripes) {
2210                         num_stripes = index;
2211                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2212                                 num_stripes /= sub_stripes;
2213                                 num_stripes *= sub_stripes;
2214                         }
2215                         looped = 1;
2216                         goto again;
2217                 }
2218                 if (!looped && max_avail > 0) {
2219                         looped = 1;
2220                         calc_size = max_avail;
2221                         goto again;
2222                 }
2223                 kfree(map);
2224                 return -ENOSPC;
2225         }
2226         map->sector_size = extent_root->sectorsize;
2227         map->stripe_len = stripe_len;
2228         map->io_align = stripe_len;
2229         map->io_width = stripe_len;
2230         map->type = type;
2231         map->num_stripes = num_stripes;
2232         map->sub_stripes = sub_stripes;
2233
2234         *map_ret = map;
2235         *stripe_size = calc_size;
2236         *num_bytes = chunk_bytes_by_type(type, calc_size,
2237                                          num_stripes, sub_stripes);
2238
2239         em = alloc_extent_map(GFP_NOFS);
2240         if (!em) {
2241                 kfree(map);
2242                 return -ENOMEM;
2243         }
2244         em->bdev = (struct block_device *)map;
2245         em->start = start;
2246         em->len = *num_bytes;
2247         em->block_start = 0;
2248         em->block_len = em->len;
2249
2250         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2251         spin_lock(&em_tree->lock);
2252         ret = add_extent_mapping(em_tree, em);
2253         spin_unlock(&em_tree->lock);
2254         BUG_ON(ret);
2255         free_extent_map(em);
2256
2257         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2258                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2259                                      start, *num_bytes);
2260         BUG_ON(ret);
2261
2262         index = 0;
2263         while (index < map->num_stripes) {
2264                 device = map->stripes[index].dev;
2265                 dev_offset = map->stripes[index].physical;
2266
2267                 ret = btrfs_alloc_dev_extent(trans, device,
2268                                 info->chunk_root->root_key.objectid,
2269                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2270                                 start, dev_offset, calc_size);
2271                 BUG_ON(ret);
2272                 index++;
2273         }
2274
2275         return 0;
2276 }
2277
2278 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2279                                 struct btrfs_root *extent_root,
2280                                 struct map_lookup *map, u64 chunk_offset,
2281                                 u64 chunk_size, u64 stripe_size)
2282 {
2283         u64 dev_offset;
2284         struct btrfs_key key;
2285         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2286         struct btrfs_device *device;
2287         struct btrfs_chunk *chunk;
2288         struct btrfs_stripe *stripe;
2289         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2290         int index = 0;
2291         int ret;
2292
2293         chunk = kzalloc(item_size, GFP_NOFS);
2294         if (!chunk)
2295                 return -ENOMEM;
2296
2297         index = 0;
2298         while (index < map->num_stripes) {
2299                 device = map->stripes[index].dev;
2300                 device->bytes_used += stripe_size;
2301                 ret = btrfs_update_device(trans, device);
2302                 BUG_ON(ret);
2303                 index++;
2304         }
2305
2306         index = 0;
2307         stripe = &chunk->stripe;
2308         while (index < map->num_stripes) {
2309                 device = map->stripes[index].dev;
2310                 dev_offset = map->stripes[index].physical;
2311
2312                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2313                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2314                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2315                 stripe++;
2316                 index++;
2317         }
2318
2319         btrfs_set_stack_chunk_length(chunk, chunk_size);
2320         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2321         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2322         btrfs_set_stack_chunk_type(chunk, map->type);
2323         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2324         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2325         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2326         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2327         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2328
2329         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2330         key.type = BTRFS_CHUNK_ITEM_KEY;
2331         key.offset = chunk_offset;
2332
2333         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2334         BUG_ON(ret);
2335
2336         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2337                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2338                                              item_size);
2339                 BUG_ON(ret);
2340         }
2341         kfree(chunk);
2342         return 0;
2343 }
2344
2345 /*
2346  * Chunk allocation falls into two parts. The first part does works
2347  * that make the new allocated chunk useable, but not do any operation
2348  * that modifies the chunk tree. The second part does the works that
2349  * require modifying the chunk tree. This division is important for the
2350  * bootstrap process of adding storage to a seed btrfs.
