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 "btrfs: 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         root->fs_info->num_tolerated_disk_barrier_failures =
1479                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1480
1481         /*
1482          * at this point, the device is zero sized.  We want to
1483          * remove it from the devices list and zero out the old super
1484          */
1485         if (clear_super) {
1486                 /* make sure this device isn't detected as part of
1487                  * the FS anymore
1488                  */
1489                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1490                 set_buffer_dirty(bh);
1491                 sync_dirty_buffer(bh);
1492         }
1493
1494         ret = 0;
1495
1496 error_brelse:
1497         brelse(bh);
1498 error_close:
1499         if (bdev)
1500                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1501 out:
1502         mutex_unlock(&uuid_mutex);
1503         return ret;
1504 error_undo:
1505         if (device->writeable) {
1506                 lock_chunks(root);
1507                 list_add(&device->dev_alloc_list,
1508                          &root->fs_info->fs_devices->alloc_list);
1509                 unlock_chunks(root);
1510                 root->fs_info->fs_devices->rw_devices++;
1511         }
1512         goto error_brelse;
1513 }
1514
1515 /*
1516  * does all the dirty work required for changing file system's UUID.
1517  */
1518 static int btrfs_prepare_sprout(struct btrfs_root *root)
1519 {
1520         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1521         struct btrfs_fs_devices *old_devices;
1522         struct btrfs_fs_devices *seed_devices;
1523         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1524         struct btrfs_device *device;
1525         u64 super_flags;
1526
1527         BUG_ON(!mutex_is_locked(&uuid_mutex));
1528         if (!fs_devices->seeding)
1529                 return -EINVAL;
1530
1531         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1532         if (!seed_devices)
1533                 return -ENOMEM;
1534
1535         old_devices = clone_fs_devices(fs_devices);
1536         if (IS_ERR(old_devices)) {
1537                 kfree(seed_devices);
1538                 return PTR_ERR(old_devices);
1539         }
1540
1541         list_add(&old_devices->list, &fs_uuids);
1542
1543         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1544         seed_devices->opened = 1;
1545         INIT_LIST_HEAD(&seed_devices->devices);
1546         INIT_LIST_HEAD(&seed_devices->alloc_list);
1547         mutex_init(&seed_devices->device_list_mutex);
1548
1549         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1550         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1551                               synchronize_rcu);
1552         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1553
1554         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1555         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1556                 device->fs_devices = seed_devices;
1557         }
1558
1559         fs_devices->seeding = 0;
1560         fs_devices->num_devices = 0;
1561         fs_devices->open_devices = 0;
1562         fs_devices->total_devices = 0;
1563         fs_devices->seed = seed_devices;
1564
1565         generate_random_uuid(fs_devices->fsid);
1566         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1567         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1568         super_flags = btrfs_super_flags(disk_super) &
1569                       ~BTRFS_SUPER_FLAG_SEEDING;
1570         btrfs_set_super_flags(disk_super, super_flags);
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * strore the expected generation for seed devices in device items.
1577  */
1578 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1579                                struct btrfs_root *root)
1580 {
1581         struct btrfs_path *path;
1582         struct extent_buffer *leaf;
1583         struct btrfs_dev_item *dev_item;
1584         struct btrfs_device *device;
1585         struct btrfs_key key;
1586         u8 fs_uuid[BTRFS_UUID_SIZE];
1587         u8 dev_uuid[BTRFS_UUID_SIZE];
1588         u64 devid;
1589         int ret;
1590
1591         path = btrfs_alloc_path();
1592         if (!path)
1593                 return -ENOMEM;
1594
1595         root = root->fs_info->chunk_root;
1596         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1597         key.offset = 0;
1598         key.type = BTRFS_DEV_ITEM_KEY;
1599
1600         while (1) {
1601                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1602                 if (ret < 0)
1603                         goto error;
1604
1605                 leaf = path->nodes[0];
1606 next_slot:
1607                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1608                         ret = btrfs_next_leaf(root, path);
1609                         if (ret > 0)
1610                                 break;
1611                         if (ret < 0)
1612                                 goto error;
1613                         leaf = path->nodes[0];
1614                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1615                         btrfs_release_path(path);
1616                         continue;
1617                 }
1618
1619                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1620                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1621                     key.type != BTRFS_DEV_ITEM_KEY)
1622                         break;
1623
1624                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1625                                           struct btrfs_dev_item);
1626                 devid = btrfs_device_id(leaf, dev_item);
1627                 read_extent_buffer(leaf, dev_uuid,
1628                                    (unsigned long)btrfs_device_uuid(dev_item),
1629                                    BTRFS_UUID_SIZE);
1630                 read_extent_buffer(leaf, fs_uuid,
1631                                    (unsigned long)btrfs_device_fsid(dev_item),
1632                                    BTRFS_UUID_SIZE);
1633                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1634                 BUG_ON(!device); /* Logic error */
1635
1636                 if (device->fs_devices->seeding) {
1637                         btrfs_set_device_generation(leaf, dev_item,
1638                                                     device->generation);
1639                         btrfs_mark_buffer_dirty(leaf);
1640                 }
1641
1642                 path->slots[0]++;
1643                 goto next_slot;
1644         }
1645         ret = 0;
1646 error:
1647         btrfs_free_path(path);
1648         return ret;
1649 }
1650
1651 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1652 {
1653         struct request_queue *q;
1654         struct btrfs_trans_handle *trans;
1655         struct btrfs_device *device;
1656         struct block_device *bdev;
1657         struct list_head *devices;
1658         struct super_block *sb = root->fs_info->sb;
1659         struct rcu_string *name;
1660         u64 total_bytes;
1661         int seeding_dev = 0;
1662         int ret = 0;
1663
1664         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1665                 return -EROFS;
1666
1667         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1668                                   root->fs_info->bdev_holder);
1669         if (IS_ERR(bdev))
1670                 return PTR_ERR(bdev);
1671
1672         if (root->fs_info->fs_devices->seeding) {
1673                 seeding_dev = 1;
1674                 down_write(&sb->s_umount);
1675                 mutex_lock(&uuid_mutex);
1676         }
1677
1678         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1679
1680         devices = &root->fs_info->fs_devices->devices;
1681         /*
1682          * we have the volume lock, so we don't need the extra
1683          * device list mutex while reading the list here.
1684          */
1685         list_for_each_entry(device, devices, dev_list) {
1686                 if (device->bdev == bdev) {
1687                         ret = -EEXIST;
1688                         goto error;
1689                 }
1690         }
1691
1692         device = kzalloc(sizeof(*device), GFP_NOFS);
1693         if (!device) {
1694                 /* we can safely leave the fs_devices entry around */
1695                 ret = -ENOMEM;
1696                 goto error;
1697         }
1698
1699         name = rcu_string_strdup(device_path, GFP_NOFS);
1700         if (!name) {
1701                 kfree(device);
1702                 ret = -ENOMEM;
1703                 goto error;
1704         }
1705         rcu_assign_pointer(device->name, name);
1706
1707         ret = find_next_devid(root, &device->devid);
1708         if (ret) {
1709                 rcu_string_free(device->name);
1710                 kfree(device);
1711                 goto error;
1712         }
1713
1714         trans = btrfs_start_transaction(root, 0);
1715         if (IS_ERR(trans)) {
1716                 rcu_string_free(device->name);
1717                 kfree(device);
1718                 ret = PTR_ERR(trans);
1719                 goto error;
1720         }
1721
1722         lock_chunks(root);
1723
1724         q = bdev_get_queue(bdev);
1725         if (blk_queue_discard(q))
1726                 device->can_discard = 1;
1727         device->writeable = 1;
1728         device->work.func = pending_bios_fn;
1729         generate_random_uuid(device->uuid);
1730         spin_lock_init(&device->io_lock);
1731         device->generation = trans->transid;
1732         device->io_width = root->sectorsize;
1733         device->io_align = root->sectorsize;
1734         device->sector_size = root->sectorsize;
1735         device->total_bytes = i_size_read(bdev->bd_inode);
1736         device->disk_total_bytes = device->total_bytes;
1737         device->dev_root = root->fs_info->dev_root;
1738         device->bdev = bdev;
1739         device->in_fs_metadata = 1;
1740         device->mode = FMODE_EXCL;
1741         set_blocksize(device->bdev, 4096);
1742
1743         if (seeding_dev) {
1744                 sb->s_flags &= ~MS_RDONLY;
1745                 ret = btrfs_prepare_sprout(root);
1746                 BUG_ON(ret); /* -ENOMEM */
1747         }
1748
1749         device->fs_devices = root->fs_info->fs_devices;
1750
1751         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1752         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1753         list_add(&device->dev_alloc_list,
1754                  &root->fs_info->fs_devices->alloc_list);
1755         root->fs_info->fs_devices->num_devices++;
1756         root->fs_info->fs_devices->open_devices++;
1757         root->fs_info->fs_devices->rw_devices++;
1758         root->fs_info->fs_devices->total_devices++;
1759         if (device->can_discard)
1760                 root->fs_info->fs_devices->num_can_discard++;
1761         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1762
1763         spin_lock(&root->fs_info->free_chunk_lock);
1764         root->fs_info->free_chunk_space += device->total_bytes;
1765         spin_unlock(&root->fs_info->free_chunk_lock);
1766
1767         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1768                 root->fs_info->fs_devices->rotating = 1;
1769
1770         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1771         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1772                                     total_bytes + device->total_bytes);
1773
1774         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1775         btrfs_set_super_num_devices(root->fs_info->super_copy,
1776                                     total_bytes + 1);
1777         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1778
1779         if (seeding_dev) {
1780                 ret = init_first_rw_device(trans, root, device);
1781                 if (ret) {
1782                         btrfs_abort_transaction(trans, root, ret);
1783                         goto error_trans;
1784                 }
1785                 ret = btrfs_finish_sprout(trans, root);
1786                 if (ret) {
1787                         btrfs_abort_transaction(trans, root, ret);
1788                         goto error_trans;
1789                 }
1790         } else {
1791                 ret = btrfs_add_device(trans, root, device);
1792                 if (ret) {
1793                         btrfs_abort_transaction(trans, root, ret);
1794                         goto error_trans;
1795                 }
1796         }
1797
1798         /*
1799          * we've got more storage, clear any full flags on the space
1800          * infos
1801          */
1802         btrfs_clear_space_info_full(root->fs_info);
1803
1804         unlock_chunks(root);
1805         root->fs_info->num_tolerated_disk_barrier_failures =
1806                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1807         ret = btrfs_commit_transaction(trans, root);
1808
1809         if (seeding_dev) {
1810                 mutex_unlock(&uuid_mutex);
1811                 up_write(&sb->s_umount);
1812
1813                 if (ret) /* transaction commit */
1814                         return ret;
1815
1816                 ret = btrfs_relocate_sys_chunks(root);
1817                 if (ret < 0)
1818                         btrfs_error(root->fs_info, ret,
1819                                     "Failed to relocate sys chunks after "
1820                                     "device initialization. This can be fixed "
1821                                     "using the \"btrfs balance\" command.");
1822                 trans = btrfs_attach_transaction(root);
1823                 if (IS_ERR(trans)) {
1824                         if (PTR_ERR(trans) == -ENOENT)
1825                                 return 0;
1826                         return PTR_ERR(trans);
1827                 }
1828                 ret = btrfs_commit_transaction(trans, root);
1829         }
1830
1831         return ret;
1832
1833 error_trans:
1834         unlock_chunks(root);
1835         btrfs_end_transaction(trans, root);
1836         rcu_string_free(device->name);
1837         kfree(device);
1838 error:
1839         blkdev_put(bdev, FMODE_EXCL);
1840         if (seeding_dev) {
1841                 mutex_unlock(&uuid_mutex);
1842                 up_write(&sb->s_umount);
1843         }
1844         return ret;
1845 }
1846
1847 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1848                                         struct btrfs_device *device)
1849 {
1850         int ret;
1851         struct btrfs_path *path;
1852         struct btrfs_root *root;
1853         struct btrfs_dev_item *dev_item;
1854         struct extent_buffer *leaf;
1855         struct btrfs_key key;
1856
1857         root = device->dev_root->fs_info->chunk_root;
1858
1859         path = btrfs_alloc_path();
1860         if (!path)
1861                 return -ENOMEM;
1862
1863         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1864         key.type = BTRFS_DEV_ITEM_KEY;
1865         key.offset = device->devid;
1866
1867         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1868         if (ret < 0)
1869                 goto out;
1870
1871         if (ret > 0) {
1872                 ret = -ENOENT;
1873                 goto out;
1874         }
1875
1876         leaf = path->nodes[0];
1877         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1878
1879         btrfs_set_device_id(leaf, dev_item, device->devid);
1880         btrfs_set_device_type(leaf, dev_item, device->type);
1881         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1882         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1883         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1884         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1885         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1886         btrfs_mark_buffer_dirty(leaf);
