Pileus Git - ~andy/linux/blob - 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/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <asm/div64.h>
  29 #include "compat.h"
  30 #include "ctree.h"
  31 #include "extent_map.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "print-tree.h"
  35 #include "volumes.h"
  36 #include "async-thread.h"
  37 #include "check-integrity.h"
  38 #include "rcu-string.h"
  39
  40 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  41                                 struct btrfs_root *root,
  42                                 struct btrfs_device *device);
  43 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  44 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  45 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  46
  47 static DEFINE_MUTEX(uuid_mutex);
  48 static LIST_HEAD(fs_uuids);
  49
  50 static void lock_chunks(struct btrfs_root *root)
  51 {
  52         mutex_lock(&root->fs_info->chunk_mutex);
  53 }
  54
  55 static void unlock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_unlock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  61 {
  62         struct btrfs_device *device;
  63         WARN_ON(fs_devices->opened);
  64         while (!list_empty(&fs_devices->devices)) {
  65                 device = list_entry(fs_devices->devices.next,
  66                                     struct btrfs_device, dev_list);
  67                 list_del(&device->dev_list);
  68                 rcu_string_free(device->name);
  69                 kfree(device);
  70         }
  71         kfree(fs_devices);
  72 }
  73
  74 void btrfs_cleanup_fs_uuids(void)
  75 {
  76         struct btrfs_fs_devices *fs_devices;
  77
  78         while (!list_empty(&fs_uuids)) {
  79                 fs_devices = list_entry(fs_uuids.next,
  80                                         struct btrfs_fs_devices, list);
  81                 list_del(&fs_devices->list);
  82                 free_fs_devices(fs_devices);
  83         }
  84 }
  85
  86 static noinline struct btrfs_device *__find_device(struct list_head *head,
  87                                                    u64 devid, u8 *uuid)
  88 {
  89         struct btrfs_device *dev;
  90
  91         list_for_each_entry(dev, head, dev_list) {
  92                 if (dev->devid == devid &&
  93                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  94                         return dev;
  95                 }
  96         }
  97         return NULL;
  98 }
  99
 100 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 101 {
 102         struct btrfs_fs_devices *fs_devices;
 103
 104         list_for_each_entry(fs_devices, &fs_uuids, list) {
 105                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 106                         return fs_devices;
 107         }
 108         return NULL;
 109 }
 110
 111 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 112                         struct bio *head, struct bio *tail)
 113 {
 114
 115         struct bio *old_head;
 116
 117         old_head = pending_bios->head;
 118         pending_bios->head = head;
 119         if (pending_bios->tail)
 120                 tail->bi_next = old_head;
 121         else
 122                 pending_bios->tail = tail;
 123 }
 124
 125 /*
 126  * we try to collect pending bios for a device so we don't get a large
 127  * number of procs sending bios down to the same device.  This greatly
 128  * improves the schedulers ability to collect and merge the bios.
 129  *
 130  * But, it also turns into a long list of bios to process and that is sure
 131  * to eventually make the worker thread block.  The solution here is to
 132  * make some progress and then put this work struct back at the end of
 133  * the list if the block device is congested.  This way, multiple devices
 134  * can make progress from a single worker thread.
 135  */
 136 static noinline void run_scheduled_bios(struct btrfs_device *device)
 137 {
 138         struct bio *pending;
 139         struct backing_dev_info *bdi;
 140         struct btrfs_fs_info *fs_info;
 141         struct btrfs_pending_bios *pending_bios;
 142         struct bio *tail;
 143         struct bio *cur;
 144         int again = 0;
 145         unsigned long num_run;
 146         unsigned long batch_run = 0;
 147         unsigned long limit;
 148         unsigned long last_waited = 0;
 149         int force_reg = 0;
 150         int sync_pending = 0;
 151         struct blk_plug plug;
 152
 153         /*
 154          * this function runs all the bios we've collected for
 155          * a particular device.  We don't want to wander off to
 156          * another device without first sending all of these down.
 157          * So, setup a plug here and finish it off before we return
 158          */
 159         blk_start_plug(&plug);
 160
 161         bdi = blk_get_backing_dev_info(device->bdev);
 162         fs_info = device->dev_root->fs_info;
 163         limit = btrfs_async_submit_limit(fs_info);
 164         limit = limit * 2 / 3;
 165
 166 loop:
 167         spin_lock(&device->io_lock);
 168
 169 loop_lock:
 170         num_run = 0;
 171
 172         /* take all the bios off the list at once and process them
 173          * later on (without the lock held).  But, remember the
 174          * tail and other pointers so the bios can be properly reinserted
 175          * into the list if we hit congestion
 176          */
 177         if (!force_reg && device->pending_sync_bios.head) {
 178                 pending_bios = &device->pending_sync_bios;
 179                 force_reg = 1;
 180         } else {
 181                 pending_bios = &device->pending_bios;
 182                 force_reg = 0;
 183         }
 184
 185         pending = pending_bios->head;
 186         tail = pending_bios->tail;
 187         WARN_ON(pending && !tail);
 188
 189         /*
 190          * if pending was null this time around, no bios need processing
 191          * at all and we can stop.  Otherwise it'll loop back up again
 192          * and do an additional check so no bios are missed.
 193          *
 194          * device->running_pending is used to synchronize with the
 195          * schedule_bio code.
 196          */
 197         if (device->pending_sync_bios.head == NULL &&
 198             device->pending_bios.head == NULL) {
 199                 again = 0;
 200                 device->running_pending = 0;
 201         } else {
 202                 again = 1;
 203                 device->running_pending = 1;
 204         }
 205
 206         pending_bios->head = NULL;
 207         pending_bios->tail = NULL;
 208
 209         spin_unlock(&device->io_lock);
 210
 211         while (pending) {
 212
 213                 rmb();
 214                 /* we want to work on both lists, but do more bios on the
 215                  * sync list than the regular list
 216                  */
 217                 if ((num_run > 32 &&
 218                     pending_bios != &device->pending_sync_bios &&
 219                     device->pending_sync_bios.head) ||
 220                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 221                     device->pending_bios.head)) {
 222                         spin_lock(&device->io_lock);
 223                         requeue_list(pending_bios, pending, tail);
 224                         goto loop_lock;
 225                 }
 226
 227                 cur = pending;
 228                 pending = pending->bi_next;
 229                 cur->bi_next = NULL;
 230
 231                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 232                     waitqueue_active(&fs_info->async_submit_wait))
 233                         wake_up(&fs_info->async_submit_wait);
 234
 235                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 236
 237                 /*
 238                  * if we're doing the sync list, record that our
 239                  * plug has some sync requests on it
 240                  *
 241                  * If we're doing the regular list and there are
 242                  * sync requests sitting around, unplug before
 243                  * we add more
 244                  */
 245                 if (pending_bios == &device->pending_sync_bios) {
 246                         sync_pending = 1;
 247                 } else if (sync_pending) {
 248                         blk_finish_plug(&plug);
 249                         blk_start_plug(&plug);
 250                         sync_pending = 0;
 251                 }
 252
 253                 btrfsic_submit_bio(cur->bi_rw, cur);
 254                 num_run++;
 255                 batch_run++;
 256                 if (need_resched())
 257                         cond_resched();
 258
 259                 /*
 260                  * we made progress, there is more work to do and the bdi
 261                  * is now congested.  Back off and let other work structs
 262                  * run instead
 263                  */
 264                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 265                     fs_info->fs_devices->open_devices > 1) {
 266                         struct io_context *ioc;
 267
 268                         ioc = current->io_context;
 269
 270                         /*
 271                          * the main goal here is that we don't want to
 272                          * block if we're going to be able to submit
 273                          * more requests without blocking.
 274                          *
 275                          * This code does two great things, it pokes into
 276                          * the elevator code from a filesystem _and_
 277                          * it makes assumptions about how batching works.
 278                          */
 279                         if (ioc && ioc->nr_batch_requests > 0 &&
 280                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 281                             (last_waited == 0 ||
 282                              ioc->last_waited == last_waited)) {
 283                                 /*
 284                                  * we want to go through our batch of
 285                                  * requests and stop.  So, we copy out
 286                                  * the ioc->last_waited time and test
 287                                  * against it before looping
 288                                  */
 289                                 last_waited = ioc->last_waited;
 290                                 if (need_resched())
 291                                         cond_resched();
 292                                 continue;
 293                         }
 294                         spin_lock(&device->io_lock);
 295                         requeue_list(pending_bios, pending, tail);
 296                         device->running_pending = 1;
 297
 298                         spin_unlock(&device->io_lock);
 299                         btrfs_requeue_work(&device->work);
 300                         goto done;
 301                 }
 302                 /* unplug every 64 requests just for good measure */
 303                 if (batch_run % 64 == 0) {
 304                         blk_finish_plug(&plug);
 305                         blk_start_plug(&plug);
 306                         sync_pending = 0;
 307                 }
 308         }
 309
 310         cond_resched();
 311         if (again)
 312                 goto loop;
 313
 314         spin_lock(&device->io_lock);
 315         if (device->pending_bios.head || device->pending_sync_bios.head)
 316                 goto loop_lock;
 317         spin_unlock(&device->io_lock);
 318
 319 done:
 320         blk_finish_plug(&plug);
 321 }
 322
 323 static void pending_bios_fn(struct btrfs_work *work)
 324 {
 325         struct btrfs_device *device;
 326
 327         device = container_of(work, struct btrfs_device, work);
 328         run_scheduled_bios(device);
 329 }
 330
 331 static noinline int device_list_add(const char *path,
 332                            struct btrfs_super_block *disk_super,
 333                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 334 {
 335         struct btrfs_device *device;
 336         struct btrfs_fs_devices *fs_devices;
 337         struct rcu_string *name;
 338         u64 found_transid = btrfs_super_generation(disk_super);
 339
 340         fs_devices = find_fsid(disk_super->fsid);
 341         if (!fs_devices) {
 342                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 343                 if (!fs_devices)
 344                         return -ENOMEM;
 345                 INIT_LIST_HEAD(&fs_devices->devices);
 346                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 347                 list_add(&fs_devices->list, &fs_uuids);
 348                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 349                 fs_devices->latest_devid = devid;
 350                 fs_devices->latest_trans = found_transid;
 351                 mutex_init(&fs_devices->device_list_mutex);
 352                 device = NULL;
 353         } else {
 354                 device = __find_device(&fs_devices->devices, devid,
 355                                        disk_super->dev_item.uuid);
 356         }
 357         if (!device) {
 358                 if (fs_devices->opened)
 359                         return -EBUSY;
 360
 361                 device = kzalloc(sizeof(*device), GFP_NOFS);
 362                 if (!device) {
 363                         /* we can safely leave the fs_devices entry around */
 364                         return -ENOMEM;
 365                 }
 366                 device->devid = devid;
 367                 device->dev_stats_valid = 0;
 368                 device->work.func = pending_bios_fn;
 369                 memcpy(device->uuid, disk_super->dev_item.uuid,
 370                        BTRFS_UUID_SIZE);
 371                 spin_lock_init(&device->io_lock);
 372
 373                 name = rcu_string_strdup(path, GFP_NOFS);
 374                 if (!name) {
 375                         kfree(device);
 376                         return -ENOMEM;
 377                 }
 378                 rcu_assign_pointer(device->name, name);
 379                 INIT_LIST_HEAD(&device->dev_alloc_list);
 380
 381                 /* init readahead state */
 382                 spin_lock_init(&device->reada_lock);
 383                 device->reada_curr_zone = NULL;
 384                 atomic_set(&device->reada_in_flight, 0);
 385                 device->reada_next = 0;
 386                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 387                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 388
 389                 mutex_lock(&fs_devices->device_list_mutex);
 390                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 391                 mutex_unlock(&fs_devices->device_list_mutex);
 392
 393                 device->fs_devices = fs_devices;
 394                 fs_devices->num_devices++;
 395         } else if (!device->name || strcmp(device->name->str, path)) {
 396                 name = rcu_string_strdup(path, GFP_NOFS);
 397                 if (!name)
 398                         return -ENOMEM;
 399                 rcu_string_free(device->name);
 400                 rcu_assign_pointer(device->name, name);
 401                 if (device->missing) {
 402                         fs_devices->missing_devices--;
 403                         device->missing = 0;
 404                 }
 405         }
 406
 407         if (found_transid > fs_devices->latest_trans) {
 408                 fs_devices->latest_devid = devid;
 409                 fs_devices->latest_trans = found_transid;
 410         }
 411         *fs_devices_ret = fs_devices;
 412         return 0;
 413 }
 414
 415 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 416 {
 417         struct btrfs_fs_devices *fs_devices;
 418         struct btrfs_device *device;
 419         struct btrfs_device *orig_dev;
 420
 421         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 422         if (!fs_devices)
 423                 return ERR_PTR(-ENOMEM);
 424
 425         INIT_LIST_HEAD(&fs_devices->devices);
 426         INIT_LIST_HEAD(&fs_devices->alloc_list);
 427         INIT_LIST_HEAD(&fs_devices->list);
 428         mutex_init(&fs_devices->device_list_mutex);
 429         fs_devices->latest_devid = orig->latest_devid;
 430         fs_devices->latest_trans = orig->latest_trans;
 431         fs_devices->total_devices = orig->total_devices;
 432         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 433
 434         /* We have held the volume lock, it is safe to get the devices. */
 435         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 436                 struct rcu_string *name;
 437
 438                 device = kzalloc(sizeof(*device), GFP_NOFS);
 439                 if (!device)
 440                         goto error;
 441
 442                 /*
 443                  * This is ok to do without rcu read locked because we hold the
 444                  * uuid mutex so nothing we touch in here is going to disappear.
