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 <linux/capability.h>
  25 #include <asm/div64.h>
  26 #include "compat.h"
  27 #include "ctree.h"
  28 #include "extent_map.h"
  29 #include "disk-io.h"
  30 #include "transaction.h"
  31 #include "print-tree.h"
  32 #include "volumes.h"
  33 #include "async-thread.h"
  34
  35 struct map_lookup {
  36         u64 type;
  37         int io_align;
  38         int io_width;
  39         int stripe_len;
  40         int sector_size;
  41         int num_stripes;
  42         int sub_stripes;
  43         struct btrfs_bio_stripe stripes[];
  44 };
  45
  46 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  47                                 struct btrfs_root *root,
  48                                 struct btrfs_device *device);
  49 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  50
  51 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  52                             (sizeof(struct btrfs_bio_stripe) * (n)))
  53
  54 static DEFINE_MUTEX(uuid_mutex);
  55 static LIST_HEAD(fs_uuids);
  56
  57 void btrfs_lock_volumes(void)
  58 {
  59         mutex_lock(&uuid_mutex);
  60 }
  61
  62 void btrfs_unlock_volumes(void)
  63 {
  64         mutex_unlock(&uuid_mutex);
  65 }
  66
  67 static void lock_chunks(struct btrfs_root *root)
  68 {
  69         mutex_lock(&root->fs_info->chunk_mutex);
  70 }
  71
  72 static void unlock_chunks(struct btrfs_root *root)
  73 {
  74         mutex_unlock(&root->fs_info->chunk_mutex);
  75 }
  76
  77 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  78 {
  79         struct btrfs_device *device;
  80         WARN_ON(fs_devices->opened);
  81         while (!list_empty(&fs_devices->devices)) {
  82                 device = list_entry(fs_devices->devices.next,
  83                                     struct btrfs_device, dev_list);
  84                 list_del(&device->dev_list);
  85                 kfree(device->name);
  86                 kfree(device);
  87         }
  88         kfree(fs_devices);
  89 }
  90
  91 int btrfs_cleanup_fs_uuids(void)
  92 {
  93         struct btrfs_fs_devices *fs_devices;
  94
  95         while (!list_empty(&fs_uuids)) {
  96                 fs_devices = list_entry(fs_uuids.next,
  97                                         struct btrfs_fs_devices, list);
  98                 list_del(&fs_devices->list);
  99                 free_fs_devices(fs_devices);
 100         }
 101         return 0;
 102 }
 103
 104 static noinline struct btrfs_device *__find_device(struct list_head *head,
 105                                                    u64 devid, u8 *uuid)
 106 {
 107         struct btrfs_device *dev;
 108
 109         list_for_each_entry(dev, head, dev_list) {
 110                 if (dev->devid == devid &&
 111                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 112                         return dev;
 113                 }
 114         }
 115         return NULL;
 116 }
 117
 118 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 119 {
 120         struct btrfs_fs_devices *fs_devices;
 121
 122         list_for_each_entry(fs_devices, &fs_uuids, list) {
 123                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 124                         return fs_devices;
 125         }
 126         return NULL;
 127 }
 128
 129 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 130                         struct bio *head, struct bio *tail)
 131 {
 132
 133         struct bio *old_head;
 134
 135         old_head = pending_bios->head;
 136         pending_bios->head = head;
 137         if (pending_bios->tail)
 138                 tail->bi_next = old_head;
 139         else
 140                 pending_bios->tail = tail;
 141 }
 142
 143 /*
 144  * we try to collect pending bios for a device so we don't get a large
 145  * number of procs sending bios down to the same device.  This greatly
 146  * improves the schedulers ability to collect and merge the bios.
 147  *
 148  * But, it also turns into a long list of bios to process and that is sure
 149  * to eventually make the worker thread block.  The solution here is to
 150  * make some progress and then put this work struct back at the end of
 151  * the list if the block device is congested.  This way, multiple devices
 152  * can make progress from a single worker thread.
 153  */
 154 static noinline int run_scheduled_bios(struct btrfs_device *device)
 155 {
 156         struct bio *pending;
 157         struct backing_dev_info *bdi;
 158         struct btrfs_fs_info *fs_info;
 159         struct btrfs_pending_bios *pending_bios;
 160         struct bio *tail;
 161         struct bio *cur;
 162         int again = 0;
 163         unsigned long num_run;
 164         unsigned long num_sync_run;
 165         unsigned long batch_run = 0;
 166         unsigned long limit;
 167         unsigned long last_waited = 0;
 168         int force_reg = 0;
 169
 170         bdi = blk_get_backing_dev_info(device->bdev);
 171         fs_info = device->dev_root->fs_info;
 172         limit = btrfs_async_submit_limit(fs_info);
 173         limit = limit * 2 / 3;
 174
 175         /* we want to make sure that every time we switch from the sync
 176          * list to the normal list, we unplug
 177          */
 178         num_sync_run = 0;
 179
 180 loop:
 181         spin_lock(&device->io_lock);
 182
 183 loop_lock:
 184         num_run = 0;
 185
 186         /* take all the bios off the list at once and process them
 187          * later on (without the lock held).  But, remember the
 188          * tail and other pointers so the bios can be properly reinserted
 189          * into the list if we hit congestion
 190          */
 191         if (!force_reg && device->pending_sync_bios.head) {
 192                 pending_bios = &device->pending_sync_bios;
 193                 force_reg = 1;
 194         } else {
 195                 pending_bios = &device->pending_bios;
 196                 force_reg = 0;
 197         }
 198
 199         pending = pending_bios->head;
 200         tail = pending_bios->tail;
 201         WARN_ON(pending && !tail);
 202
 203         /*
 204          * if pending was null this time around, no bios need processing
 205          * at all and we can stop.  Otherwise it'll loop back up again
 206          * and do an additional check so no bios are missed.
 207          *
 208          * device->running_pending is used to synchronize with the
 209          * schedule_bio code.
 210          */
 211         if (device->pending_sync_bios.head == NULL &&
 212             device->pending_bios.head == NULL) {
 213                 again = 0;
 214                 device->running_pending = 0;
 215         } else {
 216                 again = 1;
 217                 device->running_pending = 1;
 218         }
 219
 220         pending_bios->head = NULL;
 221         pending_bios->tail = NULL;
 222
 223         spin_unlock(&device->io_lock);
 224
 225         /*
 226          * if we're doing the regular priority list, make sure we unplug
 227          * for any high prio bios we've sent down
 228          */
 229         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
 230                 num_sync_run = 0;
 231                 blk_run_backing_dev(bdi, NULL);
 232         }
 233
 234         while (pending) {
 235
 236                 rmb();
 237                 /* we want to work on both lists, but do more bios on the
 238                  * sync list than the regular list
 239                  */
 240                 if ((num_run > 32 &&
 241                     pending_bios != &device->pending_sync_bios &&
 242                     device->pending_sync_bios.head) ||
 243                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 244                     device->pending_bios.head)) {
 245                         spin_lock(&device->io_lock);
 246                         requeue_list(pending_bios, pending, tail);
 247                         goto loop_lock;
 248                 }
 249
 250                 cur = pending;
 251                 pending = pending->bi_next;
 252                 cur->bi_next = NULL;
 253                 atomic_dec(&fs_info->nr_async_bios);
 254
 255                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 256                     waitqueue_active(&fs_info->async_submit_wait))
 257                         wake_up(&fs_info->async_submit_wait);
 258
 259                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 260                 submit_bio(cur->bi_rw, cur);
 261                 num_run++;
 262                 batch_run++;
 263
 264                 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
 265                         num_sync_run++;
 266
 267                 if (need_resched()) {
 268                         if (num_sync_run) {
 269                                 blk_run_backing_dev(bdi, NULL);
 270                                 num_sync_run = 0;
 271                         }
 272                         cond_resched();
 273                 }
 274
 275                 /*
 276                  * we made progress, there is more work to do and the bdi
 277                  * is now congested.  Back off and let other work structs
 278                  * run instead
 279                  */
 280                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 281                     fs_info->fs_devices->open_devices > 1) {
 282                         struct io_context *ioc;
 283
 284                         ioc = current->io_context;
 285
 286                         /*
 287                          * the main goal here is that we don't want to
 288                          * block if we're going to be able to submit
 289                          * more requests without blocking.
 290                          *
 291                          * This code does two great things, it pokes into
 292                          * the elevator code from a filesystem _and_
 293                          * it makes assumptions about how batching works.
 294                          */
 295                         if (ioc && ioc->nr_batch_requests > 0 &&
 296                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 297                             (last_waited == 0 ||
 298                              ioc->last_waited == last_waited)) {
 299                                 /*
 300                                  * we want to go through our batch of
 301                                  * requests and stop.  So, we copy out
 302                                  * the ioc->last_waited time and test
 303                                  * against it before looping
 304                                  */
 305                                 last_waited = ioc->last_waited;
 306                                 if (need_resched()) {
 307                                         if (num_sync_run) {
 308                                                 blk_run_backing_dev(bdi, NULL);
 309                                                 num_sync_run = 0;
 310                                         }
 311                                         cond_resched();
 312                                 }
 313                                 continue;
 314                         }
 315                         spin_lock(&device->io_lock);
 316                         requeue_list(pending_bios, pending, tail);
 317                         device->running_pending = 1;
 318
 319                         spin_unlock(&device->io_lock);
 320                         btrfs_requeue_work(&device->work);
 321                         goto done;
 322                 }
 323         }
 324
 325         if (num_sync_run) {
 326                 num_sync_run = 0;
 327                 blk_run_backing_dev(bdi, NULL);
 328         }
 329
 330         cond_resched();
 331         if (again)
 332                 goto loop;
 333
 334         spin_lock(&device->io_lock);
 335         if (device->pending_bios.head || device->pending_sync_bios.head)
 336                 goto loop_lock;
 337         spin_unlock(&device->io_lock);
 338
 339         /*
 340          * IO has already been through a long path to get here.  Checksumming,
 341          * async helper threads, perhaps compression.  We've done a pretty
 342          * good job of collecting a batch of IO and should just unplug
 343          * the device right away.
 344          *
 345          * This will help anyone who is waiting on the IO, they might have
 346          * already unplugged, but managed to do so before the bio they
 347          * cared about found its way down here.
