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