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 <linux/kthread.h>
  27 #include <asm/div64.h>
  28 #include "compat.h"
  29 #include "ctree.h"
  30 #include "extent_map.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "print-tree.h"
  34 #include "volumes.h"
  35 #include "async-thread.h"
  36 #include "check-integrity.h"
  37
  38 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  39                                 struct btrfs_root *root,
  40                                 struct btrfs_device *device);
  41 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  42
  43 static DEFINE_MUTEX(uuid_mutex);
  44 static LIST_HEAD(fs_uuids);
  45
  46 static void lock_chunks(struct btrfs_root *root)
  47 {
  48         mutex_lock(&root->fs_info->chunk_mutex);
  49 }
  50
  51 static void unlock_chunks(struct btrfs_root *root)
  52 {
  53         mutex_unlock(&root->fs_info->chunk_mutex);
  54 }
  55
  56 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  57 {
  58         struct btrfs_device *device;
  59         WARN_ON(fs_devices->opened);
  60         while (!list_empty(&fs_devices->devices)) {
  61                 device = list_entry(fs_devices->devices.next,
  62                                     struct btrfs_device, dev_list);
  63                 list_del(&device->dev_list);
  64                 kfree(device->name);
  65                 kfree(device);
  66         }
  67         kfree(fs_devices);
  68 }
  69
  70 int btrfs_cleanup_fs_uuids(void)
  71 {
  72         struct btrfs_fs_devices *fs_devices;
  73
  74         while (!list_empty(&fs_uuids)) {
  75                 fs_devices = list_entry(fs_uuids.next,
  76                                         struct btrfs_fs_devices, list);
  77                 list_del(&fs_devices->list);
  78                 free_fs_devices(fs_devices);
  79         }
  80         return 0;
  81 }
  82
  83 static noinline struct btrfs_device *__find_device(struct list_head *head,
  84                                                    u64 devid, u8 *uuid)
  85 {
  86         struct btrfs_device *dev;
  87
  88         list_for_each_entry(dev, head, dev_list) {
  89                 if (dev->devid == devid &&
  90                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  91                         return dev;
  92                 }
  93         }
  94         return NULL;
  95 }
  96
  97 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
  98 {
  99         struct btrfs_fs_devices *fs_devices;
 100
 101         list_for_each_entry(fs_devices, &fs_uuids, list) {
 102                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 103                         return fs_devices;
 104         }
 105         return NULL;
 106 }
 107
 108 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 109                         struct bio *head, struct bio *tail)
 110 {
 111
 112         struct bio *old_head;
 113
 114         old_head = pending_bios->head;
 115         pending_bios->head = head;
 116         if (pending_bios->tail)
 117                 tail->bi_next = old_head;
 118         else
 119                 pending_bios->tail = tail;
 120 }
 121
 122 /*
 123  * we try to collect pending bios for a device so we don't get a large
 124  * number of procs sending bios down to the same device.  This greatly
 125  * improves the schedulers ability to collect and merge the bios.
 126  *
 127  * But, it also turns into a long list of bios to process and that is sure
 128  * to eventually make the worker thread block.  The solution here is to
 129  * make some progress and then put this work struct back at the end of
 130  * the list if the block device is congested.  This way, multiple devices
 131  * can make progress from a single worker thread.
 132  */
 133 static noinline int run_scheduled_bios(struct btrfs_device *device)
 134 {
 135         struct bio *pending;
 136         struct backing_dev_info *bdi;
 137         struct btrfs_fs_info *fs_info;
 138         struct btrfs_pending_bios *pending_bios;
 139         struct bio *tail;
 140         struct bio *cur;
 141         int again = 0;
 142         unsigned long num_run;
 143         unsigned long batch_run = 0;
 144         unsigned long limit;
 145         unsigned long last_waited = 0;
 146         int force_reg = 0;
 147         int sync_pending = 0;
 148         struct blk_plug plug;
 149
 150         /*
 151          * this function runs all the bios we've collected for
 152          * a particular device.  We don't want to wander off to
 153          * another device without first sending all of these down.
 154          * So, setup a plug here and finish it off before we return
 155          */
 156         blk_start_plug(&plug);
 157
 158         bdi = blk_get_backing_dev_info(device->bdev);
 159         fs_info = device->dev_root->fs_info;
 160         limit = btrfs_async_submit_limit(fs_info);
 161         limit = limit * 2 / 3;
 162
 163 loop:
 164         spin_lock(&device->io_lock);
 165
 166 loop_lock:
 167         num_run = 0;
 168
 169         /* take all the bios off the list at once and process them
 170          * later on (without the lock held).  But, remember the
 171          * tail and other pointers so the bios can be properly reinserted
 172          * into the list if we hit congestion
 173          */
 174         if (!force_reg && device->pending_sync_bios.head) {
 175                 pending_bios = &device->pending_sync_bios;
 176                 force_reg = 1;
 177         } else {
 178                 pending_bios = &device->pending_bios;
 179                 force_reg = 0;
 180         }
 181
 182         pending = pending_bios->head;
 183         tail = pending_bios->tail;
 184         WARN_ON(pending && !tail);
 185
 186         /*
 187          * if pending was null this time around, no bios need processing
 188          * at all and we can stop.  Otherwise it'll loop back up again
 189          * and do an additional check so no bios are missed.
 190          *
 191          * device->running_pending is used to synchronize with the
 192          * schedule_bio code.
 193          */
 194         if (device->pending_sync_bios.head == NULL &&
 195             device->pending_bios.head == NULL) {
 196                 again = 0;
 197                 device->running_pending = 0;
 198         } else {
 199                 again = 1;
 200                 device->running_pending = 1;
 201         }
 202
 203         pending_bios->head = NULL;
 204         pending_bios->tail = NULL;
 205
 206         spin_unlock(&device->io_lock);
 207
 208         while (pending) {
 209
 210                 rmb();
 211                 /* we want to work on both lists, but do more bios on the
 212                  * sync list than the regular list
 213                  */
 214                 if ((num_run > 32 &&
 215                     pending_bios != &device->pending_sync_bios &&
 216                     device->pending_sync_bios.head) ||
 217                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 218                     device->pending_bios.head)) {
 219                         spin_lock(&device->io_lock);
 220                         requeue_list(pending_bios, pending, tail);
 221                         goto loop_lock;
 222                 }
 223
 224                 cur = pending;
 225                 pending = pending->bi_next;
 226                 cur->bi_next = NULL;
 227                 atomic_dec(&fs_info->nr_async_bios);
 228
 229                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 230                     waitqueue_active(&fs_info->async_submit_wait))
 231                         wake_up(&fs_info->async_submit_wait);
 232
 233                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 234
 235                 /*
 236                  * if we're doing the sync list, record that our
 237                  * plug has some sync requests on it
 238                  *
 239                  * If we're doing the regular list and there are
 240                  * sync requests sitting around, unplug before
 241                  * we add more
 242                  */
 243                 if (pending_bios == &device->pending_sync_bios) {
 244                         sync_pending = 1;
 245                 } else if (sync_pending) {
 246                         blk_finish_plug(&plug);
 247                         blk_start_plug(&plug);
 248                         sync_pending = 0;
 249                 }
 250
 251                 btrfsic_submit_bio(cur->bi_rw, cur);
 252                 num_run++;
 253                 batch_run++;
 254                 if (need_resched())
 255                         cond_resched();
 256
 257                 /*
 258                  * we made progress, there is more work to do and the bdi
 259                  * is now congested.  Back off and let other work structs
 260                  * run instead
 261                  */
 262                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 263                     fs_info->fs_devices->open_devices > 1) {
 264                         struct io_context *ioc;
 265
 266                         ioc = current->io_context;
 267
 268                         /*
 269                          * the main goal here is that we don't want to
 270                          * block if we're going to be able to submit
 271                          * more requests without blocking.
 272                          *
 273                          * This code does two great things, it pokes into
 274                          * the elevator code from a filesystem _and_
 275                          * it makes assumptions about how batching works.
 276                          */
 277                         if (ioc && ioc->nr_batch_requests > 0 &&
 278                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 279                             (last_waited == 0 ||
 280                              ioc->last_waited == last_waited)) {
 281                                 /*
 282                                  * we want to go through our batch of
 283                                  * requests and stop.  So, we copy out
 284                                  * the ioc->last_waited time and test
 285                                  * against it before looping
 286                                  */
 287                                 last_waited = ioc->last_waited;
 288                                 if (need_resched())
 289                                         cond_resched();
 290                                 continue;
 291                         }
 292                         spin_lock(&device->io_lock);
 293                         requeue_list(pending_bios, pending, tail);
 294                         device->running_pending = 1;
 295
 296                         spin_unlock(&device->io_lock);
 297                         btrfs_requeue_work(&device->work);
 298                         goto done;
 299                 }
 300                 /* unplug every 64 requests just for good measure */
 301                 if (batch_run % 64 == 0) {
 302                         blk_finish_plug(&plug);
 303                         blk_start_plug(&plug);
 304                         sync_pending = 0;
 305                 }
 306         }
 307
 308         cond_resched();
 309         if (again)
 310                 goto loop;
 311
 312         spin_lock(&device->io_lock);
 313         if (device->pending_bios.head || device->pending_sync_bios.head)
 314                 goto loop_lock;
 315         spin_unlock(&device->io_lock);
 316
 317 done:
 318         blk_finish_plug(&plug);
 319         return 0;
 320 }
 321
 322 static void pending_bios_fn(struct btrfs_work *work)
 323 {
 324         struct btrfs_device *device;
 325
 326         device = container_of(work, struct btrfs_device, work);
 327         run_scheduled_bios(device);
 328 }
 329
 330 static noinline int device_list_add(const char *path,
 331                            struct btrfs_super_block *disk_super,
 332                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 333 {
 334         struct btrfs_device *device;
 335         struct btrfs_fs_devices *fs_devices;
 336         u64 found_transid = btrfs_super_generation(disk_super);
 337         char *name;
 338
 339         fs_devices = find_fsid(disk_super->fsid);
 340         if (!fs_devices) {
 341                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 342                 if (!fs_devices)
 343                         return -ENOMEM;
 344                 INIT_LIST_HEAD(&fs_devices->devices);
 345                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 346                 list_add(&fs_devices->list, &fs_uuids);
 347                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 348                 fs_devices->latest_devid = devid;
 349                 fs_devices->latest_trans = found_transid;
 350                 mutex_init(&fs_devices->device_list_mutex);
 351                 device = NULL;
 352         } else {
 353                 device = __find_device(&fs_devices->devices, devid,
 354                                        disk_super->dev_item.uuid);
 355         }
 356         if (!device) {
 357                 if (fs_devices->opened)
 358                         return -EBUSY;
 359
 360                 device = kzalloc(sizeof(*device), GFP_NOFS);
 361                 if (!device) {
 362                         /* we can safely leave the fs_devices entry around */
 363                         return -ENOMEM;
 364                 }
 365                 device->devid = devid;
 366                 device->work.func = pending_bios_fn;
 367                 memcpy(device->uuid, disk_super->dev_item.uuid,
 368                        BTRFS_UUID_SIZE);
 369                 spin_lock_init(&device->io_lock);
 370                 device->name = kstrdup(path, GFP_NOFS);
 371                 if (!device->name) {
 372                         kfree(device);
 373                         return -ENOMEM;
 374                 }
 375                 INIT_LIST_HEAD(&device->dev_alloc_list);
 376
 377                 /* init readahead state */
 378                 spin_lock_init(&device->reada_lock);
 379                 device->reada_curr_zone = NULL;
 380                 atomic_set(&device->reada_in_flight, 0);
 381                 device->reada_next = 0;
 382                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 383                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 384
 385                 mutex_lock(&fs_devices->device_list_mutex);
 386                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 387                 mutex_unlock(&fs_devices->device_list_mutex);
 388
 389                 device->fs_devices = fs_devices;
 390                 fs_devices->num_devices++;
 391         } else if (!device->name || strcmp(device->name, path)) {
 392                 name = kstrdup(path, GFP_NOFS);
 393                 if (!name)
 394                         return -ENOMEM;
 395                 kfree(device->name);
 396                 device->name = name;
 397                 if (device->missing) {
 398                         fs_devices->missing_devices--;
 399                         device->missing = 0;
 400                 }
 401         }
 402
 403         if (found_transid > fs_devices->latest_trans) {
 404                 fs_devices->latest_devid = devid;
 405                 fs_devices->latest_trans = found_transid;
 406         }
 407         *fs_devices_ret = fs_devices;
 408         return 0;
 409 }
 410
 411 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 412 {
 413         struct btrfs_fs_devices *fs_devices;
 414         struct btrfs_device *device;
 415         struct btrfs_device *orig_dev;
 416
 417         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 418         if (!fs_devices)
 419                 return ERR_PTR(-ENOMEM);
 420
 421         INIT_LIST_HEAD(&fs_devices->devices);
 422         INIT_LIST_HEAD(&fs_devices->alloc_list);
 423         INIT_LIST_HEAD(&fs_devices->list);
 424         mutex_init(&fs_devices->device_list_mutex);
 425         fs_devices->latest_devid = orig->latest_devid;
 426         fs_devices->latest_trans = orig->latest_trans;
 427         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 428
 429         /* We have held the volume lock, it is safe to get the devices. */
 430         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 431                 device = kzalloc(sizeof(*device), GFP_NOFS);
 432                 if (!device)
 433                         goto error;
 434
 435                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 436                 if (!device->name) {
 437                         kfree(device);
 438                         goto error;
 439                 }
 440
 441                 device->devid = orig_dev->devid;
 442                 device->work.func = pending_bios_fn;
 443                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 444                 spin_lock_init(&device->io_lock);
 445                 INIT_LIST_HEAD(&device->dev_list);
 446                 INIT_LIST_HEAD(&device->dev_alloc_list);
 447
 448                 list_add(&device->dev_list, &fs_devices->devices);
 449                 device->fs_devices = fs_devices;
 450                 fs_devices->num_devices++;
 451         }
 452         return fs_devices;
 453 error:
 454         free_fs_devices(fs_devices);
 455         return ERR_PTR(-ENOMEM);
 456 }
 457
 458 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 459 {
 460         struct btrfs_device *device, *next;
 461
 462         struct block_device *latest_bdev = NULL;
 463         u64 latest_devid = 0;
 464         u64 latest_transid = 0;
 465
 466         mutex_lock(&uuid_mutex);
 467 again:
 468         /* This is the initialized path, it is safe to release the devices. */
 469         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 470                 if (device->in_fs_metadata) {
 471                         if (!latest_transid ||
 472                             device->generation > latest_transid) {
 473                                 latest_devid = device->devid;
 474                                 latest_transid = device->generation;
 475                                 latest_bdev = device->bdev;
 476                         }
 477                         continue;
 478                 }
 479
 480                 if (device->bdev) {
 481                         blkdev_put(device->bdev, device->mode);
