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 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  42                             (sizeof(struct btrfs_bio_stripe) * (n)))
  43
  44 static DEFINE_MUTEX(uuid_mutex);
  45 static LIST_HEAD(fs_uuids);
  46
  47 void btrfs_lock_volumes(void)
  48 {
  49         mutex_lock(&uuid_mutex);
  50 }
  51
  52 void btrfs_unlock_volumes(void)
  53 {
  54         mutex_unlock(&uuid_mutex);
  55 }
  56
  57 static void lock_chunks(struct btrfs_root *root)
  58 {
  59         mutex_lock(&root->fs_info->chunk_mutex);
  60 }
  61
  62 static void unlock_chunks(struct btrfs_root *root)
  63 {
  64         mutex_unlock(&root->fs_info->chunk_mutex);
  65 }
  66
  67 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  68 {
  69         struct btrfs_device *device;
  70         WARN_ON(fs_devices->opened);
  71         while (!list_empty(&fs_devices->devices)) {
  72                 device = list_entry(fs_devices->devices.next,
  73                                     struct btrfs_device, dev_list);
  74                 list_del(&device->dev_list);
  75                 kfree(device->name);
  76                 kfree(device);
  77         }
  78         kfree(fs_devices);
  79 }
  80
  81 int btrfs_cleanup_fs_uuids(void)
  82 {
  83         struct btrfs_fs_devices *fs_devices;
  84
  85         while (!list_empty(&fs_uuids)) {
  86                 fs_devices = list_entry(fs_uuids.next,
  87                                         struct btrfs_fs_devices, list);
  88                 list_del(&fs_devices->list);
  89                 free_fs_devices(fs_devices);
  90         }
  91         return 0;
  92 }
  93
  94 static noinline struct btrfs_device *__find_device(struct list_head *head,
  95                                                    u64 devid, u8 *uuid)
  96 {
  97         struct btrfs_device *dev;
  98
  99         list_for_each_entry(dev, head, dev_list) {
 100                 if (dev->devid == devid &&
 101                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 102                         return dev;
 103                 }
 104         }
 105         return NULL;
 106 }
 107
 108 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 109 {
 110         struct btrfs_fs_devices *fs_devices;
 111
 112         list_for_each_entry(fs_devices, &fs_uuids, list) {
 113                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 114                         return fs_devices;
 115         }
 116         return NULL;
 117 }
 118
 119 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 120                         struct bio *head, struct bio *tail)
 121 {
 122
 123         struct bio *old_head;
 124
 125         old_head = pending_bios->head;
 126         pending_bios->head = head;
 127         if (pending_bios->tail)
 128                 tail->bi_next = old_head;
 129         else
 130                 pending_bios->tail = tail;
 131 }
 132
 133 /*
 134  * we try to collect pending bios for a device so we don't get a large
 135  * number of procs sending bios down to the same device.  This greatly
 136  * improves the schedulers ability to collect and merge the bios.
 137  *
 138  * But, it also turns into a long list of bios to process and that is sure
 139  * to eventually make the worker thread block.  The solution here is to
 140  * make some progress and then put this work struct back at the end of
 141  * the list if the block device is congested.  This way, multiple devices
 142  * can make progress from a single worker thread.
 143  */
 144 static noinline int run_scheduled_bios(struct btrfs_device *device)
 145 {
 146         struct bio *pending;
 147         struct backing_dev_info *bdi;
 148         struct btrfs_fs_info *fs_info;
 149         struct btrfs_pending_bios *pending_bios;
 150         struct bio *tail;
 151         struct bio *cur;
 152         int again = 0;
 153         unsigned long num_run;
 154         unsigned long batch_run = 0;
 155         unsigned long limit;
 156         unsigned long last_waited = 0;
 157         int force_reg = 0;
 158
 159         bdi = blk_get_backing_dev_info(device->bdev);
 160         fs_info = device->dev_root->fs_info;
 161         limit = btrfs_async_submit_limit(fs_info);
 162         limit = limit * 2 / 3;
 163
 164 loop:
 165         spin_lock(&device->io_lock);
 166
 167 loop_lock:
 168         num_run = 0;
 169
 170         /* take all the bios off the list at once and process them
 171          * later on (without the lock held).  But, remember the
 172          * tail and other pointers so the bios can be properly reinserted
 173          * into the list if we hit congestion
 174          */
 175         if (!force_reg && device->pending_sync_bios.head) {
 176                 pending_bios = &device->pending_sync_bios;
 177                 force_reg = 1;
 178         } else {
 179                 pending_bios = &device->pending_bios;
 180                 force_reg = 0;
 181         }
 182
 183         pending = pending_bios->head;
 184         tail = pending_bios->tail;
 185         WARN_ON(pending && !tail);
 186
 187         /*
 188          * if pending was null this time around, no bios need processing
 189          * at all and we can stop.  Otherwise it'll loop back up again
 190          * and do an additional check so no bios are missed.
 191          *
 192          * device->running_pending is used to synchronize with the
 193          * schedule_bio code.
 194          */
 195         if (device->pending_sync_bios.head == NULL &&
 196             device->pending_bios.head == NULL) {
 197                 again = 0;
 198                 device->running_pending = 0;
 199         } else {
 200                 again = 1;
 201                 device->running_pending = 1;
 202         }
 203
 204         pending_bios->head = NULL;
 205         pending_bios->tail = NULL;
 206
 207         spin_unlock(&device->io_lock);
 208
 209         while (pending) {
 210
 211                 rmb();
 212                 /* we want to work on both lists, but do more bios on the
 213                  * sync list than the regular list
 214                  */
 215                 if ((num_run > 32 &&
 216                     pending_bios != &device->pending_sync_bios &&
 217                     device->pending_sync_bios.head) ||
 218                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 219                     device->pending_bios.head)) {
 220                         spin_lock(&device->io_lock);
 221                         requeue_list(pending_bios, pending, tail);
 222                         goto loop_lock;
 223                 }
 224
 225                 cur = pending;
 226                 pending = pending->bi_next;
 227                 cur->bi_next = NULL;
 228                 atomic_dec(&fs_info->nr_async_bios);
 229
 230                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 231                     waitqueue_active(&fs_info->async_submit_wait))
 232                         wake_up(&fs_info->async_submit_wait);
 233
 234                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 235
 236                 submit_bio(cur->bi_rw, cur);
 237                 num_run++;
 238                 batch_run++;
 239                 if (need_resched())
 240                         cond_resched();
 241
 242                 /*
 243                  * we made progress, there is more work to do and the bdi
 244                  * is now congested.  Back off and let other work structs
 245                  * run instead
 246                  */
 247                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 248                     fs_info->fs_devices->open_devices > 1) {
 249                         struct io_context *ioc;
 250
 251                         ioc = current->io_context;
 252
 253                         /*
 254                          * the main goal here is that we don't want to
 255                          * block if we're going to be able to submit
 256                          * more requests without blocking.
 257                          *
 258                          * This code does two great things, it pokes into
 259                          * the elevator code from a filesystem _and_
 260                          * it makes assumptions about how batching works.
 261                          */
 262                         if (ioc && ioc->nr_batch_requests > 0 &&
 263                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 264                             (last_waited == 0 ||
 265                              ioc->last_waited == last_waited)) {
 266                                 /*
 267                                  * we want to go through our batch of
 268                                  * requests and stop.  So, we copy out
 269                                  * the ioc->last_waited time and test
 270                                  * against it before looping
 271                                  */
 272                                 last_waited = ioc->last_waited;
 273                                 if (need_resched())
 274                                         cond_resched();
 275                                 continue;
 276                         }
 277                         spin_lock(&device->io_lock);
 278                         requeue_list(pending_bios, pending, tail);
 279                         device->running_pending = 1;
 280
 281                         spin_unlock(&device->io_lock);
 282                         btrfs_requeue_work(&device->work);
 283                         goto done;
 284                 }
 285         }
 286
 287         cond_resched();
 288         if (again)
 289                 goto loop;
 290
 291         spin_lock(&device->io_lock);
 292         if (device->pending_bios.head || device->pending_sync_bios.head)
 293                 goto loop_lock;
 294         spin_unlock(&device->io_lock);
 295
 296 done:
 297         return 0;
 298 }
 299
 300 static void pending_bios_fn(struct btrfs_work *work)
 301 {
 302         struct btrfs_device *device;
 303
 304         device = container_of(work, struct btrfs_device, work);
 305         run_scheduled_bios(device);
 306 }
 307
 308 static noinline int device_list_add(const char *path,
 309                            struct btrfs_super_block *disk_super,
 310                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 311 {
 312         struct btrfs_device *device;
 313         struct btrfs_fs_devices *fs_devices;
 314         u64 found_transid = btrfs_super_generation(disk_super);
 315         char *name;
 316
 317         fs_devices = find_fsid(disk_super->fsid);
 318         if (!fs_devices) {
 319                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 320                 if (!fs_devices)
 321                         return -ENOMEM;
 322                 INIT_LIST_HEAD(&fs_devices->devices);
 323                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 324                 list_add(&fs_devices->list, &fs_uuids);
 325                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 326                 fs_devices->latest_devid = devid;
 327                 fs_devices->latest_trans = found_transid;
 328                 mutex_init(&fs_devices->device_list_mutex);
 329                 device = NULL;
 330         } else {
 331                 device = __find_device(&fs_devices->devices, devid,
 332                                        disk_super->dev_item.uuid);
 333         }
 334         if (!device) {
 335                 if (fs_devices->opened)
 336                         return -EBUSY;
 337
 338                 device = kzalloc(sizeof(*device), GFP_NOFS);
 339                 if (!device) {
 340                         /* we can safely leave the fs_devices entry around */
 341                         return -ENOMEM;
 342                 }
 343                 device->devid = devid;
 344                 device->work.func = pending_bios_fn;
 345                 memcpy(device->uuid, disk_super->dev_item.uuid,
 346                        BTRFS_UUID_SIZE);
 347                 spin_lock_init(&device->io_lock);
 348                 device->name = kstrdup(path, GFP_NOFS);
 349                 if (!device->name) {
 350                         kfree(device);
 351                         return -ENOMEM;
 352                 }
 353                 INIT_LIST_HEAD(&device->dev_alloc_list);
 354
 355                 mutex_lock(&fs_devices->device_list_mutex);
 356                 list_add(&device->dev_list, &fs_devices->devices);
 357                 mutex_unlock(&fs_devices->device_list_mutex);
 358
 359                 device->fs_devices = fs_devices;
 360                 fs_devices->num_devices++;
 361         } else if (!device->name || strcmp(device->name, path)) {
 362                 name = kstrdup(path, GFP_NOFS);
 363                 if (!name)
 364                         return -ENOMEM;
 365                 kfree(device->name);
 366                 device->name = name;
 367                 if (device->missing) {
 368                         fs_devices->missing_devices--;
 369                         device->missing = 0;
 370                 }
 371         }
 372
 373         if (found_transid > fs_devices->latest_trans) {
 374                 fs_devices->latest_devid = devid;
 375                 fs_devices->latest_trans = found_transid;
 376         }
 377         *fs_devices_ret = fs_devices;
 378         return 0;
 379 }
 380
 381 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 382 {
 383         struct btrfs_fs_devices *fs_devices;
 384         struct btrfs_device *device;
 385         struct btrfs_device *orig_dev;
 386
 387         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 388         if (!fs_devices)
 389                 return ERR_PTR(-ENOMEM);
 390
 391         INIT_LIST_HEAD(&fs_devices->devices);
 392         INIT_LIST_HEAD(&fs_devices->alloc_list);
 393         INIT_LIST_HEAD(&fs_devices->list);
 394         mutex_init(&fs_devices->device_list_mutex);
 395         fs_devices->latest_devid = orig->latest_devid;
 396         fs_devices->latest_trans = orig->latest_trans;
 397         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 398
 399         mutex_lock(&orig->device_list_mutex);
 400         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 401                 device = kzalloc(sizeof(*device), GFP_NOFS);
 402                 if (!device)
 403                         goto error;
 404
 405                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 406                 if (!device->name) {
 407                         kfree(device);
 408                         goto error;
 409                 }
 410
 411                 device->devid = orig_dev->devid;
 412                 device->work.func = pending_bios_fn;
 413                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 414                 spin_lock_init(&device->io_lock);
 415                 INIT_LIST_HEAD(&device->dev_list);
 416                 INIT_LIST_HEAD(&device->dev_alloc_list);
 417
 418                 list_add(&device->dev_list, &fs_devices->devices);
 419                 device->fs_devices = fs_devices;
 420                 fs_devices->num_devices++;
 421         }
 422         mutex_unlock(&orig->device_list_mutex);
 423         return fs_devices;
 424 error:
 425         mutex_unlock(&orig->device_list_mutex);
 426         free_fs_devices(fs_devices);
 427         return ERR_PTR(-ENOMEM);
 428 }
 429
 430 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 431 {
 432         struct btrfs_device *device, *next;
 433
 434         mutex_lock(&uuid_mutex);
 435 again:
 436         mutex_lock(&fs_devices->device_list_mutex);
 437         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 438                 if (device->in_fs_metadata)
 439                         continue;
 440
 441                 if (device->bdev) {
 442                         blkdev_put(device->bdev, device->mode);
 443                         device->bdev = NULL;
 444                         fs_devices->open_devices--;
 445                 }
 446                 if (device->writeable) {