2351  */
2352 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2353                       struct btrfs_root *extent_root, u64 type)
2354 {
2355         u64 chunk_offset;
2356         u64 chunk_size;
2357         u64 stripe_size;
2358         struct map_lookup *map;
2359         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2360         int ret;
2361
2362         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2363                               &chunk_offset);
2364         if (ret)
2365                 return ret;
2366
2367         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2368                                   &stripe_size, chunk_offset, type);
2369         if (ret)
2370                 return ret;
2371
2372         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2373                                    chunk_size, stripe_size);
2374         BUG_ON(ret);
2375         return 0;
2376 }
2377
2378 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2379                                          struct btrfs_root *root,
2380                                          struct btrfs_device *device)
2381 {
2382         u64 chunk_offset;
2383         u64 sys_chunk_offset;
2384         u64 chunk_size;
2385         u64 sys_chunk_size;
2386         u64 stripe_size;
2387         u64 sys_stripe_size;
2388         u64 alloc_profile;
2389         struct map_lookup *map;
2390         struct map_lookup *sys_map;
2391         struct btrfs_fs_info *fs_info = root->fs_info;
2392         struct btrfs_root *extent_root = fs_info->extent_root;
2393         int ret;
2394
2395         ret = find_next_chunk(fs_info->chunk_root,
2396                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2397         BUG_ON(ret);
2398
2399         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2400                         (fs_info->metadata_alloc_profile &
2401                          fs_info->avail_metadata_alloc_bits);
2402         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2403
2404         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2405                                   &stripe_size, chunk_offset, alloc_profile);
2406         BUG_ON(ret);
2407
2408         sys_chunk_offset = chunk_offset + chunk_size;
2409
2410         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2411                         (fs_info->system_alloc_profile &
2412                          fs_info->avail_system_alloc_bits);
2413         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2414
2415         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2416                                   &sys_chunk_size, &sys_stripe_size,
2417                                   sys_chunk_offset, alloc_profile);
2418         BUG_ON(ret);
2419
2420         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2421         BUG_ON(ret);
2422
2423         /*
2424          * Modifying chunk tree needs allocating new blocks from both
2425          * system block group and metadata block group. So we only can
2426          * do operations require modifying the chunk tree after both
2427          * block groups were created.
2428          */
2429         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2430                                    chunk_size, stripe_size);
2431         BUG_ON(ret);
2432
2433         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2434                                    sys_chunk_offset, sys_chunk_size,
2435                                    sys_stripe_size);
2436         BUG_ON(ret);
2437         return 0;
2438 }
2439
2440 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2441 {
2442         struct extent_map *em;
2443         struct map_lookup *map;
2444         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2445         int readonly = 0;
2446         int i;
2447
2448         spin_lock(&map_tree->map_tree.lock);
2449         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2450         spin_unlock(&map_tree->map_tree.lock);
2451         if (!em)
2452                 return 1;
2453
2454         map = (struct map_lookup *)em->bdev;
2455         for (i = 0; i < map->num_stripes; i++) {
2456                 if (!map->stripes[i].dev->writeable) {
2457                         readonly = 1;
2458                         break;
2459                 }
2460         }
2461         free_extent_map(em);
2462         return readonly;
2463 }
2464
2465 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2466 {
2467         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2468 }
2469
2470 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2471 {
2472         struct extent_map *em;
2473
2474         while (1) {
2475                 spin_lock(&tree->map_tree.lock);
2476                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2477                 if (em)
2478                         remove_extent_mapping(&tree->map_tree, em);
2479                 spin_unlock(&tree->map_tree.lock);
2480                 if (!em)
2481                         break;
2482                 kfree(em->bdev);
2483                 /* once for us */
2484                 free_extent_map(em);
2485                 /* once for the tree */
2486                 free_extent_map(em);
2487         }
2488 }
2489
2490 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2491 {
2492         struct extent_map *em;
2493         struct map_lookup *map;
2494         struct extent_map_tree *em_tree = &map_tree->map_tree;
2495         int ret;
2496
2497         spin_lock(&em_tree->lock);
2498         em = lookup_extent_mapping(em_tree, logical, len);
2499         spin_unlock(&em_tree->lock);
2500         BUG_ON(!em);
2501