 445                  */
 446                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 447                 if (!name) {
 448                         kfree(device);
 449                         goto error;
 450                 }
 451                 rcu_assign_pointer(device->name, name);
 452
 453                 device->devid = orig_dev->devid;
 454                 device->work.func = pending_bios_fn;
 455                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 456                 spin_lock_init(&device->io_lock);
 457                 INIT_LIST_HEAD(&device->dev_list);
 458                 INIT_LIST_HEAD(&device->dev_alloc_list);
 459
 460                 list_add(&device->dev_list, &fs_devices->devices);
 461                 device->fs_devices = fs_devices;
 462                 fs_devices->num_devices++;
 463         }
 464         return fs_devices;
 465 error:
 466         free_fs_devices(fs_devices);
 467         return ERR_PTR(-ENOMEM);
 468 }
 469
 470 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 471 {
 472         struct btrfs_device *device, *next;
 473
 474         struct block_device *latest_bdev = NULL;
 475         u64 latest_devid = 0;
 476         u64 latest_transid = 0;
 477
 478         mutex_lock(&uuid_mutex);
 479 again:
 480         /* This is the initialized path, it is safe to release the devices. */
 481         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 482                 if (device->in_fs_metadata) {
 483                         if (!latest_transid ||
 484                             device->generation > latest_transid) {
 485                                 latest_devid = device->devid;
 486                                 latest_transid = device->generation;
 487                                 latest_bdev = device->bdev;
 488                         }
 489                         continue;
 490                 }
 491
 492                 if (device->bdev) {
 493                         blkdev_put(device->bdev, device->mode);
 494                         device->bdev = NULL;
 495                         fs_devices->open_devices--;
 496                 }
 497                 if (device->writeable) {
 498                         list_del_init(&device->dev_alloc_list);
 499                         device->writeable = 0;
 500                         fs_devices->rw_devices--;
 501                 }
 502                 list_del_init(&device->dev_list);
 503                 fs_devices->num_devices--;
 504                 rcu_string_free(device->name);
 505                 kfree(device);
 506         }
 507
 508         if (fs_devices->seed) {
 509                 fs_devices = fs_devices->seed;
 510                 goto again;
 511         }
 512
 513         fs_devices->latest_bdev = latest_bdev;
 514         fs_devices->latest_devid = latest_devid;
 515         fs_devices->latest_trans = latest_transid;
 516
 517         mutex_unlock(&uuid_mutex);
 518 }
 519
 520 static void __free_device(struct work_struct *work)
 521 {
 522         struct btrfs_device *device;
 523
 524         device = container_of(work, struct btrfs_device, rcu_work);
 525
 526         if (device->bdev)
 527                 blkdev_put(device->bdev, device->mode);
 528
 529         rcu_string_free(device->name);
 530         kfree(device);
 531 }
 532
 533 static void free_device(struct rcu_head *head)
 534 {
 535         struct btrfs_device *device;
 536
 537         device = container_of(head, struct btrfs_device, rcu);
 538
 539         INIT_WORK(&device->rcu_work, __free_device);
 540         schedule_work(&device->rcu_work);
 541 }
 542
 543 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 544 {
 545         struct btrfs_device *device;
 546
 547         if (--fs_devices->opened > 0)
 548                 return 0;
 549
 550         mutex_lock(&fs_devices->device_list_mutex);
 551         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 552                 struct btrfs_device *new_device;
 553                 struct rcu_string *name;
 554
 555                 if (device->bdev)
 556                         fs_devices->open_devices--;
 557
 558                 if (device->writeable) {
 559                         list_del_init(&device->dev_alloc_list);
 560                         fs_devices->rw_devices--;
 561                 }
 562
 563                 if (device->can_discard)
 564                         fs_devices->num_can_discard--;
 565
 566                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 567                 BUG_ON(!new_device); /* -ENOMEM */
 568                 memcpy(new_device, device, sizeof(*new_device));
 569
 570                 /* Safe because we are under uuid_mutex */
 571                 if (device->name) {
 572                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 573                         BUG_ON(device->name && !name); /* -ENOMEM */
 574                         rcu_assign_pointer(new_device->name, name);
 575                 }
 576                 new_device->bdev = NULL;
 577                 new_device->writeable = 0;
 578                 new_device->in_fs_metadata = 0;
 579                 new_device->can_discard = 0;
 580                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 581
 582                 call_rcu(&device->rcu, free_device);
 583         }
 584         mutex_unlock(&fs_devices->device_list_mutex);
 585
 586         WARN_ON(fs_devices->open_devices);
 587         WARN_ON(fs_devices->rw_devices);
 588         fs_devices->opened = 0;
 589         fs_devices->seeding = 0;
 590
 591         return 0;
 592 }
 593
 594 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 595 {
 596         struct btrfs_fs_devices *seed_devices = NULL;
 597         int ret;
 598
 599         mutex_lock(&uuid_mutex);
 600         ret = __btrfs_close_devices(fs_devices);
 601         if (!fs_devices->opened) {
 602                 seed_devices = fs_devices->seed;
 603                 fs_devices->seed = NULL;
 604         }
 605         mutex_unlock(&uuid_mutex);
 606
 607         while (seed_devices) {
 608                 fs_devices = seed_devices;
 609                 seed_devices = fs_devices->seed;
 610                 __btrfs_close_devices(fs_devices);
 611                 free_fs_devices(fs_devices);
 612         }
 613         return ret;
 614 }
 615
 616 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 617                                 fmode_t flags, void *holder)
 618 {
 619         struct request_queue *q;
 620         struct block_device *bdev;
 621         struct list_head *head = &fs_devices->devices;
 622         struct btrfs_device *device;
 623         struct block_device *latest_bdev = NULL;
 624         struct buffer_head *bh;
 625         struct btrfs_super_block *disk_super;
 626         u64 latest_devid = 0;
 627         u64 latest_transid = 0;
 628         u64 devid;
 629         int seeding = 1;
 630         int ret = 0;
 631
 632         flags |= FMODE_EXCL;
 633
 634         list_for_each_entry(device, head, dev_list) {
 635                 if (device->bdev)
 636                         continue;
 637                 if (!device->name)
 638                         continue;
 639
 640                 bdev = blkdev_get_by_path(device->name->str, flags, holder);
 641                 if (IS_ERR(bdev)) {
 642                         printk(KERN_INFO "open %s failed\n", device->name->str);
 643                         goto error;
 644                 }
 645                 filemap_write_and_wait(bdev->bd_inode->i_mapping);
 646                 invalidate_bdev(bdev);
 647                 set_blocksize(bdev, 4096);
 648
 649                 bh = btrfs_read_dev_super(bdev);
 650                 if (!bh)
 651                         goto error_close;
 652
 653                 disk_super = (struct btrfs_super_block *)bh->b_data;
 654                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 655                 if (devid != device->devid)
 656                         goto error_brelse;
 657
 658                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 659                            BTRFS_UUID_SIZE))
 660                         goto error_brelse;
 661
 662                 device->generation = btrfs_super_generation(disk_super);
 663                 if (!latest_transid || device->generation > latest_transid) {
 664                         latest_devid = devid;
 665                         latest_transid = device->generation;
 666                         latest_bdev = bdev;
 667                 }
 668
 669                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 670                         device->writeable = 0;
 671                 } else {
 672                         device->writeable = !bdev_read_only(bdev);
 673                         seeding = 0;
 674                 }
 675
 676                 q = bdev_get_queue(bdev);
 677                 if (blk_queue_discard(q)) {
 678                         device->can_discard = 1;
 679                         fs_devices->num_can_discard++;
 680                 }
 681
 682                 device->bdev = bdev;
 683                 device->in_fs_metadata = 0;
 684                 device->mode = flags;
 685
 686                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 687                         fs_devices->rotating = 1;
 688
 689                 fs_devices->open_devices++;
 690                 if (device->writeable) {
 691                         fs_devices->rw_devices++;
 692                         list_add(&device->dev_alloc_list,
 693                                  &fs_devices->alloc_list);
 694                 }
 695                 brelse(bh);
 696                 continue;
 697
 698 error_brelse:
 699                 brelse(bh);
 700 error_close:
 701                 blkdev_put(bdev, flags);
 702 error:
 703                 continue;
 704         }
 705         if (fs_devices->open_devices == 0) {
 706                 ret = -EINVAL;
 707                 goto out;
 708         }
 709         fs_devices->seeding = seeding;
 710         fs_devices->opened = 1;
 711         fs_devices->latest_bdev = latest_bdev;
 712         fs_devices->latest_devid = latest_devid;
 713         fs_devices->latest_trans = latest_transid;
 714         fs_devices->total_rw_bytes = 0;
 715 out:
 716         return ret;
 717 }
 718
 719 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 720                        fmode_t flags, void *holder)
 721 {
 722         int ret;
 723
 724         mutex_lock(&uuid_mutex);
 725         if (fs_devices->opened) {
 726                 fs_devices->opened++;
 727                 ret = 0;
 728         } else {
 729                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 730         }
 731         mutex_unlock(&uuid_mutex);
 732         return ret;
 733 }
 734
 735 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 736                           struct btrfs_fs_devices **fs_devices_ret)
 737 {
 738         struct btrfs_super_block *disk_super;
 739         struct block_device *bdev;
 740         struct buffer_head *bh;
 741         int ret;
 742         u64 devid;
 743         u64 transid;
 744         u64 total_devices;
 745
 746         flags |= FMODE_EXCL;
 747         bdev = blkdev_get_by_path(path, flags, holder);
 748
 749         if (IS_ERR(bdev)) {
 750                 ret = PTR_ERR(bdev);
 751                 goto error;
 752         }
 753
 754         mutex_lock(&uuid_mutex);
 755         ret = set_blocksize(bdev, 4096);
 756         if (ret)
 757                 goto error_close;
 758         bh = btrfs_read_dev_super(bdev);
 759         if (!bh) {
 760                 ret = -EINVAL;
 761                 goto error_close;
 762         }
 763         disk_super = (struct btrfs_super_block *)bh->b_data;
 764         devid = btrfs_stack_device_id(&disk_super->dev_item);
 765         transid = btrfs_super_generation(disk_super);
 766         total_devices = btrfs_super_num_devices(disk_super);
 767         if (disk_super->label[0])
 768                 printk(KERN_INFO "device label %s ", disk_super->label);
 769         else
 770                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 771         printk(KERN_CONT "devid %llu transid %llu %s\n",
 772                (unsigned long long)devid, (unsigned long long)transid, path);
 773         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 774         if (!ret && fs_devices_ret)
 775                 (*fs_devices_ret)->total_devices = total_devices;
 776         brelse(bh);
 777 error_close:
 778         mutex_unlock(&uuid_mutex);
 779         blkdev_put(bdev, flags);
 780 error:
 781         return ret;
 782 }
 783
 784 /* helper to account the used device space in the range */
 785 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 786                                    u64 end, u64 *length)
 787 {
 788         struct btrfs_key key;
 789         struct btrfs_root *root = device->dev_root;
 790         struct btrfs_dev_extent *dev_extent;
 791         struct btrfs_path *path;
 792         u64 extent_end;
 793         int ret;
 794         int slot;
 795         struct extent_buffer *l;
 796
 797         *length = 0;
 798
 799         if (start >= device->total_bytes)
 800                 return 0;
 801
 802         path = btrfs_alloc_path();
 803         if (!path)
 804                 return -ENOMEM;
 805         path->reada = 2;
 806
 807         key.objectid = device->devid;
 808         key.offset = start;
 809         key.type = BTRFS_DEV_EXTENT_KEY;
 810
 811         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 812         if (ret < 0)
 813                 goto out;
 814         if (ret > 0) {
 815                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 816                 if (ret < 0)
 817                         goto out;
 818         }
 819
 820         while (1) {
 821                 l = path->nodes[0];
 822                 slot = path->slots[0];
 823                 if (slot >= btrfs_header_nritems(l)) {
 824                         ret = btrfs_next_leaf(root, path);
 825                         if (ret == 0)
 826                                 continue;
 827                         if (ret < 0)
 828                                 goto out;
 829
 830                         break;
 831                 }
 832                 btrfs_item_key_to_cpu(l, &key, slot);
 833
 834                 if (key.objectid < device->devid)
 835                         goto next;
 836
 837                 if (key.objectid > device->devid)
 838                         break;
 839
 840                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 841                         goto next;
 842
 843                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 844                 extent_end = key.offset + btrfs_dev_extent_length(l,
 845                                                                   dev_extent);
 846                 if (key.offset <= start && extent_end > end) {
 847                         *length = end - start + 1;
 848                         break;
 849                 } else if (key.offset <= start && extent_end > start)
 850                         *length += extent_end - start;
 851                 else if (key.offset > start && extent_end <= end)
 852                         *length += extent_end - key.offset;
 853                 else if (key.offset > start && key.offset <= end) {
 854                         *length += end - key.offset + 1;
 855                         break;
 856                 } else if (key.offset > end)
 857                         break;
 858
 859 next:
 860                 path->slots[0]++;
 861         }
 862         ret = 0;
 863 out:
 864         btrfs_free_path(path);
 865         return ret;
 866 }
 867
 868 /*
 869  * find_free_dev_extent - find free space in the specified device
 870  * @device:     the device which we search the free space in
 871  * @num_bytes:  the size of the free space that we need
 872  * @start:      store the start of the free space.
 873  * @len:        the size of the free space. that we find, or the size of the max
 874  *              free space if we don't find suitable free space
 875  *
 876  * this uses a pretty simple search, the expectation is that it is
 877  * called very infrequently and that a given device has a small number
 878  * of extents
 879  *
 880  * @start is used to store the start of the free space if we find. But if we
 881  * don't find suitable free space, it will be used to store the start position
 882  * of the max free space.
 883  *
 884  * @len is used to store the size of the free space that we find.
 885  * But if we don't find suitable free space, it is used to store the size of
 886  * the max free space.
 887  */
 888 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 889                          u64 *start, u64 *len)
 890 {
 891         struct btrfs_key key;
 892         struct btrfs_root *root = device->dev_root;
 893         struct btrfs_dev_extent *dev_extent;
 894         struct btrfs_path *path;
 895         u64 hole_size;
 896         u64 max_hole_start;
 897         u64 max_hole_size;
 898         u64 extent_end;
 899         u64 search_start;
 900         u64 search_end = device->total_bytes;
 901         int ret;
 902         int slot;
 903         struct extent_buffer *l;
 904
 905         /* FIXME use last free of some kind */
 906
 907         /* we don't want to overwrite the superblock on the drive,
 908          * so we make sure to start at an offset of at least 1MB
 909          */
 910         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 911
 912         max_hole_start = search_start;
 913         max_hole_size = 0;
 914         hole_size = 0;
 915
 916         if (search_start >= search_end) {
 917                 ret = -ENOSPC;
 918                 goto error;
 919         }
 920
 921         path = btrfs_alloc_path();
 922         if (!path) {
 923                 ret = -ENOMEM;
 924                 goto error;
 925         }
 926         path->reada = 2;
 927
 928         key.objectid = device->devid;
 929         key.offset = search_start;
 930         key.type = BTRFS_DEV_EXTENT_KEY;
 931
 932         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 933         if (ret < 0)
 934                 goto out;
 935         if (ret > 0) {
 936                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 937                 if (ret < 0)
 938                         goto out;
 939         }
 940
 941         while (1) {
 942                 l = path->nodes[0];
 943                 slot = path->slots[0];
 944                 if (slot >= btrfs_header_nritems(l)) {
 945                         ret = btrfs_next_leaf(root, path);
 946                         if (ret == 0)
 947                                 continue;
 948                         if (ret < 0)
 949                                 goto out;
 950
 951                         break;
 952                 }
 953                 btrfs_item_key_to_cpu(l, &key, slot);
 954
 955                 if (key.objectid < device->devid)
 956                         goto next;
 957
 958                 if (key.objectid > device->devid)
 959                         break;
 960
 961                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 962                         goto next;
 963
 964                 if (key.offset > search_start) {
 965                         hole_size = key.offset - search_start;
 966
 967                         if (hole_size > max_hole_size) {
 968                                 max_hole_start = search_start;
 969                                 max_hole_size = hole_size;
 970                         }
 971
 972                         /*
 973                          * If this free space is greater than which we need,
 974                          * it must be the max free space that we have found
 975                          * until now, so max_hole_start must point to the start
 976                          * of this free space and the length of this free space
 977                          * is stored in max_hole_size. Thus, we return
 978                          * max_hole_start and max_hole_size and go back to the
 979                          * caller.
 980                          */
 981                         if (hole_size >= num_bytes) {
 982                                 ret = 0;
 983                                 goto out;
 984                         }
 985                 }
 986
 987                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 988                 extent_end = key.offset + btrfs_dev_extent_length(l,
 989                                                                   dev_extent);
 990                 if (extent_end > search_start)
 991                         search_start = extent_end;
 992 next:
 993                 path->slots[0]++;
 994                 cond_resched();
 995         }
 996
 997         /*
 998          * At this point, search_start should be the end of
 999          * allocated dev extents, and when shrinking the device,
1000          * search_end may be smaller than search_start.
1001          */
1002         if (search_end > search_start)
1003                 hole_size = search_end - search_start;
1004
1005         if (hole_size > max_hole_size) {
1006                 max_hole_start = search_start;
1007                 max_hole_size = hole_size;
1008         }
1009
1010         /* See above. */
1011         if (hole_size < num_bytes)
1012                 ret = -ENOSPC;
1013         else
1014                 ret = 0;
1015
1016 out:
1017         btrfs_free_path(path);
1018 error:
1019         *start = max_hole_start;
1020         if (len)
1021                 *len = max_hole_size;
1022         return ret;
1023 }
1024
1025 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1026                           struct btrfs_device *device,
1027                           u64 start)
1028 {
1029         int ret;
1030         struct btrfs_path *path;
1031         struct btrfs_root *root = device->dev_root;
1032         struct btrfs_key key;
1033         struct btrfs_key found_key;
1034         struct extent_buffer *leaf = NULL;
1035         struct btrfs_dev_extent *extent = NULL;
1036
1037         path = btrfs_alloc_path();
1038         if (!path)
1039                 return -ENOMEM;
1040
1041         key.objectid = device->devid;
1042         key.offset = start;
1043         key.type = BTRFS_DEV_EXTENT_KEY;
1044 again:
1045         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1046         if (ret > 0) {
1047                 ret = btrfs_previous_item(root, path, key.objectid,
1048                                           BTRFS_DEV_EXTENT_KEY);
1049                 if (ret)
1050                         goto out;
1051                 leaf = path->nodes[0];
1052                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1053                 extent = btrfs_item_ptr(leaf, path->slots[0],
1054                                         struct btrfs_dev_extent);
1055                 BUG_ON(found_key.offset > start || found_key.offset +
1056                        btrfs_dev_extent_length(leaf, extent) < start);
1057                 key = found_key;
1058                 btrfs_release_path(path);
1059                 goto again;
1060         } else if (ret == 0) {
1061                 leaf = path->nodes[0];
1062                 extent = btrfs_item_ptr(leaf, path->slots[0],
1063                                         struct btrfs_dev_extent);
1064         } else {
1065                 btrfs_error(root->fs_info, ret, "Slot search failed");
1066                 goto out;
1067         }
1068
1069         if (device->bytes_used > 0) {
1070                 u64 len = btrfs_dev_extent_length(leaf, extent);
1071                 device->bytes_used -= len;
1072                 spin_lock(&root->fs_info->free_chunk_lock);
1073                 root->fs_info->free_chunk_space += len;
1074                 spin_unlock(&root->fs_info->free_chunk_lock);
1075         }
1076         ret = btrfs_del_item(trans, root, path);
1077         if (ret) {
1078                 btrfs_error(root->fs_info, ret,
1079                             "Failed to remove dev extent item");
1080         }
1081 out:
1082         btrfs_free_path(path);
1083         return ret;
1084 }
1085
1086 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1087                            struct btrfs_device *device,
1088                            u64 chunk_tree, u64 chunk_objectid,
1089                            u64 chunk_offset, u64 start, u64 num_bytes)
1090 {
1091         int ret;
1092         struct btrfs_path *path;
1093         struct btrfs_root *root = device->dev_root;
1094         struct btrfs_dev_extent *extent;
1095         struct extent_buffer *leaf;
1096         struct btrfs_key key;
1097
1098         WARN_ON(!device->in_fs_metadata);
1099         path = btrfs_alloc_path();
1100         if (!path)
1101                 return -ENOMEM;
1102
1103         key.objectid = device->devid;
1104         key.offset = start;
1105         key.type = BTRFS_DEV_EXTENT_KEY;
1106         ret = btrfs_insert_empty_item(trans, root, path, &key,
1107                                       sizeof(*extent));
1108         if (ret)
1109                 goto out;
1110
1111         leaf = path->nodes[0];
1112         extent = btrfs_item_ptr(leaf, path->slots[0],
1113                                 struct btrfs_dev_extent);
1114         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1115         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1116         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1117
1118         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1119                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1120                     BTRFS_UUID_SIZE);
1121
1122         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1123         btrfs_mark_buffer_dirty(leaf);
1124 out:
1125         btrfs_free_path(path);
1126         return ret;
1127 }
1128
1129 static noinline int find_next_chunk(struct btrfs_root *root,
1130                                     u64 objectid, u64 *offset)
1131 {
1132         struct btrfs_path *path;
1133         int ret;
1134         struct btrfs_key key;
1135         struct btrfs_chunk *chunk;
1136         struct btrfs_key found_key;
1137
1138         path = btrfs_alloc_path();
1139         if (!path)
1140                 return -ENOMEM;
1141
1142         key.objectid = objectid;
1143         key.offset = (u64)-1;
1144         key.type = BTRFS_CHUNK_ITEM_KEY;
1145
1146         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1147         if (ret < 0)
1148                 goto error;
1149
1150         BUG_ON(ret == 0); /* Corruption */
1151
1152         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1153         if (ret) {
1154                 *offset = 0;
1155         } else {
1156                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1157                                       path->slots[0]);
1158                 if (found_key.objectid != objectid)
1159                         *offset = 0;
1160                 else {
1161                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1162                                                struct btrfs_chunk);
1163                         *offset = found_key.offset +
1164                                 btrfs_chunk_length(path->nodes[0], chunk);
1165                 }
1166         }
1167         ret = 0;
1168 error:
1169         btrfs_free_path(path);
1170         return ret;
1171 }
1172
1173 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1174 {
1175         int ret;
1176         struct btrfs_key key;
1177         struct btrfs_key found_key;
1178         struct btrfs_path *path;
1179
1180         root = root->fs_info->chunk_root;
1181
1182         path = btrfs_alloc_path();
1183         if (!path)
1184                 return -ENOMEM;
1185
1186         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1187         key.type = BTRFS_DEV_ITEM_KEY;
1188         key.offset = (u64)-1;
1189
1190         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1191         if (ret < 0)
1192                 goto error;
1193
1194         BUG_ON(ret == 0); /* Corruption */
1195
1196         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1197                                   BTRFS_DEV_ITEM_KEY);
1198         if (ret) {
1199                 *objectid = 1;
1200         } else {
1201                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1202                                       path->slots[0]);
1203                 *objectid = found_key.offset + 1;
1204         }
1205         ret = 0;
1206 error:
1207         btrfs_free_path(path);
1208         return ret;
1209 }
1210
1211 /*
1212  * the device information is stored in the chunk root
1213  * the btrfs_device struct should be fully filled in
1214  */
1215 int btrfs_add_device(struct btrfs_trans_handle *trans,
1216                      struct btrfs_root *root,
1217                      struct btrfs_device *device)
1218 {
1219         int ret;
1220         struct btrfs_path *path;
1221         struct btrfs_dev_item *dev_item;
1222         struct extent_buffer *leaf;
1223         struct btrfs_key key;
1224         unsigned long ptr;
1225
1226         root = root->fs_info->chunk_root;
1227
1228         path = btrfs_alloc_path();
1229         if (!path)
1230                 return -ENOMEM;
1231
1232         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1233         key.type = BTRFS_DEV_ITEM_KEY;
1234         key.offset = device->devid;
1235
1236         ret = btrfs_insert_empty_item(trans, root, path, &key,
1237                                       sizeof(*dev_item));
1238         if (ret)
1239                 goto out;
1240
1241         leaf = path->nodes[0];
1242         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1243
1244         btrfs_set_device_id(leaf, dev_item, device->devid);
1245         btrfs_set_device_generation(leaf, dev_item, 0);
1246         btrfs_set_device_type(leaf, dev_item, device->type);
1247         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1248         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1249         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1250         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1251         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1252         btrfs_set_device_group(leaf, dev_item, 0);
1253         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1254         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1255         btrfs_set_device_start_offset(leaf, dev_item, 0);
1256
1257         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1258         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1259         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1260         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1261         btrfs_mark_buffer_dirty(leaf);
1262
1263         ret = 0;
1264 out:
1265         btrfs_free_path(path);
1266         return ret;
1267 }
1268
1269 static int btrfs_rm_dev_item(struct btrfs_root *root,
1270                              struct btrfs_device *device)
1271 {
1272         int ret;
1273         struct btrfs_path *path;
1274         struct btrfs_key key;
1275         struct btrfs_trans_handle *trans;
1276
1277         root = root->fs_info->chunk_root;
1278
1279         path = btrfs_alloc_path();
1280         if (!path)
1281                 return -ENOMEM;
1282
1283         trans = btrfs_start_transaction(root, 0);
1284         if (IS_ERR(trans)) {
1285                 btrfs_free_path(path);
1286                 return PTR_ERR(trans);
1287         }
1288         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1289         key.type = BTRFS_DEV_ITEM_KEY;
1290         key.offset = device->devid;
1291         lock_chunks(root);
1292
1293         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1294         if (ret < 0)
1295                 goto out;
1296
1297         if (ret > 0) {
1298                 ret = -ENOENT;
1299                 goto out;
1300         }
1301
1302         ret = btrfs_del_item(trans, root, path);
1303         if (ret)
1304                 goto out;
1305 out:
1306         btrfs_free_path(path);
1307         unlock_chunks(root);
1308         btrfs_commit_transaction(trans, root);
1309         return ret;
1310 }
1311
1312 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1313 {
1314         struct btrfs_device *device;
1315         struct btrfs_device *next_device;
1316         struct block_device *bdev;
1317         struct buffer_head *bh = NULL;
1318         struct btrfs_super_block *disk_super;
1319         struct btrfs_fs_devices *cur_devices;
1320         u64 all_avail;
1321         u64 devid;
1322         u64 num_devices;
1323         u8 *dev_uuid;
1324         int ret = 0;
1325         bool clear_super = false;
1326
1327         mutex_lock(&uuid_mutex);
1328
1329         all_avail = root->fs_info->avail_data_alloc_bits |
1330                 root->fs_info->avail_system_alloc_bits |
1331                 root->fs_info->avail_metadata_alloc_bits;
1332
1333         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1334             root->fs_info->fs_devices->num_devices <= 4) {
1335                 printk(KERN_ERR "btrfs: unable to go below four devices "
1336                        "on raid10\n");
1337                 ret = -EINVAL;
1338                 goto out;
1339         }
1340
1341         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1342             root->fs_info->fs_devices->num_devices <= 2) {
1343                 printk(KERN_ERR "btrfs: unable to go below two "
1344                        "devices on raid1\n");
1345                 ret = -EINVAL;
1346                 goto out;
1347         }
1348
1349         if (strcmp(device_path, "missing") == 0) {
1350                 struct list_head *devices;
1351                 struct btrfs_device *tmp;
1352
1353                 device = NULL;
1354                 devices = &root->fs_info->fs_devices->devices;
1355                 /*
1356                  * It is safe to read the devices since the volume_mutex
1357                  * is held.