1887
1888 out:
1889         btrfs_free_path(path);
1890         return ret;
1891 }
1892
1893 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1894                       struct btrfs_device *device, u64 new_size)
1895 {
1896         struct btrfs_super_block *super_copy =
1897                 device->dev_root->fs_info->super_copy;
1898         u64 old_total = btrfs_super_total_bytes(super_copy);
1899         u64 diff = new_size - device->total_bytes;
1900
1901         if (!device->writeable)
1902                 return -EACCES;
1903         if (new_size <= device->total_bytes)
1904                 return -EINVAL;
1905
1906         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1907         device->fs_devices->total_rw_bytes += diff;
1908
1909         device->total_bytes = new_size;
1910         device->disk_total_bytes = new_size;
1911         btrfs_clear_space_info_full(device->dev_root->fs_info);
1912
1913         return btrfs_update_device(trans, device);
1914 }
1915
1916 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1917                       struct btrfs_device *device, u64 new_size)
1918 {
1919         int ret;
1920         lock_chunks(device->dev_root);
1921         ret = __btrfs_grow_device(trans, device, new_size);
1922         unlock_chunks(device->dev_root);
1923         return ret;
1924 }
1925
1926 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1927                             struct btrfs_root *root,
1928                             u64 chunk_tree, u64 chunk_objectid,
1929                             u64 chunk_offset)
1930 {
1931         int ret;
1932         struct btrfs_path *path;
1933         struct btrfs_key key;
1934
1935         root = root->fs_info->chunk_root;
1936         path = btrfs_alloc_path();
1937         if (!path)
1938                 return -ENOMEM;
1939
1940         key.objectid = chunk_objectid;
1941         key.offset = chunk_offset;
1942         key.type = BTRFS_CHUNK_ITEM_KEY;
1943
1944         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1945         if (ret < 0)
1946                 goto out;
1947         else if (ret > 0) { /* Logic error or corruption */
1948                 btrfs_error(root->fs_info, -ENOENT,
1949                             "Failed lookup while freeing chunk.");
1950                 ret = -ENOENT;
1951                 goto out;
1952         }
1953
1954         ret = btrfs_del_item(trans, root, path);
1955         if (ret < 0)
1956                 btrfs_error(root->fs_info, ret,
1957                             "Failed to delete chunk item.");
1958 out:
1959         btrfs_free_path(path);
1960         return ret;
1961 }
1962
1963 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1964                         chunk_offset)
1965 {
1966         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1967         struct btrfs_disk_key *disk_key;
1968         struct btrfs_chunk *chunk;
1969         u8 *ptr;
1970         int ret = 0;
1971         u32 num_stripes;
1972         u32 array_size;
1973         u32 len = 0;
1974         u32 cur;
1975         struct btrfs_key key;
1976
1977         array_size = btrfs_super_sys_array_size(super_copy);
1978
1979         ptr = super_copy->sys_chunk_array;
1980         cur = 0;
1981
1982         while (cur < array_size) {
1983                 disk_key = (struct btrfs_disk_key *)ptr;
1984                 btrfs_disk_key_to_cpu(&key, disk_key);
1985
1986                 len = sizeof(*disk_key);
1987
1988                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1989                         chunk = (struct btrfs_chunk *)(ptr + len);
1990                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1991                         len += btrfs_chunk_item_size(num_stripes);
1992                 } else {
1993                         ret = -EIO;
1994                         break;
1995                 }
1996                 if (key.objectid == chunk_objectid &&
1997                     key.offset == chunk_offset) {
1998                         memmove(ptr, ptr + len, array_size - (cur + len));
1999                         array_size -= len;
2000                         btrfs_set_super_sys_array_size(super_copy, array_size);
2001                 } else {
2002                         ptr += len;
2003                         cur += len;
2004                 }
2005         }
2006         return ret;
2007 }
2008
2009 static int btrfs_relocate_chunk(struct btrfs_root *root,
2010                          u64 chunk_tree, u64 chunk_objectid,
2011                          u64 chunk_offset)
2012 {
2013         struct extent_map_tree *em_tree;
2014         struct btrfs_root *extent_root;
2015         struct btrfs_trans_handle *trans;
2016         struct extent_map *em;
2017         struct map_lookup *map;
2018         int ret;
2019         int i;
2020
2021         root = root->fs_info->chunk_root;
2022         extent_root = root->fs_info->extent_root;
2023         em_tree = &root->fs_info->mapping_tree.map_tree;
2024
2025         ret = btrfs_can_relocate(extent_root, chunk_offset);
2026         if (ret)
2027                 return -ENOSPC;
2028
2029         /* step one, relocate all the extents inside this chunk */
2030         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2031         if (ret)
2032                 return ret;
2033
2034         trans = btrfs_start_transaction(root, 0);
2035         BUG_ON(IS_ERR(trans));
2036
2037         lock_chunks(root);
2038
2039         /*
2040          * step two, delete the device extents and the
2041          * chunk tree entries
2042          */
2043         read_lock(&em_tree->lock);
2044         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2045         read_unlock(&em_tree->lock);
2046
2047         BUG_ON(!em || em->start > chunk_offset ||
2048                em->start + em->len < chunk_offset);
2049         map = (struct map_lookup *)em->bdev;
2050
2051         for (i = 0; i < map->num_stripes; i++) {
2052                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2053                                             map->stripes[i].physical);
2054                 BUG_ON(ret);
2055
2056                 if (map->stripes[i].dev) {
2057                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2058                         BUG_ON(ret);
2059                 }
2060         }
2061         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2062                                chunk_offset);
2063
2064         BUG_ON(ret);
2065
2066         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2067
2068         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2069                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2070                 BUG_ON(ret);
2071         }
2072
2073         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2074         BUG_ON(ret);
2075
2076         write_lock(&em_tree->lock);
2077         remove_extent_mapping(em_tree, em);
2078         write_unlock(&em_tree->lock);
2079
2080         kfree(map);
2081         em->bdev = NULL;
2082
2083         /* once for the tree */
2084         free_extent_map(em);
2085         /* once for us */
2086         free_extent_map(em);
2087
2088         unlock_chunks(root);
2089         btrfs_end_transaction(trans, root);
2090         return 0;
2091 }
2092
2093 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2094 {
2095         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2096         struct btrfs_path *path;
2097         struct extent_buffer *leaf;
2098         struct btrfs_chunk *chunk;
2099         struct btrfs_key key;
2100         struct btrfs_key found_key;
2101         u64 chunk_tree = chunk_root->root_key.objectid;
2102         u64 chunk_type;
2103         bool retried = false;
2104         int failed = 0;
2105         int ret;
2106
2107         path = btrfs_alloc_path();
2108         if (!path)
2109                 return -ENOMEM;
2110
2111 again:
2112         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2113         key.offset = (u64)-1;
2114         key.type = BTRFS_CHUNK_ITEM_KEY;
2115
2116         while (1) {
2117                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2118                 if (ret < 0)
2119                         goto error;
2120                 BUG_ON(ret == 0); /* Corruption */
2121
2122                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2123                                           key.type);
2124                 if (ret < 0)
2125                         goto error;
2126                 if (ret > 0)
2127                         break;
2128
2129                 leaf = path->nodes[0];
2130                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2131
2132                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2133                                        struct btrfs_chunk);
2134                 chunk_type = btrfs_chunk_type(leaf, chunk);
2135                 btrfs_release_path(path);
2136
2137                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2138                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2139                                                    found_key.objectid,
2140                                                    found_key.offset);
2141                         if (ret == -ENOSPC)
2142                                 failed++;
2143                         else if (ret)
2144                                 BUG();
2145                 }
2146
2147                 if (found_key.offset == 0)
2148                         break;
2149                 key.offset = found_key.offset - 1;
2150         }
2151         ret = 0;
2152         if (failed && !retried) {
2153                 failed = 0;
2154                 retried = true;
2155                 goto again;
2156         } else if (failed && retried) {
2157                 WARN_ON(1);
2158                 ret = -ENOSPC;
2159         }
2160 error:
2161         btrfs_free_path(path);
2162         return ret;
2163 }
2164
2165 static int insert_balance_item(struct btrfs_root *root,
2166                                struct btrfs_balance_control *bctl)
2167 {
2168         struct btrfs_trans_handle *trans;
2169         struct btrfs_balance_item *item;
2170         struct btrfs_disk_balance_args disk_bargs;
2171         struct btrfs_path *path;
2172         struct extent_buffer *leaf;
2173         struct btrfs_key key;
2174         int ret, err;
2175
2176         path = btrfs_alloc_path();
2177         if (!path)
2178                 return -ENOMEM;
2179
2180         trans = btrfs_start_transaction(root, 0);
2181         if (IS_ERR(trans)) {
2182                 btrfs_free_path(path);
2183                 return PTR_ERR(trans);
2184         }
2185
2186         key.objectid = BTRFS_BALANCE_OBJECTID;
2187         key.type = BTRFS_BALANCE_ITEM_KEY;
2188         key.offset = 0;
2189
2190         ret = btrfs_insert_empty_item(trans, root, path, &key,
2191                                       sizeof(*item));
2192         if (ret)
2193                 goto out;
2194
2195         leaf = path->nodes[0];
2196         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2197
2198         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2199
2200         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2201         btrfs_set_balance_data(leaf, item, &disk_bargs);
2202         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2203         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2204         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2205         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2206
2207         btrfs_set_balance_flags(leaf, item, bctl->flags);
2208
2209         btrfs_mark_buffer_dirty(leaf);
2210 out:
2211         btrfs_free_path(path);
2212         err = btrfs_commit_transaction(trans, root);
2213         if (err && !ret)
2214                 ret = err;
2215         return ret;
2216 }
2217
2218 static int del_balance_item(struct btrfs_root *root)
2219 {
2220         struct btrfs_trans_handle *trans;
2221         struct btrfs_path *path;
2222         struct btrfs_key key;
2223         int ret, err;
2224
2225         path = btrfs_alloc_path();
2226         if (!path)
2227                 return -ENOMEM;
2228
2229         trans = btrfs_start_transaction(root, 0);
2230         if (IS_ERR(trans)) {
2231                 btrfs_free_path(path);
2232                 return PTR_ERR(trans);
2233         }
2234
2235         key.objectid = BTRFS_BALANCE_OBJECTID;
2236         key.type = BTRFS_BALANCE_ITEM_KEY;
2237         key.offset = 0;
2238
2239         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2240         if (ret < 0)
2241                 goto out;
2242         if (ret > 0) {
2243                 ret = -ENOENT;
2244                 goto out;
2245         }
2246
2247         ret = btrfs_del_item(trans, root, path);
2248 out:
2249         btrfs_free_path(path);
2250         err = btrfs_commit_transaction(trans, root);
2251         if (err && !ret)
2252                 ret = err;
2253         return ret;
2254 }
2255
2256 /*
2257  * This is a heuristic used to reduce the number of chunks balanced on
2258  * resume after balance was interrupted.
2259  */
2260 static void update_balance_args(struct btrfs_balance_control *bctl)
2261 {
2262         /*
2263          * Turn on soft mode for chunk types that were being converted.
2264          */
2265         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2266                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2267         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2268                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2269         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2270                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2271
2272         /*
2273          * Turn on usage filter if is not already used.  The idea is
2274          * that chunks that we have already balanced should be
2275          * reasonably full.  Don't do it for chunks that are being
2276          * converted - that will keep us from relocating unconverted
2277          * (albeit full) chunks.
2278          */
2279         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2280             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2281                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2282                 bctl->data.usage = 90;
2283         }
2284         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2285             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2286                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2287                 bctl->sys.usage = 90;
2288         }
2289         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2290             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2291                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2292                 bctl->meta.usage = 90;
2293         }
2294 }
2295
2296 /*
2297  * Should be called with both balance and volume mutexes held to
2298  * serialize other volume operations (add_dev/rm_dev/resize) with
2299  * restriper.  Same goes for unset_balance_control.
2300  */
2301 static void set_balance_control(struct btrfs_balance_control *bctl)
2302 {
2303         struct btrfs_fs_info *fs_info = bctl->fs_info;
2304
2305         BUG_ON(fs_info->balance_ctl);
2306
2307         spin_lock(&fs_info->balance_lock);
2308         fs_info->balance_ctl = bctl;
2309         spin_unlock(&fs_info->balance_lock);
2310 }
2311
2312 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2313 {
2314         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2315
2316         BUG_ON(!fs_info->balance_ctl);
2317
2318         spin_lock(&fs_info->balance_lock);
2319         fs_info->balance_ctl = NULL;
2320         spin_unlock(&fs_info->balance_lock);
2321
2322         kfree(bctl);
2323 }
2324
2325 /*
2326  * Balance filters.  Return 1 if chunk should be filtered out
2327  * (should not be balanced).
