fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <asm/div64.h>
  29 #include "compat.h"
  30 #include "ctree.h"
  31 #include "extent_map.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "print-tree.h"
  35 #include "volumes.h"
  36 #include "async-thread.h"
  37 #include "check-integrity.h"
  38 #include "rcu-string.h"
  39
  40 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  41                                 struct btrfs_root *root,
  42                                 struct btrfs_device *device);
  43 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  44 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  45 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  46
  47 static DEFINE_MUTEX(uuid_mutex);
  48 static LIST_HEAD(fs_uuids);
  49
  50 static void lock_chunks(struct btrfs_root *root)
  51 {
  52         mutex_lock(&root->fs_info->chunk_mutex);
  53 }
  54
  55 static void unlock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_unlock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  61 {
  62         struct btrfs_device *device;
  63         WARN_ON(fs_devices->opened);
  64         while (!list_empty(&fs_devices->devices)) {
  65                 device = list_entry(fs_devices->devices.next,
  66                                     struct btrfs_device, dev_list);
  67                 list_del(&device->dev_list);
  68                 rcu_string_free(device->name);
  69                 kfree(device);
  70         }
  71         kfree(fs_devices);
  72 }
  73
  74 void btrfs_cleanup_fs_uuids(void)
  75 {
  76         struct btrfs_fs_devices *fs_devices;
  77
  78         while (!list_empty(&fs_uuids)) {
  79                 fs_devices = list_entry(fs_uuids.next,
  80                                         struct btrfs_fs_devices, list);
  81                 list_del(&fs_devices->list);
  82                 free_fs_devices(fs_devices);
  83         }
  84 }
  85
  86 static noinline struct btrfs_device *__find_device(struct list_head *head,
  87                                                    u64 devid, u8 *uuid)
  88 {
  89         struct btrfs_device *dev;
  90
  91         list_for_each_entry(dev, head, dev_list) {
  92                 if (dev->devid == devid &&
  93                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  94                         return dev;
  95                 }
  96         }
  97         return NULL;
  98 }
  99
 100 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 101 {
 102         struct btrfs_fs_devices *fs_devices;
 103
 104         list_for_each_entry(fs_devices, &fs_uuids, list) {
 105                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 106                         return fs_devices;
 107         }
 108         return NULL;
 109 }
 110
 111 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 112                         struct bio *head, struct bio *tail)
 113 {
 114
 115         struct bio *old_head;
 116
 117         old_head = pending_bios->head;
 118         pending_bios->head = head;
 119         if (pending_bios->tail)
 120                 tail->bi_next = old_head;
 121         else
 122                 pending_bios->tail = tail;
 123 }
 124
 125 /*
 126  * we try to collect pending bios for a device so we don't get a large
 127  * number of procs sending bios down to the same device.  This greatly
 128  * improves the schedulers ability to collect and merge the bios.
 129  *
 130  * But, it also turns into a long list of bios to process and that is sure
 131  * to eventually make the worker thread block.  The solution here is to
 132  * make some progress and then put this work struct back at the end of
 133  * the list if the block device is congested.  This way, multiple devices
 134  * can make progress from a single worker thread.
 135  */
 136 static noinline void run_scheduled_bios(struct btrfs_device *device)
 137 {
 138         struct bio *pending;
 139         struct backing_dev_info *bdi;
 140         struct btrfs_fs_info *fs_info;
 141         struct btrfs_pending_bios *pending_bios;
 142         struct bio *tail;
 143         struct bio *cur;
 144         int again = 0;
 145         unsigned long num_run;
 146         unsigned long batch_run = 0;
 147         unsigned long limit;
 148         unsigned long last_waited = 0;
 149         int force_reg = 0;
 150         int sync_pending = 0;
 151         struct blk_plug plug;
 152
 153         /*
 154          * this function runs all the bios we've collected for
 155          * a particular device.  We don't want to wander off to
 156          * another device without first sending all of these down.
 157          * So, setup a plug here and finish it off before we return
 158          */
 159         blk_start_plug(&plug);
 160
 161         bdi = blk_get_backing_dev_info(device->bdev);
 162         fs_info = device->dev_root->fs_info;
 163         limit = btrfs_async_submit_limit(fs_info);
 164         limit = limit * 2 / 3;
 165
 166 loop:
 167         spin_lock(&device->io_lock);
 168
 169 loop_lock:
 170         num_run = 0;
 171
 172         /* take all the bios off the list at once and process them
 173          * later on (without the lock held).  But, remember the
 174          * tail and other pointers so the bios can be properly reinserted
 175          * into the list if we hit congestion
 176          */
 177         if (!force_reg && device->pending_sync_bios.head) {
 178                 pending_bios = &device->pending_sync_bios;
 179                 force_reg = 1;
 180         } else {
 181                 pending_bios = &device->pending_bios;
 182                 force_reg = 0;
 183         }
 184
 185         pending = pending_bios->head;
 186         tail = pending_bios->tail;
 187         WARN_ON(pending && !tail);
 188
 189         /*
 190          * if pending was null this time around, no bios need processing
 191          * at all and we can stop.  Otherwise it'll loop back up again
 192          * and do an additional check so no bios are missed.
 193          *
 194          * device->running_pending is used to synchronize with the
 195          * schedule_bio code.
 196          */
 197         if (device->pending_sync_bios.head == NULL &&
 198             device->pending_bios.head == NULL) {
 199                 again = 0;
 200                 device->running_pending = 0;
 201         } else {
 202                 again = 1;
 203                 device->running_pending = 1;
 204         }
 205
 206         pending_bios->head = NULL;
 207         pending_bios->tail = NULL;
 208
 209         spin_unlock(&device->io_lock);
 210
 211         while (pending) {
 212
 213                 rmb();
 214                 /* we want to work on both lists, but do more bios on the
 215                  * sync list than the regular list
 216                  */
 217                 if ((num_run > 32 &&
 218                     pending_bios != &device->pending_sync_bios &&
 219                     device->pending_sync_bios.head) ||
 220                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 221                     device->pending_bios.head)) {
 222                         spin_lock(&device->io_lock);
 223                         requeue_list(pending_bios, pending, tail);
 224                         goto loop_lock;
 225                 }
 226
 227                 cur = pending;
 228                 pending = pending->bi_next;
 229                 cur->bi_next = NULL;
 230
 231                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 232                     waitqueue_active(&fs_info->async_submit_wait))
 233                         wake_up(&fs_info->async_submit_wait);
 234
 235                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 236
 237                 /*
 238                  * if we're doing the sync list, record that our
 239                  * plug has some sync requests on it
 240                  *
 241                  * If we're doing the regular list and there are
 242                  * sync requests sitting around, unplug before
 243                  * we add more
 244                  */
 245                 if (pending_bios == &device->pending_sync_bios) {
 246                         sync_pending = 1;
 247                 } else if (sync_pending) {
 248                         blk_finish_plug(&plug);
 249                         blk_start_plug(&plug);
 250                         sync_pending = 0;
 251                 }
 252
 253                 btrfsic_submit_bio(cur->bi_rw, cur);
 254                 num_run++;
 255                 batch_run++;
 256                 if (need_resched())
 257                         cond_resched();
 258
 259                 /*
 260                  * we made progress, there is more work to do and the bdi
 261                  * is now congested.  Back off and let other work structs
 262                  * run instead
 263                  */
 264                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 265                     fs_info->fs_devices->open_devices > 1) {
 266                         struct io_context *ioc;
 267
 268                         ioc = current->io_context;
 269
 270                         /*
 271                          * the main goal here is that we don't want to
 272                          * block if we're going to be able to submit
 273                          * more requests without blocking.
 274                          *
 275                          * This code does two great things, it pokes into
 276                          * the elevator code from a filesystem _and_
 277                          * it makes assumptions about how batching works.
 278                          */
 279                         if (ioc && ioc->nr_batch_requests > 0 &&
 280                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 281                             (last_waited == 0 ||
 282                              ioc->last_waited == last_waited)) {
 283                                 /*
 284                                  * we want to go through our batch of
 285                                  * requests and stop.  So, we copy out
 286                                  * the ioc->last_waited time and test
 287                                  * against it before looping
 288                                  */
 289                                 last_waited = ioc->last_waited;
 290                                 if (need_resched())
 291                                         cond_resched();
 292                                 continue;
 293                         }
 294                         spin_lock(&device->io_lock);
 295                         requeue_list(pending_bios, pending, tail);
 296                         device->running_pending = 1;
 297
 298                         spin_unlock(&device->io_lock);
 299                         btrfs_requeue_work(&device->work);
 300                         goto done;
 301                 }
 302                 /* unplug every 64 requests just for good measure */
 303                 if (batch_run % 64 == 0) {
 304                         blk_finish_plug(&plug);
 305                         blk_start_plug(&plug);
 306                         sync_pending = 0;
 307                 }
 308         }
 309
 310         cond_resched();
 311         if (again)
 312                 goto loop;
 313
 314         spin_lock(&device->io_lock);
 315         if (device->pending_bios.head || device->pending_sync_bios.head)
 316                 goto loop_lock;
 317         spin_unlock(&device->io_lock);
 318
 319 done:
 320         blk_finish_plug(&plug);
 321 }
 322
 323 static void pending_bios_fn(struct btrfs_work *work)
 324 {
 325         struct btrfs_device *device;
 326
 327         device = container_of(work, struct btrfs_device, work);
 328         run_scheduled_bios(device);
 329 }
 330
 331 static noinline int device_list_add(const char *path,
 332                            struct btrfs_super_block *disk_super,
 333                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 334 {
 335         struct btrfs_device *device;
 336         struct btrfs_fs_devices *fs_devices;
 337         struct rcu_string *name;
 338         u64 found_transid = btrfs_super_generation(disk_super);
 339
 340         fs_devices = find_fsid(disk_super->fsid);
 341         if (!fs_devices) {
 342                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 343                 if (!fs_devices)
 344                         return -ENOMEM;
 345                 INIT_LIST_HEAD(&fs_devices->devices);
 346                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 347                 list_add(&fs_devices->list, &fs_uuids);
 348                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 349                 fs_devices->latest_devid = devid;
 350                 fs_devices->latest_trans = found_transid;
 351                 mutex_init(&fs_devices->device_list_mutex);
 352                 device = NULL;
 353         } else {
 354                 device = __find_device(&fs_devices->devices, devid,
 355                                        disk_super->dev_item.uuid);
 356         }
 357         if (!device) {
 358                 if (fs_devices->opened)
 359                         return -EBUSY;
 360
 361                 device = kzalloc(sizeof(*device), GFP_NOFS);
 362                 if (!device) {
 363                         /* we can safely leave the fs_devices entry around */
 364                         return -ENOMEM;
 365                 }
 366                 device->devid = devid;
 367                 device->dev_stats_valid = 0;
 368                 device->work.func = pending_bios_fn;
 369                 memcpy(device->uuid, disk_super->dev_item.uuid,
 370                        BTRFS_UUID_SIZE);
 371                 spin_lock_init(&device->io_lock);
 372
 373                 name = rcu_string_strdup(path, GFP_NOFS);
 374                 if (!name) {
 375                         kfree(device);
 376                         return -ENOMEM;
 377                 }
 378                 rcu_assign_pointer(device->name, name);
 379                 INIT_LIST_HEAD(&device->dev_alloc_list);
 380
 381                 /* init readahead state */
 382                 spin_lock_init(&device->reada_lock);
 383                 device->reada_curr_zone = NULL;
 384                 atomic_set(&device->reada_in_flight, 0);
 385                 device->reada_next = 0;
 386                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 387                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 388
 389                 mutex_lock(&fs_devices->device_list_mutex);
 390                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 391                 mutex_unlock(&fs_devices->device_list_mutex);
 392
 393                 device->fs_devices = fs_devices;
 394                 fs_devices->num_devices++;
 395         } else if (!device->name || strcmp(device->name->str, path)) {
 396                 name = rcu_string_strdup(path, GFP_NOFS);
 397                 if (!name)
 398                         return -ENOMEM;
 399                 rcu_string_free(device->name);
 400                 rcu_assign_pointer(device->name, name);
 401                 if (device->missing) {
 402                         fs_devices->missing_devices--;
 403                         device->missing = 0;
 404                 }
 405         }
 406
 407         if (found_transid > fs_devices->latest_trans) {
 408                 fs_devices->latest_devid = devid;
 409                 fs_devices->latest_trans = found_transid;
 410         }
 411         *fs_devices_ret = fs_devices;
 412         return 0;
 413 }
 414
 415 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 416 {
 417         struct btrfs_fs_devices *fs_devices;
 418         struct btrfs_device *device;
 419         struct btrfs_device *orig_dev;
 420
 421         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 422         if (!fs_devices)
 423                 return ERR_PTR(-ENOMEM);
 424
 425         INIT_LIST_HEAD(&fs_devices->devices);
 426         INIT_LIST_HEAD(&fs_devices->alloc_list);
 427         INIT_LIST_HEAD(&fs_devices->list);
 428         mutex_init(&fs_devices->device_list_mutex);
 429         fs_devices->latest_devid = orig->latest_devid;
 430         fs_devices->latest_trans = orig->latest_trans;
 431         fs_devices->total_devices = orig->total_devices;
 432         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 433
 434         /* We have held the volume lock, it is safe to get the devices. */
 435         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 436                 struct rcu_string *name;
 437
 438                 device = kzalloc(sizeof(*device), GFP_NOFS);
 439                 if (!device)
 440                         goto error;
 441
 442                 /*
 443                  * This is ok to do without rcu read locked because we hold the
 444                  * uuid mutex so nothing we touch in here is going to disappear.