2502         BUG_ON(em->start > logical || em->start + em->len < logical);
2503         map = (struct map_lookup *)em->bdev;
2504         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2505                 ret = map->num_stripes;
2506         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2507                 ret = map->sub_stripes;
2508         else
2509                 ret = 1;
2510         free_extent_map(em);
2511         return ret;
2512 }
2513
2514 static int find_live_mirror(struct map_lookup *map, int first, int num,
2515                             int optimal)
2516 {
2517         int i;
2518         if (map->stripes[optimal].dev->bdev)
2519                 return optimal;
2520         for (i = first; i < first + num; i++) {
2521                 if (map->stripes[i].dev->bdev)
2522                         return i;
2523         }
2524         /* we couldn't find one that doesn't fail.  Just return something
2525          * and the io error handling code will clean up eventually
2526          */
2527         return optimal;
2528 }
2529
2530 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2531                              u64 logical, u64 *length,
2532                              struct btrfs_multi_bio **multi_ret,
2533                              int mirror_num, struct page *unplug_page)
2534 {
2535         struct extent_map *em;
2536         struct map_lookup *map;
2537         struct extent_map_tree *em_tree = &map_tree->map_tree;
2538         u64 offset;
2539         u64 stripe_offset;
2540         u64 stripe_nr;
2541         int stripes_allocated = 8;
2542         int stripes_required = 1;
2543         int stripe_index;
2544         int i;
2545         int num_stripes;
2546         int max_errors = 0;
2547         struct btrfs_multi_bio *multi = NULL;
2548
2549         if (multi_ret && !(rw & (1 << BIO_RW)))
2550                 stripes_allocated = 1;
2551 again:
2552         if (multi_ret) {
2553                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2554                                 GFP_NOFS);
2555                 if (!multi)
2556                         return -ENOMEM;
2557
2558                 atomic_set(&multi->error, 0);
2559         }
2560
2561         spin_lock(&em_tree->lock);
2562         em = lookup_extent_mapping(em_tree, logical, *length);
2563         spin_unlock(&em_tree->lock);
2564
2565         if (!em && unplug_page)
2566                 return 0;
2567
2568         if (!em) {
2569                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2570                        (unsigned long long)logical,
2571                        (unsigned long long)*length);
2572                 BUG();
2573         }
2574
2575         BUG_ON(em->start > logical || em->start + em->len < logical);
2576         map = (struct map_lookup *)em->bdev;
2577         offset = logical - em->start;
2578
2579         if (mirror_num > map->num_stripes)
2580                 mirror_num = 0;
2581
2582         /* if our multi bio struct is too small, back off and try again */
2583         if (rw & (1 << BIO_RW)) {
2584                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2585                                  BTRFS_BLOCK_GROUP_DUP)) {
2586                         stripes_required = map->num_stripes;
2587                         max_errors = 1;
2588                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2589                         stripes_required = map->sub_stripes;
2590                         max_errors = 1;
2591                 }
2592         }
2593         if (multi_ret && (rw & (1 << BIO_RW)) &&
2594             stripes_allocated < stripes_required) {
2595                 stripes_allocated = map->num_stripes;
2596                 free_extent_map(em);
2597                 kfree(multi);
2598                 goto again;
2599         }
2600         stripe_nr = offset;
2601         /*
2602          * stripe_nr counts the total number of stripes we have to stride
2603          * to get to this block
2604          */
2605         do_div(stripe_nr, map->stripe_len);
2606
2607         stripe_offset = stripe_nr * map->stripe_len;
2608         BUG_ON(offset < stripe_offset);
2609
2610         /* stripe_offset is the offset of this block in its stripe*/
2611         stripe_offset = offset - stripe_offset;
2612
2613         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2614                          BTRFS_BLOCK_GROUP_RAID10 |
2615                          BTRFS_BLOCK_GROUP_DUP)) {
2616                 /* we limit the length of each bio to what fits in a stripe */
2617                 *length = min_t(u64, em->len - offset,
2618                               map->stripe_len - stripe_offset);
2619         } else {
2620                 *length = em->len - offset;
2621         }
2622
2623         if (!multi_ret && !unplug_page)
2624                 goto out;
2625
2626         num_stripes = 1;
2627         stripe_index = 0;
2628         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2629                 if (unplug_page || (rw & (1 << BIO_RW)))
2630                         num_stripes = map->num_stripes;
2631                 else if (mirror_num)
2632                         stripe_index = mirror_num - 1;
2633                 else {
2634                         stripe_index = find_live_mirror(map, 0,
2635                                             map->num_stripes,
2636                                             current->pid % map->num_stripes);
2637                 }
2638
2639         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2640                 if (rw & (1 << BIO_RW))
2641                         num_stripes = map->num_stripes;
2642                 else if (mirror_num)