2328  */
2329 static int chunk_profiles_filter(u64 chunk_type,
2330                                  struct btrfs_balance_args *bargs)
2331 {
2332         chunk_type = chunk_to_extended(chunk_type) &
2333                                 BTRFS_EXTENDED_PROFILE_MASK;
2334
2335         if (bargs->profiles & chunk_type)
2336                 return 0;
2337
2338         return 1;
2339 }
2340
2341 static u64 div_factor_fine(u64 num, int factor)
2342 {
2343         if (factor <= 0)
2344                 return 0;
2345         if (factor >= 100)
2346                 return num;
2347
2348         num *= factor;
2349         do_div(num, 100);
2350         return num;
2351 }
2352
2353 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2354                               struct btrfs_balance_args *bargs)
2355 {
2356         struct btrfs_block_group_cache *cache;
2357         u64 chunk_used, user_thresh;
2358         int ret = 1;
2359
2360         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2361         chunk_used = btrfs_block_group_used(&cache->item);
2362
2363         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2364         if (chunk_used < user_thresh)
2365                 ret = 0;
2366
2367         btrfs_put_block_group(cache);
2368         return ret;
2369 }
2370
2371 static int chunk_devid_filter(struct extent_buffer *leaf,
2372                               struct btrfs_chunk *chunk,
2373                               struct btrfs_balance_args *bargs)
2374 {
2375         struct btrfs_stripe *stripe;
2376         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2377         int i;
2378
2379         for (i = 0; i < num_stripes; i++) {
2380                 stripe = btrfs_stripe_nr(chunk, i);
2381                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2382                         return 0;
2383         }
2384
2385         return 1;
2386 }
2387
2388 /* [pstart, pend) */
2389 static int chunk_drange_filter(struct extent_buffer *leaf,
2390                                struct btrfs_chunk *chunk,
2391                                u64 chunk_offset,
2392                                struct btrfs_balance_args *bargs)
2393 {
2394         struct btrfs_stripe *stripe;
2395         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2396         u64 stripe_offset;
2397         u64 stripe_length;
2398         int factor;
2399         int i;
2400
2401         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2402                 return 0;
2403
2404         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2405              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2406                 factor = 2;
2407         else
2408                 factor = 1;
2409         factor = num_stripes / factor;
2410
2411         for (i = 0; i < num_stripes; i++) {
2412                 stripe = btrfs_stripe_nr(chunk, i);
2413                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2414                         continue;
2415
2416                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2417                 stripe_length = btrfs_chunk_length(leaf, chunk);
2418                 do_div(stripe_length, factor);
2419
2420                 if (stripe_offset < bargs->pend &&
2421                     stripe_offset + stripe_length > bargs->pstart)
2422                         return 0;
2423         }
2424
2425         return 1;
2426 }
2427
2428 /* [vstart, vend) */
2429 static int chunk_vrange_filter(struct extent_buffer *leaf,
2430                                struct btrfs_chunk *chunk,
2431                                u64 chunk_offset,
2432                                struct btrfs_balance_args *bargs)
2433 {
2434         if (chunk_offset < bargs->vend &&
2435             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2436                 /* at least part of the chunk is inside this vrange */
2437                 return 0;
2438
2439         return 1;
2440 }
2441
2442 static int chunk_soft_convert_filter(u64 chunk_type,
2443                                      struct btrfs_balance_args *bargs)
2444 {
2445         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2446                 return 0;
2447
2448         chunk_type = chunk_to_extended(chunk_type) &
2449                                 BTRFS_EXTENDED_PROFILE_MASK;
2450
2451         if (bargs->target == chunk_type)
2452                 return 1;
2453
2454         return 0;
2455 }
2456
2457 static int should_balance_chunk(struct btrfs_root *root,
2458                                 struct extent_buffer *leaf,
2459                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2460 {
2461         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2462         struct btrfs_balance_args *bargs = NULL;
2463         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2464
2465         /* type filter */
2466         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2467               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2468                 return 0;
2469         }
2470
2471         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2472                 bargs = &bctl->data;
2473         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2474                 bargs = &bctl->sys;
2475         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2476                 bargs = &bctl->meta;
2477
2478         /* profiles filter */
2479         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2480             chunk_profiles_filter(chunk_type, bargs)) {
2481                 return 0;
2482         }
2483
2484         /* usage filter */
2485         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2486             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2487                 return 0;
2488         }
2489
2490         /* devid filter */
2491         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2492             chunk_devid_filter(leaf, chunk, bargs)) {
2493                 return 0;
2494         }
2495
2496         /* drange filter, makes sense only with devid filter */
2497         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2498             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2499                 return 0;
2500         }
2501
2502         /* vrange filter */
2503         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2504             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2505                 return 0;
2506         }
2507
2508         /* soft profile changing mode */
2509         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2510             chunk_soft_convert_filter(chunk_type, bargs)) {
2511                 return 0;
2512         }
2513
2514         return 1;
2515 }
2516
2517 static u64 div_factor(u64 num, int factor)
2518 {
2519         if (factor == 10)
2520                 return num;
2521         num *= factor;
2522         do_div(num, 10);
2523         return num;
2524 }
2525
2526 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2527 {
2528         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2529         struct btrfs_root *chunk_root = fs_info->chunk_root;
2530         struct btrfs_root *dev_root = fs_info->dev_root;
2531         struct list_head *devices;
2532         struct btrfs_device *device;
2533         u64 old_size;
2534         u64 size_to_free;
2535         struct btrfs_chunk *chunk;
2536         struct btrfs_path *path;
2537         struct btrfs_key key;
2538         struct btrfs_key found_key;
2539         struct btrfs_trans_handle *trans;
2540         struct extent_buffer *leaf;
2541         int slot;
2542         int ret;
2543         int enospc_errors = 0;
2544         bool counting = true;
2545
2546         /* step one make some room on all the devices */
2547         devices = &fs_info->fs_devices->devices;
2548         list_for_each_entry(device, devices, dev_list) {
2549                 old_size = device->total_bytes;
2550                 size_to_free = div_factor(old_size, 1);
2551                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2552                 if (!device->writeable ||
2553                     device->total_bytes - device->bytes_used > size_to_free)
2554                         continue;
2555
2556                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2557                 if (ret == -ENOSPC)
2558                         break;
2559                 BUG_ON(ret);
2560
2561                 trans = btrfs_start_transaction(dev_root, 0);
2562                 BUG_ON(IS_ERR(trans));
2563
2564                 ret = btrfs_grow_device(trans, device, old_size);
2565                 BUG_ON(ret);
2566
2567                 btrfs_end_transaction(trans, dev_root);
2568         }
2569
2570         /* step two, relocate all the chunks */
2571         path = btrfs_alloc_path();
2572         if (!path) {
2573                 ret = -ENOMEM;
2574                 goto error;
2575         }
2576
2577         /* zero out stat counters */
2578         spin_lock(&fs_info->balance_lock);
2579         memset(&bctl->stat, 0, sizeof(bctl->stat));
2580         spin_unlock(&fs_info->balance_lock);
2581 again:
2582         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2583         key.offset = (u64)-1;
2584         key.type = BTRFS_CHUNK_ITEM_KEY;
2585
2586         while (1) {
2587                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2588                     atomic_read(&fs_info->balance_cancel_req)) {
2589                         ret = -ECANCELED;
2590                         goto error;
2591                 }
2592
2593                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2594                 if (ret < 0)
2595                         goto error;
2596
2597                 /*
2598                  * this shouldn't happen, it means the last relocate
2599                  * failed
2600                  */
2601                 if (ret == 0)
2602                         BUG(); /* FIXME break ? */
2603
2604                 ret = btrfs_previous_item(chunk_root, path, 0,
2605                                           BTRFS_CHUNK_ITEM_KEY);
2606                 if (ret) {
2607                         ret = 0;
2608                         break;
2609                 }
2610
2611                 leaf = path->nodes[0];
2612                 slot = path->slots[0];
2613                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2614
2615                 if (found_key.objectid != key.objectid)
2616                         break;
2617
2618                 /* chunk zero is special */
2619                 if (found_key.offset == 0)
2620                         break;
2621
2622                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2623
2624                 if (!counting) {
2625                         spin_lock(&fs_info->balance_lock);
2626                         bctl->stat.considered++;
2627                         spin_unlock(&fs_info->balance_lock);
2628                 }
2629
2630                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2631                                            found_key.offset);
2632                 btrfs_release_path(path);
2633                 if (!ret)
2634                         goto loop;
2635
2636                 if (counting) {
2637                         spin_lock(&fs_info->balance_lock);
2638                         bctl->stat.expected++;
2639                         spin_unlock(&fs_info->balance_lock);
2640                         goto loop;
2641                 }
2642
2643                 ret = btrfs_relocate_chunk(chunk_root,
2644                                            chunk_root->root_key.objectid,
2645                                            found_key.objectid,
2646                                            found_key.offset);
2647                 if (ret && ret != -ENOSPC)
2648                         goto error;
2649                 if (ret == -ENOSPC) {
2650                         enospc_errors++;
2651                 } else {
2652                         spin_lock(&fs_info->balance_lock);
2653                         bctl->stat.completed++;
2654                         spin_unlock(&fs_info->balance_lock);
2655                 }
2656 loop:
2657                 key.offset = found_key.offset - 1;
2658         }
2659
2660         if (counting) {
2661                 btrfs_release_path(path);
2662                 counting = false;
2663                 goto again;
2664         }
2665 error:
2666         btrfs_free_path(path);
2667         if (enospc_errors) {
2668                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2669                        enospc_errors);
2670                 if (!ret)
2671                         ret = -ENOSPC;
2672         }
2673
2674         return ret;
2675 }
2676
2677 /**
2678  * alloc_profile_is_valid - see if a given profile is valid and reduced
2679  * @flags: profile to validate
2680  * @extended: if true @flags is treated as an extended profile
2681  */
2682 static int alloc_profile_is_valid(u64 flags, int extended)
2683 {
2684         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2685                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2686
2687         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2688
2689         /* 1) check that all other bits are zeroed */
2690         if (flags & ~mask)
2691                 return 0;
2692
2693         /* 2) see if profile is reduced */
2694         if (flags == 0)
2695                 return !extended; /* "0" is valid for usual profiles */
2696
2697         /* true if exactly one bit set */
2698         return (flags & (flags - 1)) == 0;
2699 }
2700
2701 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2702 {
2703         /* cancel requested || normal exit path */
2704         return atomic_read(&fs_info->balance_cancel_req) ||
2705                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2706                  atomic_read(&fs_info->balance_cancel_req) == 0);
2707 }
2708
2709 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2710 {
2711         int ret;
2712
2713         unset_balance_control(fs_info);
2714         ret = del_balance_item(fs_info->tree_root);
2715         BUG_ON(ret);
2716 }
2717
2718 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2719                                struct btrfs_ioctl_balance_args *bargs);
2720
2721 /*
2722  * Should be called with both balance and volume mutexes held
2723  */
2724 int btrfs_balance(struct btrfs_balance_control *bctl,
2725                   struct btrfs_ioctl_balance_args *bargs)
2726 {
2727         struct btrfs_fs_info *fs_info = bctl->fs_info;
2728         u64 allowed;
2729         int mixed = 0;
2730         int ret;
2731
2732         if (btrfs_fs_closing(fs_info) ||
2733             atomic_read(&fs_info->balance_pause_req) ||
2734             atomic_read(&fs_info->balance_cancel_req)) {
2735                 ret = -EINVAL;
2736                 goto out;
2737         }
2738
2739         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2740         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2741                 mixed = 1;
2742
2743         /*
2744          * In case of mixed groups both data and meta should be picked,
2745          * and identical options should be given for both of them.