 445                  */
 446                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 447                 if (!name) {
 448                         kfree(device);
 449                         goto error;
 450                 }
 451                 rcu_assign_pointer(device->name, name);
 452
 453                 device->devid = orig_dev->devid;
 454                 device->work.func = pending_bios_fn;
 455                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 456                 spin_lock_init(&device->io_lock);
 457                 INIT_LIST_HEAD(&device->dev_list);
 458                 INIT_LIST_HEAD(&device->dev_alloc_list);
 459
 460                 list_add(&device->dev_list, &fs_devices->devices);
 461                 device->fs_devices = fs_devices;
 462                 fs_devices->num_devices++;
 463         }
 464         return fs_devices;
 465 error:
 466         free_fs_devices(fs_devices);
 467         return ERR_PTR(-ENOMEM);
 468 }
 469
 470 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 471 {
 472         struct btrfs_device *device, *next;
 473
 474         struct block_device *latest_bdev = NULL;
 475         u64 latest_devid = 0;
 476         u64 latest_transid = 0;
 477
 478         mutex_lock(&uuid_mutex);
 479 again:
 480         /* This is the initialized path, it is safe to release the devices. */
 481         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 482                 if (device->in_fs_metadata) {
 483                         if (!latest_transid ||
 484                             device->generation > latest_transid) {
 485                                 latest_devid = device->devid;
 486                                 latest_transid = device->generation;
 487                                 latest_bdev = device->bdev;
 488                         }
 489                         continue;
 490                 }
 491
 492                 if (device->bdev) {
 493                         blkdev_put(device->bdev, device->mode);
 494                         device->bdev = NULL;
 495                         fs_devices->open_devices--;
 496                 }
 497                 if (device->writeable) {
 498                         list_del_init(&device->dev_alloc_list);
 499                         device->writeable = 0;
 500                         fs_devices->rw_devices--;
 501                 }
 502                 list_del_init(&device->dev_list);
 503                 fs_devices->num_devices--;
 504                 rcu_string_free(device->name);
 505                 kfree(device);
 506         }
 507
 508         if (fs_devices->seed) {
 509                 fs_devices = fs_devices->seed;
 510                 goto again;
 511         }
 512
 513         fs_devices->latest_bdev = latest_bdev;
 514         fs_devices->latest_devid = latest_devid;
 515         fs_devices->latest_trans = latest_transid;
 516
 517         mutex_unlock(&uuid_mutex);
 518 }
 519
 520 static void __free_device(struct work_struct *work)
 521 {
 522         struct btrfs_device *device;
 523
 524         device = container_of(work, struct btrfs_device, rcu_work);
 525
 526         if (device->bdev)
 527                 blkdev_put(device->bdev, device->mode);
 528
 529         rcu_string_free(device->name);
 530         kfree(device);
 531 }
 532
 533 static void free_device(struct rcu_head *head)
 534 {
 535         struct btrfs_device *device;
 536
 537         device = container_of(head, struct btrfs_device, rcu);
 538
 539         INIT_WORK(&device->rcu_work, __free_device);
 540         schedule_work(&device->rcu_work);
 541 }
 542
 543 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 544 {
 545         struct btrfs_device *device;
 546
 547         if (--fs_devices->opened > 0)
 548                 return 0;
 549
 550         mutex_lock(&fs_devices->device_list_mutex);
 551         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 552                 struct btrfs_device *new_device;
 553                 struct rcu_string *name;
 554
 555                 if (device->bdev)
 556                         fs_devices->open_devices--;
 557
 558                 if (device->writeable) {
 559                         list_del_init(&device->dev_alloc_list);
 560                         fs_devices->rw_devices--;
 561                 }
 562
 563                 if (device->can_discard)
 564                         fs_devices->num_can_discard--;
 565
 566                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 567                 BUG_ON(!new_device); /* -ENOMEM */
 568                 memcpy(new_device, device, sizeof(*new_device));
 569
 570                 /* Safe because we are under uuid_mutex */
 571                 if (device->name) {
 572                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 573                         BUG_ON(device->name && !name); /* -ENOMEM */
 574                         rcu_assign_pointer(new_device->name, name);
 575                 }
 576                 new_device->bdev = NULL;
 577                 new_device->writeable = 0;
 578                 new_device->in_fs_metadata = 0;
 579                 new_device->can_discard = 0;
 580                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 581
 582                 call_rcu(&device->rcu, free_device);
 583         }
 584         mutex_unlock(&fs_devices->device_list_mutex);
 585
 586         WARN_ON(fs_devices->open_devices);
 587         WARN_ON(fs_devices->rw_devices);
 588         fs_devices->opened = 0;
 589         fs_devices->seeding = 0;
 590
 591         return 0;
 592 }
 593
 594 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 595 {
 596         struct btrfs_fs_devices *seed_devices = NULL;
 597         int ret;
 598
 599         mutex_lock(&uuid_mutex);
 600         ret = __btrfs_close_devices(fs_devices);
 601         if (!fs_devices->opened) {
 602                 seed_devices = fs_devices->seed;
 603                 fs_devices->seed = NULL;
 604         }
 605         mutex_unlock(&uuid_mutex);
 606
 607         while (seed_devices) {
 608                 fs_devices = seed_devices;
 609                 seed_devices = fs_devices->seed;
 610                 __btrfs_close_devices(fs_devices);
 611                 free_fs_devices(fs_devices);
 612         }
 613         return ret;
 614 }
 615
 616 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 617                                 fmode_t flags, void *holder)
 618 {
 619         struct request_queue *q;
 620         struct block_device *bdev;
 621         struct list_head *head = &fs_devices->devices;
 622         struct btrfs_device *device;
 623         struct block_device *latest_bdev = NULL;
 624         struct buffer_head *bh;
 625         struct btrfs_super_block *disk_super;
 626         u64 latest_devid = 0;
 627         u64 latest_transid = 0;
 628         u64 devid;
 629         int seeding = 1;
 630         int ret = 0;
 631
 632         flags |= FMODE_EXCL;
 633
 634         list_for_each_entry(device, head, dev_list) {
 635                 if (device->bdev)
 636                         continue;
 637                 if (!device->name)
 638                         continue;
 639
 640                 bdev = blkdev_get_by_path(device->name->str, flags, holder);
 641                 if (IS_ERR(bdev)) {
 642                         printk(KERN_INFO "open %s failed\n", device->name->str);
 643                         goto error;
 644                 }
 645                 filemap_write_and_wait(bdev->bd_inode->i_mapping);
 646                 invalidate_bdev(bdev);
 647                 set_blocksize(bdev, 4096);
 648
 649                 bh = btrfs_read_dev_super(bdev);
 650                 if (!bh)
 651                         goto error_close;
 652
 653                 disk_super = (struct btrfs_super_block *)bh->b_data;
 654                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 655                 if (devid != device->devid)
 656                         goto error_brelse;
 657
 658                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 659                            BTRFS_UUID_SIZE))
 660                         goto error_brelse;
 661
 662                 device->generation = btrfs_super_generation(disk_super);
 663                 if (!latest_transid || device->generation > latest_transid) {
 664                         latest_devid = devid;
 665                         latest_transid = device->generation;
 666                         latest_bdev = bdev;
 667                 }
 668
 669                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 670                         device->writeable = 0;
 671                 } else {
 672                         device->writeable = !bdev_read_only(bdev);
 673                         seeding = 0;
 674                 }
 675
 676                 q = bdev_get_queue(bdev);
 677                 if (blk_queue_discard(q)) {
 678                         device->can_discard = 1;
 679                         fs_devices->num_can_discard++;
 680                 }
 681
 682                 device->bdev = bdev;
 683                 device->in_fs_metadata = 0;
 684                 device->mode = flags;
 685
 686                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 687                         fs_devices->rotating = 1;
 688
 689                 fs_devices->open_devices++;
 690                 if (device->writeable) {
 691                         fs_devices->rw_devices++;
 692                         list_add(&device->dev_alloc_list,
 693                                  &fs_devices->alloc_list);
 694                 }
 695                 brelse(bh);
 696                 continue;
 697
 698 error_brelse:
 699                 brelse(bh);
 700 error_close:
 701                 blkdev_put(bdev, flags);
 702 error:
 703                 continue;
 704         }
 705         if (fs_devices->open_devices == 0) {
 706                 ret = -EINVAL;
 707                 goto out;
 708         }
 709         fs_devices->seeding = seeding;
 710         fs_devices->opened = 1;
 711         fs_devices->latest_bdev = latest_bdev;
 712         fs_devices->latest_devid = latest_devid;
 713         fs_devices->latest_trans = latest_transid;
 714         fs_devices->total_rw_bytes = 0;
 715 out:
 716         return ret;
 717 }
 718
 719 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 720                        fmode_t flags, void *holder)
 721 {
 722         int ret;
 723
 724         mutex_lock(&uuid_mutex);
 725         if (fs_devices->opened) {
 726                 fs_devices->opened++;
 727                 ret = 0;
 728         } else {
 729                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 730         }
 731         mutex_unlock(&uuid_mutex);
 732         return ret;
 733 }
 734
 735 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 736                           struct btrfs_fs_devices **fs_devices_ret)
 737 {
 738         struct btrfs_super_block *disk_super;
 739         struct block_device *bdev;
 740         struct buffer_head *bh;
 741         int ret;
 742         u64 devid;
 743         u64 transid;
 744         u64 total_devices;
 745
 746         flags |= FMODE_EXCL;
 747         bdev = blkdev_get_by_path(path, flags, holder);
 748
 749         if (IS_ERR(bdev)) {
 750                 ret = PTR_ERR(bdev);
 751                 goto error;
 752         }
 753
 754         mutex_lock(&uuid_mutex);
 755         ret = set_blocksize(bdev, 4096);
 756         if (ret)
 757                 goto error_close;
 758         bh = btrfs_read_dev_super(bdev);
 759         if (!bh) {
 760                 ret = -EINVAL;
 761                 goto error_close;
 762         }
 763         disk_super = (struct btrfs_super_block *)bh->b_data;
 764         devid = btrfs_stack_device_id(&disk_super->dev_item);
 765         transid = btrfs_super_generation(disk_super);
 766         total_devices = btrfs_super_num_devices(disk_super);
 767         if (disk_super->label[0])
 768                 printk(KERN_INFO "device label %s ", disk_super->label);
 769         else
 770                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 771         printk(KERN_CONT "devid %llu transid %llu %s\n",
 772                (unsigned long long)devid, (unsigned long long)transid, path);
 773         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 774         if (!ret && fs_devices_ret)
 775                 (*fs_devices_ret)->total_devices = total_devices;
 776         brelse(bh);
 777 error_close:
 778         mutex_unlock(&uuid_mutex);
 779         blkdev_put(bdev, flags);
 780 error:
 781         return ret;
 782 }
 783
 784 /* helper to account the used device space in the range */
 785 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 786                                    u64 end, u64 *length)
 787 {
 788         struct btrfs_key key;
 789         struct btrfs_root *root = device->dev_root;
 790         struct btrfs_dev_extent *dev_extent;
 791         struct btrfs_path *path;
 792         u64 extent_end;
 793         int ret;
 794         int slot;
 795         struct extent_buffer *l;
 796
 797         *length = 0;
 798
 799         if (start >= device->total_bytes)
 800                 return 0;
 801
 802         path = btrfs_alloc_path();
 803         if (!path)
 804                 return -ENOMEM;
 805         path->reada = 2;
 806
 807         key.objectid = device->devid;
 808         key.offset = start;
 809         key.type = BTRFS_DEV_EXTENT_KEY;
 810
 811         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 812         if (ret < 0)
 813                 goto out;
 814         if (ret > 0) {
 815                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 816                 if (ret < 0)
 817                         goto out;
 818         }
 819
 820         while (1) {
 821                 l = path->nodes[0];
 822                 slot = path->slots[0];
 823                 if (slot >= btrfs_header_nritems(l)) {
 824                         ret = btrfs_next_leaf(root, path);
 825                         if (ret == 0)
 826                                 continue;
 827                         if (ret < 0)
 828                                 goto out;
 829
 830                         break;
 831                 }
 832                 btrfs_item_key_to_cpu(l, &key, slot);
 833
 834                 if (key.objectid < device->devid)
 835                         goto next;
 836
 837                 if (key.objectid > device->devid)
 838                         break;
 839
 840                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 841                         goto next;
 842
 843                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 844                 extent_end = key.offset + btrfs_dev_extent_length(l,
 845                                                                   dev_extent);
 846                 if (key.offset <= start && extent_end > end) {
 847                         *length = end - start + 1;
 848                         break;
 849                 } else if (key.offset <= start && extent_end > start)
 850                         *length += extent_end - start;
 851                 else if (key.offset > start && extent_end <= end)
 852                         *length += extent_end - key.offset;
 853                 else if (key.offset > start && key.offset <= end) {
 854                         *length += end - key.offset + 1;
 855                         break;
 856                 } else if (key.offset > end)
 857                         break;
 858
 859 next:
 860                 path->slots[0]++;
 861         }
 862         ret = 0;
 863 out:
 864         btrfs_free_path(path);
 865         return ret;
 866 }
 867
 868 /*
 869  * find_free_dev_extent - find free space in the specified device
 870  * @device:     the device which we search the free space in
 871  * @num_bytes:  the size of the free space that we need
 872  * @start:      store the start of the free space.
 873  * @len:        the size of the free space. that we find, or the size of the max
 874  *              free space if we don't find suitable free space
 875  *
 876  * this uses a pretty simple search, the expectation is that it is
 877  * called very infrequently and that a given device has a small number
 878  * of extents
 879  *
 880  * @start is used to store the start of the free space if we find. But if we
 881  * don't find suitable free space, it will be used to store the start position
 882  * of the max free space.
 883  *
 884  * @len is used to store the size of the free space that we find.
 885  * But if we don't find suitable free space, it is used to store the size of
 886  * the max free space.
 887  */
 888 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 889                          u64 *start, u64 *len)
 890 {
 891         struct btrfs_key key;
 892         struct btrfs_root *root = device->dev_root;
 893         struct btrfs_dev_extent *dev_extent;
 894         struct btrfs_path *path;
 895         u64 hole_size;
 896         u64 max_hole_start;
 897         u64 max_hole_size;
 898         u64 extent_end;
 899         u64 search_start;
 900         u64 search_end = device->total_bytes;
 901         int ret;
 902         int slot;
 903         struct extent_buffer *l;
 904
 905         /* FIXME use last free of some kind */
 906
 907         /* we don't want to overwrite the superblock on the drive,
 908          * so we make sure to start at an offset of at least 1MB
 909          */
 910         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 911
 912         max_hole_start = search_start;
 913         max_hole_size = 0;
 914         hole_size = 0;
 915
 916         if (search_start >= search_end) {
 917                 ret = -ENOSPC;
 918                 goto error;
 919         }
 920
 921         path = btrfs_alloc_path();
 922         if (!path) {
 923                 ret = -ENOMEM;
 924                 goto error;
 925         }
 926         path->reada = 2;
 927
 928         key.objectid = device->devid;
 929         key.offset = search_start;
 930         key.type = BTRFS_DEV_EXTENT_KEY;
 931
 932         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 933         if (ret < 0)
 934                 goto out;
 935         if (ret > 0) {
 936                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 937                 if (ret < 0)
 938                         goto out;
 939         }
 940
 941         while (1) {
 942                 l = path->nodes[0];
 943                 slot = path->slots[0];
 944                 if (slot >= btrfs_header_nritems(l)) {
 945                         ret = btrfs_next_leaf(root, path);
 946                         if (ret == 0)
 947                                 continue;
 948                         if (ret < 0)
 949                                 goto out;
 950
 951                         break;
 952                 }
 953                 btrfs_item_key_to_cpu(l, &key, slot);
 954
 955                 if (key.objectid < device->devid)
 956                         goto next;
 957
 958                 if (key.objectid > device->devid)
 959                         break;
 960
 961                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 962                         goto next;
 963
 964                 if (key.offset > search_start) {
 965                         hole_size = key.offset - search_start;
 966
 967                         if (hole_size > max_hole_size) {
 968                                 max_hole_start = search_start;
 969                                 max_hole_size = hole_size;
 970                         }
 971
 972                         /*
 973                          * If this free space is greater than which we need,
 974                          * it must be the max free space that we have found
 975                          * until now, so max_hole_start must point to the start
 976                          * of this free space and the length of this free space
 977                          * is stored in max_hole_size. Thus, we return
 978                          * max_hole_start and max_hole_size and go back to the
 979                          * caller.
 980                          */
 981                         if (hole_size >= num_bytes) {
 982                                 ret = 0;
 983                                 goto out;
 984                         }
 985                 }
 986
 987                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 988                 extent_end = key.offset + btrfs_dev_extent_length(l,
 989                                                                   dev_extent);
 990                 if (extent_end > search_start)
 991                         search_start = extent_end;
 992 next:
 993                 path->slots[0]++;
 994                 cond_resched();
 995         }
 996
 997         /*
 998          * At this point, search_start should be the end of
 999          * allocated dev extents, and when shrinking the device,
1000          * search_end may be smaller than search_start.