1358                  */
1359                 list_for_each_entry(tmp, devices, dev_list) {
1360                         if (tmp->in_fs_metadata && !tmp->bdev) {
1361                                 device = tmp;
1362                                 break;
1363                         }
1364                 }
1365                 bdev = NULL;
1366                 bh = NULL;
1367                 disk_super = NULL;
1368                 if (!device) {
1369                         printk(KERN_ERR "btrfs: no missing devices found to "
1370                                "remove\n");
1371                         goto out;
1372                 }
1373         } else {
1374                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1375                                           root->fs_info->bdev_holder);
1376                 if (IS_ERR(bdev)) {
1377                         ret = PTR_ERR(bdev);
1378                         goto out;
1379                 }
1380
1381                 set_blocksize(bdev, 4096);
1382                 invalidate_bdev(bdev);
1383                 bh = btrfs_read_dev_super(bdev);
1384                 if (!bh) {
1385                         ret = -EINVAL;
1386                         goto error_close;
1387                 }
1388                 disk_super = (struct btrfs_super_block *)bh->b_data;
1389                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1390                 dev_uuid = disk_super->dev_item.uuid;
1391                 device = btrfs_find_device(root, devid, dev_uuid,
1392                                            disk_super->fsid);
1393                 if (!device) {
1394                         ret = -ENOENT;
1395                         goto error_brelse;
1396                 }
1397         }
1398
1399         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1400                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1401                        "device\n");
1402                 ret = -EINVAL;
1403                 goto error_brelse;
1404         }
1405
1406         if (device->writeable) {
1407                 lock_chunks(root);
1408                 list_del_init(&device->dev_alloc_list);
1409                 unlock_chunks(root);
1410                 root->fs_info->fs_devices->rw_devices--;
1411                 clear_super = true;
1412         }
1413
1414         ret = btrfs_shrink_device(device, 0);
1415         if (ret)
1416                 goto error_undo;
1417
1418         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1419         if (ret)
1420                 goto error_undo;
1421
1422         spin_lock(&root->fs_info->free_chunk_lock);
1423         root->fs_info->free_chunk_space = device->total_bytes -
1424                 device->bytes_used;
1425         spin_unlock(&root->fs_info->free_chunk_lock);
1426
1427         device->in_fs_metadata = 0;
1428         btrfs_scrub_cancel_dev(root, device);
1429
1430         /*
1431          * the device list mutex makes sure that we don't change
1432          * the device list while someone else is writing out all
1433          * the device supers.
1434          */
1435
1436         cur_devices = device->fs_devices;
1437         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1438         list_del_rcu(&device->dev_list);
1439
1440         device->fs_devices->num_devices--;
1441         device->fs_devices->total_devices--;
1442
1443         if (device->missing)
1444                 root->fs_info->fs_devices->missing_devices--;
1445
1446         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1447                                  struct btrfs_device, dev_list);
1448         if (device->bdev == root->fs_info->sb->s_bdev)
1449                 root->fs_info->sb->s_bdev = next_device->bdev;
1450         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1451                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1452
1453         if (device->bdev)
1454                 device->fs_devices->open_devices--;
1455
1456         call_rcu(&device->rcu, free_device);
1457         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1458
1459         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1460         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1461
1462         if (cur_devices->open_devices == 0) {
1463                 struct btrfs_fs_devices *fs_devices;
1464                 fs_devices = root->fs_info->fs_devices;
1465                 while (fs_devices) {
1466                         if (fs_devices->seed == cur_devices)
1467                                 break;
1468                         fs_devices = fs_devices->seed;
1469                 }
1470                 fs_devices->seed = cur_devices->seed;
1471                 cur_devices->seed = NULL;
1472                 lock_chunks(root);
1473                 __btrfs_close_devices(cur_devices);
1474                 unlock_chunks(root);
1475                 free_fs_devices(cur_devices);
1476         }
1477
1478         /*
1479          * at this point, the device is zero sized.  We want to
1480          * remove it from the devices list and zero out the old super
1481          */
1482         if (clear_super) {
1483                 /* make sure this device isn't detected as part of
1484                  * the FS anymore
1485                  */
1486                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1487                 set_buffer_dirty(bh);
1488                 sync_dirty_buffer(bh);
1489         }
1490
1491         ret = 0;
1492
1493 error_brelse:
1494         brelse(bh);
1495 error_close:
1496         if (bdev)
1497                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1498 out:
1499         mutex_unlock(&uuid_mutex);
1500         return ret;
1501 error_undo:
1502         if (device->writeable) {
1503                 lock_chunks(root);
1504                 list_add(&device->dev_alloc_list,
1505                          &root->fs_info->fs_devices->alloc_list);
1506                 unlock_chunks(root);
1507                 root->fs_info->fs_devices->rw_devices++;
1508         }
1509         goto error_brelse;
1510 }
1511
1512 /*
1513  * does all the dirty work required for changing file system's UUID.
1514  */
1515 static int btrfs_prepare_sprout(struct btrfs_root *root)
1516 {
1517         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1518         struct btrfs_fs_devices *old_devices;
1519         struct btrfs_fs_devices *seed_devices;
1520         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1521         struct btrfs_device *device;
1522         u64 super_flags;
1523
1524         BUG_ON(!mutex_is_locked(&uuid_mutex));
1525         if (!fs_devices->seeding)
1526                 return -EINVAL;
1527
1528         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1529         if (!seed_devices)
1530                 return -ENOMEM;
1531
1532         old_devices = clone_fs_devices(fs_devices);
1533         if (IS_ERR(old_devices)) {
1534                 kfree(seed_devices);
1535                 return PTR_ERR(old_devices);
1536         }
1537
1538         list_add(&old_devices->list, &fs_uuids);
1539
1540         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1541         seed_devices->opened = 1;
1542         INIT_LIST_HEAD(&seed_devices->devices);
1543         INIT_LIST_HEAD(&seed_devices->alloc_list);
1544         mutex_init(&seed_devices->device_list_mutex);
1545
1546         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1547         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1548                               synchronize_rcu);
1549         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1550
1551         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1552         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1553                 device->fs_devices = seed_devices;
1554         }
1555
1556         fs_devices->seeding = 0;
1557         fs_devices->num_devices = 0;
1558         fs_devices->open_devices = 0;
1559         fs_devices->total_devices = 0;
1560         fs_devices->seed = seed_devices;
1561
1562         generate_random_uuid(fs_devices->fsid);
1563         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1564         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1565         super_flags = btrfs_super_flags(disk_super) &
1566                       ~BTRFS_SUPER_FLAG_SEEDING;
1567         btrfs_set_super_flags(disk_super, super_flags);
1568
1569         return 0;
1570 }
1571
1572 /*
1573  * strore the expected generation for seed devices in device items.
1574  */
1575 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1576                                struct btrfs_root *root)
1577 {
1578         struct btrfs_path *path;
1579         struct extent_buffer *leaf;
1580         struct btrfs_dev_item *dev_item;
1581         struct btrfs_device *device;
1582         struct btrfs_key key;
1583         u8 fs_uuid[BTRFS_UUID_SIZE];
1584         u8 dev_uuid[BTRFS_UUID_SIZE];
1585         u64 devid;
1586         int ret;
1587
1588         path = btrfs_alloc_path();
1589         if (!path)
1590                 return -ENOMEM;
1591
1592         root = root->fs_info->chunk_root;
1593         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1594         key.offset = 0;
1595         key.type = BTRFS_DEV_ITEM_KEY;
1596
1597         while (1) {
1598                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1599                 if (ret < 0)
1600                         goto error;
1601
1602                 leaf = path->nodes[0];
1603 next_slot:
1604                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1605                         ret = btrfs_next_leaf(root, path);
1606                         if (ret > 0)
1607                                 break;
1608                         if (ret < 0)
1609                                 goto error;
1610                         leaf = path->nodes[0];
1611                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1612                         btrfs_release_path(path);
1613                         continue;
1614                 }
1615
1616                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1617                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1618                     key.type != BTRFS_DEV_ITEM_KEY)
1619                         break;
1620
1621                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1622                                           struct btrfs_dev_item);
1623                 devid = btrfs_device_id(leaf, dev_item);
1624                 read_extent_buffer(leaf, dev_uuid,
1625                                    (unsigned long)btrfs_device_uuid(dev_item),
1626                                    BTRFS_UUID_SIZE);
1627                 read_extent_buffer(leaf, fs_uuid,
1628                                    (unsigned long)btrfs_device_fsid(dev_item),
1629                                    BTRFS_UUID_SIZE);
1630                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1631                 BUG_ON(!device); /* Logic error */
1632
1633                 if (device->fs_devices->seeding) {
1634                         btrfs_set_device_generation(leaf, dev_item,
1635                                                     device->generation);
1636                         btrfs_mark_buffer_dirty(leaf);
1637                 }
1638
1639                 path->slots[0]++;
1640                 goto next_slot;
1641         }
1642         ret = 0;
1643 error:
1644         btrfs_free_path(path);
1645         return ret;
1646 }
1647
1648 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1649 {
1650         struct request_queue *q;
1651         struct btrfs_trans_handle *trans;
1652         struct btrfs_device *device;
1653         struct block_device *bdev;
1654         struct list_head *devices;
1655         struct super_block *sb = root->fs_info->sb;
1656         struct rcu_string *name;
1657         u64 total_bytes;
1658         int seeding_dev = 0;
1659         int ret = 0;
1660
1661         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1662                 return -EROFS;
1663
1664         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1665                                   root->fs_info->bdev_holder);
1666         if (IS_ERR(bdev))
1667                 return PTR_ERR(bdev);
1668
1669         if (root->fs_info->fs_devices->seeding) {
1670                 seeding_dev = 1;
1671                 down_write(&sb->s_umount);
1672                 mutex_lock(&uuid_mutex);
1673         }
1674
1675         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1676
1677         devices = &root->fs_info->fs_devices->devices;
1678         /*
1679          * we have the volume lock, so we don't need the extra
1680          * device list mutex while reading the list here.