2746          */
2747         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2748         if (mixed && (bctl->flags & allowed)) {
2749                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2750                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2751                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2752                         printk(KERN_ERR "btrfs: with mixed groups data and "
2753                                "metadata balance options must be the same\n");
2754                         ret = -EINVAL;
2755                         goto out;
2756                 }
2757         }
2758
2759         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2760         if (fs_info->fs_devices->num_devices == 1)
2761                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2762         else if (fs_info->fs_devices->num_devices < 4)
2763                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2764         else
2765                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2766                                 BTRFS_BLOCK_GROUP_RAID10);
2767
2768         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2769             (!alloc_profile_is_valid(bctl->data.target, 1) ||
2770              (bctl->data.target & ~allowed))) {
2771                 printk(KERN_ERR "btrfs: unable to start balance with target "
2772                        "data profile %llu\n",
2773                        (unsigned long long)bctl->data.target);
2774                 ret = -EINVAL;
2775                 goto out;
2776         }
2777         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2778             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2779              (bctl->meta.target & ~allowed))) {
2780                 printk(KERN_ERR "btrfs: unable to start balance with target "
2781                        "metadata profile %llu\n",
2782                        (unsigned long long)bctl->meta.target);
2783                 ret = -EINVAL;
2784                 goto out;
2785         }
2786         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2787             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2788              (bctl->sys.target & ~allowed))) {
2789                 printk(KERN_ERR "btrfs: unable to start balance with target "
2790                        "system profile %llu\n",
2791                        (unsigned long long)bctl->sys.target);
2792                 ret = -EINVAL;
2793                 goto out;
2794         }
2795
2796         /* allow dup'ed data chunks only in mixed mode */
2797         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2798             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2799                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2800                 ret = -EINVAL;
2801                 goto out;
2802         }
2803
2804         /* allow to reduce meta or sys integrity only if force set */
2805         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2806                         BTRFS_BLOCK_GROUP_RAID10;
2807         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2808              (fs_info->avail_system_alloc_bits & allowed) &&
2809              !(bctl->sys.target & allowed)) ||
2810             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2811              (fs_info->avail_metadata_alloc_bits & allowed) &&
2812              !(bctl->meta.target & allowed))) {
2813                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2814                         printk(KERN_INFO "btrfs: force reducing metadata "
2815                                "integrity\n");
2816                 } else {
2817                         printk(KERN_ERR "btrfs: balance will reduce metadata "
2818                                "integrity, use force if you want this\n");
2819                         ret = -EINVAL;
2820                         goto out;
2821                 }
2822         }
2823
2824         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
2825                 int num_tolerated_disk_barrier_failures;
2826                 u64 target = bctl->sys.target;
2827
2828                 num_tolerated_disk_barrier_failures =
2829                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2830                 if (num_tolerated_disk_barrier_failures > 0 &&
2831                     (target &
2832                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
2833                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
2834                         num_tolerated_disk_barrier_failures = 0;
2835                 else if (num_tolerated_disk_barrier_failures > 1 &&
2836                          (target &
2837                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
2838                         num_tolerated_disk_barrier_failures = 1;
2839
2840                 fs_info->num_tolerated_disk_barrier_failures =
2841                         num_tolerated_disk_barrier_failures;
2842         }
2843
2844         ret = insert_balance_item(fs_info->tree_root, bctl);
2845         if (ret && ret != -EEXIST)
2846                 goto out;
2847
2848         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2849                 BUG_ON(ret == -EEXIST);
2850                 set_balance_control(bctl);
2851         } else {
2852                 BUG_ON(ret != -EEXIST);
2853                 spin_lock(&fs_info->balance_lock);
2854                 update_balance_args(bctl);
2855                 spin_unlock(&fs_info->balance_lock);
2856         }
2857
2858         atomic_inc(&fs_info->balance_running);
2859         mutex_unlock(&fs_info->balance_mutex);
2860
2861         ret = __btrfs_balance(fs_info);
2862
2863         mutex_lock(&fs_info->balance_mutex);
2864         atomic_dec(&fs_info->balance_running);
2865
2866         if (bargs) {
2867                 memset(bargs, 0, sizeof(*bargs));
2868                 update_ioctl_balance_args(fs_info, 0, bargs);
2869         }
2870
2871         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2872             balance_need_close(fs_info)) {
2873                 __cancel_balance(fs_info);
2874         }
2875
2876         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
2877                 fs_info->num_tolerated_disk_barrier_failures =
2878                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2879         }
2880
2881         wake_up(&fs_info->balance_wait_q);
2882
2883         return ret;
2884 out:
2885         if (bctl->flags & BTRFS_BALANCE_RESUME)
2886                 __cancel_balance(fs_info);
2887         else
2888                 kfree(bctl);
2889         return ret;
2890 }
2891
2892 static int balance_kthread(void *data)
2893 {
2894         struct btrfs_fs_info *fs_info = data;
2895         int ret = 0;
2896
2897         mutex_lock(&fs_info->volume_mutex);
2898         mutex_lock(&fs_info->balance_mutex);
2899
2900         if (fs_info->balance_ctl) {
2901                 printk(KERN_INFO "btrfs: continuing balance\n");
2902                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
2903         }
2904
2905         mutex_unlock(&fs_info->balance_mutex);
2906         mutex_unlock(&fs_info->volume_mutex);
2907
2908         return ret;
2909 }
2910
2911 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
2912 {
2913         struct task_struct *tsk;
2914
2915         spin_lock(&fs_info->balance_lock);
2916         if (!fs_info->balance_ctl) {
2917                 spin_unlock(&fs_info->balance_lock);
2918                 return 0;
2919         }
2920         spin_unlock(&fs_info->balance_lock);
2921
2922         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2923                 printk(KERN_INFO "btrfs: force skipping balance\n");
2924                 return 0;
2925         }
2926
2927         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
2928         if (IS_ERR(tsk))
2929                 return PTR_ERR(tsk);
2930
2931         return 0;
2932 }
2933
2934 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
2935 {
2936         struct btrfs_balance_control *bctl;
2937         struct btrfs_balance_item *item;
2938         struct btrfs_disk_balance_args disk_bargs;
2939         struct btrfs_path *path;
2940         struct extent_buffer *leaf;
2941         struct btrfs_key key;
2942         int ret;
2943
2944         path = btrfs_alloc_path();
2945         if (!path)
2946                 return -ENOMEM;
2947
2948         key.objectid = BTRFS_BALANCE_OBJECTID;
2949         key.type = BTRFS_BALANCE_ITEM_KEY;
2950         key.offset = 0;
2951
2952         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2953         if (ret < 0)
2954                 goto out;
2955         if (ret > 0) { /* ret = -ENOENT; */
2956                 ret = 0;
2957                 goto out;
2958         }
2959
2960         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2961         if (!bctl) {
2962                 ret = -ENOMEM;
2963                 goto out;
2964         }
2965
2966         leaf = path->nodes[0];
2967         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2968
2969         bctl->fs_info = fs_info;
2970         bctl->flags = btrfs_balance_flags(leaf, item);
2971         bctl->flags |= BTRFS_BALANCE_RESUME;
2972
2973         btrfs_balance_data(leaf, item, &disk_bargs);
2974         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2975         btrfs_balance_meta(leaf, item, &disk_bargs);
2976         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2977         btrfs_balance_sys(leaf, item, &disk_bargs);
2978         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2979
2980         mutex_lock(&fs_info->volume_mutex);
2981         mutex_lock(&fs_info->balance_mutex);
2982
2983         set_balance_control(bctl);
2984
2985         mutex_unlock(&fs_info->balance_mutex);
2986         mutex_unlock(&fs_info->volume_mutex);
2987 out:
2988         btrfs_free_path(path);
2989         return ret;
2990 }
2991
2992 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2993 {
2994         int ret = 0;
2995
2996         mutex_lock(&fs_info->balance_mutex);
2997         if (!fs_info->balance_ctl) {
2998                 mutex_unlock(&fs_info->balance_mutex);
2999                 return -ENOTCONN;
3000         }
3001
3002         if (atomic_read(&fs_info->balance_running)) {
3003                 atomic_inc(&fs_info->balance_pause_req);
3004                 mutex_unlock(&fs_info->balance_mutex);
3005
3006                 wait_event(fs_info->balance_wait_q,
3007                            atomic_read(&fs_info->balance_running) == 0);
3008
3009                 mutex_lock(&fs_info->balance_mutex);
3010                 /* we are good with balance_ctl ripped off from under us */
3011                 BUG_ON(atomic_read(&fs_info->balance_running));
3012                 atomic_dec(&fs_info->balance_pause_req);
3013         } else {
3014                 ret = -ENOTCONN;
3015         }
3016
3017         mutex_unlock(&fs_info->balance_mutex);
3018         return ret;
3019 }
3020
3021 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3022 {
3023         mutex_lock(&fs_info->balance_mutex);
3024         if (!fs_info->balance_ctl) {
3025                 mutex_unlock(&fs_info->balance_mutex);
3026                 return -ENOTCONN;
3027         }
3028
3029         atomic_inc(&fs_info->balance_cancel_req);
3030         /*
3031          * if we are running just wait and return, balance item is
3032          * deleted in btrfs_balance in this case
3033          */
3034         if (atomic_read(&fs_info->balance_running)) {
3035                 mutex_unlock(&fs_info->balance_mutex);
3036                 wait_event(fs_info->balance_wait_q,
3037                            atomic_read(&fs_info->balance_running) == 0);
3038                 mutex_lock(&fs_info->balance_mutex);
3039         } else {
3040                 /* __cancel_balance needs volume_mutex */
3041                 mutex_unlock(&fs_info->balance_mutex);
3042                 mutex_lock(&fs_info->volume_mutex);
3043                 mutex_lock(&fs_info->balance_mutex);
3044
3045                 if (fs_info->balance_ctl)
3046                         __cancel_balance(fs_info);
3047
3048                 mutex_unlock(&fs_info->volume_mutex);
3049         }
3050
3051         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3052         atomic_dec(&fs_info->balance_cancel_req);
3053         mutex_unlock(&fs_info->balance_mutex);
3054         return 0;
3055 }
3056
3057 /*
3058  * shrinking a device means finding all of the device extents past
3059  * the new size, and then following the back refs to the chunks.
3060  * The chunk relocation code actually frees the device extent
3061  */
3062 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3063 {
3064         struct btrfs_trans_handle *trans;
3065         struct btrfs_root *root = device->dev_root;
3066         struct btrfs_dev_extent *dev_extent = NULL;
3067         struct btrfs_path *path;
3068         u64 length;
3069         u64 chunk_tree;
3070         u64 chunk_objectid;
3071         u64 chunk_offset;
3072         int ret;
3073         int slot;
3074         int failed = 0;
3075         bool retried = false;
3076         struct extent_buffer *l;
3077         struct btrfs_key key;
3078         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3079         u64 old_total = btrfs_super_total_bytes(super_copy);
3080         u64 old_size = device->total_bytes;
3081         u64 diff = device->total_bytes - new_size;
3082
3083         if (new_size >= device->total_bytes)
3084                 return -EINVAL;
3085
3086         path = btrfs_alloc_path();
3087         if (!path)
3088                 return -ENOMEM;
3089
3090         path->reada = 2;
3091
3092         lock_chunks(root);
3093
3094         device->total_bytes = new_size;
3095         if (device->writeable) {
3096                 device->fs_devices->total_rw_bytes -= diff;
3097                 spin_lock(&root->fs_info->free_chunk_lock);
3098                 root->fs_info->free_chunk_space -= diff;
3099                 spin_unlock(&root->fs_info->free_chunk_lock);
3100         }
3101         unlock_chunks(root);
3102
3103 again:
3104         key.objectid = device->devid;
3105         key.offset = (u64)-1;
3106         key.type = BTRFS_DEV_EXTENT_KEY;
3107
3108         do {
3109                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3110                 if (ret < 0)
3111                         goto done;
3112
3113                 ret = btrfs_previous_item(root, path, 0, key.type);
3114                 if (ret < 0)
3115                         goto done;
3116                 if (ret) {
3117                         ret = 0;
3118                         btrfs_release_path(path);
3119                         break;
3120                 }
3121
3122                 l = path->nodes[0];
3123                 slot = path->slots[0];
3124                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3125
3126                 if (key.objectid != device->devid) {
3127                         btrfs_release_path(path);
3128                         break;
3129                 }
3130
3131                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3132                 length = btrfs_dev_extent_length(l, dev_extent);
3133
3134                 if (key.offset + length <= new_size) {
3135                         btrfs_release_path(path);
3136                         break;
3137                 }
3138
3139                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3140                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3141                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3142                 btrfs_release_path(path);
3143
3144                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3145                                            chunk_offset);
3146                 if (ret && ret != -ENOSPC)
3147                         goto done;
3148                 if (ret == -ENOSPC)
3149                         failed++;
3150         } while (key.offset-- > 0);
3151
3152         if (failed && !retried) {
3153                 failed = 0;
3154                 retried = true;
3155                 goto again;
3156         } else if (failed && retried) {
3157                 ret = -ENOSPC;
3158                 lock_chunks(root);
3159
3160                 device->total_bytes = old_size;
3161                 if (device->writeable)
3162                         device->fs_devices->total_rw_bytes += diff;
3163                 spin_lock(&root->fs_info->free_chunk_lock);
3164                 root->fs_info->free_chunk_space += diff;
3165                 spin_unlock(&root->fs_info->free_chunk_lock);
3166                 unlock_chunks(root);
3167                 goto done;
3168         }
3169
3170         /* Shrinking succeeded, else we would be at "done". */
3171         trans = btrfs_start_transaction(root, 0);
3172         if (IS_ERR(trans)) {
3173                 ret = PTR_ERR(trans);
3174                 goto done;
3175         }
3176
3177         lock_chunks(root);
3178
3179         device->disk_total_bytes = new_size;
3180         /* Now btrfs_update_device() will change the on-disk size. */
3181         ret = btrfs_update_device(trans, device);
3182         if (ret) {
3183                 unlock_chunks(root);
3184                 btrfs_end_transaction(trans, root);
3185                 goto done;
3186         }
3187         WARN_ON(diff > old_total);
3188         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3189         unlock_chunks(root);
3190         btrfs_end_transaction(trans, root);
3191 done:
3192         btrfs_free_path(path);
3193         return ret;
3194 }
3195
3196 static int btrfs_add_system_chunk(struct btrfs_root *root,
3197                            struct btrfs_key *key,
3198                            struct btrfs_chunk *chunk, int item_size)
3199 {
3200         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3201         struct btrfs_disk_key disk_key;
3202         u32 array_size;
3203         u8 *ptr;
3204
3205         array_size = btrfs_super_sys_array_size(super_copy);
3206         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3207                 return -EFBIG;
3208
3209         ptr = super_copy->sys_chunk_array + array_size;
3210         btrfs_cpu_key_to_disk(&disk_key, key);
3211         memcpy(ptr, &disk_key, sizeof(disk_key));
3212         ptr += sizeof(disk_key);
3213         memcpy(ptr, chunk, item_size);
3214         item_size += sizeof(disk_key);
3215         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3216         return 0;
3217 }
3218
3219 /*
3220  * sort the devices in descending order by max_avail, total_avail
3221  */
3222 static int btrfs_cmp_device_info(const void *a, const void *b)
3223 {
3224         const struct btrfs_device_info *di_a = a;
3225         const struct btrfs_device_info *di_b = b;
3226
3227         if (di_a->max_avail > di_b->max_avail)
3228                 return -1;
3229         if (di_a->max_avail < di_b->max_avail)
3230                 return 1;
3231         if (di_a->total_avail > di_b->total_avail)
3232                 return -1;
3233         if (di_a->total_avail < di_b->total_avail)
3234                 return 1;
3235         return 0;
3236 }
3237
3238 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3239                                struct btrfs_root *extent_root,
3240                                struct map_lookup **map_ret,
3241                                u64 *num_bytes_out, u64 *stripe_size_out,
3242                                u64 start, u64 type)
3243 {
3244         struct btrfs_fs_info *info = extent_root->fs_info;
3245         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3246         struct list_head *cur;
3247         struct map_lookup *map = NULL;
3248         struct extent_map_tree *em_tree;
3249         struct extent_map *em;
3250         struct btrfs_device_info *devices_info = NULL;
3251         u64 total_avail;
3252         int num_stripes;        /* total number of stripes to allocate */
3253         int sub_stripes;        /* sub_stripes info for map */
3254         int dev_stripes;        /* stripes per dev */
3255         int devs_max;           /* max devs to use */
3256         int devs_min;           /* min devs needed */
3257         int devs_increment;     /* ndevs has to be a multiple of this */
3258         int ncopies;            /* how many copies to data has */
3259         int ret;
3260         u64 max_stripe_size;
3261         u64 max_chunk_size;
3262         u64 stripe_size;
3263         u64 num_bytes;
3264         int ndevs;
3265         int i;
3266         int j;
3267
3268         BUG_ON(!alloc_profile_is_valid(type, 0));
3269
3270         if (list_empty(&fs_devices->alloc_list))
3271                 return -ENOSPC;
3272
3273         sub_stripes = 1;
3274         dev_stripes = 1;
3275         devs_increment = 1;
3276         ncopies = 1;
3277         devs_max = 0;   /* 0 == as many as possible */
3278         devs_min = 1;
3279
3280         /*
3281          * define the properties of each RAID type.