1001          */
1002         if (search_end > search_start)
1003                 hole_size = search_end - search_start;
1004
1005         if (hole_size > max_hole_size) {
1006                 max_hole_start = search_start;
1007                 max_hole_size = hole_size;
1008         }
1009
1010         /* See above. */
1011         if (hole_size < num_bytes)
1012                 ret = -ENOSPC;
1013         else
1014                 ret = 0;
1015
1016 out:
1017         btrfs_free_path(path);
1018 error:
1019         *start = max_hole_start;
1020         if (len)
1021                 *len = max_hole_size;
1022         return ret;
1023 }
1024
1025 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1026                           struct btrfs_device *device,
1027                           u64 start)
1028 {
1029         int ret;
1030         struct btrfs_path *path;
1031         struct btrfs_root *root = device->dev_root;
1032         struct btrfs_key key;
1033         struct btrfs_key found_key;
1034         struct extent_buffer *leaf = NULL;
1035         struct btrfs_dev_extent *extent = NULL;
1036
1037         path = btrfs_alloc_path();
1038         if (!path)
1039                 return -ENOMEM;
1040
1041         key.objectid = device->devid;
1042         key.offset = start;
1043         key.type = BTRFS_DEV_EXTENT_KEY;
1044 again:
1045         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1046         if (ret > 0) {
1047                 ret = btrfs_previous_item(root, path, key.objectid,
1048                                           BTRFS_DEV_EXTENT_KEY);
1049                 if (ret)
1050                         goto out;
1051                 leaf = path->nodes[0];
1052                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1053                 extent = btrfs_item_ptr(leaf, path->slots[0],
1054                                         struct btrfs_dev_extent);
1055                 BUG_ON(found_key.offset > start || found_key.offset +
1056                        btrfs_dev_extent_length(leaf, extent) < start);
1057                 key = found_key;
1058                 btrfs_release_path(path);
1059                 goto again;
1060         } else if (ret == 0) {
1061                 leaf = path->nodes[0];
1062                 extent = btrfs_item_ptr(leaf, path->slots[0],
1063                                         struct btrfs_dev_extent);
1064         } else {
1065                 btrfs_error(root->fs_info, ret, "Slot search failed");
1066                 goto out;
1067         }
1068
1069         if (device->bytes_used > 0) {
1070                 u64 len = btrfs_dev_extent_length(leaf, extent);
1071                 device->bytes_used -= len;
1072                 spin_lock(&root->fs_info->free_chunk_lock);
1073                 root->fs_info->free_chunk_space += len;
1074                 spin_unlock(&root->fs_info->free_chunk_lock);
1075         }
1076         ret = btrfs_del_item(trans, root, path);
1077         if (ret) {
1078                 btrfs_error(root->fs_info, ret,
1079                             "Failed to remove dev extent item");
1080         }
1081 out:
1082         btrfs_free_path(path);
1083         return ret;
1084 }
1085
1086 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1087                            struct btrfs_device *device,
1088                            u64 chunk_tree, u64 chunk_objectid,
1089                            u64 chunk_offset, u64 start, u64 num_bytes)
1090 {
1091         int ret;
1092         struct btrfs_path *path;
1093         struct btrfs_root *root = device->dev_root;
1094         struct btrfs_dev_extent *extent;
1095         struct extent_buffer *leaf;
1096         struct btrfs_key key;
1097
1098         WARN_ON(!device->in_fs_metadata);
1099         path = btrfs_alloc_path();
1100         if (!path)
1101                 return -ENOMEM;
1102
1103         key.objectid = device->devid;
1104         key.offset = start;
1105         key.type = BTRFS_DEV_EXTENT_KEY;
1106         ret = btrfs_insert_empty_item(trans, root, path, &key,
1107                                       sizeof(*extent));
1108         if (ret)
1109                 goto out;
1110
1111         leaf = path->nodes[0];
1112         extent = btrfs_item_ptr(leaf, path->slots[0],
1113                                 struct btrfs_dev_extent);
1114         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1115         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1116         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1117
1118         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1119                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1120                     BTRFS_UUID_SIZE);
1121
1122         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1123         btrfs_mark_buffer_dirty(leaf);
1124 out:
1125         btrfs_free_path(path);
1126         return ret;
1127 }
1128
1129 static noinline int find_next_chunk(struct btrfs_root *root,
1130                                     u64 objectid, u64 *offset)
1131 {
1132         struct btrfs_path *path;
1133         int ret;
1134         struct btrfs_key key;
1135         struct btrfs_chunk *chunk;
1136         struct btrfs_key found_key;
1137
1138         path = btrfs_alloc_path();
1139         if (!path)
1140                 return -ENOMEM;
1141
1142         key.objectid = objectid;
1143         key.offset = (u64)-1;
1144         key.type = BTRFS_CHUNK_ITEM_KEY;
1145
1146         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1147         if (ret < 0)
1148                 goto error;
1149
1150         BUG_ON(ret == 0); /* Corruption */
1151
1152         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1153         if (ret) {
1154                 *offset = 0;
1155         } else {
1156                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1157                                       path->slots[0]);
1158                 if (found_key.objectid != objectid)
1159                         *offset = 0;
1160                 else {
1161                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1162                                                struct btrfs_chunk);
1163                         *offset = found_key.offset +
1164                                 btrfs_chunk_length(path->nodes[0], chunk);
1165                 }
1166         }
1167         ret = 0;
1168 error:
1169         btrfs_free_path(path);
1170         return ret;
1171 }
1172
1173 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1174 {
1175         int ret;
1176         struct btrfs_key key;
1177         struct btrfs_key found_key;
1178         struct btrfs_path *path;
1179
1180         root = root->fs_info->chunk_root;
1181
1182         path = btrfs_alloc_path();
1183         if (!path)
1184                 return -ENOMEM;
1185
1186         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1187         key.type = BTRFS_DEV_ITEM_KEY;
1188         key.offset = (u64)-1;
1189
1190         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1191         if (ret < 0)
1192                 goto error;
1193
1194         BUG_ON(ret == 0); /* Corruption */
1195
1196         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1197                                   BTRFS_DEV_ITEM_KEY);
1198         if (ret) {
1199                 *objectid = 1;
1200         } else {
1201                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1202                                       path->slots[0]);
1203                 *objectid = found_key.offset + 1;
1204         }
1205         ret = 0;
1206 error:
1207         btrfs_free_path(path);
1208         return ret;
1209 }
1210
1211 /*
1212  * the device information is stored in the chunk root
1213  * the btrfs_device struct should be fully filled in
1214  */
1215 int btrfs_add_device(struct btrfs_trans_handle *trans,
1216                      struct btrfs_root *root,
1217                      struct btrfs_device *device)
1218 {
1219         int ret;
1220         struct btrfs_path *path;
1221         struct btrfs_dev_item *dev_item;
1222         struct extent_buffer *leaf;
1223         struct btrfs_key key;
1224         unsigned long ptr;
1225
1226         root = root->fs_info->chunk_root;
1227
1228         path = btrfs_alloc_path();
1229         if (!path)
1230                 return -ENOMEM;
1231
1232         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1233         key.type = BTRFS_DEV_ITEM_KEY;
1234         key.offset = device->devid;
1235
1236         ret = btrfs_insert_empty_item(trans, root, path, &key,
1237                                       sizeof(*dev_item));
1238         if (ret)
1239                 goto out;
1240
1241         leaf = path->nodes[0];
1242         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1243
1244         btrfs_set_device_id(leaf, dev_item, device->devid);
1245         btrfs_set_device_generation(leaf, dev_item, 0);
1246         btrfs_set_device_type(leaf, dev_item, device->type);
1247         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1248         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1249         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1250         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1251         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1252         btrfs_set_device_group(leaf, dev_item, 0);
1253         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1254         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1255         btrfs_set_device_start_offset(leaf, dev_item, 0);
1256
1257         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1258         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1259         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1260         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1261         btrfs_mark_buffer_dirty(leaf);
1262
1263         ret = 0;
1264 out:
1265         btrfs_free_path(path);
1266         return ret;
1267 }
1268
1269 static int btrfs_rm_dev_item(struct btrfs_root *root,
1270                              struct btrfs_device *device)
1271 {
1272         int ret;
1273         struct btrfs_path *path;
1274         struct btrfs_key key;
1275         struct btrfs_trans_handle *trans;
1276
1277         root = root->fs_info->chunk_root;
1278
1279         path = btrfs_alloc_path();
1280         if (!path)
1281                 return -ENOMEM;
1282
1283         trans = btrfs_start_transaction(root, 0);
1284         if (IS_ERR(trans)) {
1285                 btrfs_free_path(path);
1286                 return PTR_ERR(trans);
1287         }
1288         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1289         key.type = BTRFS_DEV_ITEM_KEY;
1290         key.offset = device->devid;
1291         lock_chunks(root);
1292
1293         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1294         if (ret < 0)
1295                 goto out;
1296
1297         if (ret > 0) {
1298                 ret = -ENOENT;
1299                 goto out;
1300         }
1301
1302         ret = btrfs_del_item(trans, root, path);
1303         if (ret)
1304                 goto out;
1305 out:
1306         btrfs_free_path(path);
1307         unlock_chunks(root);
1308         btrfs_commit_transaction(trans, root);
1309         return ret;
1310 }
1311
1312 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1313 {
1314         struct btrfs_device *device;
1315         struct btrfs_device *next_device;
1316         struct block_device *bdev;
1317         struct buffer_head *bh = NULL;
1318         struct btrfs_super_block *disk_super;
1319         struct btrfs_fs_devices *cur_devices;
1320         u64 all_avail;
1321         u64 devid;
1322         u64 num_devices;
1323         u8 *dev_uuid;
1324         int ret = 0;
1325         bool clear_super = false;
1326
1327         mutex_lock(&uuid_mutex);
1328
1329         all_avail = root->fs_info->avail_data_alloc_bits |
1330                 root->fs_info->avail_system_alloc_bits |
1331                 root->fs_info->avail_metadata_alloc_bits;
1332
1333         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1334             root->fs_info->fs_devices->num_devices <= 4) {
1335                 printk(KERN_ERR "btrfs: unable to go below four devices "
1336                        "on raid10\n");
1337                 ret = -EINVAL;
1338                 goto out;
1339         }
1340
1341         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1342             root->fs_info->fs_devices->num_devices <= 2) {
1343                 printk(KERN_ERR "btrfs: unable to go below two "
1344                        "devices on raid1\n");
1345                 ret = -EINVAL;
1346                 goto out;
1347         }
1348
1349         if (strcmp(device_path, "missing") == 0) {
1350                 struct list_head *devices;
1351                 struct btrfs_device *tmp;
1352
1353                 device = NULL;
1354                 devices = &root->fs_info->fs_devices->devices;
1355                 /*
1356                  * It is safe to read the devices since the volume_mutex
1357                  * is held.
1358                  */
1359                 list_for_each_entry(tmp, devices, dev_list) {
1360                         if (tmp->in_fs_metadata && !tmp->bdev) {
1361                                 device = tmp;
1362                                 break;
1363                         }
1364                 }
1365                 bdev = NULL;
1366                 bh = NULL;
1367                 disk_super = NULL;
1368                 if (!device) {
1369                         printk(KERN_ERR "btrfs: no missing devices found to "
1370                                "remove\n");
1371                         goto out;
1372                 }
1373         } else {
1374                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1375                                           root->fs_info->bdev_holder);
1376                 if (IS_ERR(bdev)) {
1377                         ret = PTR_ERR(bdev);
1378                         goto out;
1379                 }
1380
1381                 set_blocksize(bdev, 4096);
1382                 invalidate_bdev(bdev);
1383                 bh = btrfs_read_dev_super(bdev);
1384                 if (!bh) {
1385                         ret = -EINVAL;
1386                         goto error_close;
1387                 }
1388                 disk_super = (struct btrfs_super_block *)bh->b_data;
1389                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1390                 dev_uuid = disk_super->dev_item.uuid;
1391                 device = btrfs_find_device(root, devid, dev_uuid,
1392                                            disk_super->fsid);
1393                 if (!device) {
1394                         ret = -ENOENT;
1395                         goto error_brelse;
1396                 }
1397         }
1398
1399         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1400                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1401                        "device\n");
1402                 ret = -EINVAL;
1403                 goto error_brelse;
1404         }
1405
1406         if (device->writeable) {
1407                 lock_chunks(root);
1408                 list_del_init(&device->dev_alloc_list);
1409                 unlock_chunks(root);
1410                 root->fs_info->fs_devices->rw_devices--;
1411                 clear_super = true;
1412         }
1413
1414         ret = btrfs_shrink_device(device, 0);
1415         if (ret)
1416                 goto error_undo;
1417
1418         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1419         if (ret)
1420                 goto error_undo;
1421
1422         spin_lock(&root->fs_info->free_chunk_lock);
1423         root->fs_info->free_chunk_space = device->total_bytes -
1424                 device->bytes_used;
1425         spin_unlock(&root->fs_info->free_chunk_lock);
1426
1427         device->in_fs_metadata = 0;
1428         btrfs_scrub_cancel_dev(root, device);
1429
1430         /*
1431          * the device list mutex makes sure that we don't change
1432          * the device list while someone else is writing out all
1433          * the device supers.
1434          */
1435
1436         cur_devices = device->fs_devices;
1437         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1438         list_del_rcu(&device->dev_list);
1439
1440         device->fs_devices->num_devices--;
1441         device->fs_devices->total_devices--;
1442
1443         if (device->missing)
1444                 root->fs_info->fs_devices->missing_devices--;
1445
1446         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1447                                  struct btrfs_device, dev_list);
1448         if (device->bdev == root->fs_info->sb->s_bdev)
1449                 root->fs_info->sb->s_bdev = next_device->bdev;
1450         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1451                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1452
1453         if (device->bdev)
1454                 device->fs_devices->open_devices--;
1455
1456         call_rcu(&device->rcu, free_device);
1457         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1458
1459         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1460         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1461
1462         if (cur_devices->open_devices == 0) {
1463                 struct btrfs_fs_devices *fs_devices;
1464                 fs_devices = root->fs_info->fs_devices;
1465                 while (fs_devices) {
1466                         if (fs_devices->seed == cur_devices)
1467                                 break;
1468                         fs_devices = fs_devices->seed;
1469                 }
1470                 fs_devices->seed = cur_devices->seed;
1471                 cur_devices->seed = NULL;
1472                 lock_chunks(root);
1473                 __btrfs_close_devices(cur_devices);
1474                 unlock_chunks(root);
1475                 free_fs_devices(cur_devices);
1476         }
1477
1478         /*
1479          * at this point, the device is zero sized.  We want to
1480          * remove it from the devices list and zero out the old super
1481          */
1482         if (clear_super) {
1483                 /* make sure this device isn't detected as part of
1484                  * the FS anymore
1485                  */
1486                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1487                 set_buffer_dirty(bh);
1488                 sync_dirty_buffer(bh);
1489         }
1490
1491         ret = 0;
1492
1493 error_brelse:
1494         brelse(bh);
1495 error_close:
1496         if (bdev)
1497                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1498 out:
1499         mutex_unlock(&uuid_mutex);
1500         return ret;
1501 error_undo:
1502         if (device->writeable) {
1503                 lock_chunks(root);
1504                 list_add(&device->dev_alloc_list,
1505                          &root->fs_info->fs_devices->alloc_list);
1506                 unlock_chunks(root);
1507                 root->fs_info->fs_devices->rw_devices++;
1508         }
1509         goto error_brelse;
1510 }
1511
1512 /*
1513  * does all the dirty work required for changing file system's UUID.
1514  */
1515 static int btrfs_prepare_sprout(struct btrfs_root *root)
1516 {
1517         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1518         struct btrfs_fs_devices *old_devices;
1519         struct btrfs_fs_devices *seed_devices;
1520         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1521         struct btrfs_device *device;
1522         u64 super_flags;
1523
1524         BUG_ON(!mutex_is_locked(&uuid_mutex));
1525         if (!fs_devices->seeding)
1526                 return -EINVAL;
1527
1528         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1529         if (!seed_devices)
1530                 return -ENOMEM;
1531
1532         old_devices = clone_fs_devices(fs_devices);
1533         if (IS_ERR(old_devices)) {
1534                 kfree(seed_devices);
1535                 return PTR_ERR(old_devices);
1536         }
1537
1538         list_add(&old_devices->list, &fs_uuids);
1539
1540         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1541         seed_devices->opened = 1;
1542         INIT_LIST_HEAD(&seed_devices->devices);
1543         INIT_LIST_HEAD(&seed_devices->alloc_list);
1544         mutex_init(&seed_devices->device_list_mutex);
1545
1546         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1547         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1548                               synchronize_rcu);
1549         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1550
1551         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1552         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1553                 device->fs_devices = seed_devices;
1554         }
1555
1556         fs_devices->seeding = 0;
1557         fs_devices->num_devices = 0;
1558         fs_devices->open_devices = 0;
1559         fs_devices->total_devices = 0;
1560         fs_devices->seed = seed_devices;
1561
1562         generate_random_uuid(fs_devices->fsid);
1563         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1564         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1565         super_flags = btrfs_super_flags(disk_super) &
1566                       ~BTRFS_SUPER_FLAG_SEEDING;
1567         btrfs_set_super_flags(disk_super, super_flags);
1568
1569         return 0;
1570 }
1571
1572 /*
1573  * strore the expected generation for seed devices in device items.
1574  */
1575 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1576                                struct btrfs_root *root)
1577 {
1578         struct btrfs_path *path;
1579         struct extent_buffer *leaf;
1580         struct btrfs_dev_item *dev_item;
1581         struct btrfs_device *device;
1582         struct btrfs_key key;
1583         u8 fs_uuid[BTRFS_UUID_SIZE];
1584         u8 dev_uuid[BTRFS_UUID_SIZE];
1585         u64 devid;
1586         int ret;
1587
1588         path = btrfs_alloc_path();
1589         if (!path)
1590                 return -ENOMEM;
1591
1592         root = root->fs_info->chunk_root;
1593         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1594         key.offset = 0;
1595         key.type = BTRFS_DEV_ITEM_KEY;
1596
1597         while (1) {
1598                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1599                 if (ret < 0)
1600                         goto error;
1601
1602                 leaf = path->nodes[0];
1603 next_slot:
1604                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1605                         ret = btrfs_next_leaf(root, path);
1606                         if (ret > 0)
1607                                 break;
1608                         if (ret < 0)
1609                                 goto error;
1610                         leaf = path->nodes[0];
1611                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1612                         btrfs_release_path(path);
1613                         continue;
1614                 }
1615
1616                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1617                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1618                     key.type != BTRFS_DEV_ITEM_KEY)
1619                         break;
1620
1621                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1622                                           struct btrfs_dev_item);
1623                 devid = btrfs_device_id(leaf, dev_item);
1624                 read_extent_buffer(leaf, dev_uuid,
1625                                    (unsigned long)btrfs_device_uuid(dev_item),
1626                                    BTRFS_UUID_SIZE);
1627                 read_extent_buffer(leaf, fs_uuid,
1628                                    (unsigned long)btrfs_device_fsid(dev_item),
1629                                    BTRFS_UUID_SIZE);
1630                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1631                 BUG_ON(!device); /* Logic error */
1632
1633                 if (device->fs_devices->seeding) {
1634                         btrfs_set_device_generation(leaf, dev_item,
1635                                                     device->generation);
1636                         btrfs_mark_buffer_dirty(leaf);
1637                 }
1638
1639                 path->slots[0]++;
1640                 goto next_slot;
1641         }
1642         ret = 0;
1643 error:
1644         btrfs_free_path(path);
1645         return ret;
1646 }
1647
1648 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1649 {
1650         struct request_queue *q;
1651         struct btrfs_trans_handle *trans;
1652         struct btrfs_device *device;
1653         struct block_device *bdev;
1654         struct list_head *devices;
1655         struct super_block *sb = root->fs_info->sb;
1656         struct rcu_string *name;
1657         u64 total_bytes;
1658         int seeding_dev = 0;
1659         int ret = 0;
1660
1661         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1662                 return -EROFS;
1663
1664         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1665                                   root->fs_info->bdev_holder);
1666         if (IS_ERR(bdev))
1667                 return PTR_ERR(bdev);
1668
1669         if (root->fs_info->fs_devices->seeding) {
1670                 seeding_dev = 1;
1671                 down_write(&sb->s_umount);
1672                 mutex_lock(&uuid_mutex);
1673         }
1674
1675         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1676
1677         devices = &root->fs_info->fs_devices->devices;
1678         /*
1679          * we have the volume lock, so we don't need the extra
1680          * device list mutex while reading the list here.