 482                         device->bdev = NULL;
 483                         fs_devices->open_devices--;
 484                 }
 485                 if (device->writeable) {
 486                         list_del_init(&device->dev_alloc_list);
 487                         device->writeable = 0;
 488                         fs_devices->rw_devices--;
 489                 }
 490                 list_del_init(&device->dev_list);
 491                 fs_devices->num_devices--;
 492                 kfree(device->name);
 493                 kfree(device);
 494         }
 495
 496         if (fs_devices->seed) {
 497                 fs_devices = fs_devices->seed;
 498                 goto again;
 499         }
 500
 501         fs_devices->latest_bdev = latest_bdev;
 502         fs_devices->latest_devid = latest_devid;
 503         fs_devices->latest_trans = latest_transid;
 504
 505         mutex_unlock(&uuid_mutex);
 506         return 0;
 507 }
 508
 509 static void __free_device(struct work_struct *work)
 510 {
 511         struct btrfs_device *device;
 512
 513         device = container_of(work, struct btrfs_device, rcu_work);
 514
 515         if (device->bdev)
 516                 blkdev_put(device->bdev, device->mode);
 517
 518         kfree(device->name);
 519         kfree(device);
 520 }
 521
 522 static void free_device(struct rcu_head *head)
 523 {
 524         struct btrfs_device *device;
 525
 526         device = container_of(head, struct btrfs_device, rcu);
 527
 528         INIT_WORK(&device->rcu_work, __free_device);
 529         schedule_work(&device->rcu_work);
 530 }
 531
 532 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 533 {
 534         struct btrfs_device *device;
 535
 536         if (--fs_devices->opened > 0)
 537                 return 0;
 538
 539         mutex_lock(&fs_devices->device_list_mutex);
 540         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 541                 struct btrfs_device *new_device;
 542
 543                 if (device->bdev)
 544                         fs_devices->open_devices--;
 545
 546                 if (device->writeable) {
 547                         list_del_init(&device->dev_alloc_list);
 548                         fs_devices->rw_devices--;
 549                 }
 550
 551                 if (device->can_discard)
 552                         fs_devices->num_can_discard--;
 553
 554                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 555                 BUG_ON(!new_device);
 556                 memcpy(new_device, device, sizeof(*new_device));
 557                 new_device->name = kstrdup(device->name, GFP_NOFS);
 558                 BUG_ON(device->name && !new_device->name);
 559                 new_device->bdev = NULL;
 560                 new_device->writeable = 0;
 561                 new_device->in_fs_metadata = 0;
 562                 new_device->can_discard = 0;
 563                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 564
 565                 call_rcu(&device->rcu, free_device);
 566         }
 567         mutex_unlock(&fs_devices->device_list_mutex);
 568
 569         WARN_ON(fs_devices->open_devices);
 570         WARN_ON(fs_devices->rw_devices);
 571         fs_devices->opened = 0;
 572         fs_devices->seeding = 0;
 573
 574         return 0;
 575 }
 576
 577 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 578 {
 579         struct btrfs_fs_devices *seed_devices = NULL;
 580         int ret;
 581
 582         mutex_lock(&uuid_mutex);
 583         ret = __btrfs_close_devices(fs_devices);
 584         if (!fs_devices->opened) {
 585                 seed_devices = fs_devices->seed;
 586                 fs_devices->seed = NULL;
 587         }
 588         mutex_unlock(&uuid_mutex);
 589
 590         while (seed_devices) {
 591                 fs_devices = seed_devices;
 592                 seed_devices = fs_devices->seed;
 593                 __btrfs_close_devices(fs_devices);
 594                 free_fs_devices(fs_devices);
 595         }
 596         return ret;
 597 }
 598
 599 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 600                                 fmode_t flags, void *holder)
 601 {
 602         struct request_queue *q;
 603         struct block_device *bdev;
 604         struct list_head *head = &fs_devices->devices;
 605         struct btrfs_device *device;
 606         struct block_device *latest_bdev = NULL;
 607         struct buffer_head *bh;
 608         struct btrfs_super_block *disk_super;
 609         u64 latest_devid = 0;
 610         u64 latest_transid = 0;
 611         u64 devid;
 612         int seeding = 1;
 613         int ret = 0;
 614
 615         flags |= FMODE_EXCL;
 616
 617         list_for_each_entry(device, head, dev_list) {
 618                 if (device->bdev)
 619                         continue;
 620                 if (!device->name)
 621                         continue;
 622
 623                 bdev = blkdev_get_by_path(device->name, flags, holder);
 624                 if (IS_ERR(bdev)) {
 625                         printk(KERN_INFO "open %s failed\n", device->name);
 626                         goto error;
 627                 }
 628                 set_blocksize(bdev, 4096);
 629
 630                 bh = btrfs_read_dev_super(bdev);
 631                 if (!bh)
 632                         goto error_close;
 633
 634                 disk_super = (struct btrfs_super_block *)bh->b_data;
 635                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 636                 if (devid != device->devid)
 637                         goto error_brelse;
 638
 639                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 640                            BTRFS_UUID_SIZE))
 641                         goto error_brelse;
 642
 643                 device->generation = btrfs_super_generation(disk_super);
 644                 if (!latest_transid || device->generation > latest_transid) {
 645                         latest_devid = devid;
 646                         latest_transid = device->generation;
 647                         latest_bdev = bdev;
 648                 }
 649
 650                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 651                         device->writeable = 0;
 652                 } else {
 653                         device->writeable = !bdev_read_only(bdev);
 654                         seeding = 0;
 655                 }
 656
 657                 q = bdev_get_queue(bdev);
 658                 if (blk_queue_discard(q)) {
 659                         device->can_discard = 1;
 660                         fs_devices->num_can_discard++;
 661                 }
 662
 663                 device->bdev = bdev;
 664                 device->in_fs_metadata = 0;
 665                 device->mode = flags;
 666
 667                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 668                         fs_devices->rotating = 1;
 669
 670                 fs_devices->open_devices++;
 671                 if (device->writeable) {
 672                         fs_devices->rw_devices++;
 673                         list_add(&device->dev_alloc_list,
 674                                  &fs_devices->alloc_list);
 675                 }
 676                 brelse(bh);
 677                 continue;
 678
 679 error_brelse:
 680                 brelse(bh);
 681 error_close:
 682                 blkdev_put(bdev, flags);
 683 error:
 684                 continue;
 685         }
 686         if (fs_devices->open_devices == 0) {
 687                 ret = -EINVAL;
 688                 goto out;
 689         }
 690         fs_devices->seeding = seeding;
 691         fs_devices->opened = 1;
 692         fs_devices->latest_bdev = latest_bdev;
 693         fs_devices->latest_devid = latest_devid;
 694         fs_devices->latest_trans = latest_transid;
 695         fs_devices->total_rw_bytes = 0;
 696 out:
 697         return ret;
 698 }
 699
 700 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 701                        fmode_t flags, void *holder)
 702 {
 703         int ret;
 704
 705         mutex_lock(&uuid_mutex);
 706         if (fs_devices->opened) {
 707                 fs_devices->opened++;
 708                 ret = 0;
 709         } else {
 710                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 711         }
 712         mutex_unlock(&uuid_mutex);
 713         return ret;
 714 }
 715
 716 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 717                           struct btrfs_fs_devices **fs_devices_ret)
 718 {
 719         struct btrfs_super_block *disk_super;
 720         struct block_device *bdev;
 721         struct buffer_head *bh;
 722         int ret;
 723         u64 devid;
 724         u64 transid;
 725
 726         flags |= FMODE_EXCL;
 727         bdev = blkdev_get_by_path(path, flags, holder);
 728
 729         if (IS_ERR(bdev)) {
 730                 ret = PTR_ERR(bdev);
 731                 goto error;
 732         }
 733
 734         mutex_lock(&uuid_mutex);
 735         ret = set_blocksize(bdev, 4096);
 736         if (ret)
 737                 goto error_close;
 738         bh = btrfs_read_dev_super(bdev);
 739         if (!bh) {
 740                 ret = -EINVAL;
 741                 goto error_close;
 742         }
 743         disk_super = (struct btrfs_super_block *)bh->b_data;
 744         devid = btrfs_stack_device_id(&disk_super->dev_item);
 745         transid = btrfs_super_generation(disk_super);
 746         if (disk_super->label[0])
 747                 printk(KERN_INFO "device label %s ", disk_super->label);
 748         else
 749                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 750         printk(KERN_CONT "devid %llu transid %llu %s\n",
 751                (unsigned long long)devid, (unsigned long long)transid, path);
 752         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 753
 754         brelse(bh);
 755 error_close:
 756         mutex_unlock(&uuid_mutex);
 757         blkdev_put(bdev, flags);
 758 error:
 759         return ret;
 760 }
 761
 762 /* helper to account the used device space in the range */
 763 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 764                                    u64 end, u64 *length)
 765 {
 766         struct btrfs_key key;
 767         struct btrfs_root *root = device->dev_root;
 768         struct btrfs_dev_extent *dev_extent;
 769         struct btrfs_path *path;
 770         u64 extent_end;
 771         int ret;
 772         int slot;
 773         struct extent_buffer *l;
 774
 775         *length = 0;
 776
 777         if (start >= device->total_bytes)
 778                 return 0;
 779
 780         path = btrfs_alloc_path();
 781         if (!path)
 782                 return -ENOMEM;
 783         path->reada = 2;
 784
 785         key.objectid = device->devid;
 786         key.offset = start;
 787         key.type = BTRFS_DEV_EXTENT_KEY;
 788
 789         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 790         if (ret < 0)
 791                 goto out;
 792         if (ret > 0) {
 793                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 794                 if (ret < 0)
 795                         goto out;
 796         }
 797
 798         while (1) {
 799                 l = path->nodes[0];
 800                 slot = path->slots[0];
 801                 if (slot >= btrfs_header_nritems(l)) {
 802                         ret = btrfs_next_leaf(root, path);
 803                         if (ret == 0)
 804                                 continue;
 805                         if (ret < 0)
 806                                 goto out;
 807
 808                         break;
 809                 }
 810                 btrfs_item_key_to_cpu(l, &key, slot);
 811
 812                 if (key.objectid < device->devid)
 813                         goto next;
 814
 815                 if (key.objectid > device->devid)
 816                         break;
 817
 818                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 819                         goto next;
 820
 821                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 822                 extent_end = key.offset + btrfs_dev_extent_length(l,
 823                                                                   dev_extent);
 824                 if (key.offset <= start && extent_end > end) {
 825                         *length = end - start + 1;
 826                         break;
 827                 } else if (key.offset <= start && extent_end > start)
 828                         *length += extent_end - start;
 829                 else if (key.offset > start && extent_end <= end)
 830                         *length += extent_end - key.offset;
 831                 else if (key.offset > start && key.offset <= end) {
 832                         *length += end - key.offset + 1;
 833                         break;
 834                 } else if (key.offset > end)
 835                         break;
 836
 837 next:
 838                 path->slots[0]++;
 839         }
 840         ret = 0;
 841 out:
 842         btrfs_free_path(path);
 843         return ret;
 844 }
 845
 846 /*
 847  * find_free_dev_extent - find free space in the specified device
 848  * @device:     the device which we search the free space in
 849  * @num_bytes:  the size of the free space that we need
 850  * @start:      store the start of the free space.
 851  * @len:        the size of the free space. that we find, or the size of the max
 852  *              free space if we don't find suitable free space
 853  *
 854  * this uses a pretty simple search, the expectation is that it is
 855  * called very infrequently and that a given device has a small number
 856  * of extents
 857  *
 858  * @start is used to store the start of the free space if we find. But if we
 859  * don't find suitable free space, it will be used to store the start position
 860  * of the max free space.
 861  *
 862  * @len is used to store the size of the free space that we find.
 863  * But if we don't find suitable free space, it is used to store the size of
 864  * the max free space.
 865  */
 866 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 867                          u64 *start, u64 *len)
 868 {
 869         struct btrfs_key key;
 870         struct btrfs_root *root = device->dev_root;
 871         struct btrfs_dev_extent *dev_extent;
 872         struct btrfs_path *path;
 873         u64 hole_size;
 874         u64 max_hole_start;
 875         u64 max_hole_size;
 876         u64 extent_end;
 877         u64 search_start;
 878         u64 search_end = device->total_bytes;
 879         int ret;
 880         int slot;
 881         struct extent_buffer *l;
 882
 883         /* FIXME use last free of some kind */
 884
 885         /* we don't want to overwrite the superblock on the drive,
 886          * so we make sure to start at an offset of at least 1MB
 887          */
 888         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 889
 890         max_hole_start = search_start;
 891         max_hole_size = 0;
 892         hole_size = 0;
 893
 894         if (search_start >= search_end) {
 895                 ret = -ENOSPC;
 896                 goto error;
 897         }
 898
 899         path = btrfs_alloc_path();
 900         if (!path) {
 901                 ret = -ENOMEM;
 902                 goto error;
 903         }
 904         path->reada = 2;
 905
 906         key.objectid = device->devid;
 907         key.offset = search_start;
 908         key.type = BTRFS_DEV_EXTENT_KEY;
 909
 910         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 911         if (ret < 0)
 912                 goto out;
 913         if (ret > 0) {
 914                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 915                 if (ret < 0)
 916                         goto out;
 917         }
 918
 919         while (1) {
 920                 l = path->nodes[0];
 921                 slot = path->slots[0];
 922                 if (slot >= btrfs_header_nritems(l)) {
 923                         ret = btrfs_next_leaf(root, path);
 924                         if (ret == 0)
 925                                 continue;
 926                         if (ret < 0)
 927                                 goto out;
 928
 929                         break;
 930                 }
 931                 btrfs_item_key_to_cpu(l, &key, slot);
 932
 933                 if (key.objectid < device->devid)
 934                         goto next;
 935
 936                 if (key.objectid > device->devid)
 937                         break;
 938
 939                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 940                         goto next;
 941
 942                 if (key.offset > search_start) {
 943                         hole_size = key.offset - search_start;
 944
 945                         if (hole_size > max_hole_size) {
 946                                 max_hole_start = search_start;
 947                                 max_hole_size = hole_size;
 948                         }
 949
 950                         /*
 951                          * If this free space is greater than which we need,
 952                          * it must be the max free space that we have found
 953                          * until now, so max_hole_start must point to the start
 954                          * of this free space and the length of this free space
 955                          * is stored in max_hole_size. Thus, we return
 956                          * max_hole_start and max_hole_size and go back to the
 957                          * caller.
 958                          */
 959                         if (hole_size >= num_bytes) {
 960                                 ret = 0;
 961                                 goto out;
 962                         }
 963                 }
 964
 965                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 966                 extent_end = key.offset + btrfs_dev_extent_length(l,
 967                                                                   dev_extent);
 968                 if (extent_end > search_start)
 969                         search_start = extent_end;
 970 next:
 971                 path->slots[0]++;
 972                 cond_resched();
 973         }
 974
 975         /*
 976          * At this point, search_start should be the end of
 977          * allocated dev extents, and when shrinking the device,
 978          * search_end may be smaller than search_start.
