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/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <asm/div64.h>
  27 #include "compat.h"
  28 #include "ctree.h"
  29 #include "extent_map.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32 #include "print-tree.h"
  33 #include "volumes.h"
  34 #include "async-thread.h"
  35
  36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  37                                 struct btrfs_root *root,
  38                                 struct btrfs_device *device);
  39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  40
  41 static DEFINE_MUTEX(uuid_mutex);
  42 static LIST_HEAD(fs_uuids);
  43
  44 static void lock_chunks(struct btrfs_root *root)
  45 {
  46         mutex_lock(&root->fs_info->chunk_mutex);
  47 }
  48
  49 static void unlock_chunks(struct btrfs_root *root)
  50 {
  51         mutex_unlock(&root->fs_info->chunk_mutex);
  52 }
  53
  54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  55 {
  56         struct btrfs_device *device;
  57         WARN_ON(fs_devices->opened);
  58         while (!list_empty(&fs_devices->devices)) {
  59                 device = list_entry(fs_devices->devices.next,
  60                                     struct btrfs_device, dev_list);
  61                 list_del(&device->dev_list);
  62                 kfree(device->name);
  63                 kfree(device);
  64         }
  65         kfree(fs_devices);
  66 }
  67
  68 int btrfs_cleanup_fs_uuids(void)
  69 {
  70         struct btrfs_fs_devices *fs_devices;
  71
  72         while (!list_empty(&fs_uuids)) {
  73                 fs_devices = list_entry(fs_uuids.next,
  74                                         struct btrfs_fs_devices, list);
  75                 list_del(&fs_devices->list);
  76                 free_fs_devices(fs_devices);
  77         }
  78         return 0;
  79 }
  80
  81 static noinline struct btrfs_device *__find_device(struct list_head *head,
  82                                                    u64 devid, u8 *uuid)
  83 {
  84         struct btrfs_device *dev;
  85
  86         list_for_each_entry(dev, head, dev_list) {
  87                 if (dev->devid == devid &&
  88                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  89                         return dev;
  90                 }
  91         }
  92         return NULL;
  93 }
  94
  95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
  96 {
  97         struct btrfs_fs_devices *fs_devices;
  98
  99         list_for_each_entry(fs_devices, &fs_uuids, list) {
 100                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 101                         return fs_devices;
 102         }
 103         return NULL;
 104 }
 105
 106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 107                         struct bio *head, struct bio *tail)
 108 {
 109
 110         struct bio *old_head;
 111
 112         old_head = pending_bios->head;
 113         pending_bios->head = head;
 114         if (pending_bios->tail)
 115                 tail->bi_next = old_head;
 116         else
 117                 pending_bios->tail = tail;
 118 }
 119
 120 /*
 121  * we try to collect pending bios for a device so we don't get a large
 122  * number of procs sending bios down to the same device.  This greatly
 123  * improves the schedulers ability to collect and merge the bios.
 124  *
 125  * But, it also turns into a long list of bios to process and that is sure
 126  * to eventually make the worker thread block.  The solution here is to
 127  * make some progress and then put this work struct back at the end of
 128  * the list if the block device is congested.  This way, multiple devices
 129  * can make progress from a single worker thread.
 130  */
 131 static noinline int run_scheduled_bios(struct btrfs_device *device)
 132 {
 133         struct bio *pending;
 134         struct backing_dev_info *bdi;
 135         struct btrfs_fs_info *fs_info;
 136         struct btrfs_pending_bios *pending_bios;
 137         struct bio *tail;
 138         struct bio *cur;
 139         int again = 0;
 140         unsigned long num_run;
 141         unsigned long batch_run = 0;
 142         unsigned long limit;
 143         unsigned long last_waited = 0;
 144         int force_reg = 0;
 145         struct blk_plug plug;
 146
 147         /*
 148          * this function runs all the bios we've collected for
 149          * a particular device.  We don't want to wander off to
 150          * another device without first sending all of these down.
 151          * So, setup a plug here and finish it off before we return
 152          */
 153         blk_start_plug(&plug);
 154
 155         bdi = blk_get_backing_dev_info(device->bdev);
 156         fs_info = device->dev_root->fs_info;
 157         limit = btrfs_async_submit_limit(fs_info);
 158         limit = limit * 2 / 3;
 159
 160 loop:
 161         spin_lock(&device->io_lock);
 162
 163 loop_lock:
 164         num_run = 0;
 165
 166         /* take all the bios off the list at once and process them
 167          * later on (without the lock held).  But, remember the
 168          * tail and other pointers so the bios can be properly reinserted
 169          * into the list if we hit congestion
 170          */
 171         if (!force_reg && device->pending_sync_bios.head) {
 172                 pending_bios = &device->pending_sync_bios;
 173                 force_reg = 1;
 174         } else {
 175                 pending_bios = &device->pending_bios;
 176                 force_reg = 0;
 177         }
 178
 179         pending = pending_bios->head;
 180         tail = pending_bios->tail;
 181         WARN_ON(pending && !tail);
 182
 183         /*
 184          * if pending was null this time around, no bios need processing
 185          * at all and we can stop.  Otherwise it'll loop back up again
 186          * and do an additional check so no bios are missed.
 187          *
 188          * device->running_pending is used to synchronize with the
 189          * schedule_bio code.
 190          */
 191         if (device->pending_sync_bios.head == NULL &&
 192             device->pending_bios.head == NULL) {
 193                 again = 0;
 194                 device->running_pending = 0;
 195         } else {
 196                 again = 1;
 197                 device->running_pending = 1;
 198         }
 199
 200         pending_bios->head = NULL;
 201         pending_bios->tail = NULL;
 202
 203         spin_unlock(&device->io_lock);
 204
 205         while (pending) {
 206
 207                 rmb();
 208                 /* we want to work on both lists, but do more bios on the
 209                  * sync list than the regular list
 210                  */
 211                 if ((num_run > 32 &&
 212                     pending_bios != &device->pending_sync_bios &&
 213                     device->pending_sync_bios.head) ||
 214                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 215                     device->pending_bios.head)) {
 216                         spin_lock(&device->io_lock);
 217                         requeue_list(pending_bios, pending, tail);
 218                         goto loop_lock;
 219                 }
 220
 221                 cur = pending;
 222                 pending = pending->bi_next;
 223                 cur->bi_next = NULL;
 224                 atomic_dec(&fs_info->nr_async_bios);
 225
 226                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 227                     waitqueue_active(&fs_info->async_submit_wait))
 228                         wake_up(&fs_info->async_submit_wait);
 229
 230                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 231
 232                 submit_bio(cur->bi_rw, cur);
 233                 num_run++;
 234                 batch_run++;
 235                 if (need_resched())
 236                         cond_resched();
 237
 238                 /*
 239                  * we made progress, there is more work to do and the bdi
 240                  * is now congested.  Back off and let other work structs
 241                  * run instead
 242                  */
 243                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 244                     fs_info->fs_devices->open_devices > 1) {
 245                         struct io_context *ioc;
 246
 247                         ioc = current->io_context;
 248
 249                         /*
 250                          * the main goal here is that we don't want to
 251                          * block if we're going to be able to submit
 252                          * more requests without blocking.
 253                          *
 254                          * This code does two great things, it pokes into
 255                          * the elevator code from a filesystem _and_
 256                          * it makes assumptions about how batching works.
 257                          */
 258                         if (ioc && ioc->nr_batch_requests > 0 &&
 259                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 260                             (last_waited == 0 ||
 261                              ioc->last_waited == last_waited)) {
 262                                 /*
 263                                  * we want to go through our batch of
 264                                  * requests and stop.  So, we copy out
 265                                  * the ioc->last_waited time and test
 266                                  * against it before looping
 267                                  */
 268                                 last_waited = ioc->last_waited;
 269                                 if (need_resched())
 270                                         cond_resched();
 271                                 continue;
 272                         }
 273                         spin_lock(&device->io_lock);
 274                         requeue_list(pending_bios, pending, tail);
 275                         device->running_pending = 1;
 276
 277                         spin_unlock(&device->io_lock);
 278                         btrfs_requeue_work(&device->work);
 279                         goto done;
 280                 }
 281         }
 282
 283         cond_resched();
 284         if (again)
 285                 goto loop;
 286
 287         spin_lock(&device->io_lock);
 288         if (device->pending_bios.head || device->pending_sync_bios.head)
 289                 goto loop_lock;
 290         spin_unlock(&device->io_lock);
 291
 292 done:
 293         blk_finish_plug(&plug);
 294         return 0;
 295 }
 296
 297 static void pending_bios_fn(struct btrfs_work *work)
 298 {
 299         struct btrfs_device *device;
 300
 301         device = container_of(work, struct btrfs_device, work);
 302         run_scheduled_bios(device);
 303 }
 304
 305 static noinline int device_list_add(const char *path,
 306                            struct btrfs_super_block *disk_super,
 307                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 308 {
 309         struct btrfs_device *device;
 310         struct btrfs_fs_devices *fs_devices;
 311         u64 found_transid = btrfs_super_generation(disk_super);
 312         char *name;
 313
 314         fs_devices = find_fsid(disk_super->fsid);
 315         if (!fs_devices) {
 316                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 317                 if (!fs_devices)
 318                         return -ENOMEM;
 319                 INIT_LIST_HEAD(&fs_devices->devices);
 320                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 321                 list_add(&fs_devices->list, &fs_uuids);
 322                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 323                 fs_devices->latest_devid = devid;
 324                 fs_devices->latest_trans = found_transid;
 325                 mutex_init(&fs_devices->device_list_mutex);
 326                 device = NULL;
 327         } else {
 328                 device = __find_device(&fs_devices->devices, devid,
 329                                        disk_super->dev_item.uuid);
 330         }
 331         if (!device) {
 332                 if (fs_devices->opened)
 333                         return -EBUSY;
 334
 335                 device = kzalloc(sizeof(*device), GFP_NOFS);
 336                 if (!device) {
 337                         /* we can safely leave the fs_devices entry around */
 338                         return -ENOMEM;
 339                 }
 340                 device->devid = devid;
 341                 device->work.func = pending_bios_fn;
 342                 memcpy(device->uuid, disk_super->dev_item.uuid,
 343                        BTRFS_UUID_SIZE);
 344                 spin_lock_init(&device->io_lock);
 345                 device->name = kstrdup(path, GFP_NOFS);
 346                 if (!device->name) {
 347                         kfree(device);
 348                         return -ENOMEM;
 349                 }
 350                 INIT_LIST_HEAD(&device->dev_alloc_list);
 351
 352                 mutex_lock(&fs_devices->device_list_mutex);
 353                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 354                 mutex_unlock(&fs_devices->device_list_mutex);
 355
 356                 device->fs_devices = fs_devices;
 357                 fs_devices->num_devices++;
 358         } else if (!device->name || strcmp(device->name, path)) {
 359                 name = kstrdup(path, GFP_NOFS);
 360                 if (!name)
 361                         return -ENOMEM;
 362                 kfree(device->name);
 363                 device->name = name;
 364                 if (device->missing) {
 365                         fs_devices->missing_devices--;
 366                         device->missing = 0;
 367                 }
 368         }
 369
 370         if (found_transid > fs_devices->latest_trans) {
 371                 fs_devices->latest_devid = devid;
 372                 fs_devices->latest_trans = found_transid;
 373         }
 374         *fs_devices_ret = fs_devices;
 375         return 0;
 376 }
 377
 378 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 379 {
 380         struct btrfs_fs_devices *fs_devices;
 381         struct btrfs_device *device;
 382         struct btrfs_device *orig_dev;
 383
 384         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 385         if (!fs_devices)
 386                 return ERR_PTR(-ENOMEM);
 387
 388         INIT_LIST_HEAD(&fs_devices->devices);
 389         INIT_LIST_HEAD(&fs_devices->alloc_list);
 390         INIT_LIST_HEAD(&fs_devices->list);
 391         mutex_init(&fs_devices->device_list_mutex);
 392         fs_devices->latest_devid = orig->latest_devid;
 393         fs_devices->latest_trans = orig->latest_trans;
 394         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 395
 396         /* We have held the volume lock, it is safe to get the devices. */
 397         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 398                 device = kzalloc(sizeof(*device), GFP_NOFS);
 399                 if (!device)
 400                         goto error;
 401
 402                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 403                 if (!device->name) {
 404                         kfree(device);
 405                         goto error;
 406                 }
 407
 408                 device->devid = orig_dev->devid;
 409                 device->work.func = pending_bios_fn;
 410                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 411                 spin_lock_init(&device->io_lock);
 412                 INIT_LIST_HEAD(&device->dev_list);
 413                 INIT_LIST_HEAD(&device->dev_alloc_list);
 414
 415                 list_add(&device->dev_list, &fs_devices->devices);
 416                 device->fs_devices = fs_devices;
 417                 fs_devices->num_devices++;
 418         }
 419         return fs_devices;
 420 error:
 421         free_fs_devices(fs_devices);
 422         return ERR_PTR(-ENOMEM);
 423 }
 424
 425 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 426 {
 427         struct btrfs_device *device, *next;
 428
 429         mutex_lock(&uuid_mutex);
 430 again:
 431         /* This is the initialized path, it is safe to release the devices. */
 432         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 433                 if (device->in_fs_metadata)
 434                         continue;
 435
 436                 if (device->bdev) {
 437                         blkdev_put(device->bdev, device->mode);
 438                         device->bdev = NULL;
 439                         fs_devices->open_devices--;
 440                 }
 441                 if (device->writeable) {
 442                         list_del_init(&device->dev_alloc_list);
 443                         device->writeable = 0;