1681          */
1682         list_for_each_entry(device, devices, dev_list) {
1683                 if (device->bdev == bdev) {
1684                         ret = -EEXIST;
1685                         goto error;
1686                 }
1687         }
1688
1689         device = kzalloc(sizeof(*device), GFP_NOFS);
1690         if (!device) {
1691                 /* we can safely leave the fs_devices entry around */
1692                 ret = -ENOMEM;
1693                 goto error;
1694         }
1695
1696         name = rcu_string_strdup(device_path, GFP_NOFS);
1697         if (!name) {
1698                 kfree(device);
1699                 ret = -ENOMEM;
1700                 goto error;
1701         }
1702         rcu_assign_pointer(device->name, name);
1703
1704         ret = find_next_devid(root, &device->devid);
1705         if (ret) {
1706                 rcu_string_free(device->name);
1707                 kfree(device);
1708                 goto error;
1709         }
1710
1711         trans = btrfs_start_transaction(root, 0);
1712         if (IS_ERR(trans)) {
1713                 rcu_string_free(device->name);
1714                 kfree(device);
1715                 ret = PTR_ERR(trans);
1716                 goto error;
1717         }
1718
1719         lock_chunks(root);
1720
1721         q = bdev_get_queue(bdev);
1722         if (blk_queue_discard(q))
1723                 device->can_discard = 1;
1724         device->writeable = 1;
1725         device->work.func = pending_bios_fn;
1726         generate_random_uuid(device->uuid);
1727         spin_lock_init(&device->io_lock);
1728         device->generation = trans->transid;
1729         device->io_width = root->sectorsize;
1730         device->io_align = root->sectorsize;
1731         device->sector_size = root->sectorsize;
1732         device->total_bytes = i_size_read(bdev->bd_inode);
1733         device->disk_total_bytes = device->total_bytes;
1734         device->dev_root = root->fs_info->dev_root;
1735         device->bdev = bdev;
1736         device->in_fs_metadata = 1;
1737         device->mode = FMODE_EXCL;
1738         set_blocksize(device->bdev, 4096);
1739
1740         if (seeding_dev) {
1741                 sb->s_flags &= ~MS_RDONLY;
1742                 ret = btrfs_prepare_sprout(root);
1743                 BUG_ON(ret); /* -ENOMEM */
1744         }
1745
1746         device->fs_devices = root->fs_info->fs_devices;
1747
1748         /*
1749          * we don't want write_supers to jump in here with our device
1750          * half setup
1751          */
1752         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1753         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1754         list_add(&device->dev_alloc_list,
1755                  &root->fs_info->fs_devices->alloc_list);
1756         root->fs_info->fs_devices->num_devices++;
1757         root->fs_info->fs_devices->open_devices++;
1758         root->fs_info->fs_devices->rw_devices++;
1759         root->fs_info->fs_devices->total_devices++;
1760         if (device->can_discard)
1761                 root->fs_info->fs_devices->num_can_discard++;
1762         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1763
1764         spin_lock(&root->fs_info->free_chunk_lock);
1765         root->fs_info->free_chunk_space += device->total_bytes;
1766         spin_unlock(&root->fs_info->free_chunk_lock);
1767
1768         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1769                 root->fs_info->fs_devices->rotating = 1;
1770
1771         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1772         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1773                                     total_bytes + device->total_bytes);
1774
1775         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1776         btrfs_set_super_num_devices(root->fs_info->super_copy,
1777                                     total_bytes + 1);
1778         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1779
1780         if (seeding_dev) {
1781                 ret = init_first_rw_device(trans, root, device);
1782                 if (ret)
1783                         goto error_trans;
1784                 ret = btrfs_finish_sprout(trans, root);
1785                 if (ret)
1786                         goto error_trans;
1787         } else {
1788                 ret = btrfs_add_device(trans, root, device);
1789                 if (ret)
1790                         goto error_trans;
1791         }
1792
1793         /*
1794          * we've got more storage, clear any full flags on the space
1795          * infos
1796          */
1797         btrfs_clear_space_info_full(root->fs_info);
1798
1799         unlock_chunks(root);
1800         ret = btrfs_commit_transaction(trans, root);
1801
1802         if (seeding_dev) {
1803                 mutex_unlock(&uuid_mutex);
1804                 up_write(&sb->s_umount);
1805
1806                 if (ret) /* transaction commit */
1807                         return ret;
1808
1809                 ret = btrfs_relocate_sys_chunks(root);
1810                 if (ret < 0)
1811                         btrfs_error(root->fs_info, ret,
1812                                     "Failed to relocate sys chunks after "
1813                                     "device initialization. This can be fixed "
1814                                     "using the \"btrfs balance\" command.");
1815         }
1816
1817         return ret;
1818
1819 error_trans:
1820         unlock_chunks(root);
1821         btrfs_abort_transaction(trans, root, ret);
1822         btrfs_end_transaction(trans, root);
1823         rcu_string_free(device->name);
1824         kfree(device);
1825 error:
1826         blkdev_put(bdev, FMODE_EXCL);
1827         if (seeding_dev) {
1828                 mutex_unlock(&uuid_mutex);
1829                 up_write(&sb->s_umount);
1830         }
1831         return ret;
1832 }
1833
1834 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1835                                         struct btrfs_device *device)
1836 {
1837         int ret;
1838         struct btrfs_path *path;
1839         struct btrfs_root *root;
1840         struct btrfs_dev_item *dev_item;
1841         struct extent_buffer *leaf;
1842         struct btrfs_key key;
1843
1844         root = device->dev_root->fs_info->chunk_root;
1845
1846         path = btrfs_alloc_path();
1847         if (!path)
1848                 return -ENOMEM;
1849
1850         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1851         key.type = BTRFS_DEV_ITEM_KEY;
1852         key.offset = device->devid;
1853
1854         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1855         if (ret < 0)
1856                 goto out;
1857
1858         if (ret > 0) {
1859                 ret = -ENOENT;
1860                 goto out;
1861         }
1862
1863         leaf = path->nodes[0];
1864         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1865
1866         btrfs_set_device_id(leaf, dev_item, device->devid);
1867         btrfs_set_device_type(leaf, dev_item, device->type);
1868         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1869         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1870         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1871         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1872         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1873         btrfs_mark_buffer_dirty(leaf);
1874
1875 out:
1876         btrfs_free_path(path);
1877         return ret;
1878 }
1879
1880 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1881                       struct btrfs_device *device, u64 new_size)
1882 {
1883         struct btrfs_super_block *super_copy =
1884                 device->dev_root->fs_info->super_copy;
1885         u64 old_total = btrfs_super_total_bytes(super_copy);
1886         u64 diff = new_size - device->total_bytes;
1887
1888         if (!device->writeable)
1889                 return -EACCES;
1890         if (new_size <= device->total_bytes)
1891                 return -EINVAL;
1892
1893         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1894         device->fs_devices->total_rw_bytes += diff;
1895
1896         device->total_bytes = new_size;
1897         device->disk_total_bytes = new_size;
1898         btrfs_clear_space_info_full(device->dev_root->fs_info);
1899
1900         return btrfs_update_device(trans, device);
1901 }
1902
1903 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1904                       struct btrfs_device *device, u64 new_size)
1905 {
1906         int ret;
1907         lock_chunks(device->dev_root);
1908         ret = __btrfs_grow_device(trans, device, new_size);
1909         unlock_chunks(device->dev_root);
1910         return ret;
1911 }
1912
1913 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1914                             struct btrfs_root *root,
1915                             u64 chunk_tree, u64 chunk_objectid,
1916                             u64 chunk_offset)
1917 {
1918         int ret;
1919         struct btrfs_path *path;
1920         struct btrfs_key key;
1921
1922         root = root->fs_info->chunk_root;
1923         path = btrfs_alloc_path();
1924         if (!path)
1925                 return -ENOMEM;
1926
1927         key.objectid = chunk_objectid;
1928         key.offset = chunk_offset;
1929         key.type = BTRFS_CHUNK_ITEM_KEY;
1930
1931         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1932         if (ret < 0)
1933                 goto out;
1934         else if (ret > 0) { /* Logic error or corruption */
1935                 btrfs_error(root->fs_info, -ENOENT,
1936                             "Failed lookup while freeing chunk.");
1937                 ret = -ENOENT;
1938                 goto out;
1939         }
1940
1941         ret = btrfs_del_item(trans, root, path);
1942         if (ret < 0)
1943                 btrfs_error(root->fs_info, ret,
1944                             "Failed to delete chunk item.");
1945 out:
1946         btrfs_free_path(path);
1947         return ret;
1948 }
1949
1950 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1951                         chunk_offset)
1952 {
1953         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1954         struct btrfs_disk_key *disk_key;
1955         struct btrfs_chunk *chunk;
1956         u8 *ptr;
1957         int ret = 0;
1958         u32 num_stripes;
1959         u32 array_size;
1960         u32 len = 0;
1961         u32 cur;
1962         struct btrfs_key key;
1963
1964         array_size = btrfs_super_sys_array_size(super_copy);
1965
1966         ptr = super_copy->sys_chunk_array;
1967         cur = 0;
1968
1969         while (cur < array_size) {
1970                 disk_key = (struct btrfs_disk_key *)ptr;
1971                 btrfs_disk_key_to_cpu(&key, disk_key);
1972
1973                 len = sizeof(*disk_key);
1974
1975                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1976                         chunk = (struct btrfs_chunk *)(ptr + len);
1977                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1978                         len += btrfs_chunk_item_size(num_stripes);
1979                 } else {
1980                         ret = -EIO;
1981                         break;
1982                 }
1983                 if (key.objectid == chunk_objectid &&
1984                     key.offset == chunk_offset) {
1985                         memmove(ptr, ptr + len, array_size - (cur + len));
1986                         array_size -= len;
1987                         btrfs_set_super_sys_array_size(super_copy, array_size);
1988                 } else {
1989                         ptr += len;
1990                         cur += len;
1991                 }
1992         }
1993         return ret;
1994 }
1995
1996 static int btrfs_relocate_chunk(struct btrfs_root *root,
1997                          u64 chunk_tree, u64 chunk_objectid,
1998                          u64 chunk_offset)
1999 {
2000         struct extent_map_tree *em_tree;
2001         struct btrfs_root *extent_root;
2002         struct btrfs_trans_handle *trans;
2003         struct extent_map *em;
2004         struct map_lookup *map;
2005         int ret;
2006         int i;
2007
2008         root = root->fs_info->chunk_root;
2009         extent_root = root->fs_info->extent_root;
2010         em_tree = &root->fs_info->mapping_tree.map_tree;
2011
2012         ret = btrfs_can_relocate(extent_root, chunk_offset);
2013         if (ret)
2014                 return -ENOSPC;
2015
2016         /* step one, relocate all the extents inside this chunk */
2017         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2018         if (ret)
2019                 return ret;
2020
2021         trans = btrfs_start_transaction(root, 0);
2022         BUG_ON(IS_ERR(trans));
2023
2024         lock_chunks(root);
2025
2026         /*
2027          * step two, delete the device extents and the
2028          * chunk tree entries
2029          */
2030         read_lock(&em_tree->lock);
2031         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2032         read_unlock(&em_tree->lock);
2033
2034         BUG_ON(!em || em->start > chunk_offset ||
2035                em->start + em->len < chunk_offset);
2036         map = (struct map_lookup *)em->bdev;
2037
2038         for (i = 0; i < map->num_stripes; i++) {
2039                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2040                                             map->stripes[i].physical);
2041                 BUG_ON(ret);
2042
2043                 if (map->stripes[i].dev) {
2044                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2045                         BUG_ON(ret);
2046                 }
2047         }
2048         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2049                                chunk_offset);
2050
2051         BUG_ON(ret);
2052
2053         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2054
2055         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2056                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2057                 BUG_ON(ret);
2058         }
2059
2060         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2061         BUG_ON(ret);
2062
2063         write_lock(&em_tree->lock);
2064         remove_extent_mapping(em_tree, em);
2065         write_unlock(&em_tree->lock);
2066
2067         kfree(map);
2068         em->bdev = NULL;
2069
2070         /* once for the tree */
2071         free_extent_map(em);
2072         /* once for us */
2073         free_extent_map(em);
2074
2075         unlock_chunks(root);
2076         btrfs_end_transaction(trans, root);
2077         return 0;
2078 }
2079
2080 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2081 {
2082         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2083         struct btrfs_path *path;
2084         struct extent_buffer *leaf;
2085         struct btrfs_chunk *chunk;
2086         struct btrfs_key key;
2087         struct btrfs_key found_key;
2088         u64 chunk_tree = chunk_root->root_key.objectid;
2089         u64 chunk_type;
2090         bool retried = false;
2091         int failed = 0;
2092         int ret;
2093
2094         path = btrfs_alloc_path();
2095         if (!path)
2096                 return -ENOMEM;
2097
2098 again:
2099         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2100         key.offset = (u64)-1;
2101         key.type = BTRFS_CHUNK_ITEM_KEY;
2102
2103         while (1) {
2104                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2105                 if (ret < 0)
2106                         goto error;
2107                 BUG_ON(ret == 0); /* Corruption */
2108
2109                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2110                                           key.type);
2111                 if (ret < 0)
2112                         goto error;
2113                 if (ret > 0)
2114                         break;
2115
2116                 leaf = path->nodes[0];
2117                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2118
2119                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2120                                        struct btrfs_chunk);
2121                 chunk_type = btrfs_chunk_type(leaf, chunk);
2122                 btrfs_release_path(path);
2123
2124                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2125                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2126                                                    found_key.objectid,
2127                                                    found_key.offset);
2128                         if (ret == -ENOSPC)
2129                                 failed++;
2130                         else if (ret)
2131                                 BUG();
2132                 }
2133
2134                 if (found_key.offset == 0)
2135                         break;
2136                 key.offset = found_key.offset - 1;
2137         }
2138         ret = 0;
2139         if (failed && !retried) {
2140                 failed = 0;
2141                 retried = true;
2142                 goto again;
2143         } else if (failed && retried) {
2144                 WARN_ON(1);
2145                 ret = -ENOSPC;
2146         }
2147 error:
2148         btrfs_free_path(path);
2149         return ret;
2150 }
2151
2152 static int insert_balance_item(struct btrfs_root *root,
2153                                struct btrfs_balance_control *bctl)
2154 {
2155         struct btrfs_trans_handle *trans;
2156         struct btrfs_balance_item *item;
2157         struct btrfs_disk_balance_args disk_bargs;
2158         struct btrfs_path *path;
2159         struct extent_buffer *leaf;
2160         struct btrfs_key key;
2161         int ret, err;
2162
2163         path = btrfs_alloc_path();
2164         if (!path)
2165                 return -ENOMEM;
2166
2167         trans = btrfs_start_transaction(root, 0);
2168         if (IS_ERR(trans)) {
2169                 btrfs_free_path(path);
2170                 return PTR_ERR(trans);
2171         }
2172
2173         key.objectid = BTRFS_BALANCE_OBJECTID;
2174         key.type = BTRFS_BALANCE_ITEM_KEY;
2175         key.offset = 0;
2176
2177         ret = btrfs_insert_empty_item(trans, root, path, &key,
2178                                       sizeof(*item));
2179         if (ret)
2180                 goto out;
2181
2182         leaf = path->nodes[0];
2183         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2184
2185         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2186
2187         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2188         btrfs_set_balance_data(leaf, item, &disk_bargs);
2189         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2190         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2191         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2192         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2193
2194         btrfs_set_balance_flags(leaf, item, bctl->flags);
2195
2196         btrfs_mark_buffer_dirty(leaf);
2197 out:
2198         btrfs_free_path(path);
2199         err = btrfs_commit_transaction(trans, root);
2200         if (err && !ret)
2201                 ret = err;
2202         return ret;
2203 }
2204
2205 static int del_balance_item(struct btrfs_root *root)
2206 {
2207         struct btrfs_trans_handle *trans;
2208         struct btrfs_path *path;
2209         struct btrfs_key key;
2210         int ret, err;
2211
2212         path = btrfs_alloc_path();
2213         if (!path)
2214                 return -ENOMEM;
2215
2216         trans = btrfs_start_transaction(root, 0);
2217         if (IS_ERR(trans)) {
2218                 btrfs_free_path(path);
2219                 return PTR_ERR(trans);
2220         }
2221
2222         key.objectid = BTRFS_BALANCE_OBJECTID;
2223         key.type = BTRFS_BALANCE_ITEM_KEY;
2224         key.offset = 0;
2225
2226         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2227         if (ret < 0)
2228                 goto out;
2229         if (ret > 0) {
2230                 ret = -ENOENT;
2231                 goto out;
2232         }
2233
2234         ret = btrfs_del_item(trans, root, path);
2235 out:
2236         btrfs_free_path(path);
2237         err = btrfs_commit_transaction(trans, root);
2238         if (err && !ret)
2239                 ret = err;
2240         return ret;
2241 }
2242
2243 /*
2244  * This is a heuristic used to reduce the number of chunks balanced on
2245  * resume after balance was interrupted.
2246  */
2247 static void update_balance_args(struct btrfs_balance_control *bctl)
2248 {
2249         /*
2250          * Turn on soft mode for chunk types that were being converted.
2251          */
2252         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2253                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2254         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2255                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2256         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2257                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2258
2259         /*
2260          * Turn on usage filter if is not already used.  The idea is
2261          * that chunks that we have already balanced should be
2262          * reasonably full.  Don't do it for chunks that are being
2263          * converted - that will keep us from relocating unconverted
2264          * (albeit full) chunks.
2265          */
2266         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2267             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2268                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2269                 bctl->data.usage = 90;
2270         }
2271         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2272             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2273                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2274                 bctl->sys.usage = 90;
2275         }
2276         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2277             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2278                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2279                 bctl->meta.usage = 90;
2280         }
2281 }
2282
2283 /*
2284  * Should be called with both balance and volume mutexes held to
2285  * serialize other volume operations (add_dev/rm_dev/resize) with
2286  * restriper.  Same goes for unset_balance_control.