 979          */
 980         if (search_end > search_start)
 981                 hole_size = search_end - search_start;
 982
 983         if (hole_size > max_hole_size) {
 984                 max_hole_start = search_start;
 985                 max_hole_size = hole_size;
 986         }
 987
 988         /* See above. */
 989         if (hole_size < num_bytes)
 990                 ret = -ENOSPC;
 991         else
 992                 ret = 0;
 993
 994 out:
 995         btrfs_free_path(path);
 996 error:
 997         *start = max_hole_start;
 998         if (len)
 999                 *len = max_hole_size;
1000         return ret;
1001 }
1002
1003 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1004                           struct btrfs_device *device,
1005                           u64 start)
1006 {
1007         int ret;
1008         struct btrfs_path *path;
1009         struct btrfs_root *root = device->dev_root;
1010         struct btrfs_key key;
1011         struct btrfs_key found_key;
1012         struct extent_buffer *leaf = NULL;
1013         struct btrfs_dev_extent *extent = NULL;
1014
1015         path = btrfs_alloc_path();
1016         if (!path)
1017                 return -ENOMEM;
1018
1019         key.objectid = device->devid;
1020         key.offset = start;
1021         key.type = BTRFS_DEV_EXTENT_KEY;
1022 again:
1023         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1024         if (ret > 0) {
1025                 ret = btrfs_previous_item(root, path, key.objectid,
1026                                           BTRFS_DEV_EXTENT_KEY);
1027                 if (ret)
1028                         goto out;
1029                 leaf = path->nodes[0];
1030                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1031                 extent = btrfs_item_ptr(leaf, path->slots[0],
1032                                         struct btrfs_dev_extent);
1033                 BUG_ON(found_key.offset > start || found_key.offset +
1034                        btrfs_dev_extent_length(leaf, extent) < start);
1035                 key = found_key;
1036                 btrfs_release_path(path);
1037                 goto again;
1038         } else if (ret == 0) {
1039                 leaf = path->nodes[0];
1040                 extent = btrfs_item_ptr(leaf, path->slots[0],
1041                                         struct btrfs_dev_extent);
1042         }
1043         BUG_ON(ret);
1044
1045         if (device->bytes_used > 0) {
1046                 u64 len = btrfs_dev_extent_length(leaf, extent);
1047                 device->bytes_used -= len;
1048                 spin_lock(&root->fs_info->free_chunk_lock);
1049                 root->fs_info->free_chunk_space += len;
1050                 spin_unlock(&root->fs_info->free_chunk_lock);
1051         }
1052         ret = btrfs_del_item(trans, root, path);
1053
1054 out:
1055         btrfs_free_path(path);
1056         return ret;
1057 }
1058
1059 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1060                            struct btrfs_device *device,
1061                            u64 chunk_tree, u64 chunk_objectid,
1062                            u64 chunk_offset, u64 start, u64 num_bytes)
1063 {
1064         int ret;
1065         struct btrfs_path *path;
1066         struct btrfs_root *root = device->dev_root;
1067         struct btrfs_dev_extent *extent;
1068         struct extent_buffer *leaf;
1069         struct btrfs_key key;
1070
1071         WARN_ON(!device->in_fs_metadata);
1072         path = btrfs_alloc_path();
1073         if (!path)
1074                 return -ENOMEM;
1075
1076         key.objectid = device->devid;
1077         key.offset = start;
1078         key.type = BTRFS_DEV_EXTENT_KEY;
1079         ret = btrfs_insert_empty_item(trans, root, path, &key,
1080                                       sizeof(*extent));
1081         BUG_ON(ret);
1082
1083         leaf = path->nodes[0];
1084         extent = btrfs_item_ptr(leaf, path->slots[0],
1085                                 struct btrfs_dev_extent);
1086         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1087         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1088         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1089
1090         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1091                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1092                     BTRFS_UUID_SIZE);
1093
1094         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1095         btrfs_mark_buffer_dirty(leaf);
1096         btrfs_free_path(path);
1097         return ret;
1098 }
1099
1100 static noinline int find_next_chunk(struct btrfs_root *root,
1101                                     u64 objectid, u64 *offset)
1102 {
1103         struct btrfs_path *path;
1104         int ret;
1105         struct btrfs_key key;
1106         struct btrfs_chunk *chunk;
1107         struct btrfs_key found_key;
1108
1109         path = btrfs_alloc_path();
1110         if (!path)
1111                 return -ENOMEM;
1112
1113         key.objectid = objectid;
1114         key.offset = (u64)-1;
1115         key.type = BTRFS_CHUNK_ITEM_KEY;
1116
1117         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1118         if (ret < 0)
1119                 goto error;
1120
1121         BUG_ON(ret == 0);
1122
1123         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1124         if (ret) {
1125                 *offset = 0;
1126         } else {
1127                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1128                                       path->slots[0]);
1129                 if (found_key.objectid != objectid)
1130                         *offset = 0;
1131                 else {
1132                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1133                                                struct btrfs_chunk);
1134                         *offset = found_key.offset +
1135                                 btrfs_chunk_length(path->nodes[0], chunk);
1136                 }
1137         }
1138         ret = 0;
1139 error:
1140         btrfs_free_path(path);
1141         return ret;
1142 }
1143
1144 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1145 {
1146         int ret;
1147         struct btrfs_key key;
1148         struct btrfs_key found_key;
1149         struct btrfs_path *path;
1150
1151         root = root->fs_info->chunk_root;
1152
1153         path = btrfs_alloc_path();
1154         if (!path)
1155                 return -ENOMEM;
1156
1157         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1158         key.type = BTRFS_DEV_ITEM_KEY;
1159         key.offset = (u64)-1;
1160
1161         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1162         if (ret < 0)
1163                 goto error;
1164
1165         BUG_ON(ret == 0);
1166
1167         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1168                                   BTRFS_DEV_ITEM_KEY);
1169         if (ret) {
1170                 *objectid = 1;
1171         } else {
1172                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1173                                       path->slots[0]);
1174                 *objectid = found_key.offset + 1;
1175         }
1176         ret = 0;
1177 error:
1178         btrfs_free_path(path);
1179         return ret;
1180 }
1181
1182 /*
1183  * the device information is stored in the chunk root
1184  * the btrfs_device struct should be fully filled in
1185  */
1186 int btrfs_add_device(struct btrfs_trans_handle *trans,
1187                      struct btrfs_root *root,
1188                      struct btrfs_device *device)
1189 {
1190         int ret;
1191         struct btrfs_path *path;
1192         struct btrfs_dev_item *dev_item;
1193         struct extent_buffer *leaf;
1194         struct btrfs_key key;
1195         unsigned long ptr;
1196
1197         root = root->fs_info->chunk_root;
1198
1199         path = btrfs_alloc_path();
1200         if (!path)
1201                 return -ENOMEM;
1202
1203         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1204         key.type = BTRFS_DEV_ITEM_KEY;
1205         key.offset = device->devid;
1206
1207         ret = btrfs_insert_empty_item(trans, root, path, &key,
1208                                       sizeof(*dev_item));
1209         if (ret)
1210                 goto out;
1211
1212         leaf = path->nodes[0];
1213         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1214
1215         btrfs_set_device_id(leaf, dev_item, device->devid);
1216         btrfs_set_device_generation(leaf, dev_item, 0);
1217         btrfs_set_device_type(leaf, dev_item, device->type);
1218         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1219         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1220         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1221         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1222         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1223         btrfs_set_device_group(leaf, dev_item, 0);
1224         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1225         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1226         btrfs_set_device_start_offset(leaf, dev_item, 0);
1227
1228         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1229         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1230         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1231         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1232         btrfs_mark_buffer_dirty(leaf);
1233
1234         ret = 0;
1235 out:
1236         btrfs_free_path(path);
1237         return ret;
1238 }
1239
1240 static int btrfs_rm_dev_item(struct btrfs_root *root,
1241                              struct btrfs_device *device)
1242 {
1243         int ret;
1244         struct btrfs_path *path;
1245         struct btrfs_key key;
1246         struct btrfs_trans_handle *trans;
1247
1248         root = root->fs_info->chunk_root;
1249
1250         path = btrfs_alloc_path();
1251         if (!path)
1252                 return -ENOMEM;
1253
1254         trans = btrfs_start_transaction(root, 0);
1255         if (IS_ERR(trans)) {
1256                 btrfs_free_path(path);
1257                 return PTR_ERR(trans);
1258         }
1259         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1260         key.type = BTRFS_DEV_ITEM_KEY;
1261         key.offset = device->devid;
1262         lock_chunks(root);
1263
1264         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1265         if (ret < 0)
1266                 goto out;
1267
1268         if (ret > 0) {
1269                 ret = -ENOENT;
1270                 goto out;
1271         }
1272
1273         ret = btrfs_del_item(trans, root, path);
1274         if (ret)
1275                 goto out;
1276 out:
1277         btrfs_free_path(path);
1278         unlock_chunks(root);
1279         btrfs_commit_transaction(trans, root);
1280         return ret;
1281 }
1282
1283 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1284 {
1285         struct btrfs_device *device;
1286         struct btrfs_device *next_device;
1287         struct block_device *bdev;
1288         struct buffer_head *bh = NULL;
1289         struct btrfs_super_block *disk_super;
1290         struct btrfs_fs_devices *cur_devices;
1291         u64 all_avail;
1292         u64 devid;
1293         u64 num_devices;
1294         u8 *dev_uuid;
1295         int ret = 0;
1296         bool clear_super = false;
1297
1298         mutex_lock(&uuid_mutex);
1299
1300         all_avail = root->fs_info->avail_data_alloc_bits |
1301                 root->fs_info->avail_system_alloc_bits |
1302                 root->fs_info->avail_metadata_alloc_bits;
1303
1304         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1305             root->fs_info->fs_devices->num_devices <= 4) {
1306                 printk(KERN_ERR "btrfs: unable to go below four devices "
1307                        "on raid10\n");
1308                 ret = -EINVAL;
1309                 goto out;
1310         }
1311
1312         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1313             root->fs_info->fs_devices->num_devices <= 2) {
1314                 printk(KERN_ERR "btrfs: unable to go below two "
1315                        "devices on raid1\n");
1316                 ret = -EINVAL;
1317                 goto out;
1318         }
1319
1320         if (strcmp(device_path, "missing") == 0) {
1321                 struct list_head *devices;
1322                 struct btrfs_device *tmp;
1323
1324                 device = NULL;
1325                 devices = &root->fs_info->fs_devices->devices;
1326                 /*
1327                  * It is safe to read the devices since the volume_mutex
1328                  * is held.
1329                  */
1330                 list_for_each_entry(tmp, devices, dev_list) {
1331                         if (tmp->in_fs_metadata && !tmp->bdev) {
1332                                 device = tmp;
1333                                 break;
1334                         }
1335                 }
1336                 bdev = NULL;
1337                 bh = NULL;
1338                 disk_super = NULL;
1339                 if (!device) {
1340                         printk(KERN_ERR "btrfs: no missing devices found to "
1341                                "remove\n");
1342                         goto out;
1343                 }
1344         } else {
1345                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1346                                           root->fs_info->bdev_holder);
1347                 if (IS_ERR(bdev)) {
1348                         ret = PTR_ERR(bdev);
1349                         goto out;
1350                 }
1351
1352                 set_blocksize(bdev, 4096);
1353                 bh = btrfs_read_dev_super(bdev);
1354                 if (!bh) {
1355                         ret = -EINVAL;
1356                         goto error_close;
1357                 }
1358                 disk_super = (struct btrfs_super_block *)bh->b_data;
1359                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1360                 dev_uuid = disk_super->dev_item.uuid;
1361                 device = btrfs_find_device(root, devid, dev_uuid,
1362                                            disk_super->fsid);
1363                 if (!device) {
1364                         ret = -ENOENT;
1365                         goto error_brelse;
1366                 }
1367         }
1368
1369         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1370                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1371                        "device\n");
1372                 ret = -EINVAL;
1373                 goto error_brelse;
1374         }
1375
1376         if (device->writeable) {
1377                 lock_chunks(root);
1378                 list_del_init(&device->dev_alloc_list);
1379                 unlock_chunks(root);
1380                 root->fs_info->fs_devices->rw_devices--;
1381                 clear_super = true;
1382         }
1383
1384         ret = btrfs_shrink_device(device, 0);
1385         if (ret)
1386                 goto error_undo;
1387
1388         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1389         if (ret)
1390                 goto error_undo;
1391
1392         spin_lock(&root->fs_info->free_chunk_lock);
1393         root->fs_info->free_chunk_space = device->total_bytes -
1394                 device->bytes_used;
1395         spin_unlock(&root->fs_info->free_chunk_lock);
1396
1397         device->in_fs_metadata = 0;
1398         btrfs_scrub_cancel_dev(root, device);
1399
1400         /*
1401          * the device list mutex makes sure that we don't change
1402          * the device list while someone else is writing out all
1403          * the device supers.
1404          */
1405
1406         cur_devices = device->fs_devices;
1407         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1408         list_del_rcu(&device->dev_list);
1409
1410         device->fs_devices->num_devices--;
1411
1412         if (device->missing)
1413                 root->fs_info->fs_devices->missing_devices--;
1414
1415         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1416                                  struct btrfs_device, dev_list);
1417         if (device->bdev == root->fs_info->sb->s_bdev)
1418                 root->fs_info->sb->s_bdev = next_device->bdev;
1419         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1420                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1421
1422         if (device->bdev)
1423                 device->fs_devices->open_devices--;
1424
1425         call_rcu(&device->rcu, free_device);
1426         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1427
1428         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1429         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1430
1431         if (cur_devices->open_devices == 0) {
1432                 struct btrfs_fs_devices *fs_devices;
1433                 fs_devices = root->fs_info->fs_devices;
1434                 while (fs_devices) {
1435                         if (fs_devices->seed == cur_devices)
1436                                 break;
1437                         fs_devices = fs_devices->seed;
1438                 }
1439                 fs_devices->seed = cur_devices->seed;
1440                 cur_devices->seed = NULL;
1441                 lock_chunks(root);
1442                 __btrfs_close_devices(cur_devices);
1443                 unlock_chunks(root);
1444                 free_fs_devices(cur_devices);
1445         }
1446
1447         /*
1448          * at this point, the device is zero sized.  We want to
1449          * remove it from the devices list and zero out the old super
1450          */
1451         if (clear_super) {
1452                 /* make sure this device isn't detected as part of
1453                  * the FS anymore
1454                  */
1455                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1456                 set_buffer_dirty(bh);
1457                 sync_dirty_buffer(bh);
1458         }
1459
1460         ret = 0;
1461
1462 error_brelse:
1463         brelse(bh);
1464 error_close:
1465         if (bdev)
1466                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1467 out:
1468         mutex_unlock(&uuid_mutex);
1469         return ret;
1470 error_undo:
1471         if (device->writeable) {
1472                 lock_chunks(root);
1473                 list_add(&device->dev_alloc_list,
1474                          &root->fs_info->fs_devices->alloc_list);
1475                 unlock_chunks(root);
1476                 root->fs_info->fs_devices->rw_devices++;
1477         }
1478         goto error_brelse;
1479 }
1480
1481 /*
1482  * does all the dirty work required for changing file system's UUID.
1483  */
1484 static int btrfs_prepare_sprout(struct btrfs_root *root)
1485 {
1486         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1487         struct btrfs_fs_devices *old_devices;
1488         struct btrfs_fs_devices *seed_devices;
1489         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1490         struct btrfs_device *device;
1491         u64 super_flags;
1492
1493         BUG_ON(!mutex_is_locked(&uuid_mutex));
1494         if (!fs_devices->seeding)
1495                 return -EINVAL;
1496
1497         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1498         if (!seed_devices)
1499                 return -ENOMEM;
1500
1501         old_devices = clone_fs_devices(fs_devices);
1502         if (IS_ERR(old_devices)) {
1503                 kfree(seed_devices);
1504                 return PTR_ERR(old_devices);
1505         }
1506
1507         list_add(&old_devices->list, &fs_uuids);
1508
1509         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1510         seed_devices->opened = 1;
1511         INIT_LIST_HEAD(&seed_devices->devices);
1512         INIT_LIST_HEAD(&seed_devices->alloc_list);
1513         mutex_init(&seed_devices->device_list_mutex);
1514
1515         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1516         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1517                               synchronize_rcu);
1518         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1519
1520         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1521         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1522                 device->fs_devices = seed_devices;
1523         }
1524
1525         fs_devices->seeding = 0;
1526         fs_devices->num_devices = 0;
1527         fs_devices->open_devices = 0;
1528         fs_devices->seed = seed_devices;
1529
1530         generate_random_uuid(fs_devices->fsid);
1531         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1532         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1533         super_flags = btrfs_super_flags(disk_super) &
1534                       ~BTRFS_SUPER_FLAG_SEEDING;
1535         btrfs_set_super_flags(disk_super, super_flags);
1536
1537         return 0;
1538 }
1539
1540 /*
1541  * strore the expected generation for seed devices in device items.
1542  */
1543 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1544                                struct btrfs_root *root)
1545 {
1546         struct btrfs_path *path;
1547         struct extent_buffer *leaf;
1548         struct btrfs_dev_item *dev_item;
1549         struct btrfs_device *device;
1550         struct btrfs_key key;
1551         u8 fs_uuid[BTRFS_UUID_SIZE];
1552         u8 dev_uuid[BTRFS_UUID_SIZE];
1553         u64 devid;
1554         int ret;
1555
1556         path = btrfs_alloc_path();
1557         if (!path)
1558                 return -ENOMEM;
1559
1560         root = root->fs_info->chunk_root;
1561         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1562         key.offset = 0;
1563         key.type = BTRFS_DEV_ITEM_KEY;
1564
1565         while (1) {
1566                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1567                 if (ret < 0)
1568                         goto error;
1569
1570                 leaf = path->nodes[0];
1571 next_slot:
1572                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1573                         ret = btrfs_next_leaf(root, path);
1574                         if (ret > 0)
1575                                 break;
1576                         if (ret < 0)
1577                                 goto error;
1578                         leaf = path->nodes[0];
1579                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1580                         btrfs_release_path(path);
1581                         continue;
1582                 }
1583
1584                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1585                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1586                     key.type != BTRFS_DEV_ITEM_KEY)
1587                         break;
1588
1589                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1590                                           struct btrfs_dev_item);
1591                 devid = btrfs_device_id(leaf, dev_item);
1592                 read_extent_buffer(leaf, dev_uuid,
1593                                    (unsigned long)btrfs_device_uuid(dev_item),
1594                                    BTRFS_UUID_SIZE);
1595                 read_extent_buffer(leaf, fs_uuid,
1596                                    (unsigned long)btrfs_device_fsid(dev_item),
1597                                    BTRFS_UUID_SIZE);
1598                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1599                 BUG_ON(!device);
1600
1601                 if (device->fs_devices->seeding) {
1602                         btrfs_set_device_generation(leaf, dev_item,
1603                                                     device->generation);
1604                         btrfs_mark_buffer_dirty(leaf);
1605                 }
1606
1607                 path->slots[0]++;
1608                 goto next_slot;
1609         }
1610         ret = 0;
1611 error:
1612         btrfs_free_path(path);
1613         return ret;
1614 }
1615
1616 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1617 {
1618         struct request_queue *q;
1619         struct btrfs_trans_handle *trans;
1620         struct btrfs_device *device;
1621         struct block_device *bdev;
1622         struct list_head *devices;
1623         struct super_block *sb = root->fs_info->sb;
1624         u64 total_bytes;
1625         int seeding_dev = 0;
1626         int ret = 0;
1627
1628         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1629                 return -EINVAL;
1630
1631         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1632                                   root->fs_info->bdev_holder);
1633         if (IS_ERR(bdev))
1634                 return PTR_ERR(bdev);
1635
1636         if (root->fs_info->fs_devices->seeding) {
1637                 seeding_dev = 1;
1638                 down_write(&sb->s_umount);
1639                 mutex_lock(&uuid_mutex);
1640         }
1641
1642         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1643
1644         devices = &root->fs_info->fs_devices->devices;
1645         /*
1646          * we have the volume lock, so we don't need the extra
1647          * device list mutex while reading the list here.