 348          */
 349         blk_run_backing_dev(bdi, NULL);
 350 done:
 351         return 0;
 352 }
 353
 354 static void pending_bios_fn(struct btrfs_work *work)
 355 {
 356         struct btrfs_device *device;
 357
 358         device = container_of(work, struct btrfs_device, work);
 359         run_scheduled_bios(device);
 360 }
 361
 362 static noinline int device_list_add(const char *path,
 363                            struct btrfs_super_block *disk_super,
 364                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 365 {
 366         struct btrfs_device *device;
 367         struct btrfs_fs_devices *fs_devices;
 368         u64 found_transid = btrfs_super_generation(disk_super);
 369
 370         fs_devices = find_fsid(disk_super->fsid);
 371         if (!fs_devices) {
 372                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 373                 if (!fs_devices)
 374                         return -ENOMEM;
 375                 INIT_LIST_HEAD(&fs_devices->devices);
 376                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 377                 list_add(&fs_devices->list, &fs_uuids);
 378                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 379                 fs_devices->latest_devid = devid;
 380                 fs_devices->latest_trans = found_transid;
 381                 mutex_init(&fs_devices->device_list_mutex);
 382                 device = NULL;
 383         } else {
 384                 device = __find_device(&fs_devices->devices, devid,
 385                                        disk_super->dev_item.uuid);
 386         }
 387         if (!device) {
 388                 if (fs_devices->opened)
 389                         return -EBUSY;
 390
 391                 device = kzalloc(sizeof(*device), GFP_NOFS);
 392                 if (!device) {
 393                         /* we can safely leave the fs_devices entry around */
 394                         return -ENOMEM;
 395                 }
 396                 device->devid = devid;
 397                 device->work.func = pending_bios_fn;
 398                 memcpy(device->uuid, disk_super->dev_item.uuid,
 399                        BTRFS_UUID_SIZE);
 400                 device->barriers = 1;
 401                 spin_lock_init(&device->io_lock);
 402                 device->name = kstrdup(path, GFP_NOFS);
 403                 if (!device->name) {
 404                         kfree(device);
 405                         return -ENOMEM;
 406                 }
 407                 INIT_LIST_HEAD(&device->dev_alloc_list);
 408
 409                 mutex_lock(&fs_devices->device_list_mutex);
 410                 list_add(&device->dev_list, &fs_devices->devices);
 411                 mutex_unlock(&fs_devices->device_list_mutex);
 412
 413                 device->fs_devices = fs_devices;
 414                 fs_devices->num_devices++;
 415         }
 416
 417         if (found_transid > fs_devices->latest_trans) {
 418                 fs_devices->latest_devid = devid;
 419                 fs_devices->latest_trans = found_transid;
 420         }
 421         *fs_devices_ret = fs_devices;
 422         return 0;
 423 }
 424
 425 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 426 {
 427         struct btrfs_fs_devices *fs_devices;
 428         struct btrfs_device *device;
 429         struct btrfs_device *orig_dev;
 430
 431         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 432         if (!fs_devices)
 433                 return ERR_PTR(-ENOMEM);
 434
 435         INIT_LIST_HEAD(&fs_devices->devices);
 436         INIT_LIST_HEAD(&fs_devices->alloc_list);
 437         INIT_LIST_HEAD(&fs_devices->list);
 438         mutex_init(&fs_devices->device_list_mutex);
 439         fs_devices->latest_devid = orig->latest_devid;
 440         fs_devices->latest_trans = orig->latest_trans;
 441         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 442
 443         mutex_lock(&orig->device_list_mutex);
 444         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 445                 device = kzalloc(sizeof(*device), GFP_NOFS);
 446                 if (!device)
 447                         goto error;
 448
 449                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 450                 if (!device->name) {
 451                         kfree(device);
 452                         goto error;
 453                 }
 454
 455                 device->devid = orig_dev->devid;
 456                 device->work.func = pending_bios_fn;
 457                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 458                 device->barriers = 1;
 459                 spin_lock_init(&device->io_lock);
 460                 INIT_LIST_HEAD(&device->dev_list);
 461                 INIT_LIST_HEAD(&device->dev_alloc_list);
 462
 463                 list_add(&device->dev_list, &fs_devices->devices);
 464                 device->fs_devices = fs_devices;
 465                 fs_devices->num_devices++;
 466         }
 467         mutex_unlock(&orig->device_list_mutex);
 468         return fs_devices;
 469 error:
 470         mutex_unlock(&orig->device_list_mutex);
 471         free_fs_devices(fs_devices);
 472         return ERR_PTR(-ENOMEM);
 473 }
 474
 475 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 476 {
 477         struct btrfs_device *device, *next;
 478
 479         mutex_lock(&uuid_mutex);
 480 again:
 481         mutex_lock(&fs_devices->device_list_mutex);
 482         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 483                 if (device->in_fs_metadata)
 484                         continue;
 485
 486                 if (device->bdev) {
 487                         close_bdev_exclusive(device->bdev, device->mode);
 488                         device->bdev = NULL;
 489                         fs_devices->open_devices--;
 490                 }
 491                 if (device->writeable) {
 492                         list_del_init(&device->dev_alloc_list);
 493                         device->writeable = 0;
 494                         fs_devices->rw_devices--;
 495                 }
 496                 list_del_init(&device->dev_list);
 497                 fs_devices->num_devices--;
 498                 kfree(device->name);
 499                 kfree(device);
 500         }
 501         mutex_unlock(&fs_devices->device_list_mutex);
 502
 503         if (fs_devices->seed) {
 504                 fs_devices = fs_devices->seed;
 505                 goto again;
 506         }
 507
 508         mutex_unlock(&uuid_mutex);
 509         return 0;
 510 }
 511
 512 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 513 {
 514         struct btrfs_device *device;
 515
 516         if (--fs_devices->opened > 0)
 517                 return 0;
 518
 519         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 520                 if (device->bdev) {
 521                         close_bdev_exclusive(device->bdev, device->mode);
 522                         fs_devices->open_devices--;
 523                 }
 524                 if (device->writeable) {
 525                         list_del_init(&device->dev_alloc_list);
 526                         fs_devices->rw_devices--;
 527                 }
 528
 529                 device->bdev = NULL;
 530                 device->writeable = 0;
 531                 device->in_fs_metadata = 0;
 532         }
 533         WARN_ON(fs_devices->open_devices);
 534         WARN_ON(fs_devices->rw_devices);
 535         fs_devices->opened = 0;
 536         fs_devices->seeding = 0;
 537
 538         return 0;
 539 }
 540
 541 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 542 {
 543         struct btrfs_fs_devices *seed_devices = NULL;
 544         int ret;
 545
 546         mutex_lock(&uuid_mutex);
 547         ret = __btrfs_close_devices(fs_devices);
 548         if (!fs_devices->opened) {
 549                 seed_devices = fs_devices->seed;
 550                 fs_devices->seed = NULL;
 551         }
 552         mutex_unlock(&uuid_mutex);
 553
 554         while (seed_devices) {
 555                 fs_devices = seed_devices;
 556                 seed_devices = fs_devices->seed;
 557                 __btrfs_close_devices(fs_devices);
 558                 free_fs_devices(fs_devices);
 559         }
 560         return ret;
 561 }
 562
 563 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 564                                 fmode_t flags, void *holder)
 565 {
 566         struct block_device *bdev;
 567         struct list_head *head = &fs_devices->devices;
 568         struct btrfs_device *device;
 569         struct block_device *latest_bdev = NULL;
 570         struct buffer_head *bh;
 571         struct btrfs_super_block *disk_super;
 572         u64 latest_devid = 0;
 573         u64 latest_transid = 0;
 574         u64 devid;
 575         int seeding = 1;
 576         int ret = 0;
 577
 578         list_for_each_entry(device, head, dev_list) {
 579                 if (device->bdev)
 580                         continue;
 581                 if (!device->name)
 582                         continue;
 583
 584                 bdev = open_bdev_exclusive(device->name, flags, holder);
 585                 if (IS_ERR(bdev)) {
 586                         printk(KERN_INFO "open %s failed\n", device->name);
 587                         goto error;
 588                 }
 589                 set_blocksize(bdev, 4096);
 590
 591                 bh = btrfs_read_dev_super(bdev);
 592                 if (!bh)
 593                         goto error_close;
 594
 595                 disk_super = (struct btrfs_super_block *)bh->b_data;
 596                 devid = le64_to_cpu(disk_super->dev_item.devid);
 597                 if (devid != device->devid)
 598                         goto error_brelse;
 599
 600                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 601                            BTRFS_UUID_SIZE))
 602                         goto error_brelse;
 603
 604                 device->generation = btrfs_super_generation(disk_super);
 605                 if (!latest_transid || device->generation > latest_transid) {
 606                         latest_devid = devid;
 607                         latest_transid = device->generation;
 608                         latest_bdev = bdev;
 609                 }
 610
 611                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 612                         device->writeable = 0;
 613                 } else {
 614                         device->writeable = !bdev_read_only(bdev);
 615                         seeding = 0;
 616                 }
 617
 618                 device->bdev = bdev;
 619                 device->in_fs_metadata = 0;
 620                 device->mode = flags;
 621
 622                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 623                         fs_devices->rotating = 1;
 624
 625                 fs_devices->open_devices++;
 626                 if (device->writeable) {
 627                         fs_devices->rw_devices++;
 628                         list_add(&device->dev_alloc_list,
 629                                  &fs_devices->alloc_list);
 630                 }
 631                 continue;
 632
 633 error_brelse:
 634                 brelse(bh);
 635 error_close:
 636                 close_bdev_exclusive(bdev, FMODE_READ);
 637 error:
 638                 continue;
 639         }
 640         if (fs_devices->open_devices == 0) {
 641                 ret = -EIO;
 642                 goto out;
 643         }
 644         fs_devices->seeding = seeding;
 645         fs_devices->opened = 1;
 646         fs_devices->latest_bdev = latest_bdev;
 647         fs_devices->latest_devid = latest_devid;
 648         fs_devices->latest_trans = latest_transid;
 649         fs_devices->total_rw_bytes = 0;
 650 out:
 651         return ret;
 652 }
 653
 654 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 655                        fmode_t flags, void *holder)
 656 {
 657         int ret;
 658
 659         mutex_lock(&uuid_mutex);
 660         if (fs_devices->opened) {
 661                 fs_devices->opened++;
 662                 ret = 0;
 663         } else {
 664                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 665         }
 666         mutex_unlock(&uuid_mutex);
 667         return ret;
 668 }
 669
 670 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 671                           struct btrfs_fs_devices **fs_devices_ret)
 672 {
 673         struct btrfs_super_block *disk_super;
 674         struct block_device *bdev;
 675         struct buffer_head *bh;
 676         int ret;
 677         u64 devid;
 678         u64 transid;
 679
 680         mutex_lock(&uuid_mutex);
 681
 682         bdev = open_bdev_exclusive(path, flags, holder);
 683
 684         if (IS_ERR(bdev)) {
 685                 ret = PTR_ERR(bdev);
 686                 goto error;
 687         }
 688
 689         ret = set_blocksize(bdev, 4096);
 690         if (ret)
 691                 goto error_close;
 692         bh = btrfs_read_dev_super(bdev);
 693         if (!bh) {
 694                 ret = -EIO;
 695                 goto error_close;
 696         }
 697         disk_super = (struct btrfs_super_block *)bh->b_data;
 698         devid = le64_to_cpu(disk_super->dev_item.devid);
 699         transid = btrfs_super_generation(disk_super);
 700         if (disk_super->label[0])
 701                 printk(KERN_INFO "device label %s ", disk_super->label);
 702         else {
 703                 /* FIXME, make a readl uuid parser */
 704                 printk(KERN_INFO "device fsid %llx-%llx ",
 705                        *(unsigned long long *)disk_super->fsid,
 706                        *(unsigned long long *)(disk_super->fsid + 8));
 707         }
 708         printk(KERN_CONT "devid %llu transid %llu %s\n",
 709                (unsigned long long)devid, (unsigned long long)transid, path);
 710         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 711
 712         brelse(bh);
 713 error_close:
 714         close_bdev_exclusive(bdev, flags);
 715 error:
 716         mutex_unlock(&uuid_mutex);
 717         return ret;
 718 }
 719
 720 /*
 721  * this uses a pretty simple search, the expectation is that it is
 722  * called very infrequently and that a given device has a small number
 723  * of extents
 724  */
 725 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 726                          struct btrfs_device *device, u64 num_bytes,
 727                          u64 *start, u64 *max_avail)
 728 {
 729         struct btrfs_key key;
 730         struct btrfs_root *root = device->dev_root;
 731         struct btrfs_dev_extent *dev_extent = NULL;
 732         struct btrfs_path *path;
 733         u64 hole_size = 0;
 734         u64 last_byte = 0;
 735         u64 search_start = 0;
 736         u64 search_end = device->total_bytes;
 737         int ret;
 738         int slot = 0;
 739         int start_found;
 740         struct extent_buffer *l;
 741
 742         path = btrfs_alloc_path();
 743         if (!path)
 744                 return -ENOMEM;
 745         path->reada = 2;
 746         start_found = 0;
 747
 748         /* FIXME use last free of some kind */
 749
 750         /* we don't want to overwrite the superblock on the drive,
 751          * so we make sure to start at an offset of at least 1MB
 752          */
 753         search_start = max((u64)1024 * 1024, search_start);
 754
 755         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 756                 search_start = max(root->fs_info->alloc_start, search_start);
 757
 758         key.objectid = device->devid;
 759         key.offset = search_start;
 760         key.type = BTRFS_DEV_EXTENT_KEY;
 761         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 762         if (ret < 0)
 763                 goto error;
 764         if (ret > 0) {
 765                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 766                 if (ret < 0)
 767                         goto error;
 768                 if (ret > 0)
 769                         start_found = 1;
 770         }
 771         l = path->nodes[0];
 772         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 773         while (1) {
 774                 l = path->nodes[0];
 775                 slot = path->slots[0];
 776                 if (slot >= btrfs_header_nritems(l)) {
 777                         ret = btrfs_next_leaf(root, path);
 778                         if (ret == 0)
 779                                 continue;
 780                         if (ret < 0)
 781                                 goto error;
 782 no_more_items:
 783                         if (!start_found) {
 784                                 if (search_start >= search_end) {
 785                                         ret = -ENOSPC;
 786                                         goto error;
 787                                 }
 788                                 *start = search_start;
 789                                 start_found = 1;
 790                                 goto check_pending;
 791                         }
 792                         *start = last_byte > search_start ?