2287  */
2288 static void set_balance_control(struct btrfs_balance_control *bctl)
2289 {
2290         struct btrfs_fs_info *fs_info = bctl->fs_info;
2291
2292         BUG_ON(fs_info->balance_ctl);
2293
2294         spin_lock(&fs_info->balance_lock);
2295         fs_info->balance_ctl = bctl;
2296         spin_unlock(&fs_info->balance_lock);
2297 }
2298
2299 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2300 {
2301         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2302
2303         BUG_ON(!fs_info->balance_ctl);
2304
2305         spin_lock(&fs_info->balance_lock);
2306         fs_info->balance_ctl = NULL;
2307         spin_unlock(&fs_info->balance_lock);
2308
2309         kfree(bctl);
2310 }
2311
2312 /*
2313  * Balance filters.  Return 1 if chunk should be filtered out
2314  * (should not be balanced).
2315  */
2316 static int chunk_profiles_filter(u64 chunk_type,
2317                                  struct btrfs_balance_args *bargs)
2318 {
2319         chunk_type = chunk_to_extended(chunk_type) &
2320                                 BTRFS_EXTENDED_PROFILE_MASK;
2321
2322         if (bargs->profiles & chunk_type)
2323                 return 0;
2324
2325         return 1;
2326 }
2327
2328 static u64 div_factor_fine(u64 num, int factor)
2329 {
2330         if (factor <= 0)
2331                 return 0;
2332         if (factor >= 100)
2333                 return num;
2334
2335         num *= factor;
2336         do_div(num, 100);
2337         return num;
2338 }
2339
2340 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2341                               struct btrfs_balance_args *bargs)
2342 {
2343         struct btrfs_block_group_cache *cache;
2344         u64 chunk_used, user_thresh;
2345         int ret = 1;
2346
2347         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2348         chunk_used = btrfs_block_group_used(&cache->item);
2349
2350         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2351         if (chunk_used < user_thresh)
2352                 ret = 0;
2353
2354         btrfs_put_block_group(cache);
2355         return ret;
2356 }
2357
2358 static int chunk_devid_filter(struct extent_buffer *leaf,
2359                               struct btrfs_chunk *chunk,
2360                               struct btrfs_balance_args *bargs)
2361 {
2362         struct btrfs_stripe *stripe;
2363         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2364         int i;
2365
2366         for (i = 0; i < num_stripes; i++) {
2367                 stripe = btrfs_stripe_nr(chunk, i);
2368                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2369                         return 0;
2370         }
2371
2372         return 1;
2373 }
2374
2375 /* [pstart, pend) */
2376 static int chunk_drange_filter(struct extent_buffer *leaf,
2377                                struct btrfs_chunk *chunk,
2378                                u64 chunk_offset,
2379                                struct btrfs_balance_args *bargs)
2380 {
2381         struct btrfs_stripe *stripe;
2382         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2383         u64 stripe_offset;
2384         u64 stripe_length;
2385         int factor;
2386         int i;
2387
2388         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2389                 return 0;
2390
2391         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2392              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2393                 factor = 2;
2394         else
2395                 factor = 1;
2396         factor = num_stripes / factor;
2397
2398         for (i = 0; i < num_stripes; i++) {
2399                 stripe = btrfs_stripe_nr(chunk, i);
2400                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2401                         continue;
2402
2403                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2404                 stripe_length = btrfs_chunk_length(leaf, chunk);
2405                 do_div(stripe_length, factor);
2406
2407                 if (stripe_offset < bargs->pend &&
2408                     stripe_offset + stripe_length > bargs->pstart)
2409                         return 0;
2410         }
2411
2412         return 1;
2413 }
2414
2415 /* [vstart, vend) */
2416 static int chunk_vrange_filter(struct extent_buffer *leaf,
2417                                struct btrfs_chunk *chunk,
2418                                u64 chunk_offset,
2419                                struct btrfs_balance_args *bargs)
2420 {
2421         if (chunk_offset < bargs->vend &&
2422             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2423                 /* at least part of the chunk is inside this vrange */
2424                 return 0;
2425
2426         return 1;
2427 }
2428
2429 static int chunk_soft_convert_filter(u64 chunk_type,
2430                                      struct btrfs_balance_args *bargs)
2431 {
2432         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2433                 return 0;
2434
2435         chunk_type = chunk_to_extended(chunk_type) &
2436                                 BTRFS_EXTENDED_PROFILE_MASK;
2437
2438         if (bargs->target == chunk_type)
2439                 return 1;
2440
2441         return 0;
2442 }
2443
2444 static int should_balance_chunk(struct btrfs_root *root,
2445                                 struct extent_buffer *leaf,
2446                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2447 {
2448         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2449         struct btrfs_balance_args *bargs = NULL;
2450         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2451
2452         /* type filter */
2453         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2454               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2455                 return 0;
2456         }
2457
2458         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2459                 bargs = &bctl->data;
2460         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2461                 bargs = &bctl->sys;
2462         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2463                 bargs = &bctl->meta;
2464
2465         /* profiles filter */
2466         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2467             chunk_profiles_filter(chunk_type, bargs)) {
2468                 return 0;
2469         }
2470
2471         /* usage filter */
2472         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2473             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2474                 return 0;
2475         }
2476
2477         /* devid filter */
2478         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2479             chunk_devid_filter(leaf, chunk, bargs)) {
2480                 return 0;
2481         }
2482
2483         /* drange filter, makes sense only with devid filter */
2484         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2485             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2486                 return 0;
2487         }
2488
2489         /* vrange filter */
2490         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2491             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2492                 return 0;
2493         }
2494
2495         /* soft profile changing mode */
2496         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2497             chunk_soft_convert_filter(chunk_type, bargs)) {
2498                 return 0;
2499         }
2500
2501         return 1;
2502 }
2503
2504 static u64 div_factor(u64 num, int factor)
2505 {
2506         if (factor == 10)
2507                 return num;
2508         num *= factor;
2509         do_div(num, 10);
2510         return num;
2511 }
2512
2513 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2514 {
2515         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2516         struct btrfs_root *chunk_root = fs_info->chunk_root;
2517         struct btrfs_root *dev_root = fs_info->dev_root;
2518         struct list_head *devices;
2519         struct btrfs_device *device;
2520         u64 old_size;
2521         u64 size_to_free;
2522         struct btrfs_chunk *chunk;
2523         struct btrfs_path *path;
2524         struct btrfs_key key;
2525         struct btrfs_key found_key;
2526         struct btrfs_trans_handle *trans;
2527         struct extent_buffer *leaf;
2528         int slot;
2529         int ret;
2530         int enospc_errors = 0;
2531         bool counting = true;
2532
2533         /* step one make some room on all the devices */
2534         devices = &fs_info->fs_devices->devices;
2535         list_for_each_entry(device, devices, dev_list) {
2536                 old_size = device->total_bytes;
2537                 size_to_free = div_factor(old_size, 1);
2538                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2539                 if (!device->writeable ||
2540                     device->total_bytes - device->bytes_used > size_to_free)
2541                         continue;
2542
2543                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2544                 if (ret == -ENOSPC)
2545                         break;
2546                 BUG_ON(ret);
2547
2548                 trans = btrfs_start_transaction(dev_root, 0);
2549                 BUG_ON(IS_ERR(trans));
2550
2551                 ret = btrfs_grow_device(trans, device, old_size);
2552                 BUG_ON(ret);
2553
2554                 btrfs_end_transaction(trans, dev_root);
2555         }
2556
2557         /* step two, relocate all the chunks */
2558         path = btrfs_alloc_path();
2559         if (!path) {
2560                 ret = -ENOMEM;
2561                 goto error;
2562         }
2563
2564         /* zero out stat counters */
2565         spin_lock(&fs_info->balance_lock);
2566         memset(&bctl->stat, 0, sizeof(bctl->stat));
2567         spin_unlock(&fs_info->balance_lock);
2568 again:
2569         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2570         key.offset = (u64)-1;
2571         key.type = BTRFS_CHUNK_ITEM_KEY;
2572
2573         while (1) {
2574                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2575                     atomic_read(&fs_info->balance_cancel_req)) {
2576                         ret = -ECANCELED;
2577                         goto error;
2578                 }
2579
2580                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2581                 if (ret < 0)
2582                         goto error;
2583
2584                 /*
2585                  * this shouldn't happen, it means the last relocate
2586                  * failed
2587                  */
2588                 if (ret == 0)
2589                         BUG(); /* FIXME break ? */
2590
2591                 ret = btrfs_previous_item(chunk_root, path, 0,
2592                                           BTRFS_CHUNK_ITEM_KEY);
2593                 if (ret) {
2594                         ret = 0;
2595                         break;
2596                 }
2597
2598                 leaf = path->nodes[0];
2599                 slot = path->slots[0];
2600                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2601
2602                 if (found_key.objectid != key.objectid)
2603                         break;
2604
2605                 /* chunk zero is special */
2606                 if (found_key.offset == 0)
2607                         break;
2608
2609                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2610
2611                 if (!counting) {
2612                         spin_lock(&fs_info->balance_lock);
2613                         bctl->stat.considered++;
2614                         spin_unlock(&fs_info->balance_lock);
2615                 }
2616
2617                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2618                                            found_key.offset);
2619                 btrfs_release_path(path);
2620                 if (!ret)
2621                         goto loop;
2622
2623                 if (counting) {
2624                         spin_lock(&fs_info->balance_lock);
2625                         bctl->stat.expected++;
2626                         spin_unlock(&fs_info->balance_lock);
2627                         goto loop;
2628                 }
2629
2630                 ret = btrfs_relocate_chunk(chunk_root,
2631                                            chunk_root->root_key.objectid,
2632                                            found_key.objectid,
2633                                            found_key.offset);
2634                 if (ret && ret != -ENOSPC)
2635                         goto error;
2636                 if (ret == -ENOSPC) {
2637                         enospc_errors++;
2638                 } else {
2639                         spin_lock(&fs_info->balance_lock);
2640                         bctl->stat.completed++;
2641                         spin_unlock(&fs_info->balance_lock);
2642                 }
2643 loop:
2644                 key.offset = found_key.offset - 1;
2645         }
2646
2647         if (counting) {
2648                 btrfs_release_path(path);
2649                 counting = false;
2650                 goto again;
2651         }
2652 error:
2653         btrfs_free_path(path);
2654         if (enospc_errors) {
2655                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2656                        enospc_errors);
2657                 if (!ret)
2658                         ret = -ENOSPC;
2659         }
2660
2661         return ret;
2662 }
2663
2664 /**
2665  * alloc_profile_is_valid - see if a given profile is valid and reduced
2666  * @flags: profile to validate
2667  * @extended: if true @flags is treated as an extended profile
2668  */
2669 static int alloc_profile_is_valid(u64 flags, int extended)
2670 {
2671         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2672                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2673
2674         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2675
2676         /* 1) check that all other bits are zeroed */
2677         if (flags & ~mask)
2678                 return 0;
2679
2680         /* 2) see if profile is reduced */
2681         if (flags == 0)
2682                 return !extended; /* "0" is valid for usual profiles */
2683
2684         /* true if exactly one bit set */
2685         return (flags & (flags - 1)) == 0;
2686 }
2687
2688 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2689 {
2690         /* cancel requested || normal exit path */
2691         return atomic_read(&fs_info->balance_cancel_req) ||
2692                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2693                  atomic_read(&fs_info->balance_cancel_req) == 0);
2694 }
2695
2696 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2697 {
2698         int ret;
2699
2700         unset_balance_control(fs_info);
2701         ret = del_balance_item(fs_info->tree_root);
2702         BUG_ON(ret);
2703 }
2704
2705 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2706                                struct btrfs_ioctl_balance_args *bargs);
2707
2708 /*
2709  * Should be called with both balance and volume mutexes held
2710  */
2711 int btrfs_balance(struct btrfs_balance_control *bctl,
2712                   struct btrfs_ioctl_balance_args *bargs)
2713 {
2714         struct btrfs_fs_info *fs_info = bctl->fs_info;
2715         u64 allowed;
2716         int mixed = 0;
2717         int ret;
2718
2719         if (btrfs_fs_closing(fs_info) ||
2720             atomic_read(&fs_info->balance_pause_req) ||
2721             atomic_read(&fs_info->balance_cancel_req)) {
2722                 ret = -EINVAL;
2723                 goto out;
2724         }
2725
2726         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2727         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2728                 mixed = 1;
2729
2730         /*
2731          * In case of mixed groups both data and meta should be picked,
2732          * and identical options should be given for both of them.
2733          */
2734         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2735         if (mixed && (bctl->flags & allowed)) {
2736                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2737                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2738                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2739                         printk(KERN_ERR "btrfs: with mixed groups data and "
2740                                "metadata balance options must be the same\n");
2741                         ret = -EINVAL;
2742                         goto out;
2743                 }
2744         }
2745
2746         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2747         if (fs_info->fs_devices->num_devices == 1)
2748                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2749         else if (fs_info->fs_devices->num_devices < 4)
2750                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2751         else
2752                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2753                                 BTRFS_BLOCK_GROUP_RAID10);
2754
2755         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2756             (!alloc_profile_is_valid(bctl->data.target, 1) ||
2757              (bctl->data.target & ~allowed))) {
2758                 printk(KERN_ERR "btrfs: unable to start balance with target "
2759                        "data profile %llu\n",
2760                        (unsigned long long)bctl->data.target);
2761                 ret = -EINVAL;
2762                 goto out;
2763         }
2764         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2765             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2766              (bctl->meta.target & ~allowed))) {
2767                 printk(KERN_ERR "btrfs: unable to start balance with target "
2768                        "metadata profile %llu\n",
2769                        (unsigned long long)bctl->meta.target);
2770                 ret = -EINVAL;
2771                 goto out;
2772         }
2773         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2774             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2775              (bctl->sys.target & ~allowed))) {
2776                 printk(KERN_ERR "btrfs: unable to start balance with target "
2777                        "system profile %llu\n",
2778                        (unsigned long long)bctl->sys.target);
2779                 ret = -EINVAL;
2780                 goto out;
2781         }
2782
2783         /* allow dup'ed data chunks only in mixed mode */
2784         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2785             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2786                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2787                 ret = -EINVAL;
2788                 goto out;
2789         }
2790
2791         /* allow to reduce meta or sys integrity only if force set */
2792         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2793                         BTRFS_BLOCK_GROUP_RAID10;
2794         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2795              (fs_info->avail_system_alloc_bits & allowed) &&
2796              !(bctl->sys.target & allowed)) ||
2797             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2798              (fs_info->avail_metadata_alloc_bits & allowed) &&
2799              !(bctl->meta.target & allowed))) {
2800                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2801                         printk(KERN_INFO "btrfs: force reducing metadata "
2802                                "integrity\n");
2803                 } else {
2804                         printk(KERN_ERR "btrfs: balance will reduce metadata "
2805                                "integrity, use force if you want this\n");
2806                         ret = -EINVAL;
2807                         goto out;
2808                 }
2809         }
2810
2811         ret = insert_balance_item(fs_info->tree_root, bctl);
2812         if (ret && ret != -EEXIST)
2813                 goto out;
2814
2815         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2816                 BUG_ON(ret == -EEXIST);
2817                 set_balance_control(bctl);
2818         } else {
2819                 BUG_ON(ret != -EEXIST);
2820                 spin_lock(&fs_info->balance_lock);
2821                 update_balance_args(bctl);
2822                 spin_unlock(&fs_info->balance_lock);
2823         }
2824
2825         atomic_inc(&fs_info->balance_running);
2826         mutex_unlock(&fs_info->balance_mutex);
2827
2828         ret = __btrfs_balance(fs_info);
2829
2830         mutex_lock(&fs_info->balance_mutex);
2831         atomic_dec(&fs_info->balance_running);
2832
2833         if (bargs) {
2834                 memset(bargs, 0, sizeof(*bargs));
2835                 update_ioctl_balance_args(fs_info, 0, bargs);
2836         }
2837
2838         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2839             balance_need_close(fs_info)) {
2840                 __cancel_balance(fs_info);
2841         }
2842
2843         wake_up(&fs_info->balance_wait_q);
2844
2845         return ret;
2846 out:
2847         if (bctl->flags & BTRFS_BALANCE_RESUME)
2848                 __cancel_balance(fs_info);
2849         else
2850                 kfree(bctl);
2851         return ret;
2852 }
2853
2854 static int balance_kthread(void *data)
2855 {
2856         struct btrfs_fs_info *fs_info = data;
2857         int ret = 0;
2858
2859         mutex_lock(&fs_info->volume_mutex);
2860         mutex_lock(&fs_info->balance_mutex);
2861
2862         if (fs_info->balance_ctl) {
2863                 printk(KERN_INFO "btrfs: continuing balance\n");
2864                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
2865         }
2866
2867         mutex_unlock(&fs_info->balance_mutex);
2868         mutex_unlock(&fs_info->volume_mutex);
2869
2870         return ret;
2871 }
2872
2873 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
2874 {
2875         struct task_struct *tsk;
2876
2877         spin_lock(&fs_info->balance_lock);
2878         if (!fs_info->balance_ctl) {
2879                 spin_unlock(&fs_info->balance_lock);
2880                 return 0;
2881         }
2882         spin_unlock(&fs_info->balance_lock);
2883
2884         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2885                 printk(KERN_INFO "btrfs: force skipping balance\n");
2886                 return 0;
2887         }
2888
2889         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
2890         if (IS_ERR(tsk))
2891                 return PTR_ERR(tsk);
2892
2893         return 0;
2894 }
2895
2896 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
2897 {
2898         struct btrfs_balance_control *bctl;
2899         struct btrfs_balance_item *item;
2900         struct btrfs_disk_balance_args disk_bargs;
2901         struct btrfs_path *path;
2902         struct extent_buffer *leaf;
2903         struct btrfs_key key;
2904         int ret;
2905
2906         path = btrfs_alloc_path();
2907         if (!path)
2908                 return -ENOMEM;
2909
2910         key.objectid = BTRFS_BALANCE_OBJECTID;
2911         key.type = BTRFS_BALANCE_ITEM_KEY;
2912         key.offset = 0;
2913
2914         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2915         if (ret < 0)
2916                 goto out;
2917         if (ret > 0) { /* ret = -ENOENT; */
2918                 ret = 0;
2919                 goto out;
2920         }
2921
2922         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2923         if (!bctl) {
2924                 ret = -ENOMEM;
2925                 goto out;
2926         }
2927
2928         leaf = path->nodes[0];
2929         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2930
2931         bctl->fs_info = fs_info;
2932         bctl->flags = btrfs_balance_flags(leaf, item);
2933         bctl->flags |= BTRFS_BALANCE_RESUME;
2934
2935         btrfs_balance_data(leaf, item, &disk_bargs);
2936         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2937         btrfs_balance_meta(leaf, item, &disk_bargs);
2938         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2939         btrfs_balance_sys(leaf, item, &disk_bargs);
2940         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2941
2942         mutex_lock(&fs_info->volume_mutex);
2943         mutex_lock(&fs_info->balance_mutex);
2944
2945         set_balance_control(bctl);
2946
2947         mutex_unlock(&fs_info->balance_mutex);
2948         mutex_unlock(&fs_info->volume_mutex);
2949 out:
2950         btrfs_free_path(path);
2951         return ret;
2952 }
2953
2954 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2955 {
2956         int ret = 0;
2957
2958         mutex_lock(&fs_info->balance_mutex);
2959         if (!fs_info->balance_ctl) {
2960                 mutex_unlock(&fs_info->balance_mutex);
2961                 return -ENOTCONN;
2962         }
2963
2964         if (atomic_read(&fs_info->balance_running)) {
2965                 atomic_inc(&fs_info->balance_pause_req);
2966                 mutex_unlock(&fs_info->balance_mutex);
2967
2968                 wait_event(fs_info->balance_wait_q,
2969                            atomic_read(&fs_info->balance_running) == 0);
2970
2971                 mutex_lock(&fs_info->balance_mutex);
2972                 /* we are good with balance_ctl ripped off from under us */
2973                 BUG_ON(atomic_read(&fs_info->balance_running));
2974                 atomic_dec(&fs_info->balance_pause_req);
2975         } else {
2976                 ret = -ENOTCONN;
2977         }
2978
2979         mutex_unlock(&fs_info->balance_mutex);
2980         return ret;
2981 }
2982
2983 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
2984 {
2985         mutex_lock(&fs_info->balance_mutex);
2986         if (!fs_info->balance_ctl) {
2987                 mutex_unlock(&fs_info->balance_mutex);
2988                 return -ENOTCONN;
2989         }
2990
2991         atomic_inc(&fs_info->balance_cancel_req);
2992         /*
2993          * if we are running just wait and return, balance item is
2994          * deleted in btrfs_balance in this case
2995          */
2996         if (atomic_read(&fs_info->balance_running)) {
2997                 mutex_unlock(&fs_info->balance_mutex);
2998                 wait_event(fs_info->balance_wait_q,
2999                            atomic_read(&fs_info->balance_running) == 0);
3000                 mutex_lock(&fs_info->balance_mutex);
3001         } else {
3002                 /* __cancel_balance needs volume_mutex */
3003                 mutex_unlock(&fs_info->balance_mutex);
3004                 mutex_lock(&fs_info->volume_mutex);
3005                 mutex_lock(&fs_info->balance_mutex);
3006
3007                 if (fs_info->balance_ctl)
3008                         __cancel_balance(fs_info);
3009
3010                 mutex_unlock(&fs_info->volume_mutex);
3011         }
3012
3013         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3014         atomic_dec(&fs_info->balance_cancel_req);
3015         mutex_unlock(&fs_info->balance_mutex);
3016         return 0;
3017 }
3018
3019 /*
3020  * shrinking a device means finding all of the device extents past
3021  * the new size, and then following the back refs to the chunks.