 447                         list_del_init(&device->dev_alloc_list);
 448                         device->writeable = 0;
 449                         fs_devices->rw_devices--;
 450                 }
 451                 list_del_init(&device->dev_list);
 452                 fs_devices->num_devices--;
 453                 kfree(device->name);
 454                 kfree(device);
 455         }
 456         mutex_unlock(&fs_devices->device_list_mutex);
 457
 458         if (fs_devices->seed) {
 459                 fs_devices = fs_devices->seed;
 460                 goto again;
 461         }
 462
 463         mutex_unlock(&uuid_mutex);
 464         return 0;
 465 }
 466
 467 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 468 {
 469         struct btrfs_device *device;
 470
 471         if (--fs_devices->opened > 0)
 472                 return 0;
 473
 474         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 475                 if (device->bdev) {
 476                         blkdev_put(device->bdev, device->mode);
 477                         fs_devices->open_devices--;
 478                 }
 479                 if (device->writeable) {
 480                         list_del_init(&device->dev_alloc_list);
 481                         fs_devices->rw_devices--;
 482                 }
 483
 484                 device->bdev = NULL;
 485                 device->writeable = 0;
 486                 device->in_fs_metadata = 0;
 487         }
 488         WARN_ON(fs_devices->open_devices);
 489         WARN_ON(fs_devices->rw_devices);
 490         fs_devices->opened = 0;
 491         fs_devices->seeding = 0;
 492
 493         return 0;
 494 }
 495
 496 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 497 {
 498         struct btrfs_fs_devices *seed_devices = NULL;
 499         int ret;
 500
 501         mutex_lock(&uuid_mutex);
 502         ret = __btrfs_close_devices(fs_devices);
 503         if (!fs_devices->opened) {
 504                 seed_devices = fs_devices->seed;
 505                 fs_devices->seed = NULL;
 506         }
 507         mutex_unlock(&uuid_mutex);
 508
 509         while (seed_devices) {
 510                 fs_devices = seed_devices;
 511                 seed_devices = fs_devices->seed;
 512                 __btrfs_close_devices(fs_devices);
 513                 free_fs_devices(fs_devices);
 514         }
 515         return ret;
 516 }
 517
 518 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 519                                 fmode_t flags, void *holder)
 520 {
 521         struct block_device *bdev;
 522         struct list_head *head = &fs_devices->devices;
 523         struct btrfs_device *device;
 524         struct block_device *latest_bdev = NULL;
 525         struct buffer_head *bh;
 526         struct btrfs_super_block *disk_super;
 527         u64 latest_devid = 0;
 528         u64 latest_transid = 0;
 529         u64 devid;
 530         int seeding = 1;
 531         int ret = 0;
 532
 533         flags |= FMODE_EXCL;
 534
 535         list_for_each_entry(device, head, dev_list) {
 536                 if (device->bdev)
 537                         continue;
 538                 if (!device->name)
 539                         continue;
 540
 541                 bdev = blkdev_get_by_path(device->name, flags, holder);
 542                 if (IS_ERR(bdev)) {
 543                         printk(KERN_INFO "open %s failed\n", device->name);
 544                         goto error;
 545                 }
 546                 set_blocksize(bdev, 4096);
 547
 548                 bh = btrfs_read_dev_super(bdev);
 549                 if (!bh) {
 550                         ret = -EINVAL;
 551                         goto error_close;
 552                 }
 553
 554                 disk_super = (struct btrfs_super_block *)bh->b_data;
 555                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 556                 if (devid != device->devid)
 557                         goto error_brelse;
 558
 559                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 560                            BTRFS_UUID_SIZE))
 561                         goto error_brelse;
 562
 563                 device->generation = btrfs_super_generation(disk_super);
 564                 if (!latest_transid || device->generation > latest_transid) {
 565                         latest_devid = devid;
 566                         latest_transid = device->generation;
 567                         latest_bdev = bdev;
 568                 }
 569
 570                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 571                         device->writeable = 0;
 572                 } else {
 573                         device->writeable = !bdev_read_only(bdev);
 574                         seeding = 0;
 575                 }
 576
 577                 device->bdev = bdev;
 578                 device->in_fs_metadata = 0;
 579                 device->mode = flags;
 580
 581                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 582                         fs_devices->rotating = 1;
 583
 584                 fs_devices->open_devices++;
 585                 if (device->writeable) {
 586                         fs_devices->rw_devices++;
 587                         list_add(&device->dev_alloc_list,
 588                                  &fs_devices->alloc_list);
 589                 }
 590                 continue;
 591
 592 error_brelse:
 593                 brelse(bh);
 594 error_close:
 595                 blkdev_put(bdev, flags);
 596 error:
 597                 continue;
 598         }
 599         if (fs_devices->open_devices == 0) {
 600                 ret = -EIO;
 601                 goto out;
 602         }
 603         fs_devices->seeding = seeding;
 604         fs_devices->opened = 1;
 605         fs_devices->latest_bdev = latest_bdev;
 606         fs_devices->latest_devid = latest_devid;
 607         fs_devices->latest_trans = latest_transid;
 608         fs_devices->total_rw_bytes = 0;
 609 out:
 610         return ret;
 611 }
 612
 613 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 614                        fmode_t flags, void *holder)
 615 {
 616         int ret;
 617
 618         mutex_lock(&uuid_mutex);
 619         if (fs_devices->opened) {
 620                 fs_devices->opened++;
 621                 ret = 0;
 622         } else {
 623                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 624         }
 625         mutex_unlock(&uuid_mutex);
 626         return ret;
 627 }
 628
 629 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 630                           struct btrfs_fs_devices **fs_devices_ret)
 631 {
 632         struct btrfs_super_block *disk_super;
 633         struct block_device *bdev;
 634         struct buffer_head *bh;
 635         int ret;
 636         u64 devid;
 637         u64 transid;
 638
 639         mutex_lock(&uuid_mutex);
 640
 641         flags |= FMODE_EXCL;
 642         bdev = blkdev_get_by_path(path, flags, holder);
 643
 644         if (IS_ERR(bdev)) {
 645                 ret = PTR_ERR(bdev);
 646                 goto error;
 647         }
 648
 649         ret = set_blocksize(bdev, 4096);
 650         if (ret)
 651                 goto error_close;
 652         bh = btrfs_read_dev_super(bdev);
 653         if (!bh) {
 654                 ret = -EINVAL;
 655                 goto error_close;
 656         }
 657         disk_super = (struct btrfs_super_block *)bh->b_data;
 658         devid = btrfs_stack_device_id(&disk_super->dev_item);
 659         transid = btrfs_super_generation(disk_super);
 660         if (disk_super->label[0])
 661                 printk(KERN_INFO "device label %s ", disk_super->label);
 662         else {
 663                 /* FIXME, make a readl uuid parser */
 664                 printk(KERN_INFO "device fsid %llx-%llx ",
 665                        *(unsigned long long *)disk_super->fsid,
 666                        *(unsigned long long *)(disk_super->fsid + 8));
 667         }
 668         printk(KERN_CONT "devid %llu transid %llu %s\n",
 669                (unsigned long long)devid, (unsigned long long)transid, path);
 670         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 671
 672         brelse(bh);
 673 error_close:
 674         blkdev_put(bdev, flags);
 675 error:
 676         mutex_unlock(&uuid_mutex);
 677         return ret;
 678 }
 679
 680 /* helper to account the used device space in the range */
 681 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 682                                    u64 end, u64 *length)
 683 {
 684         struct btrfs_key key;
 685         struct btrfs_root *root = device->dev_root;
 686         struct btrfs_dev_extent *dev_extent;
 687         struct btrfs_path *path;
 688         u64 extent_end;
 689         int ret;
 690         int slot;
 691         struct extent_buffer *l;
 692
 693         *length = 0;
 694
 695         if (start >= device->total_bytes)
 696                 return 0;
 697
 698         path = btrfs_alloc_path();
 699         if (!path)
 700                 return -ENOMEM;
 701         path->reada = 2;
 702
 703         key.objectid = device->devid;
 704         key.offset = start;
 705         key.type = BTRFS_DEV_EXTENT_KEY;
 706
 707         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 708         if (ret < 0)
 709                 goto out;
 710         if (ret > 0) {
 711                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 712                 if (ret < 0)
 713                         goto out;
 714         }
 715
 716         while (1) {
 717                 l = path->nodes[0];
 718                 slot = path->slots[0];
 719                 if (slot >= btrfs_header_nritems(l)) {
 720                         ret = btrfs_next_leaf(root, path);
 721                         if (ret == 0)
 722                                 continue;
 723                         if (ret < 0)
 724                                 goto out;
 725
 726                         break;
 727                 }
 728                 btrfs_item_key_to_cpu(l, &key, slot);
 729
 730                 if (key.objectid < device->devid)
 731                         goto next;
 732
 733                 if (key.objectid > device->devid)
 734                         break;
 735
 736                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 737                         goto next;
 738
 739                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 740                 extent_end = key.offset + btrfs_dev_extent_length(l,
 741                                                                   dev_extent);
 742                 if (key.offset <= start && extent_end > end) {
 743                         *length = end - start + 1;
 744                         break;
 745                 } else if (key.offset <= start && extent_end > start)
 746                         *length += extent_end - start;
 747                 else if (key.offset > start && extent_end <= end)
 748                         *length += extent_end - key.offset;
 749                 else if (key.offset > start && key.offset <= end) {
 750                         *length += end - key.offset + 1;
 751                         break;
 752                 } else if (key.offset > end)
 753                         break;
 754
 755 next:
 756                 path->slots[0]++;
 757         }
 758         ret = 0;
 759 out:
 760         btrfs_free_path(path);
 761         return ret;
 762 }
 763
 764 /*
 765  * find_free_dev_extent - find free space in the specified device
 766  * @trans:      transaction handler
 767  * @device:     the device which we search the free space in
 768  * @num_bytes:  the size of the free space that we need
 769  * @start:      store the start of the free space.
 770  * @len:        the size of the free space. that we find, or the size of the max
 771  *              free space if we don't find suitable free space
 772  *
 773  * this uses a pretty simple search, the expectation is that it is
 774  * called very infrequently and that a given device has a small number
 775  * of extents
 776  *
 777  * @start is used to store the start of the free space if we find. But if we
 778  * don't find suitable free space, it will be used to store the start position
 779  * of the max free space.
 780  *
 781  * @len is used to store the size of the free space that we find.
 782  * But if we don't find suitable free space, it is used to store the size of
 783  * the max free space.
 784  */
 785 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 786                          struct btrfs_device *device, u64 num_bytes,
 787                          u64 *start, u64 *len)
 788 {
 789         struct btrfs_key key;
 790         struct btrfs_root *root = device->dev_root;
 791         struct btrfs_dev_extent *dev_extent;
 792         struct btrfs_path *path;
 793         u64 hole_size;
 794         u64 max_hole_start;
 795         u64 max_hole_size;
 796         u64 extent_end;
 797         u64 search_start;
 798         u64 search_end = device->total_bytes;
 799         int ret;
 800         int slot;
 801         struct extent_buffer *l;
 802
 803         /* FIXME use last free of some kind */
 804
 805         /* we don't want to overwrite the superblock on the drive,
 806          * so we make sure to start at an offset of at least 1MB
 807          */
 808         search_start = 1024 * 1024;
 809
 810         if (root->fs_info->alloc_start + num_bytes <= search_end)
 811                 search_start = max(root->fs_info->alloc_start, search_start);
 812
 813         max_hole_start = search_start;
 814         max_hole_size = 0;
 815
 816         if (search_start >= search_end) {
 817                 ret = -ENOSPC;
 818                 goto error;
 819         }
 820
 821         path = btrfs_alloc_path();
 822         if (!path) {
 823                 ret = -ENOMEM;
 824                 goto error;
 825         }
 826         path->reada = 2;
 827
 828         key.objectid = device->devid;
 829         key.offset = search_start;
 830         key.type = BTRFS_DEV_EXTENT_KEY;
 831
 832         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 833         if (ret < 0)
 834                 goto out;
 835         if (ret > 0) {
 836                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 837                 if (ret < 0)
 838                         goto out;
 839         }
 840
 841         while (1) {
 842                 l = path->nodes[0];
 843                 slot = path->slots[0];
 844                 if (slot >= btrfs_header_nritems(l)) {
 845                         ret = btrfs_next_leaf(root, path);
 846                         if (ret == 0)
 847                                 continue;
 848                         if (ret < 0)
 849                                 goto out;
 850
 851                         break;
 852                 }
 853                 btrfs_item_key_to_cpu(l, &key, slot);
 854
 855                 if (key.objectid < device->devid)
 856                         goto next;
 857
 858                 if (key.objectid > device->devid)
 859                         break;
 860
 861                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 862                         goto next;
 863
 864                 if (key.offset > search_start) {
 865                         hole_size = key.offset - search_start;
 866
 867                         if (hole_size > max_hole_size) {
 868                                 max_hole_start = search_start;
 869                                 max_hole_size = hole_size;
 870                         }
 871
 872                         /*
 873                          * If this free space is greater than which we need,
 874                          * it must be the max free space that we have found
 875                          * until now, so max_hole_start must point to the start
 876                          * of this free space and the length of this free space
 877                          * is stored in max_hole_size. Thus, we return
 878                          * max_hole_start and max_hole_size and go back to the
 879                          * caller.