3282          * FIXME: move this to a global table and use it in all RAID
3283          * calculation code
3284          */
3285         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3286                 dev_stripes = 2;
3287                 ncopies = 2;
3288                 devs_max = 1;
3289         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3290                 devs_min = 2;
3291         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3292                 devs_increment = 2;
3293                 ncopies = 2;
3294                 devs_max = 2;
3295                 devs_min = 2;
3296         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3297                 sub_stripes = 2;
3298                 devs_increment = 2;
3299                 ncopies = 2;
3300                 devs_min = 4;
3301         } else {
3302                 devs_max = 1;
3303         }
3304
3305         if (type & BTRFS_BLOCK_GROUP_DATA) {
3306                 max_stripe_size = 1024 * 1024 * 1024;
3307                 max_chunk_size = 10 * max_stripe_size;
3308         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3309                 /* for larger filesystems, use larger metadata chunks */
3310                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3311                         max_stripe_size = 1024 * 1024 * 1024;
3312                 else
3313                         max_stripe_size = 256 * 1024 * 1024;
3314                 max_chunk_size = max_stripe_size;
3315         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3316                 max_stripe_size = 32 * 1024 * 1024;
3317                 max_chunk_size = 2 * max_stripe_size;
3318         } else {
3319                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3320                        type);
3321                 BUG_ON(1);
3322         }
3323
3324         /* we don't want a chunk larger than 10% of writeable space */
3325         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3326                              max_chunk_size);
3327
3328         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3329                                GFP_NOFS);
3330         if (!devices_info)
3331                 return -ENOMEM;
3332
3333         cur = fs_devices->alloc_list.next;
3334
3335         /*
3336          * in the first pass through the devices list, we gather information
3337          * about the available holes on each device.
3338          */
3339         ndevs = 0;
3340         while (cur != &fs_devices->alloc_list) {
3341                 struct btrfs_device *device;
3342                 u64 max_avail;
3343                 u64 dev_offset;
3344
3345                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3346
3347                 cur = cur->next;
3348
3349                 if (!device->writeable) {
3350                         printk(KERN_ERR
3351                                "btrfs: read-only device in alloc_list\n");
3352                         WARN_ON(1);
3353                         continue;
3354                 }
3355
3356                 if (!device->in_fs_metadata)
3357                         continue;
3358
3359                 if (device->total_bytes > device->bytes_used)
3360                         total_avail = device->total_bytes - device->bytes_used;
3361                 else
3362                         total_avail = 0;
3363
3364                 /* If there is no space on this device, skip it. */
3365                 if (total_avail == 0)
3366                         continue;
3367
3368                 ret = find_free_dev_extent(device,
3369                                            max_stripe_size * dev_stripes,
3370                                            &dev_offset, &max_avail);
3371                 if (ret && ret != -ENOSPC)
3372                         goto error;
3373
3374                 if (ret == 0)
3375                         max_avail = max_stripe_size * dev_stripes;
3376
3377                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3378                         continue;
3379
3380                 devices_info[ndevs].dev_offset = dev_offset;
3381                 devices_info[ndevs].max_avail = max_avail;
3382                 devices_info[ndevs].total_avail = total_avail;
3383                 devices_info[ndevs].dev = device;
3384                 ++ndevs;
3385         }
3386
3387         /*
3388          * now sort the devices by hole size / available space
3389          */
3390         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3391              btrfs_cmp_device_info, NULL);
3392
3393         /* round down to number of usable stripes */
3394         ndevs -= ndevs % devs_increment;
3395
3396         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3397                 ret = -ENOSPC;
3398                 goto error;
3399         }
3400
3401         if (devs_max && ndevs > devs_max)
3402                 ndevs = devs_max;
3403         /*
3404          * the primary goal is to maximize the number of stripes, so use as many
3405          * devices as possible, even if the stripes are not maximum sized.
3406          */
3407         stripe_size = devices_info[ndevs-1].max_avail;
3408         num_stripes = ndevs * dev_stripes;
3409
3410         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3411                 stripe_size = max_chunk_size * ncopies;
3412                 do_div(stripe_size, ndevs);
3413         }
3414
3415         do_div(stripe_size, dev_stripes);
3416
3417         /* align to BTRFS_STRIPE_LEN */
3418         do_div(stripe_size, BTRFS_STRIPE_LEN);
3419         stripe_size *= BTRFS_STRIPE_LEN;
3420
3421         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3422         if (!map) {
3423                 ret = -ENOMEM;
3424                 goto error;
3425         }
3426         map->num_stripes = num_stripes;
3427
3428         for (i = 0; i < ndevs; ++i) {
3429                 for (j = 0; j < dev_stripes; ++j) {
3430                         int s = i * dev_stripes + j;
3431                         map->stripes[s].dev = devices_info[i].dev;
3432                         map->stripes[s].physical = devices_info[i].dev_offset +
3433                                                    j * stripe_size;
3434                 }
3435         }
3436         map->sector_size = extent_root->sectorsize;
3437         map->stripe_len = BTRFS_STRIPE_LEN;
3438         map->io_align = BTRFS_STRIPE_LEN;
3439         map->io_width = BTRFS_STRIPE_LEN;
3440         map->type = type;
3441         map->sub_stripes = sub_stripes;
3442
3443         *map_ret = map;
3444         num_bytes = stripe_size * (num_stripes / ncopies);
3445
3446         *stripe_size_out = stripe_size;
3447         *num_bytes_out = num_bytes;
3448
3449         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3450
3451         em = alloc_extent_map();
3452         if (!em) {
3453                 ret = -ENOMEM;
3454                 goto error;
3455         }
3456         em->bdev = (struct block_device *)map;
3457         em->start = start;
3458         em->len = num_bytes;
3459         em->block_start = 0;
3460         em->block_len = em->len;
3461
3462         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3463         write_lock(&em_tree->lock);
3464         ret = add_extent_mapping(em_tree, em);
3465         write_unlock(&em_tree->lock);
3466         free_extent_map(em);
3467         if (ret)
3468                 goto error;
3469
3470         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3471                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3472                                      start, num_bytes);
3473         if (ret)
3474                 goto error;
3475
3476         for (i = 0; i < map->num_stripes; ++i) {
3477                 struct btrfs_device *device;
3478                 u64 dev_offset;
3479
3480                 device = map->stripes[i].dev;
3481                 dev_offset = map->stripes[i].physical;
3482
3483                 ret = btrfs_alloc_dev_extent(trans, device,
3484                                 info->chunk_root->root_key.objectid,
3485                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3486                                 start, dev_offset, stripe_size);
3487                 if (ret) {
3488                         btrfs_abort_transaction(trans, extent_root, ret);
3489                         goto error;
3490                 }
3491         }
3492
3493         kfree(devices_info);
3494         return 0;
3495
3496 error:
3497         kfree(map);
3498         kfree(devices_info);
3499         return ret;
3500 }
3501
3502 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3503                                 struct btrfs_root *extent_root,
3504                                 struct map_lookup *map, u64 chunk_offset,
3505                                 u64 chunk_size, u64 stripe_size)
3506 {
3507         u64 dev_offset;
3508         struct btrfs_key key;
3509         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3510         struct btrfs_device *device;
3511         struct btrfs_chunk *chunk;
3512         struct btrfs_stripe *stripe;
3513         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3514         int index = 0;
3515         int ret;
3516
3517         chunk = kzalloc(item_size, GFP_NOFS);
3518         if (!chunk)
3519                 return -ENOMEM;
3520
3521         index = 0;
3522         while (index < map->num_stripes) {
3523                 device = map->stripes[index].dev;
3524                 device->bytes_used += stripe_size;
3525                 ret = btrfs_update_device(trans, device);
3526                 if (ret)
3527                         goto out_free;
3528                 index++;
3529         }
3530
3531         spin_lock(&extent_root->fs_info->free_chunk_lock);
3532         extent_root->fs_info->free_chunk_space -= (stripe_size *
3533                                                    map->num_stripes);
3534         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3535
3536         index = 0;
3537         stripe = &chunk->stripe;
3538         while (index < map->num_stripes) {
3539                 device = map->stripes[index].dev;
3540                 dev_offset = map->stripes[index].physical;
3541
3542                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3543                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3544                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3545                 stripe++;
3546                 index++;
3547         }
3548
3549         btrfs_set_stack_chunk_length(chunk, chunk_size);
3550         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3551         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3552         btrfs_set_stack_chunk_type(chunk, map->type);
3553         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3554         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3555         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3556         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3557         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3558
3559         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3560         key.type = BTRFS_CHUNK_ITEM_KEY;
3561         key.offset = chunk_offset;
3562
3563         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3564
3565         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3566                 /*
3567                  * TODO: Cleanup of inserted chunk root in case of
3568                  * failure.
3569                  */
3570                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3571                                              item_size);
3572         }
3573
3574 out_free:
3575         kfree(chunk);
3576         return ret;
3577 }
3578
3579 /*
3580  * Chunk allocation falls into two parts. The first part does works
3581  * that make the new allocated chunk useable, but not do any operation
3582  * that modifies the chunk tree. The second part does the works that
3583  * require modifying the chunk tree. This division is important for the
3584  * bootstrap process of adding storage to a seed btrfs.
3585  */
3586 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3587                       struct btrfs_root *extent_root, u64 type)
3588 {
3589         u64 chunk_offset;
3590         u64 chunk_size;
3591         u64 stripe_size;
3592         struct map_lookup *map;
3593         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3594         int ret;
3595
3596         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3597                               &chunk_offset);
3598         if (ret)
3599                 return ret;
3600
3601         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3602                                   &stripe_size, chunk_offset, type);
3603         if (ret)
3604                 return ret;
3605
3606         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3607                                    chunk_size, stripe_size);
3608         if (ret)
3609                 return ret;
3610         return 0;
3611 }
3612
3613 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3614                                          struct btrfs_root *root,
3615                                          struct btrfs_device *device)
3616 {
3617         u64 chunk_offset;
3618         u64 sys_chunk_offset;
3619         u64 chunk_size;
3620         u64 sys_chunk_size;
3621         u64 stripe_size;
3622         u64 sys_stripe_size;
3623         u64 alloc_profile;
3624         struct map_lookup *map;
3625         struct map_lookup *sys_map;
3626         struct btrfs_fs_info *fs_info = root->fs_info;
3627         struct btrfs_root *extent_root = fs_info->extent_root;
3628         int ret;
3629
3630         ret = find_next_chunk(fs_info->chunk_root,
3631                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3632         if (ret)
3633                 return ret;
3634
3635         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3636                                 fs_info->avail_metadata_alloc_bits;
3637         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3638
3639         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3640                                   &stripe_size, chunk_offset, alloc_profile);
3641         if (ret)
3642                 return ret;
3643
3644         sys_chunk_offset = chunk_offset + chunk_size;
3645
3646         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3647                                 fs_info->avail_system_alloc_bits;
3648         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3649
3650         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3651                                   &sys_chunk_size, &sys_stripe_size,
3652                                   sys_chunk_offset, alloc_profile);
3653         if (ret) {
3654                 btrfs_abort_transaction(trans, root, ret);
3655                 goto out;
3656         }
3657
3658         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3659         if (ret) {
3660                 btrfs_abort_transaction(trans, root, ret);
3661                 goto out;
3662         }
3663
3664         /*
3665          * Modifying chunk tree needs allocating new blocks from both
3666          * system block group and metadata block group. So we only can
3667          * do operations require modifying the chunk tree after both
3668          * block groups were created.