3022  * The chunk relocation code actually frees the device extent
3023  */
3024 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3025 {
3026         struct btrfs_trans_handle *trans;
3027         struct btrfs_root *root = device->dev_root;
3028         struct btrfs_dev_extent *dev_extent = NULL;
3029         struct btrfs_path *path;
3030         u64 length;
3031         u64 chunk_tree;
3032         u64 chunk_objectid;
3033         u64 chunk_offset;
3034         int ret;
3035         int slot;
3036         int failed = 0;
3037         bool retried = false;
3038         struct extent_buffer *l;
3039         struct btrfs_key key;
3040         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3041         u64 old_total = btrfs_super_total_bytes(super_copy);
3042         u64 old_size = device->total_bytes;
3043         u64 diff = device->total_bytes - new_size;
3044
3045         if (new_size >= device->total_bytes)
3046                 return -EINVAL;
3047
3048         path = btrfs_alloc_path();
3049         if (!path)
3050                 return -ENOMEM;
3051
3052         path->reada = 2;
3053
3054         lock_chunks(root);
3055
3056         device->total_bytes = new_size;
3057         if (device->writeable) {
3058                 device->fs_devices->total_rw_bytes -= diff;
3059                 spin_lock(&root->fs_info->free_chunk_lock);
3060                 root->fs_info->free_chunk_space -= diff;
3061                 spin_unlock(&root->fs_info->free_chunk_lock);
3062         }
3063         unlock_chunks(root);
3064
3065 again:
3066         key.objectid = device->devid;
3067         key.offset = (u64)-1;
3068         key.type = BTRFS_DEV_EXTENT_KEY;
3069
3070         do {
3071                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3072                 if (ret < 0)
3073                         goto done;
3074
3075                 ret = btrfs_previous_item(root, path, 0, key.type);
3076                 if (ret < 0)
3077                         goto done;
3078                 if (ret) {
3079                         ret = 0;
3080                         btrfs_release_path(path);
3081                         break;
3082                 }
3083
3084                 l = path->nodes[0];
3085                 slot = path->slots[0];
3086                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3087
3088                 if (key.objectid != device->devid) {
3089                         btrfs_release_path(path);
3090                         break;
3091                 }
3092
3093                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3094                 length = btrfs_dev_extent_length(l, dev_extent);
3095
3096                 if (key.offset + length <= new_size) {
3097                         btrfs_release_path(path);
3098                         break;
3099                 }
3100
3101                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3102                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3103                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3104                 btrfs_release_path(path);
3105
3106                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3107                                            chunk_offset);
3108                 if (ret && ret != -ENOSPC)
3109                         goto done;
3110                 if (ret == -ENOSPC)
3111                         failed++;
3112         } while (key.offset-- > 0);
3113
3114         if (failed && !retried) {
3115                 failed = 0;
3116                 retried = true;
3117                 goto again;
3118         } else if (failed && retried) {
3119                 ret = -ENOSPC;
3120                 lock_chunks(root);
3121
3122                 device->total_bytes = old_size;
3123                 if (device->writeable)
3124                         device->fs_devices->total_rw_bytes += diff;
3125                 spin_lock(&root->fs_info->free_chunk_lock);
3126                 root->fs_info->free_chunk_space += diff;
3127                 spin_unlock(&root->fs_info->free_chunk_lock);
3128                 unlock_chunks(root);
3129                 goto done;
3130         }
3131
3132         /* Shrinking succeeded, else we would be at "done". */
3133         trans = btrfs_start_transaction(root, 0);
3134         if (IS_ERR(trans)) {
3135                 ret = PTR_ERR(trans);
3136                 goto done;
3137         }
3138
3139         lock_chunks(root);
3140
3141         device->disk_total_bytes = new_size;
3142         /* Now btrfs_update_device() will change the on-disk size. */
3143         ret = btrfs_update_device(trans, device);
3144         if (ret) {
3145                 unlock_chunks(root);
3146                 btrfs_end_transaction(trans, root);
3147                 goto done;
3148         }
3149         WARN_ON(diff > old_total);
3150         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3151         unlock_chunks(root);
3152         btrfs_end_transaction(trans, root);
3153 done:
3154         btrfs_free_path(path);
3155         return ret;
3156 }
3157
3158 static int btrfs_add_system_chunk(struct btrfs_root *root,
3159                            struct btrfs_key *key,
3160                            struct btrfs_chunk *chunk, int item_size)
3161 {
3162         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3163         struct btrfs_disk_key disk_key;
3164         u32 array_size;
3165         u8 *ptr;
3166
3167         array_size = btrfs_super_sys_array_size(super_copy);
3168         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3169                 return -EFBIG;
3170
3171         ptr = super_copy->sys_chunk_array + array_size;
3172         btrfs_cpu_key_to_disk(&disk_key, key);
3173         memcpy(ptr, &disk_key, sizeof(disk_key));
3174         ptr += sizeof(disk_key);
3175         memcpy(ptr, chunk, item_size);
3176         item_size += sizeof(disk_key);
3177         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3178         return 0;
3179 }
3180
3181 /*
3182  * sort the devices in descending order by max_avail, total_avail
3183  */
3184 static int btrfs_cmp_device_info(const void *a, const void *b)
3185 {
3186         const struct btrfs_device_info *di_a = a;
3187         const struct btrfs_device_info *di_b = b;
3188
3189         if (di_a->max_avail > di_b->max_avail)
3190                 return -1;
3191         if (di_a->max_avail < di_b->max_avail)
3192                 return 1;
3193         if (di_a->total_avail > di_b->total_avail)
3194                 return -1;
3195         if (di_a->total_avail < di_b->total_avail)
3196                 return 1;
3197         return 0;
3198 }
3199
3200 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3201                                struct btrfs_root *extent_root,
3202                                struct map_lookup **map_ret,
3203                                u64 *num_bytes_out, u64 *stripe_size_out,
3204                                u64 start, u64 type)
3205 {
3206         struct btrfs_fs_info *info = extent_root->fs_info;
3207         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3208         struct list_head *cur;
3209         struct map_lookup *map = NULL;
3210         struct extent_map_tree *em_tree;
3211         struct extent_map *em;
3212         struct btrfs_device_info *devices_info = NULL;
3213         u64 total_avail;
3214         int num_stripes;        /* total number of stripes to allocate */
3215         int sub_stripes;        /* sub_stripes info for map */
3216         int dev_stripes;        /* stripes per dev */
3217         int devs_max;           /* max devs to use */
3218         int devs_min;           /* min devs needed */
3219         int devs_increment;     /* ndevs has to be a multiple of this */
3220         int ncopies;            /* how many copies to data has */
3221         int ret;
3222         u64 max_stripe_size;
3223         u64 max_chunk_size;
3224         u64 stripe_size;
3225         u64 num_bytes;
3226         int ndevs;
3227         int i;
3228         int j;
3229
3230         BUG_ON(!alloc_profile_is_valid(type, 0));
3231
3232         if (list_empty(&fs_devices->alloc_list))
3233                 return -ENOSPC;
3234
3235         sub_stripes = 1;
3236         dev_stripes = 1;
3237         devs_increment = 1;
3238         ncopies = 1;
3239         devs_max = 0;   /* 0 == as many as possible */
3240         devs_min = 1;
3241
3242         /*
3243          * define the properties of each RAID type.
3244          * FIXME: move this to a global table and use it in all RAID
3245          * calculation code
3246          */
3247         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3248                 dev_stripes = 2;
3249                 ncopies = 2;
3250                 devs_max = 1;
3251         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3252                 devs_min = 2;
3253         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3254                 devs_increment = 2;
3255                 ncopies = 2;
3256                 devs_max = 2;
3257                 devs_min = 2;
3258         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3259                 sub_stripes = 2;
3260                 devs_increment = 2;
3261                 ncopies = 2;
3262                 devs_min = 4;
3263         } else {
3264                 devs_max = 1;
3265         }
3266
3267         if (type & BTRFS_BLOCK_GROUP_DATA) {
3268                 max_stripe_size = 1024 * 1024 * 1024;
3269                 max_chunk_size = 10 * max_stripe_size;
3270         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3271                 /* for larger filesystems, use larger metadata chunks */
3272                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3273                         max_stripe_size = 1024 * 1024 * 1024;
3274                 else
3275                         max_stripe_size = 256 * 1024 * 1024;
3276                 max_chunk_size = max_stripe_size;
3277         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3278                 max_stripe_size = 32 * 1024 * 1024;
3279                 max_chunk_size = 2 * max_stripe_size;
3280         } else {
3281                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3282                        type);
3283                 BUG_ON(1);
3284         }
3285
3286         /* we don't want a chunk larger than 10% of writeable space */
3287         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3288                              max_chunk_size);
3289
3290         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3291                                GFP_NOFS);
3292         if (!devices_info)
3293                 return -ENOMEM;
3294
3295         cur = fs_devices->alloc_list.next;
3296
3297         /*
3298          * in the first pass through the devices list, we gather information
3299          * about the available holes on each device.
3300          */
3301         ndevs = 0;
3302         while (cur != &fs_devices->alloc_list) {
3303                 struct btrfs_device *device;
3304                 u64 max_avail;
3305                 u64 dev_offset;
3306
3307                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3308
3309                 cur = cur->next;
3310
3311                 if (!device->writeable) {
3312                         printk(KERN_ERR
3313                                "btrfs: read-only device in alloc_list\n");
3314                         WARN_ON(1);
3315                         continue;
3316                 }
3317
3318                 if (!device->in_fs_metadata)
3319                         continue;
3320
3321                 if (device->total_bytes > device->bytes_used)
3322                         total_avail = device->total_bytes - device->bytes_used;
3323                 else
3324                         total_avail = 0;
3325
3326                 /* If there is no space on this device, skip it. */
3327                 if (total_avail == 0)
3328                         continue;
3329
3330                 ret = find_free_dev_extent(device,
3331                                            max_stripe_size * dev_stripes,
3332                                            &dev_offset, &max_avail);
3333                 if (ret && ret != -ENOSPC)
3334                         goto error;
3335
3336                 if (ret == 0)
3337                         max_avail = max_stripe_size * dev_stripes;
3338
3339                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3340                         continue;
3341
3342                 devices_info[ndevs].dev_offset = dev_offset;
3343                 devices_info[ndevs].max_avail = max_avail;
3344                 devices_info[ndevs].total_avail = total_avail;
3345                 devices_info[ndevs].dev = device;
3346                 ++ndevs;
3347         }
3348
3349         /*
3350          * now sort the devices by hole size / available space
3351          */
3352         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3353              btrfs_cmp_device_info, NULL);
3354
3355         /* round down to number of usable stripes */
3356         ndevs -= ndevs % devs_increment;
3357
3358         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3359                 ret = -ENOSPC;
3360                 goto error;
3361         }
3362
3363         if (devs_max && ndevs > devs_max)
3364                 ndevs = devs_max;
3365         /*
3366          * the primary goal is to maximize the number of stripes, so use as many
3367          * devices as possible, even if the stripes are not maximum sized.
3368          */
3369         stripe_size = devices_info[ndevs-1].max_avail;
3370         num_stripes = ndevs * dev_stripes;
3371
3372         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3373                 stripe_size = max_chunk_size * ncopies;
3374                 do_div(stripe_size, ndevs);
3375         }
3376
3377         do_div(stripe_size, dev_stripes);
3378
3379         /* align to BTRFS_STRIPE_LEN */
3380         do_div(stripe_size, BTRFS_STRIPE_LEN);
3381         stripe_size *= BTRFS_STRIPE_LEN;
3382
3383         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3384         if (!map) {
3385                 ret = -ENOMEM;
3386                 goto error;
3387         }
3388         map->num_stripes = num_stripes;
3389
3390         for (i = 0; i < ndevs; ++i) {
3391                 for (j = 0; j < dev_stripes; ++j) {
3392                         int s = i * dev_stripes + j;
3393                         map->stripes[s].dev = devices_info[i].dev;
3394                         map->stripes[s].physical = devices_info[i].dev_offset +
3395                                                    j * stripe_size;
3396                 }
3397         }
3398         map->sector_size = extent_root->sectorsize;
3399         map->stripe_len = BTRFS_STRIPE_LEN;
3400         map->io_align = BTRFS_STRIPE_LEN;
3401         map->io_width = BTRFS_STRIPE_LEN;
3402         map->type = type;
3403         map->sub_stripes = sub_stripes;
3404
3405         *map_ret = map;
3406         num_bytes = stripe_size * (num_stripes / ncopies);
3407
3408         *stripe_size_out = stripe_size;
3409         *num_bytes_out = num_bytes;
3410
3411         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3412
3413         em = alloc_extent_map();
3414         if (!em) {
3415                 ret = -ENOMEM;
3416                 goto error;
3417         }
3418         em->bdev = (struct block_device *)map;
3419         em->start = start;
3420         em->len = num_bytes;
3421         em->block_start = 0;
3422         em->block_len = em->len;
3423
3424         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3425         write_lock(&em_tree->lock);
3426         ret = add_extent_mapping(em_tree, em);
3427         write_unlock(&em_tree->lock);
3428         free_extent_map(em);
3429         if (ret)
3430                 goto error;
3431
3432         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3433                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3434                                      start, num_bytes);
3435         if (ret)
3436                 goto error;
3437
3438         for (i = 0; i < map->num_stripes; ++i) {
3439                 struct btrfs_device *device;
3440                 u64 dev_offset;
3441
3442                 device = map->stripes[i].dev;
3443                 dev_offset = map->stripes[i].physical;
3444
3445                 ret = btrfs_alloc_dev_extent(trans, device,
3446                                 info->chunk_root->root_key.objectid,
3447                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3448                                 start, dev_offset, stripe_size);
3449                 if (ret) {
3450                         btrfs_abort_transaction(trans, extent_root, ret);
3451                         goto error;
3452                 }
3453         }
3454
3455         kfree(devices_info);
3456         return 0;
3457
3458 error:
3459         kfree(map);
3460         kfree(devices_info);
3461         return ret;
3462 }
3463
3464 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3465                                 struct btrfs_root *extent_root,
3466                                 struct map_lookup *map, u64 chunk_offset,
3467                                 u64 chunk_size, u64 stripe_size)
3468 {
3469         u64 dev_offset;
3470         struct btrfs_key key;
3471         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3472         struct btrfs_device *device;
3473         struct btrfs_chunk *chunk;
3474         struct btrfs_stripe *stripe;
3475         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3476         int index = 0;
3477         int ret;
3478
3479         chunk = kzalloc(item_size, GFP_NOFS);
3480         if (!chunk)
3481                 return -ENOMEM;
3482
3483         index = 0;
3484         while (index < map->num_stripes) {
3485                 device = map->stripes[index].dev;
3486                 device->bytes_used += stripe_size;
3487                 ret = btrfs_update_device(trans, device);
3488                 if (ret)
3489                         goto out_free;
3490                 index++;
3491         }
3492
3493         spin_lock(&extent_root->fs_info->free_chunk_lock);
3494         extent_root->fs_info->free_chunk_space -= (stripe_size *
3495                                                    map->num_stripes);
3496         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3497
3498         index = 0;
3499         stripe = &chunk->stripe;
3500         while (index < map->num_stripes) {
3501                 device = map->stripes[index].dev;
3502                 dev_offset = map->stripes[index].physical;
3503
3504                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3505                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3506                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3507                 stripe++;
3508                 index++;
3509         }
3510
3511         btrfs_set_stack_chunk_length(chunk, chunk_size);
3512         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3513         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3514         btrfs_set_stack_chunk_type(chunk, map->type);
3515         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3516         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3517         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3518         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3519         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3520
3521         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3522         key.type = BTRFS_CHUNK_ITEM_KEY;
3523         key.offset = chunk_offset;
3524
3525         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3526
3527         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3528                 /*
3529                  * TODO: Cleanup of inserted chunk root in case of
3530                  * failure.