 793                                 last_byte : search_start;
 794                         if (search_end <= *start) {
 795                                 ret = -ENOSPC;
 796                                 goto error;
 797                         }
 798                         goto check_pending;
 799                 }
 800                 btrfs_item_key_to_cpu(l, &key, slot);
 801
 802                 if (key.objectid < device->devid)
 803                         goto next;
 804
 805                 if (key.objectid > device->devid)
 806                         goto no_more_items;
 807
 808                 if (key.offset >= search_start && key.offset > last_byte &&
 809                     start_found) {
 810                         if (last_byte < search_start)
 811                                 last_byte = search_start;
 812                         hole_size = key.offset - last_byte;
 813
 814                         if (hole_size > *max_avail)
 815                                 *max_avail = hole_size;
 816
 817                         if (key.offset > last_byte &&
 818                             hole_size >= num_bytes) {
 819                                 *start = last_byte;
 820                                 goto check_pending;
 821                         }
 822                 }
 823                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 824                         goto next;
 825
 826                 start_found = 1;
 827                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 828                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 829 next:
 830                 path->slots[0]++;
 831                 cond_resched();
 832         }
 833 check_pending:
 834         /* we have to make sure we didn't find an extent that has already
 835          * been allocated by the map tree or the original allocation
 836          */
 837         BUG_ON(*start < search_start);
 838
 839         if (*start + num_bytes > search_end) {
 840                 ret = -ENOSPC;
 841                 goto error;
 842         }
 843         /* check for pending inserts here */
 844         ret = 0;
 845
 846 error:
 847         btrfs_free_path(path);
 848         return ret;
 849 }
 850
 851 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 852                           struct btrfs_device *device,
 853                           u64 start)
 854 {
 855         int ret;
 856         struct btrfs_path *path;
 857         struct btrfs_root *root = device->dev_root;
 858         struct btrfs_key key;
 859         struct btrfs_key found_key;
 860         struct extent_buffer *leaf = NULL;
 861         struct btrfs_dev_extent *extent = NULL;
 862
 863         path = btrfs_alloc_path();
 864         if (!path)
 865                 return -ENOMEM;
 866
 867         key.objectid = device->devid;
 868         key.offset = start;
 869         key.type = BTRFS_DEV_EXTENT_KEY;
 870
 871         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 872         if (ret > 0) {
 873                 ret = btrfs_previous_item(root, path, key.objectid,
 874                                           BTRFS_DEV_EXTENT_KEY);
 875                 BUG_ON(ret);
 876                 leaf = path->nodes[0];
 877                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 878                 extent = btrfs_item_ptr(leaf, path->slots[0],
 879                                         struct btrfs_dev_extent);
 880                 BUG_ON(found_key.offset > start || found_key.offset +
 881                        btrfs_dev_extent_length(leaf, extent) < start);
 882                 ret = 0;
 883         } else if (ret == 0) {
 884                 leaf = path->nodes[0];
 885                 extent = btrfs_item_ptr(leaf, path->slots[0],
 886                                         struct btrfs_dev_extent);
 887         }
 888         BUG_ON(ret);
 889
 890         if (device->bytes_used > 0)
 891                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 892         ret = btrfs_del_item(trans, root, path);
 893         BUG_ON(ret);
 894
 895         btrfs_free_path(path);
 896         return ret;
 897 }
 898
 899 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 900                            struct btrfs_device *device,
 901                            u64 chunk_tree, u64 chunk_objectid,
 902                            u64 chunk_offset, u64 start, u64 num_bytes)
 903 {
 904         int ret;
 905         struct btrfs_path *path;
 906         struct btrfs_root *root = device->dev_root;
 907         struct btrfs_dev_extent *extent;
 908         struct extent_buffer *leaf;
 909         struct btrfs_key key;
 910
 911         WARN_ON(!device->in_fs_metadata);
 912         path = btrfs_alloc_path();
 913         if (!path)
 914                 return -ENOMEM;
 915
 916         key.objectid = device->devid;
 917         key.offset = start;
 918         key.type = BTRFS_DEV_EXTENT_KEY;
 919         ret = btrfs_insert_empty_item(trans, root, path, &key,
 920                                       sizeof(*extent));
 921         BUG_ON(ret);
 922
 923         leaf = path->nodes[0];
 924         extent = btrfs_item_ptr(leaf, path->slots[0],
 925                                 struct btrfs_dev_extent);
 926         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 927         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 928         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 929
 930         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 931                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 932                     BTRFS_UUID_SIZE);
 933
 934         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 935         btrfs_mark_buffer_dirty(leaf);
 936         btrfs_free_path(path);
 937         return ret;
 938 }
 939
 940 static noinline int find_next_chunk(struct btrfs_root *root,
 941                                     u64 objectid, u64 *offset)
 942 {
 943         struct btrfs_path *path;
 944         int ret;
 945         struct btrfs_key key;
 946         struct btrfs_chunk *chunk;
 947         struct btrfs_key found_key;
 948
 949         path = btrfs_alloc_path();
 950         BUG_ON(!path);
 951
 952         key.objectid = objectid;
 953         key.offset = (u64)-1;
 954         key.type = BTRFS_CHUNK_ITEM_KEY;
 955
 956         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 957         if (ret < 0)
 958                 goto error;
 959
 960         BUG_ON(ret == 0);
 961
 962         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 963         if (ret) {
 964                 *offset = 0;
 965         } else {
 966                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 967                                       path->slots[0]);
 968                 if (found_key.objectid != objectid)
 969                         *offset = 0;
 970                 else {
 971                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 972                                                struct btrfs_chunk);
 973                         *offset = found_key.offset +
 974                                 btrfs_chunk_length(path->nodes[0], chunk);
 975                 }
 976         }
 977         ret = 0;
 978 error:
 979         btrfs_free_path(path);
 980         return ret;
 981 }
 982
 983 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
 984 {
 985         int ret;
 986         struct btrfs_key key;
 987         struct btrfs_key found_key;
 988         struct btrfs_path *path;
 989
 990         root = root->fs_info->chunk_root;
 991
 992         path = btrfs_alloc_path();
 993         if (!path)
 994                 return -ENOMEM;
 995
 996         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 997         key.type = BTRFS_DEV_ITEM_KEY;
 998         key.offset = (u64)-1;
 999
1000         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1001         if (ret < 0)
1002                 goto error;
1003
1004         BUG_ON(ret == 0);
1005
1006         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1007                                   BTRFS_DEV_ITEM_KEY);
1008         if (ret) {
1009                 *objectid = 1;
1010         } else {
1011                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1012                                       path->slots[0]);
1013                 *objectid = found_key.offset + 1;
1014         }
1015         ret = 0;
1016 error:
1017         btrfs_free_path(path);
1018         return ret;
1019 }
1020
1021 /*
1022  * the device information is stored in the chunk root
1023  * the btrfs_device struct should be fully filled in
1024  */
1025 int btrfs_add_device(struct btrfs_trans_handle *trans,
1026                      struct btrfs_root *root,
1027                      struct btrfs_device *device)
1028 {
1029         int ret;
1030         struct btrfs_path *path;
1031         struct btrfs_dev_item *dev_item;
1032         struct extent_buffer *leaf;
1033         struct btrfs_key key;
1034         unsigned long ptr;
1035
1036         root = root->fs_info->chunk_root;
1037
1038         path = btrfs_alloc_path();
1039         if (!path)
1040                 return -ENOMEM;
1041
1042         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1043         key.type = BTRFS_DEV_ITEM_KEY;
1044         key.offset = device->devid;
1045
1046         ret = btrfs_insert_empty_item(trans, root, path, &key,
1047                                       sizeof(*dev_item));
1048         if (ret)
1049                 goto out;
1050
1051         leaf = path->nodes[0];
1052         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1053
1054         btrfs_set_device_id(leaf, dev_item, device->devid);
1055         btrfs_set_device_generation(leaf, dev_item, 0);
1056         btrfs_set_device_type(leaf, dev_item, device->type);
1057         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1058         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1059         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1060         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1061         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1062         btrfs_set_device_group(leaf, dev_item, 0);
1063         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1064         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1065         btrfs_set_device_start_offset(leaf, dev_item, 0);
1066
1067         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1068         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1069         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1070         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1071         btrfs_mark_buffer_dirty(leaf);
1072
1073         ret = 0;
1074 out:
1075         btrfs_free_path(path);
1076         return ret;
1077 }
1078
1079 static int btrfs_rm_dev_item(struct btrfs_root *root,
1080                              struct btrfs_device *device)
1081 {
1082         int ret;
1083         struct btrfs_path *path;
1084         struct btrfs_key key;
1085         struct btrfs_trans_handle *trans;
1086
1087         root = root->fs_info->chunk_root;
1088
1089         path = btrfs_alloc_path();
1090         if (!path)
1091                 return -ENOMEM;
1092
1093         trans = btrfs_start_transaction(root, 1);
1094         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1095         key.type = BTRFS_DEV_ITEM_KEY;
1096         key.offset = device->devid;
1097         lock_chunks(root);
1098
1099         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1100         if (ret < 0)
1101                 goto out;
1102
1103         if (ret > 0) {
1104                 ret = -ENOENT;
1105                 goto out;
1106         }
1107
1108         ret = btrfs_del_item(trans, root, path);
1109         if (ret)
1110                 goto out;
1111 out:
1112         btrfs_free_path(path);
1113         unlock_chunks(root);
1114         btrfs_commit_transaction(trans, root);
1115         return ret;
1116 }
1117
1118 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1119 {
1120         struct btrfs_device *device;
1121         struct btrfs_device *next_device;
1122         struct block_device *bdev;
1123         struct buffer_head *bh = NULL;
1124         struct btrfs_super_block *disk_super;
1125         u64 all_avail;
1126         u64 devid;
1127         u64 num_devices;
1128         u8 *dev_uuid;
1129         int ret = 0;
1130
1131         mutex_lock(&uuid_mutex);
1132         mutex_lock(&root->fs_info->volume_mutex);
1133
1134         all_avail = root->fs_info->avail_data_alloc_bits |
1135                 root->fs_info->avail_system_alloc_bits |
1136                 root->fs_info->avail_metadata_alloc_bits;
1137
1138         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1139             root->fs_info->fs_devices->num_devices <= 4) {
1140                 printk(KERN_ERR "btrfs: unable to go below four devices "
1141                        "on raid10\n");
1142                 ret = -EINVAL;
1143                 goto out;
1144         }
1145
1146         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1147             root->fs_info->fs_devices->num_devices <= 2) {
1148                 printk(KERN_ERR "btrfs: unable to go below two "
1149                        "devices on raid1\n");
1150                 ret = -EINVAL;
1151                 goto out;
1152         }
1153
1154         if (strcmp(device_path, "missing") == 0) {
1155                 struct list_head *devices;
1156                 struct btrfs_device *tmp;
1157
1158                 device = NULL;
1159                 devices = &root->fs_info->fs_devices->devices;
1160                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1161                 list_for_each_entry(tmp, devices, dev_list) {
1162                         if (tmp->in_fs_metadata && !tmp->bdev) {
1163                                 device = tmp;
1164                                 break;
1165                         }
1166                 }
1167                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1168                 bdev = NULL;
1169                 bh = NULL;
1170                 disk_super = NULL;
1171                 if (!device) {
1172                         printk(KERN_ERR "btrfs: no missing devices found to "
1173                                "remove\n");
1174                         goto out;
1175                 }
1176         } else {
1177                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1178                                       root->fs_info->bdev_holder);
1179                 if (IS_ERR(bdev)) {
1180                         ret = PTR_ERR(bdev);
1181                         goto out;
1182                 }
1183
1184                 set_blocksize(bdev, 4096);
1185                 bh = btrfs_read_dev_super(bdev);
1186                 if (!bh) {
1187                         ret = -EIO;
1188                         goto error_close;
1189                 }
1190                 disk_super = (struct btrfs_super_block *)bh->b_data;
1191                 devid = le64_to_cpu(disk_super->dev_item.devid);
1192                 dev_uuid = disk_super->dev_item.uuid;
1193                 device = btrfs_find_device(root, devid, dev_uuid,
1194                                            disk_super->fsid);
1195                 if (!device) {
1196                         ret = -ENOENT;
1197                         goto error_brelse;
1198                 }
1199         }
1200
1201         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1202                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1203                        "device\n");
1204                 ret = -EINVAL;
1205                 goto error_brelse;
1206         }
1207
1208         if (device->writeable) {
1209                 list_del_init(&device->dev_alloc_list);
1210                 root->fs_info->fs_devices->rw_devices--;
1211         }
1212
1213         ret = btrfs_shrink_device(device, 0);
1214         if (ret)
1215                 goto error_brelse;
1216
1217         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1218         if (ret)
1219                 goto error_brelse;
1220
1221         device->in_fs_metadata = 0;
1222
1223         /*
1224          * the device list mutex makes sure that we don't change
1225          * the device list while someone else is writing out all
1226          * the device supers.
1227          */
1228         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1229         list_del_init(&device->dev_list);
1230         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1231
1232         device->fs_devices->num_devices--;
1233
1234         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1235                                  struct btrfs_device, dev_list);
1236         if (device->bdev == root->fs_info->sb->s_bdev)
1237                 root->fs_info->sb->s_bdev = next_device->bdev;
1238         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1239                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1240
1241         if (device->bdev) {
1242                 close_bdev_exclusive(device->bdev, device->mode);
1243                 device->bdev = NULL;
1244                 device->fs_devices->open_devices--;
1245         }
1246
1247         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1248         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1249
1250         if (device->fs_devices->open_devices == 0) {
1251                 struct btrfs_fs_devices *fs_devices;
1252                 fs_devices = root->fs_info->fs_devices;
1253                 while (fs_devices) {
1254                         if (fs_devices->seed == device->fs_devices)
1255                                 break;
1256                         fs_devices = fs_devices->seed;
1257                 }
1258                 fs_devices->seed = device->fs_devices->seed;
1259                 device->fs_devices->seed = NULL;
1260                 __btrfs_close_devices(device->fs_devices);
1261                 free_fs_devices(device->fs_devices);
1262         }
1263
1264         /*
1265          * at this point, the device is zero sized.  We want to
1266          * remove it from the devices list and zero out the old super
1267          */
1268         if (device->writeable) {
1269                 /* make sure this device isn't detected as part of
1270                  * the FS anymore
1271                  */
1272                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1273                 set_buffer_dirty(bh);
1274                 sync_dirty_buffer(bh);
1275         }
1276
1277         kfree(device->name);
1278         kfree(device);
1279         ret = 0;
1280
1281 error_brelse:
1282         brelse(bh);
1283 error_close:
1284         if (bdev)
1285                 close_bdev_exclusive(bdev, FMODE_READ);
1286 out:
1287         mutex_unlock(&root->fs_info->volume_mutex);
1288         mutex_unlock(&uuid_mutex);
1289         return ret;
1290 }
1291
1292 /*
1293  * does all the dirty work required for changing file system's UUID.