 880                          */
 881                         if (hole_size >= num_bytes) {
 882                                 ret = 0;
 883                                 goto out;
 884                         }
 885                 }
 886
 887                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 888                 extent_end = key.offset + btrfs_dev_extent_length(l,
 889                                                                   dev_extent);
 890                 if (extent_end > search_start)
 891                         search_start = extent_end;
 892 next:
 893                 path->slots[0]++;
 894                 cond_resched();
 895         }
 896
 897         hole_size = search_end- search_start;
 898         if (hole_size > max_hole_size) {
 899                 max_hole_start = search_start;
 900                 max_hole_size = hole_size;
 901         }
 902
 903         /* See above. */
 904         if (hole_size < num_bytes)
 905                 ret = -ENOSPC;
 906         else
 907                 ret = 0;
 908
 909 out:
 910         btrfs_free_path(path);
 911 error:
 912         *start = max_hole_start;
 913         if (len)
 914                 *len = max_hole_size;
 915         return ret;
 916 }
 917
 918 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 919                           struct btrfs_device *device,
 920                           u64 start)
 921 {
 922         int ret;
 923         struct btrfs_path *path;
 924         struct btrfs_root *root = device->dev_root;
 925         struct btrfs_key key;
 926         struct btrfs_key found_key;
 927         struct extent_buffer *leaf = NULL;
 928         struct btrfs_dev_extent *extent = NULL;
 929
 930         path = btrfs_alloc_path();
 931         if (!path)
 932                 return -ENOMEM;
 933
 934         key.objectid = device->devid;
 935         key.offset = start;
 936         key.type = BTRFS_DEV_EXTENT_KEY;
 937
 938         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 939         if (ret > 0) {
 940                 ret = btrfs_previous_item(root, path, key.objectid,
 941                                           BTRFS_DEV_EXTENT_KEY);
 942                 BUG_ON(ret);
 943                 leaf = path->nodes[0];
 944                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 945                 extent = btrfs_item_ptr(leaf, path->slots[0],
 946                                         struct btrfs_dev_extent);
 947                 BUG_ON(found_key.offset > start || found_key.offset +
 948                        btrfs_dev_extent_length(leaf, extent) < start);
 949                 ret = 0;
 950         } else if (ret == 0) {
 951                 leaf = path->nodes[0];
 952                 extent = btrfs_item_ptr(leaf, path->slots[0],
 953                                         struct btrfs_dev_extent);
 954         }
 955         BUG_ON(ret);
 956
 957         if (device->bytes_used > 0)
 958                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 959         ret = btrfs_del_item(trans, root, path);
 960         BUG_ON(ret);
 961
 962         btrfs_free_path(path);
 963         return ret;
 964 }
 965
 966 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 967                            struct btrfs_device *device,
 968                            u64 chunk_tree, u64 chunk_objectid,
 969                            u64 chunk_offset, u64 start, u64 num_bytes)
 970 {
 971         int ret;
 972         struct btrfs_path *path;
 973         struct btrfs_root *root = device->dev_root;
 974         struct btrfs_dev_extent *extent;
 975         struct extent_buffer *leaf;
 976         struct btrfs_key key;
 977
 978         WARN_ON(!device->in_fs_metadata);
 979         path = btrfs_alloc_path();
 980         if (!path)
 981                 return -ENOMEM;
 982
 983         key.objectid = device->devid;
 984         key.offset = start;
 985         key.type = BTRFS_DEV_EXTENT_KEY;
 986         ret = btrfs_insert_empty_item(trans, root, path, &key,
 987                                       sizeof(*extent));
 988         BUG_ON(ret);
 989
 990         leaf = path->nodes[0];
 991         extent = btrfs_item_ptr(leaf, path->slots[0],
 992                                 struct btrfs_dev_extent);
 993         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 994         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 995         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 996
 997         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 998                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 999                     BTRFS_UUID_SIZE);
1000
1001         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1002         btrfs_mark_buffer_dirty(leaf);
1003         btrfs_free_path(path);
1004         return ret;
1005 }
1006
1007 static noinline int find_next_chunk(struct btrfs_root *root,
1008                                     u64 objectid, u64 *offset)
1009 {
1010         struct btrfs_path *path;
1011         int ret;
1012         struct btrfs_key key;
1013         struct btrfs_chunk *chunk;
1014         struct btrfs_key found_key;
1015
1016         path = btrfs_alloc_path();
1017         BUG_ON(!path);
1018
1019         key.objectid = objectid;
1020         key.offset = (u64)-1;
1021         key.type = BTRFS_CHUNK_ITEM_KEY;
1022
1023         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1024         if (ret < 0)
1025                 goto error;
1026
1027         BUG_ON(ret == 0);
1028
1029         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1030         if (ret) {
1031                 *offset = 0;
1032         } else {
1033                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1034                                       path->slots[0]);
1035                 if (found_key.objectid != objectid)
1036                         *offset = 0;
1037                 else {
1038                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1039                                                struct btrfs_chunk);
1040                         *offset = found_key.offset +
1041                                 btrfs_chunk_length(path->nodes[0], chunk);
1042                 }
1043         }
1044         ret = 0;
1045 error:
1046         btrfs_free_path(path);
1047         return ret;
1048 }
1049
1050 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1051 {
1052         int ret;
1053         struct btrfs_key key;
1054         struct btrfs_key found_key;
1055         struct btrfs_path *path;
1056
1057         root = root->fs_info->chunk_root;
1058
1059         path = btrfs_alloc_path();
1060         if (!path)
1061                 return -ENOMEM;
1062
1063         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1064         key.type = BTRFS_DEV_ITEM_KEY;
1065         key.offset = (u64)-1;
1066
1067         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1068         if (ret < 0)
1069                 goto error;
1070
1071         BUG_ON(ret == 0);
1072
1073         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1074                                   BTRFS_DEV_ITEM_KEY);
1075         if (ret) {
1076                 *objectid = 1;
1077         } else {
1078                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1079                                       path->slots[0]);
1080                 *objectid = found_key.offset + 1;
1081         }
1082         ret = 0;
1083 error:
1084         btrfs_free_path(path);
1085         return ret;
1086 }
1087
1088 /*
1089  * the device information is stored in the chunk root
1090  * the btrfs_device struct should be fully filled in
1091  */
1092 int btrfs_add_device(struct btrfs_trans_handle *trans,
1093                      struct btrfs_root *root,
1094                      struct btrfs_device *device)
1095 {
1096         int ret;
1097         struct btrfs_path *path;
1098         struct btrfs_dev_item *dev_item;
1099         struct extent_buffer *leaf;
1100         struct btrfs_key key;
1101         unsigned long ptr;
1102
1103         root = root->fs_info->chunk_root;
1104
1105         path = btrfs_alloc_path();
1106         if (!path)
1107                 return -ENOMEM;
1108
1109         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1110         key.type = BTRFS_DEV_ITEM_KEY;
1111         key.offset = device->devid;
1112
1113         ret = btrfs_insert_empty_item(trans, root, path, &key,
1114                                       sizeof(*dev_item));
1115         if (ret)
1116                 goto out;
1117
1118         leaf = path->nodes[0];
1119         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1120
1121         btrfs_set_device_id(leaf, dev_item, device->devid);
1122         btrfs_set_device_generation(leaf, dev_item, 0);
1123         btrfs_set_device_type(leaf, dev_item, device->type);
1124         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1125         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1126         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1127         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1128         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1129         btrfs_set_device_group(leaf, dev_item, 0);
1130         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1131         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1132         btrfs_set_device_start_offset(leaf, dev_item, 0);
1133
1134         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1135         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1136         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1137         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1138         btrfs_mark_buffer_dirty(leaf);
1139
1140         ret = 0;
1141 out:
1142         btrfs_free_path(path);
1143         return ret;
1144 }
1145
1146 static int btrfs_rm_dev_item(struct btrfs_root *root,
1147                              struct btrfs_device *device)
1148 {
1149         int ret;
1150         struct btrfs_path *path;
1151         struct btrfs_key key;
1152         struct btrfs_trans_handle *trans;
1153
1154         root = root->fs_info->chunk_root;
1155
1156         path = btrfs_alloc_path();
1157         if (!path)
1158                 return -ENOMEM;
1159
1160         trans = btrfs_start_transaction(root, 0);
1161         if (IS_ERR(trans)) {
1162                 btrfs_free_path(path);
1163                 return PTR_ERR(trans);
1164         }
1165         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1166         key.type = BTRFS_DEV_ITEM_KEY;
1167         key.offset = device->devid;
1168         lock_chunks(root);
1169
1170         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1171         if (ret < 0)
1172                 goto out;
1173
1174         if (ret > 0) {
1175                 ret = -ENOENT;
1176                 goto out;
1177         }
1178
1179         ret = btrfs_del_item(trans, root, path);
1180         if (ret)
1181                 goto out;
1182 out:
1183         btrfs_free_path(path);
1184         unlock_chunks(root);
1185         btrfs_commit_transaction(trans, root);
1186         return ret;
1187 }
1188
1189 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1190 {
1191         struct btrfs_device *device;
1192         struct btrfs_device *next_device;
1193         struct block_device *bdev;
1194         struct buffer_head *bh = NULL;
1195         struct btrfs_super_block *disk_super;
1196         u64 all_avail;
1197         u64 devid;
1198         u64 num_devices;
1199         u8 *dev_uuid;
1200         int ret = 0;
1201
1202         mutex_lock(&uuid_mutex);
1203         mutex_lock(&root->fs_info->volume_mutex);
1204
1205         all_avail = root->fs_info->avail_data_alloc_bits |
1206                 root->fs_info->avail_system_alloc_bits |
1207                 root->fs_info->avail_metadata_alloc_bits;
1208
1209         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1210             root->fs_info->fs_devices->num_devices <= 4) {
1211                 printk(KERN_ERR "btrfs: unable to go below four devices "
1212                        "on raid10\n");
1213                 ret = -EINVAL;
1214                 goto out;
1215         }
1216
1217         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1218             root->fs_info->fs_devices->num_devices <= 2) {
1219                 printk(KERN_ERR "btrfs: unable to go below two "
1220                        "devices on raid1\n");
1221                 ret = -EINVAL;
1222                 goto out;
1223         }
1224
1225         if (strcmp(device_path, "missing") == 0) {
1226                 struct list_head *devices;
1227                 struct btrfs_device *tmp;
1228
1229                 device = NULL;
1230                 devices = &root->fs_info->fs_devices->devices;
1231                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1232                 list_for_each_entry(tmp, devices, dev_list) {
1233                         if (tmp->in_fs_metadata && !tmp->bdev) {
1234                                 device = tmp;
1235                                 break;
1236                         }
1237                 }
1238                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1239                 bdev = NULL;
1240                 bh = NULL;
1241                 disk_super = NULL;
1242                 if (!device) {
1243                         printk(KERN_ERR "btrfs: no missing devices found to "
1244                                "remove\n");
1245                         goto out;
1246                 }
1247         } else {
1248                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1249                                           root->fs_info->bdev_holder);
1250                 if (IS_ERR(bdev)) {
1251                         ret = PTR_ERR(bdev);
1252                         goto out;
1253                 }
1254
1255                 set_blocksize(bdev, 4096);
1256                 bh = btrfs_read_dev_super(bdev);
1257                 if (!bh) {
1258                         ret = -EINVAL;
1259                         goto error_close;
1260                 }
1261                 disk_super = (struct btrfs_super_block *)bh->b_data;
1262                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1263                 dev_uuid = disk_super->dev_item.uuid;
1264                 device = btrfs_find_device(root, devid, dev_uuid,
1265                                            disk_super->fsid);
1266                 if (!device) {
1267                         ret = -ENOENT;
1268                         goto error_brelse;
1269                 }
1270         }
1271
1272         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1273                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1274                        "device\n");
1275                 ret = -EINVAL;
1276                 goto error_brelse;
1277         }
1278
1279         if (device->writeable) {
1280                 list_del_init(&device->dev_alloc_list);
1281                 root->fs_info->fs_devices->rw_devices--;
1282         }
1283
1284         ret = btrfs_shrink_device(device, 0);
1285         if (ret)
1286                 goto error_undo;
1287
1288         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1289         if (ret)
1290                 goto error_undo;
1291
1292         device->in_fs_metadata = 0;
1293
1294         /*
1295          * the device list mutex makes sure that we don't change
1296          * the device list while someone else is writing out all
1297          * the device supers.
1298          */
1299         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1300         list_del_init(&device->dev_list);
1301         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1302
1303         device->fs_devices->num_devices--;
1304
1305         if (device->missing)
1306                 root->fs_info->fs_devices->missing_devices--;
1307
1308         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1309                                  struct btrfs_device, dev_list);
1310         if (device->bdev == root->fs_info->sb->s_bdev)
1311                 root->fs_info->sb->s_bdev = next_device->bdev;
1312         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1313                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1314
1315         if (device->bdev) {
1316                 blkdev_put(device->bdev, device->mode);
1317                 device->bdev = NULL;
1318                 device->fs_devices->open_devices--;
1319         }
1320
1321         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1322         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1323
1324         if (device->fs_devices->open_devices == 0) {
1325                 struct btrfs_fs_devices *fs_devices;
1326                 fs_devices = root->fs_info->fs_devices;
1327                 while (fs_devices) {
1328                         if (fs_devices->seed == device->fs_devices)
1329                                 break;
1330                         fs_devices = fs_devices->seed;
1331                 }
1332                 fs_devices->seed = device->fs_devices->seed;
1333                 device->fs_devices->seed = NULL;
1334                 __btrfs_close_devices(device->fs_devices);
1335                 free_fs_devices(device->fs_devices);
1336         }
1337
1338         /*
1339          * at this point, the device is zero sized.  We want to
1340          * remove it from the devices list and zero out the old super
1341          */
1342         if (device->writeable) {
1343                 /* make sure this device isn't detected as part of
1344                  * the FS anymore
1345                  */
1346                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1347                 set_buffer_dirty(bh);
1348                 sync_dirty_buffer(bh);
1349         }
1350
1351         kfree(device->name);
1352         kfree(device);
1353         ret = 0;
1354
1355 error_brelse:
1356         brelse(bh);
1357 error_close:
1358         if (bdev)
1359                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1360 out:
1361         mutex_unlock(&root->fs_info->volume_mutex);
1362         mutex_unlock(&uuid_mutex);
1363         return ret;
1364 error_undo:
1365         if (device->writeable) {
1366                 list_add(&device->dev_alloc_list,
1367                          &root->fs_info->fs_devices->alloc_list);
1368                 root->fs_info->fs_devices->rw_devices++;
1369         }
1370         goto error_brelse;
1371 }
1372
1373 /*
1374  * does all the dirty work required for changing file system's UUID.
1375  */
1376 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1377                                 struct btrfs_root *root)
1378 {
1379         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1380         struct btrfs_fs_devices *old_devices;
1381         struct btrfs_fs_devices *seed_devices;
1382         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1383         struct btrfs_device *device;
1384         u64 super_flags;
1385
1386         BUG_ON(!mutex_is_locked(&uuid_mutex));
1387         if (!fs_devices->seeding)
1388                 return -EINVAL;
1389
1390         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1391         if (!seed_devices)
1392                 return -ENOMEM;
1393
1394         old_devices = clone_fs_devices(fs_devices);
1395         if (IS_ERR(old_devices)) {
1396                 kfree(seed_devices);
1397                 return PTR_ERR(old_devices);
1398         }
1399
1400         list_add(&old_devices->list, &fs_uuids);
1401
1402         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1403         seed_devices->opened = 1;
1404         INIT_LIST_HEAD(&seed_devices->devices);
1405         INIT_LIST_HEAD(&seed_devices->alloc_list);
1406         mutex_init(&seed_devices->device_list_mutex);
1407         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1408         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1409         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1410                 device->fs_devices = seed_devices;
1411         }
1412
1413         fs_devices->seeding = 0;
1414         fs_devices->num_devices = 0;
1415         fs_devices->open_devices = 0;
1416         fs_devices->seed = seed_devices;
1417
1418         generate_random_uuid(fs_devices->fsid);
1419         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1420         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1421         super_flags = btrfs_super_flags(disk_super) &
1422                       ~BTRFS_SUPER_FLAG_SEEDING;
1423         btrfs_set_super_flags(disk_super, super_flags);
1424
1425         return 0;
1426 }
1427
1428 /*
1429  * strore the expected generation for seed devices in device items.