1681          */
1682         list_for_each_entry(device, devices, dev_list) {
1683                 if (device->bdev == bdev) {
1684                         ret = -EEXIST;
1685                         goto error;
1686                 }
1687         }
1688
1689         device = kzalloc(sizeof(*device), GFP_NOFS);
1690         if (!device) {
1691                 /* we can safely leave the fs_devices entry around */
1692                 ret = -ENOMEM;
1693                 goto error;
1694         }
1695
1696         name = rcu_string_strdup(device_path, GFP_NOFS);
1697         if (!name) {
1698                 kfree(device);
1699                 ret = -ENOMEM;
1700                 goto error;
1701         }
1702         rcu_assign_pointer(device->name, name);
1703
1704         ret = find_next_devid(root, &device->devid);
1705         if (ret) {
1706                 rcu_string_free(device->name);
1707                 kfree(device);
1708                 goto error;
1709         }
1710
1711         trans = btrfs_start_transaction(root, 0);
1712         if (IS_ERR(trans)) {
1713                 rcu_string_free(device->name);
1714                 kfree(device);
1715                 ret = PTR_ERR(trans);
1716                 goto error;
1717         }
1718
1719         lock_chunks(root);
1720
1721         q = bdev_get_queue(bdev);
1722         if (blk_queue_discard(q))
1723                 device->can_discard = 1;
1724         device->writeable = 1;
1725         device->work.func = pending_bios_fn;
1726         generate_random_uuid(device->uuid);
1727         spin_lock_init(&device->io_lock);
1728         device->generation = trans->transid;
1729         device->io_width = root->sectorsize;
1730         device->io_align = root->sectorsize;
1731         device->sector_size = root->sectorsize;
1732         device->total_bytes = i_size_read(bdev->bd_inode);
1733         device->disk_total_bytes = device->total_bytes;
1734         device->dev_root = root->fs_info->dev_root;
1735         device->bdev = bdev;
1736         device->in_fs_metadata = 1;
1737         device->mode = FMODE_EXCL;
1738         set_blocksize(device->bdev, 4096);
1739
1740         if (seeding_dev) {
1741                 sb->s_flags &= ~MS_RDONLY;
1742                 ret = btrfs_prepare_sprout(root);
1743                 BUG_ON(ret); /* -ENOMEM */
1744         }
1745
1746         device->fs_devices = root->fs_info->fs_devices;
1747
1748         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1749         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1750         list_add(&device->dev_alloc_list,
1751                  &root->fs_info->fs_devices->alloc_list);
1752         root->fs_info->fs_devices->num_devices++;
1753         root->fs_info->fs_devices->open_devices++;
1754         root->fs_info->fs_devices->rw_devices++;
1755         root->fs_info->fs_devices->total_devices++;
1756         if (device->can_discard)
1757                 root->fs_info->fs_devices->num_can_discard++;
1758         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1759
1760         spin_lock(&root->fs_info->free_chunk_lock);
1761         root->fs_info->free_chunk_space += device->total_bytes;
1762         spin_unlock(&root->fs_info->free_chunk_lock);
1763
1764         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1765                 root->fs_info->fs_devices->rotating = 1;
1766
1767         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1768         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1769                                     total_bytes + device->total_bytes);
1770
1771         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1772         btrfs_set_super_num_devices(root->fs_info->super_copy,
1773                                     total_bytes + 1);
1774         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1775
1776         if (seeding_dev) {
1777                 ret = init_first_rw_device(trans, root, device);
1778                 if (ret)
1779                         goto error_trans;
1780                 ret = btrfs_finish_sprout(trans, root);
1781                 if (ret)
1782                         goto error_trans;
1783         } else {
1784                 ret = btrfs_add_device(trans, root, device);
1785                 if (ret)
1786                         goto error_trans;
1787         }
1788
1789         /*
1790          * we've got more storage, clear any full flags on the space
1791          * infos
1792          */
1793         btrfs_clear_space_info_full(root->fs_info);
1794
1795         unlock_chunks(root);
1796         ret = btrfs_commit_transaction(trans, root);
1797
1798         if (seeding_dev) {
1799                 mutex_unlock(&uuid_mutex);
1800                 up_write(&sb->s_umount);
1801
1802                 if (ret) /* transaction commit */
1803                         return ret;
1804
1805                 ret = btrfs_relocate_sys_chunks(root);
1806                 if (ret < 0)
1807                         btrfs_error(root->fs_info, ret,
1808                                     "Failed to relocate sys chunks after "
1809                                     "device initialization. This can be fixed "
1810                                     "using the \"btrfs balance\" command.");
1811         }
1812
1813         return ret;
1814
1815 error_trans:
1816         unlock_chunks(root);
1817         btrfs_abort_transaction(trans, root, ret);
1818         btrfs_end_transaction(trans, root);
1819         rcu_string_free(device->name);
1820         kfree(device);
1821 error:
1822         blkdev_put(bdev, FMODE_EXCL);
1823         if (seeding_dev) {
1824                 mutex_unlock(&uuid_mutex);
1825                 up_write(&sb->s_umount);
1826         }
1827         return ret;
1828 }
1829
1830 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1831                                         struct btrfs_device *device)
1832 {
1833         int ret;
1834         struct btrfs_path *path;
1835         struct btrfs_root *root;
1836         struct btrfs_dev_item *dev_item;
1837         struct extent_buffer *leaf;
1838         struct btrfs_key key;
1839
1840         root = device->dev_root->fs_info->chunk_root;
1841
1842         path = btrfs_alloc_path();
1843         if (!path)
1844                 return -ENOMEM;
1845
1846         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1847         key.type = BTRFS_DEV_ITEM_KEY;
1848         key.offset = device->devid;
1849
1850         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1851         if (ret < 0)
1852                 goto out;
1853
1854         if (ret > 0) {
1855                 ret = -ENOENT;
1856                 goto out;
1857         }
1858
1859         leaf = path->nodes[0];
1860         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1861
1862         btrfs_set_device_id(leaf, dev_item, device->devid);
1863         btrfs_set_device_type(leaf, dev_item, device->type);
1864         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1865         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1866         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1867         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1868         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1869         btrfs_mark_buffer_dirty(leaf);
1870
1871 out:
1872         btrfs_free_path(path);
1873         return ret;
1874 }
1875
1876 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1877                       struct btrfs_device *device, u64 new_size)
1878 {
1879         struct btrfs_super_block *super_copy =
1880                 device->dev_root->fs_info->super_copy;
1881         u64 old_total = btrfs_super_total_bytes(super_copy);
1882         u64 diff = new_size - device->total_bytes;
1883
1884         if (!device->writeable)
1885                 return -EACCES;
1886         if (new_size <= device->total_bytes)
1887                 return -EINVAL;
1888
1889         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1890         device->fs_devices->total_rw_bytes += diff;
1891
1892         device->total_bytes = new_size;
1893         device->disk_total_bytes = new_size;
1894         btrfs_clear_space_info_full(device->dev_root->fs_info);
1895
1896         return btrfs_update_device(trans, device);
1897 }
1898
1899 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1900                       struct btrfs_device *device, u64 new_size)
1901 {
1902         int ret;
1903         lock_chunks(device->dev_root);
1904         ret = __btrfs_grow_device(trans, device, new_size);
1905         unlock_chunks(device->dev_root);
1906         return ret;
1907 }
1908
1909 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1910                             struct btrfs_root *root,
1911                             u64 chunk_tree, u64 chunk_objectid,
1912                             u64 chunk_offset)
1913 {
1914         int ret;
1915         struct btrfs_path *path;
1916         struct btrfs_key key;
1917
1918         root = root->fs_info->chunk_root;
1919         path = btrfs_alloc_path();
1920         if (!path)
1921                 return -ENOMEM;
1922
1923         key.objectid = chunk_objectid;
1924         key.offset = chunk_offset;
1925         key.type = BTRFS_CHUNK_ITEM_KEY;
1926
1927         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1928         if (ret < 0)
1929                 goto out;
1930         else if (ret > 0) { /* Logic error or corruption */
1931                 btrfs_error(root->fs_info, -ENOENT,
1932                             "Failed lookup while freeing chunk.");
1933                 ret = -ENOENT;
1934                 goto out;
1935         }
1936
1937         ret = btrfs_del_item(trans, root, path);
1938         if (ret < 0)
1939                 btrfs_error(root->fs_info, ret,
1940                             "Failed to delete chunk item.");
1941 out:
1942         btrfs_free_path(path);
1943         return ret;
1944 }
1945
1946 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1947                         chunk_offset)
1948 {
1949         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1950         struct btrfs_disk_key *disk_key;
1951         struct btrfs_chunk *chunk;
1952         u8 *ptr;
1953         int ret = 0;
1954         u32 num_stripes;
1955         u32 array_size;
1956         u32 len = 0;
1957         u32 cur;
1958         struct btrfs_key key;
1959
1960         array_size = btrfs_super_sys_array_size(super_copy);
1961
1962         ptr = super_copy->sys_chunk_array;
1963         cur = 0;
1964
1965         while (cur < array_size) {
1966                 disk_key = (struct btrfs_disk_key *)ptr;
1967                 btrfs_disk_key_to_cpu(&key, disk_key);
1968
1969                 len = sizeof(*disk_key);
1970
1971                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1972                         chunk = (struct btrfs_chunk *)(ptr + len);
1973                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1974                         len += btrfs_chunk_item_size(num_stripes);
1975                 } else {
1976                         ret = -EIO;
1977                         break;
1978                 }
1979                 if (key.objectid == chunk_objectid &&
1980                     key.offset == chunk_offset) {
1981                         memmove(ptr, ptr + len, array_size - (cur + len));
1982                         array_size -= len;
1983                         btrfs_set_super_sys_array_size(super_copy, array_size);
1984                 } else {
1985                         ptr += len;
1986                         cur += len;
1987                 }
1988         }
1989         return ret;
1990 }
1991
1992 static int btrfs_relocate_chunk(struct btrfs_root *root,
1993                          u64 chunk_tree, u64 chunk_objectid,
1994                          u64 chunk_offset)
1995 {
1996         struct extent_map_tree *em_tree;
1997         struct btrfs_root *extent_root;
1998         struct btrfs_trans_handle *trans;
1999         struct extent_map *em;
2000         struct map_lookup *map;
2001         int ret;
2002         int i;
2003
2004         root = root->fs_info->chunk_root;
2005         extent_root = root->fs_info->extent_root;
2006         em_tree = &root->fs_info->mapping_tree.map_tree;
2007
2008         ret = btrfs_can_relocate(extent_root, chunk_offset);
2009         if (ret)
2010                 return -ENOSPC;
2011
2012         /* step one, relocate all the extents inside this chunk */
2013         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2014         if (ret)
2015                 return ret;
2016
2017         trans = btrfs_start_transaction(root, 0);
2018         BUG_ON(IS_ERR(trans));
2019
2020         lock_chunks(root);
2021
2022         /*
2023          * step two, delete the device extents and the
2024          * chunk tree entries
2025          */
2026         read_lock(&em_tree->lock);
2027         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2028         read_unlock(&em_tree->lock);
2029
2030         BUG_ON(!em || em->start > chunk_offset ||
2031                em->start + em->len < chunk_offset);
2032         map = (struct map_lookup *)em->bdev;
2033
2034         for (i = 0; i < map->num_stripes; i++) {
2035                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2036                                             map->stripes[i].physical);
2037                 BUG_ON(ret);
2038
2039                 if (map->stripes[i].dev) {
2040                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2041                         BUG_ON(ret);
2042                 }
2043         }
2044         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2045                                chunk_offset);
2046
2047         BUG_ON(ret);
2048
2049         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2050
2051         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2052                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2053                 BUG_ON(ret);
2054         }
2055
2056         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2057         BUG_ON(ret);
2058
2059         write_lock(&em_tree->lock);
2060         remove_extent_mapping(em_tree, em);
2061         write_unlock(&em_tree->lock);
2062
2063         kfree(map);
2064         em->bdev = NULL;
2065
2066         /* once for the tree */
2067         free_extent_map(em);
2068         /* once for us */
2069         free_extent_map(em);
2070
2071         unlock_chunks(root);
2072         btrfs_end_transaction(trans, root);
2073         return 0;
2074 }
2075
2076 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2077 {
2078         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2079         struct btrfs_path *path;
2080         struct extent_buffer *leaf;
2081         struct btrfs_chunk *chunk;
2082         struct btrfs_key key;
2083         struct btrfs_key found_key;
2084         u64 chunk_tree = chunk_root->root_key.objectid;
2085         u64 chunk_type;
2086         bool retried = false;
2087         int failed = 0;
2088         int ret;
2089
2090         path = btrfs_alloc_path();
2091         if (!path)
2092                 return -ENOMEM;
2093
2094 again:
2095         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2096         key.offset = (u64)-1;
2097         key.type = BTRFS_CHUNK_ITEM_KEY;
2098
2099         while (1) {
2100                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2101                 if (ret < 0)
2102                         goto error;
2103                 BUG_ON(ret == 0); /* Corruption */
2104
2105                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2106                                           key.type);
2107                 if (ret < 0)
2108                         goto error;
2109                 if (ret > 0)
2110                         break;
2111
2112                 leaf = path->nodes[0];
2113                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2114
2115                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2116                                        struct btrfs_chunk);
2117                 chunk_type = btrfs_chunk_type(leaf, chunk);
2118                 btrfs_release_path(path);
2119
2120                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2121                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2122                                                    found_key.objectid,
2123                                                    found_key.offset);
2124                         if (ret == -ENOSPC)
2125                                 failed++;
2126                         else if (ret)
2127                                 BUG();
2128                 }
2129
2130                 if (found_key.offset == 0)
2131                         break;
2132                 key.offset = found_key.offset - 1;
2133         }
2134         ret = 0;
2135         if (failed && !retried) {
2136                 failed = 0;
2137                 retried = true;
2138                 goto again;
2139         } else if (failed && retried) {
2140                 WARN_ON(1);
2141                 ret = -ENOSPC;
2142         }
2143 error:
2144         btrfs_free_path(path);
2145         return ret;
2146 }
2147
2148 static int insert_balance_item(struct btrfs_root *root,
2149                                struct btrfs_balance_control *bctl)
2150 {
2151         struct btrfs_trans_handle *trans;
2152         struct btrfs_balance_item *item;
2153         struct btrfs_disk_balance_args disk_bargs;
2154         struct btrfs_path *path;
2155         struct extent_buffer *leaf;
2156         struct btrfs_key key;
2157         int ret, err;
2158
2159         path = btrfs_alloc_path();
2160         if (!path)
2161                 return -ENOMEM;
2162
2163         trans = btrfs_start_transaction(root, 0);
2164         if (IS_ERR(trans)) {
2165                 btrfs_free_path(path);
2166                 return PTR_ERR(trans);
2167         }
2168
2169         key.objectid = BTRFS_BALANCE_OBJECTID;
2170         key.type = BTRFS_BALANCE_ITEM_KEY;
2171         key.offset = 0;
2172
2173         ret = btrfs_insert_empty_item(trans, root, path, &key,
2174                                       sizeof(*item));
2175         if (ret)
2176                 goto out;
2177
2178         leaf = path->nodes[0];
2179         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2180
2181         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2182
2183         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2184         btrfs_set_balance_data(leaf, item, &disk_bargs);
2185         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2186         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2187         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2188         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2189
2190         btrfs_set_balance_flags(leaf, item, bctl->flags);
2191
2192         btrfs_mark_buffer_dirty(leaf);
2193 out:
2194         btrfs_free_path(path);
2195         err = btrfs_commit_transaction(trans, root);
2196         if (err && !ret)
2197                 ret = err;
2198         return ret;
2199 }
2200
2201 static int del_balance_item(struct btrfs_root *root)
2202 {
2203         struct btrfs_trans_handle *trans;
2204         struct btrfs_path *path;
2205         struct btrfs_key key;
2206         int ret, err;
2207
2208         path = btrfs_alloc_path();
2209         if (!path)
2210                 return -ENOMEM;
2211
2212         trans = btrfs_start_transaction(root, 0);
2213         if (IS_ERR(trans)) {
2214                 btrfs_free_path(path);
2215                 return PTR_ERR(trans);
2216         }
2217
2218         key.objectid = BTRFS_BALANCE_OBJECTID;
2219         key.type = BTRFS_BALANCE_ITEM_KEY;
2220         key.offset = 0;
2221
2222         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2223         if (ret < 0)
2224                 goto out;
2225         if (ret > 0) {
2226                 ret = -ENOENT;
2227                 goto out;
2228         }
2229
2230         ret = btrfs_del_item(trans, root, path);
2231 out:
2232         btrfs_free_path(path);
2233         err = btrfs_commit_transaction(trans, root);
2234         if (err && !ret)
2235                 ret = err;
2236         return ret;
2237 }
2238
2239 /*
2240  * This is a heuristic used to reduce the number of chunks balanced on
2241  * resume after balance was interrupted.
2242  */
2243 static void update_balance_args(struct btrfs_balance_control *bctl)
2244 {
2245         /*
2246          * Turn on soft mode for chunk types that were being converted.
2247          */
2248         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2249                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2250         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2251                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2252         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2253                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2254
2255         /*
2256          * Turn on usage filter if is not already used.  The idea is
2257          * that chunks that we have already balanced should be
2258          * reasonably full.  Don't do it for chunks that are being
2259          * converted - that will keep us from relocating unconverted
2260          * (albeit full) chunks.
2261          */
2262         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2263             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2264                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2265                 bctl->data.usage = 90;
2266         }
2267         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2268             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2269                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2270                 bctl->sys.usage = 90;
2271         }
2272         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2273             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2274                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2275                 bctl->meta.usage = 90;
2276         }
2277 }
2278
2279 /*
2280  * Should be called with both balance and volume mutexes held to
2281  * serialize other volume operations (add_dev/rm_dev/resize) with
2282  * restriper.  Same goes for unset_balance_control.
2283  */
2284 static void set_balance_control(struct btrfs_balance_control *bctl)
2285 {
2286         struct btrfs_fs_info *fs_info = bctl->fs_info;
2287
2288         BUG_ON(fs_info->balance_ctl);
2289
2290         spin_lock(&fs_info->balance_lock);
2291         fs_info->balance_ctl = bctl;
2292         spin_unlock(&fs_info->balance_lock);
2293 }
2294
2295 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2296 {
2297         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2298
2299         BUG_ON(!fs_info->balance_ctl);
2300
2301         spin_lock(&fs_info->balance_lock);
2302         fs_info->balance_ctl = NULL;
2303         spin_unlock(&fs_info->balance_lock);
2304
2305         kfree(bctl);
2306 }
2307
2308 /*
2309  * Balance filters.  Return 1 if chunk should be filtered out
2310  * (should not be balanced).