3669          */
3670         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3671                                    chunk_size, stripe_size);
3672         if (ret) {
3673                 btrfs_abort_transaction(trans, root, ret);
3674                 goto out;
3675         }
3676
3677         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3678                                    sys_chunk_offset, sys_chunk_size,
3679                                    sys_stripe_size);
3680         if (ret)
3681                 btrfs_abort_transaction(trans, root, ret);
3682
3683 out:
3684
3685         return ret;
3686 }
3687
3688 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3689 {
3690         struct extent_map *em;
3691         struct map_lookup *map;
3692         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3693         int readonly = 0;
3694         int i;
3695
3696         read_lock(&map_tree->map_tree.lock);
3697         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3698         read_unlock(&map_tree->map_tree.lock);
3699         if (!em)
3700                 return 1;
3701
3702         if (btrfs_test_opt(root, DEGRADED)) {
3703                 free_extent_map(em);
3704                 return 0;
3705         }
3706
3707         map = (struct map_lookup *)em->bdev;
3708         for (i = 0; i < map->num_stripes; i++) {
3709                 if (!map->stripes[i].dev->writeable) {
3710                         readonly = 1;
3711                         break;
3712                 }
3713         }
3714         free_extent_map(em);
3715         return readonly;
3716 }
3717
3718 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3719 {
3720         extent_map_tree_init(&tree->map_tree);
3721 }
3722
3723 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3724 {
3725         struct extent_map *em;
3726
3727         while (1) {
3728                 write_lock(&tree->map_tree.lock);
3729                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3730                 if (em)
3731                         remove_extent_mapping(&tree->map_tree, em);
3732                 write_unlock(&tree->map_tree.lock);
3733                 if (!em)
3734                         break;
3735                 kfree(em->bdev);
3736                 /* once for us */
3737                 free_extent_map(em);
3738                 /* once for the tree */
3739                 free_extent_map(em);
3740         }
3741 }
3742
3743 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3744 {
3745         struct extent_map *em;
3746         struct map_lookup *map;
3747         struct extent_map_tree *em_tree = &map_tree->map_tree;
3748         int ret;
3749
3750         read_lock(&em_tree->lock);
3751         em = lookup_extent_mapping(em_tree, logical, len);
3752         read_unlock(&em_tree->lock);
3753         BUG_ON(!em);
3754
3755         BUG_ON(em->start > logical || em->start + em->len < logical);
3756         map = (struct map_lookup *)em->bdev;
3757         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3758                 ret = map->num_stripes;
3759         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3760                 ret = map->sub_stripes;
3761         else
3762                 ret = 1;
3763         free_extent_map(em);
3764         return ret;
3765 }
3766
3767 static int find_live_mirror(struct map_lookup *map, int first, int num,
3768                             int optimal)
3769 {
3770         int i;
3771         if (map->stripes[optimal].dev->bdev)
3772                 return optimal;
3773         for (i = first; i < first + num; i++) {
3774                 if (map->stripes[i].dev->bdev)
3775                         return i;
3776         }
3777         /* we couldn't find one that doesn't fail.  Just return something
3778          * and the io error handling code will clean up eventually
3779          */
3780         return optimal;
3781 }
3782
3783 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3784                              u64 logical, u64 *length,
3785                              struct btrfs_bio **bbio_ret,
3786                              int mirror_num)
3787 {
3788         struct extent_map *em;
3789         struct map_lookup *map;
3790         struct extent_map_tree *em_tree = &map_tree->map_tree;
3791         u64 offset;
3792         u64 stripe_offset;
3793         u64 stripe_end_offset;
3794         u64 stripe_nr;
3795         u64 stripe_nr_orig;
3796         u64 stripe_nr_end;
3797         int stripe_index;
3798         int i;
3799         int ret = 0;
3800         int num_stripes;
3801         int max_errors = 0;
3802         struct btrfs_bio *bbio = NULL;
3803
3804         read_lock(&em_tree->lock);
3805         em = lookup_extent_mapping(em_tree, logical, *length);
3806         read_unlock(&em_tree->lock);
3807
3808         if (!em) {
3809                 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
3810                        (unsigned long long)logical,
3811                        (unsigned long long)*length);
3812                 BUG();
3813         }
3814
3815         BUG_ON(em->start > logical || em->start + em->len < logical);
3816         map = (struct map_lookup *)em->bdev;
3817         offset = logical - em->start;
3818
3819         if (mirror_num > map->num_stripes)
3820                 mirror_num = 0;
3821
3822         stripe_nr = offset;
3823         /*
3824          * stripe_nr counts the total number of stripes we have to stride
3825          * to get to this block
3826          */
3827         do_div(stripe_nr, map->stripe_len);
3828
3829         stripe_offset = stripe_nr * map->stripe_len;
3830         BUG_ON(offset < stripe_offset);
3831
3832         /* stripe_offset is the offset of this block in its stripe*/
3833         stripe_offset = offset - stripe_offset;
3834
3835         if (rw & REQ_DISCARD)
3836                 *length = min_t(u64, em->len - offset, *length);
3837         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3838                 /* we limit the length of each bio to what fits in a stripe */
3839                 *length = min_t(u64, em->len - offset,
3840                                 map->stripe_len - stripe_offset);
3841         } else {
3842                 *length = em->len - offset;
3843         }
3844
3845         if (!bbio_ret)
3846                 goto out;
3847
3848         num_stripes = 1;
3849         stripe_index = 0;
3850         stripe_nr_orig = stripe_nr;
3851         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3852                         (~(map->stripe_len - 1));
3853         do_div(stripe_nr_end, map->stripe_len);
3854         stripe_end_offset = stripe_nr_end * map->stripe_len -
3855                             (offset + *length);
3856         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3857                 if (rw & REQ_DISCARD)
3858                         num_stripes = min_t(u64, map->num_stripes,
3859                                             stripe_nr_end - stripe_nr_orig);
3860                 stripe_index = do_div(stripe_nr, map->num_stripes);
3861         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3862                 if (rw & (REQ_WRITE | REQ_DISCARD))
3863                         num_stripes = map->num_stripes;
3864                 else if (mirror_num)
3865                         stripe_index = mirror_num - 1;
3866                 else {
3867                         stripe_index = find_live_mirror(map, 0,
3868                                             map->num_stripes,
3869                                             current->pid % map->num_stripes);
3870                         mirror_num = stripe_index + 1;
3871                 }
3872
3873         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3874                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3875                         num_stripes = map->num_stripes;
3876                 } else if (mirror_num) {
3877                         stripe_index = mirror_num - 1;
3878                 } else {
3879                         mirror_num = 1;
3880                 }
3881
3882         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3883                 int factor = map->num_stripes / map->sub_stripes;
3884
3885                 stripe_index = do_div(stripe_nr, factor);
3886                 stripe_index *= map->sub_stripes;
3887
3888                 if (rw & REQ_WRITE)
3889                         num_stripes = map->sub_stripes;
3890                 else if (rw & REQ_DISCARD)
3891                         num_stripes = min_t(u64, map->sub_stripes *
3892                                             (stripe_nr_end - stripe_nr_orig),
3893                                             map->num_stripes);
3894                 else if (mirror_num)
3895                         stripe_index += mirror_num - 1;
3896                 else {
3897                         int old_stripe_index = stripe_index;
3898                         stripe_index = find_live_mirror(map, stripe_index,
3899                                               map->sub_stripes, stripe_index +
3900                                               current->pid % map->sub_stripes);
3901                         mirror_num = stripe_index - old_stripe_index + 1;
3902                 }
3903         } else {
3904                 /*
3905                  * after this do_div call, stripe_nr is the number of stripes
3906                  * on this device we have to walk to find the data, and
3907                  * stripe_index is the number of our device in the stripe array
3908                  */
3909                 stripe_index = do_div(stripe_nr, map->num_stripes);
3910                 mirror_num = stripe_index + 1;
3911         }
3912         BUG_ON(stripe_index >= map->num_stripes);
3913
3914         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3915         if (!bbio) {
3916                 ret = -ENOMEM;
3917                 goto out;
3918         }
3919         atomic_set(&bbio->error, 0);
3920
3921         if (rw & REQ_DISCARD) {
3922                 int factor = 0;
3923                 int sub_stripes = 0;
3924                 u64 stripes_per_dev = 0;
3925                 u32 remaining_stripes = 0;
3926                 u32 last_stripe = 0;
3927
3928                 if (map->type &
3929                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3930                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3931                                 sub_stripes = 1;
3932                         else
3933                                 sub_stripes = map->sub_stripes;
3934
3935                         factor = map->num_stripes / sub_stripes;
3936                         stripes_per_dev = div_u64_rem(stripe_nr_end -
3937                                                       stripe_nr_orig,
3938                                                       factor,
3939                                                       &remaining_stripes);
3940                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
3941                         last_stripe *= sub_stripes;
3942                 }
3943
3944                 for (i = 0; i < num_stripes; i++) {
3945                         bbio->stripes[i].physical =
3946                                 map->stripes[stripe_index].physical +
3947                                 stripe_offset + stripe_nr * map->stripe_len;
3948                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3949
3950                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3951                                          BTRFS_BLOCK_GROUP_RAID10)) {
3952                                 bbio->stripes[i].length = stripes_per_dev *
3953                                                           map->stripe_len;
3954
3955                                 if (i / sub_stripes < remaining_stripes)
3956                                         bbio->stripes[i].length +=
3957                                                 map->stripe_len;
3958
3959                                 /*
3960                                  * Special for the first stripe and
3961                                  * the last stripe:
3962                                  *
3963                                  * |-------|...|-------|
3964                                  *     |----------|
3965                                  *    off     end_off
3966                                  */
3967                                 if (i < sub_stripes)
3968                                         bbio->stripes[i].length -=
3969                                                 stripe_offset;
3970
3971                                 if (stripe_index >= last_stripe &&
3972                                     stripe_index <= (last_stripe +
3973                                                      sub_stripes - 1))
3974                                         bbio->stripes[i].length -=
3975                                                 stripe_end_offset;
3976
3977                                 if (i == sub_stripes - 1)
3978                                         stripe_offset = 0;
3979                         } else
3980                                 bbio->stripes[i].length = *length;
3981
3982                         stripe_index++;
3983                         if (stripe_index == map->num_stripes) {
3984                                 /* This could only happen for RAID0/10 */
3985                                 stripe_index = 0;
3986                                 stripe_nr++;
3987                         }
3988                 }
3989         } else {
3990                 for (i = 0; i < num_stripes; i++) {
3991                         bbio->stripes[i].physical =
3992                                 map->stripes[stripe_index].physical +
3993                                 stripe_offset +
3994                                 stripe_nr * map->stripe_len;
3995                         bbio->stripes[i].dev =
3996                                 map->stripes[stripe_index].dev;
3997                         stripe_index++;
3998                 }
3999         }
4000
4001         if (rw & REQ_WRITE) {
4002                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4003                                  BTRFS_BLOCK_GROUP_RAID10 |
4004                                  BTRFS_BLOCK_GROUP_DUP)) {
4005                         max_errors = 1;
4006                 }
4007         }
4008
4009         *bbio_ret = bbio;
4010         bbio->num_stripes = num_stripes;
4011         bbio->max_errors = max_errors;
4012         bbio->mirror_num = mirror_num;
4013 out:
4014         free_extent_map(em);
4015         return ret;
4016 }
4017
4018 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
4019                       u64 logical, u64 *length,
4020                       struct btrfs_bio **bbio_ret, int mirror_num)
4021 {
4022         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
4023                                  mirror_num);
4024 }
4025
4026 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4027                      u64 chunk_start, u64 physical, u64 devid,
4028                      u64 **logical, int *naddrs, int *stripe_len)
4029 {
4030         struct extent_map_tree *em_tree = &map_tree->map_tree;
4031         struct extent_map *em;
4032         struct map_lookup *map;
4033         u64 *buf;
4034         u64 bytenr;
4035         u64 length;
4036         u64 stripe_nr;
4037         int i, j, nr = 0;
4038
4039         read_lock(&em_tree->lock);
4040         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4041         read_unlock(&em_tree->lock);
4042
4043         BUG_ON(!em || em->start != chunk_start);
4044         map = (struct map_lookup *)em->bdev;
4045
4046         length = em->len;
4047         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4048                 do_div(length, map->num_stripes / map->sub_stripes);
4049         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4050                 do_div(length, map->num_stripes);
4051
4052         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4053         BUG_ON(!buf); /* -ENOMEM */
4054
4055         for (i = 0; i < map->num_stripes; i++) {
4056                 if (devid && map->stripes[i].dev->devid != devid)
4057                         continue;
4058                 if (map->stripes[i].physical > physical ||
4059                     map->stripes[i].physical + length <= physical)
4060                         continue;
4061
4062                 stripe_nr = physical - map->stripes[i].physical;
4063                 do_div(stripe_nr, map->stripe_len);
4064
4065                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4066                         stripe_nr = stripe_nr * map->num_stripes + i;
4067                         do_div(stripe_nr, map->sub_stripes);
4068                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4069                         stripe_nr = stripe_nr * map->num_stripes + i;
4070                 }
4071                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4072                 WARN_ON(nr >= map->num_stripes);
4073                 for (j = 0; j < nr; j++) {
4074                         if (buf[j] == bytenr)
4075                                 break;
4076                 }
4077                 if (j == nr) {
4078                         WARN_ON(nr >= map->num_stripes);
4079                         buf[nr++] = bytenr;
4080                 }
4081         }
4082
4083         *logical = buf;
4084         *naddrs = nr;
4085         *stripe_len = map->stripe_len;
4086
4087         free_extent_map(em);
4088         return 0;
4089 }
4090
4091 static void *merge_stripe_index_into_bio_private(void *bi_private,
4092                                                  unsigned int stripe_index)
4093 {
4094         /*
4095          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4096          * at most 1.
4097          * The alternative solution (instead of stealing bits from the
4098          * pointer) would be to allocate an intermediate structure
4099          * that contains the old private pointer plus the stripe_index.