 444                         fs_devices->rw_devices--;
 445                 }
 446                 list_del_init(&device->dev_list);
 447                 fs_devices->num_devices--;
 448                 kfree(device->name);
 449                 kfree(device);
 450         }
 451
 452         if (fs_devices->seed) {
 453                 fs_devices = fs_devices->seed;
 454                 goto again;
 455         }
 456
 457         mutex_unlock(&uuid_mutex);
 458         return 0;
 459 }
 460
 461 static void __free_device(struct work_struct *work)
 462 {
 463         struct btrfs_device *device;
 464
 465         device = container_of(work, struct btrfs_device, rcu_work);
 466
 467         if (device->bdev)
 468                 blkdev_put(device->bdev, device->mode);
 469
 470         kfree(device->name);
 471         kfree(device);
 472 }
 473
 474 static void free_device(struct rcu_head *head)
 475 {
 476         struct btrfs_device *device;
 477
 478         device = container_of(head, struct btrfs_device, rcu);
 479
 480         INIT_WORK(&device->rcu_work, __free_device);
 481         schedule_work(&device->rcu_work);
 482 }
 483
 484 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 485 {
 486         struct btrfs_device *device;
 487
 488         if (--fs_devices->opened > 0)
 489                 return 0;
 490
 491         mutex_lock(&fs_devices->device_list_mutex);
 492         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 493                 struct btrfs_device *new_device;
 494
 495                 if (device->bdev)
 496                         fs_devices->open_devices--;
 497
 498                 if (device->writeable) {
 499                         list_del_init(&device->dev_alloc_list);
 500                         fs_devices->rw_devices--;
 501                 }
 502
 503                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 504                 BUG_ON(!new_device);
 505                 memcpy(new_device, device, sizeof(*new_device));
 506                 new_device->name = kstrdup(device->name, GFP_NOFS);
 507                 BUG_ON(device->name && !new_device->name);
 508                 new_device->bdev = NULL;
 509                 new_device->writeable = 0;
 510                 new_device->in_fs_metadata = 0;
 511                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 512
 513                 call_rcu(&device->rcu, free_device);
 514         }
 515         mutex_unlock(&fs_devices->device_list_mutex);
 516
 517         WARN_ON(fs_devices->open_devices);
 518         WARN_ON(fs_devices->rw_devices);
 519         fs_devices->opened = 0;
 520         fs_devices->seeding = 0;
 521
 522         return 0;
 523 }
 524
 525 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 526 {
 527         struct btrfs_fs_devices *seed_devices = NULL;
 528         int ret;
 529
 530         mutex_lock(&uuid_mutex);
 531         ret = __btrfs_close_devices(fs_devices);
 532         if (!fs_devices->opened) {
 533                 seed_devices = fs_devices->seed;
 534                 fs_devices->seed = NULL;
 535         }
 536         mutex_unlock(&uuid_mutex);
 537
 538         while (seed_devices) {
 539                 fs_devices = seed_devices;
 540                 seed_devices = fs_devices->seed;
 541                 __btrfs_close_devices(fs_devices);
 542                 free_fs_devices(fs_devices);
 543         }
 544         return ret;
 545 }
 546
 547 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 548                                 fmode_t flags, void *holder)
 549 {
 550         struct block_device *bdev;
 551         struct list_head *head = &fs_devices->devices;
 552         struct btrfs_device *device;
 553         struct block_device *latest_bdev = NULL;
 554         struct buffer_head *bh;
 555         struct btrfs_super_block *disk_super;
 556         u64 latest_devid = 0;
 557         u64 latest_transid = 0;
 558         u64 devid;
 559         int seeding = 1;
 560         int ret = 0;
 561
 562         flags |= FMODE_EXCL;
 563
 564         list_for_each_entry(device, head, dev_list) {
 565                 if (device->bdev)
 566                         continue;
 567                 if (!device->name)
 568                         continue;
 569
 570                 bdev = blkdev_get_by_path(device->name, flags, holder);
 571                 if (IS_ERR(bdev)) {
 572                         printk(KERN_INFO "open %s failed\n", device->name);
 573                         goto error;
 574                 }
 575                 set_blocksize(bdev, 4096);
 576
 577                 bh = btrfs_read_dev_super(bdev);
 578                 if (!bh) {
 579                         ret = -EINVAL;
 580                         goto error_close;
 581                 }
 582
 583                 disk_super = (struct btrfs_super_block *)bh->b_data;
 584                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 585                 if (devid != device->devid)
 586                         goto error_brelse;
 587
 588                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 589                            BTRFS_UUID_SIZE))
 590                         goto error_brelse;
 591
 592                 device->generation = btrfs_super_generation(disk_super);
 593                 if (!latest_transid || device->generation > latest_transid) {
 594                         latest_devid = devid;
 595                         latest_transid = device->generation;
 596                         latest_bdev = bdev;
 597                 }
 598
 599                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 600                         device->writeable = 0;
 601                 } else {
 602                         device->writeable = !bdev_read_only(bdev);
 603                         seeding = 0;
 604                 }
 605
 606                 device->bdev = bdev;
 607                 device->in_fs_metadata = 0;
 608                 device->mode = flags;
 609
 610                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 611                         fs_devices->rotating = 1;
 612
 613                 fs_devices->open_devices++;
 614                 if (device->writeable) {
 615                         fs_devices->rw_devices++;
 616                         list_add(&device->dev_alloc_list,
 617                                  &fs_devices->alloc_list);
 618                 }
 619                 brelse(bh);
 620                 continue;
 621
 622 error_brelse:
 623                 brelse(bh);
 624 error_close:
 625                 blkdev_put(bdev, flags);
 626 error:
 627                 continue;
 628         }
 629         if (fs_devices->open_devices == 0) {
 630                 ret = -EIO;
 631                 goto out;
 632         }
 633         fs_devices->seeding = seeding;
 634         fs_devices->opened = 1;
 635         fs_devices->latest_bdev = latest_bdev;
 636         fs_devices->latest_devid = latest_devid;
 637         fs_devices->latest_trans = latest_transid;
 638         fs_devices->total_rw_bytes = 0;
 639 out:
 640         return ret;
 641 }
 642
 643 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 644                        fmode_t flags, void *holder)
 645 {
 646         int ret;
 647
 648         mutex_lock(&uuid_mutex);
 649         if (fs_devices->opened) {
 650                 fs_devices->opened++;
 651                 ret = 0;
 652         } else {
 653                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 654         }
 655         mutex_unlock(&uuid_mutex);
 656         return ret;
 657 }
 658
 659 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 660                           struct btrfs_fs_devices **fs_devices_ret)
 661 {
 662         struct btrfs_super_block *disk_super;
 663         struct block_device *bdev;
 664         struct buffer_head *bh;
 665         int ret;
 666         u64 devid;
 667         u64 transid;
 668
 669         mutex_lock(&uuid_mutex);
 670
 671         flags |= FMODE_EXCL;
 672         bdev = blkdev_get_by_path(path, flags, holder);
 673
 674         if (IS_ERR(bdev)) {
 675                 ret = PTR_ERR(bdev);
 676                 goto error;
 677         }
 678
 679         ret = set_blocksize(bdev, 4096);
 680         if (ret)
 681                 goto error_close;
 682         bh = btrfs_read_dev_super(bdev);
 683         if (!bh) {
 684                 ret = -EINVAL;
 685                 goto error_close;
 686         }
 687         disk_super = (struct btrfs_super_block *)bh->b_data;
 688         devid = btrfs_stack_device_id(&disk_super->dev_item);
 689         transid = btrfs_super_generation(disk_super);
 690         if (disk_super->label[0])
 691                 printk(KERN_INFO "device label %s ", disk_super->label);
 692         else
 693                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 694         printk(KERN_CONT "devid %llu transid %llu %s\n",
 695                (unsigned long long)devid, (unsigned long long)transid, path);
 696         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 697
 698         brelse(bh);
 699 error_close:
 700         blkdev_put(bdev, flags);
 701 error:
 702         mutex_unlock(&uuid_mutex);
 703         return ret;
 704 }
 705
 706 /* helper to account the used device space in the range */
 707 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 708                                    u64 end, u64 *length)
 709 {
 710         struct btrfs_key key;
 711         struct btrfs_root *root = device->dev_root;
 712         struct btrfs_dev_extent *dev_extent;
 713         struct btrfs_path *path;
 714         u64 extent_end;
 715         int ret;
 716         int slot;
 717         struct extent_buffer *l;
 718
 719         *length = 0;
 720
 721         if (start >= device->total_bytes)
 722                 return 0;
 723
 724         path = btrfs_alloc_path();
 725         if (!path)
 726                 return -ENOMEM;
 727         path->reada = 2;
 728
 729         key.objectid = device->devid;
 730         key.offset = start;
 731         key.type = BTRFS_DEV_EXTENT_KEY;
 732
 733         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 734         if (ret < 0)
 735                 goto out;
 736         if (ret > 0) {
 737                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 738                 if (ret < 0)
 739                         goto out;
 740         }
 741
 742         while (1) {
 743                 l = path->nodes[0];
 744                 slot = path->slots[0];
 745                 if (slot >= btrfs_header_nritems(l)) {
 746                         ret = btrfs_next_leaf(root, path);
 747                         if (ret == 0)
 748                                 continue;
 749                         if (ret < 0)
 750                                 goto out;
 751
 752                         break;
 753                 }
 754                 btrfs_item_key_to_cpu(l, &key, slot);
 755
 756                 if (key.objectid < device->devid)
 757                         goto next;
 758
 759                 if (key.objectid > device->devid)
 760                         break;
 761
 762                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 763                         goto next;
 764
 765                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 766                 extent_end = key.offset + btrfs_dev_extent_length(l,
 767                                                                   dev_extent);
 768                 if (key.offset <= start && extent_end > end) {
 769                         *length = end - start + 1;
 770                         break;
 771                 } else if (key.offset <= start && extent_end > start)
 772                         *length += extent_end - start;
 773                 else if (key.offset > start && extent_end <= end)
 774                         *length += extent_end - key.offset;
 775                 else if (key.offset > start && key.offset <= end) {
 776                         *length += end - key.offset + 1;
 777                         break;
 778                 } else if (key.offset > end)
 779                         break;
 780
 781 next:
 782                 path->slots[0]++;
 783         }
 784         ret = 0;
 785 out:
 786         btrfs_free_path(path);
 787         return ret;
 788 }
 789
 790 /*
 791  * find_free_dev_extent - find free space in the specified device
 792  * @trans:      transaction handler
 793  * @device:     the device which we search the free space in
 794  * @num_bytes:  the size of the free space that we need
 795  * @start:      store the start of the free space.
 796  * @len:        the size of the free space. that we find, or the size of the max
 797  *              free space if we don't find suitable free space
 798  *
 799  * this uses a pretty simple search, the expectation is that it is
 800  * called very infrequently and that a given device has a small number
 801  * of extents
 802  *
 803  * @start is used to store the start of the free space if we find. But if we
 804  * don't find suitable free space, it will be used to store the start position
 805  * of the max free space.
 806  *
 807  * @len is used to store the size of the free space that we find.
 808  * But if we don't find suitable free space, it is used to store the size of
 809  * the max free space.
 810  */
 811 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 812                          struct btrfs_device *device, u64 num_bytes,
 813                          u64 *start, u64 *len)
 814 {
 815         struct btrfs_key key;
 816         struct btrfs_root *root = device->dev_root;
 817         struct btrfs_dev_extent *dev_extent;
 818         struct btrfs_path *path;
 819         u64 hole_size;
 820         u64 max_hole_start;
 821         u64 max_hole_size;
 822         u64 extent_end;
 823         u64 search_start;
 824         u64 search_end = device->total_bytes;
 825         int ret;
 826         int slot;
 827         struct extent_buffer *l;
 828
 829         /* FIXME use last free of some kind */
 830
 831         /* we don't want to overwrite the superblock on the drive,
 832          * so we make sure to start at an offset of at least 1MB
 833          */
 834         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 835
 836         max_hole_start = search_start;
 837         max_hole_size = 0;
 838
 839         if (search_start >= search_end) {
 840                 ret = -ENOSPC;
 841                 goto error;
 842         }
 843
 844         path = btrfs_alloc_path();
 845         if (!path) {
 846                 ret = -ENOMEM;
 847                 goto error;
 848         }
 849         path->reada = 2;
 850
 851         key.objectid = device->devid;
 852         key.offset = search_start;
 853         key.type = BTRFS_DEV_EXTENT_KEY;
 854
 855         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 856         if (ret < 0)
 857                 goto out;
 858         if (ret > 0) {
 859                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 860                 if (ret < 0)
 861                         goto out;
 862         }
 863
 864         while (1) {
 865                 l = path->nodes[0];
 866                 slot = path->slots[0];
 867                 if (slot >= btrfs_header_nritems(l)) {
 868                         ret = btrfs_next_leaf(root, path);
 869                         if (ret == 0)
 870                                 continue;
 871                         if (ret < 0)
 872                                 goto out;
 873
 874                         break;
 875                 }
 876                 btrfs_item_key_to_cpu(l, &key, slot);
 877
 878                 if (key.objectid < device->devid)
 879                         goto next;
 880
 881                 if (key.objectid > device->devid)
 882                         break;
 883
 884                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 885                         goto next;
 886
 887                 if (key.offset > search_start) {
 888                         hole_size = key.offset - search_start;
 889
 890                         if (hole_size > max_hole_size) {
 891                                 max_hole_start = search_start;
 892                                 max_hole_size = hole_size;
 893                         }
 894
 895                         /*
 896                          * If this free space is greater than which we need,
 897                          * it must be the max free space that we have found
 898                          * until now, so max_hole_start must point to the start
 899                          * of this free space and the length of this free space
 900                          * is stored in max_hole_size. Thus, we return
 901                          * max_hole_start and max_hole_size and go back to the
 902                          * caller.