2643                         stripe_index = mirror_num - 1;
2644
2645         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2646                 int factor = map->num_stripes / map->sub_stripes;
2647
2648                 stripe_index = do_div(stripe_nr, factor);
2649                 stripe_index *= map->sub_stripes;
2650
2651                 if (unplug_page || (rw & (1 << BIO_RW)))
2652                         num_stripes = map->sub_stripes;
2653                 else if (mirror_num)
2654                         stripe_index += mirror_num - 1;
2655                 else {
2656                         stripe_index = find_live_mirror(map, stripe_index,
2657                                               map->sub_stripes, stripe_index +
2658                                               current->pid % map->sub_stripes);
2659                 }
2660         } else {
2661                 /*
2662                  * after this do_div call, stripe_nr is the number of stripes
2663                  * on this device we have to walk to find the data, and
2664                  * stripe_index is the number of our device in the stripe array
2665                  */
2666                 stripe_index = do_div(stripe_nr, map->num_stripes);
2667         }
2668         BUG_ON(stripe_index >= map->num_stripes);
2669
2670         for (i = 0; i < num_stripes; i++) {
2671                 if (unplug_page) {
2672                         struct btrfs_device *device;
2673                         struct backing_dev_info *bdi;
2674
2675                         device = map->stripes[stripe_index].dev;
2676                         if (device->bdev) {
2677                                 bdi = blk_get_backing_dev_info(device->bdev);
2678                                 if (bdi->unplug_io_fn)
2679                                         bdi->unplug_io_fn(bdi, unplug_page);
2680                         }
2681                 } else {
2682                         multi->stripes[i].physical =
2683                                 map->stripes[stripe_index].physical +
2684                                 stripe_offset + stripe_nr * map->stripe_len;
2685                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2686                 }
2687                 stripe_index++;
2688         }
2689         if (multi_ret) {
2690                 *multi_ret = multi;
2691                 multi->num_stripes = num_stripes;
2692                 multi->max_errors = max_errors;
2693         }
2694 out:
2695         free_extent_map(em);
2696         return 0;
2697 }
2698
2699 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2700                       u64 logical, u64 *length,
2701                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2702 {
2703         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2704                                  mirror_num, NULL);
2705 }
2706
2707 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2708                      u64 chunk_start, u64 physical, u64 devid,
2709                      u64 **logical, int *naddrs, int *stripe_len)
2710 {
2711         struct extent_map_tree *em_tree = &map_tree->map_tree;
2712         struct extent_map *em;
2713         struct map_lookup *map;
2714         u64 *buf;
2715         u64 bytenr;
2716         u64 length;
2717         u64 stripe_nr;
2718         int i, j, nr = 0;
2719
2720         spin_lock(&em_tree->lock);
2721         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2722         spin_unlock(&em_tree->lock);
2723
2724         BUG_ON(!em || em->start != chunk_start);
2725         map = (struct map_lookup *)em->bdev;
2726
2727         length = em->len;
2728         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2729                 do_div(length, map->num_stripes / map->sub_stripes);
2730         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2731                 do_div(length, map->num_stripes);
2732
2733         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2734         BUG_ON(!buf);
2735
2736         for (i = 0; i < map->num_stripes; i++) {
2737                 if (devid && map->stripes[i].dev->devid != devid)
2738                         continue;
2739                 if (map->stripes[i].physical > physical ||
2740                     map->stripes[i].physical + length <= physical)
2741                         continue;
2742
2743                 stripe_nr = physical - map->stripes[i].physical;
2744                 do_div(stripe_nr, map->stripe_len);
2745
2746                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2747                         stripe_nr = stripe_nr * map->num_stripes + i;
2748                         do_div(stripe_nr, map->sub_stripes);
2749                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2750                         stripe_nr = stripe_nr * map->num_stripes + i;
2751                 }
2752                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2753                 WARN_ON(nr >= map->num_stripes);
2754                 for (j = 0; j < nr; j++) {
2755                         if (buf[j] == bytenr)
2756                                 break;
2757                 }
2758                 if (j == nr) {
2759                         WARN_ON(nr >= map->num_stripes);
2760                         buf[nr++] = bytenr;
2761                 }
2762         }
2763
2764         for (i = 0; i > nr; i++) {
2765                 struct btrfs_multi_bio *multi;
2766                 struct btrfs_bio_stripe *stripe;
2767                 int ret;
2768
2769                 length = 1;
2770                 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2771                                       &length, &multi, 0);