3531                  */
3532                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3533                                              item_size);
3534         }
3535
3536 out_free:
3537         kfree(chunk);
3538         return ret;
3539 }
3540
3541 /*
3542  * Chunk allocation falls into two parts. The first part does works
3543  * that make the new allocated chunk useable, but not do any operation
3544  * that modifies the chunk tree. The second part does the works that
3545  * require modifying the chunk tree. This division is important for the
3546  * bootstrap process of adding storage to a seed btrfs.
3547  */
3548 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3549                       struct btrfs_root *extent_root, u64 type)
3550 {
3551         u64 chunk_offset;
3552         u64 chunk_size;
3553         u64 stripe_size;
3554         struct map_lookup *map;
3555         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3556         int ret;
3557
3558         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3559                               &chunk_offset);
3560         if (ret)
3561                 return ret;
3562
3563         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3564                                   &stripe_size, chunk_offset, type);
3565         if (ret)
3566                 return ret;
3567
3568         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3569                                    chunk_size, stripe_size);
3570         if (ret)
3571                 return ret;
3572         return 0;
3573 }
3574
3575 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3576                                          struct btrfs_root *root,
3577                                          struct btrfs_device *device)
3578 {
3579         u64 chunk_offset;
3580         u64 sys_chunk_offset;
3581         u64 chunk_size;
3582         u64 sys_chunk_size;
3583         u64 stripe_size;
3584         u64 sys_stripe_size;
3585         u64 alloc_profile;
3586         struct map_lookup *map;
3587         struct map_lookup *sys_map;
3588         struct btrfs_fs_info *fs_info = root->fs_info;
3589         struct btrfs_root *extent_root = fs_info->extent_root;
3590         int ret;
3591
3592         ret = find_next_chunk(fs_info->chunk_root,
3593                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3594         if (ret)
3595                 return ret;
3596
3597         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3598                                 fs_info->avail_metadata_alloc_bits;
3599         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3600
3601         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3602                                   &stripe_size, chunk_offset, alloc_profile);
3603         if (ret)
3604                 return ret;
3605
3606         sys_chunk_offset = chunk_offset + chunk_size;
3607
3608         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3609                                 fs_info->avail_system_alloc_bits;
3610         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3611
3612         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3613                                   &sys_chunk_size, &sys_stripe_size,
3614                                   sys_chunk_offset, alloc_profile);
3615         if (ret)
3616                 goto abort;
3617
3618         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3619         if (ret)
3620                 goto abort;
3621
3622         /*
3623          * Modifying chunk tree needs allocating new blocks from both
3624          * system block group and metadata block group. So we only can
3625          * do operations require modifying the chunk tree after both
3626          * block groups were created.
3627          */
3628         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3629                                    chunk_size, stripe_size);
3630         if (ret)
3631                 goto abort;
3632
3633         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3634                                    sys_chunk_offset, sys_chunk_size,
3635                                    sys_stripe_size);
3636         if (ret)
3637                 goto abort;
3638
3639         return 0;
3640
3641 abort:
3642         btrfs_abort_transaction(trans, root, ret);
3643         return ret;
3644 }
3645
3646 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3647 {
3648         struct extent_map *em;
3649         struct map_lookup *map;
3650         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3651         int readonly = 0;
3652         int i;
3653
3654         read_lock(&map_tree->map_tree.lock);
3655         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3656         read_unlock(&map_tree->map_tree.lock);
3657         if (!em)
3658                 return 1;
3659
3660         if (btrfs_test_opt(root, DEGRADED)) {
3661                 free_extent_map(em);
3662                 return 0;
3663         }
3664
3665         map = (struct map_lookup *)em->bdev;
3666         for (i = 0; i < map->num_stripes; i++) {
3667                 if (!map->stripes[i].dev->writeable) {
3668                         readonly = 1;
3669                         break;
3670                 }
3671         }
3672         free_extent_map(em);
3673         return readonly;
3674 }
3675
3676 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3677 {
3678         extent_map_tree_init(&tree->map_tree);
3679 }
3680
3681 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3682 {
3683         struct extent_map *em;
3684
3685         while (1) {
3686                 write_lock(&tree->map_tree.lock);
3687                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3688                 if (em)
3689                         remove_extent_mapping(&tree->map_tree, em);
3690                 write_unlock(&tree->map_tree.lock);
3691                 if (!em)
3692                         break;
3693                 kfree(em->bdev);
3694                 /* once for us */
3695                 free_extent_map(em);
3696                 /* once for the tree */
3697                 free_extent_map(em);
3698         }
3699 }
3700
3701 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3702 {
3703         struct extent_map *em;
3704         struct map_lookup *map;
3705         struct extent_map_tree *em_tree = &map_tree->map_tree;
3706         int ret;
3707
3708         read_lock(&em_tree->lock);
3709         em = lookup_extent_mapping(em_tree, logical, len);
3710         read_unlock(&em_tree->lock);
3711         BUG_ON(!em);
3712
3713         BUG_ON(em->start > logical || em->start + em->len < logical);
3714         map = (struct map_lookup *)em->bdev;
3715         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3716                 ret = map->num_stripes;
3717         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3718                 ret = map->sub_stripes;
3719         else
3720                 ret = 1;
3721         free_extent_map(em);
3722         return ret;
3723 }
3724
3725 static int find_live_mirror(struct map_lookup *map, int first, int num,
3726                             int optimal)
3727 {
3728         int i;
3729         if (map->stripes[optimal].dev->bdev)
3730                 return optimal;
3731         for (i = first; i < first + num; i++) {
3732                 if (map->stripes[i].dev->bdev)
3733                         return i;
3734         }
3735         /* we couldn't find one that doesn't fail.  Just return something
3736          * and the io error handling code will clean up eventually
3737          */
3738         return optimal;
3739 }
3740
3741 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3742                              u64 logical, u64 *length,
3743                              struct btrfs_bio **bbio_ret,
3744                              int mirror_num)
3745 {
3746         struct extent_map *em;
3747         struct map_lookup *map;
3748         struct extent_map_tree *em_tree = &map_tree->map_tree;
3749         u64 offset;
3750         u64 stripe_offset;
3751         u64 stripe_end_offset;
3752         u64 stripe_nr;
3753         u64 stripe_nr_orig;
3754         u64 stripe_nr_end;
3755         int stripe_index;
3756         int i;
3757         int ret = 0;
3758         int num_stripes;
3759         int max_errors = 0;
3760         struct btrfs_bio *bbio = NULL;
3761
3762         read_lock(&em_tree->lock);
3763         em = lookup_extent_mapping(em_tree, logical, *length);
3764         read_unlock(&em_tree->lock);
3765
3766         if (!em) {
3767                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3768                        (unsigned long long)logical,
3769                        (unsigned long long)*length);
3770                 BUG();
3771         }
3772
3773         BUG_ON(em->start > logical || em->start + em->len < logical);
3774         map = (struct map_lookup *)em->bdev;
3775         offset = logical - em->start;
3776
3777         if (mirror_num > map->num_stripes)
3778                 mirror_num = 0;
3779
3780         stripe_nr = offset;
3781         /*
3782          * stripe_nr counts the total number of stripes we have to stride
3783          * to get to this block
3784          */
3785         do_div(stripe_nr, map->stripe_len);
3786
3787         stripe_offset = stripe_nr * map->stripe_len;
3788         BUG_ON(offset < stripe_offset);
3789
3790         /* stripe_offset is the offset of this block in its stripe*/
3791         stripe_offset = offset - stripe_offset;
3792
3793         if (rw & REQ_DISCARD)
3794                 *length = min_t(u64, em->len - offset, *length);
3795         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3796                 /* we limit the length of each bio to what fits in a stripe */
3797                 *length = min_t(u64, em->len - offset,
3798                                 map->stripe_len - stripe_offset);
3799         } else {
3800                 *length = em->len - offset;
3801         }
3802
3803         if (!bbio_ret)
3804                 goto out;
3805
3806         num_stripes = 1;
3807         stripe_index = 0;
3808         stripe_nr_orig = stripe_nr;
3809         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3810                         (~(map->stripe_len - 1));
3811         do_div(stripe_nr_end, map->stripe_len);
3812         stripe_end_offset = stripe_nr_end * map->stripe_len -
3813                             (offset + *length);
3814         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3815                 if (rw & REQ_DISCARD)
3816                         num_stripes = min_t(u64, map->num_stripes,
3817                                             stripe_nr_end - stripe_nr_orig);
3818                 stripe_index = do_div(stripe_nr, map->num_stripes);
3819         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3820                 if (rw & (REQ_WRITE | REQ_DISCARD))
3821                         num_stripes = map->num_stripes;
3822                 else if (mirror_num)
3823                         stripe_index = mirror_num - 1;
3824                 else {
3825                         stripe_index = find_live_mirror(map, 0,
3826                                             map->num_stripes,
3827                                             current->pid % map->num_stripes);
3828                         mirror_num = stripe_index + 1;
3829                 }
3830
3831         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3832                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3833                         num_stripes = map->num_stripes;
3834                 } else if (mirror_num) {
3835                         stripe_index = mirror_num - 1;
3836                 } else {
3837                         mirror_num = 1;
3838                 }
3839
3840         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3841                 int factor = map->num_stripes / map->sub_stripes;
3842
3843                 stripe_index = do_div(stripe_nr, factor);
3844                 stripe_index *= map->sub_stripes;
3845
3846                 if (rw & REQ_WRITE)
3847                         num_stripes = map->sub_stripes;
3848                 else if (rw & REQ_DISCARD)
3849                         num_stripes = min_t(u64, map->sub_stripes *
3850                                             (stripe_nr_end - stripe_nr_orig),
3851                                             map->num_stripes);
3852                 else if (mirror_num)
3853                         stripe_index += mirror_num - 1;
3854                 else {
3855                         int old_stripe_index = stripe_index;
3856                         stripe_index = find_live_mirror(map, stripe_index,
3857                                               map->sub_stripes, stripe_index +
3858                                               current->pid % map->sub_stripes);
3859                         mirror_num = stripe_index - old_stripe_index + 1;
3860                 }
3861         } else {
3862                 /*
3863                  * after this do_div call, stripe_nr is the number of stripes
3864                  * on this device we have to walk to find the data, and
3865                  * stripe_index is the number of our device in the stripe array
3866                  */
3867                 stripe_index = do_div(stripe_nr, map->num_stripes);
3868                 mirror_num = stripe_index + 1;
3869         }
3870         BUG_ON(stripe_index >= map->num_stripes);
3871
3872         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3873         if (!bbio) {
3874                 ret = -ENOMEM;
3875                 goto out;
3876         }
3877         atomic_set(&bbio->error, 0);
3878
3879         if (rw & REQ_DISCARD) {
3880                 int factor = 0;
3881                 int sub_stripes = 0;
3882                 u64 stripes_per_dev = 0;
3883                 u32 remaining_stripes = 0;
3884                 u32 last_stripe = 0;
3885
3886                 if (map->type &
3887                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3888                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3889                                 sub_stripes = 1;
3890                         else
3891                                 sub_stripes = map->sub_stripes;
3892
3893                         factor = map->num_stripes / sub_stripes;
3894                         stripes_per_dev = div_u64_rem(stripe_nr_end -
3895                                                       stripe_nr_orig,
3896                                                       factor,
3897                                                       &remaining_stripes);
3898                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
3899                         last_stripe *= sub_stripes;
3900                 }
3901
3902                 for (i = 0; i < num_stripes; i++) {
3903                         bbio->stripes[i].physical =
3904                                 map->stripes[stripe_index].physical +
3905                                 stripe_offset + stripe_nr * map->stripe_len;
3906                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3907
3908                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3909                                          BTRFS_BLOCK_GROUP_RAID10)) {
3910                                 bbio->stripes[i].length = stripes_per_dev *
3911                                                           map->stripe_len;
3912
3913                                 if (i / sub_stripes < remaining_stripes)
3914                                         bbio->stripes[i].length +=
3915                                                 map->stripe_len;
3916
3917                                 /*
3918                                  * Special for the first stripe and
3919                                  * the last stripe:
3920                                  *
3921                                  * |-------|...|-------|
3922                                  *     |----------|
3923                                  *    off     end_off
3924                                  */
3925                                 if (i < sub_stripes)
3926                                         bbio->stripes[i].length -=
3927                                                 stripe_offset;
3928
3929                                 if (stripe_index >= last_stripe &&
3930                                     stripe_index <= (last_stripe +
3931                                                      sub_stripes - 1))
3932                                         bbio->stripes[i].length -=
3933                                                 stripe_end_offset;
3934
3935                                 if (i == sub_stripes - 1)
3936                                         stripe_offset = 0;
3937                         } else
3938                                 bbio->stripes[i].length = *length;
3939
3940                         stripe_index++;
3941                         if (stripe_index == map->num_stripes) {
3942                                 /* This could only happen for RAID0/10 */
3943                                 stripe_index = 0;
3944                                 stripe_nr++;
3945                         }
3946                 }
3947         } else {
3948                 for (i = 0; i < num_stripes; i++) {
3949                         bbio->stripes[i].physical =
3950                                 map->stripes[stripe_index].physical +
3951                                 stripe_offset +
3952                                 stripe_nr * map->stripe_len;
3953                         bbio->stripes[i].dev =
3954                                 map->stripes[stripe_index].dev;
3955                         stripe_index++;
3956                 }
3957         }
3958
3959         if (rw & REQ_WRITE) {
3960                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3961                                  BTRFS_BLOCK_GROUP_RAID10 |
3962                                  BTRFS_BLOCK_GROUP_DUP)) {
3963                         max_errors = 1;
3964                 }
3965         }
3966
3967         *bbio_ret = bbio;
3968         bbio->num_stripes = num_stripes;
3969         bbio->max_errors = max_errors;
3970         bbio->mirror_num = mirror_num;
3971 out:
3972         free_extent_map(em);
3973         return ret;
3974 }
3975
3976 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3977                       u64 logical, u64 *length,
3978                       struct btrfs_bio **bbio_ret, int mirror_num)
3979 {
3980         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3981                                  mirror_num);
3982 }
3983
3984 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3985                      u64 chunk_start, u64 physical, u64 devid,
3986                      u64 **logical, int *naddrs, int *stripe_len)
3987 {
3988         struct extent_map_tree *em_tree = &map_tree->map_tree;
3989         struct extent_map *em;
3990         struct map_lookup *map;
3991         u64 *buf;
3992         u64 bytenr;
3993         u64 length;
3994         u64 stripe_nr;
3995         int i, j, nr = 0;
3996
3997         read_lock(&em_tree->lock);
3998         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3999         read_unlock(&em_tree->lock);
4000
4001         BUG_ON(!em || em->start != chunk_start);
4002         map = (struct map_lookup *)em->bdev;
4003
4004         length = em->len;
4005         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4006                 do_div(length, map->num_stripes / map->sub_stripes);
4007         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4008                 do_div(length, map->num_stripes);
4009
4010         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4011         BUG_ON(!buf); /* -ENOMEM */
4012
4013         for (i = 0; i < map->num_stripes; i++) {
4014                 if (devid && map->stripes[i].dev->devid != devid)
4015                         continue;
4016                 if (map->stripes[i].physical > physical ||
4017                     map->stripes[i].physical + length <= physical)
4018                         continue;
4019
4020                 stripe_nr = physical - map->stripes[i].physical;
4021                 do_div(stripe_nr, map->stripe_len);
4022
4023                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4024                         stripe_nr = stripe_nr * map->num_stripes + i;
4025                         do_div(stripe_nr, map->sub_stripes);
4026                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4027                         stripe_nr = stripe_nr * map->num_stripes + i;
4028                 }
4029                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4030                 WARN_ON(nr >= map->num_stripes);
4031                 for (j = 0; j < nr; j++) {
4032                         if (buf[j] == bytenr)
4033                                 break;
4034                 }
4035                 if (j == nr) {
4036                         WARN_ON(nr >= map->num_stripes);
4037                         buf[nr++] = bytenr;
4038                 }
4039         }
4040
4041         *logical = buf;
4042         *naddrs = nr;
4043         *stripe_len = map->stripe_len;
4044
4045         free_extent_map(em);
4046         return 0;
4047 }
4048
4049 static void *merge_stripe_index_into_bio_private(void *bi_private,
4050                                                  unsigned int stripe_index)
4051 {
4052         /*
4053          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4054          * at most 1.