 903                          */
 904                         if (hole_size >= num_bytes) {
 905                                 ret = 0;
 906                                 goto out;
 907                         }
 908                 }
 909
 910                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 911                 extent_end = key.offset + btrfs_dev_extent_length(l,
 912                                                                   dev_extent);
 913                 if (extent_end > search_start)
 914                         search_start = extent_end;
 915 next:
 916                 path->slots[0]++;
 917                 cond_resched();
 918         }
 919
 920         hole_size = search_end- search_start;
 921         if (hole_size > max_hole_size) {
 922                 max_hole_start = search_start;
 923                 max_hole_size = hole_size;
 924         }
 925
 926         /* See above. */
 927         if (hole_size < num_bytes)
 928                 ret = -ENOSPC;
 929         else
 930                 ret = 0;
 931
 932 out:
 933         btrfs_free_path(path);
 934 error:
 935         *start = max_hole_start;
 936         if (len)
 937                 *len = max_hole_size;
 938         return ret;
 939 }
 940
 941 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 942                           struct btrfs_device *device,
 943                           u64 start)
 944 {
 945         int ret;
 946         struct btrfs_path *path;
 947         struct btrfs_root *root = device->dev_root;
 948         struct btrfs_key key;
 949         struct btrfs_key found_key;
 950         struct extent_buffer *leaf = NULL;
 951         struct btrfs_dev_extent *extent = NULL;
 952
 953         path = btrfs_alloc_path();
 954         if (!path)
 955                 return -ENOMEM;
 956
 957         key.objectid = device->devid;
 958         key.offset = start;
 959         key.type = BTRFS_DEV_EXTENT_KEY;
 960
 961         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 962         if (ret > 0) {
 963                 ret = btrfs_previous_item(root, path, key.objectid,
 964                                           BTRFS_DEV_EXTENT_KEY);
 965                 if (ret)
 966                         goto out;
 967                 leaf = path->nodes[0];
 968                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 969                 extent = btrfs_item_ptr(leaf, path->slots[0],
 970                                         struct btrfs_dev_extent);
 971                 BUG_ON(found_key.offset > start || found_key.offset +
 972                        btrfs_dev_extent_length(leaf, extent) < start);
 973         } else if (ret == 0) {
 974                 leaf = path->nodes[0];
 975                 extent = btrfs_item_ptr(leaf, path->slots[0],
 976                                         struct btrfs_dev_extent);
 977         }
 978         BUG_ON(ret);
 979
 980         if (device->bytes_used > 0)
 981                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 982         ret = btrfs_del_item(trans, root, path);
 983
 984 out:
 985         btrfs_free_path(path);
 986         return ret;
 987 }
 988
 989 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 990                            struct btrfs_device *device,
 991                            u64 chunk_tree, u64 chunk_objectid,
 992                            u64 chunk_offset, u64 start, u64 num_bytes)
 993 {
 994         int ret;
 995         struct btrfs_path *path;
 996         struct btrfs_root *root = device->dev_root;
 997         struct btrfs_dev_extent *extent;
 998         struct extent_buffer *leaf;
 999         struct btrfs_key key;
1000
1001         WARN_ON(!device->in_fs_metadata);
1002         path = btrfs_alloc_path();
1003         if (!path)
1004                 return -ENOMEM;
1005
1006         key.objectid = device->devid;
1007         key.offset = start;
1008         key.type = BTRFS_DEV_EXTENT_KEY;
1009         ret = btrfs_insert_empty_item(trans, root, path, &key,
1010                                       sizeof(*extent));
1011         BUG_ON(ret);
1012
1013         leaf = path->nodes[0];
1014         extent = btrfs_item_ptr(leaf, path->slots[0],
1015                                 struct btrfs_dev_extent);
1016         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1017         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1018         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1019
1020         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1021                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1022                     BTRFS_UUID_SIZE);
1023
1024         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1025         btrfs_mark_buffer_dirty(leaf);
1026         btrfs_free_path(path);
1027         return ret;
1028 }
1029
1030 static noinline int find_next_chunk(struct btrfs_root *root,
1031                                     u64 objectid, u64 *offset)
1032 {
1033         struct btrfs_path *path;
1034         int ret;
1035         struct btrfs_key key;
1036         struct btrfs_chunk *chunk;
1037         struct btrfs_key found_key;
1038
1039         path = btrfs_alloc_path();
1040         BUG_ON(!path);
1041
1042         key.objectid = objectid;
1043         key.offset = (u64)-1;
1044         key.type = BTRFS_CHUNK_ITEM_KEY;
1045
1046         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1047         if (ret < 0)
1048                 goto error;
1049
1050         BUG_ON(ret == 0);
1051
1052         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1053         if (ret) {
1054                 *offset = 0;
1055         } else {
1056                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1057                                       path->slots[0]);
1058                 if (found_key.objectid != objectid)
1059                         *offset = 0;
1060                 else {
1061                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1062                                                struct btrfs_chunk);
1063                         *offset = found_key.offset +
1064                                 btrfs_chunk_length(path->nodes[0], chunk);
1065                 }
1066         }
1067         ret = 0;
1068 error:
1069         btrfs_free_path(path);
1070         return ret;
1071 }
1072
1073 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1074 {
1075         int ret;
1076         struct btrfs_key key;
1077         struct btrfs_key found_key;
1078         struct btrfs_path *path;
1079
1080         root = root->fs_info->chunk_root;
1081
1082         path = btrfs_alloc_path();
1083         if (!path)
1084                 return -ENOMEM;
1085
1086         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1087         key.type = BTRFS_DEV_ITEM_KEY;
1088         key.offset = (u64)-1;
1089
1090         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1091         if (ret < 0)
1092                 goto error;
1093
1094         BUG_ON(ret == 0);
1095
1096         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1097                                   BTRFS_DEV_ITEM_KEY);
1098         if (ret) {
1099                 *objectid = 1;
1100         } else {
1101                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1102                                       path->slots[0]);
1103                 *objectid = found_key.offset + 1;
1104         }
1105         ret = 0;
1106 error:
1107         btrfs_free_path(path);
1108         return ret;
1109 }
1110
1111 /*
1112  * the device information is stored in the chunk root
1113  * the btrfs_device struct should be fully filled in
1114  */
1115 int btrfs_add_device(struct btrfs_trans_handle *trans,
1116                      struct btrfs_root *root,
1117                      struct btrfs_device *device)
1118 {
1119         int ret;
1120         struct btrfs_path *path;
1121         struct btrfs_dev_item *dev_item;
1122         struct extent_buffer *leaf;
1123         struct btrfs_key key;
1124         unsigned long ptr;
1125
1126         root = root->fs_info->chunk_root;
1127
1128         path = btrfs_alloc_path();
1129         if (!path)
1130                 return -ENOMEM;
1131
1132         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1133         key.type = BTRFS_DEV_ITEM_KEY;
1134         key.offset = device->devid;
1135
1136         ret = btrfs_insert_empty_item(trans, root, path, &key,
1137                                       sizeof(*dev_item));
1138         if (ret)
1139                 goto out;
1140
1141         leaf = path->nodes[0];
1142         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1143
1144         btrfs_set_device_id(leaf, dev_item, device->devid);
1145         btrfs_set_device_generation(leaf, dev_item, 0);
1146         btrfs_set_device_type(leaf, dev_item, device->type);
1147         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1148         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1149         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1150         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1151         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1152         btrfs_set_device_group(leaf, dev_item, 0);
1153         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1154         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1155         btrfs_set_device_start_offset(leaf, dev_item, 0);
1156
1157         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1158         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1159         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1160         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1161         btrfs_mark_buffer_dirty(leaf);
1162
1163         ret = 0;
1164 out:
1165         btrfs_free_path(path);
1166         return ret;
1167 }
1168
1169 static int btrfs_rm_dev_item(struct btrfs_root *root,
1170                              struct btrfs_device *device)
1171 {
1172         int ret;
1173         struct btrfs_path *path;
1174         struct btrfs_key key;
1175         struct btrfs_trans_handle *trans;
1176
1177         root = root->fs_info->chunk_root;
1178
1179         path = btrfs_alloc_path();
1180         if (!path)
1181                 return -ENOMEM;
1182
1183         trans = btrfs_start_transaction(root, 0);
1184         if (IS_ERR(trans)) {
1185                 btrfs_free_path(path);
1186                 return PTR_ERR(trans);
1187         }
1188         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1189         key.type = BTRFS_DEV_ITEM_KEY;
1190         key.offset = device->devid;
1191         lock_chunks(root);
1192
1193         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1194         if (ret < 0)
1195                 goto out;
1196
1197         if (ret > 0) {
1198                 ret = -ENOENT;
1199                 goto out;
1200         }
1201
1202         ret = btrfs_del_item(trans, root, path);
1203         if (ret)
1204                 goto out;
1205 out:
1206         btrfs_free_path(path);
1207         unlock_chunks(root);
1208         btrfs_commit_transaction(trans, root);
1209         return ret;
1210 }
1211
1212 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1213 {
1214         struct btrfs_device *device;
1215         struct btrfs_device *next_device;
1216         struct block_device *bdev;
1217         struct buffer_head *bh = NULL;
1218         struct btrfs_super_block *disk_super;
1219         struct btrfs_fs_devices *cur_devices;
1220         u64 all_avail;
1221         u64 devid;
1222         u64 num_devices;
1223         u8 *dev_uuid;
1224         int ret = 0;
1225         bool clear_super = false;
1226
1227         mutex_lock(&uuid_mutex);
1228         mutex_lock(&root->fs_info->volume_mutex);
1229
1230         all_avail = root->fs_info->avail_data_alloc_bits |
1231                 root->fs_info->avail_system_alloc_bits |
1232                 root->fs_info->avail_metadata_alloc_bits;
1233
1234         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1235             root->fs_info->fs_devices->num_devices <= 4) {
1236                 printk(KERN_ERR "btrfs: unable to go below four devices "
1237                        "on raid10\n");
1238                 ret = -EINVAL;
1239                 goto out;
1240         }
1241
1242         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1243             root->fs_info->fs_devices->num_devices <= 2) {
1244                 printk(KERN_ERR "btrfs: unable to go below two "
1245                        "devices on raid1\n");
1246                 ret = -EINVAL;
1247                 goto out;
1248         }
1249
1250         if (strcmp(device_path, "missing") == 0) {
1251                 struct list_head *devices;
1252                 struct btrfs_device *tmp;
1253
1254                 device = NULL;
1255                 devices = &root->fs_info->fs_devices->devices;
1256                 /*
1257                  * It is safe to read the devices since the volume_mutex
1258                  * is held.
1259                  */
1260                 list_for_each_entry(tmp, devices, dev_list) {
1261                         if (tmp->in_fs_metadata && !tmp->bdev) {
1262                                 device = tmp;
1263                                 break;
1264                         }
1265                 }
1266                 bdev = NULL;
1267                 bh = NULL;
1268                 disk_super = NULL;
1269                 if (!device) {
1270                         printk(KERN_ERR "btrfs: no missing devices found to "
1271                                "remove\n");
1272                         goto out;
1273                 }
1274         } else {
1275                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1276                                           root->fs_info->bdev_holder);
1277                 if (IS_ERR(bdev)) {
1278                         ret = PTR_ERR(bdev);
1279                         goto out;
1280                 }
1281
1282                 set_blocksize(bdev, 4096);
1283                 bh = btrfs_read_dev_super(bdev);
1284                 if (!bh) {
1285                         ret = -EINVAL;
1286                         goto error_close;
1287                 }
1288                 disk_super = (struct btrfs_super_block *)bh->b_data;
1289                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1290                 dev_uuid = disk_super->dev_item.uuid;
1291                 device = btrfs_find_device(root, devid, dev_uuid,
1292                                            disk_super->fsid);
1293                 if (!device) {
1294                         ret = -ENOENT;
1295                         goto error_brelse;
1296                 }
1297         }
1298
1299         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1300                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1301                        "device\n");
1302                 ret = -EINVAL;
1303                 goto error_brelse;
1304         }
1305
1306         if (device->writeable) {
1307                 lock_chunks(root);
1308                 list_del_init(&device->dev_alloc_list);
1309                 unlock_chunks(root);
1310                 root->fs_info->fs_devices->rw_devices--;
1311                 clear_super = true;
1312         }
1313
1314         ret = btrfs_shrink_device(device, 0);
1315         if (ret)
1316                 goto error_undo;
1317
1318         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1319         if (ret)
1320                 goto error_undo;
1321
1322         device->in_fs_metadata = 0;
1323         btrfs_scrub_cancel_dev(root, device);
1324
1325         /*
1326          * the device list mutex makes sure that we don't change
1327          * the device list while someone else is writing out all
1328          * the device supers.
1329          */
1330
1331         cur_devices = device->fs_devices;
1332         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1333         list_del_rcu(&device->dev_list);
1334
1335         device->fs_devices->num_devices--;
1336
1337         if (device->missing)
1338                 root->fs_info->fs_devices->missing_devices--;
1339
1340         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1341                                  struct btrfs_device, dev_list);
1342         if (device->bdev == root->fs_info->sb->s_bdev)
1343                 root->fs_info->sb->s_bdev = next_device->bdev;
1344         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1345                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1346
1347         if (device->bdev)
1348                 device->fs_devices->open_devices--;
1349
1350         call_rcu(&device->rcu, free_device);
1351         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1352
1353         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1354         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1355
1356         if (cur_devices->open_devices == 0) {
1357                 struct btrfs_fs_devices *fs_devices;
1358                 fs_devices = root->fs_info->fs_devices;
1359                 while (fs_devices) {
1360                         if (fs_devices->seed == cur_devices)
1361                                 break;
1362                         fs_devices = fs_devices->seed;
1363                 }
1364                 fs_devices->seed = cur_devices->seed;
1365                 cur_devices->seed = NULL;
1366                 lock_chunks(root);
1367                 __btrfs_close_devices(cur_devices);
1368                 unlock_chunks(root);
1369                 free_fs_devices(cur_devices);
1370         }
1371
1372         /*
1373          * at this point, the device is zero sized.  We want to
1374          * remove it from the devices list and zero out the old super
1375          */
1376         if (clear_super) {
1377                 /* make sure this device isn't detected as part of
1378                  * the FS anymore
1379                  */
1380                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1381                 set_buffer_dirty(bh);
1382                 sync_dirty_buffer(bh);
1383         }
1384
1385         ret = 0;
1386
1387 error_brelse:
1388         brelse(bh);
1389 error_close:
1390         if (bdev)
1391                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1392 out:
1393         mutex_unlock(&root->fs_info->volume_mutex);
1394         mutex_unlock(&uuid_mutex);
1395         return ret;
1396 error_undo:
1397         if (device->writeable) {
1398                 lock_chunks(root);
1399                 list_add(&device->dev_alloc_list,
1400                          &root->fs_info->fs_devices->alloc_list);
1401                 unlock_chunks(root);
1402                 root->fs_info->fs_devices->rw_devices++;
1403         }
1404         goto error_brelse;
1405 }
1406
1407 /*
1408  * does all the dirty work required for changing file system's UUID.