2772                 BUG_ON(ret);
2773
2774                 stripe = multi->stripes;
2775                 for (j = 0; j < multi->num_stripes; j++) {
2776                         if (stripe->physical >= physical &&
2777                             physical < stripe->physical + length)
2778                                 break;
2779                 }
2780                 BUG_ON(j >= multi->num_stripes);
2781                 kfree(multi);
2782         }
2783
2784         *logical = buf;
2785         *naddrs = nr;
2786         *stripe_len = map->stripe_len;
2787
2788         free_extent_map(em);
2789         return 0;
2790 }
2791
2792 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2793                       u64 logical, struct page *page)
2794 {
2795         u64 length = PAGE_CACHE_SIZE;
2796         return __btrfs_map_block(map_tree, READ, logical, &length,
2797                                  NULL, 0, page);
2798 }
2799
2800 static void end_bio_multi_stripe(struct bio *bio, int err)
2801 {
2802         struct btrfs_multi_bio *multi = bio->bi_private;
2803         int is_orig_bio = 0;
2804
2805         if (err)
2806                 atomic_inc(&multi->error);
2807
2808         if (bio == multi->orig_bio)
2809                 is_orig_bio = 1;
2810
2811         if (atomic_dec_and_test(&multi->stripes_pending)) {
2812                 if (!is_orig_bio) {
2813                         bio_put(bio);
2814                         bio = multi->orig_bio;
2815                 }
2816                 bio->bi_private = multi->private;
2817                 bio->bi_end_io = multi->end_io;
2818                 /* only send an error to the higher layers if it is
2819                  * beyond the tolerance of the multi-bio
2820                  */
2821                 if (atomic_read(&multi->error) > multi->max_errors) {
2822                         err = -EIO;
2823                 } else if (err) {
2824                         /*
2825                          * this bio is actually up to date, we didn't
2826                          * go over the max number of errors
2827                          */
2828                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2829                         err = 0;
2830                 }
2831                 kfree(multi);
2832
2833                 bio_endio(bio, err);
2834         } else if (!is_orig_bio) {
2835                 bio_put(bio);
2836         }
2837 }
2838
2839 struct async_sched {
2840         struct bio *bio;
2841         int rw;
2842         struct btrfs_fs_info *info;
2843         struct btrfs_work work;
2844 };
2845
2846 /*
2847  * see run_scheduled_bios for a description of why bios are collected for
2848  * async submit.
2849  *
2850  * This will add one bio to the pending list for a device and make sure
2851  * the work struct is scheduled.
2852  */
2853 static noinline int schedule_bio(struct btrfs_root *root,
2854                                  struct btrfs_device *device,
2855                                  int rw, struct bio *bio)
2856 {
2857         int should_queue = 1;
2858         struct btrfs_pending_bios *pending_bios;
2859
2860         /* don't bother with additional async steps for reads, right now */
2861         if (!(rw & (1 << BIO_RW))) {
2862                 bio_get(bio);
2863                 submit_bio(rw, bio);
2864                 bio_put(bio);
2865                 return 0;
2866         }
2867
2868         /*
2869          * nr_async_bios allows us to reliably return congestion to the
2870          * higher layers.  Otherwise, the async bio makes it appear we have
2871          * made progress against dirty pages when we've really just put it
2872          * on a queue for later
2873          */
2874         atomic_inc(&root->fs_info->nr_async_bios);
2875         WARN_ON(bio->bi_next);
2876         bio->bi_next = NULL;
2877         bio->bi_rw |= rw;
2878
2879         spin_lock(&device->io_lock);
2880         if (bio_sync(bio))
2881                 pending_bios = &device->pending_sync_bios;
2882         else
2883                 pending_bios = &device->pending_bios;
2884
2885         if (pending_bios->tail)
2886                 pending_bios->tail->bi_next = bio;
2887
2888         pending_bios->tail = bio;
2889         if (!pending_bios->head)
2890                 pending_bios->head = bio;
2891         if (device->running_pending)
2892                 should_queue = 0;
2893
2894         spin_unlock(&device->io_lock);
2895
2896         if (should_queue)
2897                 btrfs_queue_worker(&root->fs_info->submit_workers,
2898                                    &device->work);
2899         return 0;
2900 }
2901
2902 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2903                   int mirror_num, int async_submit)
2904 {
2905         struct btrfs_mapping_tree *map_tree;
2906         struct btrfs_device *dev;
2907         struct bio *first_bio = bio;
2908         u64 logical = (u64)bio->bi_sector << 9;
2909         u64 length = 0;
2910         u64 map_length;
2911         struct btrfs_multi_bio *multi = NULL;
2912         int ret;
2913         int dev_nr = 0;
2914         int total_devs = 1;
2915
2916         length = bio->bi_size;
2917         map_tree = &root->fs_info->mapping_tree;
2918         map_length = length;
2919
2920         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2921                               mirror_num);
2922         BUG_ON(ret);
2923
2924         total_devs = multi->num_stripes;
2925         if (map_length < length) {
2926                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2927                        "len %llu\n", (unsigned long long)logical,