1294  */
1295 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1296                                 struct btrfs_root *root)
1297 {
1298         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1299         struct btrfs_fs_devices *old_devices;
1300         struct btrfs_fs_devices *seed_devices;
1301         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1302         struct btrfs_device *device;
1303         u64 super_flags;
1304
1305         BUG_ON(!mutex_is_locked(&uuid_mutex));
1306         if (!fs_devices->seeding)
1307                 return -EINVAL;
1308
1309         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1310         if (!seed_devices)
1311                 return -ENOMEM;
1312
1313         old_devices = clone_fs_devices(fs_devices);
1314         if (IS_ERR(old_devices)) {
1315                 kfree(seed_devices);
1316                 return PTR_ERR(old_devices);
1317         }
1318
1319         list_add(&old_devices->list, &fs_uuids);
1320
1321         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1322         seed_devices->opened = 1;
1323         INIT_LIST_HEAD(&seed_devices->devices);
1324         INIT_LIST_HEAD(&seed_devices->alloc_list);
1325         mutex_init(&seed_devices->device_list_mutex);
1326         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1327         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1328         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1329                 device->fs_devices = seed_devices;
1330         }
1331
1332         fs_devices->seeding = 0;
1333         fs_devices->num_devices = 0;
1334         fs_devices->open_devices = 0;
1335         fs_devices->seed = seed_devices;
1336
1337         generate_random_uuid(fs_devices->fsid);
1338         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1339         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1340         super_flags = btrfs_super_flags(disk_super) &
1341                       ~BTRFS_SUPER_FLAG_SEEDING;
1342         btrfs_set_super_flags(disk_super, super_flags);
1343
1344         return 0;
1345 }
1346
1347 /*
1348  * strore the expected generation for seed devices in device items.
1349  */
1350 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1351                                struct btrfs_root *root)
1352 {
1353         struct btrfs_path *path;
1354         struct extent_buffer *leaf;
1355         struct btrfs_dev_item *dev_item;
1356         struct btrfs_device *device;
1357         struct btrfs_key key;
1358         u8 fs_uuid[BTRFS_UUID_SIZE];
1359         u8 dev_uuid[BTRFS_UUID_SIZE];
1360         u64 devid;
1361         int ret;
1362
1363         path = btrfs_alloc_path();
1364         if (!path)
1365                 return -ENOMEM;
1366
1367         root = root->fs_info->chunk_root;
1368         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1369         key.offset = 0;
1370         key.type = BTRFS_DEV_ITEM_KEY;
1371
1372         while (1) {
1373                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1374                 if (ret < 0)
1375                         goto error;
1376
1377                 leaf = path->nodes[0];
1378 next_slot:
1379                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1380                         ret = btrfs_next_leaf(root, path);
1381                         if (ret > 0)
1382                                 break;
1383                         if (ret < 0)
1384                                 goto error;
1385                         leaf = path->nodes[0];
1386                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1387                         btrfs_release_path(root, path);
1388                         continue;
1389                 }
1390
1391                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1392                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1393                     key.type != BTRFS_DEV_ITEM_KEY)
1394                         break;
1395
1396                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1397                                           struct btrfs_dev_item);
1398                 devid = btrfs_device_id(leaf, dev_item);
1399                 read_extent_buffer(leaf, dev_uuid,
1400                                    (unsigned long)btrfs_device_uuid(dev_item),
1401                                    BTRFS_UUID_SIZE);
1402                 read_extent_buffer(leaf, fs_uuid,
1403                                    (unsigned long)btrfs_device_fsid(dev_item),
1404                                    BTRFS_UUID_SIZE);
1405                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1406                 BUG_ON(!device);
1407
1408                 if (device->fs_devices->seeding) {
1409                         btrfs_set_device_generation(leaf, dev_item,
1410                                                     device->generation);
1411                         btrfs_mark_buffer_dirty(leaf);
1412                 }
1413
1414                 path->slots[0]++;
1415                 goto next_slot;
1416         }
1417         ret = 0;
1418 error:
1419         btrfs_free_path(path);
1420         return ret;
1421 }
1422
1423 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1424 {
1425         struct btrfs_trans_handle *trans;
1426         struct btrfs_device *device;
1427         struct block_device *bdev;
1428         struct list_head *devices;
1429         struct super_block *sb = root->fs_info->sb;
1430         u64 total_bytes;
1431         int seeding_dev = 0;
1432         int ret = 0;
1433
1434         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1435                 return -EINVAL;
1436
1437         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1438         if (IS_ERR(bdev))
1439                 return PTR_ERR(bdev);
1440
1441         if (root->fs_info->fs_devices->seeding) {
1442                 seeding_dev = 1;
1443                 down_write(&sb->s_umount);
1444                 mutex_lock(&uuid_mutex);
1445         }
1446
1447         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1448         mutex_lock(&root->fs_info->volume_mutex);
1449
1450         devices = &root->fs_info->fs_devices->devices;
1451         /*
1452          * we have the volume lock, so we don't need the extra
1453          * device list mutex while reading the list here.
1454          */
1455         list_for_each_entry(device, devices, dev_list) {
1456                 if (device->bdev == bdev) {
1457                         ret = -EEXIST;
1458                         goto error;
1459                 }
1460         }
1461
1462         device = kzalloc(sizeof(*device), GFP_NOFS);
1463         if (!device) {
1464                 /* we can safely leave the fs_devices entry around */
1465                 ret = -ENOMEM;
1466                 goto error;
1467         }
1468
1469         device->name = kstrdup(device_path, GFP_NOFS);
1470         if (!device->name) {
1471                 kfree(device);
1472                 ret = -ENOMEM;
1473                 goto error;
1474         }
1475
1476         ret = find_next_devid(root, &device->devid);
1477         if (ret) {
1478                 kfree(device);
1479                 goto error;
1480         }
1481
1482         trans = btrfs_start_transaction(root, 1);
1483         lock_chunks(root);
1484
1485         device->barriers = 1;
1486         device->writeable = 1;
1487         device->work.func = pending_bios_fn;
1488         generate_random_uuid(device->uuid);
1489         spin_lock_init(&device->io_lock);
1490         device->generation = trans->transid;
1491         device->io_width = root->sectorsize;
1492         device->io_align = root->sectorsize;
1493         device->sector_size = root->sectorsize;
1494         device->total_bytes = i_size_read(bdev->bd_inode);
1495         device->disk_total_bytes = device->total_bytes;
1496         device->dev_root = root->fs_info->dev_root;
1497         device->bdev = bdev;
1498         device->in_fs_metadata = 1;
1499         device->mode = 0;
1500         set_blocksize(device->bdev, 4096);
1501
1502         if (seeding_dev) {
1503                 sb->s_flags &= ~MS_RDONLY;
1504                 ret = btrfs_prepare_sprout(trans, root);
1505                 BUG_ON(ret);
1506         }
1507
1508         device->fs_devices = root->fs_info->fs_devices;
1509
1510         /*
1511          * we don't want write_supers to jump in here with our device
1512          * half setup
1513          */
1514         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1515         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1516         list_add(&device->dev_alloc_list,
1517                  &root->fs_info->fs_devices->alloc_list);
1518         root->fs_info->fs_devices->num_devices++;
1519         root->fs_info->fs_devices->open_devices++;
1520         root->fs_info->fs_devices->rw_devices++;
1521         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1522
1523         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1524                 root->fs_info->fs_devices->rotating = 1;
1525
1526         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1527         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1528                                     total_bytes + device->total_bytes);
1529
1530         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1531         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1532                                     total_bytes + 1);
1533         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1534
1535         if (seeding_dev) {
1536                 ret = init_first_rw_device(trans, root, device);
1537                 BUG_ON(ret);
1538                 ret = btrfs_finish_sprout(trans, root);
1539                 BUG_ON(ret);
1540         } else {
1541                 ret = btrfs_add_device(trans, root, device);
1542         }
1543
1544         /*
1545          * we've got more storage, clear any full flags on the space
1546          * infos
1547          */
1548         btrfs_clear_space_info_full(root->fs_info);
1549
1550         unlock_chunks(root);
1551         btrfs_commit_transaction(trans, root);
1552
1553         if (seeding_dev) {
1554                 mutex_unlock(&uuid_mutex);
1555                 up_write(&sb->s_umount);
1556
1557                 ret = btrfs_relocate_sys_chunks(root);
1558                 BUG_ON(ret);
1559         }
1560 out:
1561         mutex_unlock(&root->fs_info->volume_mutex);
1562         return ret;
1563 error:
1564         close_bdev_exclusive(bdev, 0);
1565         if (seeding_dev) {
1566                 mutex_unlock(&uuid_mutex);
1567                 up_write(&sb->s_umount);
1568         }
1569         goto out;
1570 }
1571
1572 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1573                                         struct btrfs_device *device)
1574 {
1575         int ret;
1576         struct btrfs_path *path;
1577         struct btrfs_root *root;
1578         struct btrfs_dev_item *dev_item;
1579         struct extent_buffer *leaf;
1580         struct btrfs_key key;
1581
1582         root = device->dev_root->fs_info->chunk_root;
1583
1584         path = btrfs_alloc_path();
1585         if (!path)
1586                 return -ENOMEM;
1587
1588         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1589         key.type = BTRFS_DEV_ITEM_KEY;
1590         key.offset = device->devid;
1591
1592         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1593         if (ret < 0)
1594                 goto out;
1595
1596         if (ret > 0) {
1597                 ret = -ENOENT;
1598                 goto out;
1599         }
1600
1601         leaf = path->nodes[0];
1602         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1603
1604         btrfs_set_device_id(leaf, dev_item, device->devid);
1605         btrfs_set_device_type(leaf, dev_item, device->type);
1606         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1607         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1608         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1609         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1610         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1611         btrfs_mark_buffer_dirty(leaf);
1612
1613 out:
1614         btrfs_free_path(path);
1615         return ret;
1616 }
1617
1618 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1619                       struct btrfs_device *device, u64 new_size)
1620 {
1621         struct btrfs_super_block *super_copy =
1622                 &device->dev_root->fs_info->super_copy;
1623         u64 old_total = btrfs_super_total_bytes(super_copy);
1624         u64 diff = new_size - device->total_bytes;
1625
1626         if (!device->writeable)
1627                 return -EACCES;
1628         if (new_size <= device->total_bytes)
1629                 return -EINVAL;
1630
1631         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1632         device->fs_devices->total_rw_bytes += diff;
1633
1634         device->total_bytes = new_size;
1635         device->disk_total_bytes = new_size;
1636         btrfs_clear_space_info_full(device->dev_root->fs_info);
1637
1638         return btrfs_update_device(trans, device);
1639 }
1640
1641 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1642                       struct btrfs_device *device, u64 new_size)
1643 {
1644         int ret;
1645         lock_chunks(device->dev_root);
1646         ret = __btrfs_grow_device(trans, device, new_size);
1647         unlock_chunks(device->dev_root);
1648         return ret;
1649 }
1650
1651 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1652                             struct btrfs_root *root,
1653                             u64 chunk_tree, u64 chunk_objectid,
1654                             u64 chunk_offset)
1655 {
1656         int ret;
1657         struct btrfs_path *path;
1658         struct btrfs_key key;
1659
1660         root = root->fs_info->chunk_root;
1661         path = btrfs_alloc_path();
1662         if (!path)
1663                 return -ENOMEM;
1664
1665         key.objectid = chunk_objectid;
1666         key.offset = chunk_offset;
1667         key.type = BTRFS_CHUNK_ITEM_KEY;
1668
1669         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1670         BUG_ON(ret);
1671
1672         ret = btrfs_del_item(trans, root, path);
1673         BUG_ON(ret);
1674
1675         btrfs_free_path(path);
1676         return 0;
1677 }
1678
1679 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1680                         chunk_offset)
1681 {
1682         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1683         struct btrfs_disk_key *disk_key;
1684         struct btrfs_chunk *chunk;
1685         u8 *ptr;
1686         int ret = 0;
1687         u32 num_stripes;
1688         u32 array_size;
1689         u32 len = 0;
1690         u32 cur;
1691         struct btrfs_key key;
1692
1693         array_size = btrfs_super_sys_array_size(super_copy);
1694
1695         ptr = super_copy->sys_chunk_array;
1696         cur = 0;
1697
1698         while (cur < array_size) {
1699                 disk_key = (struct btrfs_disk_key *)ptr;
1700                 btrfs_disk_key_to_cpu(&key, disk_key);
1701
1702                 len = sizeof(*disk_key);
1703
1704                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1705                         chunk = (struct btrfs_chunk *)(ptr + len);
1706                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1707                         len += btrfs_chunk_item_size(num_stripes);
1708                 } else {
1709                         ret = -EIO;
1710                         break;
1711                 }
1712                 if (key.objectid == chunk_objectid &&
1713                     key.offset == chunk_offset) {