1409  */
1410 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1411                                 struct btrfs_root *root)
1412 {
1413         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1414         struct btrfs_fs_devices *old_devices;
1415         struct btrfs_fs_devices *seed_devices;
1416         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1417         struct btrfs_device *device;
1418         u64 super_flags;
1419
1420         BUG_ON(!mutex_is_locked(&uuid_mutex));
1421         if (!fs_devices->seeding)
1422                 return -EINVAL;
1423
1424         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1425         if (!seed_devices)
1426                 return -ENOMEM;
1427
1428         old_devices = clone_fs_devices(fs_devices);
1429         if (IS_ERR(old_devices)) {
1430                 kfree(seed_devices);
1431                 return PTR_ERR(old_devices);
1432         }
1433
1434         list_add(&old_devices->list, &fs_uuids);
1435
1436         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1437         seed_devices->opened = 1;
1438         INIT_LIST_HEAD(&seed_devices->devices);
1439         INIT_LIST_HEAD(&seed_devices->alloc_list);
1440         mutex_init(&seed_devices->device_list_mutex);
1441
1442         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1443         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1444                               synchronize_rcu);
1445         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1446
1447         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1448         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1449                 device->fs_devices = seed_devices;
1450         }
1451
1452         fs_devices->seeding = 0;
1453         fs_devices->num_devices = 0;
1454         fs_devices->open_devices = 0;
1455         fs_devices->seed = seed_devices;
1456
1457         generate_random_uuid(fs_devices->fsid);
1458         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1459         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1460         super_flags = btrfs_super_flags(disk_super) &
1461                       ~BTRFS_SUPER_FLAG_SEEDING;
1462         btrfs_set_super_flags(disk_super, super_flags);
1463
1464         return 0;
1465 }
1466
1467 /*
1468  * strore the expected generation for seed devices in device items.
1469  */
1470 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1471                                struct btrfs_root *root)
1472 {
1473         struct btrfs_path *path;
1474         struct extent_buffer *leaf;
1475         struct btrfs_dev_item *dev_item;
1476         struct btrfs_device *device;
1477         struct btrfs_key key;
1478         u8 fs_uuid[BTRFS_UUID_SIZE];
1479         u8 dev_uuid[BTRFS_UUID_SIZE];
1480         u64 devid;
1481         int ret;
1482
1483         path = btrfs_alloc_path();
1484         if (!path)
1485                 return -ENOMEM;
1486
1487         root = root->fs_info->chunk_root;
1488         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1489         key.offset = 0;
1490         key.type = BTRFS_DEV_ITEM_KEY;
1491
1492         while (1) {
1493                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1494                 if (ret < 0)
1495                         goto error;
1496
1497                 leaf = path->nodes[0];
1498 next_slot:
1499                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1500                         ret = btrfs_next_leaf(root, path);
1501                         if (ret > 0)
1502                                 break;
1503                         if (ret < 0)
1504                                 goto error;
1505                         leaf = path->nodes[0];
1506                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1507                         btrfs_release_path(path);
1508                         continue;
1509                 }
1510
1511                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1512                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1513                     key.type != BTRFS_DEV_ITEM_KEY)
1514                         break;
1515
1516                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1517                                           struct btrfs_dev_item);
1518                 devid = btrfs_device_id(leaf, dev_item);
1519                 read_extent_buffer(leaf, dev_uuid,
1520                                    (unsigned long)btrfs_device_uuid(dev_item),
1521                                    BTRFS_UUID_SIZE);
1522                 read_extent_buffer(leaf, fs_uuid,
1523                                    (unsigned long)btrfs_device_fsid(dev_item),
1524                                    BTRFS_UUID_SIZE);
1525                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1526                 BUG_ON(!device);
1527
1528                 if (device->fs_devices->seeding) {
1529                         btrfs_set_device_generation(leaf, dev_item,
1530                                                     device->generation);
1531                         btrfs_mark_buffer_dirty(leaf);
1532                 }
1533
1534                 path->slots[0]++;
1535                 goto next_slot;
1536         }
1537         ret = 0;
1538 error:
1539         btrfs_free_path(path);
1540         return ret;
1541 }
1542
1543 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1544 {
1545         struct btrfs_trans_handle *trans;
1546         struct btrfs_device *device;
1547         struct block_device *bdev;
1548         struct list_head *devices;
1549         struct super_block *sb = root->fs_info->sb;
1550         u64 total_bytes;
1551         int seeding_dev = 0;
1552         int ret = 0;
1553
1554         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1555                 return -EINVAL;
1556
1557         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1558                                   root->fs_info->bdev_holder);
1559         if (IS_ERR(bdev))
1560                 return PTR_ERR(bdev);
1561
1562         if (root->fs_info->fs_devices->seeding) {
1563                 seeding_dev = 1;
1564                 down_write(&sb->s_umount);
1565                 mutex_lock(&uuid_mutex);
1566         }
1567
1568         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1569         mutex_lock(&root->fs_info->volume_mutex);
1570
1571         devices = &root->fs_info->fs_devices->devices;
1572         /*
1573          * we have the volume lock, so we don't need the extra
1574          * device list mutex while reading the list here.
1575          */
1576         list_for_each_entry(device, devices, dev_list) {
1577                 if (device->bdev == bdev) {
1578                         ret = -EEXIST;
1579                         goto error;
1580                 }
1581         }
1582
1583         device = kzalloc(sizeof(*device), GFP_NOFS);
1584         if (!device) {
1585                 /* we can safely leave the fs_devices entry around */
1586                 ret = -ENOMEM;
1587                 goto error;
1588         }
1589
1590         device->name = kstrdup(device_path, GFP_NOFS);
1591         if (!device->name) {
1592                 kfree(device);
1593                 ret = -ENOMEM;
1594                 goto error;
1595         }
1596
1597         ret = find_next_devid(root, &device->devid);
1598         if (ret) {
1599                 kfree(device->name);
1600                 kfree(device);
1601                 goto error;
1602         }
1603
1604         trans = btrfs_start_transaction(root, 0);
1605         if (IS_ERR(trans)) {
1606                 kfree(device->name);
1607                 kfree(device);
1608                 ret = PTR_ERR(trans);
1609                 goto error;
1610         }
1611
1612         lock_chunks(root);
1613
1614         device->writeable = 1;
1615         device->work.func = pending_bios_fn;
1616         generate_random_uuid(device->uuid);
1617         spin_lock_init(&device->io_lock);
1618         device->generation = trans->transid;
1619         device->io_width = root->sectorsize;
1620         device->io_align = root->sectorsize;
1621         device->sector_size = root->sectorsize;
1622         device->total_bytes = i_size_read(bdev->bd_inode);
1623         device->disk_total_bytes = device->total_bytes;
1624         device->dev_root = root->fs_info->dev_root;
1625         device->bdev = bdev;
1626         device->in_fs_metadata = 1;
1627         device->mode = FMODE_EXCL;
1628         set_blocksize(device->bdev, 4096);
1629
1630         if (seeding_dev) {
1631                 sb->s_flags &= ~MS_RDONLY;
1632                 ret = btrfs_prepare_sprout(trans, root);
1633                 BUG_ON(ret);
1634         }
1635
1636         device->fs_devices = root->fs_info->fs_devices;
1637
1638         /*
1639          * we don't want write_supers to jump in here with our device
1640          * half setup
1641          */
1642         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1643         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1644         list_add(&device->dev_alloc_list,
1645                  &root->fs_info->fs_devices->alloc_list);
1646         root->fs_info->fs_devices->num_devices++;
1647         root->fs_info->fs_devices->open_devices++;
1648         root->fs_info->fs_devices->rw_devices++;
1649         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1650
1651         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1652                 root->fs_info->fs_devices->rotating = 1;
1653
1654         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1655         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1656                                     total_bytes + device->total_bytes);
1657
1658         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1659         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1660                                     total_bytes + 1);
1661         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1662
1663         if (seeding_dev) {
1664                 ret = init_first_rw_device(trans, root, device);
1665                 BUG_ON(ret);
1666                 ret = btrfs_finish_sprout(trans, root);
1667                 BUG_ON(ret);
1668         } else {
1669                 ret = btrfs_add_device(trans, root, device);
1670         }
1671
1672         /*
1673          * we've got more storage, clear any full flags on the space
1674          * infos
1675          */
1676         btrfs_clear_space_info_full(root->fs_info);
1677
1678         unlock_chunks(root);
1679         btrfs_commit_transaction(trans, root);
1680
1681         if (seeding_dev) {
1682                 mutex_unlock(&uuid_mutex);
1683                 up_write(&sb->s_umount);
1684
1685                 ret = btrfs_relocate_sys_chunks(root);
1686                 BUG_ON(ret);
1687         }
1688 out:
1689         mutex_unlock(&root->fs_info->volume_mutex);
1690         return ret;
1691 error:
1692         blkdev_put(bdev, FMODE_EXCL);
1693         if (seeding_dev) {
1694                 mutex_unlock(&uuid_mutex);
1695                 up_write(&sb->s_umount);
1696         }
1697         goto out;
1698 }
1699
1700 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1701                                         struct btrfs_device *device)
1702 {
1703         int ret;
1704         struct btrfs_path *path;
1705         struct btrfs_root *root;
1706         struct btrfs_dev_item *dev_item;
1707         struct extent_buffer *leaf;
1708         struct btrfs_key key;
1709
1710         root = device->dev_root->fs_info->chunk_root;
1711
1712         path = btrfs_alloc_path();
1713         if (!path)
1714                 return -ENOMEM;
1715
1716         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1717         key.type = BTRFS_DEV_ITEM_KEY;
1718         key.offset = device->devid;
1719
1720         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1721         if (ret < 0)
1722                 goto out;
1723
1724         if (ret > 0) {
1725                 ret = -ENOENT;
1726                 goto out;
1727         }
1728
1729         leaf = path->nodes[0];
1730         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1731
1732         btrfs_set_device_id(leaf, dev_item, device->devid);
1733         btrfs_set_device_type(leaf, dev_item, device->type);
1734         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1735         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1736         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1737         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1738         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1739         btrfs_mark_buffer_dirty(leaf);
1740
1741 out:
1742         btrfs_free_path(path);
1743         return ret;
1744 }
1745
1746 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1747                       struct btrfs_device *device, u64 new_size)
1748 {
1749         struct btrfs_super_block *super_copy =
1750                 &device->dev_root->fs_info->super_copy;
1751         u64 old_total = btrfs_super_total_bytes(super_copy);
1752         u64 diff = new_size - device->total_bytes;
1753
1754         if (!device->writeable)
1755                 return -EACCES;
1756         if (new_size <= device->total_bytes)
1757                 return -EINVAL;
1758
1759         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1760         device->fs_devices->total_rw_bytes += diff;
1761
1762         device->total_bytes = new_size;
1763         device->disk_total_bytes = new_size;
1764         btrfs_clear_space_info_full(device->dev_root->fs_info);
1765
1766         return btrfs_update_device(trans, device);
1767 }
1768
1769 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1770                       struct btrfs_device *device, u64 new_size)
1771 {
1772         int ret;
1773         lock_chunks(device->dev_root);
1774         ret = __btrfs_grow_device(trans, device, new_size);
1775         unlock_chunks(device->dev_root);
1776         return ret;
1777 }
1778
1779 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1780                             struct btrfs_root *root,
1781                             u64 chunk_tree, u64 chunk_objectid,
1782                             u64 chunk_offset)
1783 {
1784         int ret;
1785         struct btrfs_path *path;
1786         struct btrfs_key key;
1787
1788         root = root->fs_info->chunk_root;
1789         path = btrfs_alloc_path();
1790         if (!path)
1791                 return -ENOMEM;
1792
1793         key.objectid = chunk_objectid;
1794         key.offset = chunk_offset;
1795         key.type = BTRFS_CHUNK_ITEM_KEY;
1796
1797         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1798         BUG_ON(ret);
1799
1800         ret = btrfs_del_item(trans, root, path);
1801
1802         btrfs_free_path(path);
1803         return ret;
1804 }
1805
1806 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1807                         chunk_offset)
1808 {
1809         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1810         struct btrfs_disk_key *disk_key;
1811         struct btrfs_chunk *chunk;
1812         u8 *ptr;
1813         int ret = 0;
1814         u32 num_stripes;
1815         u32 array_size;
1816         u32 len = 0;
1817         u32 cur;
1818         struct btrfs_key key;
1819
1820         array_size = btrfs_super_sys_array_size(super_copy);
1821
1822         ptr = super_copy->sys_chunk_array;
1823         cur = 0;
1824
1825         while (cur < array_size) {
1826                 disk_key = (struct btrfs_disk_key *)ptr;
1827                 btrfs_disk_key_to_cpu(&key, disk_key);
1828
1829                 len = sizeof(*disk_key);
1830
1831                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1832                         chunk = (struct btrfs_chunk *)(ptr + len);
1833                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1834                         len += btrfs_chunk_item_size(num_stripes);
1835                 } else {
1836                         ret = -EIO;
1837                         break;
1838                 }
1839                 if (key.objectid == chunk_objectid &&
1840                     key.offset == chunk_offset) {
1841                         memmove(ptr, ptr + len, array_size - (cur + len));
1842                         array_size -= len;
1843                         btrfs_set_super_sys_array_size(super_copy, array_size);
1844                 } else {
1845                         ptr += len;
1846                         cur += len;
1847                 }
1848         }
1849         return ret;
1850 }
1851
1852 static int btrfs_relocate_chunk(struct btrfs_root *root,
1853                          u64 chunk_tree, u64 chunk_objectid,
1854                          u64 chunk_offset)
1855 {
1856         struct extent_map_tree *em_tree;
1857         struct btrfs_root *extent_root;
1858         struct btrfs_trans_handle *trans;
1859         struct extent_map *em;
1860         struct map_lookup *map;
1861         int ret;
1862         int i;
1863
1864         root = root->fs_info->chunk_root;
1865         extent_root = root->fs_info->extent_root;
1866         em_tree = &root->fs_info->mapping_tree.map_tree;
1867