2311  */
2312 static int chunk_profiles_filter(u64 chunk_type,
2313                                  struct btrfs_balance_args *bargs)
2314 {
2315         chunk_type = chunk_to_extended(chunk_type) &
2316                                 BTRFS_EXTENDED_PROFILE_MASK;
2317
2318         if (bargs->profiles & chunk_type)
2319                 return 0;
2320
2321         return 1;
2322 }
2323
2324 static u64 div_factor_fine(u64 num, int factor)
2325 {
2326         if (factor <= 0)
2327                 return 0;
2328         if (factor >= 100)
2329                 return num;
2330
2331         num *= factor;
2332         do_div(num, 100);
2333         return num;
2334 }
2335
2336 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2337                               struct btrfs_balance_args *bargs)
2338 {
2339         struct btrfs_block_group_cache *cache;
2340         u64 chunk_used, user_thresh;
2341         int ret = 1;
2342
2343         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2344         chunk_used = btrfs_block_group_used(&cache->item);
2345
2346         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2347         if (chunk_used < user_thresh)
2348                 ret = 0;
2349
2350         btrfs_put_block_group(cache);
2351         return ret;
2352 }
2353
2354 static int chunk_devid_filter(struct extent_buffer *leaf,
2355                               struct btrfs_chunk *chunk,
2356                               struct btrfs_balance_args *bargs)
2357 {
2358         struct btrfs_stripe *stripe;
2359         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2360         int i;
2361
2362         for (i = 0; i < num_stripes; i++) {
2363                 stripe = btrfs_stripe_nr(chunk, i);
2364                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2365                         return 0;
2366         }
2367
2368         return 1;
2369 }
2370
2371 /* [pstart, pend) */
2372 static int chunk_drange_filter(struct extent_buffer *leaf,
2373                                struct btrfs_chunk *chunk,
2374                                u64 chunk_offset,
2375                                struct btrfs_balance_args *bargs)
2376 {
2377         struct btrfs_stripe *stripe;
2378         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2379         u64 stripe_offset;
2380         u64 stripe_length;
2381         int factor;
2382         int i;
2383
2384         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2385                 return 0;
2386
2387         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2388              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2389                 factor = 2;
2390         else
2391                 factor = 1;
2392         factor = num_stripes / factor;
2393
2394         for (i = 0; i < num_stripes; i++) {
2395                 stripe = btrfs_stripe_nr(chunk, i);
2396                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2397                         continue;
2398
2399                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2400                 stripe_length = btrfs_chunk_length(leaf, chunk);
2401                 do_div(stripe_length, factor);
2402
2403                 if (stripe_offset < bargs->pend &&
2404                     stripe_offset + stripe_length > bargs->pstart)
2405                         return 0;
2406         }
2407
2408         return 1;
2409 }
2410
2411 /* [vstart, vend) */
2412 static int chunk_vrange_filter(struct extent_buffer *leaf,
2413                                struct btrfs_chunk *chunk,
2414                                u64 chunk_offset,
2415                                struct btrfs_balance_args *bargs)
2416 {
2417         if (chunk_offset < bargs->vend &&
2418             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2419                 /* at least part of the chunk is inside this vrange */
2420                 return 0;
2421
2422         return 1;
2423 }
2424
2425 static int chunk_soft_convert_filter(u64 chunk_type,
2426                                      struct btrfs_balance_args *bargs)
2427 {
2428         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2429                 return 0;
2430
2431         chunk_type = chunk_to_extended(chunk_type) &
2432                                 BTRFS_EXTENDED_PROFILE_MASK;
2433
2434         if (bargs->target == chunk_type)
2435                 return 1;
2436
2437         return 0;
2438 }
2439
2440 static int should_balance_chunk(struct btrfs_root *root,
2441                                 struct extent_buffer *leaf,
2442                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2443 {
2444         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2445         struct btrfs_balance_args *bargs = NULL;
2446         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2447
2448         /* type filter */
2449         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2450               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2451                 return 0;
2452         }
2453
2454         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2455                 bargs = &bctl->data;
2456         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2457                 bargs = &bctl->sys;
2458         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2459                 bargs = &bctl->meta;
2460
2461         /* profiles filter */
2462         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2463             chunk_profiles_filter(chunk_type, bargs)) {
2464                 return 0;
2465         }
2466
2467         /* usage filter */
2468         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2469             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2470                 return 0;
2471         }
2472
2473         /* devid filter */
2474         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2475             chunk_devid_filter(leaf, chunk, bargs)) {
2476                 return 0;
2477         }
2478
2479         /* drange filter, makes sense only with devid filter */
2480         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2481             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2482                 return 0;
2483         }
2484
2485         /* vrange filter */
2486         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2487             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2488                 return 0;
2489         }
2490
2491         /* soft profile changing mode */
2492         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2493             chunk_soft_convert_filter(chunk_type, bargs)) {
2494                 return 0;
2495         }
2496
2497         return 1;
2498 }
2499
2500 static u64 div_factor(u64 num, int factor)
2501 {
2502         if (factor == 10)
2503                 return num;
2504         num *= factor;
2505         do_div(num, 10);
2506         return num;
2507 }
2508
2509 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2510 {
2511         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2512         struct btrfs_root *chunk_root = fs_info->chunk_root;
2513         struct btrfs_root *dev_root = fs_info->dev_root;
2514         struct list_head *devices;
2515         struct btrfs_device *device;
2516         u64 old_size;
2517         u64 size_to_free;
2518         struct btrfs_chunk *chunk;
2519         struct btrfs_path *path;
2520         struct btrfs_key key;
2521         struct btrfs_key found_key;
2522         struct btrfs_trans_handle *trans;
2523         struct extent_buffer *leaf;
2524         int slot;
2525         int ret;
2526         int enospc_errors = 0;
2527         bool counting = true;
2528
2529         /* step one make some room on all the devices */
2530         devices = &fs_info->fs_devices->devices;
2531         list_for_each_entry(device, devices, dev_list) {
2532                 old_size = device->total_bytes;
2533                 size_to_free = div_factor(old_size, 1);
2534                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2535                 if (!device->writeable ||
2536                     device->total_bytes - device->bytes_used > size_to_free)
2537                         continue;
2538
2539                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2540                 if (ret == -ENOSPC)
2541                         break;
2542                 BUG_ON(ret);
2543
2544                 trans = btrfs_start_transaction(dev_root, 0);
2545                 BUG_ON(IS_ERR(trans));
2546
2547                 ret = btrfs_grow_device(trans, device, old_size);
2548                 BUG_ON(ret);
2549
2550                 btrfs_end_transaction(trans, dev_root);
2551         }
2552
2553         /* step two, relocate all the chunks */
2554         path = btrfs_alloc_path();
2555         if (!path) {
2556                 ret = -ENOMEM;
2557                 goto error;
2558         }
2559
2560         /* zero out stat counters */
2561         spin_lock(&fs_info->balance_lock);
2562         memset(&bctl->stat, 0, sizeof(bctl->stat));
2563         spin_unlock(&fs_info->balance_lock);
2564 again:
2565         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2566         key.offset = (u64)-1;
2567         key.type = BTRFS_CHUNK_ITEM_KEY;
2568
2569         while (1) {
2570                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2571                     atomic_read(&fs_info->balance_cancel_req)) {
2572                         ret = -ECANCELED;
2573                         goto error;
2574                 }
2575
2576                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2577                 if (ret < 0)
2578                         goto error;
2579
2580                 /*
2581                  * this shouldn't happen, it means the last relocate
2582                  * failed
2583                  */
2584                 if (ret == 0)
2585                         BUG(); /* FIXME break ? */
2586
2587                 ret = btrfs_previous_item(chunk_root, path, 0,
2588                                           BTRFS_CHUNK_ITEM_KEY);
2589                 if (ret) {
2590                         ret = 0;
2591                         break;
2592                 }
2593
2594                 leaf = path->nodes[0];
2595                 slot = path->slots[0];
2596                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2597
2598                 if (found_key.objectid != key.objectid)
2599                         break;
2600
2601                 /* chunk zero is special */
2602                 if (found_key.offset == 0)
2603                         break;
2604
2605                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2606
2607                 if (!counting) {
2608                         spin_lock(&fs_info->balance_lock);
2609                         bctl->stat.considered++;
2610                         spin_unlock(&fs_info->balance_lock);
2611                 }
2612
2613                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2614                                            found_key.offset);
2615                 btrfs_release_path(path);
2616                 if (!ret)
2617                         goto loop;
2618
2619                 if (counting) {
2620                         spin_lock(&fs_info->balance_lock);
2621                         bctl->stat.expected++;
2622                         spin_unlock(&fs_info->balance_lock);
2623                         goto loop;
2624                 }
2625
2626                 ret = btrfs_relocate_chunk(chunk_root,
2627                                            chunk_root->root_key.objectid,
2628                                            found_key.objectid,
2629                                            found_key.offset);
2630                 if (ret && ret != -ENOSPC)
2631                         goto error;
2632                 if (ret == -ENOSPC) {
2633                         enospc_errors++;
2634                 } else {
2635                         spin_lock(&fs_info->balance_lock);
2636                         bctl->stat.completed++;
2637                         spin_unlock(&fs_info->balance_lock);
2638                 }
2639 loop:
2640                 key.offset = found_key.offset - 1;
2641         }
2642
2643         if (counting) {
2644                 btrfs_release_path(path);
2645                 counting = false;
2646                 goto again;
2647         }
2648 error:
2649         btrfs_free_path(path);
2650         if (enospc_errors) {
2651                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2652                        enospc_errors);
2653                 if (!ret)
2654                         ret = -ENOSPC;
2655         }
2656
2657         return ret;
2658 }
2659
2660 /**
2661  * alloc_profile_is_valid - see if a given profile is valid and reduced
2662  * @flags: profile to validate
2663  * @extended: if true @flags is treated as an extended profile
2664  */
2665 static int alloc_profile_is_valid(u64 flags, int extended)
2666 {
2667         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2668                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2669
2670         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2671
2672         /* 1) check that all other bits are zeroed */
2673         if (flags & ~mask)
2674                 return 0;
2675
2676         /* 2) see if profile is reduced */
2677         if (flags == 0)
2678                 return !extended; /* "0" is valid for usual profiles */
2679
2680         /* true if exactly one bit set */
2681         return (flags & (flags - 1)) == 0;
2682 }
2683
2684 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2685 {
2686         /* cancel requested || normal exit path */
2687         return atomic_read(&fs_info->balance_cancel_req) ||
2688                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2689                  atomic_read(&fs_info->balance_cancel_req) == 0);
2690 }
2691
2692 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2693 {
2694         int ret;
2695
2696         unset_balance_control(fs_info);
2697         ret = del_balance_item(fs_info->tree_root);
2698         BUG_ON(ret);
2699 }
2700
2701 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2702                                struct btrfs_ioctl_balance_args *bargs);
2703
2704 /*
2705  * Should be called with both balance and volume mutexes held
2706  */
2707 int btrfs_balance(struct btrfs_balance_control *bctl,
2708                   struct btrfs_ioctl_balance_args *bargs)
2709 {
2710         struct btrfs_fs_info *fs_info = bctl->fs_info;
2711         u64 allowed;
2712         int mixed = 0;
2713         int ret;
2714
2715         if (btrfs_fs_closing(fs_info) ||
2716             atomic_read(&fs_info->balance_pause_req) ||
2717             atomic_read(&fs_info->balance_cancel_req)) {
2718                 ret = -EINVAL;
2719                 goto out;
2720         }
2721
2722         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2723         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2724                 mixed = 1;
2725
2726         /*
2727          * In case of mixed groups both data and meta should be picked,
2728          * and identical options should be given for both of them.
2729          */
2730         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2731         if (mixed && (bctl->flags & allowed)) {
2732                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2733                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2734                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2735                         printk(KERN_ERR "btrfs: with mixed groups data and "
2736                                "metadata balance options must be the same\n");
2737                         ret = -EINVAL;
2738                         goto out;
2739                 }
2740         }
2741
2742         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2743         if (fs_info->fs_devices->num_devices == 1)
2744                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2745         else if (fs_info->fs_devices->num_devices < 4)
2746                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2747         else
2748                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2749                                 BTRFS_BLOCK_GROUP_RAID10);
2750
2751         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2752             (!alloc_profile_is_valid(bctl->data.target, 1) ||
2753              (bctl->data.target & ~allowed))) {
2754                 printk(KERN_ERR "btrfs: unable to start balance with target "
2755                        "data profile %llu\n",
2756                        (unsigned long long)bctl->data.target);
2757                 ret = -EINVAL;
2758                 goto out;
2759         }
2760         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2761             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2762              (bctl->meta.target & ~allowed))) {
2763                 printk(KERN_ERR "btrfs: unable to start balance with target "
2764                        "metadata profile %llu\n",
2765                        (unsigned long long)bctl->meta.target);
2766                 ret = -EINVAL;
2767                 goto out;
2768         }
2769         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2770             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2771              (bctl->sys.target & ~allowed))) {
2772                 printk(KERN_ERR "btrfs: unable to start balance with target "
2773                        "system profile %llu\n",
2774                        (unsigned long long)bctl->sys.target);
2775                 ret = -EINVAL;
2776                 goto out;
2777         }
2778
2779         /* allow dup'ed data chunks only in mixed mode */
2780         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2781             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2782                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2783                 ret = -EINVAL;
2784                 goto out;
2785         }
2786
2787         /* allow to reduce meta or sys integrity only if force set */
2788         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2789                         BTRFS_BLOCK_GROUP_RAID10;
2790         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2791              (fs_info->avail_system_alloc_bits & allowed) &&
2792              !(bctl->sys.target & allowed)) ||
2793             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2794              (fs_info->avail_metadata_alloc_bits & allowed) &&
2795              !(bctl->meta.target & allowed))) {
2796                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2797                         printk(KERN_INFO "btrfs: force reducing metadata "
2798                                "integrity\n");
2799                 } else {
2800                         printk(KERN_ERR "btrfs: balance will reduce metadata "
2801                                "integrity, use force if you want this\n");
2802                         ret = -EINVAL;
2803                         goto out;
2804                 }
2805         }
2806
2807         ret = insert_balance_item(fs_info->tree_root, bctl);
2808         if (ret && ret != -EEXIST)
2809                 goto out;
2810
2811         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2812                 BUG_ON(ret == -EEXIST);
2813                 set_balance_control(bctl);
2814         } else {
2815                 BUG_ON(ret != -EEXIST);
2816                 spin_lock(&fs_info->balance_lock);
2817                 update_balance_args(bctl);
2818                 spin_unlock(&fs_info->balance_lock);
2819         }
2820
2821         atomic_inc(&fs_info->balance_running);
2822         mutex_unlock(&fs_info->balance_mutex);
2823
2824         ret = __btrfs_balance(fs_info);
2825
2826         mutex_lock(&fs_info->balance_mutex);
2827         atomic_dec(&fs_info->balance_running);
2828
2829         if (bargs) {
2830                 memset(bargs, 0, sizeof(*bargs));
2831                 update_ioctl_balance_args(fs_info, 0, bargs);
2832         }
2833
2834         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2835             balance_need_close(fs_info)) {
2836                 __cancel_balance(fs_info);
2837         }
2838
2839         wake_up(&fs_info->balance_wait_q);
2840
2841         return ret;
2842 out:
2843         if (bctl->flags & BTRFS_BALANCE_RESUME)
2844                 __cancel_balance(fs_info);
2845         else
2846                 kfree(bctl);
2847         return ret;
2848 }
2849
2850 static int balance_kthread(void *data)
2851 {
2852         struct btrfs_fs_info *fs_info = data;
2853         int ret = 0;
2854
2855         mutex_lock(&fs_info->volume_mutex);
2856         mutex_lock(&fs_info->balance_mutex);
2857
2858         if (fs_info->balance_ctl) {
2859                 printk(KERN_INFO "btrfs: continuing balance\n");
2860                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
2861         }
2862
2863         mutex_unlock(&fs_info->balance_mutex);
2864         mutex_unlock(&fs_info->volume_mutex);
2865
2866         return ret;
2867 }
2868
2869 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
2870 {
2871         struct task_struct *tsk;
2872
2873         spin_lock(&fs_info->balance_lock);
2874         if (!fs_info->balance_ctl) {
2875                 spin_unlock(&fs_info->balance_lock);
2876                 return 0;
2877         }
2878         spin_unlock(&fs_info->balance_lock);
2879
2880         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2881                 printk(KERN_INFO "btrfs: force skipping balance\n");
2882                 return 0;
2883         }
2884
2885         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
2886         if (IS_ERR(tsk))
2887                 return PTR_ERR(tsk);
2888
2889         return 0;
2890 }
2891
2892 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
2893 {
2894         struct btrfs_balance_control *bctl;
2895         struct btrfs_balance_item *item;
2896         struct btrfs_disk_balance_args disk_bargs;
2897         struct btrfs_path *path;
2898         struct extent_buffer *leaf;
2899         struct btrfs_key key;
2900         int ret;
2901
2902         path = btrfs_alloc_path();
2903         if (!path)
2904                 return -ENOMEM;
2905
2906         key.objectid = BTRFS_BALANCE_OBJECTID;
2907         key.type = BTRFS_BALANCE_ITEM_KEY;
2908         key.offset = 0;
2909
2910         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2911         if (ret < 0)
2912                 goto out;
2913         if (ret > 0) { /* ret = -ENOENT; */
2914                 ret = 0;
2915                 goto out;
2916         }
2917
2918         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2919         if (!bctl) {
2920                 ret = -ENOMEM;
2921                 goto out;
2922         }
2923
2924         leaf = path->nodes[0];
2925         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2926
2927         bctl->fs_info = fs_info;
2928         bctl->flags = btrfs_balance_flags(leaf, item);
2929         bctl->flags |= BTRFS_BALANCE_RESUME;
2930
2931         btrfs_balance_data(leaf, item, &disk_bargs);
2932         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2933         btrfs_balance_meta(leaf, item, &disk_bargs);
2934         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2935         btrfs_balance_sys(leaf, item, &disk_bargs);
2936         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2937
2938         mutex_lock(&fs_info->volume_mutex);
2939         mutex_lock(&fs_info->balance_mutex);
2940
2941         set_balance_control(bctl);
2942
2943         mutex_unlock(&fs_info->balance_mutex);
2944         mutex_unlock(&fs_info->volume_mutex);
2945 out:
2946         btrfs_free_path(path);
2947         return ret;
2948 }
2949
2950 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2951 {
2952         int ret = 0;
2953
2954         mutex_lock(&fs_info->balance_mutex);
2955         if (!fs_info->balance_ctl) {
2956                 mutex_unlock(&fs_info->balance_mutex);
2957                 return -ENOTCONN;
2958         }
2959
2960         if (atomic_read(&fs_info->balance_running)) {
2961                 atomic_inc(&fs_info->balance_pause_req);
2962                 mutex_unlock(&fs_info->balance_mutex);
2963
2964                 wait_event(fs_info->balance_wait_q,
2965                            atomic_read(&fs_info->balance_running) == 0);
2966
2967                 mutex_lock(&fs_info->balance_mutex);
2968                 /* we are good with balance_ctl ripped off from under us */
2969                 BUG_ON(atomic_read(&fs_info->balance_running));
2970                 atomic_dec(&fs_info->balance_pause_req);
2971         } else {
2972                 ret = -ENOTCONN;
2973         }
2974
2975         mutex_unlock(&fs_info->balance_mutex);
2976         return ret;
2977 }
2978
2979 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
2980 {
2981         mutex_lock(&fs_info->balance_mutex);
2982         if (!fs_info->balance_ctl) {
2983                 mutex_unlock(&fs_info->balance_mutex);
2984                 return -ENOTCONN;
2985         }
2986
2987         atomic_inc(&fs_info->balance_cancel_req);
2988         /*
2989          * if we are running just wait and return, balance item is
2990          * deleted in btrfs_balance in this case
2991          */
2992         if (atomic_read(&fs_info->balance_running)) {
2993                 mutex_unlock(&fs_info->balance_mutex);
2994                 wait_event(fs_info->balance_wait_q,
2995                            atomic_read(&fs_info->balance_running) == 0);
2996                 mutex_lock(&fs_info->balance_mutex);
2997         } else {
2998                 /* __cancel_balance needs volume_mutex */
2999                 mutex_unlock(&fs_info->balance_mutex);
3000                 mutex_lock(&fs_info->volume_mutex);
3001                 mutex_lock(&fs_info->balance_mutex);
3002
3003                 if (fs_info->balance_ctl)
3004                         __cancel_balance(fs_info);
3005
3006                 mutex_unlock(&fs_info->volume_mutex);
3007         }
3008
3009         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3010         atomic_dec(&fs_info->balance_cancel_req);
3011         mutex_unlock(&fs_info->balance_mutex);
3012         return 0;
3013 }
3014
3015 /*
3016  * shrinking a device means finding all of the device extents past
3017  * the new size, and then following the back refs to the chunks.