1430  */
1431 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1432                                struct btrfs_root *root)
1433 {
1434         struct btrfs_path *path;
1435         struct extent_buffer *leaf;
1436         struct btrfs_dev_item *dev_item;
1437         struct btrfs_device *device;
1438         struct btrfs_key key;
1439         u8 fs_uuid[BTRFS_UUID_SIZE];
1440         u8 dev_uuid[BTRFS_UUID_SIZE];
1441         u64 devid;
1442         int ret;
1443
1444         path = btrfs_alloc_path();
1445         if (!path)
1446                 return -ENOMEM;
1447
1448         root = root->fs_info->chunk_root;
1449         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1450         key.offset = 0;
1451         key.type = BTRFS_DEV_ITEM_KEY;
1452
1453         while (1) {
1454                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1455                 if (ret < 0)
1456                         goto error;
1457
1458                 leaf = path->nodes[0];
1459 next_slot:
1460                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1461                         ret = btrfs_next_leaf(root, path);
1462                         if (ret > 0)
1463                                 break;
1464                         if (ret < 0)
1465                                 goto error;
1466                         leaf = path->nodes[0];
1467                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1468                         btrfs_release_path(root, path);
1469                         continue;
1470                 }
1471
1472                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1473                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1474                     key.type != BTRFS_DEV_ITEM_KEY)
1475                         break;
1476
1477                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1478                                           struct btrfs_dev_item);
1479                 devid = btrfs_device_id(leaf, dev_item);
1480                 read_extent_buffer(leaf, dev_uuid,
1481                                    (unsigned long)btrfs_device_uuid(dev_item),
1482                                    BTRFS_UUID_SIZE);
1483                 read_extent_buffer(leaf, fs_uuid,
1484                                    (unsigned long)btrfs_device_fsid(dev_item),
1485                                    BTRFS_UUID_SIZE);
1486                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1487                 BUG_ON(!device);
1488
1489                 if (device->fs_devices->seeding) {
1490                         btrfs_set_device_generation(leaf, dev_item,
1491                                                     device->generation);
1492                         btrfs_mark_buffer_dirty(leaf);
1493                 }
1494
1495                 path->slots[0]++;
1496                 goto next_slot;
1497         }
1498         ret = 0;
1499 error:
1500         btrfs_free_path(path);
1501         return ret;
1502 }
1503
1504 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1505 {
1506         struct btrfs_trans_handle *trans;
1507         struct btrfs_device *device;
1508         struct block_device *bdev;
1509         struct list_head *devices;
1510         struct super_block *sb = root->fs_info->sb;
1511         u64 total_bytes;
1512         int seeding_dev = 0;
1513         int ret = 0;
1514
1515         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1516                 return -EINVAL;
1517
1518         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1519                                   root->fs_info->bdev_holder);
1520         if (IS_ERR(bdev))
1521                 return PTR_ERR(bdev);
1522
1523         if (root->fs_info->fs_devices->seeding) {
1524                 seeding_dev = 1;
1525                 down_write(&sb->s_umount);
1526                 mutex_lock(&uuid_mutex);
1527         }
1528
1529         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1530         mutex_lock(&root->fs_info->volume_mutex);
1531
1532         devices = &root->fs_info->fs_devices->devices;
1533         /*
1534          * we have the volume lock, so we don't need the extra
1535          * device list mutex while reading the list here.
1536          */
1537         list_for_each_entry(device, devices, dev_list) {
1538                 if (device->bdev == bdev) {
1539                         ret = -EEXIST;
1540                         goto error;
1541                 }
1542         }
1543
1544         device = kzalloc(sizeof(*device), GFP_NOFS);
1545         if (!device) {
1546                 /* we can safely leave the fs_devices entry around */
1547                 ret = -ENOMEM;
1548                 goto error;
1549         }
1550
1551         device->name = kstrdup(device_path, GFP_NOFS);
1552         if (!device->name) {
1553                 kfree(device);
1554                 ret = -ENOMEM;
1555                 goto error;
1556         }
1557
1558         ret = find_next_devid(root, &device->devid);
1559         if (ret) {
1560                 kfree(device->name);
1561                 kfree(device);
1562                 goto error;
1563         }
1564
1565         trans = btrfs_start_transaction(root, 0);
1566         if (IS_ERR(trans)) {
1567                 kfree(device->name);
1568                 kfree(device);
1569                 ret = PTR_ERR(trans);
1570                 goto error;
1571         }
1572
1573         lock_chunks(root);
1574
1575         device->writeable = 1;
1576         device->work.func = pending_bios_fn;
1577         generate_random_uuid(device->uuid);
1578         spin_lock_init(&device->io_lock);
1579         device->generation = trans->transid;
1580         device->io_width = root->sectorsize;
1581         device->io_align = root->sectorsize;
1582         device->sector_size = root->sectorsize;
1583         device->total_bytes = i_size_read(bdev->bd_inode);
1584         device->disk_total_bytes = device->total_bytes;
1585         device->dev_root = root->fs_info->dev_root;
1586         device->bdev = bdev;
1587         device->in_fs_metadata = 1;
1588         device->mode = FMODE_EXCL;
1589         set_blocksize(device->bdev, 4096);
1590
1591         if (seeding_dev) {
1592                 sb->s_flags &= ~MS_RDONLY;
1593                 ret = btrfs_prepare_sprout(trans, root);
1594                 BUG_ON(ret);
1595         }
1596
1597         device->fs_devices = root->fs_info->fs_devices;
1598
1599         /*
1600          * we don't want write_supers to jump in here with our device
1601          * half setup
1602          */
1603         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1604         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1605         list_add(&device->dev_alloc_list,
1606                  &root->fs_info->fs_devices->alloc_list);
1607         root->fs_info->fs_devices->num_devices++;
1608         root->fs_info->fs_devices->open_devices++;
1609         root->fs_info->fs_devices->rw_devices++;
1610         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1611
1612         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1613                 root->fs_info->fs_devices->rotating = 1;
1614
1615         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1616         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1617                                     total_bytes + device->total_bytes);
1618
1619         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1620         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1621                                     total_bytes + 1);
1622         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1623
1624         if (seeding_dev) {
1625                 ret = init_first_rw_device(trans, root, device);
1626                 BUG_ON(ret);
1627                 ret = btrfs_finish_sprout(trans, root);
1628                 BUG_ON(ret);
1629         } else {
1630                 ret = btrfs_add_device(trans, root, device);
1631         }
1632
1633         /*
1634          * we've got more storage, clear any full flags on the space
1635          * infos
1636          */
1637         btrfs_clear_space_info_full(root->fs_info);
1638
1639         unlock_chunks(root);
1640         btrfs_commit_transaction(trans, root);
1641
1642         if (seeding_dev) {
1643                 mutex_unlock(&uuid_mutex);
1644                 up_write(&sb->s_umount);
1645
1646                 ret = btrfs_relocate_sys_chunks(root);
1647                 BUG_ON(ret);
1648         }
1649 out:
1650         mutex_unlock(&root->fs_info->volume_mutex);
1651         return ret;
1652 error:
1653         blkdev_put(bdev, FMODE_EXCL);
1654         if (seeding_dev) {
1655                 mutex_unlock(&uuid_mutex);
1656                 up_write(&sb->s_umount);
1657         }
1658         goto out;
1659 }
1660
1661 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1662                                         struct btrfs_device *device)
1663 {
1664         int ret;
1665         struct btrfs_path *path;
1666         struct btrfs_root *root;
1667         struct btrfs_dev_item *dev_item;
1668         struct extent_buffer *leaf;
1669         struct btrfs_key key;
1670
1671         root = device->dev_root->fs_info->chunk_root;
1672
1673         path = btrfs_alloc_path();
1674         if (!path)
1675                 return -ENOMEM;
1676
1677         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1678         key.type = BTRFS_DEV_ITEM_KEY;
1679         key.offset = device->devid;
1680
1681         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1682         if (ret < 0)
1683                 goto out;
1684
1685         if (ret > 0) {
1686                 ret = -ENOENT;
1687                 goto out;
1688         }
1689
1690         leaf = path->nodes[0];
1691         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1692
1693         btrfs_set_device_id(leaf, dev_item, device->devid);
1694         btrfs_set_device_type(leaf, dev_item, device->type);
1695         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1696         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1697         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1698         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1699         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1700         btrfs_mark_buffer_dirty(leaf);
1701
1702 out:
1703         btrfs_free_path(path);
1704         return ret;
1705 }
1706
1707 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1708                       struct btrfs_device *device, u64 new_size)
1709 {
1710         struct btrfs_super_block *super_copy =
1711                 &device->dev_root->fs_info->super_copy;
1712         u64 old_total = btrfs_super_total_bytes(super_copy);
1713         u64 diff = new_size - device->total_bytes;
1714
1715         if (!device->writeable)
1716                 return -EACCES;
1717         if (new_size <= device->total_bytes)
1718                 return -EINVAL;
1719
1720         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1721         device->fs_devices->total_rw_bytes += diff;
1722
1723         device->total_bytes = new_size;
1724         device->disk_total_bytes = new_size;
1725         btrfs_clear_space_info_full(device->dev_root->fs_info);
1726
1727         return btrfs_update_device(trans, device);
1728 }
1729
1730 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1731                       struct btrfs_device *device, u64 new_size)
1732 {
1733         int ret;
1734         lock_chunks(device->dev_root);
1735         ret = __btrfs_grow_device(trans, device, new_size);
1736         unlock_chunks(device->dev_root);
1737         return ret;
1738 }
1739
1740 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1741                             struct btrfs_root *root,
1742                             u64 chunk_tree, u64 chunk_objectid,
1743                             u64 chunk_offset)
1744 {
1745         int ret;
1746         struct btrfs_path *path;
1747         struct btrfs_key key;
1748
1749         root = root->fs_info->chunk_root;
1750         path = btrfs_alloc_path();
1751         if (!path)
1752                 return -ENOMEM;
1753
1754         key.objectid = chunk_objectid;
1755         key.offset = chunk_offset;
1756         key.type = BTRFS_CHUNK_ITEM_KEY;
1757
1758         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1759         BUG_ON(ret);
1760
1761         ret = btrfs_del_item(trans, root, path);
1762         BUG_ON(ret);
1763
1764         btrfs_free_path(path);
1765         return 0;
1766 }
1767
1768 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1769                         chunk_offset)
1770 {
1771         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1772         struct btrfs_disk_key *disk_key;
1773         struct btrfs_chunk *chunk;
1774         u8 *ptr;
1775         int ret = 0;
1776         u32 num_stripes;
1777         u32 array_size;
1778         u32 len = 0;
1779         u32 cur;
1780         struct btrfs_key key;
1781
1782         array_size = btrfs_super_sys_array_size(super_copy);
1783
1784         ptr = super_copy->sys_chunk_array;
1785         cur = 0;
1786
1787         while (cur < array_size) {
1788                 disk_key = (struct btrfs_disk_key *)ptr;
1789                 btrfs_disk_key_to_cpu(&key, disk_key);
1790
1791                 len = sizeof(*disk_key);
1792
1793                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1794                         chunk = (struct btrfs_chunk *)(ptr + len);
1795                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1796                         len += btrfs_chunk_item_size(num_stripes);
1797                 } else {
1798                         ret = -EIO;
1799                         break;
1800                 }
1801                 if (key.objectid == chunk_objectid &&
1802                     key.offset == chunk_offset) {
1803                         memmove(ptr, ptr + len, array_size - (cur + len));
1804                         array_size -= len;
1805                         btrfs_set_super_sys_array_size(super_copy, array_size);
1806                 } else {
1807                         ptr += len;
1808                         cur += len;
1809                 }
1810         }
1811         return ret;
1812 }
1813
1814 static int btrfs_relocate_chunk(struct btrfs_root *root,
1815                          u64 chunk_tree, u64 chunk_objectid,
1816                          u64 chunk_offset)
1817 {
1818         struct extent_map_tree *em_tree;
1819         struct btrfs_root *extent_root;
1820         struct btrfs_trans_handle *trans;
1821         struct extent_map *em;
1822         struct map_lookup *map;
1823         int ret;
1824         int i;
1825
1826         root = root->fs_info->chunk_root;
1827         extent_root = root->fs_info->extent_root;
1828         em_tree = &root->fs_info->mapping_tree.map_tree;
1829
1830         ret = btrfs_can_relocate(extent_root, chunk_offset);
1831         if (ret)
1832                 return -ENOSPC;
1833
1834         /* step one, relocate all the extents inside this chunk */
1835         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1836         if (ret)
1837                 return ret;
1838
1839         trans = btrfs_start_transaction(root, 0);
1840         BUG_ON(IS_ERR(trans));
1841
1842         lock_chunks(root);
1843
1844         /*
1845          * step two, delete the device extents and the
1846          * chunk tree entries
1847          */
1848         read_lock(&em_tree->lock);
1849         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1850         read_unlock(&em_tree->lock);
1851
1852         BUG_ON(em->start > chunk_offset ||
1853                em->start + em->len < chunk_offset);
1854         map = (struct map_lookup *)em->bdev;
1855
1856         for (i = 0; i < map->num_stripes; i++) {
1857                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1858                                             map->stripes[i].physical);
1859                 BUG_ON(ret);
1860
1861                 if (map->stripes[i].dev) {
1862                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1863                         BUG_ON(ret);
1864                 }
1865         }
1866         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1867                                chunk_offset);
1868
1869         BUG_ON(ret);
1870
1871         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1872
1873         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1874                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1875                 BUG_ON(ret);
1876         }
1877
1878         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1879         BUG_ON(ret);
1880
1881         write_lock(&em_tree->lock);
1882         remove_extent_mapping(em_tree, em);
1883         write_unlock(&em_tree->lock);
1884
1885         kfree(map);
1886         em->bdev = NULL;
1887
1888         /* once for the tree */
1889         free_extent_map(em);
1890         /* once for us */
1891         free_extent_map(em);
1892
1893         unlock_chunks(root);
1894         btrfs_end_transaction(trans, root);
1895         return 0;
1896 }
1897
1898 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1899 {
1900         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1901         struct btrfs_path *path;
1902         struct extent_buffer *leaf;
1903         struct btrfs_chunk *chunk;
1904         struct btrfs_key key;
1905         struct btrfs_key found_key;
1906         u64 chunk_tree = chunk_root->root_key.objectid;
1907         u64 chunk_type;
1908         bool retried = false;
1909         int failed = 0;
1910         int ret;
1911
1912         path = btrfs_alloc_path();
1913         if (!path)
1914                 return -ENOMEM;
1915
1916 again:
1917         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1918         key.offset = (u64)-1;
1919         key.type = BTRFS_CHUNK_ITEM_KEY;
1920
1921         while (1) {
1922                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1923                 if (ret < 0)
1924                         goto error;
1925                 BUG_ON(ret == 0);
1926
1927                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1928                                           key.type);
1929                 if (ret < 0)
1930                         goto error;
1931                 if (ret > 0)
1932                         break;
1933
1934                 leaf = path->nodes[0];
1935                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1936
1937                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1938                                        struct btrfs_chunk);
1939                 chunk_type = btrfs_chunk_type(leaf, chunk);
1940                 btrfs_release_path(chunk_root, path);
1941
1942                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1943                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1944                                                    found_key.objectid,
1945                                                    found_key.offset);
1946                         if (ret == -ENOSPC)
1947                                 failed++;
1948                         else if (ret)
1949                                 BUG();
1950                 }
1951
1952                 if (found_key.offset == 0)
1953                         break;
1954                 key.offset = found_key.offset - 1;
1955         }
1956         ret = 0;
1957         if (failed && !retried) {
1958                 failed = 0;
1959                 retried = true;
1960                 goto again;
1961         } else if (failed && retried) {
1962                 WARN_ON(1);
1963                 ret = -ENOSPC;
1964         }
1965 error:
1966         btrfs_free_path(path);
1967         return ret;
1968 }
1969
1970 static u64 div_factor(u64 num, int factor)
1971 {
1972         if (factor == 10)
1973                 return num;
1974         num *= factor;
1975         do_div(num, 10);
1976         return num;
1977 }
1978
1979 int btrfs_balance(struct btrfs_root *dev_root)
1980 {
1981         int ret;
1982         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1983         struct btrfs_device *device;
1984         u64 old_size;
1985         u64 size_to_free;
1986         struct btrfs_path *path;
1987         struct btrfs_key key;
1988         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1989         struct btrfs_trans_handle *trans;
1990         struct btrfs_key found_key;
1991
1992         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1993                 return -EROFS;
1994
1995         if (!capable(CAP_SYS_ADMIN))
1996                 return -EPERM;
1997
1998         mutex_lock(&dev_root->fs_info->volume_mutex);
1999         dev_root = dev_root->fs_info->dev_root;
2000
2001         /* step one make some room on all the devices */
2002         list_for_each_entry(device, devices, dev_list) {
2003                 old_size = device->total_bytes;
2004                 size_to_free = div_factor(old_size, 1);
2005                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2006                 if (!device->writeable ||
2007                     device->total_bytes - device->bytes_used > size_to_free)
2008                         continue;
2009
2010                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2011                 if (ret == -ENOSPC)
2012                         break;
2013                 BUG_ON(ret);
2014
2015                 trans = btrfs_start_transaction(dev_root, 0);
2016                 BUG_ON(IS_ERR(trans));
2017
2018                 ret = btrfs_grow_device(trans, device, old_size);
2019                 BUG_ON(ret);
2020
2021                 btrfs_end_transaction(trans, dev_root);
2022         }
2023
2024         /* step two, relocate all the chunks */
2025         path = btrfs_alloc_path();
2026         BUG_ON(!path);
2027
2028         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2029         key.offset = (u64)-1;
2030         key.type = BTRFS_CHUNK_ITEM_KEY;
2031
2032         while (1) {
2033                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2034                 if (ret < 0)
2035                         goto error;
2036
2037                 /*
2038                  * this shouldn't happen, it means the last relocate
2039                  * failed
2040                  */
2041                 if (ret == 0)
2042                         break;
2043
2044                 ret = btrfs_previous_item(chunk_root, path, 0,
2045                                           BTRFS_CHUNK_ITEM_KEY);
2046                 if (ret)
2047                         break;
2048
2049                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2050                                       path->slots[0]);
2051                 if (found_key.objectid != key.objectid)
2052                         break;
2053
2054                 /* chunk zero is special */
2055                 if (found_key.offset == 0)
2056                         break;
2057
2058                 btrfs_release_path(chunk_root, path);
2059                 ret = btrfs_relocate_chunk(chunk_root,
2060                                            chunk_root->root_key.objectid,
2061                                            found_key.objectid,
2062                                            found_key.offset);
2063                 BUG_ON(ret && ret != -ENOSPC);
2064                 key.offset = found_key.offset - 1;
2065         }
2066         ret = 0;
2067 error:
2068         btrfs_free_path(path);
2069         mutex_unlock(&dev_root->fs_info->volume_mutex);
2070         return ret;
2071 }
2072
2073 /*
2074  * shrinking a device means finding all of the device extents past
2075  * the new size, and then following the back refs to the chunks.