1868         ret = btrfs_can_relocate(extent_root, chunk_offset);
1869         if (ret)
1870                 return -ENOSPC;
1871
1872         /* step one, relocate all the extents inside this chunk */
1873         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1874         if (ret)
1875                 return ret;
1876
1877         trans = btrfs_start_transaction(root, 0);
1878         BUG_ON(IS_ERR(trans));
1879
1880         lock_chunks(root);
1881
1882         /*
1883          * step two, delete the device extents and the
1884          * chunk tree entries
1885          */
1886         read_lock(&em_tree->lock);
1887         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1888         read_unlock(&em_tree->lock);
1889
1890         BUG_ON(em->start > chunk_offset ||
1891                em->start + em->len < chunk_offset);
1892         map = (struct map_lookup *)em->bdev;
1893
1894         for (i = 0; i < map->num_stripes; i++) {
1895                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1896                                             map->stripes[i].physical);
1897                 BUG_ON(ret);
1898
1899                 if (map->stripes[i].dev) {
1900                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1901                         BUG_ON(ret);
1902                 }
1903         }
1904         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1905                                chunk_offset);
1906
1907         BUG_ON(ret);
1908
1909         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1910
1911         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1912                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1913                 BUG_ON(ret);
1914         }
1915
1916         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1917         BUG_ON(ret);
1918
1919         write_lock(&em_tree->lock);
1920         remove_extent_mapping(em_tree, em);
1921         write_unlock(&em_tree->lock);
1922
1923         kfree(map);
1924         em->bdev = NULL;
1925
1926         /* once for the tree */
1927         free_extent_map(em);
1928         /* once for us */
1929         free_extent_map(em);
1930
1931         unlock_chunks(root);
1932         btrfs_end_transaction(trans, root);
1933         return 0;
1934 }
1935
1936 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1937 {
1938         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1939         struct btrfs_path *path;
1940         struct extent_buffer *leaf;
1941         struct btrfs_chunk *chunk;
1942         struct btrfs_key key;
1943         struct btrfs_key found_key;
1944         u64 chunk_tree = chunk_root->root_key.objectid;
1945         u64 chunk_type;
1946         bool retried = false;
1947         int failed = 0;
1948         int ret;
1949
1950         path = btrfs_alloc_path();
1951         if (!path)
1952                 return -ENOMEM;
1953
1954 again:
1955         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1956         key.offset = (u64)-1;
1957         key.type = BTRFS_CHUNK_ITEM_KEY;
1958
1959         while (1) {
1960                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1961                 if (ret < 0)
1962                         goto error;
1963                 BUG_ON(ret == 0);
1964
1965                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1966                                           key.type);
1967                 if (ret < 0)
1968                         goto error;
1969                 if (ret > 0)
1970                         break;
1971
1972                 leaf = path->nodes[0];
1973                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1974
1975                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1976                                        struct btrfs_chunk);
1977                 chunk_type = btrfs_chunk_type(leaf, chunk);
1978                 btrfs_release_path(path);
1979
1980                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1981                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1982                                                    found_key.objectid,
1983                                                    found_key.offset);
1984                         if (ret == -ENOSPC)
1985                                 failed++;
1986                         else if (ret)
1987                                 BUG();
1988                 }
1989
1990                 if (found_key.offset == 0)
1991                         break;
1992                 key.offset = found_key.offset - 1;
1993         }
1994         ret = 0;
1995         if (failed && !retried) {
1996                 failed = 0;
1997                 retried = true;
1998                 goto again;
1999         } else if (failed && retried) {
2000                 WARN_ON(1);
2001                 ret = -ENOSPC;
2002         }
2003 error:
2004         btrfs_free_path(path);
2005         return ret;
2006 }
2007
2008 static u64 div_factor(u64 num, int factor)
2009 {
2010         if (factor == 10)
2011                 return num;
2012         num *= factor;
2013         do_div(num, 10);
2014         return num;
2015 }
2016
2017 int btrfs_balance(struct btrfs_root *dev_root)
2018 {
2019         int ret;
2020         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2021         struct btrfs_device *device;
2022         u64 old_size;
2023         u64 size_to_free;
2024         struct btrfs_path *path;
2025         struct btrfs_key key;
2026         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2027         struct btrfs_trans_handle *trans;
2028         struct btrfs_key found_key;
2029
2030         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2031                 return -EROFS;
2032
2033         if (!capable(CAP_SYS_ADMIN))
2034                 return -EPERM;
2035
2036         mutex_lock(&dev_root->fs_info->volume_mutex);
2037         dev_root = dev_root->fs_info->dev_root;
2038
2039         /* step one make some room on all the devices */
2040         list_for_each_entry(device, devices, dev_list) {
2041                 old_size = device->total_bytes;
2042                 size_to_free = div_factor(old_size, 1);
2043                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2044                 if (!device->writeable ||
2045                     device->total_bytes - device->bytes_used > size_to_free)
2046                         continue;
2047
2048                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2049                 if (ret == -ENOSPC)
2050                         break;
2051                 BUG_ON(ret);
2052
2053                 trans = btrfs_start_transaction(dev_root, 0);
2054                 BUG_ON(IS_ERR(trans));
2055
2056                 ret = btrfs_grow_device(trans, device, old_size);
2057                 BUG_ON(ret);
2058
2059                 btrfs_end_transaction(trans, dev_root);
2060         }
2061
2062         /* step two, relocate all the chunks */
2063         path = btrfs_alloc_path();
2064         BUG_ON(!path);
2065
2066         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2067         key.offset = (u64)-1;
2068         key.type = BTRFS_CHUNK_ITEM_KEY;
2069
2070         while (1) {
2071                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2072                 if (ret < 0)
2073                         goto error;
2074
2075                 /*
2076                  * this shouldn't happen, it means the last relocate
2077                  * failed
2078                  */
2079                 if (ret == 0)
2080                         break;
2081
2082                 ret = btrfs_previous_item(chunk_root, path, 0,
2083                                           BTRFS_CHUNK_ITEM_KEY);
2084                 if (ret)
2085                         break;
2086
2087                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2088                                       path->slots[0]);
2089                 if (found_key.objectid != key.objectid)
2090                         break;
2091
2092                 /* chunk zero is special */
2093                 if (found_key.offset == 0)
2094                         break;
2095
2096                 btrfs_release_path(path);
2097                 ret = btrfs_relocate_chunk(chunk_root,
2098                                            chunk_root->root_key.objectid,
2099                                            found_key.objectid,
2100                                            found_key.offset);
2101                 BUG_ON(ret && ret != -ENOSPC);
2102                 key.offset = found_key.offset - 1;
2103         }
2104         ret = 0;
2105 error:
2106         btrfs_free_path(path);
2107         mutex_unlock(&dev_root->fs_info->volume_mutex);
2108         return ret;
2109 }
2110
2111 /*
2112  * shrinking a device means finding all of the device extents past
2113  * the new size, and then following the back refs to the chunks.
2114  * The chunk relocation code actually frees the device extent
2115  */
2116 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2117 {
2118         struct btrfs_trans_handle *trans;
2119         struct btrfs_root *root = device->dev_root;
2120         struct btrfs_dev_extent *dev_extent = NULL;
2121         struct btrfs_path *path;
2122         u64 length;
2123         u64 chunk_tree;
2124         u64 chunk_objectid;
2125         u64 chunk_offset;
2126         int ret;
2127         int slot;
2128         int failed = 0;
2129         bool retried = false;
2130         struct extent_buffer *l;
2131         struct btrfs_key key;
2132         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2133         u64 old_total = btrfs_super_total_bytes(super_copy);
2134         u64 old_size = device->total_bytes;
2135         u64 diff = device->total_bytes - new_size;
2136
2137         if (new_size >= device->total_bytes)
2138                 return -EINVAL;
2139
2140         path = btrfs_alloc_path();
2141         if (!path)
2142                 return -ENOMEM;
2143
2144         path->reada = 2;
2145
2146         lock_chunks(root);
2147
2148         device->total_bytes = new_size;
2149         if (device->writeable)
2150                 device->fs_devices->total_rw_bytes -= diff;
2151         unlock_chunks(root);
2152
2153 again:
2154         key.objectid = device->devid;
2155         key.offset = (u64)-1;
2156         key.type = BTRFS_DEV_EXTENT_KEY;
2157
2158         while (1) {
2159                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2160                 if (ret < 0)
2161                         goto done;
2162
2163                 ret = btrfs_previous_item(root, path, 0, key.type);
2164                 if (ret < 0)
2165                         goto done;
2166                 if (ret) {
2167                         ret = 0;
2168                         btrfs_release_path(path);
2169                         break;
2170                 }
2171
2172                 l = path->nodes[0];
2173                 slot = path->slots[0];
2174                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2175
2176                 if (key.objectid != device->devid) {
2177                         btrfs_release_path(path);
2178                         break;
2179                 }
2180
2181                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2182                 length = btrfs_dev_extent_length(l, dev_extent);
2183
2184                 if (key.offset + length <= new_size) {
2185                         btrfs_release_path(path);
2186                         break;
2187                 }
2188
2189                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2190                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2191                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2192                 btrfs_release_path(path);
2193
2194                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2195                                            chunk_offset);
2196                 if (ret && ret != -ENOSPC)
2197                         goto done;
2198                 if (ret == -ENOSPC)
2199                         failed++;
2200                 key.offset -= 1;
2201         }
2202
2203         if (failed && !retried) {
2204                 failed = 0;
2205                 retried = true;
2206                 goto again;
2207         } else if (failed && retried) {
2208                 ret = -ENOSPC;
2209                 lock_chunks(root);
2210
2211                 device->total_bytes = old_size;
2212                 if (device->writeable)
2213                         device->fs_devices->total_rw_bytes += diff;
2214                 unlock_chunks(root);
2215                 goto done;
2216         }
2217
2218         /* Shrinking succeeded, else we would be at "done". */
2219         trans = btrfs_start_transaction(root, 0);
2220         if (IS_ERR(trans)) {
2221                 ret = PTR_ERR(trans);
2222                 goto done;
2223         }
2224
2225         lock_chunks(root);
2226
2227         device->disk_total_bytes = new_size;
2228         /* Now btrfs_update_device() will change the on-disk size. */
2229         ret = btrfs_update_device(trans, device);
2230         if (ret) {
2231                 unlock_chunks(root);
2232                 btrfs_end_transaction(trans, root);
2233                 goto done;
2234         }
2235         WARN_ON(diff > old_total);
2236         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2237         unlock_chunks(root);
2238         btrfs_end_transaction(trans, root);
2239 done:
2240         btrfs_free_path(path);
2241         return ret;
2242 }
2243
2244 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2245                            struct btrfs_root *root,
2246                            struct btrfs_key *key,
2247                            struct btrfs_chunk *chunk, int item_size)
2248 {
2249         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2250         struct btrfs_disk_key disk_key;
2251         u32 array_size;
2252         u8 *ptr;
2253
2254         array_size = btrfs_super_sys_array_size(super_copy);
2255         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2256                 return -EFBIG;
2257
2258         ptr = super_copy->sys_chunk_array + array_size;
2259         btrfs_cpu_key_to_disk(&disk_key, key);
2260         memcpy(ptr, &disk_key, sizeof(disk_key));
2261         ptr += sizeof(disk_key);
2262         memcpy(ptr, chunk, item_size);
2263         item_size += sizeof(disk_key);
2264         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2265         return 0;
2266 }
2267
2268 /*
2269  * sort the devices in descending order by max_avail, total_avail
2270  */
2271 static int btrfs_cmp_device_info(const void *a, const void *b)
2272 {
2273         const struct btrfs_device_info *di_a = a;
2274         const struct btrfs_device_info *di_b = b;
2275
2276         if (di_a->max_avail > di_b->max_avail)
2277                 return -1;
2278         if (di_a->max_avail < di_b->max_avail)
2279                 return 1;
2280         if (di_a->total_avail > di_b->total_avail)
2281                 return -1;
2282         if (di_a->total_avail < di_b->total_avail)
2283                 return 1;
2284         return 0;
2285 }
2286
2287 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2288                                struct btrfs_root *extent_root,
2289                                struct map_lookup **map_ret,
2290                                u64 *num_bytes_out, u64 *stripe_size_out,
2291                                u64 start, u64 type)
2292 {
2293         struct btrfs_fs_info *info = extent_root->fs_info;
2294         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2295         struct list_head *cur;
2296         struct map_lookup *map = NULL;
2297         struct extent_map_tree *em_tree;
2298         struct extent_map *em;
2299         struct btrfs_device_info *devices_info = NULL;
2300         u64 total_avail;
2301         int num_stripes;        /* total number of stripes to allocate */
2302         int sub_stripes;        /* sub_stripes info for map */
2303         int dev_stripes;        /* stripes per dev */
2304         int devs_max;           /* max devs to use */
2305         int devs_min;           /* min devs needed */
2306         int devs_increment;     /* ndevs has to be a multiple of this */
2307         int ncopies;            /* how many copies to data has */
2308         int ret;
2309         u64 max_stripe_size;
2310         u64 max_chunk_size;
2311         u64 stripe_size;
2312         u64 num_bytes;
2313         int ndevs;
2314         int i;
2315         int j;
2316
2317         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2318             (type & BTRFS_BLOCK_GROUP_DUP)) {
2319                 WARN_ON(1);
2320                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2321         }
2322
2323         if (list_empty(&fs_devices->alloc_list))
2324                 return -ENOSPC;
2325
2326         sub_stripes = 1;
2327         dev_stripes = 1;
2328         devs_increment = 1;
2329         ncopies = 1;
2330         devs_max = 0;   /* 0 == as many as possible */
2331         devs_min = 1;
2332
2333         /*
2334          * define the properties of each RAID type.