3018  * The chunk relocation code actually frees the device extent
3019  */
3020 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3021 {
3022         struct btrfs_trans_handle *trans;
3023         struct btrfs_root *root = device->dev_root;
3024         struct btrfs_dev_extent *dev_extent = NULL;
3025         struct btrfs_path *path;
3026         u64 length;
3027         u64 chunk_tree;
3028         u64 chunk_objectid;
3029         u64 chunk_offset;
3030         int ret;
3031         int slot;
3032         int failed = 0;
3033         bool retried = false;
3034         struct extent_buffer *l;
3035         struct btrfs_key key;
3036         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3037         u64 old_total = btrfs_super_total_bytes(super_copy);
3038         u64 old_size = device->total_bytes;
3039         u64 diff = device->total_bytes - new_size;
3040
3041         if (new_size >= device->total_bytes)
3042                 return -EINVAL;
3043
3044         path = btrfs_alloc_path();
3045         if (!path)
3046                 return -ENOMEM;
3047
3048         path->reada = 2;
3049
3050         lock_chunks(root);
3051
3052         device->total_bytes = new_size;
3053         if (device->writeable) {
3054                 device->fs_devices->total_rw_bytes -= diff;
3055                 spin_lock(&root->fs_info->free_chunk_lock);
3056                 root->fs_info->free_chunk_space -= diff;
3057                 spin_unlock(&root->fs_info->free_chunk_lock);
3058         }
3059         unlock_chunks(root);
3060
3061 again:
3062         key.objectid = device->devid;
3063         key.offset = (u64)-1;
3064         key.type = BTRFS_DEV_EXTENT_KEY;
3065
3066         do {
3067                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3068                 if (ret < 0)
3069                         goto done;
3070
3071                 ret = btrfs_previous_item(root, path, 0, key.type);
3072                 if (ret < 0)
3073                         goto done;
3074                 if (ret) {
3075                         ret = 0;
3076                         btrfs_release_path(path);
3077                         break;
3078                 }
3079
3080                 l = path->nodes[0];
3081                 slot = path->slots[0];
3082                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3083
3084                 if (key.objectid != device->devid) {
3085                         btrfs_release_path(path);
3086                         break;
3087                 }
3088
3089                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3090                 length = btrfs_dev_extent_length(l, dev_extent);
3091
3092                 if (key.offset + length <= new_size) {
3093                         btrfs_release_path(path);
3094                         break;
3095                 }
3096
3097                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3098                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3099                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3100                 btrfs_release_path(path);
3101
3102                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3103                                            chunk_offset);
3104                 if (ret && ret != -ENOSPC)
3105                         goto done;
3106                 if (ret == -ENOSPC)
3107                         failed++;
3108         } while (key.offset-- > 0);
3109
3110         if (failed && !retried) {
3111                 failed = 0;
3112                 retried = true;
3113                 goto again;
3114         } else if (failed && retried) {
3115                 ret = -ENOSPC;
3116                 lock_chunks(root);
3117
3118                 device->total_bytes = old_size;
3119                 if (device->writeable)
3120                         device->fs_devices->total_rw_bytes += diff;
3121                 spin_lock(&root->fs_info->free_chunk_lock);
3122                 root->fs_info->free_chunk_space += diff;
3123                 spin_unlock(&root->fs_info->free_chunk_lock);
3124                 unlock_chunks(root);
3125                 goto done;
3126         }
3127
3128         /* Shrinking succeeded, else we would be at "done". */
3129         trans = btrfs_start_transaction(root, 0);
3130         if (IS_ERR(trans)) {
3131                 ret = PTR_ERR(trans);
3132                 goto done;
3133         }
3134
3135         lock_chunks(root);
3136
3137         device->disk_total_bytes = new_size;
3138         /* Now btrfs_update_device() will change the on-disk size. */
3139         ret = btrfs_update_device(trans, device);
3140         if (ret) {
3141                 unlock_chunks(root);
3142                 btrfs_end_transaction(trans, root);
3143                 goto done;
3144         }
3145         WARN_ON(diff > old_total);
3146         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3147         unlock_chunks(root);
3148         btrfs_end_transaction(trans, root);
3149 done:
3150         btrfs_free_path(path);
3151         return ret;
3152 }
3153
3154 static int btrfs_add_system_chunk(struct btrfs_root *root,
3155                            struct btrfs_key *key,
3156                            struct btrfs_chunk *chunk, int item_size)
3157 {
3158         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3159         struct btrfs_disk_key disk_key;
3160         u32 array_size;
3161         u8 *ptr;
3162
3163         array_size = btrfs_super_sys_array_size(super_copy);
3164         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3165                 return -EFBIG;
3166
3167         ptr = super_copy->sys_chunk_array + array_size;
3168         btrfs_cpu_key_to_disk(&disk_key, key);
3169         memcpy(ptr, &disk_key, sizeof(disk_key));
3170         ptr += sizeof(disk_key);
3171         memcpy(ptr, chunk, item_size);
3172         item_size += sizeof(disk_key);
3173         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3174         return 0;
3175 }
3176
3177 /*
3178  * sort the devices in descending order by max_avail, total_avail
3179  */
3180 static int btrfs_cmp_device_info(const void *a, const void *b)
3181 {
3182         const struct btrfs_device_info *di_a = a;
3183         const struct btrfs_device_info *di_b = b;
3184
3185         if (di_a->max_avail > di_b->max_avail)
3186                 return -1;
3187         if (di_a->max_avail < di_b->max_avail)
3188                 return 1;
3189         if (di_a->total_avail > di_b->total_avail)
3190                 return -1;
3191         if (di_a->total_avail < di_b->total_avail)
3192                 return 1;
3193         return 0;
3194 }
3195
3196 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3197                                struct btrfs_root *extent_root,
3198                                struct map_lookup **map_ret,
3199                                u64 *num_bytes_out, u64 *stripe_size_out,
3200                                u64 start, u64 type)
3201 {
3202         struct btrfs_fs_info *info = extent_root->fs_info;
3203         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3204         struct list_head *cur;
3205         struct map_lookup *map = NULL;
3206         struct extent_map_tree *em_tree;
3207         struct extent_map *em;
3208         struct btrfs_device_info *devices_info = NULL;
3209         u64 total_avail;
3210         int num_stripes;        /* total number of stripes to allocate */
3211         int sub_stripes;        /* sub_stripes info for map */
3212         int dev_stripes;        /* stripes per dev */
3213         int devs_max;           /* max devs to use */
3214         int devs_min;           /* min devs needed */
3215         int devs_increment;     /* ndevs has to be a multiple of this */
3216         int ncopies;            /* how many copies to data has */
3217         int ret;
3218         u64 max_stripe_size;
3219         u64 max_chunk_size;
3220         u64 stripe_size;
3221         u64 num_bytes;
3222         int ndevs;
3223         int i;
3224         int j;
3225
3226         BUG_ON(!alloc_profile_is_valid(type, 0));
3227
3228         if (list_empty(&fs_devices->alloc_list))
3229                 return -ENOSPC;
3230
3231         sub_stripes = 1;
3232         dev_stripes = 1;
3233         devs_increment = 1;
3234         ncopies = 1;
3235         devs_max = 0;   /* 0 == as many as possible */
3236         devs_min = 1;
3237
3238         /*
3239          * define the properties of each RAID type.
3240          * FIXME: move this to a global table and use it in all RAID
3241          * calculation code
3242          */
3243         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3244                 dev_stripes = 2;
3245                 ncopies = 2;
3246                 devs_max = 1;
3247         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3248                 devs_min = 2;
3249         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3250                 devs_increment = 2;
3251                 ncopies = 2;
3252                 devs_max = 2;
3253                 devs_min = 2;
3254         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3255                 sub_stripes = 2;
3256                 devs_increment = 2;
3257                 ncopies = 2;
3258                 devs_min = 4;
3259         } else {
3260                 devs_max = 1;
3261         }
3262
3263         if (type & BTRFS_BLOCK_GROUP_DATA) {
3264                 max_stripe_size = 1024 * 1024 * 1024;
3265                 max_chunk_size = 10 * max_stripe_size;
3266         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3267                 /* for larger filesystems, use larger metadata chunks */
3268                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3269                         max_stripe_size = 1024 * 1024 * 1024;
3270                 else
3271                         max_stripe_size = 256 * 1024 * 1024;
3272                 max_chunk_size = max_stripe_size;
3273         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3274                 max_stripe_size = 32 * 1024 * 1024;
3275                 max_chunk_size = 2 * max_stripe_size;
3276         } else {
3277                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3278                        type);
3279                 BUG_ON(1);
3280         }
3281
3282         /* we don't want a chunk larger than 10% of writeable space */
3283         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3284                              max_chunk_size);
3285
3286         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3287                                GFP_NOFS);
3288         if (!devices_info)
3289                 return -ENOMEM;
3290
3291         cur = fs_devices->alloc_list.next;
3292
3293         /*
3294          * in the first pass through the devices list, we gather information
3295          * about the available holes on each device.
3296          */
3297         ndevs = 0;
3298         while (cur != &fs_devices->alloc_list) {
3299                 struct btrfs_device *device;
3300                 u64 max_avail;
3301                 u64 dev_offset;
3302
3303                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3304
3305                 cur = cur->next;
3306
3307                 if (!device->writeable) {
3308                         printk(KERN_ERR
3309                                "btrfs: read-only device in alloc_list\n");
3310                         WARN_ON(1);
3311                         continue;
3312                 }
3313
3314                 if (!device->in_fs_metadata)
3315                         continue;
3316
3317                 if (device->total_bytes > device->bytes_used)
3318                         total_avail = device->total_bytes - device->bytes_used;
3319                 else
3320                         total_avail = 0;
3321
3322                 /* If there is no space on this device, skip it. */
3323                 if (total_avail == 0)
3324                         continue;
3325
3326                 ret = find_free_dev_extent(device,
3327                                            max_stripe_size * dev_stripes,
3328                                            &dev_offset, &max_avail);
3329                 if (ret && ret != -ENOSPC)
3330                         goto error;
3331
3332                 if (ret == 0)
3333                         max_avail = max_stripe_size * dev_stripes;
3334
3335                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3336                         continue;
3337
3338                 devices_info[ndevs].dev_offset = dev_offset;
3339                 devices_info[ndevs].max_avail = max_avail;
3340                 devices_info[ndevs].total_avail = total_avail;
3341                 devices_info[ndevs].dev = device;
3342                 ++ndevs;
3343         }
3344
3345         /*
3346          * now sort the devices by hole size / available space
3347          */
3348         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3349              btrfs_cmp_device_info, NULL);
3350
3351         /* round down to number of usable stripes */
3352         ndevs -= ndevs % devs_increment;
3353
3354         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3355                 ret = -ENOSPC;
3356                 goto error;
3357         }
3358
3359         if (devs_max && ndevs > devs_max)
3360                 ndevs = devs_max;
3361         /*
3362          * the primary goal is to maximize the number of stripes, so use as many
3363          * devices as possible, even if the stripes are not maximum sized.
3364          */
3365         stripe_size = devices_info[ndevs-1].max_avail;
3366         num_stripes = ndevs * dev_stripes;
3367
3368         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3369                 stripe_size = max_chunk_size * ncopies;
3370                 do_div(stripe_size, ndevs);
3371         }
3372
3373         do_div(stripe_size, dev_stripes);
3374
3375         /* align to BTRFS_STRIPE_LEN */
3376         do_div(stripe_size, BTRFS_STRIPE_LEN);
3377         stripe_size *= BTRFS_STRIPE_LEN;
3378
3379         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3380         if (!map) {
3381                 ret = -ENOMEM;
3382                 goto error;
3383         }
3384         map->num_stripes = num_stripes;
3385
3386         for (i = 0; i < ndevs; ++i) {
3387                 for (j = 0; j < dev_stripes; ++j) {
3388                         int s = i * dev_stripes + j;
3389                         map->stripes[s].dev = devices_info[i].dev;
3390                         map->stripes[s].physical = devices_info[i].dev_offset +
3391                                                    j * stripe_size;
3392                 }
3393         }
3394         map->sector_size = extent_root->sectorsize;
3395         map->stripe_len = BTRFS_STRIPE_LEN;
3396         map->io_align = BTRFS_STRIPE_LEN;
3397         map->io_width = BTRFS_STRIPE_LEN;
3398         map->type = type;
3399         map->sub_stripes = sub_stripes;
3400
3401         *map_ret = map;
3402         num_bytes = stripe_size * (num_stripes / ncopies);
3403
3404         *stripe_size_out = stripe_size;
3405         *num_bytes_out = num_bytes;
3406
3407         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3408
3409         em = alloc_extent_map();
3410         if (!em) {
3411                 ret = -ENOMEM;
3412                 goto error;
3413         }
3414         em->bdev = (struct block_device *)map;
3415         em->start = start;
3416         em->len = num_bytes;
3417         em->block_start = 0;
3418         em->block_len = em->len;
3419
3420         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3421         write_lock(&em_tree->lock);
3422         ret = add_extent_mapping(em_tree, em);
3423         write_unlock(&em_tree->lock);
3424         free_extent_map(em);
3425         if (ret)
3426                 goto error;
3427
3428         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3429                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3430                                      start, num_bytes);
3431         if (ret)
3432                 goto error;
3433
3434         for (i = 0; i < map->num_stripes; ++i) {
3435                 struct btrfs_device *device;
3436                 u64 dev_offset;
3437
3438                 device = map->stripes[i].dev;
3439                 dev_offset = map->stripes[i].physical;
3440
3441                 ret = btrfs_alloc_dev_extent(trans, device,
3442                                 info->chunk_root->root_key.objectid,
3443                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3444                                 start, dev_offset, stripe_size);
3445                 if (ret) {
3446                         btrfs_abort_transaction(trans, extent_root, ret);
3447                         goto error;
3448                 }
3449         }
3450
3451         kfree(devices_info);
3452         return 0;
3453
3454 error:
3455         kfree(map);
3456         kfree(devices_info);
3457         return ret;
3458 }
3459
3460 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3461                                 struct btrfs_root *extent_root,
3462                                 struct map_lookup *map, u64 chunk_offset,
3463                                 u64 chunk_size, u64 stripe_size)
3464 {
3465         u64 dev_offset;
3466         struct btrfs_key key;
3467         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3468         struct btrfs_device *device;
3469         struct btrfs_chunk *chunk;
3470         struct btrfs_stripe *stripe;
3471         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3472         int index = 0;
3473         int ret;
3474
3475         chunk = kzalloc(item_size, GFP_NOFS);
3476         if (!chunk)
3477                 return -ENOMEM;
3478
3479         index = 0;
3480         while (index < map->num_stripes) {
3481                 device = map->stripes[index].dev;
3482                 device->bytes_used += stripe_size;
3483                 ret = btrfs_update_device(trans, device);
3484                 if (ret)
3485                         goto out_free;
3486                 index++;
3487         }
3488
3489         spin_lock(&extent_root->fs_info->free_chunk_lock);
3490         extent_root->fs_info->free_chunk_space -= (stripe_size *
3491                                                    map->num_stripes);
3492         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3493
3494         index = 0;
3495         stripe = &chunk->stripe;
3496         while (index < map->num_stripes) {
3497                 device = map->stripes[index].dev;
3498                 dev_offset = map->stripes[index].physical;
3499
3500                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3501                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3502                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3503                 stripe++;
3504                 index++;
3505         }
3506
3507         btrfs_set_stack_chunk_length(chunk, chunk_size);
3508         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3509         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3510         btrfs_set_stack_chunk_type(chunk, map->type);
3511         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3512         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3513         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3514         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3515         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3516
3517         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3518         key.type = BTRFS_CHUNK_ITEM_KEY;
3519         key.offset = chunk_offset;
3520
3521         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3522
3523         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3524                 /*
3525                  * TODO: Cleanup of inserted chunk root in case of
3526                  * failure.
3527                  */
3528                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3529                                              item_size);
3530         }
3531
3532 out_free:
3533         kfree(chunk);
3534         return ret;
3535 }
3536
3537 /*
3538  * Chunk allocation falls into two parts. The first part does works
3539  * that make the new allocated chunk useable, but not do any operation
3540  * that modifies the chunk tree. The second part does the works that
3541  * require modifying the chunk tree. This division is important for the
3542  * bootstrap process of adding storage to a seed btrfs.
3543  */
3544 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3545                       struct btrfs_root *extent_root, u64 type)
3546 {
3547         u64 chunk_offset;
3548         u64 chunk_size;
3549         u64 stripe_size;
3550         struct map_lookup *map;
3551         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3552         int ret;
3553
3554         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3555                               &chunk_offset);
3556         if (ret)
3557                 return ret;
3558
3559         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3560                                   &stripe_size, chunk_offset, type);
3561         if (ret)
3562                 return ret;
3563
3564         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3565                                    chunk_size, stripe_size);
3566         if (ret)
3567                 return ret;
3568         return 0;
3569 }
3570
3571 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3572                                          struct btrfs_root *root,
3573                                          struct btrfs_device *device)
3574 {
3575         u64 chunk_offset;
3576         u64 sys_chunk_offset;
3577         u64 chunk_size;
3578         u64 sys_chunk_size;
3579         u64 stripe_size;
3580         u64 sys_stripe_size;
3581         u64 alloc_profile;
3582         struct map_lookup *map;
3583         struct map_lookup *sys_map;
3584         struct btrfs_fs_info *fs_info = root->fs_info;
3585         struct btrfs_root *extent_root = fs_info->extent_root;
3586         int ret;
3587
3588         ret = find_next_chunk(fs_info->chunk_root,
3589                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3590         if (ret)
3591                 return ret;
3592
3593         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3594                                 fs_info->avail_metadata_alloc_bits;
3595         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3596
3597         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3598                                   &stripe_size, chunk_offset, alloc_profile);
3599         if (ret)
3600                 return ret;
3601
3602         sys_chunk_offset = chunk_offset + chunk_size;
3603
3604         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3605                                 fs_info->avail_system_alloc_bits;
3606         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3607
3608         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3609                                   &sys_chunk_size, &sys_stripe_size,
3610                                   sys_chunk_offset, alloc_profile);
3611         if (ret)
3612                 goto abort;
3613
3614         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3615         if (ret)
3616                 goto abort;
3617
3618         /*
3619          * Modifying chunk tree needs allocating new blocks from both
3620          * system block group and metadata block group. So we only can
3621          * do operations require modifying the chunk tree after both
3622          * block groups were created.