4100          */
4101         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4102         BUG_ON(stripe_index > 3);
4103         return (void *)(((uintptr_t)bi_private) | stripe_index);
4104 }
4105
4106 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4107 {
4108         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4109 }
4110
4111 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4112 {
4113         return (unsigned int)((uintptr_t)bi_private) & 3;
4114 }
4115
4116 static void btrfs_end_bio(struct bio *bio, int err)
4117 {
4118         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4119         int is_orig_bio = 0;
4120
4121         if (err) {
4122                 atomic_inc(&bbio->error);
4123                 if (err == -EIO || err == -EREMOTEIO) {
4124                         unsigned int stripe_index =
4125                                 extract_stripe_index_from_bio_private(
4126                                         bio->bi_private);
4127                         struct btrfs_device *dev;
4128
4129                         BUG_ON(stripe_index >= bbio->num_stripes);
4130                         dev = bbio->stripes[stripe_index].dev;
4131                         if (dev->bdev) {
4132                                 if (bio->bi_rw & WRITE)
4133                                         btrfs_dev_stat_inc(dev,
4134                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4135                                 else
4136                                         btrfs_dev_stat_inc(dev,
4137                                                 BTRFS_DEV_STAT_READ_ERRS);
4138                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4139                                         btrfs_dev_stat_inc(dev,
4140                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4141                                 btrfs_dev_stat_print_on_error(dev);
4142                         }
4143                 }
4144         }
4145
4146         if (bio == bbio->orig_bio)
4147                 is_orig_bio = 1;
4148
4149         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4150                 if (!is_orig_bio) {
4151                         bio_put(bio);
4152                         bio = bbio->orig_bio;
4153                 }
4154                 bio->bi_private = bbio->private;
4155                 bio->bi_end_io = bbio->end_io;
4156                 bio->bi_bdev = (struct block_device *)
4157                                         (unsigned long)bbio->mirror_num;
4158                 /* only send an error to the higher layers if it is
4159                  * beyond the tolerance of the multi-bio
4160                  */
4161                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4162                         err = -EIO;
4163                 } else {
4164                         /*
4165                          * this bio is actually up to date, we didn't
4166                          * go over the max number of errors
4167                          */
4168                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4169                         err = 0;
4170                 }
4171                 kfree(bbio);
4172
4173                 bio_endio(bio, err);
4174         } else if (!is_orig_bio) {
4175                 bio_put(bio);
4176         }
4177 }
4178
4179 struct async_sched {
4180         struct bio *bio;
4181         int rw;
4182         struct btrfs_fs_info *info;
4183         struct btrfs_work work;
4184 };
4185
4186 /*
4187  * see run_scheduled_bios for a description of why bios are collected for
4188  * async submit.
4189  *
4190  * This will add one bio to the pending list for a device and make sure
4191  * the work struct is scheduled.
4192  */
4193 static noinline void schedule_bio(struct btrfs_root *root,
4194                                  struct btrfs_device *device,
4195                                  int rw, struct bio *bio)
4196 {
4197         int should_queue = 1;
4198         struct btrfs_pending_bios *pending_bios;
4199
4200         /* don't bother with additional async steps for reads, right now */
4201         if (!(rw & REQ_WRITE)) {
4202                 bio_get(bio);
4203                 btrfsic_submit_bio(rw, bio);
4204                 bio_put(bio);
4205                 return;
4206         }
4207
4208         /*
4209          * nr_async_bios allows us to reliably return congestion to the
4210          * higher layers.  Otherwise, the async bio makes it appear we have
4211          * made progress against dirty pages when we've really just put it
4212          * on a queue for later
4213          */
4214         atomic_inc(&root->fs_info->nr_async_bios);
4215         WARN_ON(bio->bi_next);
4216         bio->bi_next = NULL;
4217         bio->bi_rw |= rw;
4218
4219         spin_lock(&device->io_lock);
4220         if (bio->bi_rw & REQ_SYNC)
4221                 pending_bios = &device->pending_sync_bios;
4222         else
4223                 pending_bios = &device->pending_bios;
4224
4225         if (pending_bios->tail)
4226                 pending_bios->tail->bi_next = bio;
4227
4228         pending_bios->tail = bio;
4229         if (!pending_bios->head)
4230                 pending_bios->head = bio;
4231         if (device->running_pending)
4232                 should_queue = 0;
4233
4234         spin_unlock(&device->io_lock);
4235
4236         if (should_queue)
4237                 btrfs_queue_worker(&root->fs_info->submit_workers,
4238                                    &device->work);
4239 }
4240
4241 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4242                   int mirror_num, int async_submit)
4243 {
4244         struct btrfs_mapping_tree *map_tree;
4245         struct btrfs_device *dev;
4246         struct bio *first_bio = bio;
4247         u64 logical = (u64)bio->bi_sector << 9;
4248         u64 length = 0;
4249         u64 map_length;
4250         int ret;
4251         int dev_nr = 0;
4252         int total_devs = 1;
4253         struct btrfs_bio *bbio = NULL;
4254
4255         length = bio->bi_size;
4256         map_tree = &root->fs_info->mapping_tree;
4257         map_length = length;
4258
4259         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4260                               mirror_num);
4261         if (ret) /* -ENOMEM */
4262                 return ret;
4263
4264         total_devs = bbio->num_stripes;
4265         if (map_length < length) {
4266                 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4267                        "len %llu\n", (unsigned long long)logical,
4268                        (unsigned long long)length,
4269                        (unsigned long long)map_length);
4270                 BUG();
4271         }
4272
4273         bbio->orig_bio = first_bio;
4274         bbio->private = first_bio->bi_private;
4275         bbio->end_io = first_bio->bi_end_io;
4276         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4277
4278         while (dev_nr < total_devs) {
4279                 if (dev_nr < total_devs - 1) {
4280                         bio = bio_clone(first_bio, GFP_NOFS);
4281                         BUG_ON(!bio); /* -ENOMEM */
4282                 } else {
4283                         bio = first_bio;
4284                 }
4285                 bio->bi_private = bbio;
4286                 bio->bi_private = merge_stripe_index_into_bio_private(
4287                                 bio->bi_private, (unsigned int)dev_nr);
4288                 bio->bi_end_io = btrfs_end_bio;
4289                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4290                 dev = bbio->stripes[dev_nr].dev;
4291                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4292 #ifdef DEBUG
4293                         struct rcu_string *name;
4294
4295                         rcu_read_lock();
4296                         name = rcu_dereference(dev->name);
4297                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4298                                  "(%s id %llu), size=%u\n", rw,
4299                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4300                                  name->str, dev->devid, bio->bi_size);
4301                         rcu_read_unlock();
4302 #endif
4303                         bio->bi_bdev = dev->bdev;
4304                         if (async_submit)
4305                                 schedule_bio(root, dev, rw, bio);
4306                         else
4307                                 btrfsic_submit_bio(rw, bio);
4308                 } else {
4309                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4310                         bio->bi_sector = logical >> 9;
4311                         bio_endio(bio, -EIO);
4312                 }
4313                 dev_nr++;
4314         }
4315         return 0;
4316 }
4317
4318 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4319                                        u8 *uuid, u8 *fsid)
4320 {
4321         struct btrfs_device *device;
4322         struct btrfs_fs_devices *cur_devices;
4323
4324         cur_devices = root->fs_info->fs_devices;
4325         while (cur_devices) {
4326                 if (!fsid ||
4327                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4328                         device = __find_device(&cur_devices->devices,
4329                                                devid, uuid);
4330                         if (device)
4331                                 return device;
4332                 }
4333                 cur_devices = cur_devices->seed;
4334         }
4335         return NULL;
4336 }
4337
4338 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4339                                             u64 devid, u8 *dev_uuid)
4340 {
4341         struct btrfs_device *device;
4342         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4343
4344         device = kzalloc(sizeof(*device), GFP_NOFS);
4345         if (!device)
4346                 return NULL;
4347         list_add(&device->dev_list,
4348                  &fs_devices->devices);
4349         device->dev_root = root->fs_info->dev_root;
4350         device->devid = devid;
4351         device->work.func = pending_bios_fn;
4352         device->fs_devices = fs_devices;
4353         device->missing = 1;
4354         fs_devices->num_devices++;
4355         fs_devices->missing_devices++;
4356         spin_lock_init(&device->io_lock);
4357         INIT_LIST_HEAD(&device->dev_alloc_list);
4358         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4359         return device;
4360 }
4361
4362 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4363                           struct extent_buffer *leaf,
4364                           struct btrfs_chunk *chunk)
4365 {
4366         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4367         struct map_lookup *map;
4368         struct extent_map *em;
4369         u64 logical;
4370         u64 length;
4371         u64 devid;
4372         u8 uuid[BTRFS_UUID_SIZE];
4373         int num_stripes;
4374         int ret;
4375         int i;
4376
4377         logical = key->offset;
4378         length = btrfs_chunk_length(leaf, chunk);
4379
4380         read_lock(&map_tree->map_tree.lock);
4381         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4382         read_unlock(&map_tree->map_tree.lock);
4383
4384         /* already mapped? */
4385         if (em && em->start <= logical && em->start + em->len > logical) {
4386                 free_extent_map(em);
4387                 return 0;
4388         } else if (em) {
4389                 free_extent_map(em);
4390         }
4391
4392         em = alloc_extent_map();
4393         if (!em)
4394                 return -ENOMEM;
4395         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4396         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4397         if (!map) {
4398                 free_extent_map(em);
4399                 return -ENOMEM;
4400         }
4401
4402         em->bdev = (struct block_device *)map;
4403         em->start = logical;
4404         em->len = length;
4405         em->block_start = 0;
4406         em->block_len = em->len;
4407
4408         map->num_stripes = num_stripes;
4409         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4410         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4411         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4412         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4413         map->type = btrfs_chunk_type(leaf, chunk);
4414         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4415         for (i = 0; i < num_stripes; i++) {
4416                 map->stripes[i].physical =
4417                         btrfs_stripe_offset_nr(leaf, chunk, i);
4418                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4419                 read_extent_buffer(leaf, uuid, (unsigned long)
4420                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4421                                    BTRFS_UUID_SIZE);
4422                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4423                                                         NULL);
4424                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4425                         kfree(map);
4426                         free_extent_map(em);
4427                         return -EIO;
4428                 }
4429                 if (!map->stripes[i].dev) {
4430                         map->stripes[i].dev =
4431                                 add_missing_dev(root, devid, uuid);
4432                         if (!map->stripes[i].dev) {
4433                                 kfree(map);
4434                                 free_extent_map(em);
4435                                 return -EIO;
4436                         }
4437                 }
4438                 map->stripes[i].dev->in_fs_metadata = 1;
4439         }
4440
4441         write_lock(&map_tree->map_tree.lock);
4442         ret = add_extent_mapping(&map_tree->map_tree, em);
4443         write_unlock(&map_tree->map_tree.lock);
4444         BUG_ON(ret); /* Tree corruption */
4445         free_extent_map(em);
4446
4447         return 0;
4448 }
4449
4450 static void fill_device_from_item(struct extent_buffer *leaf,
4451                                  struct btrfs_dev_item *dev_item,
4452                                  struct btrfs_device *device)
4453 {
4454         unsigned long ptr;
4455
4456         device->devid = btrfs_device_id(leaf, dev_item);
4457         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4458         device->total_bytes = device->disk_total_bytes;
4459         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4460         device->type = btrfs_device_type(leaf, dev_item);
4461         device->io_align = btrfs_device_io_align(leaf, dev_item);
4462         device->io_width = btrfs_device_io_width(leaf, dev_item);
4463         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4464
4465         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4466         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4467 }
4468
4469 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4470 {
4471         struct btrfs_fs_devices *fs_devices;
4472         int ret;
4473
4474         BUG_ON(!mutex_is_locked(&uuid_mutex));
4475
4476         fs_devices = root->fs_info->fs_devices->seed;
4477         while (fs_devices) {
4478                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4479                         ret = 0;
4480                         goto out;
4481                 }
4482                 fs_devices = fs_devices->seed;
4483         }
4484
4485         fs_devices = find_fsid(fsid);
4486         if (!fs_devices) {
4487                 ret = -ENOENT;
4488                 goto out;
4489         }
4490
4491         fs_devices = clone_fs_devices(fs_devices);
4492         if (IS_ERR(fs_devices)) {
4493                 ret = PTR_ERR(fs_devices);
4494                 goto out;
4495         }
4496
4497         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4498                                    root->fs_info->bdev_holder);
4499         if (ret) {
4500                 free_fs_devices(fs_devices);
4501                 goto out;
4502         }
4503
4504         if (!fs_devices->seeding) {
4505                 __btrfs_close_devices(fs_devices);
4506                 free_fs_devices(fs_devices);
4507                 ret = -EINVAL;
4508                 goto out;
4509         }
4510
4511         fs_devices->seed = root->fs_info->fs_devices->seed;
4512         root->fs_info->fs_devices->seed = fs_devices;
4513 out:
4514         return ret;
4515 }
4516
4517 static int read_one_dev(struct btrfs_root *root,
4518                         struct extent_buffer *leaf,
4519                         struct btrfs_dev_item *dev_item)
4520 {
4521         struct btrfs_device *device;
4522         u64 devid;
4523         int ret;
4524         u8 fs_uuid[BTRFS_UUID_SIZE];
4525         u8 dev_uuid[BTRFS_UUID_SIZE];
4526
4527         devid = btrfs_device_id(leaf, dev_item);
4528         read_extent_buffer(leaf, dev_uuid,
4529                            (unsigned long)btrfs_device_uuid(dev_item),
4530                            BTRFS_UUID_SIZE);
4531         read_extent_buffer(leaf, fs_uuid,
4532                            (unsigned long)btrfs_device_fsid(dev_item),
4533                            BTRFS_UUID_SIZE);
4534
4535         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4536                 ret = open_seed_devices(root, fs_uuid);
4537                 if (ret && !btrfs_test_opt(root, DEGRADED))
4538                         return ret;
4539         }
4540
4541         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4542         if (!device || !device->bdev) {
4543                 if (!btrfs_test_opt(root, DEGRADED))
4544                         return -EIO;
4545
4546                 if (!device) {
4547                         printk(KERN_WARNING "warning devid %llu missing\n",
4548                                (unsigned long long)devid);
4549                         device = add_missing_dev(root, devid, dev_uuid);
4550                         if (!device)
4551                                 return -ENOMEM;
4552                 } else if (!device->missing) {
4553                         /*
4554                          * this happens when a device that was properly setup
4555                          * in the device info lists suddenly goes bad.