1648          */
1649         list_for_each_entry(device, devices, dev_list) {
1650                 if (device->bdev == bdev) {
1651                         ret = -EEXIST;
1652                         goto error;
1653                 }
1654         }
1655
1656         device = kzalloc(sizeof(*device), GFP_NOFS);
1657         if (!device) {
1658                 /* we can safely leave the fs_devices entry around */
1659                 ret = -ENOMEM;
1660                 goto error;
1661         }
1662
1663         device->name = kstrdup(device_path, GFP_NOFS);
1664         if (!device->name) {
1665                 kfree(device);
1666                 ret = -ENOMEM;
1667                 goto error;
1668         }
1669
1670         ret = find_next_devid(root, &device->devid);
1671         if (ret) {
1672                 kfree(device->name);
1673                 kfree(device);
1674                 goto error;
1675         }
1676
1677         trans = btrfs_start_transaction(root, 0);
1678         if (IS_ERR(trans)) {
1679                 kfree(device->name);
1680                 kfree(device);
1681                 ret = PTR_ERR(trans);
1682                 goto error;
1683         }
1684
1685         lock_chunks(root);
1686
1687         q = bdev_get_queue(bdev);
1688         if (blk_queue_discard(q))
1689                 device->can_discard = 1;
1690         device->writeable = 1;
1691         device->work.func = pending_bios_fn;
1692         generate_random_uuid(device->uuid);
1693         spin_lock_init(&device->io_lock);
1694         device->generation = trans->transid;
1695         device->io_width = root->sectorsize;
1696         device->io_align = root->sectorsize;
1697         device->sector_size = root->sectorsize;
1698         device->total_bytes = i_size_read(bdev->bd_inode);
1699         device->disk_total_bytes = device->total_bytes;
1700         device->dev_root = root->fs_info->dev_root;
1701         device->bdev = bdev;
1702         device->in_fs_metadata = 1;
1703         device->mode = FMODE_EXCL;
1704         set_blocksize(device->bdev, 4096);
1705
1706         if (seeding_dev) {
1707                 sb->s_flags &= ~MS_RDONLY;
1708                 ret = btrfs_prepare_sprout(root);
1709                 BUG_ON(ret);
1710         }
1711
1712         device->fs_devices = root->fs_info->fs_devices;
1713
1714         /*
1715          * we don't want write_supers to jump in here with our device
1716          * half setup
1717          */
1718         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1719         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1720         list_add(&device->dev_alloc_list,
1721                  &root->fs_info->fs_devices->alloc_list);
1722         root->fs_info->fs_devices->num_devices++;
1723         root->fs_info->fs_devices->open_devices++;
1724         root->fs_info->fs_devices->rw_devices++;
1725         if (device->can_discard)
1726                 root->fs_info->fs_devices->num_can_discard++;
1727         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1728
1729         spin_lock(&root->fs_info->free_chunk_lock);
1730         root->fs_info->free_chunk_space += device->total_bytes;
1731         spin_unlock(&root->fs_info->free_chunk_lock);
1732
1733         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1734                 root->fs_info->fs_devices->rotating = 1;
1735
1736         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1737         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1738                                     total_bytes + device->total_bytes);
1739
1740         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1741         btrfs_set_super_num_devices(root->fs_info->super_copy,
1742                                     total_bytes + 1);
1743         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1744
1745         if (seeding_dev) {
1746                 ret = init_first_rw_device(trans, root, device);
1747                 BUG_ON(ret);
1748                 ret = btrfs_finish_sprout(trans, root);
1749                 BUG_ON(ret);
1750         } else {
1751                 ret = btrfs_add_device(trans, root, device);
1752         }
1753
1754         /*
1755          * we've got more storage, clear any full flags on the space
1756          * infos
1757          */
1758         btrfs_clear_space_info_full(root->fs_info);
1759
1760         unlock_chunks(root);
1761         btrfs_commit_transaction(trans, root);
1762
1763         if (seeding_dev) {
1764                 mutex_unlock(&uuid_mutex);
1765                 up_write(&sb->s_umount);
1766
1767                 ret = btrfs_relocate_sys_chunks(root);
1768                 BUG_ON(ret);
1769         }
1770
1771         return ret;
1772 error:
1773         blkdev_put(bdev, FMODE_EXCL);
1774         if (seeding_dev) {
1775                 mutex_unlock(&uuid_mutex);
1776                 up_write(&sb->s_umount);
1777         }
1778         return ret;
1779 }
1780
1781 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1782                                         struct btrfs_device *device)
1783 {
1784         int ret;
1785         struct btrfs_path *path;
1786         struct btrfs_root *root;
1787         struct btrfs_dev_item *dev_item;
1788         struct extent_buffer *leaf;
1789         struct btrfs_key key;
1790
1791         root = device->dev_root->fs_info->chunk_root;
1792
1793         path = btrfs_alloc_path();
1794         if (!path)
1795                 return -ENOMEM;
1796
1797         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1798         key.type = BTRFS_DEV_ITEM_KEY;
1799         key.offset = device->devid;
1800
1801         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1802         if (ret < 0)
1803                 goto out;
1804
1805         if (ret > 0) {
1806                 ret = -ENOENT;
1807                 goto out;
1808         }
1809
1810         leaf = path->nodes[0];
1811         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1812
1813         btrfs_set_device_id(leaf, dev_item, device->devid);
1814         btrfs_set_device_type(leaf, dev_item, device->type);
1815         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1816         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1817         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1818         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1819         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1820         btrfs_mark_buffer_dirty(leaf);
1821
1822 out:
1823         btrfs_free_path(path);
1824         return ret;
1825 }
1826
1827 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1828                       struct btrfs_device *device, u64 new_size)
1829 {
1830         struct btrfs_super_block *super_copy =
1831                 device->dev_root->fs_info->super_copy;
1832         u64 old_total = btrfs_super_total_bytes(super_copy);
1833         u64 diff = new_size - device->total_bytes;
1834
1835         if (!device->writeable)
1836                 return -EACCES;
1837         if (new_size <= device->total_bytes)
1838                 return -EINVAL;
1839
1840         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1841         device->fs_devices->total_rw_bytes += diff;
1842
1843         device->total_bytes = new_size;
1844         device->disk_total_bytes = new_size;
1845         btrfs_clear_space_info_full(device->dev_root->fs_info);
1846
1847         return btrfs_update_device(trans, device);
1848 }
1849
1850 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1851                       struct btrfs_device *device, u64 new_size)
1852 {
1853         int ret;
1854         lock_chunks(device->dev_root);
1855         ret = __btrfs_grow_device(trans, device, new_size);
1856         unlock_chunks(device->dev_root);
1857         return ret;
1858 }
1859
1860 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1861                             struct btrfs_root *root,
1862                             u64 chunk_tree, u64 chunk_objectid,
1863                             u64 chunk_offset)
1864 {
1865         int ret;
1866         struct btrfs_path *path;
1867         struct btrfs_key key;
1868
1869         root = root->fs_info->chunk_root;
1870         path = btrfs_alloc_path();
1871         if (!path)
1872                 return -ENOMEM;
1873
1874         key.objectid = chunk_objectid;
1875         key.offset = chunk_offset;
1876         key.type = BTRFS_CHUNK_ITEM_KEY;
1877
1878         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1879         BUG_ON(ret);
1880
1881         ret = btrfs_del_item(trans, root, path);
1882
1883         btrfs_free_path(path);
1884         return ret;
1885 }
1886
1887 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1888                         chunk_offset)
1889 {
1890         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1891         struct btrfs_disk_key *disk_key;
1892         struct btrfs_chunk *chunk;
1893         u8 *ptr;
1894         int ret = 0;
1895         u32 num_stripes;
1896         u32 array_size;
1897         u32 len = 0;
1898         u32 cur;
1899         struct btrfs_key key;
1900
1901         array_size = btrfs_super_sys_array_size(super_copy);
1902
1903         ptr = super_copy->sys_chunk_array;
1904         cur = 0;
1905
1906         while (cur < array_size) {
1907                 disk_key = (struct btrfs_disk_key *)ptr;
1908                 btrfs_disk_key_to_cpu(&key, disk_key);
1909
1910                 len = sizeof(*disk_key);
1911
1912                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1913                         chunk = (struct btrfs_chunk *)(ptr + len);
1914                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1915                         len += btrfs_chunk_item_size(num_stripes);
1916                 } else {
1917                         ret = -EIO;
1918                         break;
1919                 }
1920                 if (key.objectid == chunk_objectid &&
1921                     key.offset == chunk_offset) {
1922                         memmove(ptr, ptr + len, array_size - (cur + len));
1923                         array_size -= len;
1924                         btrfs_set_super_sys_array_size(super_copy, array_size);
1925                 } else {
1926                         ptr += len;
1927                         cur += len;
1928                 }
1929         }
1930         return ret;
1931 }
1932
1933 static int btrfs_relocate_chunk(struct btrfs_root *root,
1934                          u64 chunk_tree, u64 chunk_objectid,
1935                          u64 chunk_offset)
1936 {
1937         struct extent_map_tree *em_tree;
1938         struct btrfs_root *extent_root;
1939         struct btrfs_trans_handle *trans;
1940         struct extent_map *em;
1941         struct map_lookup *map;
1942         int ret;
1943         int i;
1944
1945         root = root->fs_info->chunk_root;
1946         extent_root = root->fs_info->extent_root;
1947         em_tree = &root->fs_info->mapping_tree.map_tree;
1948
1949         ret = btrfs_can_relocate(extent_root, chunk_offset);
1950         if (ret)
1951                 return -ENOSPC;
1952
1953         /* step one, relocate all the extents inside this chunk */
1954         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1955         if (ret)
1956                 return ret;
1957
1958         trans = btrfs_start_transaction(root, 0);
1959         BUG_ON(IS_ERR(trans));
1960
1961         lock_chunks(root);
1962
1963         /*
1964          * step two, delete the device extents and the
1965          * chunk tree entries
1966          */
1967         read_lock(&em_tree->lock);
1968         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1969         read_unlock(&em_tree->lock);
1970
1971         BUG_ON(!em || em->start > chunk_offset ||
1972                em->start + em->len < chunk_offset);
1973         map = (struct map_lookup *)em->bdev;
1974
1975         for (i = 0; i < map->num_stripes; i++) {
1976                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1977                                             map->stripes[i].physical);
1978                 BUG_ON(ret);
1979
1980                 if (map->stripes[i].dev) {
1981                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1982                         BUG_ON(ret);
1983                 }
1984         }
1985         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1986                                chunk_offset);
1987
1988         BUG_ON(ret);
1989
1990         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1991
1992         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1993                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1994                 BUG_ON(ret);
1995         }
1996
1997         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1998         BUG_ON(ret);
1999
2000         write_lock(&em_tree->lock);
2001         remove_extent_mapping(em_tree, em);
2002         write_unlock(&em_tree->lock);
2003
2004         kfree(map);
2005         em->bdev = NULL;
2006
2007         /* once for the tree */
2008         free_extent_map(em);
2009         /* once for us */
2010         free_extent_map(em);
2011
2012         unlock_chunks(root);
2013         btrfs_end_transaction(trans, root);
2014         return 0;
2015 }
2016
2017 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2018 {
2019         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2020         struct btrfs_path *path;
2021         struct extent_buffer *leaf;
2022         struct btrfs_chunk *chunk;
2023         struct btrfs_key key;
2024         struct btrfs_key found_key;
2025         u64 chunk_tree = chunk_root->root_key.objectid;
2026         u64 chunk_type;
2027         bool retried = false;
2028         int failed = 0;
2029         int ret;
2030
2031         path = btrfs_alloc_path();
2032         if (!path)
2033                 return -ENOMEM;
2034
2035 again:
2036         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2037         key.offset = (u64)-1;
2038         key.type = BTRFS_CHUNK_ITEM_KEY;
2039
2040         while (1) {
2041                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2042                 if (ret < 0)
2043                         goto error;
2044                 BUG_ON(ret == 0);
2045
2046                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2047                                           key.type);
2048                 if (ret < 0)
2049                         goto error;
2050                 if (ret > 0)
2051                         break;
2052
2053                 leaf = path->nodes[0];
2054                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2055
2056                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2057                                        struct btrfs_chunk);
2058                 chunk_type = btrfs_chunk_type(leaf, chunk);
2059                 btrfs_release_path(path);
2060
2061                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2062                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2063                                                    found_key.objectid,
2064                                                    found_key.offset);
2065                         if (ret == -ENOSPC)
2066                                 failed++;
2067                         else if (ret)
2068                                 BUG();
2069                 }
2070
2071                 if (found_key.offset == 0)
2072                         break;
2073                 key.offset = found_key.offset - 1;
2074         }
2075         ret = 0;
2076         if (failed && !retried) {
2077                 failed = 0;
2078                 retried = true;
2079                 goto again;
2080         } else if (failed && retried) {
2081                 WARN_ON(1);
2082                 ret = -ENOSPC;
2083         }
2084 error:
2085         btrfs_free_path(path);
2086         return ret;
2087 }
2088
2089 static int insert_balance_item(struct btrfs_root *root,
2090                                struct btrfs_balance_control *bctl)
2091 {
2092         struct btrfs_trans_handle *trans;
2093         struct btrfs_balance_item *item;
2094         struct btrfs_disk_balance_args disk_bargs;
2095         struct btrfs_path *path;
2096         struct extent_buffer *leaf;
2097         struct btrfs_key key;
2098         int ret, err;
2099
2100         path = btrfs_alloc_path();
2101         if (!path)
2102                 return -ENOMEM;
2103
2104         trans = btrfs_start_transaction(root, 0);
2105         if (IS_ERR(trans)) {
2106                 btrfs_free_path(path);
2107                 return PTR_ERR(trans);
2108         }
2109
2110         key.objectid = BTRFS_BALANCE_OBJECTID;
2111         key.type = BTRFS_BALANCE_ITEM_KEY;
2112         key.offset = 0;
2113
2114         ret = btrfs_insert_empty_item(trans, root, path, &key,
2115                                       sizeof(*item));
2116         if (ret)
2117                 goto out;
2118
2119         leaf = path->nodes[0];
2120         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2121
2122         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2123
2124         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2125         btrfs_set_balance_data(leaf, item, &disk_bargs);
2126         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2127         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2128         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2129         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2130
2131         btrfs_set_balance_flags(leaf, item, bctl->flags);
2132
2133         btrfs_mark_buffer_dirty(leaf);
2134 out:
2135         btrfs_free_path(path);
2136         err = btrfs_commit_transaction(trans, root);
2137         if (err && !ret)
2138                 ret = err;
2139         return ret;
2140 }
2141
2142 static int del_balance_item(struct btrfs_root *root)
2143 {
2144         struct btrfs_trans_handle *trans;
2145         struct btrfs_path *path;
2146         struct btrfs_key key;
2147         int ret, err;
2148
2149         path = btrfs_alloc_path();
2150         if (!path)
2151                 return -ENOMEM;
2152
2153         trans = btrfs_start_transaction(root, 0);
2154         if (IS_ERR(trans)) {
2155                 btrfs_free_path(path);
2156                 return PTR_ERR(trans);
2157         }
2158
2159         key.objectid = BTRFS_BALANCE_OBJECTID;
2160         key.type = BTRFS_BALANCE_ITEM_KEY;
2161         key.offset = 0;
2162
2163         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2164         if (ret < 0)
2165                 goto out;
2166         if (ret > 0) {
2167                 ret = -ENOENT;
2168                 goto out;
2169         }
2170
2171         ret = btrfs_del_item(trans, root, path);
2172 out:
2173         btrfs_free_path(path);
2174         err = btrfs_commit_transaction(trans, root);
2175         if (err && !ret)
2176                 ret = err;
2177         return ret;
2178 }
2179
2180 /*
2181  * This is a heuristic used to reduce the number of chunks balanced on
2182  * resume after balance was interrupted.
2183  */
2184 static void update_balance_args(struct btrfs_balance_control *bctl)
2185 {
2186         /*
2187          * Turn on soft mode for chunk types that were being converted.
2188          */
2189         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2190                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2191         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2192                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2193         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2194                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2195
2196         /*
2197          * Turn on usage filter if is not already used.  The idea is
2198          * that chunks that we have already balanced should be
2199          * reasonably full.  Don't do it for chunks that are being
2200          * converted - that will keep us from relocating unconverted
2201          * (albeit full) chunks.
2202          */
2203         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2204             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2205                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2206                 bctl->data.usage = 90;
2207         }
2208         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2209             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2210                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2211                 bctl->sys.usage = 90;
2212         }
2213         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2214             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2215                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2216                 bctl->meta.usage = 90;
2217         }
2218 }
2219
2220 /*
2221  * Should be called with both balance and volume mutexes held to
2222  * serialize other volume operations (add_dev/rm_dev/resize) with
2223  * restriper.  Same goes for unset_balance_control.
2224  */
2225 static void set_balance_control(struct btrfs_balance_control *bctl)
2226 {
2227         struct btrfs_fs_info *fs_info = bctl->fs_info;
2228
2229         BUG_ON(fs_info->balance_ctl);
2230
2231         spin_lock(&fs_info->balance_lock);
2232         fs_info->balance_ctl = bctl;
2233         spin_unlock(&fs_info->balance_lock);
2234 }
2235
2236 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2237 {
2238         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2239
2240         BUG_ON(!fs_info->balance_ctl);
2241
2242         spin_lock(&fs_info->balance_lock);
2243         fs_info->balance_ctl = NULL;
2244         spin_unlock(&fs_info->balance_lock);
2245
2246         kfree(bctl);
2247 }
2248
2249 /*
2250  * Balance filters.  Return 1 if chunk should be filtered out
2251  * (should not be balanced).