2928                        (unsigned long long)length,
2929                        (unsigned long long)map_length);
2930                 BUG();
2931         }
2932         multi->end_io = first_bio->bi_end_io;
2933         multi->private = first_bio->bi_private;
2934         multi->orig_bio = first_bio;
2935         atomic_set(&multi->stripes_pending, multi->num_stripes);
2936
2937         while (dev_nr < total_devs) {
2938                 if (total_devs > 1) {
2939                         if (dev_nr < total_devs - 1) {
2940                                 bio = bio_clone(first_bio, GFP_NOFS);
2941                                 BUG_ON(!bio);
2942                         } else {
2943                                 bio = first_bio;
2944                         }
2945                         bio->bi_private = multi;
2946                         bio->bi_end_io = end_bio_multi_stripe;
2947                 }
2948                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2949                 dev = multi->stripes[dev_nr].dev;
2950                 BUG_ON(rw == WRITE && !dev->writeable);
2951                 if (dev && dev->bdev) {
2952                         bio->bi_bdev = dev->bdev;
2953                         if (async_submit)
2954                                 schedule_bio(root, dev, rw, bio);
2955                         else
2956                                 submit_bio(rw, bio);
2957                 } else {
2958                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2959                         bio->bi_sector = logical >> 9;
2960                         bio_endio(bio, -EIO);
2961                 }
2962                 dev_nr++;
2963         }
2964         if (total_devs == 1)
2965                 kfree(multi);
2966         return 0;
2967 }
2968
2969 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2970                                        u8 *uuid, u8 *fsid)
2971 {
2972         struct btrfs_device *device;
2973         struct btrfs_fs_devices *cur_devices;
2974
2975         cur_devices = root->fs_info->fs_devices;
2976         while (cur_devices) {
2977                 if (!fsid ||
2978                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2979                         device = __find_device(&cur_devices->devices,
2980                                                devid, uuid);
2981                         if (device)
2982                                 return device;
2983                 }
2984                 cur_devices = cur_devices->seed;
2985         }
2986         return NULL;
2987 }
2988
2989 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2990                                             u64 devid, u8 *dev_uuid)
2991 {
2992         struct btrfs_device *device;
2993         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2994
2995         device = kzalloc(sizeof(*device), GFP_NOFS);
2996         if (!device)
2997                 return NULL;
2998         list_add(&device->dev_list,
2999                  &fs_devices->devices);
3000         device->barriers = 1;
3001         device->dev_root = root->fs_info->dev_root;
3002         device->devid = devid;
3003         device->work.func = pending_bios_fn;
3004         device->fs_devices = fs_devices;
3005         fs_devices->num_devices++;
3006         spin_lock_init(&device->io_lock);
3007         INIT_LIST_HEAD(&device->dev_alloc_list);
3008         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3009         return device;
3010 }
3011
3012 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3013                           struct extent_buffer *leaf,
3014                           struct btrfs_chunk *chunk)
3015 {
3016         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3017         struct map_lookup *map;
3018         struct extent_map *em;
3019         u64 logical;
3020         u64 length;
3021         u64 devid;
3022         u8 uuid[BTRFS_UUID_SIZE];
3023         int num_stripes;
3024         int ret;
3025         int i;
3026
3027         logical = key->offset;
3028         length = btrfs_chunk_length(leaf, chunk);
3029
3030         spin_lock(&map_tree->map_tree.lock);
3031         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3032         spin_unlock(&map_tree->map_tree.lock);
3033
3034         /* already mapped? */
3035         if (em && em->start <= logical && em->start + em->len > logical) {
3036                 free_extent_map(em);
3037                 return 0;
3038         } else if (em) {
3039                 free_extent_map(em);
3040         }
3041
3042         em = alloc_extent_map(GFP_NOFS);
3043         if (!em)
3044                 return -ENOMEM;
3045         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3046         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3047         if (!map) {
3048                 free_extent_map(em);
3049                 return -ENOMEM;
3050         }
3051
3052         em->bdev = (struct block_device *)map;
3053         em->start = logical;
3054         em->len = length;
3055         em->block_start = 0;
3056         em->block_len = em->len;
3057
3058         map->num_stripes = num_stripes;
3059         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3060         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3061         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3062         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3063         map->type = btrfs_chunk_type(leaf, chunk);
3064         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3065         for (i = 0; i < num_stripes; i++) {
3066                 map->stripes[i].physical =
3067                         btrfs_stripe_offset_nr(leaf, chunk, i);