3623          */
3624         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3625                                    chunk_size, stripe_size);
3626         if (ret)
3627                 goto abort;
3628
3629         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3630                                    sys_chunk_offset, sys_chunk_size,
3631                                    sys_stripe_size);
3632         if (ret)
3633                 goto abort;
3634
3635         return 0;
3636
3637 abort:
3638         btrfs_abort_transaction(trans, root, ret);
3639         return ret;
3640 }
3641
3642 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3643 {
3644         struct extent_map *em;
3645         struct map_lookup *map;
3646         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3647         int readonly = 0;
3648         int i;
3649
3650         read_lock(&map_tree->map_tree.lock);
3651         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3652         read_unlock(&map_tree->map_tree.lock);
3653         if (!em)
3654                 return 1;
3655
3656         if (btrfs_test_opt(root, DEGRADED)) {
3657                 free_extent_map(em);
3658                 return 0;
3659         }
3660
3661         map = (struct map_lookup *)em->bdev;
3662         for (i = 0; i < map->num_stripes; i++) {
3663                 if (!map->stripes[i].dev->writeable) {
3664                         readonly = 1;
3665                         break;
3666                 }
3667         }
3668         free_extent_map(em);
3669         return readonly;
3670 }
3671
3672 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3673 {
3674         extent_map_tree_init(&tree->map_tree);
3675 }
3676
3677 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3678 {
3679         struct extent_map *em;
3680
3681         while (1) {
3682                 write_lock(&tree->map_tree.lock);
3683                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3684                 if (em)
3685                         remove_extent_mapping(&tree->map_tree, em);
3686                 write_unlock(&tree->map_tree.lock);
3687                 if (!em)
3688                         break;
3689                 kfree(em->bdev);
3690                 /* once for us */
3691                 free_extent_map(em);
3692                 /* once for the tree */
3693                 free_extent_map(em);
3694         }
3695 }
3696
3697 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3698 {
3699         struct extent_map *em;
3700         struct map_lookup *map;
3701         struct extent_map_tree *em_tree = &map_tree->map_tree;
3702         int ret;
3703
3704         read_lock(&em_tree->lock);
3705         em = lookup_extent_mapping(em_tree, logical, len);
3706         read_unlock(&em_tree->lock);
3707         BUG_ON(!em);
3708
3709         BUG_ON(em->start > logical || em->start + em->len < logical);
3710         map = (struct map_lookup *)em->bdev;
3711         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3712                 ret = map->num_stripes;
3713         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3714                 ret = map->sub_stripes;
3715         else
3716                 ret = 1;
3717         free_extent_map(em);
3718         return ret;
3719 }
3720
3721 static int find_live_mirror(struct map_lookup *map, int first, int num,
3722                             int optimal)
3723 {
3724         int i;
3725         if (map->stripes[optimal].dev->bdev)
3726                 return optimal;
3727         for (i = first; i < first + num; i++) {
3728                 if (map->stripes[i].dev->bdev)
3729                         return i;
3730         }
3731         /* we couldn't find one that doesn't fail.  Just return something
3732          * and the io error handling code will clean up eventually
3733          */
3734         return optimal;
3735 }
3736
3737 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3738                              u64 logical, u64 *length,
3739                              struct btrfs_bio **bbio_ret,
3740                              int mirror_num)
3741 {
3742         struct extent_map *em;
3743         struct map_lookup *map;
3744         struct extent_map_tree *em_tree = &map_tree->map_tree;
3745         u64 offset;
3746         u64 stripe_offset;
3747         u64 stripe_end_offset;
3748         u64 stripe_nr;
3749         u64 stripe_nr_orig;
3750         u64 stripe_nr_end;
3751         int stripe_index;
3752         int i;
3753         int ret = 0;
3754         int num_stripes;
3755         int max_errors = 0;
3756         struct btrfs_bio *bbio = NULL;
3757
3758         read_lock(&em_tree->lock);
3759         em = lookup_extent_mapping(em_tree, logical, *length);
3760         read_unlock(&em_tree->lock);
3761
3762         if (!em) {
3763                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3764                        (unsigned long long)logical,
3765                        (unsigned long long)*length);
3766                 BUG();
3767         }
3768
3769         BUG_ON(em->start > logical || em->start + em->len < logical);
3770         map = (struct map_lookup *)em->bdev;
3771         offset = logical - em->start;
3772
3773         if (mirror_num > map->num_stripes)
3774                 mirror_num = 0;
3775
3776         stripe_nr = offset;
3777         /*
3778          * stripe_nr counts the total number of stripes we have to stride
3779          * to get to this block
3780          */
3781         do_div(stripe_nr, map->stripe_len);
3782
3783         stripe_offset = stripe_nr * map->stripe_len;
3784         BUG_ON(offset < stripe_offset);
3785
3786         /* stripe_offset is the offset of this block in its stripe*/
3787         stripe_offset = offset - stripe_offset;
3788
3789         if (rw & REQ_DISCARD)
3790                 *length = min_t(u64, em->len - offset, *length);
3791         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3792                 /* we limit the length of each bio to what fits in a stripe */
3793                 *length = min_t(u64, em->len - offset,
3794                                 map->stripe_len - stripe_offset);
3795         } else {
3796                 *length = em->len - offset;
3797         }
3798
3799         if (!bbio_ret)
3800                 goto out;
3801
3802         num_stripes = 1;
3803         stripe_index = 0;
3804         stripe_nr_orig = stripe_nr;
3805         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3806                         (~(map->stripe_len - 1));
3807         do_div(stripe_nr_end, map->stripe_len);
3808         stripe_end_offset = stripe_nr_end * map->stripe_len -
3809                             (offset + *length);
3810         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3811                 if (rw & REQ_DISCARD)
3812                         num_stripes = min_t(u64, map->num_stripes,
3813                                             stripe_nr_end - stripe_nr_orig);
3814                 stripe_index = do_div(stripe_nr, map->num_stripes);
3815         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3816                 if (rw & (REQ_WRITE | REQ_DISCARD))
3817                         num_stripes = map->num_stripes;
3818                 else if (mirror_num)
3819                         stripe_index = mirror_num - 1;
3820                 else {
3821                         stripe_index = find_live_mirror(map, 0,
3822                                             map->num_stripes,
3823                                             current->pid % map->num_stripes);
3824                         mirror_num = stripe_index + 1;
3825                 }
3826
3827         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3828                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3829                         num_stripes = map->num_stripes;
3830                 } else if (mirror_num) {
3831                         stripe_index = mirror_num - 1;
3832                 } else {
3833                         mirror_num = 1;
3834                 }
3835
3836         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3837                 int factor = map->num_stripes / map->sub_stripes;
3838
3839                 stripe_index = do_div(stripe_nr, factor);
3840                 stripe_index *= map->sub_stripes;
3841
3842                 if (rw & REQ_WRITE)
3843                         num_stripes = map->sub_stripes;
3844                 else if (rw & REQ_DISCARD)
3845                         num_stripes = min_t(u64, map->sub_stripes *
3846                                             (stripe_nr_end - stripe_nr_orig),
3847                                             map->num_stripes);
3848                 else if (mirror_num)
3849                         stripe_index += mirror_num - 1;
3850                 else {
3851                         int old_stripe_index = stripe_index;
3852                         stripe_index = find_live_mirror(map, stripe_index,
3853                                               map->sub_stripes, stripe_index +
3854                                               current->pid % map->sub_stripes);
3855                         mirror_num = stripe_index - old_stripe_index + 1;
3856                 }
3857         } else {
3858                 /*
3859                  * after this do_div call, stripe_nr is the number of stripes
3860                  * on this device we have to walk to find the data, and
3861                  * stripe_index is the number of our device in the stripe array
3862                  */
3863                 stripe_index = do_div(stripe_nr, map->num_stripes);
3864                 mirror_num = stripe_index + 1;
3865         }
3866         BUG_ON(stripe_index >= map->num_stripes);
3867
3868         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3869         if (!bbio) {
3870                 ret = -ENOMEM;
3871                 goto out;
3872         }
3873         atomic_set(&bbio->error, 0);
3874
3875         if (rw & REQ_DISCARD) {
3876                 int factor = 0;
3877                 int sub_stripes = 0;
3878                 u64 stripes_per_dev = 0;
3879                 u32 remaining_stripes = 0;
3880                 u32 last_stripe = 0;
3881
3882                 if (map->type &
3883                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3884                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3885                                 sub_stripes = 1;
3886                         else
3887                                 sub_stripes = map->sub_stripes;
3888
3889                         factor = map->num_stripes / sub_stripes;
3890                         stripes_per_dev = div_u64_rem(stripe_nr_end -
3891                                                       stripe_nr_orig,
3892                                                       factor,
3893                                                       &remaining_stripes);
3894                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
3895                         last_stripe *= sub_stripes;
3896                 }
3897
3898                 for (i = 0; i < num_stripes; i++) {
3899                         bbio->stripes[i].physical =
3900                                 map->stripes[stripe_index].physical +
3901                                 stripe_offset + stripe_nr * map->stripe_len;
3902                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3903
3904                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3905                                          BTRFS_BLOCK_GROUP_RAID10)) {
3906                                 bbio->stripes[i].length = stripes_per_dev *
3907                                                           map->stripe_len;
3908
3909                                 if (i / sub_stripes < remaining_stripes)
3910                                         bbio->stripes[i].length +=
3911                                                 map->stripe_len;
3912
3913                                 /*
3914                                  * Special for the first stripe and
3915                                  * the last stripe:
3916                                  *
3917                                  * |-------|...|-------|
3918                                  *     |----------|
3919                                  *    off     end_off
3920                                  */
3921                                 if (i < sub_stripes)
3922                                         bbio->stripes[i].length -=
3923                                                 stripe_offset;
3924
3925                                 if (stripe_index >= last_stripe &&
3926                                     stripe_index <= (last_stripe +
3927                                                      sub_stripes - 1))
3928                                         bbio->stripes[i].length -=
3929                                                 stripe_end_offset;
3930
3931                                 if (i == sub_stripes - 1)
3932                                         stripe_offset = 0;
3933                         } else
3934                                 bbio->stripes[i].length = *length;
3935
3936                         stripe_index++;
3937                         if (stripe_index == map->num_stripes) {
3938                                 /* This could only happen for RAID0/10 */
3939                                 stripe_index = 0;
3940                                 stripe_nr++;
3941                         }
3942                 }
3943         } else {
3944                 for (i = 0; i < num_stripes; i++) {
3945                         bbio->stripes[i].physical =
3946                                 map->stripes[stripe_index].physical +
3947                                 stripe_offset +
3948                                 stripe_nr * map->stripe_len;
3949                         bbio->stripes[i].dev =
3950                                 map->stripes[stripe_index].dev;
3951                         stripe_index++;
3952                 }
3953         }
3954
3955         if (rw & REQ_WRITE) {
3956                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3957                                  BTRFS_BLOCK_GROUP_RAID10 |
3958                                  BTRFS_BLOCK_GROUP_DUP)) {
3959                         max_errors = 1;
3960                 }
3961         }
3962
3963         *bbio_ret = bbio;
3964         bbio->num_stripes = num_stripes;
3965         bbio->max_errors = max_errors;
3966         bbio->mirror_num = mirror_num;
3967 out:
3968         free_extent_map(em);
3969         return ret;
3970 }
3971
3972 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3973                       u64 logical, u64 *length,
3974                       struct btrfs_bio **bbio_ret, int mirror_num)
3975 {
3976         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3977                                  mirror_num);
3978 }
3979
3980 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3981                      u64 chunk_start, u64 physical, u64 devid,
3982                      u64 **logical, int *naddrs, int *stripe_len)
3983 {
3984         struct extent_map_tree *em_tree = &map_tree->map_tree;
3985         struct extent_map *em;
3986         struct map_lookup *map;
3987         u64 *buf;
3988         u64 bytenr;
3989         u64 length;
3990         u64 stripe_nr;
3991         int i, j, nr = 0;
3992
3993         read_lock(&em_tree->lock);
3994         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3995         read_unlock(&em_tree->lock);
3996
3997         BUG_ON(!em || em->start != chunk_start);
3998         map = (struct map_lookup *)em->bdev;
3999
4000         length = em->len;
4001         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4002                 do_div(length, map->num_stripes / map->sub_stripes);
4003         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4004                 do_div(length, map->num_stripes);
4005
4006         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4007         BUG_ON(!buf); /* -ENOMEM */
4008
4009         for (i = 0; i < map->num_stripes; i++) {
4010                 if (devid && map->stripes[i].dev->devid != devid)
4011                         continue;
4012                 if (map->stripes[i].physical > physical ||
4013                     map->stripes[i].physical + length <= physical)
4014                         continue;
4015
4016                 stripe_nr = physical - map->stripes[i].physical;
4017                 do_div(stripe_nr, map->stripe_len);
4018
4019                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4020                         stripe_nr = stripe_nr * map->num_stripes + i;
4021                         do_div(stripe_nr, map->sub_stripes);
4022                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4023                         stripe_nr = stripe_nr * map->num_stripes + i;
4024                 }
4025                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4026                 WARN_ON(nr >= map->num_stripes);
4027                 for (j = 0; j < nr; j++) {
4028                         if (buf[j] == bytenr)
4029                                 break;
4030                 }
4031                 if (j == nr) {
4032                         WARN_ON(nr >= map->num_stripes);
4033                         buf[nr++] = bytenr;
4034                 }
4035         }
4036
4037         *logical = buf;
4038         *naddrs = nr;
4039         *stripe_len = map->stripe_len;
4040
4041         free_extent_map(em);
4042         return 0;
4043 }
4044
4045 static void *merge_stripe_index_into_bio_private(void *bi_private,
4046                                                  unsigned int stripe_index)
4047 {
4048         /*
4049          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4050          * at most 1.
4051          * The alternative solution (instead of stealing bits from the
4052          * pointer) would be to allocate an intermediate structure
4053          * that contains the old private pointer plus the stripe_index.
4054          */
4055         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4056         BUG_ON(stripe_index > 3);
4057         return (void *)(((uintptr_t)bi_private) | stripe_index);
4058 }
4059
4060 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4061 {
4062         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4063 }
4064
4065 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4066 {
4067         return (unsigned int)((uintptr_t)bi_private) & 3;
4068 }
4069
4070 static void btrfs_end_bio(struct bio *bio, int err)
4071 {
4072         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4073         int is_orig_bio = 0;
4074
4075         if (err) {
4076                 atomic_inc(&bbio->error);
4077                 if (err == -EIO || err == -EREMOTEIO) {
4078                         unsigned int stripe_index =
4079                                 extract_stripe_index_from_bio_private(
4080                                         bio->bi_private);
4081                         struct btrfs_device *dev;
4082
4083                         BUG_ON(stripe_index >= bbio->num_stripes);
4084                         dev = bbio->stripes[stripe_index].dev;
4085                         if (dev->bdev) {
4086                                 if (bio->bi_rw & WRITE)
4087                                         btrfs_dev_stat_inc(dev,
4088                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4089                                 else
4090                                         btrfs_dev_stat_inc(dev,
4091                                                 BTRFS_DEV_STAT_READ_ERRS);
4092                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4093                                         btrfs_dev_stat_inc(dev,
4094                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4095                                 btrfs_dev_stat_print_on_error(dev);
4096                         }
4097                 }
4098         }
4099
4100         if (bio == bbio->orig_bio)
4101                 is_orig_bio = 1;
4102
4103         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4104                 if (!is_orig_bio) {
4105                         bio_put(bio);
4106                         bio = bbio->orig_bio;
4107                 }
4108                 bio->bi_private = bbio->private;
4109                 bio->bi_end_io = bbio->end_io;
4110                 bio->bi_bdev = (struct block_device *)
4111                                         (unsigned long)bbio->mirror_num;
4112                 /* only send an error to the higher layers if it is
4113                  * beyond the tolerance of the multi-bio
4114                  */
4115                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4116                         err = -EIO;
4117                 } else {
4118                         /*
4119                          * this bio is actually up to date, we didn't
4120                          * go over the max number of errors
4121                          */
4122                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4123                         err = 0;
4124                 }
4125                 kfree(bbio);
4126
4127                 bio_endio(bio, err);
4128         } else if (!is_orig_bio) {
4129                 bio_put(bio);
4130         }
4131 }
4132
4133 struct async_sched {
4134         struct bio *bio;
4135         int rw;
4136         struct btrfs_fs_info *info;
4137         struct btrfs_work work;
4138 };
4139
4140 /*
4141  * see run_scheduled_bios for a description of why bios are collected for
4142  * async submit.
4143  *
4144  * This will add one bio to the pending list for a device and make sure
4145  * the work struct is scheduled.