2252  */
2253 static int chunk_profiles_filter(u64 chunk_type,
2254                                  struct btrfs_balance_args *bargs)
2255 {
2256         chunk_type = chunk_to_extended(chunk_type) &
2257                                 BTRFS_EXTENDED_PROFILE_MASK;
2258
2259         if (bargs->profiles & chunk_type)
2260                 return 0;
2261
2262         return 1;
2263 }
2264
2265 static u64 div_factor_fine(u64 num, int factor)
2266 {
2267         if (factor <= 0)
2268                 return 0;
2269         if (factor >= 100)
2270                 return num;
2271
2272         num *= factor;
2273         do_div(num, 100);
2274         return num;
2275 }
2276
2277 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2278                               struct btrfs_balance_args *bargs)
2279 {
2280         struct btrfs_block_group_cache *cache;
2281         u64 chunk_used, user_thresh;
2282         int ret = 1;
2283
2284         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2285         chunk_used = btrfs_block_group_used(&cache->item);
2286
2287         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2288         if (chunk_used < user_thresh)
2289                 ret = 0;
2290
2291         btrfs_put_block_group(cache);
2292         return ret;
2293 }
2294
2295 static int chunk_devid_filter(struct extent_buffer *leaf,
2296                               struct btrfs_chunk *chunk,
2297                               struct btrfs_balance_args *bargs)
2298 {
2299         struct btrfs_stripe *stripe;
2300         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2301         int i;
2302
2303         for (i = 0; i < num_stripes; i++) {
2304                 stripe = btrfs_stripe_nr(chunk, i);
2305                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2306                         return 0;
2307         }
2308
2309         return 1;
2310 }
2311
2312 /* [pstart, pend) */
2313 static int chunk_drange_filter(struct extent_buffer *leaf,
2314                                struct btrfs_chunk *chunk,
2315                                u64 chunk_offset,
2316                                struct btrfs_balance_args *bargs)
2317 {
2318         struct btrfs_stripe *stripe;
2319         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2320         u64 stripe_offset;
2321         u64 stripe_length;
2322         int factor;
2323         int i;
2324
2325         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2326                 return 0;
2327
2328         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2329              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2330                 factor = 2;
2331         else
2332                 factor = 1;
2333         factor = num_stripes / factor;
2334
2335         for (i = 0; i < num_stripes; i++) {
2336                 stripe = btrfs_stripe_nr(chunk, i);
2337                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2338                         continue;
2339
2340                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2341                 stripe_length = btrfs_chunk_length(leaf, chunk);
2342                 do_div(stripe_length, factor);
2343
2344                 if (stripe_offset < bargs->pend &&
2345                     stripe_offset + stripe_length > bargs->pstart)
2346                         return 0;
2347         }
2348
2349         return 1;
2350 }
2351
2352 /* [vstart, vend) */
2353 static int chunk_vrange_filter(struct extent_buffer *leaf,
2354                                struct btrfs_chunk *chunk,
2355                                u64 chunk_offset,
2356                                struct btrfs_balance_args *bargs)
2357 {
2358         if (chunk_offset < bargs->vend &&
2359             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2360                 /* at least part of the chunk is inside this vrange */
2361                 return 0;
2362
2363         return 1;
2364 }
2365
2366 static int chunk_soft_convert_filter(u64 chunk_type,
2367                                      struct btrfs_balance_args *bargs)
2368 {
2369         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2370                 return 0;
2371
2372         chunk_type = chunk_to_extended(chunk_type) &
2373                                 BTRFS_EXTENDED_PROFILE_MASK;
2374
2375         if (bargs->target == chunk_type)
2376                 return 1;
2377
2378         return 0;
2379 }
2380
2381 static int should_balance_chunk(struct btrfs_root *root,
2382                                 struct extent_buffer *leaf,
2383                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2384 {
2385         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2386         struct btrfs_balance_args *bargs = NULL;
2387         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2388
2389         /* type filter */
2390         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2391               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2392                 return 0;
2393         }
2394
2395         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2396                 bargs = &bctl->data;
2397         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2398                 bargs = &bctl->sys;
2399         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2400                 bargs = &bctl->meta;
2401
2402         /* profiles filter */
2403         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2404             chunk_profiles_filter(chunk_type, bargs)) {
2405                 return 0;
2406         }
2407
2408         /* usage filter */
2409         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2410             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2411                 return 0;
2412         }
2413
2414         /* devid filter */
2415         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2416             chunk_devid_filter(leaf, chunk, bargs)) {
2417                 return 0;
2418         }
2419
2420         /* drange filter, makes sense only with devid filter */
2421         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2422             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2423                 return 0;
2424         }
2425
2426         /* vrange filter */
2427         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2428             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2429                 return 0;
2430         }
2431
2432         /* soft profile changing mode */
2433         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2434             chunk_soft_convert_filter(chunk_type, bargs)) {
2435                 return 0;
2436         }
2437
2438         return 1;
2439 }
2440
2441 static u64 div_factor(u64 num, int factor)
2442 {
2443         if (factor == 10)
2444                 return num;
2445         num *= factor;
2446         do_div(num, 10);
2447         return num;
2448 }
2449
2450 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2451 {
2452         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2453         struct btrfs_root *chunk_root = fs_info->chunk_root;
2454         struct btrfs_root *dev_root = fs_info->dev_root;
2455         struct list_head *devices;
2456         struct btrfs_device *device;
2457         u64 old_size;
2458         u64 size_to_free;
2459         struct btrfs_chunk *chunk;
2460         struct btrfs_path *path;
2461         struct btrfs_key key;
2462         struct btrfs_key found_key;
2463         struct btrfs_trans_handle *trans;
2464         struct extent_buffer *leaf;
2465         int slot;
2466         int ret;
2467         int enospc_errors = 0;
2468         bool counting = true;
2469
2470         /* step one make some room on all the devices */
2471         devices = &fs_info->fs_devices->devices;
2472         list_for_each_entry(device, devices, dev_list) {
2473                 old_size = device->total_bytes;
2474                 size_to_free = div_factor(old_size, 1);
2475                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2476                 if (!device->writeable ||
2477                     device->total_bytes - device->bytes_used > size_to_free)
2478                         continue;
2479
2480                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2481                 if (ret == -ENOSPC)
2482                         break;
2483                 BUG_ON(ret);
2484
2485                 trans = btrfs_start_transaction(dev_root, 0);
2486                 BUG_ON(IS_ERR(trans));
2487
2488                 ret = btrfs_grow_device(trans, device, old_size);
2489                 BUG_ON(ret);
2490
2491                 btrfs_end_transaction(trans, dev_root);
2492         }
2493
2494         /* step two, relocate all the chunks */
2495         path = btrfs_alloc_path();
2496         if (!path) {
2497                 ret = -ENOMEM;
2498                 goto error;
2499         }
2500
2501         /* zero out stat counters */
2502         spin_lock(&fs_info->balance_lock);
2503         memset(&bctl->stat, 0, sizeof(bctl->stat));
2504         spin_unlock(&fs_info->balance_lock);
2505 again:
2506         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2507         key.offset = (u64)-1;
2508         key.type = BTRFS_CHUNK_ITEM_KEY;
2509
2510         while (1) {
2511                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2512                     atomic_read(&fs_info->balance_cancel_req)) {
2513                         ret = -ECANCELED;
2514                         goto error;
2515                 }
2516
2517                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2518                 if (ret < 0)
2519                         goto error;
2520
2521                 /*
2522                  * this shouldn't happen, it means the last relocate
2523                  * failed
2524                  */
2525                 if (ret == 0)
2526                         BUG(); /* FIXME break ? */
2527
2528                 ret = btrfs_previous_item(chunk_root, path, 0,
2529                                           BTRFS_CHUNK_ITEM_KEY);
2530                 if (ret) {
2531                         ret = 0;
2532                         break;
2533                 }
2534
2535                 leaf = path->nodes[0];
2536                 slot = path->slots[0];
2537                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2538
2539                 if (found_key.objectid != key.objectid)
2540                         break;
2541
2542                 /* chunk zero is special */
2543                 if (found_key.offset == 0)
2544                         break;
2545
2546                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2547
2548                 if (!counting) {
2549                         spin_lock(&fs_info->balance_lock);
2550                         bctl->stat.considered++;
2551                         spin_unlock(&fs_info->balance_lock);
2552                 }
2553
2554                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2555                                            found_key.offset);
2556                 btrfs_release_path(path);
2557                 if (!ret)
2558                         goto loop;
2559
2560                 if (counting) {
2561                         spin_lock(&fs_info->balance_lock);
2562                         bctl->stat.expected++;
2563                         spin_unlock(&fs_info->balance_lock);
2564                         goto loop;
2565                 }
2566
2567                 ret = btrfs_relocate_chunk(chunk_root,
2568                                            chunk_root->root_key.objectid,
2569                                            found_key.objectid,
2570                                            found_key.offset);
2571                 if (ret && ret != -ENOSPC)
2572                         goto error;
2573                 if (ret == -ENOSPC) {
2574                         enospc_errors++;
2575                 } else {
2576                         spin_lock(&fs_info->balance_lock);
2577                         bctl->stat.completed++;
2578                         spin_unlock(&fs_info->balance_lock);
2579                 }
2580 loop:
2581                 key.offset = found_key.offset - 1;
2582         }
2583
2584         if (counting) {
2585                 btrfs_release_path(path);
2586                 counting = false;
2587                 goto again;
2588         }
2589 error:
2590         btrfs_free_path(path);
2591         if (enospc_errors) {
2592                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2593                        enospc_errors);
2594                 if (!ret)
2595                         ret = -ENOSPC;
2596         }
2597
2598         return ret;
2599 }
2600
2601 /**
2602  * alloc_profile_is_valid - see if a given profile is valid and reduced
2603  * @flags: profile to validate
2604  * @extended: if true @flags is treated as an extended profile
2605  */
2606 static int alloc_profile_is_valid(u64 flags, int extended)
2607 {
2608         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2609                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2610
2611         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2612
2613         /* 1) check that all other bits are zeroed */
2614         if (flags & ~mask)
2615                 return 0;
2616
2617         /* 2) see if profile is reduced */
2618         if (flags == 0)
2619                 return !extended; /* "0" is valid for usual profiles */
2620
2621         /* true if exactly one bit set */
2622         return (flags & (flags - 1)) == 0;
2623 }
2624
2625 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2626 {
2627         /* cancel requested || normal exit path */
2628         return atomic_read(&fs_info->balance_cancel_req) ||
2629                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2630                  atomic_read(&fs_info->balance_cancel_req) == 0);
2631 }
2632
2633 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2634 {
2635         int ret;
2636
2637         unset_balance_control(fs_info);
2638         ret = del_balance_item(fs_info->tree_root);
2639         BUG_ON(ret);
2640 }
2641
2642 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2643                                struct btrfs_ioctl_balance_args *bargs);
2644
2645 /*
2646  * Should be called with both balance and volume mutexes held
2647  */
2648 int btrfs_balance(struct btrfs_balance_control *bctl,
2649                   struct btrfs_ioctl_balance_args *bargs)
2650 {
2651         struct btrfs_fs_info *fs_info = bctl->fs_info;
2652         u64 allowed;
2653         int mixed = 0;
2654         int ret;
2655
2656         if (btrfs_fs_closing(fs_info) ||
2657             atomic_read(&fs_info->balance_pause_req) ||
2658             atomic_read(&fs_info->balance_cancel_req)) {
2659                 ret = -EINVAL;
2660                 goto out;
2661         }
2662
2663         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2664         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2665                 mixed = 1;
2666
2667         /*
2668          * In case of mixed groups both data and meta should be picked,
2669          * and identical options should be given for both of them.
2670          */
2671         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2672         if (mixed && (bctl->flags & allowed)) {
2673                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2674                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2675                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2676                         printk(KERN_ERR "btrfs: with mixed groups data and "
2677                                "metadata balance options must be the same\n");
2678                         ret = -EINVAL;
2679                         goto out;
2680                 }
2681         }
2682
2683         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2684         if (fs_info->fs_devices->num_devices == 1)
2685                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2686         else if (fs_info->fs_devices->num_devices < 4)
2687                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2688         else
2689                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2690                                 BTRFS_BLOCK_GROUP_RAID10);
2691
2692         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2693             (!alloc_profile_is_valid(bctl->data.target, 1) ||
2694              (bctl->data.target & ~allowed))) {
2695                 printk(KERN_ERR "btrfs: unable to start balance with target "
2696                        "data profile %llu\n",
2697                        (unsigned long long)bctl->data.target);
2698                 ret = -EINVAL;
2699                 goto out;
2700         }
2701         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2702             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2703              (bctl->meta.target & ~allowed))) {
2704                 printk(KERN_ERR "btrfs: unable to start balance with target "
2705                        "metadata profile %llu\n",
2706                        (unsigned long long)bctl->meta.target);
2707                 ret = -EINVAL;
2708                 goto out;
2709         }
2710         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2711             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2712              (bctl->sys.target & ~allowed))) {
2713                 printk(KERN_ERR "btrfs: unable to start balance with target "
2714                        "system profile %llu\n",
2715                        (unsigned long long)bctl->sys.target);
2716                 ret = -EINVAL;
2717                 goto out;
2718         }
2719
2720         /* allow dup'ed data chunks only in mixed mode */
2721         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2722             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2723                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2724                 ret = -EINVAL;
2725                 goto out;
2726         }
2727
2728         /* allow to reduce meta or sys integrity only if force set */
2729         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2730                         BTRFS_BLOCK_GROUP_RAID10;
2731         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2732              (fs_info->avail_system_alloc_bits & allowed) &&
2733              !(bctl->sys.target & allowed)) ||
2734             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2735              (fs_info->avail_metadata_alloc_bits & allowed) &&
2736              !(bctl->meta.target & allowed))) {
2737                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2738                         printk(KERN_INFO "btrfs: force reducing metadata "
2739                                "integrity\n");
2740                 } else {
2741                         printk(KERN_ERR "btrfs: balance will reduce metadata "
2742                                "integrity, use force if you want this\n");
2743                         ret = -EINVAL;
2744                         goto out;
2745                 }
2746         }
2747
2748         ret = insert_balance_item(fs_info->tree_root, bctl);
2749         if (ret && ret != -EEXIST)
2750                 goto out;
2751
2752         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2753                 BUG_ON(ret == -EEXIST);
2754                 set_balance_control(bctl);
2755         } else {
2756                 BUG_ON(ret != -EEXIST);
2757                 spin_lock(&fs_info->balance_lock);
2758                 update_balance_args(bctl);
2759                 spin_unlock(&fs_info->balance_lock);
2760         }
2761
2762         atomic_inc(&fs_info->balance_running);
2763         mutex_unlock(&fs_info->balance_mutex);
2764
2765         ret = __btrfs_balance(fs_info);
2766
2767         mutex_lock(&fs_info->balance_mutex);
2768         atomic_dec(&fs_info->balance_running);
2769
2770         if (bargs) {
2771                 memset(bargs, 0, sizeof(*bargs));
2772                 update_ioctl_balance_args(fs_info, 0, bargs);
2773         }
2774
2775         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2776             balance_need_close(fs_info)) {
2777                 __cancel_balance(fs_info);
2778         }
2779
2780         wake_up(&fs_info->balance_wait_q);
2781
2782         return ret;
2783 out:
2784         if (bctl->flags & BTRFS_BALANCE_RESUME)
2785                 __cancel_balance(fs_info);
2786         else
2787                 kfree(bctl);
2788         return ret;
2789 }
2790
2791 static int balance_kthread(void *data)
2792 {
2793         struct btrfs_balance_control *bctl =
2794                         (struct btrfs_balance_control *)data;
2795         struct btrfs_fs_info *fs_info = bctl->fs_info;
2796         int ret = 0;
2797
2798         mutex_lock(&fs_info->volume_mutex);
2799         mutex_lock(&fs_info->balance_mutex);
2800
2801         set_balance_control(bctl);
2802
2803         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2804                 printk(KERN_INFO "btrfs: force skipping balance\n");
2805         } else {
2806                 printk(KERN_INFO "btrfs: continuing balance\n");
2807                 ret = btrfs_balance(bctl, NULL);
2808         }
2809
2810         mutex_unlock(&fs_info->balance_mutex);
2811         mutex_unlock(&fs_info->volume_mutex);
2812         return ret;
2813 }
2814
2815 int btrfs_recover_balance(struct btrfs_root *tree_root)
2816 {
2817         struct task_struct *tsk;
2818         struct btrfs_balance_control *bctl;
2819         struct btrfs_balance_item *item;
2820         struct btrfs_disk_balance_args disk_bargs;
2821         struct btrfs_path *path;
2822         struct extent_buffer *leaf;
2823         struct btrfs_key key;
2824         int ret;
2825
2826         path = btrfs_alloc_path();
2827         if (!path)
2828                 return -ENOMEM;
2829
2830         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2831         if (!bctl) {
2832                 ret = -ENOMEM;
2833                 goto out;
2834         }
2835
2836         key.objectid = BTRFS_BALANCE_OBJECTID;
2837         key.type = BTRFS_BALANCE_ITEM_KEY;
2838         key.offset = 0;
2839
2840         ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2841         if (ret < 0)
2842                 goto out_bctl;
2843         if (ret > 0) { /* ret = -ENOENT; */
2844                 ret = 0;
2845                 goto out_bctl;
2846         }
2847
2848         leaf = path->nodes[0];
2849         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2850
2851         bctl->fs_info = tree_root->fs_info;
2852         bctl->flags = btrfs_balance_flags(leaf, item) | BTRFS_BALANCE_RESUME;
2853
2854         btrfs_balance_data(leaf, item, &disk_bargs);
2855         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2856         btrfs_balance_meta(leaf, item, &disk_bargs);
2857         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2858         btrfs_balance_sys(leaf, item, &disk_bargs);
2859         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2860
2861         tsk = kthread_run(balance_kthread, bctl, "btrfs-balance");
2862         if (IS_ERR(tsk))
2863                 ret = PTR_ERR(tsk);
2864         else
2865                 goto out;
2866
2867 out_bctl:
2868         kfree(bctl);
2869 out:
2870         btrfs_free_path(path);
2871         return ret;
2872 }
2873
2874 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2875 {
2876         int ret = 0;
2877
2878         mutex_lock(&fs_info->balance_mutex);
2879         if (!fs_info->balance_ctl) {
2880                 mutex_unlock(&fs_info->balance_mutex);
2881                 return -ENOTCONN;
2882         }
2883
2884         if (atomic_read(&fs_info->balance_running)) {
2885                 atomic_inc(&fs_info->balance_pause_req);
2886                 mutex_unlock(&fs_info->balance_mutex);
2887
2888                 wait_event(fs_info->balance_wait_q,
2889                            atomic_read(&fs_info->balance_running) == 0);
2890
2891                 mutex_lock(&fs_info->balance_mutex);
2892                 /* we are good with balance_ctl ripped off from under us */
2893                 BUG_ON(atomic_read(&fs_info->balance_running));
2894                 atomic_dec(&fs_info->balance_pause_req);
2895         } else {
2896                 ret = -ENOTCONN;
2897         }
2898
2899         mutex_unlock(&fs_info->balance_mutex);
2900         return ret;
2901 }
2902
2903 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
2904 {
2905         mutex_lock(&fs_info->balance_mutex);
2906         if (!fs_info->balance_ctl) {
2907                 mutex_unlock(&fs_info->balance_mutex);
2908                 return -ENOTCONN;
2909         }
2910
2911         atomic_inc(&fs_info->balance_cancel_req);
2912         /*
2913          * if we are running just wait and return, balance item is
2914          * deleted in btrfs_balance in this case
2915          */
2916         if (atomic_read(&fs_info->balance_running)) {
2917                 mutex_unlock(&fs_info->balance_mutex);
2918                 wait_event(fs_info->balance_wait_q,
2919                            atomic_read(&fs_info->balance_running) == 0);
2920                 mutex_lock(&fs_info->balance_mutex);
2921         } else {
2922                 /* __cancel_balance needs volume_mutex */
2923                 mutex_unlock(&fs_info->balance_mutex);
2924                 mutex_lock(&fs_info->volume_mutex);
2925                 mutex_lock(&fs_info->balance_mutex);
2926
2927                 if (fs_info->balance_ctl)
2928                         __cancel_balance(fs_info);
2929
2930                 mutex_unlock(&fs_info->volume_mutex);
2931         }
2932
2933         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
2934         atomic_dec(&fs_info->balance_cancel_req);
2935         mutex_unlock(&fs_info->balance_mutex);
2936         return 0;
2937 }
2938
2939 /*
2940  * shrinking a device means finding all of the device extents past
2941  * the new size, and then following the back refs to the chunks.