4556                          * device->bdev is NULL, and so we have to set
4557                          * device->missing to one here
4558                          */
4559                         root->fs_info->fs_devices->missing_devices++;
4560                         device->missing = 1;
4561                 }
4562         }
4563
4564         if (device->fs_devices != root->fs_info->fs_devices) {
4565                 BUG_ON(device->writeable);
4566                 if (device->generation !=
4567                     btrfs_device_generation(leaf, dev_item))
4568                         return -EINVAL;
4569         }
4570
4571         fill_device_from_item(leaf, dev_item, device);
4572         device->dev_root = root->fs_info->dev_root;
4573         device->in_fs_metadata = 1;
4574         if (device->writeable) {
4575                 device->fs_devices->total_rw_bytes += device->total_bytes;
4576                 spin_lock(&root->fs_info->free_chunk_lock);
4577                 root->fs_info->free_chunk_space += device->total_bytes -
4578                         device->bytes_used;
4579                 spin_unlock(&root->fs_info->free_chunk_lock);
4580         }
4581         ret = 0;
4582         return ret;
4583 }
4584
4585 int btrfs_read_sys_array(struct btrfs_root *root)
4586 {
4587         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4588         struct extent_buffer *sb;
4589         struct btrfs_disk_key *disk_key;
4590         struct btrfs_chunk *chunk;
4591         u8 *ptr;
4592         unsigned long sb_ptr;
4593         int ret = 0;
4594         u32 num_stripes;
4595         u32 array_size;
4596         u32 len = 0;
4597         u32 cur;
4598         struct btrfs_key key;
4599
4600         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4601                                           BTRFS_SUPER_INFO_SIZE);
4602         if (!sb)
4603                 return -ENOMEM;
4604         btrfs_set_buffer_uptodate(sb);
4605         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4606         /*
4607          * The sb extent buffer is artifical and just used to read the system array.
4608          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4609          * pages up-to-date when the page is larger: extent does not cover the
4610          * whole page and consequently check_page_uptodate does not find all
4611          * the page's extents up-to-date (the hole beyond sb),
4612          * write_extent_buffer then triggers a WARN_ON.
4613          *
4614          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4615          * but sb spans only this function. Add an explicit SetPageUptodate call
4616          * to silence the warning eg. on PowerPC 64.
4617          */
4618         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4619                 SetPageUptodate(sb->pages[0]);
4620
4621         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4622         array_size = btrfs_super_sys_array_size(super_copy);
4623
4624         ptr = super_copy->sys_chunk_array;
4625         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4626         cur = 0;
4627
4628         while (cur < array_size) {
4629                 disk_key = (struct btrfs_disk_key *)ptr;
4630                 btrfs_disk_key_to_cpu(&key, disk_key);
4631
4632                 len = sizeof(*disk_key); ptr += len;
4633                 sb_ptr += len;
4634                 cur += len;
4635
4636                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4637                         chunk = (struct btrfs_chunk *)sb_ptr;
4638                         ret = read_one_chunk(root, &key, sb, chunk);
4639                         if (ret)
4640                                 break;
4641                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4642                         len = btrfs_chunk_item_size(num_stripes);
4643                 } else {
4644                         ret = -EIO;
4645                         break;
4646                 }
4647                 ptr += len;
4648                 sb_ptr += len;
4649                 cur += len;
4650         }
4651         free_extent_buffer(sb);
4652         return ret;
4653 }
4654
4655 int btrfs_read_chunk_tree(struct btrfs_root *root)
4656 {
4657         struct btrfs_path *path;
4658         struct extent_buffer *leaf;
4659         struct btrfs_key key;
4660         struct btrfs_key found_key;
4661         int ret;
4662         int slot;
4663
4664         root = root->fs_info->chunk_root;
4665
4666         path = btrfs_alloc_path();
4667         if (!path)
4668                 return -ENOMEM;
4669
4670         mutex_lock(&uuid_mutex);
4671         lock_chunks(root);
4672
4673         /* first we search for all of the device items, and then we
4674          * read in all of the chunk items.  This way we can create chunk
4675          * mappings that reference all of the devices that are afound
4676          */
4677         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4678         key.offset = 0;
4679         key.type = 0;
4680 again:
4681         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4682         if (ret < 0)
4683                 goto error;
4684         while (1) {
4685                 leaf = path->nodes[0];
4686                 slot = path->slots[0];
4687                 if (slot >= btrfs_header_nritems(leaf)) {
4688                         ret = btrfs_next_leaf(root, path);
4689                         if (ret == 0)
4690                                 continue;
4691                         if (ret < 0)
4692                                 goto error;
4693                         break;
4694                 }
4695                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4696                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4697                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4698                                 break;
4699                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4700                                 struct btrfs_dev_item *dev_item;
4701                                 dev_item = btrfs_item_ptr(leaf, slot,
4702                                                   struct btrfs_dev_item);
4703                                 ret = read_one_dev(root, leaf, dev_item);
4704                                 if (ret)
4705                                         goto error;
4706                         }
4707                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4708                         struct btrfs_chunk *chunk;
4709                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4710                         ret = read_one_chunk(root, &found_key, leaf, chunk);
4711                         if (ret)
4712                                 goto error;
4713                 }
4714                 path->slots[0]++;
4715         }
4716         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4717                 key.objectid = 0;
4718                 btrfs_release_path(path);
4719                 goto again;
4720         }
4721         ret = 0;
4722 error:
4723         unlock_chunks(root);
4724         mutex_unlock(&uuid_mutex);
4725
4726         btrfs_free_path(path);
4727         return ret;
4728 }
4729
4730 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
4731 {
4732         int i;
4733
4734         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4735                 btrfs_dev_stat_reset(dev, i);
4736 }
4737
4738 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
4739 {
4740         struct btrfs_key key;
4741         struct btrfs_key found_key;
4742         struct btrfs_root *dev_root = fs_info->dev_root;
4743         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4744         struct extent_buffer *eb;
4745         int slot;
4746         int ret = 0;
4747         struct btrfs_device *device;
4748         struct btrfs_path *path = NULL;
4749         int i;
4750
4751         path = btrfs_alloc_path();
4752         if (!path) {
4753                 ret = -ENOMEM;
4754                 goto out;
4755         }
4756
4757         mutex_lock(&fs_devices->device_list_mutex);
4758         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4759                 int item_size;
4760                 struct btrfs_dev_stats_item *ptr;
4761
4762                 key.objectid = 0;
4763                 key.type = BTRFS_DEV_STATS_KEY;
4764                 key.offset = device->devid;
4765                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
4766                 if (ret) {
4767                         __btrfs_reset_dev_stats(device);
4768                         device->dev_stats_valid = 1;
4769                         btrfs_release_path(path);
4770                         continue;
4771                 }
4772                 slot = path->slots[0];
4773                 eb = path->nodes[0];
4774                 btrfs_item_key_to_cpu(eb, &found_key, slot);
4775                 item_size = btrfs_item_size_nr(eb, slot);
4776
4777                 ptr = btrfs_item_ptr(eb, slot,
4778                                      struct btrfs_dev_stats_item);
4779
4780                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4781                         if (item_size >= (1 + i) * sizeof(__le64))
4782                                 btrfs_dev_stat_set(device, i,
4783                                         btrfs_dev_stats_value(eb, ptr, i));
4784                         else
4785                                 btrfs_dev_stat_reset(device, i);
4786                 }
4787
4788                 device->dev_stats_valid = 1;
4789                 btrfs_dev_stat_print_on_load(device);
4790                 btrfs_release_path(path);
4791         }
4792         mutex_unlock(&fs_devices->device_list_mutex);
4793
4794 out:
4795         btrfs_free_path(path);
4796         return ret < 0 ? ret : 0;
4797 }
4798
4799 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
4800                                 struct btrfs_root *dev_root,
4801                                 struct btrfs_device *device)
4802 {
4803         struct btrfs_path *path;
4804         struct btrfs_key key;
4805         struct extent_buffer *eb;
4806         struct btrfs_dev_stats_item *ptr;
4807         int ret;
4808         int i;
4809
4810         key.objectid = 0;
4811         key.type = BTRFS_DEV_STATS_KEY;
4812         key.offset = device->devid;
4813
4814         path = btrfs_alloc_path();
4815         BUG_ON(!path);
4816         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
4817         if (ret < 0) {
4818                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
4819                               ret, rcu_str_deref(device->name));
4820                 goto out;
4821         }
4822
4823         if (ret == 0 &&
4824             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
4825                 /* need to delete old one and insert a new one */
4826                 ret = btrfs_del_item(trans, dev_root, path);
4827                 if (ret != 0) {
4828                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
4829                                       rcu_str_deref(device->name), ret);
4830                         goto out;
4831                 }
4832                 ret = 1;
4833         }
4834
4835         if (ret == 1) {
4836                 /* need to insert a new item */
4837                 btrfs_release_path(path);
4838                 ret = btrfs_insert_empty_item(trans, dev_root, path,
4839                                               &key, sizeof(*ptr));
4840                 if (ret < 0) {
4841                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
4842                                       rcu_str_deref(device->name), ret);
4843                         goto out;
4844                 }
4845         }
4846
4847         eb = path->nodes[0];
4848         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
4849         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4850                 btrfs_set_dev_stats_value(eb, ptr, i,
4851                                           btrfs_dev_stat_read(device, i));
4852         btrfs_mark_buffer_dirty(eb);
4853
4854 out:
4855         btrfs_free_path(path);
4856         return ret;
4857 }
4858
4859 /*
4860  * called from commit_transaction. Writes all changed device stats to disk.
4861  */
4862 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
4863                         struct btrfs_fs_info *fs_info)
4864 {
4865         struct btrfs_root *dev_root = fs_info->dev_root;
4866         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4867         struct btrfs_device *device;
4868         int ret = 0;
4869
4870         mutex_lock(&fs_devices->device_list_mutex);
4871         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4872                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
4873                         continue;
4874
4875                 ret = update_dev_stat_item(trans, dev_root, device);
4876                 if (!ret)
4877                         device->dev_stats_dirty = 0;
4878         }
4879         mutex_unlock(&fs_devices->device_list_mutex);
4880
4881         return ret;
4882 }
4883
4884 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
4885 {
4886         btrfs_dev_stat_inc(dev, index);
4887         btrfs_dev_stat_print_on_error(dev);
4888 }
4889
4890 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
4891 {
4892         if (!dev->dev_stats_valid)
4893                 return;
4894         printk_ratelimited_in_rcu(KERN_ERR
4895                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4896                            rcu_str_deref(dev->name),
4897                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4898                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4899                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4900                            btrfs_dev_stat_read(dev,
4901                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
4902                            btrfs_dev_stat_read(dev,
4903                                                BTRFS_DEV_STAT_GENERATION_ERRS));
4904 }
4905
4906 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
4907 {
4908         int i;
4909
4910         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4911                 if (btrfs_dev_stat_read(dev, i) != 0)
4912                         break;
4913         if (i == BTRFS_DEV_STAT_VALUES_MAX)
4914                 return; /* all values == 0, suppress message */
4915
4916         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4917                rcu_str_deref(dev->name),
4918                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4919                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4920                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4921                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
4922                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
4923 }
4924
4925 int btrfs_get_dev_stats(struct btrfs_root *root,
4926                         struct btrfs_ioctl_get_dev_stats *stats)
4927 {
4928         struct btrfs_device *dev;
4929         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4930         int i;
4931
4932         mutex_lock(&fs_devices->device_list_mutex);
4933         dev = btrfs_find_device(root, stats->devid, NULL, NULL);
4934         mutex_unlock(&fs_devices->device_list_mutex);
4935
4936         if (!dev) {
4937                 printk(KERN_WARNING
4938                        "btrfs: get dev_stats failed, device not found\n");
4939                 return -ENODEV;
4940         } else if (!dev->dev_stats_valid) {
4941                 printk(KERN_WARNING
4942                        "btrfs: get dev_stats failed, not yet valid\n");
4943                 return -ENODEV;
4944         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
4945                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4946                         if (stats->nr_items > i)
4947                                 stats->values[i] =
4948                                         btrfs_dev_stat_read_and_reset(dev, i);
4949                         else
4950                                 btrfs_dev_stat_reset(dev, i);
4951                 }
4952         } else {
4953                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4954                         if (stats->nr_items > i)
4955                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
4956         }
4957         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
4958                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
4959         return 0;
4960 }