2335          * FIXME: move this to a global table and use it in all RAID
2336          * calculation code
2337          */
2338         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2339                 dev_stripes = 2;
2340                 ncopies = 2;
2341                 devs_max = 1;
2342         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2343                 devs_min = 2;
2344         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2345                 devs_increment = 2;
2346                 ncopies = 2;
2347                 devs_max = 2;
2348                 devs_min = 2;
2349         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2350                 sub_stripes = 2;
2351                 devs_increment = 2;
2352                 ncopies = 2;
2353                 devs_min = 4;
2354         } else {
2355                 devs_max = 1;
2356         }
2357
2358         if (type & BTRFS_BLOCK_GROUP_DATA) {
2359                 max_stripe_size = 1024 * 1024 * 1024;
2360                 max_chunk_size = 10 * max_stripe_size;
2361         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2362                 max_stripe_size = 256 * 1024 * 1024;
2363                 max_chunk_size = max_stripe_size;
2364         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2365                 max_stripe_size = 8 * 1024 * 1024;
2366                 max_chunk_size = 2 * max_stripe_size;
2367         } else {
2368                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2369                        type);
2370                 BUG_ON(1);
2371         }
2372
2373         /* we don't want a chunk larger than 10% of writeable space */
2374         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2375                              max_chunk_size);
2376
2377         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2378                                GFP_NOFS);
2379         if (!devices_info)
2380                 return -ENOMEM;
2381
2382         cur = fs_devices->alloc_list.next;
2383
2384         /*
2385          * in the first pass through the devices list, we gather information
2386          * about the available holes on each device.
2387          */
2388         ndevs = 0;
2389         while (cur != &fs_devices->alloc_list) {
2390                 struct btrfs_device *device;
2391                 u64 max_avail;
2392                 u64 dev_offset;
2393
2394                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2395
2396                 cur = cur->next;
2397
2398                 if (!device->writeable) {
2399                         printk(KERN_ERR
2400                                "btrfs: read-only device in alloc_list\n");
2401                         WARN_ON(1);
2402                         continue;
2403                 }
2404
2405                 if (!device->in_fs_metadata)
2406                         continue;
2407
2408                 if (device->total_bytes > device->bytes_used)
2409                         total_avail = device->total_bytes - device->bytes_used;
2410                 else
2411                         total_avail = 0;
2412                 /* avail is off by max(alloc_start, 1MB), but that is the same
2413                  * for all devices, so it doesn't hurt the sorting later on
2414                  */
2415
2416                 ret = find_free_dev_extent(trans, device,
2417                                            max_stripe_size * dev_stripes,
2418                                            &dev_offset, &max_avail);
2419                 if (ret && ret != -ENOSPC)
2420                         goto error;
2421
2422                 if (ret == 0)
2423                         max_avail = max_stripe_size * dev_stripes;
2424
2425                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2426                         continue;
2427
2428                 devices_info[ndevs].dev_offset = dev_offset;
2429                 devices_info[ndevs].max_avail = max_avail;
2430                 devices_info[ndevs].total_avail = total_avail;
2431                 devices_info[ndevs].dev = device;
2432                 ++ndevs;
2433         }
2434
2435         /*
2436          * now sort the devices by hole size / available space
2437          */
2438         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2439              btrfs_cmp_device_info, NULL);
2440
2441         /* round down to number of usable stripes */
2442         ndevs -= ndevs % devs_increment;
2443
2444         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2445                 ret = -ENOSPC;
2446                 goto error;
2447         }
2448
2449         if (devs_max && ndevs > devs_max)
2450                 ndevs = devs_max;
2451         /*
2452          * the primary goal is to maximize the number of stripes, so use as many
2453          * devices as possible, even if the stripes are not maximum sized.
2454          */
2455         stripe_size = devices_info[ndevs-1].max_avail;
2456         num_stripes = ndevs * dev_stripes;
2457
2458         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2459                 stripe_size = max_chunk_size * ncopies;
2460                 do_div(stripe_size, num_stripes);
2461         }
2462
2463         do_div(stripe_size, dev_stripes);
2464         do_div(stripe_size, BTRFS_STRIPE_LEN);
2465         stripe_size *= BTRFS_STRIPE_LEN;
2466
2467         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2468         if (!map) {
2469                 ret = -ENOMEM;
2470                 goto error;
2471         }
2472         map->num_stripes = num_stripes;
2473
2474         for (i = 0; i < ndevs; ++i) {
2475                 for (j = 0; j < dev_stripes; ++j) {
2476                         int s = i * dev_stripes + j;
2477                         map->stripes[s].dev = devices_info[i].dev;
2478                         map->stripes[s].physical = devices_info[i].dev_offset +
2479                                                    j * stripe_size;
2480                 }
2481         }
2482         map->sector_size = extent_root->sectorsize;
2483         map->stripe_len = BTRFS_STRIPE_LEN;
2484         map->io_align = BTRFS_STRIPE_LEN;
2485         map->io_width = BTRFS_STRIPE_LEN;
2486         map->type = type;
2487         map->sub_stripes = sub_stripes;
2488
2489         *map_ret = map;
2490         num_bytes = stripe_size * (num_stripes / ncopies);
2491
2492         *stripe_size_out = stripe_size;
2493         *num_bytes_out = num_bytes;
2494
2495         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2496
2497         em = alloc_extent_map();
2498         if (!em) {
2499                 ret = -ENOMEM;
2500                 goto error;
2501         }
2502         em->bdev = (struct block_device *)map;
2503         em->start = start;
2504         em->len = num_bytes;
2505         em->block_start = 0;
2506         em->block_len = em->len;
2507
2508         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2509         write_lock(&em_tree->lock);
2510         ret = add_extent_mapping(em_tree, em);
2511         write_unlock(&em_tree->lock);
2512         BUG_ON(ret);
2513         free_extent_map(em);
2514
2515         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2516                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2517                                      start, num_bytes);
2518         BUG_ON(ret);
2519
2520         for (i = 0; i < map->num_stripes; ++i) {
2521                 struct btrfs_device *device;
2522                 u64 dev_offset;
2523
2524                 device = map->stripes[i].dev;
2525                 dev_offset = map->stripes[i].physical;
2526
2527                 ret = btrfs_alloc_dev_extent(trans, device,
2528                                 info->chunk_root->root_key.objectid,
2529                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2530                                 start, dev_offset, stripe_size);
2531                 BUG_ON(ret);
2532         }
2533
2534         kfree(devices_info);
2535         return 0;
2536
2537 error:
2538         kfree(map);
2539         kfree(devices_info);
2540         return ret;
2541 }
2542
2543 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2544                                 struct btrfs_root *extent_root,
2545                                 struct map_lookup *map, u64 chunk_offset,
2546                                 u64 chunk_size, u64 stripe_size)
2547 {
2548         u64 dev_offset;
2549         struct btrfs_key key;
2550         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2551         struct btrfs_device *device;
2552         struct btrfs_chunk *chunk;
2553         struct btrfs_stripe *stripe;
2554         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2555         int index = 0;
2556         int ret;
2557
2558         chunk = kzalloc(item_size, GFP_NOFS);
2559         if (!chunk)
2560                 return -ENOMEM;
2561
2562         index = 0;
2563         while (index < map->num_stripes) {
2564                 device = map->stripes[index].dev;
2565                 device->bytes_used += stripe_size;
2566                 ret = btrfs_update_device(trans, device);
2567                 BUG_ON(ret);
2568                 index++;
2569         }
2570
2571         index = 0;
2572         stripe = &chunk->stripe;
2573         while (index < map->num_stripes) {
2574                 device = map->stripes[index].dev;
2575                 dev_offset = map->stripes[index].physical;
2576
2577                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2578                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2579                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2580                 stripe++;
2581                 index++;
2582         }
2583
2584         btrfs_set_stack_chunk_length(chunk, chunk_size);
2585         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2586         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2587         btrfs_set_stack_chunk_type(chunk, map->type);
2588         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2589         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2590         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2591         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2592         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2593
2594         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2595         key.type = BTRFS_CHUNK_ITEM_KEY;
2596         key.offset = chunk_offset;
2597
2598         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2599         BUG_ON(ret);
2600
2601         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2602                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2603                                              item_size);
2604                 BUG_ON(ret);
2605         }
2606
2607         kfree(chunk);
2608         return 0;
2609 }
2610
2611 /*
2612  * Chunk allocation falls into two parts. The first part does works
2613  * that make the new allocated chunk useable, but not do any operation
2614  * that modifies the chunk tree. The second part does the works that
2615  * require modifying the chunk tree. This division is important for the
2616  * bootstrap process of adding storage to a seed btrfs.
2617  */
2618 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2619                       struct btrfs_root *extent_root, u64 type)
2620 {
2621         u64 chunk_offset;
2622         u64 chunk_size;
2623         u64 stripe_size;
2624         struct map_lookup *map;
2625         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2626         int ret;
2627
2628         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2629                               &chunk_offset);
2630         if (ret)
2631                 return ret;
2632
2633         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2634                                   &stripe_size, chunk_offset, type);
2635         if (ret)
2636                 return ret;
2637
2638         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2639                                    chunk_size, stripe_size);
2640         BUG_ON(ret);
2641         return 0;
2642 }
2643
2644 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2645                                          struct btrfs_root *root,
2646                                          struct btrfs_device *device)
2647 {
2648         u64 chunk_offset;
2649         u64 sys_chunk_offset;
2650         u64 chunk_size;
2651         u64 sys_chunk_size;
2652         u64 stripe_size;
2653         u64 sys_stripe_size;
2654         u64 alloc_profile;
2655         struct map_lookup *map;
2656         struct map_lookup *sys_map;
2657         struct btrfs_fs_info *fs_info = root->fs_info;
2658         struct btrfs_root *extent_root = fs_info->extent_root;
2659         int ret;
2660
2661         ret = find_next_chunk(fs_info->chunk_root,
2662                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2663         BUG_ON(ret);
2664
2665         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2666                         (fs_info->metadata_alloc_profile &
2667                          fs_info->avail_metadata_alloc_bits);
2668         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2669
2670         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2671                                   &stripe_size, chunk_offset, alloc_profile);
2672         BUG_ON(ret);
2673
2674         sys_chunk_offset = chunk_offset + chunk_size;
2675
2676         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2677                         (fs_info->system_alloc_profile &
2678                          fs_info->avail_system_alloc_bits);
2679         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2680
2681         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2682                                   &sys_chunk_size, &sys_stripe_size,
2683                                   sys_chunk_offset, alloc_profile);
2684         BUG_ON(ret);
2685
2686         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2687         BUG_ON(ret);
2688
2689         /*
2690          * Modifying chunk tree needs allocating new blocks from both
2691          * system block group and metadata block group. So we only can
2692          * do operations require modifying the chunk tree after both
2693          * block groups were created.