3068                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3069                 read_extent_buffer(leaf, uuid, (unsigned long)
3070                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3071                                    BTRFS_UUID_SIZE);
3072                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3073                                                         NULL);
3074                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3075                         kfree(map);
3076                         free_extent_map(em);
3077                         return -EIO;
3078                 }
3079                 if (!map->stripes[i].dev) {
3080                         map->stripes[i].dev =
3081                                 add_missing_dev(root, devid, uuid);
3082                         if (!map->stripes[i].dev) {
3083                                 kfree(map);
3084                                 free_extent_map(em);
3085                                 return -EIO;
3086                         }
3087                 }
3088                 map->stripes[i].dev->in_fs_metadata = 1;
3089         }
3090
3091         spin_lock(&map_tree->map_tree.lock);
3092         ret = add_extent_mapping(&map_tree->map_tree, em);
3093         spin_unlock(&map_tree->map_tree.lock);
3094         BUG_ON(ret);
3095         free_extent_map(em);
3096
3097         return 0;
3098 }
3099
3100 static int fill_device_from_item(struct extent_buffer *leaf,
3101                                  struct btrfs_dev_item *dev_item,
3102                                  struct btrfs_device *device)
3103 {
3104         unsigned long ptr;
3105
3106         device->devid = btrfs_device_id(leaf, dev_item);
3107         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3108         device->total_bytes = device->disk_total_bytes;
3109         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3110         device->type = btrfs_device_type(leaf, dev_item);
3111         device->io_align = btrfs_device_io_align(leaf, dev_item);
3112         device->io_width = btrfs_device_io_width(leaf, dev_item);
3113         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3114
3115         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3116         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3117
3118         return 0;
3119 }
3120
3121 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3122 {
3123         struct btrfs_fs_devices *fs_devices;
3124         int ret;
3125
3126         mutex_lock(&uuid_mutex);
3127
3128         fs_devices = root->fs_info->fs_devices->seed;
3129         while (fs_devices) {
3130                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3131                         ret = 0;
3132                         goto out;
3133                 }
3134                 fs_devices = fs_devices->seed;
3135         }
3136
3137         fs_devices = find_fsid(fsid);
3138         if (!fs_devices) {
3139                 ret = -ENOENT;
3140                 goto out;
3141         }
3142
3143         fs_devices = clone_fs_devices(fs_devices);
3144         if (IS_ERR(fs_devices)) {
3145                 ret = PTR_ERR(fs_devices);
3146                 goto out;
3147         }
3148
3149         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3150                                    root->fs_info->bdev_holder);
3151         if (ret)
3152                 goto out;
3153
3154         if (!fs_devices->seeding) {
3155                 __btrfs_close_devices(fs_devices);
3156                 free_fs_devices(fs_devices);
3157                 ret = -EINVAL;
3158                 goto out;
3159         }
3160
3161         fs_devices->seed = root->fs_info->fs_devices->seed;
3162         root->fs_info->fs_devices->seed = fs_devices;
3163 out:
3164         mutex_unlock(&uuid_mutex);
3165         return ret;
3166 }
3167
3168 static int read_one_dev(struct btrfs_root *root,
3169                         struct extent_buffer *leaf,
3170                         struct btrfs_dev_item *dev_item)
3171 {
3172         struct btrfs_device *device;
3173         u64 devid;
3174         int ret;
3175         u8 fs_uuid[BTRFS_UUID_SIZE];
3176         u8 dev_uuid[BTRFS_UUID_SIZE];
3177
3178         devid = btrfs_device_id(leaf, dev_item);
3179         read_extent_buffer(leaf, dev_uuid,
3180                            (unsigned long)btrfs_device_uuid(dev_item),
3181                            BTRFS_UUID_SIZE);
3182         read_extent_buffer(leaf, fs_uuid,
3183                            (unsigned long)btrfs_device_fsid(dev_item),
3184                            BTRFS_UUID_SIZE);
3185
3186         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3187                 ret = open_seed_devices(root, fs_uuid);
3188                 if (ret && !btrfs_test_opt(root, DEGRADED))
3189                         return ret;
3190         }
3191
3192         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3193         if (!device || !device->bdev) {
3194                 if (!btrfs_test_opt(root, DEGRADED))
3195                         return -EIO;
3196
3197                 if (!device) {
3198                         printk(KERN_WARNING "warning devid %llu missing\n",
3199                                (unsigned long long)devid);
3200                         device = add_missing_dev(root, devid, dev_uuid);
3201                         if (!device)
3202                                 return -ENOMEM;
3203                 }
3204         }
3205
3206         if (device->fs_devices != root->fs_info->fs_devices) {
3207                 BUG_ON(device->writeable);
3208                 if (device->generation !=
3209                     btrfs_device_generation(leaf, dev_item))