1714                         memmove(ptr, ptr + len, array_size - (cur + len));
1715                         array_size -= len;
1716                         btrfs_set_super_sys_array_size(super_copy, array_size);
1717                 } else {
1718                         ptr += len;
1719                         cur += len;
1720                 }
1721         }
1722         return ret;
1723 }
1724
1725 static int btrfs_relocate_chunk(struct btrfs_root *root,
1726                          u64 chunk_tree, u64 chunk_objectid,
1727                          u64 chunk_offset)
1728 {
1729         struct extent_map_tree *em_tree;
1730         struct btrfs_root *extent_root;
1731         struct btrfs_trans_handle *trans;
1732         struct extent_map *em;
1733         struct map_lookup *map;
1734         int ret;
1735         int i;
1736
1737         root = root->fs_info->chunk_root;
1738         extent_root = root->fs_info->extent_root;
1739         em_tree = &root->fs_info->mapping_tree.map_tree;
1740
1741         ret = btrfs_can_relocate(extent_root, chunk_offset);
1742         if (ret)
1743                 return -ENOSPC;
1744
1745         /* step one, relocate all the extents inside this chunk */
1746         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1747         BUG_ON(ret);
1748
1749         trans = btrfs_start_transaction(root, 1);
1750         BUG_ON(!trans);
1751
1752         lock_chunks(root);
1753
1754         /*
1755          * step two, delete the device extents and the
1756          * chunk tree entries
1757          */
1758         read_lock(&em_tree->lock);
1759         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1760         read_unlock(&em_tree->lock);
1761
1762         BUG_ON(em->start > chunk_offset ||
1763                em->start + em->len < chunk_offset);
1764         map = (struct map_lookup *)em->bdev;
1765
1766         for (i = 0; i < map->num_stripes; i++) {
1767                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1768                                             map->stripes[i].physical);
1769                 BUG_ON(ret);
1770
1771                 if (map->stripes[i].dev) {
1772                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1773                         BUG_ON(ret);
1774                 }
1775         }
1776         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1777                                chunk_offset);
1778
1779         BUG_ON(ret);
1780
1781         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1782                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1783                 BUG_ON(ret);
1784         }
1785
1786         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1787         BUG_ON(ret);
1788
1789         write_lock(&em_tree->lock);
1790         remove_extent_mapping(em_tree, em);
1791         write_unlock(&em_tree->lock);
1792
1793         kfree(map);
1794         em->bdev = NULL;
1795
1796         /* once for the tree */
1797         free_extent_map(em);
1798         /* once for us */
1799         free_extent_map(em);
1800
1801         unlock_chunks(root);
1802         btrfs_end_transaction(trans, root);
1803         return 0;
1804 }
1805
1806 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1807 {
1808         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1809         struct btrfs_path *path;
1810         struct extent_buffer *leaf;
1811         struct btrfs_chunk *chunk;
1812         struct btrfs_key key;
1813         struct btrfs_key found_key;
1814         u64 chunk_tree = chunk_root->root_key.objectid;
1815         u64 chunk_type;
1816         bool retried = false;
1817         int failed = 0;
1818         int ret;
1819
1820         path = btrfs_alloc_path();
1821         if (!path)
1822                 return -ENOMEM;
1823
1824 again:
1825         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1826         key.offset = (u64)-1;
1827         key.type = BTRFS_CHUNK_ITEM_KEY;
1828
1829         while (1) {
1830                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1831                 if (ret < 0)
1832                         goto error;
1833                 BUG_ON(ret == 0);
1834
1835                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1836                                           key.type);
1837                 if (ret < 0)
1838                         goto error;
1839                 if (ret > 0)
1840                         break;
1841
1842                 leaf = path->nodes[0];
1843                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1844
1845                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1846                                        struct btrfs_chunk);
1847                 chunk_type = btrfs_chunk_type(leaf, chunk);
1848                 btrfs_release_path(chunk_root, path);
1849
1850                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1851                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1852                                                    found_key.objectid,
1853                                                    found_key.offset);
1854                         if (ret == -ENOSPC)
1855                                 failed++;
1856                         else if (ret)
1857                                 BUG();
1858                 }
1859
1860                 if (found_key.offset == 0)
1861                         break;
1862                 key.offset = found_key.offset - 1;
1863         }
1864         ret = 0;
1865         if (failed && !retried) {
1866                 failed = 0;
1867                 retried = true;
1868                 goto again;
1869         } else if (failed && retried) {
1870                 WARN_ON(1);
1871                 ret = -ENOSPC;
1872         }
1873 error:
1874         btrfs_free_path(path);
1875         return ret;
1876 }
1877
1878 static u64 div_factor(u64 num, int factor)
1879 {
1880         if (factor == 10)
1881                 return num;
1882         num *= factor;
1883         do_div(num, 10);
1884         return num;
1885 }
1886
1887 int btrfs_balance(struct btrfs_root *dev_root)
1888 {
1889         int ret;
1890         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1891         struct btrfs_device *device;
1892         u64 old_size;
1893         u64 size_to_free;
1894         struct btrfs_path *path;
1895         struct btrfs_key key;
1896         struct btrfs_chunk *chunk;
1897         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1898         struct btrfs_trans_handle *trans;
1899         struct btrfs_key found_key;
1900
1901         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1902                 return -EROFS;
1903
1904         if (!capable(CAP_SYS_ADMIN))
1905                 return -EPERM;
1906
1907         mutex_lock(&dev_root->fs_info->volume_mutex);
1908         dev_root = dev_root->fs_info->dev_root;
1909
1910         /* step one make some room on all the devices */
1911         list_for_each_entry(device, devices, dev_list) {
1912                 old_size = device->total_bytes;
1913                 size_to_free = div_factor(old_size, 1);
1914                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1915                 if (!device->writeable ||
1916                     device->total_bytes - device->bytes_used > size_to_free)
1917                         continue;
1918
1919                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1920                 if (ret == -ENOSPC)
1921                         break;
1922                 BUG_ON(ret);
1923
1924                 trans = btrfs_start_transaction(dev_root, 1);
1925                 BUG_ON(!trans);
1926
1927                 ret = btrfs_grow_device(trans, device, old_size);
1928                 BUG_ON(ret);
1929
1930                 btrfs_end_transaction(trans, dev_root);
1931         }
1932
1933         /* step two, relocate all the chunks */
1934         path = btrfs_alloc_path();
1935         BUG_ON(!path);
1936
1937         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1938         key.offset = (u64)-1;
1939         key.type = BTRFS_CHUNK_ITEM_KEY;
1940
1941         while (1) {
1942                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1943                 if (ret < 0)
1944                         goto error;
1945
1946                 /*
1947                  * this shouldn't happen, it means the last relocate
1948                  * failed
1949                  */
1950                 if (ret == 0)
1951                         break;
1952
1953                 ret = btrfs_previous_item(chunk_root, path, 0,
1954                                           BTRFS_CHUNK_ITEM_KEY);
1955                 if (ret)
1956                         break;
1957
1958                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1959                                       path->slots[0]);
1960                 if (found_key.objectid != key.objectid)
1961                         break;
1962
1963                 chunk = btrfs_item_ptr(path->nodes[0],
1964                                        path->slots[0],
1965                                        struct btrfs_chunk);
1966                 /* chunk zero is special */
1967                 if (found_key.offset == 0)
1968                         break;
1969
1970                 btrfs_release_path(chunk_root, path);
1971                 ret = btrfs_relocate_chunk(chunk_root,
1972                                            chunk_root->root_key.objectid,
1973                                            found_key.objectid,
1974                                            found_key.offset);
1975                 BUG_ON(ret && ret != -ENOSPC);
1976                 key.offset = found_key.offset - 1;
1977         }
1978         ret = 0;
1979 error:
1980         btrfs_free_path(path);
1981         mutex_unlock(&dev_root->fs_info->volume_mutex);
1982         return ret;
1983 }
1984
1985 /*
1986  * shrinking a device means finding all of the device extents past
1987  * the new size, and then following the back refs to the chunks.
1988  * The chunk relocation code actually frees the device extent
1989  */
1990 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1991 {
1992         struct btrfs_trans_handle *trans;
1993         struct btrfs_root *root = device->dev_root;
1994         struct btrfs_dev_extent *dev_extent = NULL;
1995         struct btrfs_path *path;
1996         u64 length;
1997         u64 chunk_tree;
1998         u64 chunk_objectid;
1999         u64 chunk_offset;
2000         int ret;
2001         int slot;
2002         int failed = 0;
2003         bool retried = false;
2004         struct extent_buffer *l;
2005         struct btrfs_key key;
2006         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2007         u64 old_total = btrfs_super_total_bytes(super_copy);
2008         u64 old_size = device->total_bytes;
2009         u64 diff = device->total_bytes - new_size;
2010
2011         if (new_size >= device->total_bytes)
2012                 return -EINVAL;
2013
2014         path = btrfs_alloc_path();
2015         if (!path)
2016                 return -ENOMEM;
2017
2018         path->reada = 2;
2019
2020         lock_chunks(root);
2021
2022         device->total_bytes = new_size;
2023         if (device->writeable)
2024                 device->fs_devices->total_rw_bytes -= diff;
2025         unlock_chunks(root);
2026
2027 again:
2028         key.objectid = device->devid;
2029         key.offset = (u64)-1;
2030         key.type = BTRFS_DEV_EXTENT_KEY;
2031
2032         while (1) {
2033                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2034                 if (ret < 0)
2035                         goto done;
2036
2037                 ret = btrfs_previous_item(root, path, 0, key.type);
2038                 if (ret < 0)
2039                         goto done;
2040                 if (ret) {
2041                         ret = 0;
2042                         btrfs_release_path(root, path);
2043                         break;
2044                 }
2045
2046                 l = path->nodes[0];
2047                 slot = path->slots[0];
2048                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2049
2050                 if (key.objectid != device->devid) {
2051                         btrfs_release_path(root, path);
2052                         break;
2053                 }
2054
2055                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2056                 length = btrfs_dev_extent_length(l, dev_extent);
2057
2058                 if (key.offset + length <= new_size) {
2059                         btrfs_release_path(root, path);
2060                         break;
2061                 }
2062
2063                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2064                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2065                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2066                 btrfs_release_path(root, path);
2067
2068                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2069                                            chunk_offset);
2070                 if (ret && ret != -ENOSPC)
2071                         goto done;
2072                 if (ret == -ENOSPC)
2073                         failed++;
2074                 key.offset -= 1;
2075         }
2076
2077         if (failed && !retried) {
2078                 failed = 0;
2079                 retried = true;
2080                 goto again;
2081         } else if (failed && retried) {
2082                 ret = -ENOSPC;
2083                 lock_chunks(root);
2084
2085                 device->total_bytes = old_size;
2086                 if (device->writeable)
2087                         device->fs_devices->total_rw_bytes += diff;
2088                 unlock_chunks(root);
2089                 goto done;
2090         }
2091
2092         /* Shrinking succeeded, else we would be at "done". */
2093         trans = btrfs_start_transaction(root, 1);
2094         if (!trans) {
2095                 ret = -ENOMEM;
2096                 goto done;
2097         }
2098         lock_chunks(root);
2099
2100         device->disk_total_bytes = new_size;
2101         /* Now btrfs_update_device() will change the on-disk size. */
2102         ret = btrfs_update_device(trans, device);
2103         if (ret) {
2104                 unlock_chunks(root);
2105                 btrfs_end_transaction(trans, root);
2106                 goto done;
2107         }
2108         WARN_ON(diff > old_total);
2109         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2110         unlock_chunks(root);
2111         btrfs_end_transaction(trans, root);
2112 done:
2113         btrfs_free_path(path);
2114         return ret;
2115 }
2116
2117 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2118                            struct btrfs_root *root,
2119                            struct btrfs_key *key,
2120                            struct btrfs_chunk *chunk, int item_size)
2121 {
2122         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2123         struct btrfs_disk_key disk_key;
2124         u32 array_size;
2125         u8 *ptr;
2126
2127         array_size = btrfs_super_sys_array_size(super_copy);
2128         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2129                 return -EFBIG;
2130
2131         ptr = super_copy->sys_chunk_array + array_size;
2132         btrfs_cpu_key_to_disk(&disk_key, key);
2133         memcpy(ptr, &disk_key, sizeof(disk_key));
2134         ptr += sizeof(disk_key);
2135         memcpy(ptr, chunk, item_size);
2136         item_size += sizeof(disk_key);
2137         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2138         return 0;
2139 }
2140
2141 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2142                                         int num_stripes, int sub_stripes)