2694          */
2695         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2696                                    chunk_size, stripe_size);
2697         BUG_ON(ret);
2698
2699         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2700                                    sys_chunk_offset, sys_chunk_size,
2701                                    sys_stripe_size);
2702         BUG_ON(ret);
2703         return 0;
2704 }
2705
2706 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2707 {
2708         struct extent_map *em;
2709         struct map_lookup *map;
2710         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2711         int readonly = 0;
2712         int i;
2713
2714         read_lock(&map_tree->map_tree.lock);
2715         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2716         read_unlock(&map_tree->map_tree.lock);
2717         if (!em)
2718                 return 1;
2719
2720         if (btrfs_test_opt(root, DEGRADED)) {
2721                 free_extent_map(em);
2722                 return 0;
2723         }
2724
2725         map = (struct map_lookup *)em->bdev;
2726         for (i = 0; i < map->num_stripes; i++) {
2727                 if (!map->stripes[i].dev->writeable) {
2728                         readonly = 1;
2729                         break;
2730                 }
2731         }
2732         free_extent_map(em);
2733         return readonly;
2734 }
2735
2736 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2737 {
2738         extent_map_tree_init(&tree->map_tree);
2739 }
2740
2741 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2742 {
2743         struct extent_map *em;
2744
2745         while (1) {
2746                 write_lock(&tree->map_tree.lock);
2747                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2748                 if (em)
2749                         remove_extent_mapping(&tree->map_tree, em);
2750                 write_unlock(&tree->map_tree.lock);
2751                 if (!em)
2752                         break;
2753                 kfree(em->bdev);
2754                 /* once for us */
2755                 free_extent_map(em);
2756                 /* once for the tree */
2757                 free_extent_map(em);
2758         }
2759 }
2760
2761 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2762 {
2763         struct extent_map *em;
2764         struct map_lookup *map;
2765         struct extent_map_tree *em_tree = &map_tree->map_tree;
2766         int ret;
2767
2768         read_lock(&em_tree->lock);
2769         em = lookup_extent_mapping(em_tree, logical, len);
2770         read_unlock(&em_tree->lock);
2771         BUG_ON(!em);
2772
2773         BUG_ON(em->start > logical || em->start + em->len < logical);
2774         map = (struct map_lookup *)em->bdev;
2775         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2776                 ret = map->num_stripes;
2777         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2778                 ret = map->sub_stripes;
2779         else
2780                 ret = 1;
2781         free_extent_map(em);
2782         return ret;
2783 }
2784
2785 static int find_live_mirror(struct map_lookup *map, int first, int num,
2786                             int optimal)
2787 {
2788         int i;
2789         if (map->stripes[optimal].dev->bdev)
2790                 return optimal;
2791         for (i = first; i < first + num; i++) {
2792                 if (map->stripes[i].dev->bdev)
2793                         return i;
2794         }
2795         /* we couldn't find one that doesn't fail.  Just return something
2796          * and the io error handling code will clean up eventually
2797          */
2798         return optimal;
2799 }
2800
2801 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2802                              u64 logical, u64 *length,
2803                              struct btrfs_multi_bio **multi_ret,
2804                              int mirror_num)
2805 {
2806         struct extent_map *em;
2807         struct map_lookup *map;
2808         struct extent_map_tree *em_tree = &map_tree->map_tree;
2809         u64 offset;
2810         u64 stripe_offset;
2811         u64 stripe_end_offset;
2812         u64 stripe_nr;
2813         u64 stripe_nr_orig;
2814         u64 stripe_nr_end;
2815         int stripes_allocated = 8;
2816         int stripes_required = 1;
2817         int stripe_index;
2818         int i;
2819         int num_stripes;
2820         int max_errors = 0;
2821         struct btrfs_multi_bio *multi = NULL;
2822
2823         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2824                 stripes_allocated = 1;
2825 again:
2826         if (multi_ret) {
2827                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2828                                 GFP_NOFS);
2829                 if (!multi)
2830                         return -ENOMEM;
2831
2832                 atomic_set(&multi->error, 0);
2833         }
2834
2835         read_lock(&em_tree->lock);
2836         em = lookup_extent_mapping(em_tree, logical, *length);
2837         read_unlock(&em_tree->lock);
2838
2839         if (!em) {
2840                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2841                        (unsigned long long)logical,
2842                        (unsigned long long)*length);
2843                 BUG();
2844         }
2845
2846         BUG_ON(em->start > logical || em->start + em->len < logical);
2847         map = (struct map_lookup *)em->bdev;
2848         offset = logical - em->start;
2849
2850         if (mirror_num > map->num_stripes)
2851                 mirror_num = 0;
2852
2853         /* if our multi bio struct is too small, back off and try again */
2854         if (rw & REQ_WRITE) {
2855                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2856                                  BTRFS_BLOCK_GROUP_DUP)) {
2857                         stripes_required = map->num_stripes;
2858                         max_errors = 1;
2859                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2860                         stripes_required = map->sub_stripes;
2861                         max_errors = 1;
2862                 }
2863         }
2864         if (rw & REQ_DISCARD) {
2865                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2866                                  BTRFS_BLOCK_GROUP_RAID1 |
2867                                  BTRFS_BLOCK_GROUP_DUP |
2868                                  BTRFS_BLOCK_GROUP_RAID10)) {
2869                         stripes_required = map->num_stripes;
2870                 }
2871         }
2872         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2873             stripes_allocated < stripes_required) {
2874                 stripes_allocated = map->num_stripes;
2875                 free_extent_map(em);
2876                 kfree(multi);
2877                 goto again;
2878         }
2879         stripe_nr = offset;
2880         /*
2881          * stripe_nr counts the total number of stripes we have to stride
2882          * to get to this block
2883          */
2884         do_div(stripe_nr, map->stripe_len);
2885
2886         stripe_offset = stripe_nr * map->stripe_len;
2887         BUG_ON(offset < stripe_offset);
2888
2889         /* stripe_offset is the offset of this block in its stripe*/
2890         stripe_offset = offset - stripe_offset;
2891
2892         if (rw & REQ_DISCARD)
2893                 *length = min_t(u64, em->len - offset, *length);
2894         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2895                               BTRFS_BLOCK_GROUP_RAID1 |
2896                               BTRFS_BLOCK_GROUP_RAID10 |
2897                               BTRFS_BLOCK_GROUP_DUP)) {
2898                 /* we limit the length of each bio to what fits in a stripe */
2899                 *length = min_t(u64, em->len - offset,
2900                                 map->stripe_len - stripe_offset);
2901         } else {
2902                 *length = em->len - offset;
2903         }
2904
2905         if (!multi_ret)
2906                 goto out;
2907
2908         num_stripes = 1;
2909         stripe_index = 0;
2910         stripe_nr_orig = stripe_nr;
2911         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2912                         (~(map->stripe_len - 1));
2913         do_div(stripe_nr_end, map->stripe_len);
2914         stripe_end_offset = stripe_nr_end * map->stripe_len -
2915                             (offset + *length);
2916         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2917                 if (rw & REQ_DISCARD)
2918                         num_stripes = min_t(u64, map->num_stripes,
2919                                             stripe_nr_end - stripe_nr_orig);
2920                 stripe_index = do_div(stripe_nr, map->num_stripes);
2921         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2922                 if (rw & (REQ_WRITE | REQ_DISCARD))
2923                         num_stripes = map->num_stripes;
2924                 else if (mirror_num)
2925                         stripe_index = mirror_num - 1;
2926                 else {
2927                         stripe_index = find_live_mirror(map, 0,
2928                                             map->num_stripes,
2929                                             current->pid % map->num_stripes);
2930                 }
2931
2932         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2933                 if (rw & (REQ_WRITE | REQ_DISCARD))
2934                         num_stripes = map->num_stripes;
2935                 else if (mirror_num)
2936                         stripe_index = mirror_num - 1;
2937
2938         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2939                 int factor = map->num_stripes / map->sub_stripes;
2940
2941                 stripe_index = do_div(stripe_nr, factor);
2942                 stripe_index *= map->sub_stripes;
2943
2944                 if (rw & REQ_WRITE)
2945                         num_stripes = map->sub_stripes;
2946                 else if (rw & REQ_DISCARD)
2947                         num_stripes = min_t(u64, map->sub_stripes *
2948                                             (stripe_nr_end - stripe_nr_orig),
2949                                             map->num_stripes);
2950                 else if (mirror_num)
2951                         stripe_index += mirror_num - 1;
2952                 else {
2953                         stripe_index = find_live_mirror(map, stripe_index,
2954                                               map->sub_stripes, stripe_index +
2955                                               current->pid % map->sub_stripes);
2956                 }
2957         } else {
2958                 /*
2959                  * after this do_div call, stripe_nr is the number of stripes
2960                  * on this device we have to walk to find the data, and
2961                  * stripe_index is the number of our device in the stripe array
2962                  */
2963                 stripe_index = do_div(stripe_nr, map->num_stripes);
2964         }
2965         BUG_ON(stripe_index >= map->num_stripes);
2966
2967         if (rw & REQ_DISCARD) {
2968                 for (i = 0; i < num_stripes; i++) {
2969                         multi->stripes[i].physical =
2970                                 map->stripes[stripe_index].physical +
2971                                 stripe_offset + stripe_nr * map->stripe_len;
2972                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2973
2974                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2975                                 u64 stripes;
2976                                 u32 last_stripe = 0;
2977                                 int j;
2978
2979                                 div_u64_rem(stripe_nr_end - 1,
2980                                             map->num_stripes,
2981                                             &last_stripe);
2982
2983                                 for (j = 0; j < map->num_stripes; j++) {
2984                                         u32 test;
2985
2986                                         div_u64_rem(stripe_nr_end - 1 - j,
2987                                                     map->num_stripes, &test);
2988                                         if (test == stripe_index)
2989                                                 break;
2990                                 }
2991                                 stripes = stripe_nr_end - 1 - j;
2992                                 do_div(stripes, map->num_stripes);
2993                                 multi->stripes[i].length = map->stripe_len *
2994                                         (stripes - stripe_nr + 1);
2995
2996                                 if (i == 0) {
2997                                         multi->stripes[i].length -=
2998                                                 stripe_offset;
2999                                         stripe_offset = 0;
3000                                 }
3001                                 if (stripe_index == last_stripe)
3002                                         multi->stripes[i].length -=
3003                                                 stripe_end_offset;
3004                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3005                                 u64 stripes;
3006                                 int j;
3007                                 int factor = map->num_stripes /
3008                                              map->sub_stripes;
3009                                 u32 last_stripe = 0;
3010
3011                                 div_u64_rem(stripe_nr_end - 1,
3012                                             factor, &last_stripe);
3013                                 last_stripe *= map->sub_stripes;
3014
3015                                 for (j = 0; j < factor; j++) {
3016                                         u32 test;
3017
3018                                         div_u64_rem(stripe_nr_end - 1 - j,
3019                                                     factor, &test);
3020
3021                                         if (test ==
3022                                             stripe_index / map->sub_stripes)
3023                                                 break;
3024                                 }
3025                                 stripes = stripe_nr_end - 1 - j;
3026                                 do_div(stripes, factor);
3027                                 multi->stripes[i].length = map->stripe_len *
3028                                         (stripes - stripe_nr + 1);
3029
3030                                 if (i < map->sub_stripes) {
3031                                         multi->stripes[i].length -=
3032                                                 stripe_offset;
3033                                         if (i == map->sub_stripes - 1)
3034                                                 stripe_offset = 0;
3035                                 }
3036                                 if (stripe_index >= last_stripe &&
3037                                     stripe_index <= (last_stripe +
3038                                                      map->sub_stripes - 1)) {
3039                                         multi->stripes[i].length -=
3040                                                 stripe_end_offset;
3041                                 }
3042                         } else
3043                                 multi->stripes[i].length = *length;
3044
3045                         stripe_index++;
3046                         if (stripe_index == map->num_stripes) {
3047                                 /* This could only happen for RAID0/10 */
3048                                 stripe_index = 0;
3049                                 stripe_nr++;
3050                         }
3051                 }
3052         } else {
3053                 for (i = 0; i < num_stripes; i++) {
3054                         multi->stripes[i].physical =
3055                                 map->stripes[stripe_index].physical +
3056                                 stripe_offset +
3057                                 stripe_nr * map->stripe_len;
3058                         multi->stripes[i].dev =
3059                                 map->stripes[stripe_index].dev;
3060                         stripe_index++;
3061                 }
3062         }
3063         if (multi_ret) {
3064                 *multi_ret = multi;
3065                 multi->num_stripes = num_stripes;
3066                 multi->max_errors = max_errors;
3067         }
3068 out:
3069         free_extent_map(em);
3070         return 0;
3071 }
3072
3073 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3074                       u64 logical, u64 *length,
3075                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3076 {
3077         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3078                                  mirror_num);
3079 }
3080
3081 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3082                      u64 chunk_start, u64 physical, u64 devid,
3083                      u64 **logical, int *naddrs, int *stripe_len)
3084 {
3085         struct extent_map_tree *em_tree = &map_tree->map_tree;
3086         struct extent_map *em;
3087         struct map_lookup *map;
3088         u64 *buf;
3089         u64 bytenr;
3090         u64 length;
3091         u64 stripe_nr;
3092         int i, j, nr = 0;
3093
3094         read_lock(&em_tree->lock);
3095         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3096         read_unlock(&em_tree->lock);
3097
3098         BUG_ON(!em || em->start != chunk_start);
3099         map = (struct map_lookup *)em->bdev;
3100
3101         length = em->len;
3102         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3103                 do_div(length, map->num_stripes / map->sub_stripes);
3104         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3105                 do_div(length, map->num_stripes);
3106
3107         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3108         BUG_ON(!buf);
3109
3110         for (i = 0; i < map->num_stripes; i++) {
3111                 if (devid && map->stripes[i].dev->devid != devid)
3112                         continue;
3113                 if (map->stripes[i].physical > physical ||
3114                     map->stripes[i].physical + length <= physical)
3115                         continue;
3116
3117                 stripe_nr = physical - map->stripes[i].physical;
3118                 do_div(stripe_nr, map->stripe_len);
3119
3120                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3121                         stripe_nr = stripe_nr * map->num_stripes + i;
3122                         do_div(stripe_nr, map->sub_stripes);
3123                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3124                         stripe_nr = stripe_nr * map->num_stripes + i;
3125                 }
3126                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3127                 WARN_ON(nr >= map->num_stripes);
3128                 for (j = 0; j < nr; j++) {
3129                         if (buf[j] == bytenr)
3130                                 break;
3131                 }
3132                 if (j == nr) {
3133                         WARN_ON(nr >= map->num_stripes);
3134                         buf[nr++] = bytenr;
3135                 }
3136         }
3137
3138         *logical = buf;
3139         *naddrs = nr;
3140         *stripe_len = map->stripe_len;
3141
3142         free_extent_map(em);
3143         return 0;
3144 }
3145
3146 static void end_bio_multi_stripe(struct bio *bio, int err)
3147 {
3148         struct btrfs_multi_bio *multi = bio->bi_private;
3149         int is_orig_bio = 0;
3150
3151         if (err)
3152                 atomic_inc(&multi->error);
3153
3154         if (bio == multi->orig_bio)
3155                 is_orig_bio = 1;
3156
3157         if (atomic_dec_and_test(&multi->stripes_pending)) {
3158                 if (!is_orig_bio) {
3159                         bio_put(bio);
3160                         bio = multi->orig_bio;
3161                 }
3162                 bio->bi_private = multi->private;
3163                 bio->bi_end_io = multi->end_io;
3164                 /* only send an error to the higher layers if it is
3165                  * beyond the tolerance of the multi-bio
3166                  */
3167                 if (atomic_read(&multi->error) > multi->max_errors) {
3168                         err = -EIO;
3169                 } else if (err) {
3170                         /*
3171                          * this bio is actually up to date, we didn't
3172                          * go over the max number of errors
3173                          */
3174                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3175                         err = 0;
3176                 }
3177                 kfree(multi);
3178
3179                 bio_endio(bio, err);
3180         } else if (!is_orig_bio) {
3181                 bio_put(bio);
3182         }
3183 }
3184
3185 struct async_sched {
3186         struct bio *bio;
3187         int rw;
3188         struct btrfs_fs_info *info;
3189         struct btrfs_work work;
3190 };
3191
3192 /*
3193  * see run_scheduled_bios for a description of why bios are collected for
3194  * async submit.
3195  *
3196  * This will add one bio to the pending list for a device and make sure
3197  * the work struct is scheduled.