2942  * The chunk relocation code actually frees the device extent
2943  */
2944 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2945 {
2946         struct btrfs_trans_handle *trans;
2947         struct btrfs_root *root = device->dev_root;
2948         struct btrfs_dev_extent *dev_extent = NULL;
2949         struct btrfs_path *path;
2950         u64 length;
2951         u64 chunk_tree;
2952         u64 chunk_objectid;
2953         u64 chunk_offset;
2954         int ret;
2955         int slot;
2956         int failed = 0;
2957         bool retried = false;
2958         struct extent_buffer *l;
2959         struct btrfs_key key;
2960         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2961         u64 old_total = btrfs_super_total_bytes(super_copy);
2962         u64 old_size = device->total_bytes;
2963         u64 diff = device->total_bytes - new_size;
2964
2965         if (new_size >= device->total_bytes)
2966                 return -EINVAL;
2967
2968         path = btrfs_alloc_path();
2969         if (!path)
2970                 return -ENOMEM;
2971
2972         path->reada = 2;
2973
2974         lock_chunks(root);
2975
2976         device->total_bytes = new_size;
2977         if (device->writeable) {
2978                 device->fs_devices->total_rw_bytes -= diff;
2979                 spin_lock(&root->fs_info->free_chunk_lock);
2980                 root->fs_info->free_chunk_space -= diff;
2981                 spin_unlock(&root->fs_info->free_chunk_lock);
2982         }
2983         unlock_chunks(root);
2984
2985 again:
2986         key.objectid = device->devid;
2987         key.offset = (u64)-1;
2988         key.type = BTRFS_DEV_EXTENT_KEY;
2989
2990         while (1) {
2991                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2992                 if (ret < 0)
2993                         goto done;
2994
2995                 ret = btrfs_previous_item(root, path, 0, key.type);
2996                 if (ret < 0)
2997                         goto done;
2998                 if (ret) {
2999                         ret = 0;
3000                         btrfs_release_path(path);
3001                         break;
3002                 }
3003
3004                 l = path->nodes[0];
3005                 slot = path->slots[0];
3006                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3007
3008                 if (key.objectid != device->devid) {
3009                         btrfs_release_path(path);
3010                         break;
3011                 }
3012
3013                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3014                 length = btrfs_dev_extent_length(l, dev_extent);
3015
3016                 if (key.offset + length <= new_size) {
3017                         btrfs_release_path(path);
3018                         break;
3019                 }
3020
3021                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3022                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3023                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3024                 btrfs_release_path(path);
3025
3026                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3027                                            chunk_offset);
3028                 if (ret && ret != -ENOSPC)
3029                         goto done;
3030                 if (ret == -ENOSPC)
3031                         failed++;
3032                 key.offset -= 1;
3033         }
3034
3035         if (failed && !retried) {
3036                 failed = 0;
3037                 retried = true;
3038                 goto again;
3039         } else if (failed && retried) {
3040                 ret = -ENOSPC;
3041                 lock_chunks(root);
3042
3043                 device->total_bytes = old_size;
3044                 if (device->writeable)
3045                         device->fs_devices->total_rw_bytes += diff;
3046                 spin_lock(&root->fs_info->free_chunk_lock);
3047                 root->fs_info->free_chunk_space += diff;
3048                 spin_unlock(&root->fs_info->free_chunk_lock);
3049                 unlock_chunks(root);
3050                 goto done;
3051         }
3052
3053         /* Shrinking succeeded, else we would be at "done". */
3054         trans = btrfs_start_transaction(root, 0);
3055         if (IS_ERR(trans)) {
3056                 ret = PTR_ERR(trans);
3057                 goto done;
3058         }
3059
3060         lock_chunks(root);
3061
3062         device->disk_total_bytes = new_size;
3063         /* Now btrfs_update_device() will change the on-disk size. */
3064         ret = btrfs_update_device(trans, device);
3065         if (ret) {
3066                 unlock_chunks(root);
3067                 btrfs_end_transaction(trans, root);
3068                 goto done;
3069         }
3070         WARN_ON(diff > old_total);
3071         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3072         unlock_chunks(root);
3073         btrfs_end_transaction(trans, root);
3074 done:
3075         btrfs_free_path(path);
3076         return ret;
3077 }
3078
3079 static int btrfs_add_system_chunk(struct btrfs_root *root,
3080                            struct btrfs_key *key,
3081                            struct btrfs_chunk *chunk, int item_size)
3082 {
3083         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3084         struct btrfs_disk_key disk_key;
3085         u32 array_size;
3086         u8 *ptr;
3087
3088         array_size = btrfs_super_sys_array_size(super_copy);
3089         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3090                 return -EFBIG;
3091
3092         ptr = super_copy->sys_chunk_array + array_size;
3093         btrfs_cpu_key_to_disk(&disk_key, key);
3094         memcpy(ptr, &disk_key, sizeof(disk_key));
3095         ptr += sizeof(disk_key);
3096         memcpy(ptr, chunk, item_size);
3097         item_size += sizeof(disk_key);
3098         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3099         return 0;
3100 }
3101
3102 /*
3103  * sort the devices in descending order by max_avail, total_avail
3104  */
3105 static int btrfs_cmp_device_info(const void *a, const void *b)
3106 {
3107         const struct btrfs_device_info *di_a = a;
3108         const struct btrfs_device_info *di_b = b;
3109
3110         if (di_a->max_avail > di_b->max_avail)
3111                 return -1;
3112         if (di_a->max_avail < di_b->max_avail)
3113                 return 1;
3114         if (di_a->total_avail > di_b->total_avail)
3115                 return -1;
3116         if (di_a->total_avail < di_b->total_avail)
3117                 return 1;
3118         return 0;
3119 }
3120
3121 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3122                                struct btrfs_root *extent_root,
3123                                struct map_lookup **map_ret,
3124                                u64 *num_bytes_out, u64 *stripe_size_out,
3125                                u64 start, u64 type)
3126 {
3127         struct btrfs_fs_info *info = extent_root->fs_info;
3128         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3129         struct list_head *cur;
3130         struct map_lookup *map = NULL;
3131         struct extent_map_tree *em_tree;
3132         struct extent_map *em;
3133         struct btrfs_device_info *devices_info = NULL;
3134         u64 total_avail;
3135         int num_stripes;        /* total number of stripes to allocate */
3136         int sub_stripes;        /* sub_stripes info for map */
3137         int dev_stripes;        /* stripes per dev */
3138         int devs_max;           /* max devs to use */
3139         int devs_min;           /* min devs needed */
3140         int devs_increment;     /* ndevs has to be a multiple of this */
3141         int ncopies;            /* how many copies to data has */
3142         int ret;
3143         u64 max_stripe_size;
3144         u64 max_chunk_size;
3145         u64 stripe_size;
3146         u64 num_bytes;
3147         int ndevs;
3148         int i;
3149         int j;
3150
3151         BUG_ON(!alloc_profile_is_valid(type, 0));
3152
3153         if (list_empty(&fs_devices->alloc_list))
3154                 return -ENOSPC;
3155
3156         sub_stripes = 1;
3157         dev_stripes = 1;
3158         devs_increment = 1;
3159         ncopies = 1;
3160         devs_max = 0;   /* 0 == as many as possible */
3161         devs_min = 1;
3162
3163         /*
3164          * define the properties of each RAID type.
3165          * FIXME: move this to a global table and use it in all RAID
3166          * calculation code
3167          */
3168         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3169                 dev_stripes = 2;
3170                 ncopies = 2;
3171                 devs_max = 1;
3172         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3173                 devs_min = 2;
3174         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3175                 devs_increment = 2;
3176                 ncopies = 2;
3177                 devs_max = 2;
3178                 devs_min = 2;
3179         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3180                 sub_stripes = 2;
3181                 devs_increment = 2;
3182                 ncopies = 2;
3183                 devs_min = 4;
3184         } else {
3185                 devs_max = 1;
3186         }
3187
3188         if (type & BTRFS_BLOCK_GROUP_DATA) {
3189                 max_stripe_size = 1024 * 1024 * 1024;
3190                 max_chunk_size = 10 * max_stripe_size;
3191         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3192                 /* for larger filesystems, use larger metadata chunks */
3193                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3194                         max_stripe_size = 1024 * 1024 * 1024;
3195                 else
3196                         max_stripe_size = 256 * 1024 * 1024;
3197                 max_chunk_size = max_stripe_size;
3198         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3199                 max_stripe_size = 32 * 1024 * 1024;
3200                 max_chunk_size = 2 * max_stripe_size;
3201         } else {
3202                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3203                        type);
3204                 BUG_ON(1);
3205         }
3206
3207         /* we don't want a chunk larger than 10% of writeable space */
3208         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3209                              max_chunk_size);
3210
3211         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3212                                GFP_NOFS);
3213         if (!devices_info)
3214                 return -ENOMEM;
3215
3216         cur = fs_devices->alloc_list.next;
3217
3218         /*
3219          * in the first pass through the devices list, we gather information
3220          * about the available holes on each device.
3221          */
3222         ndevs = 0;
3223         while (cur != &fs_devices->alloc_list) {
3224                 struct btrfs_device *device;
3225                 u64 max_avail;
3226                 u64 dev_offset;
3227
3228                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3229
3230                 cur = cur->next;
3231
3232                 if (!device->writeable) {
3233                         printk(KERN_ERR
3234                                "btrfs: read-only device in alloc_list\n");
3235                         WARN_ON(1);
3236                         continue;
3237                 }
3238
3239                 if (!device->in_fs_metadata)
3240                         continue;
3241
3242                 if (device->total_bytes > device->bytes_used)
3243                         total_avail = device->total_bytes - device->bytes_used;
3244                 else
3245                         total_avail = 0;
3246
3247                 /* If there is no space on this device, skip it. */
3248                 if (total_avail == 0)
3249                         continue;
3250
3251                 ret = find_free_dev_extent(device,
3252                                            max_stripe_size * dev_stripes,
3253                                            &dev_offset, &max_avail);
3254                 if (ret && ret != -ENOSPC)
3255                         goto error;
3256
3257                 if (ret == 0)
3258                         max_avail = max_stripe_size * dev_stripes;
3259
3260                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3261                         continue;
3262
3263                 devices_info[ndevs].dev_offset = dev_offset;
3264                 devices_info[ndevs].max_avail = max_avail;
3265                 devices_info[ndevs].total_avail = total_avail;
3266                 devices_info[ndevs].dev = device;
3267                 ++ndevs;
3268         }
3269
3270         /*
3271          * now sort the devices by hole size / available space
3272          */
3273         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3274              btrfs_cmp_device_info, NULL);
3275
3276         /* round down to number of usable stripes */
3277         ndevs -= ndevs % devs_increment;
3278
3279         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3280                 ret = -ENOSPC;
3281                 goto error;
3282         }
3283
3284         if (devs_max && ndevs > devs_max)
3285                 ndevs = devs_max;
3286         /*
3287          * the primary goal is to maximize the number of stripes, so use as many
3288          * devices as possible, even if the stripes are not maximum sized.
3289          */
3290         stripe_size = devices_info[ndevs-1].max_avail;
3291         num_stripes = ndevs * dev_stripes;
3292
3293         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
3294                 stripe_size = max_chunk_size * ncopies;
3295                 do_div(stripe_size, num_stripes);
3296         }
3297
3298         do_div(stripe_size, dev_stripes);
3299         do_div(stripe_size, BTRFS_STRIPE_LEN);
3300         stripe_size *= BTRFS_STRIPE_LEN;
3301
3302         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3303         if (!map) {
3304                 ret = -ENOMEM;
3305                 goto error;
3306         }
3307         map->num_stripes = num_stripes;
3308
3309         for (i = 0; i < ndevs; ++i) {
3310                 for (j = 0; j < dev_stripes; ++j) {
3311                         int s = i * dev_stripes + j;
3312                         map->stripes[s].dev = devices_info[i].dev;
3313                         map->stripes[s].physical = devices_info[i].dev_offset +
3314                                                    j * stripe_size;
3315                 }
3316         }
3317         map->sector_size = extent_root->sectorsize;
3318         map->stripe_len = BTRFS_STRIPE_LEN;
3319         map->io_align = BTRFS_STRIPE_LEN;
3320         map->io_width = BTRFS_STRIPE_LEN;
3321         map->type = type;
3322         map->sub_stripes = sub_stripes;
3323
3324         *map_ret = map;
3325         num_bytes = stripe_size * (num_stripes / ncopies);
3326
3327         *stripe_size_out = stripe_size;
3328         *num_bytes_out = num_bytes;
3329
3330         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3331
3332         em = alloc_extent_map();
3333         if (!em) {
3334                 ret = -ENOMEM;
3335                 goto error;
3336         }
3337         em->bdev = (struct block_device *)map;
3338         em->start = start;
3339         em->len = num_bytes;
3340         em->block_start = 0;
3341         em->block_len = em->len;
3342
3343         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3344         write_lock(&em_tree->lock);
3345         ret = add_extent_mapping(em_tree, em);
3346         write_unlock(&em_tree->lock);
3347         BUG_ON(ret);
3348         free_extent_map(em);
3349
3350         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3351                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3352                                      start, num_bytes);
3353         BUG_ON(ret);
3354
3355         for (i = 0; i < map->num_stripes; ++i) {
3356                 struct btrfs_device *device;
3357                 u64 dev_offset;
3358
3359                 device = map->stripes[i].dev;
3360                 dev_offset = map->stripes[i].physical;
3361
3362                 ret = btrfs_alloc_dev_extent(trans, device,
3363                                 info->chunk_root->root_key.objectid,
3364                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3365                                 start, dev_offset, stripe_size);
3366                 BUG_ON(ret);
3367         }
3368
3369         kfree(devices_info);
3370         return 0;
3371
3372 error:
3373         kfree(map);
3374         kfree(devices_info);
3375         return ret;
3376 }
3377
3378 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3379                                 struct btrfs_root *extent_root,
3380                                 struct map_lookup *map, u64 chunk_offset,
3381                                 u64 chunk_size, u64 stripe_size)
3382 {
3383         u64 dev_offset;
3384         struct btrfs_key key;
3385         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3386         struct btrfs_device *device;
3387         struct btrfs_chunk *chunk;
3388         struct btrfs_stripe *stripe;
3389         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3390         int index = 0;
3391         int ret;
3392
3393         chunk = kzalloc(item_size, GFP_NOFS);
3394         if (!chunk)
3395                 return -ENOMEM;
3396
3397         index = 0;
3398         while (index < map->num_stripes) {
3399                 device = map->stripes[index].dev;
3400                 device->bytes_used += stripe_size;
3401                 ret = btrfs_update_device(trans, device);
3402                 BUG_ON(ret);
3403                 index++;
3404         }
3405
3406         spin_lock(&extent_root->fs_info->free_chunk_lock);
3407         extent_root->fs_info->free_chunk_space -= (stripe_size *
3408                                                    map->num_stripes);
3409         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3410
3411         index = 0;
3412         stripe = &chunk->stripe;
3413         while (index < map->num_stripes) {
3414                 device = map->stripes[index].dev;
3415                 dev_offset = map->stripes[index].physical;
3416
3417                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3418                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3419                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3420                 stripe++;
3421                 index++;
3422         }
3423
3424         btrfs_set_stack_chunk_length(chunk, chunk_size);
3425         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3426         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3427         btrfs_set_stack_chunk_type(chunk, map->type);
3428         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3429         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3430         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3431         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3432         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3433
3434         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3435         key.type = BTRFS_CHUNK_ITEM_KEY;
3436         key.offset = chunk_offset;
3437
3438         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3439         BUG_ON(ret);
3440
3441         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3442                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3443                                              item_size);
3444                 BUG_ON(ret);
3445         }
3446
3447         kfree(chunk);
3448         return 0;
3449 }
3450
3451 /*
3452  * Chunk allocation falls into two parts. The first part does works
3453  * that make the new allocated chunk useable, but not do any operation
3454  * that modifies the chunk tree. The second part does the works that
3455  * require modifying the chunk tree. This division is important for the
3456  * bootstrap process of adding storage to a seed btrfs.