2076  * The chunk relocation code actually frees the device extent
2077  */
2078 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2079 {
2080         struct btrfs_trans_handle *trans;
2081         struct btrfs_root *root = device->dev_root;
2082         struct btrfs_dev_extent *dev_extent = NULL;
2083         struct btrfs_path *path;
2084         u64 length;
2085         u64 chunk_tree;
2086         u64 chunk_objectid;
2087         u64 chunk_offset;
2088         int ret;
2089         int slot;
2090         int failed = 0;
2091         bool retried = false;
2092         struct extent_buffer *l;
2093         struct btrfs_key key;
2094         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2095         u64 old_total = btrfs_super_total_bytes(super_copy);
2096         u64 old_size = device->total_bytes;
2097         u64 diff = device->total_bytes - new_size;
2098
2099         if (new_size >= device->total_bytes)
2100                 return -EINVAL;
2101
2102         path = btrfs_alloc_path();
2103         if (!path)
2104                 return -ENOMEM;
2105
2106         path->reada = 2;
2107
2108         lock_chunks(root);
2109
2110         device->total_bytes = new_size;
2111         if (device->writeable)
2112                 device->fs_devices->total_rw_bytes -= diff;
2113         unlock_chunks(root);
2114
2115 again:
2116         key.objectid = device->devid;
2117         key.offset = (u64)-1;
2118         key.type = BTRFS_DEV_EXTENT_KEY;
2119
2120         while (1) {
2121                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2122                 if (ret < 0)
2123                         goto done;
2124
2125                 ret = btrfs_previous_item(root, path, 0, key.type);
2126                 if (ret < 0)
2127                         goto done;
2128                 if (ret) {
2129                         ret = 0;
2130                         btrfs_release_path(root, path);
2131                         break;
2132                 }
2133
2134                 l = path->nodes[0];
2135                 slot = path->slots[0];
2136                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2137
2138                 if (key.objectid != device->devid) {
2139                         btrfs_release_path(root, path);
2140                         break;
2141                 }
2142
2143                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2144                 length = btrfs_dev_extent_length(l, dev_extent);
2145
2146                 if (key.offset + length <= new_size) {
2147                         btrfs_release_path(root, path);
2148                         break;
2149                 }
2150
2151                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2152                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2153                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2154                 btrfs_release_path(root, path);
2155
2156                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2157                                            chunk_offset);
2158                 if (ret && ret != -ENOSPC)
2159                         goto done;
2160                 if (ret == -ENOSPC)
2161                         failed++;
2162                 key.offset -= 1;
2163         }
2164
2165         if (failed && !retried) {
2166                 failed = 0;
2167                 retried = true;
2168                 goto again;
2169         } else if (failed && retried) {
2170                 ret = -ENOSPC;
2171                 lock_chunks(root);
2172
2173                 device->total_bytes = old_size;
2174                 if (device->writeable)
2175                         device->fs_devices->total_rw_bytes += diff;
2176                 unlock_chunks(root);
2177                 goto done;
2178         }
2179
2180         /* Shrinking succeeded, else we would be at "done". */
2181         trans = btrfs_start_transaction(root, 0);
2182         if (IS_ERR(trans)) {
2183                 ret = PTR_ERR(trans);
2184                 goto done;
2185         }
2186
2187         lock_chunks(root);
2188
2189         device->disk_total_bytes = new_size;
2190         /* Now btrfs_update_device() will change the on-disk size. */
2191         ret = btrfs_update_device(trans, device);
2192         if (ret) {
2193                 unlock_chunks(root);
2194                 btrfs_end_transaction(trans, root);
2195                 goto done;
2196         }
2197         WARN_ON(diff > old_total);
2198         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2199         unlock_chunks(root);
2200         btrfs_end_transaction(trans, root);
2201 done:
2202         btrfs_free_path(path);
2203         return ret;
2204 }
2205
2206 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2207                            struct btrfs_root *root,
2208                            struct btrfs_key *key,
2209                            struct btrfs_chunk *chunk, int item_size)
2210 {
2211         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2212         struct btrfs_disk_key disk_key;
2213         u32 array_size;
2214         u8 *ptr;
2215
2216         array_size = btrfs_super_sys_array_size(super_copy);
2217         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2218                 return -EFBIG;
2219
2220         ptr = super_copy->sys_chunk_array + array_size;
2221         btrfs_cpu_key_to_disk(&disk_key, key);
2222         memcpy(ptr, &disk_key, sizeof(disk_key));
2223         ptr += sizeof(disk_key);
2224         memcpy(ptr, chunk, item_size);
2225         item_size += sizeof(disk_key);
2226         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2227         return 0;
2228 }
2229
2230 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2231                                         int num_stripes, int sub_stripes)
2232 {
2233         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2234                 return calc_size;
2235         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2236                 return calc_size * (num_stripes / sub_stripes);
2237         else
2238                 return calc_size * num_stripes;
2239 }
2240
2241 /* Used to sort the devices by max_avail(descending sort) */
2242 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2243 {
2244         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2245             ((struct btrfs_device_info *)dev_info2)->max_avail)
2246                 return -1;
2247         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2248                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2249                 return 1;
2250         else
2251                 return 0;
2252 }
2253
2254 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2255                                  int *num_stripes, int *min_stripes,
2256                                  int *sub_stripes)
2257 {
2258         *num_stripes = 1;
2259         *min_stripes = 1;
2260         *sub_stripes = 0;
2261
2262         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2263                 *num_stripes = fs_devices->rw_devices;
2264                 *min_stripes = 2;
2265         }
2266         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2267                 *num_stripes = 2;
2268                 *min_stripes = 2;
2269         }
2270         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2271                 if (fs_devices->rw_devices < 2)
2272                         return -ENOSPC;
2273                 *num_stripes = 2;
2274                 *min_stripes = 2;
2275         }
2276         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2277                 *num_stripes = fs_devices->rw_devices;
2278                 if (*num_stripes < 4)
2279                         return -ENOSPC;
2280                 *num_stripes &= ~(u32)1;
2281                 *sub_stripes = 2;
2282                 *min_stripes = 4;
2283         }
2284
2285         return 0;
2286 }
2287
2288 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2289                                     u64 proposed_size, u64 type,
2290                                     int num_stripes, int small_stripe)
2291 {
2292         int min_stripe_size = 1 * 1024 * 1024;
2293         u64 calc_size = proposed_size;
2294         u64 max_chunk_size = calc_size;
2295         int ncopies = 1;
2296
2297         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2298                     BTRFS_BLOCK_GROUP_DUP |
2299                     BTRFS_BLOCK_GROUP_RAID10))
2300                 ncopies = 2;
2301
2302         if (type & BTRFS_BLOCK_GROUP_DATA) {
2303                 max_chunk_size = 10 * calc_size;
2304                 min_stripe_size = 64 * 1024 * 1024;
2305         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2306                 max_chunk_size = 256 * 1024 * 1024;
2307                 min_stripe_size = 32 * 1024 * 1024;
2308         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2309                 calc_size = 8 * 1024 * 1024;
2310                 max_chunk_size = calc_size * 2;
2311                 min_stripe_size = 1 * 1024 * 1024;
2312         }
2313
2314         /* we don't want a chunk larger than 10% of writeable space */
2315         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2316                              max_chunk_size);
2317
2318         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2319                 calc_size = max_chunk_size * ncopies;
2320                 do_div(calc_size, num_stripes);
2321                 do_div(calc_size, BTRFS_STRIPE_LEN);
2322                 calc_size *= BTRFS_STRIPE_LEN;
2323         }
2324
2325         /* we don't want tiny stripes */
2326         if (!small_stripe)
2327                 calc_size = max_t(u64, min_stripe_size, calc_size);
2328
2329         /*
2330          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2331          * we end up with something bigger than a stripe
2332          */
2333         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2334
2335         do_div(calc_size, BTRFS_STRIPE_LEN);
2336         calc_size *= BTRFS_STRIPE_LEN;
2337
2338         return calc_size;
2339 }
2340
2341 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2342                                                       int num_stripes)
2343 {
2344         struct map_lookup *new;
2345         size_t len = map_lookup_size(num_stripes);
2346
2347         BUG_ON(map->num_stripes < num_stripes);
2348
2349         if (map->num_stripes == num_stripes)
2350                 return map;
2351
2352         new = kmalloc(len, GFP_NOFS);
2353         if (!new) {
2354                 /* just change map->num_stripes */
2355                 map->num_stripes = num_stripes;
2356                 return map;
2357         }
2358
2359         memcpy(new, map, len);
2360         new->num_stripes = num_stripes;
2361         kfree(map);
2362         return new;
2363 }
2364
2365 /*
2366  * helper to allocate device space from btrfs_device_info, in which we stored
2367  * max free space information of every device. It is used when we can not
2368  * allocate chunks by default size.
2369  *
2370  * By this helper, we can allocate a new chunk as larger as possible.
2371  */
2372 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2373                                     struct btrfs_fs_devices *fs_devices,
2374                                     struct btrfs_device_info *devices,
2375                                     int nr_device, u64 type,
2376                                     struct map_lookup **map_lookup,
2377                                     int min_stripes, u64 *stripe_size)
2378 {
2379         int i, index, sort_again = 0;
2380         int min_devices = min_stripes;
2381         u64 max_avail, min_free;
2382         struct map_lookup *map = *map_lookup;
2383         int ret;
2384
2385         if (nr_device < min_stripes)
2386                 return -ENOSPC;
2387
2388         btrfs_descending_sort_devices(devices, nr_device);
2389
2390         max_avail = devices[0].max_avail;
2391         if (!max_avail)
2392                 return -ENOSPC;
2393
2394         for (i = 0; i < nr_device; i++) {
2395                 /*
2396                  * if dev_offset = 0, it means the free space of this device
2397                  * is less than what we need, and we didn't search max avail
2398                  * extent on this device, so do it now.