2143 {
2144         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2145                 return calc_size;
2146         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2147                 return calc_size * (num_stripes / sub_stripes);
2148         else
2149                 return calc_size * num_stripes;
2150 }
2151
2152 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2153                                struct btrfs_root *extent_root,
2154                                struct map_lookup **map_ret,
2155                                u64 *num_bytes, u64 *stripe_size,
2156                                u64 start, u64 type)
2157 {
2158         struct btrfs_fs_info *info = extent_root->fs_info;
2159         struct btrfs_device *device = NULL;
2160         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2161         struct list_head *cur;
2162         struct map_lookup *map = NULL;
2163         struct extent_map_tree *em_tree;
2164         struct extent_map *em;
2165         struct list_head private_devs;
2166         int min_stripe_size = 1 * 1024 * 1024;
2167         u64 calc_size = 1024 * 1024 * 1024;
2168         u64 max_chunk_size = calc_size;
2169         u64 min_free;
2170         u64 avail;
2171         u64 max_avail = 0;
2172         u64 dev_offset;
2173         int num_stripes = 1;
2174         int min_stripes = 1;
2175         int sub_stripes = 0;
2176         int looped = 0;
2177         int ret;
2178         int index;
2179         int stripe_len = 64 * 1024;
2180
2181         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2182             (type & BTRFS_BLOCK_GROUP_DUP)) {
2183                 WARN_ON(1);
2184                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2185         }
2186         if (list_empty(&fs_devices->alloc_list))
2187                 return -ENOSPC;
2188
2189         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2190                 num_stripes = fs_devices->rw_devices;
2191                 min_stripes = 2;
2192         }
2193         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2194                 num_stripes = 2;
2195                 min_stripes = 2;
2196         }
2197         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2198                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2199                 if (num_stripes < 2)
2200                         return -ENOSPC;
2201                 min_stripes = 2;
2202         }
2203         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2204                 num_stripes = fs_devices->rw_devices;
2205                 if (num_stripes < 4)
2206                         return -ENOSPC;
2207                 num_stripes &= ~(u32)1;
2208                 sub_stripes = 2;
2209                 min_stripes = 4;
2210         }
2211
2212         if (type & BTRFS_BLOCK_GROUP_DATA) {
2213                 max_chunk_size = 10 * calc_size;
2214                 min_stripe_size = 64 * 1024 * 1024;
2215         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2216                 max_chunk_size = 256 * 1024 * 1024;
2217                 min_stripe_size = 32 * 1024 * 1024;
2218         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2219                 calc_size = 8 * 1024 * 1024;
2220                 max_chunk_size = calc_size * 2;
2221                 min_stripe_size = 1 * 1024 * 1024;
2222         }
2223
2224         /* we don't want a chunk larger than 10% of writeable space */
2225         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2226                              max_chunk_size);
2227
2228 again:
2229         max_avail = 0;
2230         if (!map || map->num_stripes != num_stripes) {
2231                 kfree(map);
2232                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2233                 if (!map)
2234                         return -ENOMEM;
2235                 map->num_stripes = num_stripes;
2236         }
2237
2238         if (calc_size * num_stripes > max_chunk_size) {
2239                 calc_size = max_chunk_size;
2240                 do_div(calc_size, num_stripes);
2241                 do_div(calc_size, stripe_len);
2242                 calc_size *= stripe_len;
2243         }
2244         /* we don't want tiny stripes */
2245         calc_size = max_t(u64, min_stripe_size, calc_size);
2246
2247         do_div(calc_size, stripe_len);
2248         calc_size *= stripe_len;
2249
2250         cur = fs_devices->alloc_list.next;
2251         index = 0;
2252
2253         if (type & BTRFS_BLOCK_GROUP_DUP)
2254                 min_free = calc_size * 2;
2255         else
2256                 min_free = calc_size;
2257
2258         /*
2259          * we add 1MB because we never use the first 1MB of the device, unless
2260          * we've looped, then we are likely allocating the maximum amount of
2261          * space left already
2262          */
2263         if (!looped)
2264                 min_free += 1024 * 1024;
2265
2266         INIT_LIST_HEAD(&private_devs);
2267         while (index < num_stripes) {
2268                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2269                 BUG_ON(!device->writeable);
2270                 if (device->total_bytes > device->bytes_used)
2271                         avail = device->total_bytes - device->bytes_used;
2272                 else
2273                         avail = 0;
2274                 cur = cur->next;
2275
2276                 if (device->in_fs_metadata && avail >= min_free) {
2277                         ret = find_free_dev_extent(trans, device,
2278                                                    min_free, &dev_offset,
2279                                                    &max_avail);
2280                         if (ret == 0) {
2281                                 list_move_tail(&device->dev_alloc_list,
2282                                                &private_devs);
2283                                 map->stripes[index].dev = device;
2284                                 map->stripes[index].physical = dev_offset;
2285                                 index++;
2286                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2287                                         map->stripes[index].dev = device;
2288                                         map->stripes[index].physical =
2289                                                 dev_offset + calc_size;
2290                                         index++;
2291                                 }
2292                         }
2293                 } else if (device->in_fs_metadata && avail > max_avail)
2294                         max_avail = avail;
2295                 if (cur == &fs_devices->alloc_list)
2296                         break;
2297         }
2298         list_splice(&private_devs, &fs_devices->alloc_list);
2299         if (index < num_stripes) {
2300                 if (index >= min_stripes) {
2301                         num_stripes = index;
2302                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2303                                 num_stripes /= sub_stripes;
2304                                 num_stripes *= sub_stripes;
2305                         }
2306                         looped = 1;
2307                         goto again;
2308                 }
2309                 if (!looped && max_avail > 0) {
2310                         looped = 1;
2311                         calc_size = max_avail;
2312                         goto again;
2313                 }
2314                 kfree(map);
2315                 return -ENOSPC;
2316         }
2317         map->sector_size = extent_root->sectorsize;
2318         map->stripe_len = stripe_len;
2319         map->io_align = stripe_len;
2320         map->io_width = stripe_len;
2321         map->type = type;
2322         map->num_stripes = num_stripes;
2323         map->sub_stripes = sub_stripes;
2324
2325         *map_ret = map;
2326         *stripe_size = calc_size;
2327         *num_bytes = chunk_bytes_by_type(type, calc_size,
2328                                          num_stripes, sub_stripes);
2329
2330         em = alloc_extent_map(GFP_NOFS);
2331         if (!em) {
2332                 kfree(map);
2333                 return -ENOMEM;
2334         }
2335         em->bdev = (struct block_device *)map;
2336         em->start = start;
2337         em->len = *num_bytes;
2338         em->block_start = 0;
2339         em->block_len = em->len;
2340
2341         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2342         write_lock(&em_tree->lock);
2343         ret = add_extent_mapping(em_tree, em);
2344         write_unlock(&em_tree->lock);
2345         BUG_ON(ret);
2346         free_extent_map(em);
2347
2348         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2349                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2350                                      start, *num_bytes);
2351         BUG_ON(ret);
2352
2353         index = 0;
2354         while (index < map->num_stripes) {
2355                 device = map->stripes[index].dev;
2356                 dev_offset = map->stripes[index].physical;
2357
2358                 ret = btrfs_alloc_dev_extent(trans, device,
2359                                 info->chunk_root->root_key.objectid,
2360                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2361                                 start, dev_offset, calc_size);
2362                 BUG_ON(ret);
2363                 index++;
2364         }
2365
2366         return 0;
2367 }
2368
2369 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2370                                 struct btrfs_root *extent_root,
2371                                 struct map_lookup *map, u64 chunk_offset,
2372                                 u64 chunk_size, u64 stripe_size)
2373 {
2374         u64 dev_offset;
2375         struct btrfs_key key;
2376         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2377         struct btrfs_device *device;
2378         struct btrfs_chunk *chunk;
2379         struct btrfs_stripe *stripe;
2380         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2381         int index = 0;
2382         int ret;
2383
2384         chunk = kzalloc(item_size, GFP_NOFS);
2385         if (!chunk)
2386                 return -ENOMEM;
2387
2388         index = 0;
2389         while (index < map->num_stripes) {
2390                 device = map->stripes[index].dev;
2391                 device->bytes_used += stripe_size;
2392                 ret = btrfs_update_device(trans, device);
2393                 BUG_ON(ret);
2394                 index++;
2395         }
2396
2397         index = 0;
2398         stripe = &chunk->stripe;
2399         while (index < map->num_stripes) {
2400                 device = map->stripes[index].dev;
2401                 dev_offset = map->stripes[index].physical;
2402
2403                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2404                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2405                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2406                 stripe++;
2407                 index++;
2408         }
2409
2410         btrfs_set_stack_chunk_length(chunk, chunk_size);
2411         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2412         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2413         btrfs_set_stack_chunk_type(chunk, map->type);
2414         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2415         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2416         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2417         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2418         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2419
2420         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2421         key.type = BTRFS_CHUNK_ITEM_KEY;
2422         key.offset = chunk_offset;
2423
2424         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2425         BUG_ON(ret);
2426
2427         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2428                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2429                                              item_size);
2430                 BUG_ON(ret);
2431         }
2432         kfree(chunk);
2433         return 0;
2434 }
2435
2436 /*
2437  * Chunk allocation falls into two parts. The first part does works
2438  * that make the new allocated chunk useable, but not do any operation
2439  * that modifies the chunk tree. The second part does the works that
2440  * require modifying the chunk tree. This division is important for the
2441  * bootstrap process of adding storage to a seed btrfs.
2442  */
2443 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2444                       struct btrfs_root *extent_root, u64 type)
2445 {
2446         u64 chunk_offset;
2447         u64 chunk_size;
2448         u64 stripe_size;
2449         struct map_lookup *map;
2450         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2451         int ret;
2452
2453         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2454                               &chunk_offset);
2455         if (ret)
2456                 return ret;
2457
2458         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2459                                   &stripe_size, chunk_offset, type);
2460         if (ret)
2461                 return ret;
2462
2463         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2464                                    chunk_size, stripe_size);
2465         BUG_ON(ret);
2466         return 0;
2467 }
2468
2469 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2470                                          struct btrfs_root *root,
2471                                          struct btrfs_device *device)
2472 {
2473         u64 chunk_offset;
2474         u64 sys_chunk_offset;
2475         u64 chunk_size;
2476         u64 sys_chunk_size;
2477         u64 stripe_size;
2478         u64 sys_stripe_size;
2479         u64 alloc_profile;
2480         struct map_lookup *map;
2481         struct map_lookup *sys_map;
2482         struct btrfs_fs_info *fs_info = root->fs_info;
2483         struct btrfs_root *extent_root = fs_info->extent_root;
2484         int ret;
2485
2486         ret = find_next_chunk(fs_info->chunk_root,
2487                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2488         BUG_ON(ret);
2489
2490         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2491                         (fs_info->metadata_alloc_profile &
2492                          fs_info->avail_metadata_alloc_bits);
2493         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2494
2495         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2496                                   &stripe_size, chunk_offset, alloc_profile);
2497         BUG_ON(ret);
2498
2499         sys_chunk_offset = chunk_offset + chunk_size;
2500
2501         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2502                         (fs_info->system_alloc_profile &
2503                          fs_info->avail_system_alloc_bits);
2504         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2505
2506         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2507                                   &sys_chunk_size, &sys_stripe_size,
2508                                   sys_chunk_offset, alloc_profile);
2509         BUG_ON(ret);
2510
2511         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2512         BUG_ON(ret);
2513
2514         /*
2515          * Modifying chunk tree needs allocating new blocks from both
2516          * system block group and metadata block group. So we only can
2517          * do operations require modifying the chunk tree after both
2518          * block groups were created.