3457  */
3458 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3459                       struct btrfs_root *extent_root, u64 type)
3460 {
3461         u64 chunk_offset;
3462         u64 chunk_size;
3463         u64 stripe_size;
3464         struct map_lookup *map;
3465         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3466         int ret;
3467
3468         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3469                               &chunk_offset);
3470         if (ret)
3471                 return ret;
3472
3473         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3474                                   &stripe_size, chunk_offset, type);
3475         if (ret)
3476                 return ret;
3477
3478         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3479                                    chunk_size, stripe_size);
3480         BUG_ON(ret);
3481         return 0;
3482 }
3483
3484 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3485                                          struct btrfs_root *root,
3486                                          struct btrfs_device *device)
3487 {
3488         u64 chunk_offset;
3489         u64 sys_chunk_offset;
3490         u64 chunk_size;
3491         u64 sys_chunk_size;
3492         u64 stripe_size;
3493         u64 sys_stripe_size;
3494         u64 alloc_profile;
3495         struct map_lookup *map;
3496         struct map_lookup *sys_map;
3497         struct btrfs_fs_info *fs_info = root->fs_info;
3498         struct btrfs_root *extent_root = fs_info->extent_root;
3499         int ret;
3500
3501         ret = find_next_chunk(fs_info->chunk_root,
3502                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3503         if (ret)
3504                 return ret;
3505
3506         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3507                                 fs_info->avail_metadata_alloc_bits;
3508         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3509
3510         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3511                                   &stripe_size, chunk_offset, alloc_profile);
3512         BUG_ON(ret);
3513
3514         sys_chunk_offset = chunk_offset + chunk_size;
3515
3516         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3517                                 fs_info->avail_system_alloc_bits;
3518         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3519
3520         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3521                                   &sys_chunk_size, &sys_stripe_size,
3522                                   sys_chunk_offset, alloc_profile);
3523         BUG_ON(ret);
3524
3525         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3526         BUG_ON(ret);
3527
3528         /*
3529          * Modifying chunk tree needs allocating new blocks from both
3530          * system block group and metadata block group. So we only can
3531          * do operations require modifying the chunk tree after both
3532          * block groups were created.
3533          */
3534         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3535                                    chunk_size, stripe_size);
3536         BUG_ON(ret);
3537
3538         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3539                                    sys_chunk_offset, sys_chunk_size,
3540                                    sys_stripe_size);
3541         BUG_ON(ret);
3542         return 0;
3543 }
3544
3545 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3546 {
3547         struct extent_map *em;
3548         struct map_lookup *map;
3549         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3550         int readonly = 0;
3551         int i;
3552
3553         read_lock(&map_tree->map_tree.lock);
3554         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3555         read_unlock(&map_tree->map_tree.lock);
3556         if (!em)
3557                 return 1;
3558
3559         if (btrfs_test_opt(root, DEGRADED)) {
3560                 free_extent_map(em);
3561                 return 0;
3562         }
3563
3564         map = (struct map_lookup *)em->bdev;
3565         for (i = 0; i < map->num_stripes; i++) {
3566                 if (!map->stripes[i].dev->writeable) {
3567                         readonly = 1;
3568                         break;
3569                 }
3570         }
3571         free_extent_map(em);
3572         return readonly;
3573 }
3574
3575 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3576 {
3577         extent_map_tree_init(&tree->map_tree);
3578 }
3579
3580 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3581 {
3582         struct extent_map *em;
3583
3584         while (1) {
3585                 write_lock(&tree->map_tree.lock);
3586                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3587                 if (em)
3588                         remove_extent_mapping(&tree->map_tree, em);
3589                 write_unlock(&tree->map_tree.lock);
3590                 if (!em)
3591                         break;
3592                 kfree(em->bdev);
3593                 /* once for us */
3594                 free_extent_map(em);
3595                 /* once for the tree */
3596                 free_extent_map(em);
3597         }
3598 }
3599
3600 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3601 {
3602         struct extent_map *em;
3603         struct map_lookup *map;
3604         struct extent_map_tree *em_tree = &map_tree->map_tree;
3605         int ret;
3606
3607         read_lock(&em_tree->lock);
3608         em = lookup_extent_mapping(em_tree, logical, len);
3609         read_unlock(&em_tree->lock);
3610         BUG_ON(!em);
3611
3612         BUG_ON(em->start > logical || em->start + em->len < logical);
3613         map = (struct map_lookup *)em->bdev;
3614         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3615                 ret = map->num_stripes;
3616         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3617                 ret = map->sub_stripes;
3618         else
3619                 ret = 1;
3620         free_extent_map(em);
3621         return ret;
3622 }
3623
3624 static int find_live_mirror(struct map_lookup *map, int first, int num,
3625                             int optimal)
3626 {
3627         int i;
3628         if (map->stripes[optimal].dev->bdev)
3629                 return optimal;
3630         for (i = first; i < first + num; i++) {
3631                 if (map->stripes[i].dev->bdev)
3632                         return i;
3633         }
3634         /* we couldn't find one that doesn't fail.  Just return something
3635          * and the io error handling code will clean up eventually
3636          */
3637         return optimal;
3638 }
3639
3640 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3641                              u64 logical, u64 *length,
3642                              struct btrfs_bio **bbio_ret,
3643                              int mirror_num)
3644 {
3645         struct extent_map *em;
3646         struct map_lookup *map;
3647         struct extent_map_tree *em_tree = &map_tree->map_tree;
3648         u64 offset;
3649         u64 stripe_offset;
3650         u64 stripe_end_offset;
3651         u64 stripe_nr;
3652         u64 stripe_nr_orig;
3653         u64 stripe_nr_end;
3654         int stripe_index;
3655         int i;
3656         int ret = 0;
3657         int num_stripes;
3658         int max_errors = 0;
3659         struct btrfs_bio *bbio = NULL;
3660
3661         read_lock(&em_tree->lock);
3662         em = lookup_extent_mapping(em_tree, logical, *length);
3663         read_unlock(&em_tree->lock);
3664
3665         if (!em) {
3666                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3667                        (unsigned long long)logical,
3668                        (unsigned long long)*length);
3669                 BUG();
3670         }
3671
3672         BUG_ON(em->start > logical || em->start + em->len < logical);
3673         map = (struct map_lookup *)em->bdev;
3674         offset = logical - em->start;
3675
3676         if (mirror_num > map->num_stripes)
3677                 mirror_num = 0;
3678
3679         stripe_nr = offset;
3680         /*
3681          * stripe_nr counts the total number of stripes we have to stride
3682          * to get to this block
3683          */
3684         do_div(stripe_nr, map->stripe_len);
3685
3686         stripe_offset = stripe_nr * map->stripe_len;
3687         BUG_ON(offset < stripe_offset);
3688
3689         /* stripe_offset is the offset of this block in its stripe*/
3690         stripe_offset = offset - stripe_offset;
3691
3692         if (rw & REQ_DISCARD)
3693                 *length = min_t(u64, em->len - offset, *length);
3694         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3695                 /* we limit the length of each bio to what fits in a stripe */
3696                 *length = min_t(u64, em->len - offset,
3697                                 map->stripe_len - stripe_offset);
3698         } else {
3699                 *length = em->len - offset;
3700         }
3701
3702         if (!bbio_ret)
3703                 goto out;
3704
3705         num_stripes = 1;
3706         stripe_index = 0;
3707         stripe_nr_orig = stripe_nr;
3708         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3709                         (~(map->stripe_len - 1));
3710         do_div(stripe_nr_end, map->stripe_len);
3711         stripe_end_offset = stripe_nr_end * map->stripe_len -
3712                             (offset + *length);
3713         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3714                 if (rw & REQ_DISCARD)
3715                         num_stripes = min_t(u64, map->num_stripes,
3716                                             stripe_nr_end - stripe_nr_orig);
3717                 stripe_index = do_div(stripe_nr, map->num_stripes);
3718         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3719                 if (rw & (REQ_WRITE | REQ_DISCARD))
3720                         num_stripes = map->num_stripes;
3721                 else if (mirror_num)
3722                         stripe_index = mirror_num - 1;
3723                 else {
3724                         stripe_index = find_live_mirror(map, 0,
3725                                             map->num_stripes,
3726                                             current->pid % map->num_stripes);
3727                         mirror_num = stripe_index + 1;
3728                 }
3729
3730         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3731                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3732                         num_stripes = map->num_stripes;
3733                 } else if (mirror_num) {
3734                         stripe_index = mirror_num - 1;
3735                 } else {
3736                         mirror_num = 1;
3737                 }
3738
3739         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3740                 int factor = map->num_stripes / map->sub_stripes;
3741
3742                 stripe_index = do_div(stripe_nr, factor);
3743                 stripe_index *= map->sub_stripes;
3744
3745                 if (rw & REQ_WRITE)
3746                         num_stripes = map->sub_stripes;
3747                 else if (rw & REQ_DISCARD)
3748                         num_stripes = min_t(u64, map->sub_stripes *
3749                                             (stripe_nr_end - stripe_nr_orig),
3750                                             map->num_stripes);
3751                 else if (mirror_num)
3752                         stripe_index += mirror_num - 1;
3753                 else {
3754                         stripe_index = find_live_mirror(map, stripe_index,
3755                                               map->sub_stripes, stripe_index +
3756                                               current->pid % map->sub_stripes);
3757                         mirror_num = stripe_index + 1;
3758                 }
3759         } else {
3760                 /*
3761                  * after this do_div call, stripe_nr is the number of stripes
3762                  * on this device we have to walk to find the data, and
3763                  * stripe_index is the number of our device in the stripe array
3764                  */
3765                 stripe_index = do_div(stripe_nr, map->num_stripes);
3766                 mirror_num = stripe_index + 1;
3767         }
3768         BUG_ON(stripe_index >= map->num_stripes);
3769
3770         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3771         if (!bbio) {
3772                 ret = -ENOMEM;
3773                 goto out;
3774         }
3775         atomic_set(&bbio->error, 0);
3776
3777         if (rw & REQ_DISCARD) {
3778                 int factor = 0;
3779                 int sub_stripes = 0;
3780                 u64 stripes_per_dev = 0;
3781                 u32 remaining_stripes = 0;
3782
3783                 if (map->type &
3784                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3785                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3786                                 sub_stripes = 1;
3787                         else
3788                                 sub_stripes = map->sub_stripes;
3789
3790                         factor = map->num_stripes / sub_stripes;
3791                         stripes_per_dev = div_u64_rem(stripe_nr_end -
3792                                                       stripe_nr_orig,
3793                                                       factor,
3794                                                       &remaining_stripes);
3795                 }
3796
3797                 for (i = 0; i < num_stripes; i++) {
3798                         bbio->stripes[i].physical =
3799                                 map->stripes[stripe_index].physical +
3800                                 stripe_offset + stripe_nr * map->stripe_len;
3801                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3802
3803                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3804                                          BTRFS_BLOCK_GROUP_RAID10)) {
3805                                 bbio->stripes[i].length = stripes_per_dev *
3806                                                           map->stripe_len;
3807                                 if (i / sub_stripes < remaining_stripes)
3808                                         bbio->stripes[i].length +=
3809                                                 map->stripe_len;
3810                                 if (i < sub_stripes)
3811                                         bbio->stripes[i].length -=
3812                                                 stripe_offset;
3813                                 if ((i / sub_stripes + 1) %
3814                                     sub_stripes == remaining_stripes)
3815                                         bbio->stripes[i].length -=
3816                                                 stripe_end_offset;
3817                                 if (i == sub_stripes - 1)
3818                                         stripe_offset = 0;
3819                         } else
3820                                 bbio->stripes[i].length = *length;
3821
3822                         stripe_index++;
3823                         if (stripe_index == map->num_stripes) {
3824                                 /* This could only happen for RAID0/10 */
3825                                 stripe_index = 0;
3826                                 stripe_nr++;
3827                         }
3828                 }
3829         } else {
3830                 for (i = 0; i < num_stripes; i++) {
3831                         bbio->stripes[i].physical =
3832                                 map->stripes[stripe_index].physical +
3833                                 stripe_offset +
3834                                 stripe_nr * map->stripe_len;
3835                         bbio->stripes[i].dev =
3836                                 map->stripes[stripe_index].dev;
3837                         stripe_index++;
3838                 }
3839         }
3840
3841         if (rw & REQ_WRITE) {
3842                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3843                                  BTRFS_BLOCK_GROUP_RAID10 |
3844                                  BTRFS_BLOCK_GROUP_DUP)) {
3845                         max_errors = 1;
3846                 }
3847         }
3848
3849         *bbio_ret = bbio;
3850         bbio->num_stripes = num_stripes;
3851         bbio->max_errors = max_errors;
3852         bbio->mirror_num = mirror_num;
3853 out:
3854         free_extent_map(em);
3855         return ret;
3856 }
3857
3858 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3859                       u64 logical, u64 *length,
3860                       struct btrfs_bio **bbio_ret, int mirror_num)
3861 {
3862         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3863                                  mirror_num);
3864 }
3865
3866 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3867                      u64 chunk_start, u64 physical, u64 devid,
3868                      u64 **logical, int *naddrs, int *stripe_len)
3869 {
3870         struct extent_map_tree *em_tree = &map_tree->map_tree;
3871         struct extent_map *em;
3872         struct map_lookup *map;
3873         u64 *buf;
3874         u64 bytenr;
3875         u64 length;
3876         u64 stripe_nr;
3877         int i, j, nr = 0;
3878
3879         read_lock(&em_tree->lock);
3880         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3881         read_unlock(&em_tree->lock);
3882
3883         BUG_ON(!em || em->start != chunk_start);
3884         map = (struct map_lookup *)em->bdev;
3885
3886         length = em->len;
3887         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3888                 do_div(length, map->num_stripes / map->sub_stripes);
3889         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3890                 do_div(length, map->num_stripes);
3891
3892         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3893         BUG_ON(!buf);
3894
3895         for (i = 0; i < map->num_stripes; i++) {
3896                 if (devid && map->stripes[i].dev->devid != devid)
3897                         continue;
3898                 if (map->stripes[i].physical > physical ||
3899                     map->stripes[i].physical + length <= physical)
3900                         continue;
3901
3902                 stripe_nr = physical - map->stripes[i].physical;
3903                 do_div(stripe_nr, map->stripe_len);
3904
3905                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3906                         stripe_nr = stripe_nr * map->num_stripes + i;
3907                         do_div(stripe_nr, map->sub_stripes);
3908                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3909                         stripe_nr = stripe_nr * map->num_stripes + i;
3910                 }
3911                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3912                 WARN_ON(nr >= map->num_stripes);
3913                 for (j = 0; j < nr; j++) {
3914                         if (buf[j] == bytenr)
3915                                 break;
3916                 }
3917                 if (j == nr) {
3918                         WARN_ON(nr >= map->num_stripes);
3919                         buf[nr++] = bytenr;
3920                 }
3921         }
3922
3923         *logical = buf;
3924         *naddrs = nr;
3925         *stripe_len = map->stripe_len;
3926
3927         free_extent_map(em);
3928         return 0;
3929 }
3930
3931 static void btrfs_end_bio(struct bio *bio, int err)
3932 {
3933         struct btrfs_bio *bbio = bio->bi_private;
3934         int is_orig_bio = 0;
3935
3936         if (err)
3937                 atomic_inc(&bbio->error);
3938
3939         if (bio == bbio->orig_bio)
3940                 is_orig_bio = 1;
3941
3942         if (atomic_dec_and_test(&bbio->stripes_pending)) {
3943                 if (!is_orig_bio) {
3944                         bio_put(bio);
3945                         bio = bbio->orig_bio;
3946                 }
3947                 bio->bi_private = bbio->private;
3948                 bio->bi_end_io = bbio->end_io;
3949                 bio->bi_bdev = (struct block_device *)
3950                                         (unsigned long)bbio->mirror_num;
3951                 /* only send an error to the higher layers if it is
3952                  * beyond the tolerance of the multi-bio
3953                  */
3954                 if (atomic_read(&bbio->error) > bbio->max_errors) {
3955                         err = -EIO;
3956                 } else {
3957                         /*
3958                          * this bio is actually up to date, we didn't
3959                          * go over the max number of errors
3960                          */
3961                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3962                         err = 0;
3963                 }
3964                 kfree(bbio);
3965
3966                 bio_endio(bio, err);
3967         } else if (!is_orig_bio) {
3968                 bio_put(bio);
3969         }
3970 }
3971
3972 struct async_sched {
3973         struct bio *bio;
3974         int rw;
3975         struct btrfs_fs_info *info;
3976         struct btrfs_work work;
3977 };
3978
3979 /*
3980  * see run_scheduled_bios for a description of why bios are collected for
3981  * async submit.