2399                  */
2400                 if (!devices[i].dev_offset) {
2401                         ret = find_free_dev_extent(trans, devices[i].dev,
2402                                                    max_avail,
2403                                                    &devices[i].dev_offset,
2404                                                    &devices[i].max_avail);
2405                         if (ret != 0 && ret != -ENOSPC)
2406                                 return ret;
2407                         sort_again = 1;
2408                 }
2409         }
2410
2411         /* we update the max avail free extent of each devices, sort again */
2412         if (sort_again)
2413                 btrfs_descending_sort_devices(devices, nr_device);
2414
2415         if (type & BTRFS_BLOCK_GROUP_DUP)
2416                 min_devices = 1;
2417
2418         if (!devices[min_devices - 1].max_avail)
2419                 return -ENOSPC;
2420
2421         max_avail = devices[min_devices - 1].max_avail;
2422         if (type & BTRFS_BLOCK_GROUP_DUP)
2423                 do_div(max_avail, 2);
2424
2425         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2426                                              min_stripes, 1);
2427         if (type & BTRFS_BLOCK_GROUP_DUP)
2428                 min_free = max_avail * 2;
2429         else
2430                 min_free = max_avail;
2431
2432         if (min_free > devices[min_devices - 1].max_avail)
2433                 return -ENOSPC;
2434
2435         map = __shrink_map_lookup_stripes(map, min_stripes);
2436         *stripe_size = max_avail;
2437
2438         index = 0;
2439         for (i = 0; i < min_stripes; i++) {
2440                 map->stripes[i].dev = devices[index].dev;
2441                 map->stripes[i].physical = devices[index].dev_offset;
2442                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2443                         i++;
2444                         map->stripes[i].dev = devices[index].dev;
2445                         map->stripes[i].physical = devices[index].dev_offset +
2446                                                    max_avail;
2447                 }
2448                 index++;
2449         }
2450         *map_lookup = map;
2451
2452         return 0;
2453 }
2454
2455 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2456                                struct btrfs_root *extent_root,
2457                                struct map_lookup **map_ret,
2458                                u64 *num_bytes, u64 *stripe_size,
2459                                u64 start, u64 type)
2460 {
2461         struct btrfs_fs_info *info = extent_root->fs_info;
2462         struct btrfs_device *device = NULL;
2463         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2464         struct list_head *cur;
2465         struct map_lookup *map;
2466         struct extent_map_tree *em_tree;
2467         struct extent_map *em;
2468         struct btrfs_device_info *devices_info;
2469         struct list_head private_devs;
2470         u64 calc_size = 1024 * 1024 * 1024;
2471         u64 min_free;
2472         u64 avail;
2473         u64 dev_offset;
2474         int num_stripes;
2475         int min_stripes;
2476         int sub_stripes;
2477         int min_devices;        /* the min number of devices we need */
2478         int i;
2479         int ret;
2480         int index;
2481
2482         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2483             (type & BTRFS_BLOCK_GROUP_DUP)) {
2484                 WARN_ON(1);
2485                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2486         }
2487         if (list_empty(&fs_devices->alloc_list))
2488                 return -ENOSPC;
2489
2490         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2491                                     &min_stripes, &sub_stripes);
2492         if (ret)
2493                 return ret;
2494
2495         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2496                                GFP_NOFS);
2497         if (!devices_info)
2498                 return -ENOMEM;
2499
2500         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2501         if (!map) {
2502                 ret = -ENOMEM;
2503                 goto error;
2504         }
2505         map->num_stripes = num_stripes;
2506
2507         cur = fs_devices->alloc_list.next;
2508         index = 0;
2509         i = 0;
2510
2511         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2512                                              num_stripes, 0);
2513
2514         if (type & BTRFS_BLOCK_GROUP_DUP) {
2515                 min_free = calc_size * 2;
2516                 min_devices = 1;
2517         } else {
2518                 min_free = calc_size;
2519                 min_devices = min_stripes;
2520         }
2521
2522         INIT_LIST_HEAD(&private_devs);
2523         while (index < num_stripes) {
2524                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2525                 BUG_ON(!device->writeable);
2526                 if (device->total_bytes > device->bytes_used)
2527                         avail = device->total_bytes - device->bytes_used;
2528                 else
2529                         avail = 0;
2530                 cur = cur->next;
2531
2532                 if (device->in_fs_metadata && avail >= min_free) {
2533                         ret = find_free_dev_extent(trans, device, min_free,
2534                                                    &devices_info[i].dev_offset,
2535                                                    &devices_info[i].max_avail);
2536                         if (ret == 0) {
2537                                 list_move_tail(&device->dev_alloc_list,
2538                                                &private_devs);
2539                                 map->stripes[index].dev = device;
2540                                 map->stripes[index].physical =
2541                                                 devices_info[i].dev_offset;
2542                                 index++;
2543                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2544                                         map->stripes[index].dev = device;
2545                                         map->stripes[index].physical =
2546                                                 devices_info[i].dev_offset +
2547                                                 calc_size;
2548                                         index++;
2549                                 }
2550                         } else if (ret != -ENOSPC)
2551                                 goto error;
2552
2553                         devices_info[i].dev = device;
2554                         i++;
2555                 } else if (device->in_fs_metadata &&
2556                            avail >= BTRFS_STRIPE_LEN) {
2557                         devices_info[i].dev = device;
2558                         devices_info[i].max_avail = avail;
2559                         i++;
2560                 }
2561
2562                 if (cur == &fs_devices->alloc_list)
2563                         break;
2564         }
2565
2566         list_splice(&private_devs, &fs_devices->alloc_list);
2567         if (index < num_stripes) {
2568                 if (index >= min_stripes) {
2569                         num_stripes = index;
2570                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2571                                 num_stripes /= sub_stripes;
2572                                 num_stripes *= sub_stripes;
2573                         }
2574
2575                         map = __shrink_map_lookup_stripes(map, num_stripes);
2576                 } else if (i >= min_devices) {
2577                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2578                                                        devices_info, i, type,
2579                                                        &map, min_stripes,
2580                                                        &calc_size);
2581                         if (ret)
2582                                 goto error;
2583                 } else {
2584                         ret = -ENOSPC;
2585                         goto error;
2586                 }
2587         }
2588         map->sector_size = extent_root->sectorsize;
2589         map->stripe_len = BTRFS_STRIPE_LEN;
2590         map->io_align = BTRFS_STRIPE_LEN;
2591         map->io_width = BTRFS_STRIPE_LEN;
2592         map->type = type;
2593         map->sub_stripes = sub_stripes;
2594
2595         *map_ret = map;
2596         *stripe_size = calc_size;
2597         *num_bytes = chunk_bytes_by_type(type, calc_size,
2598                                          map->num_stripes, sub_stripes);
2599
2600         trace_btrfs_chunk_alloc(info->chunk_root, map, start, *num_bytes);
2601
2602         em = alloc_extent_map(GFP_NOFS);
2603         if (!em) {
2604                 ret = -ENOMEM;
2605                 goto error;
2606         }
2607         em->bdev = (struct block_device *)map;
2608         em->start = start;
2609         em->len = *num_bytes;
2610         em->block_start = 0;
2611         em->block_len = em->len;
2612
2613         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2614         write_lock(&em_tree->lock);
2615         ret = add_extent_mapping(em_tree, em);
2616         write_unlock(&em_tree->lock);
2617         BUG_ON(ret);
2618         free_extent_map(em);
2619
2620         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2621                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2622                                      start, *num_bytes);
2623         BUG_ON(ret);
2624
2625         index = 0;
2626         while (index < map->num_stripes) {
2627                 device = map->stripes[index].dev;
2628                 dev_offset = map->stripes[index].physical;
2629
2630                 ret = btrfs_alloc_dev_extent(trans, device,
2631                                 info->chunk_root->root_key.objectid,
2632                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2633                                 start, dev_offset, calc_size);
2634                 BUG_ON(ret);
2635                 index++;
2636         }
2637
2638         kfree(devices_info);
2639         return 0;
2640
2641 error:
2642         kfree(map);
2643         kfree(devices_info);
2644         return ret;
2645 }
2646
2647 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2648                                 struct btrfs_root *extent_root,
2649                                 struct map_lookup *map, u64 chunk_offset,
2650                                 u64 chunk_size, u64 stripe_size)
2651 {
2652         u64 dev_offset;
2653         struct btrfs_key key;
2654         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2655         struct btrfs_device *device;
2656         struct btrfs_chunk *chunk;
2657         struct btrfs_stripe *stripe;
2658         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2659         int index = 0;
2660         int ret;
2661
2662         chunk = kzalloc(item_size, GFP_NOFS);
2663         if (!chunk)
2664                 return -ENOMEM;
2665
2666         index = 0;
2667         while (index < map->num_stripes) {
2668                 device = map->stripes[index].dev;
2669                 device->bytes_used += stripe_size;
2670                 ret = btrfs_update_device(trans, device);
2671                 BUG_ON(ret);
2672                 index++;
2673         }
2674
2675         index = 0;
2676         stripe = &chunk->stripe;
2677         while (index < map->num_stripes) {
2678                 device = map->stripes[index].dev;
2679                 dev_offset = map->stripes[index].physical;
2680
2681                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2682                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2683                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2684                 stripe++;
2685                 index++;
2686         }
2687
2688         btrfs_set_stack_chunk_length(chunk, chunk_size);
2689         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2690         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2691         btrfs_set_stack_chunk_type(chunk, map->type);
2692         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2693         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2694         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2695         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2696         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2697
2698         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2699         key.type = BTRFS_CHUNK_ITEM_KEY;
2700         key.offset = chunk_offset;
2701
2702         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2703         BUG_ON(ret);
2704
2705         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2706                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2707                                              item_size);
2708                 BUG_ON(ret);
2709         }
2710
2711         kfree(chunk);
2712         return 0;
2713 }
2714
2715 /*
2716  * Chunk allocation falls into two parts. The first part does works
2717  * that make the new allocated chunk useable, but not do any operation
2718  * that modifies the chunk tree. The second part does the works that
2719  * require modifying the chunk tree. This division is important for the
2720  * bootstrap process of adding storage to a seed btrfs.
2721  */
2722 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2723                       struct btrfs_root *extent_root, u64 type)
2724 {
2725         u64 chunk_offset;
2726         u64 chunk_size;
2727         u64 stripe_size;
2728         struct map_lookup *map;
2729         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2730         int ret;
2731
2732         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2733                               &chunk_offset);
2734         if (ret)
2735                 return ret;
2736
2737         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2738                                   &stripe_size, chunk_offset, type);
2739         if (ret)
2740                 return ret;
2741
2742         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2743                                    chunk_size, stripe_size);
2744         BUG_ON(ret);
2745         return 0;
2746 }
2747
2748 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2749                                          struct btrfs_root *root,
2750                                          struct btrfs_device *device)
2751 {
2752         u64 chunk_offset;
2753         u64 sys_chunk_offset;
2754         u64 chunk_size;
2755         u64 sys_chunk_size;
2756         u64 stripe_size;
2757         u64 sys_stripe_size;
2758         u64 alloc_profile;
2759         struct map_lookup *map;
2760         struct map_lookup *sys_map;
2761         struct btrfs_fs_info *fs_info = root->fs_info;
2762         struct btrfs_root *extent_root = fs_info->extent_root;
2763         int ret;
2764
2765         ret = find_next_chunk(fs_info->chunk_root,
2766                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2767         BUG_ON(ret);
2768
2769         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2770                         (fs_info->metadata_alloc_profile &
2771                          fs_info->avail_metadata_alloc_bits);
2772         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2773
2774         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2775                                   &stripe_size, chunk_offset, alloc_profile);
2776         BUG_ON(ret);
2777
2778         sys_chunk_offset = chunk_offset + chunk_size;
2779
2780         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2781                         (fs_info->system_alloc_profile &
2782                          fs_info->avail_system_alloc_bits);
2783         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2784
2785         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2786                                   &sys_chunk_size, &sys_stripe_size,
2787                                   sys_chunk_offset, alloc_profile);
2788         BUG_ON(ret);
2789
2790         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2791         BUG_ON(ret);
2792
2793         /*
2794          * Modifying chunk tree needs allocating new blocks from both
2795          * system block group and metadata block group. So we only can
2796          * do operations require modifying the chunk tree after both
2797          * block groups were created.