4146  */
4147 static noinline void schedule_bio(struct btrfs_root *root,
4148                                  struct btrfs_device *device,
4149                                  int rw, struct bio *bio)
4150 {
4151         int should_queue = 1;
4152         struct btrfs_pending_bios *pending_bios;
4153
4154         /* don't bother with additional async steps for reads, right now */
4155         if (!(rw & REQ_WRITE)) {
4156                 bio_get(bio);
4157                 btrfsic_submit_bio(rw, bio);
4158                 bio_put(bio);
4159                 return;
4160         }
4161
4162         /*
4163          * nr_async_bios allows us to reliably return congestion to the
4164          * higher layers.  Otherwise, the async bio makes it appear we have
4165          * made progress against dirty pages when we've really just put it
4166          * on a queue for later
4167          */
4168         atomic_inc(&root->fs_info->nr_async_bios);
4169         WARN_ON(bio->bi_next);
4170         bio->bi_next = NULL;
4171         bio->bi_rw |= rw;
4172
4173         spin_lock(&device->io_lock);
4174         if (bio->bi_rw & REQ_SYNC)
4175                 pending_bios = &device->pending_sync_bios;
4176         else
4177                 pending_bios = &device->pending_bios;
4178
4179         if (pending_bios->tail)
4180                 pending_bios->tail->bi_next = bio;
4181
4182         pending_bios->tail = bio;
4183         if (!pending_bios->head)
4184                 pending_bios->head = bio;
4185         if (device->running_pending)
4186                 should_queue = 0;
4187
4188         spin_unlock(&device->io_lock);
4189
4190         if (should_queue)
4191                 btrfs_queue_worker(&root->fs_info->submit_workers,
4192                                    &device->work);
4193 }
4194
4195 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4196                   int mirror_num, int async_submit)
4197 {
4198         struct btrfs_mapping_tree *map_tree;
4199         struct btrfs_device *dev;
4200         struct bio *first_bio = bio;
4201         u64 logical = (u64)bio->bi_sector << 9;
4202         u64 length = 0;
4203         u64 map_length;
4204         int ret;
4205         int dev_nr = 0;
4206         int total_devs = 1;
4207         struct btrfs_bio *bbio = NULL;
4208
4209         length = bio->bi_size;
4210         map_tree = &root->fs_info->mapping_tree;
4211         map_length = length;
4212
4213         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4214                               mirror_num);
4215         if (ret) /* -ENOMEM */
4216                 return ret;
4217
4218         total_devs = bbio->num_stripes;
4219         if (map_length < length) {
4220                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
4221                        "len %llu\n", (unsigned long long)logical,
4222                        (unsigned long long)length,
4223                        (unsigned long long)map_length);
4224                 BUG();
4225         }
4226
4227         bbio->orig_bio = first_bio;
4228         bbio->private = first_bio->bi_private;
4229         bbio->end_io = first_bio->bi_end_io;
4230         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4231
4232         while (dev_nr < total_devs) {
4233                 if (dev_nr < total_devs - 1) {
4234                         bio = bio_clone(first_bio, GFP_NOFS);
4235                         BUG_ON(!bio); /* -ENOMEM */
4236                 } else {
4237                         bio = first_bio;
4238                 }
4239                 bio->bi_private = bbio;
4240                 bio->bi_private = merge_stripe_index_into_bio_private(
4241                                 bio->bi_private, (unsigned int)dev_nr);
4242                 bio->bi_end_io = btrfs_end_bio;
4243                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4244                 dev = bbio->stripes[dev_nr].dev;
4245                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4246 #ifdef DEBUG
4247                         struct rcu_string *name;
4248
4249                         rcu_read_lock();
4250                         name = rcu_dereference(dev->name);
4251                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4252                                  "(%s id %llu), size=%u\n", rw,
4253                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4254                                  name->str, dev->devid, bio->bi_size);
4255                         rcu_read_unlock();
4256 #endif
4257                         bio->bi_bdev = dev->bdev;
4258                         if (async_submit)
4259                                 schedule_bio(root, dev, rw, bio);
4260                         else
4261                                 btrfsic_submit_bio(rw, bio);
4262                 } else {
4263                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4264                         bio->bi_sector = logical >> 9;
4265                         bio_endio(bio, -EIO);
4266                 }
4267                 dev_nr++;
4268         }
4269         return 0;
4270 }
4271
4272 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4273                                        u8 *uuid, u8 *fsid)
4274 {
4275         struct btrfs_device *device;
4276         struct btrfs_fs_devices *cur_devices;
4277
4278         cur_devices = root->fs_info->fs_devices;
4279         while (cur_devices) {
4280                 if (!fsid ||
4281                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4282                         device = __find_device(&cur_devices->devices,
4283                                                devid, uuid);
4284                         if (device)
4285                                 return device;
4286                 }
4287                 cur_devices = cur_devices->seed;
4288         }
4289         return NULL;
4290 }
4291
4292 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4293                                             u64 devid, u8 *dev_uuid)
4294 {
4295         struct btrfs_device *device;
4296         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4297
4298         device = kzalloc(sizeof(*device), GFP_NOFS);
4299         if (!device)
4300                 return NULL;
4301         list_add(&device->dev_list,
4302                  &fs_devices->devices);
4303         device->dev_root = root->fs_info->dev_root;
4304         device->devid = devid;
4305         device->work.func = pending_bios_fn;
4306         device->fs_devices = fs_devices;
4307         device->missing = 1;
4308         fs_devices->num_devices++;
4309         fs_devices->missing_devices++;
4310         spin_lock_init(&device->io_lock);
4311         INIT_LIST_HEAD(&device->dev_alloc_list);
4312         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4313         return device;
4314 }
4315
4316 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4317                           struct extent_buffer *leaf,
4318                           struct btrfs_chunk *chunk)
4319 {
4320         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4321         struct map_lookup *map;
4322         struct extent_map *em;
4323         u64 logical;
4324         u64 length;
4325         u64 devid;
4326         u8 uuid[BTRFS_UUID_SIZE];
4327         int num_stripes;
4328         int ret;
4329         int i;
4330
4331         logical = key->offset;
4332         length = btrfs_chunk_length(leaf, chunk);
4333
4334         read_lock(&map_tree->map_tree.lock);
4335         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4336         read_unlock(&map_tree->map_tree.lock);
4337
4338         /* already mapped? */
4339         if (em && em->start <= logical && em->start + em->len > logical) {
4340                 free_extent_map(em);
4341                 return 0;
4342         } else if (em) {
4343                 free_extent_map(em);
4344         }
4345
4346         em = alloc_extent_map();
4347         if (!em)
4348                 return -ENOMEM;
4349         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4350         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4351         if (!map) {
4352                 free_extent_map(em);
4353                 return -ENOMEM;
4354         }
4355
4356         em->bdev = (struct block_device *)map;
4357         em->start = logical;
4358         em->len = length;
4359         em->block_start = 0;
4360         em->block_len = em->len;
4361
4362         map->num_stripes = num_stripes;
4363         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4364         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4365         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4366         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4367         map->type = btrfs_chunk_type(leaf, chunk);
4368         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4369         for (i = 0; i < num_stripes; i++) {
4370                 map->stripes[i].physical =
4371                         btrfs_stripe_offset_nr(leaf, chunk, i);
4372                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4373                 read_extent_buffer(leaf, uuid, (unsigned long)
4374                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4375                                    BTRFS_UUID_SIZE);
4376                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4377                                                         NULL);
4378                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4379                         kfree(map);
4380                         free_extent_map(em);
4381                         return -EIO;
4382                 }
4383                 if (!map->stripes[i].dev) {
4384                         map->stripes[i].dev =
4385                                 add_missing_dev(root, devid, uuid);
4386                         if (!map->stripes[i].dev) {
4387                                 kfree(map);
4388                                 free_extent_map(em);
4389                                 return -EIO;
4390                         }
4391                 }
4392                 map->stripes[i].dev->in_fs_metadata = 1;
4393         }
4394
4395         write_lock(&map_tree->map_tree.lock);
4396         ret = add_extent_mapping(&map_tree->map_tree, em);
4397         write_unlock(&map_tree->map_tree.lock);
4398         BUG_ON(ret); /* Tree corruption */
4399         free_extent_map(em);
4400
4401         return 0;
4402 }
4403
4404 static void fill_device_from_item(struct extent_buffer *leaf,
4405                                  struct btrfs_dev_item *dev_item,
4406                                  struct btrfs_device *device)
4407 {
4408         unsigned long ptr;
4409
4410         device->devid = btrfs_device_id(leaf, dev_item);
4411         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4412         device->total_bytes = device->disk_total_bytes;
4413         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4414         device->type = btrfs_device_type(leaf, dev_item);
4415         device->io_align = btrfs_device_io_align(leaf, dev_item);
4416         device->io_width = btrfs_device_io_width(leaf, dev_item);
4417         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4418
4419         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4420         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4421 }
4422
4423 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4424 {
4425         struct btrfs_fs_devices *fs_devices;
4426         int ret;
4427
4428         BUG_ON(!mutex_is_locked(&uuid_mutex));
4429
4430         fs_devices = root->fs_info->fs_devices->seed;
4431         while (fs_devices) {
4432                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4433                         ret = 0;
4434                         goto out;
4435                 }
4436                 fs_devices = fs_devices->seed;
4437         }
4438
4439         fs_devices = find_fsid(fsid);
4440         if (!fs_devices) {
4441                 ret = -ENOENT;
4442                 goto out;
4443         }
4444
4445         fs_devices = clone_fs_devices(fs_devices);
4446         if (IS_ERR(fs_devices)) {
4447                 ret = PTR_ERR(fs_devices);
4448                 goto out;
4449         }
4450
4451         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4452                                    root->fs_info->bdev_holder);
4453         if (ret) {
4454                 free_fs_devices(fs_devices);
4455                 goto out;
4456         }
4457
4458         if (!fs_devices->seeding) {
4459                 __btrfs_close_devices(fs_devices);
4460                 free_fs_devices(fs_devices);
4461                 ret = -EINVAL;
4462                 goto out;
4463         }
4464
4465         fs_devices->seed = root->fs_info->fs_devices->seed;
4466         root->fs_info->fs_devices->seed = fs_devices;
4467 out:
4468         return ret;
4469 }
4470
4471 static int read_one_dev(struct btrfs_root *root,
4472                         struct extent_buffer *leaf,
4473                         struct btrfs_dev_item *dev_item)
4474 {
4475         struct btrfs_device *device;
4476         u64 devid;
4477         int ret;
4478         u8 fs_uuid[BTRFS_UUID_SIZE];
4479         u8 dev_uuid[BTRFS_UUID_SIZE];
4480
4481         devid = btrfs_device_id(leaf, dev_item);
4482         read_extent_buffer(leaf, dev_uuid,
4483                            (unsigned long)btrfs_device_uuid(dev_item),
4484                            BTRFS_UUID_SIZE);
4485         read_extent_buffer(leaf, fs_uuid,
4486                            (unsigned long)btrfs_device_fsid(dev_item),
4487                            BTRFS_UUID_SIZE);
4488
4489         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4490                 ret = open_seed_devices(root, fs_uuid);
4491                 if (ret && !btrfs_test_opt(root, DEGRADED))
4492                         return ret;
4493         }
4494
4495         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4496         if (!device || !device->bdev) {
4497                 if (!btrfs_test_opt(root, DEGRADED))
4498                         return -EIO;
4499
4500                 if (!device) {
4501                         printk(KERN_WARNING "warning devid %llu missing\n",
4502                                (unsigned long long)devid);
4503                         device = add_missing_dev(root, devid, dev_uuid);
4504                         if (!device)
4505                                 return -ENOMEM;
4506                 } else if (!device->missing) {
4507                         /*
4508                          * this happens when a device that was properly setup
4509                          * in the device info lists suddenly goes bad.
4510                          * device->bdev is NULL, and so we have to set
4511                          * device->missing to one here
4512                          */
4513                         root->fs_info->fs_devices->missing_devices++;
4514                         device->missing = 1;
4515                 }
4516         }
4517
4518         if (device->fs_devices != root->fs_info->fs_devices) {
4519                 BUG_ON(device->writeable);
4520                 if (device->generation !=
4521                     btrfs_device_generation(leaf, dev_item))
4522                         return -EINVAL;
4523         }
4524
4525         fill_device_from_item(leaf, dev_item, device);
4526         device->dev_root = root->fs_info->dev_root;
4527         device->in_fs_metadata = 1;
4528         if (device->writeable) {
4529                 device->fs_devices->total_rw_bytes += device->total_bytes;
4530                 spin_lock(&root->fs_info->free_chunk_lock);
4531                 root->fs_info->free_chunk_space += device->total_bytes -
4532                         device->bytes_used;
4533                 spin_unlock(&root->fs_info->free_chunk_lock);
4534         }
4535         ret = 0;
4536         return ret;
4537 }
4538
4539 int btrfs_read_sys_array(struct btrfs_root *root)
4540 {
4541         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4542         struct extent_buffer *sb;
4543         struct btrfs_disk_key *disk_key;
4544         struct btrfs_chunk *chunk;
4545         u8 *ptr;
4546         unsigned long sb_ptr;
4547         int ret = 0;
4548         u32 num_stripes;
4549         u32 array_size;
4550         u32 len = 0;
4551         u32 cur;
4552         struct btrfs_key key;
4553
4554         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4555                                           BTRFS_SUPER_INFO_SIZE);
4556         if (!sb)
4557                 return -ENOMEM;
4558         btrfs_set_buffer_uptodate(sb);
4559         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4560         /*
4561          * The sb extent buffer is artifical and just used to read the system array.
4562          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4563          * pages up-to-date when the page is larger: extent does not cover the
4564          * whole page and consequently check_page_uptodate does not find all
4565          * the page's extents up-to-date (the hole beyond sb),
4566          * write_extent_buffer then triggers a WARN_ON.
4567          *
4568          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4569          * but sb spans only this function. Add an explicit SetPageUptodate call
4570          * to silence the warning eg. on PowerPC 64.
4571          */
4572         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4573                 SetPageUptodate(sb->pages[0]);
4574
4575         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4576         array_size = btrfs_super_sys_array_size(super_copy);
4577
4578         ptr = super_copy->sys_chunk_array;
4579         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4580         cur = 0;
4581
4582         while (cur < array_size) {
4583                 disk_key = (struct btrfs_disk_key *)ptr;
4584                 btrfs_disk_key_to_cpu(&key, disk_key);
4585
4586                 len = sizeof(*disk_key); ptr += len;
4587                 sb_ptr += len;
4588                 cur += len;
4589
4590                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4591                         chunk = (struct btrfs_chunk *)sb_ptr;
4592                         ret = read_one_chunk(root, &key, sb, chunk);
4593                         if (ret)
4594                                 break;
4595                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4596                         len = btrfs_chunk_item_size(num_stripes);
4597                 } else {
4598                         ret = -EIO;
4599                         break;
4600                 }
4601                 ptr += len;
4602                 sb_ptr += len;
4603                 cur += len;
4604         }
4605         free_extent_buffer(sb);
4606         return ret;
4607 }
4608
4609 int btrfs_read_chunk_tree(struct btrfs_root *root)
4610 {
4611         struct btrfs_path *path;
4612         struct extent_buffer *leaf;
4613         struct btrfs_key key;
4614         struct btrfs_key found_key;
4615         int ret;
4616         int slot;
4617
4618         root = root->fs_info->chunk_root;
4619
4620         path = btrfs_alloc_path();
4621         if (!path)
4622                 return -ENOMEM;
4623
4624         mutex_lock(&uuid_mutex);
4625         lock_chunks(root);
4626
4627         /* first we search for all of the device items, and then we
4628          * read in all of the chunk items.  This way we can create chunk
4629          * mappings that reference all of the devices that are afound
4630          */
4631         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4632         key.offset = 0;
4633         key.type = 0;
4634 again:
4635         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4636         if (ret < 0)
4637                 goto error;
4638         while (1) {
4639                 leaf = path->nodes[0];
4640                 slot = path->slots[0];
4641                 if (slot >= btrfs_header_nritems(leaf)) {
4642                         ret = btrfs_next_leaf(root, path);
4643                         if (ret == 0)
4644                                 continue;
4645                         if (ret < 0)
4646                                 goto error;
4647                         break;
4648                 }
4649                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4650                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4651                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4652                                 break;
4653                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4654                                 struct btrfs_dev_item *dev_item;
4655                                 dev_item = btrfs_item_ptr(leaf, slot,
4656                                                   struct btrfs_dev_item);
4657                                 ret = read_one_dev(root, leaf, dev_item);
4658                                 if (ret)
4659                                         goto error;
4660                         }
4661                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4662                         struct btrfs_chunk *chunk;
4663                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4664                         ret = read_one_chunk(root, &found_key, leaf, chunk);
4665                         if (ret)
4666                                 goto error;
4667                 }
4668                 path->slots[0]++;
4669         }
4670         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4671                 key.objectid = 0;
4672                 btrfs_release_path(path);
4673                 goto again;
4674         }
4675         ret = 0;
4676 error:
4677         unlock_chunks(root);
4678         mutex_unlock(&uuid_mutex);
4679
4680         btrfs_free_path(path);
4681         return ret;
4682 }
4683
4684 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
4685 {
4686         int i;
4687
4688         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4689                 btrfs_dev_stat_reset(dev, i);
4690 }
4691
4692 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
4693 {
4694         struct btrfs_key key;
4695         struct btrfs_key found_key;
4696         struct btrfs_root *dev_root = fs_info->dev_root;
4697         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4698         struct extent_buffer *eb;
4699         int slot;
4700         int ret = 0;
4701         struct btrfs_device *device;
4702         struct btrfs_path *path = NULL;
4703         int i;
4704
4705         path = btrfs_alloc_path();
4706         if (!path) {
4707                 ret = -ENOMEM;
4708                 goto out;
4709         }
4710
4711         mutex_lock(&fs_devices->device_list_mutex);
4712         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4713                 int item_size;
4714                 struct btrfs_dev_stats_item *ptr;
4715
4716                 key.objectid = 0;
4717                 key.type = BTRFS_DEV_STATS_KEY;
4718                 key.offset = device->devid;
4719                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
4720                 if (ret) {
4721                         __btrfs_reset_dev_stats(device);
4722                         device->dev_stats_valid = 1;
4723                         btrfs_release_path(path);
4724                         continue;
4725                 }
4726                 slot = path->slots[0];
4727                 eb = path->nodes[0];
4728                 btrfs_item_key_to_cpu(eb, &found_key, slot);
4729                 item_size = btrfs_item_size_nr(eb, slot);
4730
4731                 ptr = btrfs_item_ptr(eb, slot,
4732                                      struct btrfs_dev_stats_item);
4733
4734                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4735                         if (item_size >= (1 + i) * sizeof(__le64))
4736                                 btrfs_dev_stat_set(device, i,
4737                                         btrfs_dev_stats_value(eb, ptr, i));
4738                         else
4739                                 btrfs_dev_stat_reset(device, i);
4740                 }
4741
4742                 device->dev_stats_valid = 1;
4743                 btrfs_dev_stat_print_on_load(device);
4744                 btrfs_release_path(path);
4745         }
4746         mutex_unlock(&fs_devices->device_list_mutex);
4747
4748 out:
4749         btrfs_free_path(path);
4750         return ret < 0 ? ret : 0;
4751 }
4752
4753 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
4754                                 struct btrfs_root *dev_root,
4755                                 struct btrfs_device *device)
4756 {
4757         struct btrfs_path *path;
4758         struct btrfs_key key;
4759         struct extent_buffer *eb;
4760         struct btrfs_dev_stats_item *ptr;
4761         int ret;
4762         int i;
4763
4764         key.objectid = 0;
4765         key.type = BTRFS_DEV_STATS_KEY;
4766         key.offset = device->devid;
4767
4768         path = btrfs_alloc_path();
4769         BUG_ON(!path);
4770         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
4771         if (ret < 0) {
4772                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
4773                               ret, rcu_str_deref(device->name));
4774                 goto out;
4775         }
4776
4777         if (ret == 0 &&
4778             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
4779                 /* need to delete old one and insert a new one */
4780                 ret = btrfs_del_item(trans, dev_root, path);
4781                 if (ret != 0) {
4782                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
4783                                       rcu_str_deref(device->name), ret);
4784                         goto out;
4785                 }
4786                 ret = 1;
4787         }
4788
4789         if (ret == 1) {
4790                 /* need to insert a new item */
4791                 btrfs_release_path(path);
4792                 ret = btrfs_insert_empty_item(trans, dev_root, path,
4793                                               &key, sizeof(*ptr));
4794                 if (ret < 0) {
4795                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
4796                                       rcu_str_deref(device->name), ret);
4797                         goto out;
4798                 }
4799         }
4800
4801         eb = path->nodes[0];
4802         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
4803         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4804                 btrfs_set_dev_stats_value(eb, ptr, i,
4805                                           btrfs_dev_stat_read(device, i));
4806         btrfs_mark_buffer_dirty(eb);
4807
4808 out:
4809         btrfs_free_path(path);
4810         return ret;
4811 }
4812
4813 /*
4814  * called from commit_transaction. Writes all changed device stats to disk.
4815  */
4816 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
4817                         struct btrfs_fs_info *fs_info)
4818 {
4819         struct btrfs_root *dev_root = fs_info->dev_root;
4820         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4821         struct btrfs_device *device;
4822         int ret = 0;
4823
4824         mutex_lock(&fs_devices->device_list_mutex);
4825         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4826                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
4827                         continue;
4828
4829                 ret = update_dev_stat_item(trans, dev_root, device);
4830                 if (!ret)
4831                         device->dev_stats_dirty = 0;
4832         }
4833         mutex_unlock(&fs_devices->device_list_mutex);
4834
4835         return ret;
4836 }
4837
4838 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
4839 {
4840         btrfs_dev_stat_inc(dev, index);
4841         btrfs_dev_stat_print_on_error(dev);
4842 }
4843
4844 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
4845 {
4846         if (!dev->dev_stats_valid)
4847                 return;
4848         printk_ratelimited_in_rcu(KERN_ERR
4849                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4850                            rcu_str_deref(dev->name),
4851                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4852                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4853                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4854                            btrfs_dev_stat_read(dev,
4855                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
4856                            btrfs_dev_stat_read(dev,
4857                                                BTRFS_DEV_STAT_GENERATION_ERRS));
4858 }
4859
4860 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
4861 {
4862         int i;
4863
4864         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4865                 if (btrfs_dev_stat_read(dev, i) != 0)
4866                         break;
4867         if (i == BTRFS_DEV_STAT_VALUES_MAX)
4868                 return; /* all values == 0, suppress message */
4869
4870         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4871                rcu_str_deref(dev->name),
4872                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4873                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4874                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4875                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
4876                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
4877 }
4878
4879 int btrfs_get_dev_stats(struct btrfs_root *root,
4880                         struct btrfs_ioctl_get_dev_stats *stats)
4881 {
4882         struct btrfs_device *dev;
4883         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4884         int i;
4885
4886         mutex_lock(&fs_devices->device_list_mutex);
4887         dev = btrfs_find_device(root, stats->devid, NULL, NULL);
4888         mutex_unlock(&fs_devices->device_list_mutex);
4889
4890         if (!dev) {
4891                 printk(KERN_WARNING
4892                        "btrfs: get dev_stats failed, device not found\n");
4893                 return -ENODEV;
4894         } else if (!dev->dev_stats_valid) {
4895                 printk(KERN_WARNING
4896                        "btrfs: get dev_stats failed, not yet valid\n");
4897                 return -ENODEV;
4898         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
4899                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4900                         if (stats->nr_items > i)
4901                                 stats->values[i] =
4902                                         btrfs_dev_stat_read_and_reset(dev, i);
4903                         else
4904                                 btrfs_dev_stat_reset(dev, i);
4905                 }
4906         } else {
4907                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4908                         if (stats->nr_items > i)
4909                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
4910         }
4911         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
4912                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
4913         return 0;
4914 }