3982  *
3983  * This will add one bio to the pending list for a device and make sure
3984  * the work struct is scheduled.
3985  */
3986 static noinline int schedule_bio(struct btrfs_root *root,
3987                                  struct btrfs_device *device,
3988                                  int rw, struct bio *bio)
3989 {
3990         int should_queue = 1;
3991         struct btrfs_pending_bios *pending_bios;
3992
3993         /* don't bother with additional async steps for reads, right now */
3994         if (!(rw & REQ_WRITE)) {
3995                 bio_get(bio);
3996                 btrfsic_submit_bio(rw, bio);
3997                 bio_put(bio);
3998                 return 0;
3999         }
4000
4001         /*
4002          * nr_async_bios allows us to reliably return congestion to the
4003          * higher layers.  Otherwise, the async bio makes it appear we have
4004          * made progress against dirty pages when we've really just put it
4005          * on a queue for later
4006          */
4007         atomic_inc(&root->fs_info->nr_async_bios);
4008         WARN_ON(bio->bi_next);
4009         bio->bi_next = NULL;
4010         bio->bi_rw |= rw;
4011
4012         spin_lock(&device->io_lock);
4013         if (bio->bi_rw & REQ_SYNC)
4014                 pending_bios = &device->pending_sync_bios;
4015         else
4016                 pending_bios = &device->pending_bios;
4017
4018         if (pending_bios->tail)
4019                 pending_bios->tail->bi_next = bio;
4020
4021         pending_bios->tail = bio;
4022         if (!pending_bios->head)
4023                 pending_bios->head = bio;
4024         if (device->running_pending)
4025                 should_queue = 0;
4026
4027         spin_unlock(&device->io_lock);
4028
4029         if (should_queue)
4030                 btrfs_queue_worker(&root->fs_info->submit_workers,
4031                                    &device->work);
4032         return 0;
4033 }
4034
4035 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4036                   int mirror_num, int async_submit)
4037 {
4038         struct btrfs_mapping_tree *map_tree;
4039         struct btrfs_device *dev;
4040         struct bio *first_bio = bio;
4041         u64 logical = (u64)bio->bi_sector << 9;
4042         u64 length = 0;
4043         u64 map_length;
4044         int ret;
4045         int dev_nr = 0;
4046         int total_devs = 1;
4047         struct btrfs_bio *bbio = NULL;
4048
4049         length = bio->bi_size;
4050         map_tree = &root->fs_info->mapping_tree;
4051         map_length = length;
4052
4053         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4054                               mirror_num);
4055         BUG_ON(ret);
4056
4057         total_devs = bbio->num_stripes;
4058         if (map_length < length) {
4059                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
4060                        "len %llu\n", (unsigned long long)logical,
4061                        (unsigned long long)length,
4062                        (unsigned long long)map_length);
4063                 BUG();
4064         }
4065
4066         bbio->orig_bio = first_bio;
4067         bbio->private = first_bio->bi_private;
4068         bbio->end_io = first_bio->bi_end_io;
4069         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4070
4071         while (dev_nr < total_devs) {
4072                 if (dev_nr < total_devs - 1) {
4073                         bio = bio_clone(first_bio, GFP_NOFS);
4074                         BUG_ON(!bio);
4075                 } else {
4076                         bio = first_bio;
4077                 }
4078                 bio->bi_private = bbio;
4079                 bio->bi_end_io = btrfs_end_bio;
4080                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4081                 dev = bbio->stripes[dev_nr].dev;
4082                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4083                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4084                                  "(%s id %llu), size=%u\n", rw,
4085                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4086                                  dev->name, dev->devid, bio->bi_size);
4087                         bio->bi_bdev = dev->bdev;
4088                         if (async_submit)
4089                                 schedule_bio(root, dev, rw, bio);
4090                         else
4091                                 btrfsic_submit_bio(rw, bio);
4092                 } else {
4093                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4094                         bio->bi_sector = logical >> 9;
4095                         bio_endio(bio, -EIO);
4096                 }
4097                 dev_nr++;
4098         }
4099         return 0;
4100 }
4101
4102 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4103                                        u8 *uuid, u8 *fsid)
4104 {
4105         struct btrfs_device *device;
4106         struct btrfs_fs_devices *cur_devices;
4107
4108         cur_devices = root->fs_info->fs_devices;
4109         while (cur_devices) {
4110                 if (!fsid ||
4111                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4112                         device = __find_device(&cur_devices->devices,
4113                                                devid, uuid);
4114                         if (device)
4115                                 return device;
4116                 }
4117                 cur_devices = cur_devices->seed;
4118         }
4119         return NULL;
4120 }
4121
4122 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4123                                             u64 devid, u8 *dev_uuid)
4124 {
4125         struct btrfs_device *device;
4126         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4127
4128         device = kzalloc(sizeof(*device), GFP_NOFS);
4129         if (!device)
4130                 return NULL;
4131         list_add(&device->dev_list,
4132                  &fs_devices->devices);
4133         device->dev_root = root->fs_info->dev_root;
4134         device->devid = devid;
4135         device->work.func = pending_bios_fn;
4136         device->fs_devices = fs_devices;
4137         device->missing = 1;
4138         fs_devices->num_devices++;
4139         fs_devices->missing_devices++;
4140         spin_lock_init(&device->io_lock);
4141         INIT_LIST_HEAD(&device->dev_alloc_list);
4142         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4143         return device;
4144 }
4145
4146 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4147                           struct extent_buffer *leaf,
4148                           struct btrfs_chunk *chunk)
4149 {
4150         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4151         struct map_lookup *map;
4152         struct extent_map *em;
4153         u64 logical;
4154         u64 length;
4155         u64 devid;
4156         u8 uuid[BTRFS_UUID_SIZE];
4157         int num_stripes;
4158         int ret;
4159         int i;
4160
4161         logical = key->offset;
4162         length = btrfs_chunk_length(leaf, chunk);
4163
4164         read_lock(&map_tree->map_tree.lock);
4165         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4166         read_unlock(&map_tree->map_tree.lock);
4167
4168         /* already mapped? */
4169         if (em && em->start <= logical && em->start + em->len > logical) {
4170                 free_extent_map(em);
4171                 return 0;
4172         } else if (em) {
4173                 free_extent_map(em);
4174         }
4175
4176         em = alloc_extent_map();
4177         if (!em)
4178                 return -ENOMEM;
4179         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4180         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4181         if (!map) {
4182                 free_extent_map(em);
4183                 return -ENOMEM;
4184         }
4185
4186         em->bdev = (struct block_device *)map;
4187         em->start = logical;
4188         em->len = length;
4189         em->block_start = 0;
4190         em->block_len = em->len;
4191
4192         map->num_stripes = num_stripes;
4193         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4194         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4195         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4196         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4197         map->type = btrfs_chunk_type(leaf, chunk);
4198         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4199         for (i = 0; i < num_stripes; i++) {
4200                 map->stripes[i].physical =
4201                         btrfs_stripe_offset_nr(leaf, chunk, i);
4202                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4203                 read_extent_buffer(leaf, uuid, (unsigned long)
4204                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4205                                    BTRFS_UUID_SIZE);
4206                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4207                                                         NULL);
4208                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4209                         kfree(map);
4210                         free_extent_map(em);
4211                         return -EIO;
4212                 }
4213                 if (!map->stripes[i].dev) {
4214                         map->stripes[i].dev =
4215                                 add_missing_dev(root, devid, uuid);
4216                         if (!map->stripes[i].dev) {
4217                                 kfree(map);
4218                                 free_extent_map(em);
4219                                 return -EIO;
4220                         }
4221                 }
4222                 map->stripes[i].dev->in_fs_metadata = 1;
4223         }
4224
4225         write_lock(&map_tree->map_tree.lock);
4226         ret = add_extent_mapping(&map_tree->map_tree, em);
4227         write_unlock(&map_tree->map_tree.lock);
4228         BUG_ON(ret);
4229         free_extent_map(em);
4230
4231         return 0;
4232 }
4233
4234 static int fill_device_from_item(struct extent_buffer *leaf,
4235                                  struct btrfs_dev_item *dev_item,
4236                                  struct btrfs_device *device)
4237 {
4238         unsigned long ptr;
4239
4240         device->devid = btrfs_device_id(leaf, dev_item);
4241         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4242         device->total_bytes = device->disk_total_bytes;
4243         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4244         device->type = btrfs_device_type(leaf, dev_item);
4245         device->io_align = btrfs_device_io_align(leaf, dev_item);
4246         device->io_width = btrfs_device_io_width(leaf, dev_item);
4247         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4248
4249         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4250         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4251
4252         return 0;
4253 }
4254
4255 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4256 {
4257         struct btrfs_fs_devices *fs_devices;
4258         int ret;
4259
4260         BUG_ON(!mutex_is_locked(&uuid_mutex));
4261
4262         fs_devices = root->fs_info->fs_devices->seed;
4263         while (fs_devices) {
4264                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4265                         ret = 0;
4266                         goto out;
4267                 }
4268                 fs_devices = fs_devices->seed;
4269         }
4270
4271         fs_devices = find_fsid(fsid);
4272         if (!fs_devices) {
4273                 ret = -ENOENT;
4274                 goto out;
4275         }
4276
4277         fs_devices = clone_fs_devices(fs_devices);
4278         if (IS_ERR(fs_devices)) {
4279                 ret = PTR_ERR(fs_devices);
4280                 goto out;
4281         }
4282
4283         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4284                                    root->fs_info->bdev_holder);
4285         if (ret)
4286                 goto out;
4287
4288         if (!fs_devices->seeding) {
4289                 __btrfs_close_devices(fs_devices);
4290                 free_fs_devices(fs_devices);
4291                 ret = -EINVAL;
4292                 goto out;
4293         }
4294
4295         fs_devices->seed = root->fs_info->fs_devices->seed;
4296         root->fs_info->fs_devices->seed = fs_devices;
4297 out:
4298         return ret;
4299 }
4300
4301 static int read_one_dev(struct btrfs_root *root,
4302                         struct extent_buffer *leaf,
4303                         struct btrfs_dev_item *dev_item)
4304 {
4305         struct btrfs_device *device;
4306         u64 devid;
4307         int ret;
4308         u8 fs_uuid[BTRFS_UUID_SIZE];
4309         u8 dev_uuid[BTRFS_UUID_SIZE];
4310
4311         devid = btrfs_device_id(leaf, dev_item);
4312         read_extent_buffer(leaf, dev_uuid,
4313                            (unsigned long)btrfs_device_uuid(dev_item),
4314                            BTRFS_UUID_SIZE);
4315         read_extent_buffer(leaf, fs_uuid,
4316                            (unsigned long)btrfs_device_fsid(dev_item),
4317                            BTRFS_UUID_SIZE);
4318
4319         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4320                 ret = open_seed_devices(root, fs_uuid);
4321                 if (ret && !btrfs_test_opt(root, DEGRADED))
4322                         return ret;
4323         }
4324
4325         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4326         if (!device || !device->bdev) {
4327                 if (!btrfs_test_opt(root, DEGRADED))
4328                         return -EIO;
4329
4330                 if (!device) {
4331                         printk(KERN_WARNING "warning devid %llu missing\n",
4332                                (unsigned long long)devid);
4333                         device = add_missing_dev(root, devid, dev_uuid);
4334                         if (!device)
4335                                 return -ENOMEM;
4336                 } else if (!device->missing) {
4337                         /*
4338                          * this happens when a device that was properly setup
4339                          * in the device info lists suddenly goes bad.
4340                          * device->bdev is NULL, and so we have to set
4341                          * device->missing to one here
4342                          */
4343                         root->fs_info->fs_devices->missing_devices++;
4344                         device->missing = 1;
4345                 }
4346         }
4347
4348         if (device->fs_devices != root->fs_info->fs_devices) {
4349                 BUG_ON(device->writeable);
4350                 if (device->generation !=
4351                     btrfs_device_generation(leaf, dev_item))
4352                         return -EINVAL;
4353         }
4354
4355         fill_device_from_item(leaf, dev_item, device);
4356         device->dev_root = root->fs_info->dev_root;
4357         device->in_fs_metadata = 1;
4358         if (device->writeable) {
4359                 device->fs_devices->total_rw_bytes += device->total_bytes;
4360                 spin_lock(&root->fs_info->free_chunk_lock);
4361                 root->fs_info->free_chunk_space += device->total_bytes -
4362                         device->bytes_used;
4363                 spin_unlock(&root->fs_info->free_chunk_lock);
4364         }
4365         ret = 0;
4366         return ret;
4367 }
4368
4369 int btrfs_read_sys_array(struct btrfs_root *root)
4370 {
4371         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4372         struct extent_buffer *sb;
4373         struct btrfs_disk_key *disk_key;
4374         struct btrfs_chunk *chunk;
4375         u8 *ptr;
4376         unsigned long sb_ptr;
4377         int ret = 0;
4378         u32 num_stripes;
4379         u32 array_size;
4380         u32 len = 0;
4381         u32 cur;
4382         struct btrfs_key key;
4383
4384         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4385                                           BTRFS_SUPER_INFO_SIZE);
4386         if (!sb)
4387                 return -ENOMEM;
4388         btrfs_set_buffer_uptodate(sb);
4389         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4390         /*
4391          * The sb extent buffer is artifical and just used to read the system array.
4392          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4393          * pages up-to-date when the page is larger: extent does not cover the
4394          * whole page and consequently check_page_uptodate does not find all
4395          * the page's extents up-to-date (the hole beyond sb),
4396          * write_extent_buffer then triggers a WARN_ON.
4397          *
4398          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4399          * but sb spans only this function. Add an explicit SetPageUptodate call
4400          * to silence the warning eg. on PowerPC 64.
4401          */
4402         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4403                 SetPageUptodate(sb->pages[0]);
4404
4405         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4406         array_size = btrfs_super_sys_array_size(super_copy);
4407
4408         ptr = super_copy->sys_chunk_array;
4409         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4410         cur = 0;
4411
4412         while (cur < array_size) {
4413                 disk_key = (struct btrfs_disk_key *)ptr;
4414                 btrfs_disk_key_to_cpu(&key, disk_key);
4415
4416                 len = sizeof(*disk_key); ptr += len;
4417                 sb_ptr += len;
4418                 cur += len;
4419
4420                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4421                         chunk = (struct btrfs_chunk *)sb_ptr;
4422                         ret = read_one_chunk(root, &key, sb, chunk);
4423                         if (ret)
4424                                 break;
4425                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4426                         len = btrfs_chunk_item_size(num_stripes);
4427                 } else {
4428                         ret = -EIO;
4429                         break;
4430                 }
4431                 ptr += len;
4432                 sb_ptr += len;
4433                 cur += len;
4434         }
4435         free_extent_buffer(sb);
4436         return ret;
4437 }
4438
4439 int btrfs_read_chunk_tree(struct btrfs_root *root)
4440 {
4441         struct btrfs_path *path;
4442         struct extent_buffer *leaf;
4443         struct btrfs_key key;
4444         struct btrfs_key found_key;
4445         int ret;
4446         int slot;
4447
4448         root = root->fs_info->chunk_root;
4449
4450         path = btrfs_alloc_path();
4451         if (!path)
4452                 return -ENOMEM;
4453
4454         mutex_lock(&uuid_mutex);
4455         lock_chunks(root);
4456
4457         /* first we search for all of the device items, and then we
4458          * read in all of the chunk items.  This way we can create chunk
4459          * mappings that reference all of the devices that are afound
4460          */
4461         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4462         key.offset = 0;
4463         key.type = 0;
4464 again:
4465         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4466         if (ret < 0)
4467                 goto error;
4468         while (1) {
4469                 leaf = path->nodes[0];
4470                 slot = path->slots[0];
4471                 if (slot >= btrfs_header_nritems(leaf)) {
4472                         ret = btrfs_next_leaf(root, path);
4473                         if (ret == 0)
4474                                 continue;
4475                         if (ret < 0)
4476                                 goto error;
4477                         break;
4478                 }
4479                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4480                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4481                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4482                                 break;
4483                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4484                                 struct btrfs_dev_item *dev_item;
4485                                 dev_item = btrfs_item_ptr(leaf, slot,
4486                                                   struct btrfs_dev_item);
4487                                 ret = read_one_dev(root, leaf, dev_item);
4488                                 if (ret)
4489                                         goto error;
4490                         }
4491                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4492                         struct btrfs_chunk *chunk;
4493                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4494                         ret = read_one_chunk(root, &found_key, leaf, chunk);
4495                         if (ret)
4496                                 goto error;
4497                 }
4498                 path->slots[0]++;
4499         }
4500         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4501                 key.objectid = 0;
4502                 btrfs_release_path(path);
4503                 goto again;
4504         }
4505         ret = 0;
4506 error:
4507         unlock_chunks(root);
4508         mutex_unlock(&uuid_mutex);
4509
4510         btrfs_free_path(path);
4511         return ret;
4512 }