2798          */
2799         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2800                                    chunk_size, stripe_size);
2801         BUG_ON(ret);
2802
2803         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2804                                    sys_chunk_offset, sys_chunk_size,
2805                                    sys_stripe_size);
2806         BUG_ON(ret);
2807         return 0;
2808 }
2809
2810 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2811 {
2812         struct extent_map *em;
2813         struct map_lookup *map;
2814         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2815         int readonly = 0;
2816         int i;
2817
2818         read_lock(&map_tree->map_tree.lock);
2819         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2820         read_unlock(&map_tree->map_tree.lock);
2821         if (!em)
2822                 return 1;
2823
2824         if (btrfs_test_opt(root, DEGRADED)) {
2825                 free_extent_map(em);
2826                 return 0;
2827         }
2828
2829         map = (struct map_lookup *)em->bdev;
2830         for (i = 0; i < map->num_stripes; i++) {
2831                 if (!map->stripes[i].dev->writeable) {
2832                         readonly = 1;
2833                         break;
2834                 }
2835         }
2836         free_extent_map(em);
2837         return readonly;
2838 }
2839
2840 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2841 {
2842         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2843 }
2844
2845 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2846 {
2847         struct extent_map *em;
2848
2849         while (1) {
2850                 write_lock(&tree->map_tree.lock);
2851                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2852                 if (em)
2853                         remove_extent_mapping(&tree->map_tree, em);
2854                 write_unlock(&tree->map_tree.lock);
2855                 if (!em)
2856                         break;
2857                 kfree(em->bdev);
2858                 /* once for us */
2859                 free_extent_map(em);
2860                 /* once for the tree */
2861                 free_extent_map(em);
2862         }
2863 }
2864
2865 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2866 {
2867         struct extent_map *em;
2868         struct map_lookup *map;
2869         struct extent_map_tree *em_tree = &map_tree->map_tree;
2870         int ret;
2871
2872         read_lock(&em_tree->lock);
2873         em = lookup_extent_mapping(em_tree, logical, len);
2874         read_unlock(&em_tree->lock);
2875         BUG_ON(!em);
2876
2877         BUG_ON(em->start > logical || em->start + em->len < logical);
2878         map = (struct map_lookup *)em->bdev;
2879         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2880                 ret = map->num_stripes;
2881         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2882                 ret = map->sub_stripes;
2883         else
2884                 ret = 1;
2885         free_extent_map(em);
2886         return ret;
2887 }
2888
2889 static int find_live_mirror(struct map_lookup *map, int first, int num,
2890                             int optimal)
2891 {
2892         int i;
2893         if (map->stripes[optimal].dev->bdev)
2894                 return optimal;
2895         for (i = first; i < first + num; i++) {
2896                 if (map->stripes[i].dev->bdev)
2897                         return i;
2898         }
2899         /* we couldn't find one that doesn't fail.  Just return something
2900          * and the io error handling code will clean up eventually
2901          */
2902         return optimal;
2903 }
2904
2905 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2906                              u64 logical, u64 *length,
2907                              struct btrfs_multi_bio **multi_ret,
2908                              int mirror_num)
2909 {
2910         struct extent_map *em;
2911         struct map_lookup *map;
2912         struct extent_map_tree *em_tree = &map_tree->map_tree;
2913         u64 offset;
2914         u64 stripe_offset;
2915         u64 stripe_end_offset;
2916         u64 stripe_nr;
2917         u64 stripe_nr_orig;
2918         u64 stripe_nr_end;
2919         int stripes_allocated = 8;
2920         int stripes_required = 1;
2921         int stripe_index;
2922         int i;
2923         int num_stripes;
2924         int max_errors = 0;
2925         struct btrfs_multi_bio *multi = NULL;
2926
2927         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2928                 stripes_allocated = 1;
2929 again:
2930         if (multi_ret) {
2931                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2932                                 GFP_NOFS);
2933                 if (!multi)
2934                         return -ENOMEM;
2935
2936                 atomic_set(&multi->error, 0);
2937         }
2938
2939         read_lock(&em_tree->lock);
2940         em = lookup_extent_mapping(em_tree, logical, *length);
2941         read_unlock(&em_tree->lock);
2942
2943         if (!em) {
2944                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2945                        (unsigned long long)logical,
2946                        (unsigned long long)*length);
2947                 BUG();
2948         }
2949
2950         BUG_ON(em->start > logical || em->start + em->len < logical);
2951         map = (struct map_lookup *)em->bdev;
2952         offset = logical - em->start;
2953
2954         if (mirror_num > map->num_stripes)
2955                 mirror_num = 0;
2956
2957         /* if our multi bio struct is too small, back off and try again */
2958         if (rw & REQ_WRITE) {
2959                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2960                                  BTRFS_BLOCK_GROUP_DUP)) {
2961                         stripes_required = map->num_stripes;
2962                         max_errors = 1;
2963                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2964                         stripes_required = map->sub_stripes;
2965                         max_errors = 1;
2966                 }
2967         }
2968         if (rw & REQ_DISCARD) {
2969                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2970                                  BTRFS_BLOCK_GROUP_RAID1 |
2971                                  BTRFS_BLOCK_GROUP_DUP |
2972                                  BTRFS_BLOCK_GROUP_RAID10)) {
2973                         stripes_required = map->num_stripes;
2974                 }
2975         }
2976         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2977             stripes_allocated < stripes_required) {
2978                 stripes_allocated = map->num_stripes;
2979                 free_extent_map(em);
2980                 kfree(multi);
2981                 goto again;
2982         }
2983         stripe_nr = offset;
2984         /*
2985          * stripe_nr counts the total number of stripes we have to stride
2986          * to get to this block
2987          */
2988         do_div(stripe_nr, map->stripe_len);
2989
2990         stripe_offset = stripe_nr * map->stripe_len;
2991         BUG_ON(offset < stripe_offset);
2992
2993         /* stripe_offset is the offset of this block in its stripe*/
2994         stripe_offset = offset - stripe_offset;
2995
2996         if (rw & REQ_DISCARD)
2997                 *length = min_t(u64, em->len - offset, *length);
2998         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2999                               BTRFS_BLOCK_GROUP_RAID1 |
3000                               BTRFS_BLOCK_GROUP_RAID10 |
3001                               BTRFS_BLOCK_GROUP_DUP)) {
3002                 /* we limit the length of each bio to what fits in a stripe */
3003                 *length = min_t(u64, em->len - offset,
3004                                 map->stripe_len - stripe_offset);
3005         } else {
3006                 *length = em->len - offset;
3007         }
3008
3009         if (!multi_ret)
3010                 goto out;
3011
3012         num_stripes = 1;
3013         stripe_index = 0;
3014         stripe_nr_orig = stripe_nr;
3015         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3016                         (~(map->stripe_len - 1));
3017         do_div(stripe_nr_end, map->stripe_len);
3018         stripe_end_offset = stripe_nr_end * map->stripe_len -
3019                             (offset + *length);
3020         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3021                 if (rw & REQ_DISCARD)
3022                         num_stripes = min_t(u64, map->num_stripes,
3023                                             stripe_nr_end - stripe_nr_orig);
3024                 stripe_index = do_div(stripe_nr, map->num_stripes);
3025         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3026                 if (rw & (REQ_WRITE | REQ_DISCARD))
3027                         num_stripes = map->num_stripes;
3028                 else if (mirror_num)
3029                         stripe_index = mirror_num - 1;
3030                 else {
3031                         stripe_index = find_live_mirror(map, 0,
3032                                             map->num_stripes,
3033                                             current->pid % map->num_stripes);
3034                 }
3035
3036         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3037                 if (rw & (REQ_WRITE | REQ_DISCARD))
3038                         num_stripes = map->num_stripes;
3039                 else if (mirror_num)
3040                         stripe_index = mirror_num - 1;
3041
3042         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3043                 int factor = map->num_stripes / map->sub_stripes;
3044
3045                 stripe_index = do_div(stripe_nr, factor);
3046                 stripe_index *= map->sub_stripes;
3047
3048                 if (rw & REQ_WRITE)
3049                         num_stripes = map->sub_stripes;
3050                 else if (rw & REQ_DISCARD)
3051                         num_stripes = min_t(u64, map->sub_stripes *
3052                                             (stripe_nr_end - stripe_nr_orig),
3053                                             map->num_stripes);
3054                 else if (mirror_num)
3055                         stripe_index += mirror_num - 1;
3056                 else {
3057                         stripe_index = find_live_mirror(map, stripe_index,
3058                                               map->sub_stripes, stripe_index +
3059                                               current->pid % map->sub_stripes);
3060                 }
3061         } else {
3062                 /*
3063                  * after this do_div call, stripe_nr is the number of stripes
3064                  * on this device we have to walk to find the data, and
3065                  * stripe_index is the number of our device in the stripe array
3066                  */
3067                 stripe_index = do_div(stripe_nr, map->num_stripes);
3068         }
3069         BUG_ON(stripe_index >= map->num_stripes);
3070
3071         if (rw & REQ_DISCARD) {
3072                 for (i = 0; i < num_stripes; i++) {
3073                         multi->stripes[i].physical =
3074                                 map->stripes[stripe_index].physical +
3075                                 stripe_offset + stripe_nr * map->stripe_len;
3076                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3077
3078                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3079                                 u64 stripes;
3080                                 u32 last_stripe = 0;
3081                                 int j;
3082
3083                                 div_u64_rem(stripe_nr_end - 1,
3084                                             map->num_stripes,
3085                                             &last_stripe);
3086
3087                                 for (j = 0; j < map->num_stripes; j++) {
3088                                         u32 test;
3089
3090                                         div_u64_rem(stripe_nr_end - 1 - j,
3091                                                     map->num_stripes, &test);
3092                                         if (test == stripe_index)
3093                                                 break;
3094                                 }
3095                                 stripes = stripe_nr_end - 1 - j;
3096                                 do_div(stripes, map->num_stripes);
3097                                 multi->stripes[i].length = map->stripe_len *
3098                                         (stripes - stripe_nr + 1);
3099
3100                                 if (i == 0) {
3101                                         multi->stripes[i].length -=
3102                                                 stripe_offset;
3103                                         stripe_offset = 0;
3104                                 }
3105                                 if (stripe_index == last_stripe)
3106                                         multi->stripes[i].length -=
3107                                                 stripe_end_offset;
3108                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3109                                 u64 stripes;
3110                                 int j;
3111                                 int factor = map->num_stripes /
3112                                              map->sub_stripes;
3113                                 u32 last_stripe = 0;
3114
3115                                 div_u64_rem(stripe_nr_end - 1,
3116                                             factor, &last_stripe);
3117                                 last_stripe *= map->sub_stripes;
3118
3119                                 for (j = 0; j < factor; j++) {
3120                                         u32 test;
3121
3122                                         div_u64_rem(stripe_nr_end - 1 - j,
3123                                                     factor, &test);
3124
3125                                         if (test ==
3126                                             stripe_index / map->sub_stripes)
3127                                                 break;
3128                                 }
3129                                 stripes = stripe_nr_end - 1 - j;
3130                                 do_div(stripes, factor);
3131                                 multi->stripes[i].length = map->stripe_len *
3132                                         (stripes - stripe_nr + 1);
3133
3134                                 if (i < map->sub_stripes) {
3135                                         multi->stripes[i].length -=
3136                                                 stripe_offset;
3137                                         if (i == map->sub_stripes - 1)
3138                                                 stripe_offset = 0;
3139                                 }
3140                                 if (stripe_index >= last_stripe &&
3141                                     stripe_index <= (last_stripe +
3142                                                      map->sub_stripes - 1)) {
3143                                         multi->stripes[i].length -=
3144                                                 stripe_end_offset;
3145                                 }
3146                         } else
3147                                 multi->stripes[i].length = *length;
3148
3149                         stripe_index++;
3150                         if (stripe_index == map->num_stripes) {
3151                                 /* This could only happen for RAID0/10 */
3152                                 stripe_index = 0;
3153                                 stripe_nr++;
3154                         }
3155                 }
3156         } else {
3157                 for (i = 0; i < num_stripes; i++) {
3158                         multi->stripes[i].physical =
3159                                 map->stripes[stripe_index].physical +
3160                                 stripe_offset +
3161                                 stripe_nr * map->stripe_len;
3162                         multi->stripes[i].dev =
3163                                 map->stripes[stripe_index].dev;
3164                         stripe_index++;
3165                 }
3166         }
3167         if (multi_ret) {
3168                 *multi_ret = multi;
3169                 multi->num_stripes = num_stripes;
3170                 multi->max_errors = max_errors;
3171         }
3172 out:
3173         free_extent_map(em);
3174         return 0;
3175 }
3176
3177 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3178                       u64 logical, u64 *length,
3179                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3180 {
3181         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3182                                  mirror_num);
3183 }
3184
3185 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3186                      u64 chunk_start, u64 physical, u64 devid,
3187                      u64 **logical, int *naddrs, int *stripe_len)
3188 {
3189         struct extent_map_tree *em_tree = &map_tree->map_tree;
3190         struct extent_map *em;
3191         struct map_lookup *map;
3192         u64 *buf;
3193         u64 bytenr;
3194         u64 length;
3195         u64 stripe_nr;
3196         int i, j, nr = 0;
3197
3198         read_lock(&em_tree->lock);
3199         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3200         read_unlock(&em_tree->lock);
3201
3202         BUG_ON(!em || em->start != chunk_start);
3203         map = (struct map_lookup *)em->bdev;
3204
3205         length = em->len;
3206         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3207                 do_div(length, map->num_stripes / map->sub_stripes);
3208         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3209                 do_div(length, map->num_stripes);
3210
3211         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3212         BUG_ON(!buf);
3213
3214         for (i = 0; i < map->num_stripes; i++) {
3215                 if (devid && map->stripes[i].dev->devid != devid)
3216                         continue;
3217                 if (map->stripes[i].physical > physical ||
3218                     map->stripes[i].physical + length <= physical)
3219                         continue;
3220
3221                 stripe_nr = physical - map->stripes[i].physical;
3222                 do_div(stripe_nr, map->stripe_len);
3223
3224                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3225                         stripe_nr = stripe_nr * map->num_stripes + i;
3226                         do_div(stripe_nr, map->sub_stripes);
3227                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3228                         stripe_nr = stripe_nr * map->num_stripes + i;
3229                 }
3230                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3231                 WARN_ON(nr >= map->num_stripes);
3232                 for (j = 0; j < nr; j++) {
3233                         if (buf[j] == bytenr)
3234                                 break;
3235                 }
3236                 if (j == nr) {
3237                         WARN_ON(nr >= map->num_stripes);
3238                         buf[nr++] = bytenr;
3239                 }
3240         }
3241
3242         *logical = buf;
3243         *naddrs = nr;
3244         *stripe_len = map->stripe_len;
3245
3246         free_extent_map(em);
3247         return 0;
3248 }
3249
3250 static void end_bio_multi_stripe(struct bio *bio, int err)
3251 {
3252         struct btrfs_multi_bio *multi = bio->bi_private;
3253         int is_orig_bio = 0;
3254
3255         if (err)
3256                 atomic_inc(&multi->error);
3257
3258         if (bio == multi->orig_bio)
3259                 is_orig_bio = 1;
3260
3261         if (atomic_dec_and_test(&multi->stripes_pending)) {
3262                 if (!is_orig_bio) {
3263                         bio_put(bio);
3264                         bio = multi->orig_bio;
3265                 }
3266                 bio->bi_private = multi->private;
3267                 bio->bi_end_io = multi->end_io;
3268                 /* only send an error to the higher layers if it is
3269                  * beyond the tolerance of the multi-bio
3270                  */
3271                 if (atomic_read(&multi->error) > multi->max_errors) {
3272                         err = -EIO;
3273                 } else if (err) {
3274                         /*
3275                          * this bio is actually up to date, we didn't
3276                          * go over the max number of errors
3277                          */
3278                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3279                         err = 0;
3280                 }
3281                 kfree(multi);
3282
3283                 bio_endio(bio, err);
3284         } else if (!is_orig_bio) {
3285                 bio_put(bio);
3286         }
3287 }
3288
3289 struct async_sched {
3290         struct bio *bio;
3291         int rw;
3292         struct btrfs_fs_info *info;
3293         struct btrfs_work work;
3294 };
3295
3296 /*
3297  * see run_scheduled_bios for a description of why bios are collected for
3298  * async submit.
3299  *
3300  * This will add one bio to the pending list for a device and make sure
3301  * the work struct is scheduled.