2519          */
2520         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2521                                    chunk_size, stripe_size);
2522         BUG_ON(ret);
2523
2524         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2525                                    sys_chunk_offset, sys_chunk_size,
2526                                    sys_stripe_size);
2527         BUG_ON(ret);
2528         return 0;
2529 }
2530
2531 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2532 {
2533         struct extent_map *em;
2534         struct map_lookup *map;
2535         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2536         int readonly = 0;
2537         int i;
2538
2539         read_lock(&map_tree->map_tree.lock);
2540         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2541         read_unlock(&map_tree->map_tree.lock);
2542         if (!em)
2543                 return 1;
2544
2545         if (btrfs_test_opt(root, DEGRADED)) {
2546                 free_extent_map(em);
2547                 return 0;
2548         }
2549
2550         map = (struct map_lookup *)em->bdev;
2551         for (i = 0; i < map->num_stripes; i++) {
2552                 if (!map->stripes[i].dev->writeable) {
2553                         readonly = 1;
2554                         break;
2555                 }
2556         }
2557         free_extent_map(em);
2558         return readonly;
2559 }
2560
2561 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2562 {
2563         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2564 }
2565
2566 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2567 {
2568         struct extent_map *em;
2569
2570         while (1) {
2571                 write_lock(&tree->map_tree.lock);
2572                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2573                 if (em)
2574                         remove_extent_mapping(&tree->map_tree, em);
2575                 write_unlock(&tree->map_tree.lock);
2576                 if (!em)
2577                         break;
2578                 kfree(em->bdev);
2579                 /* once for us */
2580                 free_extent_map(em);
2581                 /* once for the tree */
2582                 free_extent_map(em);
2583         }
2584 }
2585
2586 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2587 {
2588         struct extent_map *em;
2589         struct map_lookup *map;
2590         struct extent_map_tree *em_tree = &map_tree->map_tree;
2591         int ret;
2592
2593         read_lock(&em_tree->lock);
2594         em = lookup_extent_mapping(em_tree, logical, len);
2595         read_unlock(&em_tree->lock);
2596         BUG_ON(!em);
2597
2598         BUG_ON(em->start > logical || em->start + em->len < logical);
2599         map = (struct map_lookup *)em->bdev;
2600         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2601                 ret = map->num_stripes;
2602         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2603                 ret = map->sub_stripes;
2604         else
2605                 ret = 1;
2606         free_extent_map(em);
2607         return ret;
2608 }
2609
2610 static int find_live_mirror(struct map_lookup *map, int first, int num,
2611                             int optimal)
2612 {
2613         int i;
2614         if (map->stripes[optimal].dev->bdev)
2615                 return optimal;
2616         for (i = first; i < first + num; i++) {
2617                 if (map->stripes[i].dev->bdev)
2618                         return i;
2619         }
2620         /* we couldn't find one that doesn't fail.  Just return something
2621          * and the io error handling code will clean up eventually
2622          */
2623         return optimal;
2624 }
2625
2626 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2627                              u64 logical, u64 *length,
2628                              struct btrfs_multi_bio **multi_ret,
2629                              int mirror_num, struct page *unplug_page)
2630 {
2631         struct extent_map *em;
2632         struct map_lookup *map;
2633         struct extent_map_tree *em_tree = &map_tree->map_tree;
2634         u64 offset;
2635         u64 stripe_offset;
2636         u64 stripe_nr;
2637         int stripes_allocated = 8;
2638         int stripes_required = 1;
2639         int stripe_index;
2640         int i;
2641         int num_stripes;
2642         int max_errors = 0;
2643         struct btrfs_multi_bio *multi = NULL;
2644
2645         if (multi_ret && !(rw & (1 << BIO_RW)))
2646                 stripes_allocated = 1;
2647 again:
2648         if (multi_ret) {
2649                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2650                                 GFP_NOFS);
2651                 if (!multi)
2652                         return -ENOMEM;
2653
2654                 atomic_set(&multi->error, 0);
2655         }
2656
2657         read_lock(&em_tree->lock);
2658         em = lookup_extent_mapping(em_tree, logical, *length);
2659         read_unlock(&em_tree->lock);
2660
2661         if (!em && unplug_page) {
2662                 kfree(multi);
2663                 return 0;
2664         }
2665
2666         if (!em) {
2667                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2668                        (unsigned long long)logical,
2669                        (unsigned long long)*length);
2670                 BUG();
2671         }
2672
2673         BUG_ON(em->start > logical || em->start + em->len < logical);
2674         map = (struct map_lookup *)em->bdev;
2675         offset = logical - em->start;
2676
2677         if (mirror_num > map->num_stripes)
2678                 mirror_num = 0;
2679
2680         /* if our multi bio struct is too small, back off and try again */
2681         if (rw & (1 << BIO_RW)) {
2682                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2683                                  BTRFS_BLOCK_GROUP_DUP)) {
2684                         stripes_required = map->num_stripes;
2685                         max_errors = 1;
2686                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2687                         stripes_required = map->sub_stripes;
2688                         max_errors = 1;
2689                 }
2690         }
2691         if (multi_ret && (rw & (1 << BIO_RW)) &&
2692             stripes_allocated < stripes_required) {
2693                 stripes_allocated = map->num_stripes;
2694                 free_extent_map(em);
2695                 kfree(multi);
2696                 goto again;
2697         }
2698         stripe_nr = offset;
2699         /*
2700          * stripe_nr counts the total number of stripes we have to stride
2701          * to get to this block
2702          */
2703         do_div(stripe_nr, map->stripe_len);
2704
2705         stripe_offset = stripe_nr * map->stripe_len;
2706         BUG_ON(offset < stripe_offset);
2707
2708         /* stripe_offset is the offset of this block in its stripe*/
2709         stripe_offset = offset - stripe_offset;
2710
2711         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2712                          BTRFS_BLOCK_GROUP_RAID10 |
2713                          BTRFS_BLOCK_GROUP_DUP)) {
2714                 /* we limit the length of each bio to what fits in a stripe */
2715                 *length = min_t(u64, em->len - offset,
2716                               map->stripe_len - stripe_offset);
2717         } else {
2718                 *length = em->len - offset;
2719         }
2720
2721         if (!multi_ret && !unplug_page)
2722                 goto out;
2723
2724         num_stripes = 1;
2725         stripe_index = 0;
2726         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2727                 if (unplug_page || (rw & (1 << BIO_RW)))
2728                         num_stripes = map->num_stripes;
2729                 else if (mirror_num)
2730                         stripe_index = mirror_num - 1;
2731                 else {
2732                         stripe_index = find_live_mirror(map, 0,
2733                                             map->num_stripes,
2734                                             current->pid % map->num_stripes);
2735                 }
2736
2737         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2738                 if (rw & (1 << BIO_RW))
2739                         num_stripes = map->num_stripes;
2740                 else if (mirror_num)
2741                         stripe_index = mirror_num - 1;
2742
2743         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2744                 int factor = map->num_stripes / map->sub_stripes;
2745
2746                 stripe_index = do_div(stripe_nr, factor);
2747                 stripe_index *= map->sub_stripes;
2748
2749                 if (unplug_page || (rw & (1 << BIO_RW)))
2750                         num_stripes = map->sub_stripes;
2751                 else if (mirror_num)
2752                         stripe_index += mirror_num - 1;
2753                 else {
2754                         stripe_index = find_live_mirror(map, stripe_index,
2755                                               map->sub_stripes, stripe_index +
2756                                               current->pid % map->sub_stripes);
2757                 }
2758         } else {
2759                 /*
2760                  * after this do_div call, stripe_nr is the number of stripes
2761                  * on this device we have to walk to find the data, and
2762                  * stripe_index is the number of our device in the stripe array
2763                  */
2764                 stripe_index = do_div(stripe_nr, map->num_stripes);
2765         }
2766         BUG_ON(stripe_index >= map->num_stripes);
2767
2768         for (i = 0; i < num_stripes; i++) {
2769                 if (unplug_page) {
2770                         struct btrfs_device *device;
2771                         struct backing_dev_info *bdi;
2772
2773                         device = map->stripes[stripe_index].dev;
2774                         if (device->bdev) {
2775                                 bdi = blk_get_backing_dev_info(device->bdev);
2776                                 if (bdi->unplug_io_fn)
2777                                         bdi->unplug_io_fn(bdi, unplug_page);
2778                         }
2779                 } else {
2780                         multi->stripes[i].physical =
2781                                 map->stripes[stripe_index].physical +
2782                                 stripe_offset + stripe_nr * map->stripe_len;
2783                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2784                 }
2785                 stripe_index++;
2786         }
2787         if (multi_ret) {
2788                 *multi_ret = multi;
2789                 multi->num_stripes = num_stripes;
2790                 multi->max_errors = max_errors;
2791         }
2792 out:
2793         free_extent_map(em);
2794         return 0;
2795 }
2796
2797 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2798                       u64 logical, u64 *length,
2799                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2800 {
2801         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2802                                  mirror_num, NULL);
2803 }
2804
2805 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2806                      u64 chunk_start, u64 physical, u64 devid,
2807                      u64 **logical, int *naddrs, int *stripe_len)
2808 {
2809         struct extent_map_tree *em_tree = &map_tree->map_tree;
2810         struct extent_map *em;
2811         struct map_lookup *map;
2812         u64 *buf;
2813         u64 bytenr;
2814         u64 length;
2815         u64 stripe_nr;
2816         int i, j, nr = 0;
2817
2818         read_lock(&em_tree->lock);
2819         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2820         read_unlock(&em_tree->lock);
2821
2822         BUG_ON(!em || em->start != chunk_start);
2823         map = (struct map_lookup *)em->bdev;
2824
2825         length = em->len;
2826         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2827                 do_div(length, map->num_stripes / map->sub_stripes);
2828         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2829                 do_div(length, map->num_stripes);
2830
2831         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2832         BUG_ON(!buf);
2833
2834         for (i = 0; i < map->num_stripes; i++) {
2835                 if (devid && map->stripes[i].dev->devid != devid)
2836                         continue;
2837                 if (map->stripes[i].physical > physical ||
2838                     map->stripes[i].physical + length <= physical)
2839                         continue;
2840
2841                 stripe_nr = physical - map->stripes[i].physical;
2842                 do_div(stripe_nr, map->stripe_len);
2843
2844                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2845                         stripe_nr = stripe_nr * map->num_stripes + i;
2846                         do_div(stripe_nr, map->sub_stripes);
2847                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2848                         stripe_nr = stripe_nr * map->num_stripes + i;
2849                 }
2850                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2851                 WARN_ON(nr >= map->num_stripes);
2852                 for (j = 0; j < nr; j++) {
2853                         if (buf[j] == bytenr)
2854                                 break;
2855                 }
2856                 if (j == nr) {
2857                         WARN_ON(nr >= map->num_stripes);
2858                         buf[nr++] = bytenr;
2859                 }
2860         }
2861
2862         *logical = buf;
2863         *naddrs = nr;
2864         *stripe_len = map->stripe_len;
2865
2866         free_extent_map(em);
2867         return 0;
2868 }
2869
2870 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2871                       u64 logical, struct page *page)
2872 {
2873         u64 length = PAGE_CACHE_SIZE;
2874         return __btrfs_map_block(map_tree, READ, logical, &length,
2875                                  NULL, 0, page);
2876 }
2877
2878 static void end_bio_multi_stripe(struct bio *bio, int err)
2879 {
2880         struct btrfs_multi_bio *multi = bio->bi_private;
2881         int is_orig_bio = 0;
2882
2883         if (err)
2884                 atomic_inc(&multi->error);
2885
2886         if (bio == multi->orig_bio)
2887                 is_orig_bio = 1;
2888
2889         if (atomic_dec_and_test(&multi->stripes_pending)) {
2890                 if (!is_orig_bio) {
2891                         bio_put(bio);
2892                         bio = multi->orig_bio;
2893                 }
2894                 bio->bi_private = multi->private;
2895                 bio->bi_end_io = multi->end_io;
2896                 /* only send an error to the higher layers if it is
2897                  * beyond the tolerance of the multi-bio
2898                  */
2899                 if (atomic_read(&multi->error) > multi->max_errors) {
2900                         err = -EIO;
2901                 } else if (err) {
2902                         /*
2903                          * this bio is actually up to date, we didn't
2904                          * go over the max number of errors
2905                          */
2906                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2907                         err = 0;
2908                 }
2909                 kfree(multi);
2910
2911                 bio_endio(bio, err);
2912         } else if (!is_orig_bio) {
2913                 bio_put(bio);
2914         }
2915 }
2916
2917 struct async_sched {
2918         struct bio *bio;
2919         int rw;
2920         struct btrfs_fs_info *info;
2921         struct btrfs_work work;
2922 };
2923
2924 /*
2925  * see run_scheduled_bios for a description of why bios are collected for
2926  * async submit.
2927  *
2928  * This will add one bio to the pending list for a device and make sure
2929  * the work struct is scheduled.