3210                         return -EINVAL;
3211         }
3212
3213         fill_device_from_item(leaf, dev_item, device);
3214         device->dev_root = root->fs_info->dev_root;
3215         device->in_fs_metadata = 1;
3216         if (device->writeable)
3217                 device->fs_devices->total_rw_bytes += device->total_bytes;
3218         ret = 0;
3219         return ret;
3220 }
3221
3222 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3223 {
3224         struct btrfs_dev_item *dev_item;
3225
3226         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3227                                                      dev_item);
3228         return read_one_dev(root, buf, dev_item);
3229 }
3230
3231 int btrfs_read_sys_array(struct btrfs_root *root)
3232 {
3233         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3234         struct extent_buffer *sb;
3235         struct btrfs_disk_key *disk_key;
3236         struct btrfs_chunk *chunk;
3237         u8 *ptr;
3238         unsigned long sb_ptr;
3239         int ret = 0;
3240         u32 num_stripes;
3241         u32 array_size;
3242         u32 len = 0;
3243         u32 cur;
3244         struct btrfs_key key;
3245
3246         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3247                                           BTRFS_SUPER_INFO_SIZE);
3248         if (!sb)
3249                 return -ENOMEM;
3250         btrfs_set_buffer_uptodate(sb);
3251         btrfs_set_buffer_lockdep_class(sb, 0);
3252
3253         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3254         array_size = btrfs_super_sys_array_size(super_copy);
3255
3256         ptr = super_copy->sys_chunk_array;
3257         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3258         cur = 0;
3259
3260         while (cur < array_size) {
3261                 disk_key = (struct btrfs_disk_key *)ptr;
3262                 btrfs_disk_key_to_cpu(&key, disk_key);
3263
3264                 len = sizeof(*disk_key); ptr += len;
3265                 sb_ptr += len;
3266                 cur += len;
3267
3268                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3269                         chunk = (struct btrfs_chunk *)sb_ptr;
3270                         ret = read_one_chunk(root, &key, sb, chunk);
3271                         if (ret)
3272                                 break;
3273                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3274                         len = btrfs_chunk_item_size(num_stripes);
3275                 } else {
3276                         ret = -EIO;
3277                         break;
3278                 }
3279                 ptr += len;
3280                 sb_ptr += len;
3281                 cur += len;
3282         }
3283         free_extent_buffer(sb);
3284         return ret;
3285 }
3286
3287 int btrfs_read_chunk_tree(struct btrfs_root *root)
3288 {
3289         struct btrfs_path *path;
3290         struct extent_buffer *leaf;
3291         struct btrfs_key key;
3292         struct btrfs_key found_key;
3293         int ret;
3294         int slot;
3295
3296         root = root->fs_info->chunk_root;
3297
3298         path = btrfs_alloc_path();
3299         if (!path)
3300                 return -ENOMEM;
3301
3302         /* first we search for all of the device items, and then we
3303          * read in all of the chunk items.  This way we can create chunk
3304          * mappings that reference all of the devices that are afound
3305          */
3306         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3307         key.offset = 0;
3308         key.type = 0;
3309 again:
3310         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3311         while (1) {
3312                 leaf = path->nodes[0];
3313                 slot = path->slots[0];
3314                 if (slot >= btrfs_header_nritems(leaf)) {
3315                         ret = btrfs_next_leaf(root, path);
3316                         if (ret == 0)
3317                                 continue;
3318                         if (ret < 0)
3319                                 goto error;
3320                         break;
3321                 }
3322                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3323                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3324                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3325                                 break;
3326                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3327                                 struct btrfs_dev_item *dev_item;
3328                                 dev_item = btrfs_item_ptr(leaf, slot,
3329                                                   struct btrfs_dev_item);
3330                                 ret = read_one_dev(root, leaf, dev_item);
3331                                 if (ret)
3332                                         goto error;
3333                         }
3334                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3335                         struct btrfs_chunk *chunk;
3336                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3337                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3338                         if (ret)
3339                                 goto error;
3340                 }
3341                 path->slots[0]++;
3342         }
3343         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3344                 key.objectid = 0;
3345                 btrfs_release_path(root, path);
3346                 goto again;
3347         }
3348         ret = 0;
3349 error:
3350         btrfs_free_path(path);
3351         return ret;
3352 }