3302  */
3303 static noinline int schedule_bio(struct btrfs_root *root,
3304                                  struct btrfs_device *device,
3305                                  int rw, struct bio *bio)
3306 {
3307         int should_queue = 1;
3308         struct btrfs_pending_bios *pending_bios;
3309
3310         /* don't bother with additional async steps for reads, right now */
3311         if (!(rw & REQ_WRITE)) {
3312                 bio_get(bio);
3313                 submit_bio(rw, bio);
3314                 bio_put(bio);
3315                 return 0;
3316         }
3317
3318         /*
3319          * nr_async_bios allows us to reliably return congestion to the
3320          * higher layers.  Otherwise, the async bio makes it appear we have
3321          * made progress against dirty pages when we've really just put it
3322          * on a queue for later
3323          */
3324         atomic_inc(&root->fs_info->nr_async_bios);
3325         WARN_ON(bio->bi_next);
3326         bio->bi_next = NULL;
3327         bio->bi_rw |= rw;
3328
3329         spin_lock(&device->io_lock);
3330         if (bio->bi_rw & REQ_SYNC)
3331                 pending_bios = &device->pending_sync_bios;
3332         else
3333                 pending_bios = &device->pending_bios;
3334
3335         if (pending_bios->tail)
3336                 pending_bios->tail->bi_next = bio;
3337
3338         pending_bios->tail = bio;
3339         if (!pending_bios->head)
3340                 pending_bios->head = bio;
3341         if (device->running_pending)
3342                 should_queue = 0;
3343
3344         spin_unlock(&device->io_lock);
3345
3346         if (should_queue)
3347                 btrfs_queue_worker(&root->fs_info->submit_workers,
3348                                    &device->work);
3349         return 0;
3350 }
3351
3352 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3353                   int mirror_num, int async_submit)
3354 {
3355         struct btrfs_mapping_tree *map_tree;
3356         struct btrfs_device *dev;
3357         struct bio *first_bio = bio;
3358         u64 logical = (u64)bio->bi_sector << 9;
3359         u64 length = 0;
3360         u64 map_length;
3361         struct btrfs_multi_bio *multi = NULL;
3362         int ret;
3363         int dev_nr = 0;
3364         int total_devs = 1;
3365
3366         length = bio->bi_size;
3367         map_tree = &root->fs_info->mapping_tree;
3368         map_length = length;
3369
3370         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3371                               mirror_num);
3372         BUG_ON(ret);
3373
3374         total_devs = multi->num_stripes;
3375         if (map_length < length) {
3376                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3377                        "len %llu\n", (unsigned long long)logical,
3378                        (unsigned long long)length,
3379                        (unsigned long long)map_length);
3380                 BUG();
3381         }
3382         multi->end_io = first_bio->bi_end_io;
3383         multi->private = first_bio->bi_private;
3384         multi->orig_bio = first_bio;
3385         atomic_set(&multi->stripes_pending, multi->num_stripes);
3386
3387         while (dev_nr < total_devs) {
3388                 if (total_devs > 1) {
3389                         if (dev_nr < total_devs - 1) {
3390                                 bio = bio_clone(first_bio, GFP_NOFS);
3391                                 BUG_ON(!bio);
3392                         } else {
3393                                 bio = first_bio;
3394                         }
3395                         bio->bi_private = multi;
3396                         bio->bi_end_io = end_bio_multi_stripe;
3397                 }
3398                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3399                 dev = multi->stripes[dev_nr].dev;
3400                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3401                         bio->bi_bdev = dev->bdev;
3402                         if (async_submit)
3403                                 schedule_bio(root, dev, rw, bio);
3404                         else
3405                                 submit_bio(rw, bio);
3406                 } else {
3407                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3408                         bio->bi_sector = logical >> 9;
3409                         bio_endio(bio, -EIO);
3410                 }
3411                 dev_nr++;
3412         }
3413         if (total_devs == 1)
3414                 kfree(multi);
3415         return 0;
3416 }
3417
3418 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3419                                        u8 *uuid, u8 *fsid)
3420 {
3421         struct btrfs_device *device;
3422         struct btrfs_fs_devices *cur_devices;
3423
3424         cur_devices = root->fs_info->fs_devices;
3425         while (cur_devices) {
3426                 if (!fsid ||
3427                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3428                         device = __find_device(&cur_devices->devices,
3429                                                devid, uuid);
3430                         if (device)
3431                                 return device;
3432                 }
3433                 cur_devices = cur_devices->seed;
3434         }
3435         return NULL;
3436 }
3437
3438 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3439                                             u64 devid, u8 *dev_uuid)
3440 {
3441         struct btrfs_device *device;
3442         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3443
3444         device = kzalloc(sizeof(*device), GFP_NOFS);
3445         if (!device)
3446                 return NULL;
3447         list_add(&device->dev_list,
3448                  &fs_devices->devices);
3449         device->dev_root = root->fs_info->dev_root;
3450         device->devid = devid;
3451         device->work.func = pending_bios_fn;
3452         device->fs_devices = fs_devices;
3453         device->missing = 1;
3454         fs_devices->num_devices++;
3455         fs_devices->missing_devices++;
3456         spin_lock_init(&device->io_lock);
3457         INIT_LIST_HEAD(&device->dev_alloc_list);
3458         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3459         return device;
3460 }
3461
3462 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3463                           struct extent_buffer *leaf,
3464                           struct btrfs_chunk *chunk)
3465 {
3466         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3467         struct map_lookup *map;
3468         struct extent_map *em;
3469         u64 logical;
3470         u64 length;
3471         u64 devid;
3472         u8 uuid[BTRFS_UUID_SIZE];
3473         int num_stripes;
3474         int ret;
3475         int i;
3476
3477         logical = key->offset;
3478         length = btrfs_chunk_length(leaf, chunk);
3479
3480         read_lock(&map_tree->map_tree.lock);
3481         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3482         read_unlock(&map_tree->map_tree.lock);
3483
3484         /* already mapped? */
3485         if (em && em->start <= logical && em->start + em->len > logical) {
3486                 free_extent_map(em);
3487                 return 0;
3488         } else if (em) {
3489                 free_extent_map(em);
3490         }
3491
3492         em = alloc_extent_map(GFP_NOFS);
3493         if (!em)
3494                 return -ENOMEM;
3495         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3496         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3497         if (!map) {
3498                 free_extent_map(em);
3499                 return -ENOMEM;
3500         }
3501
3502         em->bdev = (struct block_device *)map;
3503         em->start = logical;
3504         em->len = length;
3505         em->block_start = 0;
3506         em->block_len = em->len;
3507
3508         map->num_stripes = num_stripes;
3509         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3510         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3511         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3512         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3513         map->type = btrfs_chunk_type(leaf, chunk);
3514         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3515         for (i = 0; i < num_stripes; i++) {
3516                 map->stripes[i].physical =
3517                         btrfs_stripe_offset_nr(leaf, chunk, i);
3518                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3519                 read_extent_buffer(leaf, uuid, (unsigned long)
3520                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3521                                    BTRFS_UUID_SIZE);
3522                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3523                                                         NULL);
3524                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3525                         kfree(map);
3526                         free_extent_map(em);
3527                         return -EIO;
3528                 }
3529                 if (!map->stripes[i].dev) {
3530                         map->stripes[i].dev =
3531                                 add_missing_dev(root, devid, uuid);
3532                         if (!map->stripes[i].dev) {
3533                                 kfree(map);
3534                                 free_extent_map(em);
3535                                 return -EIO;
3536                         }
3537                 }
3538                 map->stripes[i].dev->in_fs_metadata = 1;
3539         }
3540
3541         write_lock(&map_tree->map_tree.lock);
3542         ret = add_extent_mapping(&map_tree->map_tree, em);
3543         write_unlock(&map_tree->map_tree.lock);
3544         BUG_ON(ret);
3545         free_extent_map(em);
3546
3547         return 0;
3548 }
3549
3550 static int fill_device_from_item(struct extent_buffer *leaf,
3551                                  struct btrfs_dev_item *dev_item,
3552                                  struct btrfs_device *device)
3553 {
3554         unsigned long ptr;
3555
3556         device->devid = btrfs_device_id(leaf, dev_item);
3557         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3558         device->total_bytes = device->disk_total_bytes;
3559         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3560         device->type = btrfs_device_type(leaf, dev_item);
3561         device->io_align = btrfs_device_io_align(leaf, dev_item);
3562         device->io_width = btrfs_device_io_width(leaf, dev_item);
3563         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3564
3565         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3566         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3567
3568         return 0;
3569 }
3570
3571 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3572 {
3573         struct btrfs_fs_devices *fs_devices;
3574         int ret;
3575
3576         mutex_lock(&uuid_mutex);
3577
3578         fs_devices = root->fs_info->fs_devices->seed;
3579         while (fs_devices) {
3580                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3581                         ret = 0;
3582                         goto out;
3583                 }
3584                 fs_devices = fs_devices->seed;
3585         }
3586
3587         fs_devices = find_fsid(fsid);
3588         if (!fs_devices) {
3589                 ret = -ENOENT;
3590                 goto out;
3591         }
3592
3593         fs_devices = clone_fs_devices(fs_devices);
3594         if (IS_ERR(fs_devices)) {
3595                 ret = PTR_ERR(fs_devices);
3596                 goto out;
3597         }
3598
3599         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3600                                    root->fs_info->bdev_holder);
3601         if (ret)
3602                 goto out;
3603
3604         if (!fs_devices->seeding) {
3605                 __btrfs_close_devices(fs_devices);
3606                 free_fs_devices(fs_devices);
3607                 ret = -EINVAL;
3608                 goto out;
3609         }
3610
3611         fs_devices->seed = root->fs_info->fs_devices->seed;
3612         root->fs_info->fs_devices->seed = fs_devices;
3613 out:
3614         mutex_unlock(&uuid_mutex);
3615         return ret;
3616 }
3617
3618 static int read_one_dev(struct btrfs_root *root,
3619                         struct extent_buffer *leaf,
3620                         struct btrfs_dev_item *dev_item)
3621 {
3622         struct btrfs_device *device;
3623         u64 devid;
3624         int ret;
3625         u8 fs_uuid[BTRFS_UUID_SIZE];
3626         u8 dev_uuid[BTRFS_UUID_SIZE];
3627
3628         devid = btrfs_device_id(leaf, dev_item);
3629         read_extent_buffer(leaf, dev_uuid,
3630                            (unsigned long)btrfs_device_uuid(dev_item),
3631                            BTRFS_UUID_SIZE);
3632         read_extent_buffer(leaf, fs_uuid,
3633                            (unsigned long)btrfs_device_fsid(dev_item),
3634                            BTRFS_UUID_SIZE);
3635
3636         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3637                 ret = open_seed_devices(root, fs_uuid);
3638                 if (ret && !btrfs_test_opt(root, DEGRADED))
3639                         return ret;
3640         }
3641
3642         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3643         if (!device || !device->bdev) {
3644                 if (!btrfs_test_opt(root, DEGRADED))
3645                         return -EIO;
3646
3647                 if (!device) {
3648                         printk(KERN_WARNING "warning devid %llu missing\n",
3649                                (unsigned long long)devid);
3650                         device = add_missing_dev(root, devid, dev_uuid);
3651                         if (!device)
3652                                 return -ENOMEM;
3653                 } else if (!device->missing) {
3654                         /*
3655                          * this happens when a device that was properly setup
3656                          * in the device info lists suddenly goes bad.
3657                          * device->bdev is NULL, and so we have to set
3658                          * device->missing to one here
3659                          */
3660                         root->fs_info->fs_devices->missing_devices++;
3661                         device->missing = 1;
3662                 }
3663         }
3664
3665         if (device->fs_devices != root->fs_info->fs_devices) {
3666                 BUG_ON(device->writeable);
3667                 if (device->generation !=
3668                     btrfs_device_generation(leaf, dev_item))
3669                         return -EINVAL;
3670         }
3671
3672         fill_device_from_item(leaf, dev_item, device);
3673         device->dev_root = root->fs_info->dev_root;
3674         device->in_fs_metadata = 1;
3675         if (device->writeable)
3676                 device->fs_devices->total_rw_bytes += device->total_bytes;
3677         ret = 0;
3678         return ret;
3679 }
3680
3681 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3682 {
3683         struct btrfs_dev_item *dev_item;
3684
3685         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3686                                                      dev_item);
3687         return read_one_dev(root, buf, dev_item);
3688 }
3689
3690 int btrfs_read_sys_array(struct btrfs_root *root)
3691 {
3692         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3693         struct extent_buffer *sb;
3694         struct btrfs_disk_key *disk_key;
3695         struct btrfs_chunk *chunk;
3696         u8 *ptr;
3697         unsigned long sb_ptr;
3698         int ret = 0;
3699         u32 num_stripes;
3700         u32 array_size;
3701         u32 len = 0;
3702         u32 cur;
3703         struct btrfs_key key;
3704
3705         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3706                                           BTRFS_SUPER_INFO_SIZE);
3707         if (!sb)
3708                 return -ENOMEM;
3709         btrfs_set_buffer_uptodate(sb);
3710         btrfs_set_buffer_lockdep_class(sb, 0);
3711
3712         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3713         array_size = btrfs_super_sys_array_size(super_copy);
3714
3715         ptr = super_copy->sys_chunk_array;
3716         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3717         cur = 0;
3718
3719         while (cur < array_size) {
3720                 disk_key = (struct btrfs_disk_key *)ptr;
3721                 btrfs_disk_key_to_cpu(&key, disk_key);
3722
3723                 len = sizeof(*disk_key); ptr += len;
3724                 sb_ptr += len;
3725                 cur += len;
3726
3727                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3728                         chunk = (struct btrfs_chunk *)sb_ptr;
3729                         ret = read_one_chunk(root, &key, sb, chunk);
3730                         if (ret)
3731                                 break;
3732                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3733                         len = btrfs_chunk_item_size(num_stripes);
3734                 } else {
3735                         ret = -EIO;
3736                         break;
3737                 }
3738                 ptr += len;
3739                 sb_ptr += len;
3740                 cur += len;
3741         }
3742         free_extent_buffer(sb);
3743         return ret;
3744 }
3745
3746 int btrfs_read_chunk_tree(struct btrfs_root *root)
3747 {
3748         struct btrfs_path *path;
3749         struct extent_buffer *leaf;
3750         struct btrfs_key key;
3751         struct btrfs_key found_key;
3752         int ret;
3753         int slot;
3754
3755         root = root->fs_info->chunk_root;
3756
3757         path = btrfs_alloc_path();
3758         if (!path)
3759                 return -ENOMEM;
3760
3761         /* first we search for all of the device items, and then we
3762          * read in all of the chunk items.  This way we can create chunk
3763          * mappings that reference all of the devices that are afound
3764          */
3765         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3766         key.offset = 0;
3767         key.type = 0;
3768 again:
3769         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3770         if (ret < 0)
3771                 goto error;
3772         while (1) {
3773                 leaf = path->nodes[0];
3774                 slot = path->slots[0];
3775                 if (slot >= btrfs_header_nritems(leaf)) {
3776                         ret = btrfs_next_leaf(root, path);
3777                         if (ret == 0)
3778                                 continue;
3779                         if (ret < 0)
3780                                 goto error;
3781                         break;
3782                 }
3783                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3784                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3785                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3786                                 break;
3787                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3788                                 struct btrfs_dev_item *dev_item;
3789                                 dev_item = btrfs_item_ptr(leaf, slot,
3790                                                   struct btrfs_dev_item);
3791                                 ret = read_one_dev(root, leaf, dev_item);
3792                                 if (ret)
3793                                         goto error;
3794                         }
3795                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3796                         struct btrfs_chunk *chunk;
3797                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3798                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3799                         if (ret)
3800                                 goto error;
3801                 }
3802                 path->slots[0]++;
3803         }
3804         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3805                 key.objectid = 0;
3806                 btrfs_release_path(root, path);
3807                 goto again;
3808         }
3809         ret = 0;
3810 error:
3811         btrfs_free_path(path);
3812         return ret;
3813 }