2930  */
2931 static noinline int schedule_bio(struct btrfs_root *root,
2932                                  struct btrfs_device *device,
2933                                  int rw, struct bio *bio)
2934 {
2935         int should_queue = 1;
2936         struct btrfs_pending_bios *pending_bios;
2937
2938         /* don't bother with additional async steps for reads, right now */
2939         if (!(rw & (1 << BIO_RW))) {
2940                 bio_get(bio);
2941                 submit_bio(rw, bio);
2942                 bio_put(bio);
2943                 return 0;
2944         }
2945
2946         /*
2947          * nr_async_bios allows us to reliably return congestion to the
2948          * higher layers.  Otherwise, the async bio makes it appear we have
2949          * made progress against dirty pages when we've really just put it
2950          * on a queue for later
2951          */
2952         atomic_inc(&root->fs_info->nr_async_bios);
2953         WARN_ON(bio->bi_next);
2954         bio->bi_next = NULL;
2955         bio->bi_rw |= rw;
2956
2957         spin_lock(&device->io_lock);
2958         if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
2959                 pending_bios = &device->pending_sync_bios;
2960         else
2961                 pending_bios = &device->pending_bios;
2962
2963         if (pending_bios->tail)
2964                 pending_bios->tail->bi_next = bio;
2965
2966         pending_bios->tail = bio;
2967         if (!pending_bios->head)
2968                 pending_bios->head = bio;
2969         if (device->running_pending)
2970                 should_queue = 0;
2971
2972         spin_unlock(&device->io_lock);
2973
2974         if (should_queue)
2975                 btrfs_queue_worker(&root->fs_info->submit_workers,
2976                                    &device->work);
2977         return 0;
2978 }
2979
2980 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2981                   int mirror_num, int async_submit)
2982 {
2983         struct btrfs_mapping_tree *map_tree;
2984         struct btrfs_device *dev;
2985         struct bio *first_bio = bio;
2986         u64 logical = (u64)bio->bi_sector << 9;
2987         u64 length = 0;
2988         u64 map_length;
2989         struct btrfs_multi_bio *multi = NULL;
2990         int ret;
2991         int dev_nr = 0;
2992         int total_devs = 1;
2993
2994         length = bio->bi_size;
2995         map_tree = &root->fs_info->mapping_tree;
2996         map_length = length;
2997
2998         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2999                               mirror_num);
3000         BUG_ON(ret);
3001
3002         total_devs = multi->num_stripes;
3003         if (map_length < length) {
3004                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3005                        "len %llu\n", (unsigned long long)logical,
3006                        (unsigned long long)length,
3007                        (unsigned long long)map_length);
3008                 BUG();
3009         }
3010         multi->end_io = first_bio->bi_end_io;
3011         multi->private = first_bio->bi_private;
3012         multi->orig_bio = first_bio;
3013         atomic_set(&multi->stripes_pending, multi->num_stripes);
3014
3015         while (dev_nr < total_devs) {
3016                 if (total_devs > 1) {
3017                         if (dev_nr < total_devs - 1) {
3018                                 bio = bio_clone(first_bio, GFP_NOFS);
3019                                 BUG_ON(!bio);
3020                         } else {
3021                                 bio = first_bio;
3022                         }
3023                         bio->bi_private = multi;
3024                         bio->bi_end_io = end_bio_multi_stripe;
3025                 }
3026                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3027                 dev = multi->stripes[dev_nr].dev;
3028                 BUG_ON(rw == WRITE && !dev->writeable);
3029                 if (dev && dev->bdev) {
3030                         bio->bi_bdev = dev->bdev;
3031                         if (async_submit)
3032                                 schedule_bio(root, dev, rw, bio);
3033                         else
3034                                 submit_bio(rw, bio);
3035                 } else {
3036                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3037                         bio->bi_sector = logical >> 9;
3038                         bio_endio(bio, -EIO);
3039                 }
3040                 dev_nr++;
3041         }
3042         if (total_devs == 1)
3043                 kfree(multi);
3044         return 0;
3045 }
3046
3047 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3048                                        u8 *uuid, u8 *fsid)
3049 {
3050         struct btrfs_device *device;
3051         struct btrfs_fs_devices *cur_devices;
3052
3053         cur_devices = root->fs_info->fs_devices;
3054         while (cur_devices) {
3055                 if (!fsid ||
3056                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3057                         device = __find_device(&cur_devices->devices,
3058                                                devid, uuid);
3059                         if (device)
3060                                 return device;
3061                 }
3062                 cur_devices = cur_devices->seed;
3063         }
3064         return NULL;
3065 }
3066
3067 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3068                                             u64 devid, u8 *dev_uuid)
3069 {
3070         struct btrfs_device *device;
3071         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3072
3073         device = kzalloc(sizeof(*device), GFP_NOFS);
3074         if (!device)
3075                 return NULL;
3076         list_add(&device->dev_list,
3077                  &fs_devices->devices);
3078         device->barriers = 1;
3079         device->dev_root = root->fs_info->dev_root;
3080         device->devid = devid;
3081         device->work.func = pending_bios_fn;
3082         device->fs_devices = fs_devices;
3083         fs_devices->num_devices++;
3084         spin_lock_init(&device->io_lock);
3085         INIT_LIST_HEAD(&device->dev_alloc_list);
3086         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3087         return device;
3088 }
3089
3090 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3091                           struct extent_buffer *leaf,
3092                           struct btrfs_chunk *chunk)
3093 {
3094         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3095         struct map_lookup *map;
3096         struct extent_map *em;
3097         u64 logical;
3098         u64 length;
3099         u64 devid;
3100         u8 uuid[BTRFS_UUID_SIZE];
3101         int num_stripes;
3102         int ret;
3103         int i;
3104
3105         logical = key->offset;
3106         length = btrfs_chunk_length(leaf, chunk);
3107
3108         read_lock(&map_tree->map_tree.lock);
3109         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3110         read_unlock(&map_tree->map_tree.lock);
3111
3112         /* already mapped? */
3113         if (em && em->start <= logical && em->start + em->len > logical) {
3114                 free_extent_map(em);
3115                 return 0;
3116         } else if (em) {
3117                 free_extent_map(em);
3118         }
3119
3120         em = alloc_extent_map(GFP_NOFS);
3121         if (!em)
3122                 return -ENOMEM;
3123         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3124         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3125         if (!map) {
3126                 free_extent_map(em);
3127                 return -ENOMEM;
3128         }
3129
3130         em->bdev = (struct block_device *)map;
3131         em->start = logical;
3132         em->len = length;
3133         em->block_start = 0;
3134         em->block_len = em->len;
3135
3136         map->num_stripes = num_stripes;
3137         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3138         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3139         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3140         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3141         map->type = btrfs_chunk_type(leaf, chunk);
3142         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3143         for (i = 0; i < num_stripes; i++) {
3144                 map->stripes[i].physical =
3145                         btrfs_stripe_offset_nr(leaf, chunk, i);
3146                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3147                 read_extent_buffer(leaf, uuid, (unsigned long)
3148                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3149                                    BTRFS_UUID_SIZE);
3150                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3151                                                         NULL);
3152                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3153                         kfree(map);
3154                         free_extent_map(em);
3155                         return -EIO;
3156                 }
3157                 if (!map->stripes[i].dev) {
3158                         map->stripes[i].dev =
3159                                 add_missing_dev(root, devid, uuid);
3160                         if (!map->stripes[i].dev) {
3161                                 kfree(map);
3162                                 free_extent_map(em);
3163                                 return -EIO;
3164                         }
3165                 }
3166                 map->stripes[i].dev->in_fs_metadata = 1;
3167         }
3168
3169         write_lock(&map_tree->map_tree.lock);
3170         ret = add_extent_mapping(&map_tree->map_tree, em);
3171         write_unlock(&map_tree->map_tree.lock);
3172         BUG_ON(ret);
3173         free_extent_map(em);
3174
3175         return 0;
3176 }
3177
3178 static int fill_device_from_item(struct extent_buffer *leaf,
3179                                  struct btrfs_dev_item *dev_item,
3180                                  struct btrfs_device *device)
3181 {
3182         unsigned long ptr;
3183
3184         device->devid = btrfs_device_id(leaf, dev_item);
3185         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3186         device->total_bytes = device->disk_total_bytes;
3187         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3188         device->type = btrfs_device_type(leaf, dev_item);
3189         device->io_align = btrfs_device_io_align(leaf, dev_item);
3190         device->io_width = btrfs_device_io_width(leaf, dev_item);
3191         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3192
3193         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3194         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3195
3196         return 0;
3197 }
3198
3199 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3200 {
3201         struct btrfs_fs_devices *fs_devices;
3202         int ret;
3203
3204         mutex_lock(&uuid_mutex);
3205
3206         fs_devices = root->fs_info->fs_devices->seed;
3207         while (fs_devices) {
3208                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3209                         ret = 0;
3210                         goto out;
3211                 }
3212                 fs_devices = fs_devices->seed;
3213         }
3214
3215         fs_devices = find_fsid(fsid);
3216         if (!fs_devices) {
3217                 ret = -ENOENT;
3218                 goto out;
3219         }
3220
3221         fs_devices = clone_fs_devices(fs_devices);
3222         if (IS_ERR(fs_devices)) {
3223                 ret = PTR_ERR(fs_devices);
3224                 goto out;
3225         }
3226
3227         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3228                                    root->fs_info->bdev_holder);
3229         if (ret)
3230                 goto out;
3231
3232         if (!fs_devices->seeding) {
3233                 __btrfs_close_devices(fs_devices);
3234                 free_fs_devices(fs_devices);
3235                 ret = -EINVAL;
3236                 goto out;
3237         }
3238
3239         fs_devices->seed = root->fs_info->fs_devices->seed;
3240         root->fs_info->fs_devices->seed = fs_devices;
3241 out:
3242         mutex_unlock(&uuid_mutex);
3243         return ret;
3244 }
3245
3246 static int read_one_dev(struct btrfs_root *root,
3247                         struct extent_buffer *leaf,
3248                         struct btrfs_dev_item *dev_item)
3249 {
3250         struct btrfs_device *device;
3251         u64 devid;
3252         int ret;
3253         u8 fs_uuid[BTRFS_UUID_SIZE];
3254         u8 dev_uuid[BTRFS_UUID_SIZE];
3255
3256         devid = btrfs_device_id(leaf, dev_item);
3257         read_extent_buffer(leaf, dev_uuid,
3258                            (unsigned long)btrfs_device_uuid(dev_item),
3259                            BTRFS_UUID_SIZE);
3260         read_extent_buffer(leaf, fs_uuid,
3261                            (unsigned long)btrfs_device_fsid(dev_item),
3262                            BTRFS_UUID_SIZE);
3263
3264         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3265                 ret = open_seed_devices(root, fs_uuid);
3266                 if (ret && !btrfs_test_opt(root, DEGRADED))
3267                         return ret;
3268         }
3269
3270         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3271         if (!device || !device->bdev) {
3272                 if (!btrfs_test_opt(root, DEGRADED))
3273                         return -EIO;
3274
3275                 if (!device) {
3276                         printk(KERN_WARNING "warning devid %llu missing\n",
3277                                (unsigned long long)devid);
3278                         device = add_missing_dev(root, devid, dev_uuid);
3279                         if (!device)
3280                                 return -ENOMEM;
3281                 }
3282         }
3283
3284         if (device->fs_devices != root->fs_info->fs_devices) {
3285                 BUG_ON(device->writeable);
3286                 if (device->generation !=
3287                     btrfs_device_generation(leaf, dev_item))
3288                         return -EINVAL;
3289         }
3290
3291         fill_device_from_item(leaf, dev_item, device);
3292         device->dev_root = root->fs_info->dev_root;
3293         device->in_fs_metadata = 1;
3294         if (device->writeable)
3295                 device->fs_devices->total_rw_bytes += device->total_bytes;
3296         ret = 0;
3297         return ret;
3298 }
3299
3300 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3301 {
3302         struct btrfs_dev_item *dev_item;
3303
3304         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3305                                                      dev_item);
3306         return read_one_dev(root, buf, dev_item);
3307 }
3308
3309 int btrfs_read_sys_array(struct btrfs_root *root)
3310 {
3311         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3312         struct extent_buffer *sb;
3313         struct btrfs_disk_key *disk_key;
3314         struct btrfs_chunk *chunk;
3315         u8 *ptr;
3316         unsigned long sb_ptr;
3317         int ret = 0;
3318         u32 num_stripes;
3319         u32 array_size;
3320         u32 len = 0;
3321         u32 cur;
3322         struct btrfs_key key;
3323
3324         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3325                                           BTRFS_SUPER_INFO_SIZE);
3326         if (!sb)
3327                 return -ENOMEM;
3328         btrfs_set_buffer_uptodate(sb);
3329         btrfs_set_buffer_lockdep_class(sb, 0);
3330
3331         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3332         array_size = btrfs_super_sys_array_size(super_copy);
3333
3334         ptr = super_copy->sys_chunk_array;
3335         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3336         cur = 0;
3337
3338         while (cur < array_size) {
3339                 disk_key = (struct btrfs_disk_key *)ptr;
3340                 btrfs_disk_key_to_cpu(&key, disk_key);
3341
3342                 len = sizeof(*disk_key); ptr += len;
3343                 sb_ptr += len;
3344                 cur += len;
3345
3346                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3347                         chunk = (struct btrfs_chunk *)sb_ptr;
3348                         ret = read_one_chunk(root, &key, sb, chunk);
3349                         if (ret)
3350                                 break;
3351                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3352                         len = btrfs_chunk_item_size(num_stripes);
3353                 } else {
3354                         ret = -EIO;
3355                         break;
3356                 }
3357                 ptr += len;
3358                 sb_ptr += len;
3359                 cur += len;
3360         }
3361         free_extent_buffer(sb);
3362         return ret;
3363 }
3364
3365 int btrfs_read_chunk_tree(struct btrfs_root *root)
3366 {
3367         struct btrfs_path *path;
3368         struct extent_buffer *leaf;
3369         struct btrfs_key key;
3370         struct btrfs_key found_key;
3371         int ret;
3372         int slot;
3373
3374         root = root->fs_info->chunk_root;
3375
3376         path = btrfs_alloc_path();
3377         if (!path)
3378                 return -ENOMEM;
3379
3380         /* first we search for all of the device items, and then we
3381          * read in all of the chunk items.  This way we can create chunk
3382          * mappings that reference all of the devices that are afound
3383          */
3384         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3385         key.offset = 0;
3386         key.type = 0;
3387 again:
3388         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3389         while (1) {
3390                 leaf = path->nodes[0];
3391                 slot = path->slots[0];
3392                 if (slot >= btrfs_header_nritems(leaf)) {
3393                         ret = btrfs_next_leaf(root, path);
3394                         if (ret == 0)
3395                                 continue;
3396                         if (ret < 0)
3397                                 goto error;
3398                         break;
3399                 }
3400                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3401                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3402                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3403                                 break;
3404                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3405                                 struct btrfs_dev_item *dev_item;
3406                                 dev_item = btrfs_item_ptr(leaf, slot,
3407                                                   struct btrfs_dev_item);
3408                                 ret = read_one_dev(root, leaf, dev_item);
3409                                 if (ret)
3410                                         goto error;
3411                         }
3412                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3413                         struct btrfs_chunk *chunk;
3414                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3415                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3416                         if (ret)
3417                                 goto error;
3418                 }
3419                 path->slots[0]++;
3420         }
3421         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3422                 key.objectid = 0;
3423                 btrfs_release_path(root, path);
3424                 goto again;
3425         }
3426         ret = 0;
3427 error:
3428         btrfs_free_path(path);
3429         return ret;
3430 }