3198  */
3199 static noinline int schedule_bio(struct btrfs_root *root,
3200                                  struct btrfs_device *device,
3201                                  int rw, struct bio *bio)
3202 {
3203         int should_queue = 1;
3204         struct btrfs_pending_bios *pending_bios;
3205
3206         /* don't bother with additional async steps for reads, right now */
3207         if (!(rw & REQ_WRITE)) {
3208                 bio_get(bio);
3209                 submit_bio(rw, bio);
3210                 bio_put(bio);
3211                 return 0;
3212         }
3213
3214         /*
3215          * nr_async_bios allows us to reliably return congestion to the
3216          * higher layers.  Otherwise, the async bio makes it appear we have
3217          * made progress against dirty pages when we've really just put it
3218          * on a queue for later
3219          */
3220         atomic_inc(&root->fs_info->nr_async_bios);
3221         WARN_ON(bio->bi_next);
3222         bio->bi_next = NULL;
3223         bio->bi_rw |= rw;
3224
3225         spin_lock(&device->io_lock);
3226         if (bio->bi_rw & REQ_SYNC)
3227                 pending_bios = &device->pending_sync_bios;
3228         else
3229                 pending_bios = &device->pending_bios;
3230
3231         if (pending_bios->tail)
3232                 pending_bios->tail->bi_next = bio;
3233
3234         pending_bios->tail = bio;
3235         if (!pending_bios->head)
3236                 pending_bios->head = bio;
3237         if (device->running_pending)
3238                 should_queue = 0;
3239
3240         spin_unlock(&device->io_lock);
3241
3242         if (should_queue)
3243                 btrfs_queue_worker(&root->fs_info->submit_workers,
3244                                    &device->work);
3245         return 0;
3246 }
3247
3248 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3249                   int mirror_num, int async_submit)
3250 {
3251         struct btrfs_mapping_tree *map_tree;
3252         struct btrfs_device *dev;
3253         struct bio *first_bio = bio;
3254         u64 logical = (u64)bio->bi_sector << 9;
3255         u64 length = 0;
3256         u64 map_length;
3257         struct btrfs_multi_bio *multi = NULL;
3258         int ret;
3259         int dev_nr = 0;
3260         int total_devs = 1;
3261
3262         length = bio->bi_size;
3263         map_tree = &root->fs_info->mapping_tree;
3264         map_length = length;
3265
3266         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3267                               mirror_num);
3268         BUG_ON(ret);
3269
3270         total_devs = multi->num_stripes;
3271         if (map_length < length) {
3272                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3273                        "len %llu\n", (unsigned long long)logical,
3274                        (unsigned long long)length,
3275                        (unsigned long long)map_length);
3276                 BUG();
3277         }
3278         multi->end_io = first_bio->bi_end_io;
3279         multi->private = first_bio->bi_private;
3280         multi->orig_bio = first_bio;
3281         atomic_set(&multi->stripes_pending, multi->num_stripes);
3282
3283         while (dev_nr < total_devs) {
3284                 if (total_devs > 1) {
3285                         if (dev_nr < total_devs - 1) {
3286                                 bio = bio_clone(first_bio, GFP_NOFS);
3287                                 BUG_ON(!bio);
3288                         } else {
3289                                 bio = first_bio;
3290                         }
3291                         bio->bi_private = multi;
3292                         bio->bi_end_io = end_bio_multi_stripe;
3293                 }
3294                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3295                 dev = multi->stripes[dev_nr].dev;
3296                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3297                         bio->bi_bdev = dev->bdev;
3298                         if (async_submit)
3299                                 schedule_bio(root, dev, rw, bio);
3300                         else
3301                                 submit_bio(rw, bio);
3302                 } else {
3303                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3304                         bio->bi_sector = logical >> 9;
3305                         bio_endio(bio, -EIO);
3306                 }
3307                 dev_nr++;
3308         }
3309         if (total_devs == 1)
3310                 kfree(multi);
3311         return 0;
3312 }
3313
3314 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3315                                        u8 *uuid, u8 *fsid)
3316 {
3317         struct btrfs_device *device;
3318         struct btrfs_fs_devices *cur_devices;
3319
3320         cur_devices = root->fs_info->fs_devices;
3321         while (cur_devices) {
3322                 if (!fsid ||
3323                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3324                         device = __find_device(&cur_devices->devices,
3325                                                devid, uuid);
3326                         if (device)
3327                                 return device;
3328                 }
3329                 cur_devices = cur_devices->seed;
3330         }
3331         return NULL;
3332 }
3333
3334 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3335                                             u64 devid, u8 *dev_uuid)
3336 {
3337         struct btrfs_device *device;
3338         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3339
3340         device = kzalloc(sizeof(*device), GFP_NOFS);
3341         if (!device)
3342                 return NULL;
3343         list_add(&device->dev_list,
3344                  &fs_devices->devices);
3345         device->dev_root = root->fs_info->dev_root;
3346         device->devid = devid;
3347         device->work.func = pending_bios_fn;
3348         device->fs_devices = fs_devices;
3349         device->missing = 1;
3350         fs_devices->num_devices++;
3351         fs_devices->missing_devices++;
3352         spin_lock_init(&device->io_lock);
3353         INIT_LIST_HEAD(&device->dev_alloc_list);
3354         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3355         return device;
3356 }
3357
3358 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3359                           struct extent_buffer *leaf,
3360                           struct btrfs_chunk *chunk)
3361 {
3362         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3363         struct map_lookup *map;
3364         struct extent_map *em;
3365         u64 logical;
3366         u64 length;
3367         u64 devid;
3368         u8 uuid[BTRFS_UUID_SIZE];
3369         int num_stripes;
3370         int ret;
3371         int i;
3372
3373         logical = key->offset;
3374         length = btrfs_chunk_length(leaf, chunk);
3375
3376         read_lock(&map_tree->map_tree.lock);
3377         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3378         read_unlock(&map_tree->map_tree.lock);
3379
3380         /* already mapped? */
3381         if (em && em->start <= logical && em->start + em->len > logical) {
3382                 free_extent_map(em);
3383                 return 0;
3384         } else if (em) {
3385                 free_extent_map(em);
3386         }
3387
3388         em = alloc_extent_map();
3389         if (!em)
3390                 return -ENOMEM;
3391         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3392         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3393         if (!map) {
3394                 free_extent_map(em);
3395                 return -ENOMEM;
3396         }
3397
3398         em->bdev = (struct block_device *)map;
3399         em->start = logical;
3400         em->len = length;
3401         em->block_start = 0;
3402         em->block_len = em->len;
3403
3404         map->num_stripes = num_stripes;
3405         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3406         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3407         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3408         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3409         map->type = btrfs_chunk_type(leaf, chunk);
3410         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3411         for (i = 0; i < num_stripes; i++) {
3412                 map->stripes[i].physical =
3413                         btrfs_stripe_offset_nr(leaf, chunk, i);
3414                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3415                 read_extent_buffer(leaf, uuid, (unsigned long)
3416                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3417                                    BTRFS_UUID_SIZE);
3418                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3419                                                         NULL);
3420                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3421                         kfree(map);
3422                         free_extent_map(em);
3423                         return -EIO;
3424                 }
3425                 if (!map->stripes[i].dev) {
3426                         map->stripes[i].dev =
3427                                 add_missing_dev(root, devid, uuid);
3428                         if (!map->stripes[i].dev) {
3429                                 kfree(map);
3430                                 free_extent_map(em);
3431                                 return -EIO;
3432                         }
3433                 }
3434                 map->stripes[i].dev->in_fs_metadata = 1;
3435         }
3436
3437         write_lock(&map_tree->map_tree.lock);
3438         ret = add_extent_mapping(&map_tree->map_tree, em);
3439         write_unlock(&map_tree->map_tree.lock);
3440         BUG_ON(ret);
3441         free_extent_map(em);
3442
3443         return 0;
3444 }
3445
3446 static int fill_device_from_item(struct extent_buffer *leaf,
3447                                  struct btrfs_dev_item *dev_item,
3448                                  struct btrfs_device *device)
3449 {
3450         unsigned long ptr;
3451
3452         device->devid = btrfs_device_id(leaf, dev_item);
3453         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3454         device->total_bytes = device->disk_total_bytes;
3455         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3456         device->type = btrfs_device_type(leaf, dev_item);
3457         device->io_align = btrfs_device_io_align(leaf, dev_item);
3458         device->io_width = btrfs_device_io_width(leaf, dev_item);
3459         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3460
3461         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3462         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3463
3464         return 0;
3465 }
3466
3467 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3468 {
3469         struct btrfs_fs_devices *fs_devices;
3470         int ret;
3471
3472         mutex_lock(&uuid_mutex);
3473
3474         fs_devices = root->fs_info->fs_devices->seed;
3475         while (fs_devices) {
3476                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3477                         ret = 0;
3478                         goto out;
3479                 }
3480                 fs_devices = fs_devices->seed;
3481         }
3482
3483         fs_devices = find_fsid(fsid);
3484         if (!fs_devices) {
3485                 ret = -ENOENT;
3486                 goto out;
3487         }
3488
3489         fs_devices = clone_fs_devices(fs_devices);
3490         if (IS_ERR(fs_devices)) {
3491                 ret = PTR_ERR(fs_devices);
3492                 goto out;
3493         }
3494
3495         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3496                                    root->fs_info->bdev_holder);
3497         if (ret)
3498                 goto out;
3499
3500         if (!fs_devices->seeding) {
3501                 __btrfs_close_devices(fs_devices);
3502                 free_fs_devices(fs_devices);
3503                 ret = -EINVAL;
3504                 goto out;
3505         }
3506
3507         fs_devices->seed = root->fs_info->fs_devices->seed;
3508         root->fs_info->fs_devices->seed = fs_devices;
3509 out:
3510         mutex_unlock(&uuid_mutex);
3511         return ret;
3512 }
3513
3514 static int read_one_dev(struct btrfs_root *root,
3515                         struct extent_buffer *leaf,
3516                         struct btrfs_dev_item *dev_item)
3517 {
3518         struct btrfs_device *device;
3519         u64 devid;
3520         int ret;
3521         u8 fs_uuid[BTRFS_UUID_SIZE];
3522         u8 dev_uuid[BTRFS_UUID_SIZE];
3523
3524         devid = btrfs_device_id(leaf, dev_item);
3525         read_extent_buffer(leaf, dev_uuid,
3526                            (unsigned long)btrfs_device_uuid(dev_item),
3527                            BTRFS_UUID_SIZE);
3528         read_extent_buffer(leaf, fs_uuid,
3529                            (unsigned long)btrfs_device_fsid(dev_item),
3530                            BTRFS_UUID_SIZE);
3531
3532         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3533                 ret = open_seed_devices(root, fs_uuid);
3534                 if (ret && !btrfs_test_opt(root, DEGRADED))
3535                         return ret;
3536         }
3537
3538         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3539         if (!device || !device->bdev) {
3540                 if (!btrfs_test_opt(root, DEGRADED))
3541                         return -EIO;
3542
3543                 if (!device) {
3544                         printk(KERN_WARNING "warning devid %llu missing\n",
3545                                (unsigned long long)devid);
3546                         device = add_missing_dev(root, devid, dev_uuid);
3547                         if (!device)
3548                                 return -ENOMEM;
3549                 } else if (!device->missing) {
3550                         /*
3551                          * this happens when a device that was properly setup
3552                          * in the device info lists suddenly goes bad.
3553                          * device->bdev is NULL, and so we have to set
3554                          * device->missing to one here
3555                          */
3556                         root->fs_info->fs_devices->missing_devices++;
3557                         device->missing = 1;
3558                 }
3559         }
3560
3561         if (device->fs_devices != root->fs_info->fs_devices) {
3562                 BUG_ON(device->writeable);
3563                 if (device->generation !=
3564                     btrfs_device_generation(leaf, dev_item))
3565                         return -EINVAL;
3566         }
3567
3568         fill_device_from_item(leaf, dev_item, device);
3569         device->dev_root = root->fs_info->dev_root;
3570         device->in_fs_metadata = 1;
3571         if (device->writeable)
3572                 device->fs_devices->total_rw_bytes += device->total_bytes;
3573         ret = 0;
3574         return ret;
3575 }
3576
3577 int btrfs_read_sys_array(struct btrfs_root *root)
3578 {
3579         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3580         struct extent_buffer *sb;
3581         struct btrfs_disk_key *disk_key;
3582         struct btrfs_chunk *chunk;
3583         u8 *ptr;
3584         unsigned long sb_ptr;
3585         int ret = 0;
3586         u32 num_stripes;
3587         u32 array_size;
3588         u32 len = 0;
3589         u32 cur;
3590         struct btrfs_key key;
3591
3592         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3593                                           BTRFS_SUPER_INFO_SIZE);
3594         if (!sb)
3595                 return -ENOMEM;
3596         btrfs_set_buffer_uptodate(sb);
3597         btrfs_set_buffer_lockdep_class(sb, 0);
3598
3599         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3600         array_size = btrfs_super_sys_array_size(super_copy);
3601
3602         ptr = super_copy->sys_chunk_array;
3603         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3604         cur = 0;
3605
3606         while (cur < array_size) {
3607                 disk_key = (struct btrfs_disk_key *)ptr;
3608                 btrfs_disk_key_to_cpu(&key, disk_key);
3609
3610                 len = sizeof(*disk_key); ptr += len;
3611                 sb_ptr += len;
3612                 cur += len;
3613
3614                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3615                         chunk = (struct btrfs_chunk *)sb_ptr;
3616                         ret = read_one_chunk(root, &key, sb, chunk);
3617                         if (ret)
3618                                 break;
3619                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3620                         len = btrfs_chunk_item_size(num_stripes);
3621                 } else {
3622                         ret = -EIO;
3623                         break;
3624                 }
3625                 ptr += len;
3626                 sb_ptr += len;
3627                 cur += len;
3628         }
3629         free_extent_buffer(sb);
3630         return ret;
3631 }
3632
3633 int btrfs_read_chunk_tree(struct btrfs_root *root)
3634 {
3635         struct btrfs_path *path;
3636         struct extent_buffer *leaf;
3637         struct btrfs_key key;
3638         struct btrfs_key found_key;
3639         int ret;
3640         int slot;
3641
3642         root = root->fs_info->chunk_root;
3643
3644         path = btrfs_alloc_path();
3645         if (!path)
3646                 return -ENOMEM;
3647
3648         /* first we search for all of the device items, and then we
3649          * read in all of the chunk items.  This way we can create chunk
3650          * mappings that reference all of the devices that are afound
3651          */
3652         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3653         key.offset = 0;
3654         key.type = 0;
3655 again:
3656         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3657         if (ret < 0)
3658                 goto error;
3659         while (1) {
3660                 leaf = path->nodes[0];
3661                 slot = path->slots[0];
3662                 if (slot >= btrfs_header_nritems(leaf)) {
3663                         ret = btrfs_next_leaf(root, path);
3664                         if (ret == 0)
3665                                 continue;
3666                         if (ret < 0)
3667                                 goto error;
3668                         break;
3669                 }
3670                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3671                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3672                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3673                                 break;
3674                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3675                                 struct btrfs_dev_item *dev_item;
3676                                 dev_item = btrfs_item_ptr(leaf, slot,
3677                                                   struct btrfs_dev_item);
3678                                 ret = read_one_dev(root, leaf, dev_item);
3679                                 if (ret)
3680                                         goto error;
3681                         }
3682                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3683                         struct btrfs_chunk *chunk;
3684                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3685                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3686                         if (ret)
3687                                 goto error;
3688                 }
3689                 path->slots[0]++;
3690         }
3691         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3692                 key.objectid = 0;
3693                 btrfs_release_path(path);
3694                 goto again;
3695         }
3696         ret = 0;
3697 error:
3698         btrfs_free_path(path);
3699         return ret;
3700 }