4055          * The alternative solution (instead of stealing bits from the
4056          * pointer) would be to allocate an intermediate structure
4057          * that contains the old private pointer plus the stripe_index.
4058          */
4059         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4060         BUG_ON(stripe_index > 3);
4061         return (void *)(((uintptr_t)bi_private) | stripe_index);
4062 }
4063
4064 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4065 {
4066         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4067 }
4068
4069 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4070 {
4071         return (unsigned int)((uintptr_t)bi_private) & 3;
4072 }
4073
4074 static void btrfs_end_bio(struct bio *bio, int err)
4075 {
4076         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4077         int is_orig_bio = 0;
4078
4079         if (err) {
4080                 atomic_inc(&bbio->error);
4081                 if (err == -EIO || err == -EREMOTEIO) {
4082                         unsigned int stripe_index =
4083                                 extract_stripe_index_from_bio_private(
4084                                         bio->bi_private);
4085                         struct btrfs_device *dev;
4086
4087                         BUG_ON(stripe_index >= bbio->num_stripes);
4088                         dev = bbio->stripes[stripe_index].dev;
4089                         if (dev->bdev) {
4090                                 if (bio->bi_rw & WRITE)
4091                                         btrfs_dev_stat_inc(dev,
4092                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4093                                 else
4094                                         btrfs_dev_stat_inc(dev,
4095                                                 BTRFS_DEV_STAT_READ_ERRS);
4096                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4097                                         btrfs_dev_stat_inc(dev,
4098                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4099                                 btrfs_dev_stat_print_on_error(dev);
4100                         }
4101                 }
4102         }
4103
4104         if (bio == bbio->orig_bio)
4105                 is_orig_bio = 1;
4106
4107         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4108                 if (!is_orig_bio) {
4109                         bio_put(bio);
4110                         bio = bbio->orig_bio;
4111                 }
4112                 bio->bi_private = bbio->private;
4113                 bio->bi_end_io = bbio->end_io;
4114                 bio->bi_bdev = (struct block_device *)
4115                                         (unsigned long)bbio->mirror_num;
4116                 /* only send an error to the higher layers if it is
4117                  * beyond the tolerance of the multi-bio
4118                  */
4119                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4120                         err = -EIO;
4121                 } else {
4122                         /*
4123                          * this bio is actually up to date, we didn't
4124                          * go over the max number of errors
4125                          */
4126                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4127                         err = 0;
4128                 }
4129                 kfree(bbio);
4130
4131                 bio_endio(bio, err);
4132         } else if (!is_orig_bio) {
4133                 bio_put(bio);
4134         }
4135 }
4136
4137 struct async_sched {
4138         struct bio *bio;
4139         int rw;
4140         struct btrfs_fs_info *info;
4141         struct btrfs_work work;
4142 };
4143
4144 /*
4145  * see run_scheduled_bios for a description of why bios are collected for
4146  * async submit.
4147  *
4148  * This will add one bio to the pending list for a device and make sure
4149  * the work struct is scheduled.
4150  */
4151 static noinline void schedule_bio(struct btrfs_root *root,
4152                                  struct btrfs_device *device,
4153                                  int rw, struct bio *bio)
4154 {
4155         int should_queue = 1;
4156         struct btrfs_pending_bios *pending_bios;
4157
4158         /* don't bother with additional async steps for reads, right now */
4159         if (!(rw & REQ_WRITE)) {
4160                 bio_get(bio);
4161                 btrfsic_submit_bio(rw, bio);
4162                 bio_put(bio);
4163                 return;
4164         }
4165
4166         /*
4167          * nr_async_bios allows us to reliably return congestion to the
4168          * higher layers.  Otherwise, the async bio makes it appear we have
4169          * made progress against dirty pages when we've really just put it
4170          * on a queue for later
4171          */
4172         atomic_inc(&root->fs_info->nr_async_bios);
4173         WARN_ON(bio->bi_next);
4174         bio->bi_next = NULL;
4175         bio->bi_rw |= rw;
4176
4177         spin_lock(&device->io_lock);
4178         if (bio->bi_rw & REQ_SYNC)
4179                 pending_bios = &device->pending_sync_bios;
4180         else
4181                 pending_bios = &device->pending_bios;
4182
4183         if (pending_bios->tail)
4184                 pending_bios->tail->bi_next = bio;
4185
4186         pending_bios->tail = bio;
4187         if (!pending_bios->head)
4188                 pending_bios->head = bio;
4189         if (device->running_pending)
4190                 should_queue = 0;
4191
4192         spin_unlock(&device->io_lock);
4193
4194         if (should_queue)
4195                 btrfs_queue_worker(&root->fs_info->submit_workers,
4196                                    &device->work);
4197 }
4198
4199 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4200                   int mirror_num, int async_submit)
4201 {
4202         struct btrfs_mapping_tree *map_tree;
4203         struct btrfs_device *dev;
4204         struct bio *first_bio = bio;
4205         u64 logical = (u64)bio->bi_sector << 9;
4206         u64 length = 0;
4207         u64 map_length;
4208         int ret;
4209         int dev_nr = 0;
4210         int total_devs = 1;
4211         struct btrfs_bio *bbio = NULL;
4212
4213         length = bio->bi_size;
4214         map_tree = &root->fs_info->mapping_tree;
4215         map_length = length;
4216
4217         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4218                               mirror_num);
4219         if (ret) /* -ENOMEM */
4220                 return ret;
4221
4222         total_devs = bbio->num_stripes;
4223         if (map_length < length) {
4224                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
4225                        "len %llu\n", (unsigned long long)logical,
4226                        (unsigned long long)length,
4227                        (unsigned long long)map_length);
4228                 BUG();
4229         }
4230
4231         bbio->orig_bio = first_bio;
4232         bbio->private = first_bio->bi_private;
4233         bbio->end_io = first_bio->bi_end_io;
4234         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4235
4236         while (dev_nr < total_devs) {
4237                 if (dev_nr < total_devs - 1) {
4238                         bio = bio_clone(first_bio, GFP_NOFS);
4239                         BUG_ON(!bio); /* -ENOMEM */
4240                 } else {
4241                         bio = first_bio;
4242                 }
4243                 bio->bi_private = bbio;
4244                 bio->bi_private = merge_stripe_index_into_bio_private(
4245                                 bio->bi_private, (unsigned int)dev_nr);
4246                 bio->bi_end_io = btrfs_end_bio;
4247                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4248                 dev = bbio->stripes[dev_nr].dev;
4249                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4250 #ifdef DEBUG
4251                         struct rcu_string *name;
4252
4253                         rcu_read_lock();
4254                         name = rcu_dereference(dev->name);
4255                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4256                                  "(%s id %llu), size=%u\n", rw,
4257                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4258                                  name->str, dev->devid, bio->bi_size);
4259                         rcu_read_unlock();
4260 #endif
4261                         bio->bi_bdev = dev->bdev;
4262                         if (async_submit)
4263                                 schedule_bio(root, dev, rw, bio);
4264                         else
4265                                 btrfsic_submit_bio(rw, bio);
4266                 } else {
4267                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4268                         bio->bi_sector = logical >> 9;
4269                         bio_endio(bio, -EIO);
4270                 }
4271                 dev_nr++;
4272         }
4273         return 0;
4274 }
4275
4276 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4277                                        u8 *uuid, u8 *fsid)
4278 {
4279         struct btrfs_device *device;
4280         struct btrfs_fs_devices *cur_devices;
4281
4282         cur_devices = root->fs_info->fs_devices;
4283         while (cur_devices) {
4284                 if (!fsid ||
4285                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4286                         device = __find_device(&cur_devices->devices,
4287                                                devid, uuid);
4288                         if (device)
4289                                 return device;
4290                 }
4291                 cur_devices = cur_devices->seed;
4292         }
4293         return NULL;
4294 }
4295
4296 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4297                                             u64 devid, u8 *dev_uuid)
4298 {
4299         struct btrfs_device *device;
4300         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4301
4302         device = kzalloc(sizeof(*device), GFP_NOFS);
4303         if (!device)
4304                 return NULL;
4305         list_add(&device->dev_list,
4306                  &fs_devices->devices);
4307         device->dev_root = root->fs_info->dev_root;
4308         device->devid = devid;
4309         device->work.func = pending_bios_fn;
4310         device->fs_devices = fs_devices;
4311         device->missing = 1;
4312         fs_devices->num_devices++;
4313         fs_devices->missing_devices++;
4314         spin_lock_init(&device->io_lock);
4315         INIT_LIST_HEAD(&device->dev_alloc_list);
4316         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4317         return device;
4318 }
4319
4320 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4321                           struct extent_buffer *leaf,
4322                           struct btrfs_chunk *chunk)
4323 {
4324         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4325         struct map_lookup *map;
4326         struct extent_map *em;
4327         u64 logical;
4328         u64 length;
4329         u64 devid;
4330         u8 uuid[BTRFS_UUID_SIZE];
4331         int num_stripes;
4332         int ret;
4333         int i;
4334
4335         logical = key->offset;
4336         length = btrfs_chunk_length(leaf, chunk);
4337
4338         read_lock(&map_tree->map_tree.lock);
4339         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4340         read_unlock(&map_tree->map_tree.lock);
4341
4342         /* already mapped? */
4343         if (em && em->start <= logical && em->start + em->len > logical) {
4344                 free_extent_map(em);
4345                 return 0;
4346         } else if (em) {
4347                 free_extent_map(em);
4348         }
4349
4350         em = alloc_extent_map();
4351         if (!em)
4352                 return -ENOMEM;
4353         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4354         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4355         if (!map) {
4356                 free_extent_map(em);
4357                 return -ENOMEM;
4358         }
4359
4360         em->bdev = (struct block_device *)map;
4361         em->start = logical;
4362         em->len = length;
4363         em->block_start = 0;
4364         em->block_len = em->len;
4365
4366         map->num_stripes = num_stripes;
4367         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4368         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4369         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4370         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4371         map->type = btrfs_chunk_type(leaf, chunk);
4372         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4373         for (i = 0; i < num_stripes; i++) {
4374                 map->stripes[i].physical =
4375                         btrfs_stripe_offset_nr(leaf, chunk, i);
4376                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4377                 read_extent_buffer(leaf, uuid, (unsigned long)
4378                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4379                                    BTRFS_UUID_SIZE);
4380                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4381                                                         NULL);
4382                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4383                         kfree(map);
4384                         free_extent_map(em);
4385                         return -EIO;
4386                 }
4387                 if (!map->stripes[i].dev) {
4388                         map->stripes[i].dev =
4389                                 add_missing_dev(root, devid, uuid);
4390                         if (!map->stripes[i].dev) {
4391                                 kfree(map);
4392                                 free_extent_map(em);
4393                                 return -EIO;
4394                         }
4395                 }
4396                 map->stripes[i].dev->in_fs_metadata = 1;
4397         }
4398
4399         write_lock(&map_tree->map_tree.lock);
4400         ret = add_extent_mapping(&map_tree->map_tree, em);
4401         write_unlock(&map_tree->map_tree.lock);
4402         BUG_ON(ret); /* Tree corruption */
4403         free_extent_map(em);
4404
4405         return 0;
4406 }
4407
4408 static void fill_device_from_item(struct extent_buffer *leaf,
4409                                  struct btrfs_dev_item *dev_item,
4410                                  struct btrfs_device *device)
4411 {
4412         unsigned long ptr;
4413
4414         device->devid = btrfs_device_id(leaf, dev_item);
4415         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4416         device->total_bytes = device->disk_total_bytes;
4417         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4418         device->type = btrfs_device_type(leaf, dev_item);
4419         device->io_align = btrfs_device_io_align(leaf, dev_item);
4420         device->io_width = btrfs_device_io_width(leaf, dev_item);
4421         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4422
4423         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4424         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4425 }
4426
4427 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4428 {
4429         struct btrfs_fs_devices *fs_devices;
4430         int ret;
4431
4432         BUG_ON(!mutex_is_locked(&uuid_mutex));
4433
4434         fs_devices = root->fs_info->fs_devices->seed;
4435         while (fs_devices) {
4436                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4437                         ret = 0;
4438                         goto out;
4439                 }
4440                 fs_devices = fs_devices->seed;
4441         }
4442
4443         fs_devices = find_fsid(fsid);
4444         if (!fs_devices) {
4445                 ret = -ENOENT;
4446                 goto out;
4447         }
4448
4449         fs_devices = clone_fs_devices(fs_devices);
4450         if (IS_ERR(fs_devices)) {
4451                 ret = PTR_ERR(fs_devices);
4452                 goto out;
4453         }
4454
4455         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4456                                    root->fs_info->bdev_holder);
4457         if (ret) {
4458                 free_fs_devices(fs_devices);
4459                 goto out;
4460         }
4461
4462         if (!fs_devices->seeding) {
4463                 __btrfs_close_devices(fs_devices);
4464                 free_fs_devices(fs_devices);
4465                 ret = -EINVAL;
4466                 goto out;
4467         }
4468
4469         fs_devices->seed = root->fs_info->fs_devices->seed;
4470         root->fs_info->fs_devices->seed = fs_devices;
4471 out:
4472         return ret;
4473 }
4474
4475 static int read_one_dev(struct btrfs_root *root,
4476                         struct extent_buffer *leaf,
4477                         struct btrfs_dev_item *dev_item)
4478 {
4479         struct btrfs_device *device;
4480         u64 devid;
4481         int ret;
4482         u8 fs_uuid[BTRFS_UUID_SIZE];
4483         u8 dev_uuid[BTRFS_UUID_SIZE];
4484
4485         devid = btrfs_device_id(leaf, dev_item);
4486         read_extent_buffer(leaf, dev_uuid,
4487                            (unsigned long)btrfs_device_uuid(dev_item),
4488                            BTRFS_UUID_SIZE);
4489         read_extent_buffer(leaf, fs_uuid,
4490                            (unsigned long)btrfs_device_fsid(dev_item),
4491                            BTRFS_UUID_SIZE);
4492
4493         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4494                 ret = open_seed_devices(root, fs_uuid);
4495                 if (ret && !btrfs_test_opt(root, DEGRADED))
4496                         return ret;
4497         }
4498
4499         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4500         if (!device || !device->bdev) {
4501                 if (!btrfs_test_opt(root, DEGRADED))
4502                         return -EIO;
4503
4504                 if (!device) {
4505                         printk(KERN_WARNING "warning devid %llu missing\n",
4506                                (unsigned long long)devid);
4507                         device = add_missing_dev(root, devid, dev_uuid);
4508                         if (!device)
4509                                 return -ENOMEM;
4510                 } else if (!device->missing) {
4511                         /*
4512                          * this happens when a device that was properly setup
4513                          * in the device info lists suddenly goes bad.
4514                          * device->bdev is NULL, and so we have to set
4515                          * device->missing to one here
4516                          */
4517                         root->fs_info->fs_devices->missing_devices++;
4518                         device->missing = 1;
4519                 }
4520         }
4521
4522         if (device->fs_devices != root->fs_info->fs_devices) {
4523                 BUG_ON(device->writeable);
4524                 if (device->generation !=
4525                     btrfs_device_generation(leaf, dev_item))
4526                         return -EINVAL;
4527         }
4528
4529         fill_device_from_item(leaf, dev_item, device);
4530         device->dev_root = root->fs_info->dev_root;
4531         device->in_fs_metadata = 1;
4532         if (device->writeable) {
4533                 device->fs_devices->total_rw_bytes += device->total_bytes;
4534                 spin_lock(&root->fs_info->free_chunk_lock);
4535                 root->fs_info->free_chunk_space += device->total_bytes -
4536                         device->bytes_used;
4537                 spin_unlock(&root->fs_info->free_chunk_lock);
4538         }
4539         ret = 0;
4540         return ret;
4541 }
4542
4543 int btrfs_read_sys_array(struct btrfs_root *root)
4544 {
4545         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4546         struct extent_buffer *sb;
4547         struct btrfs_disk_key *disk_key;
4548         struct btrfs_chunk *chunk;
4549         u8 *ptr;
4550         unsigned long sb_ptr;
4551         int ret = 0;
4552         u32 num_stripes;
4553         u32 array_size;
4554         u32 len = 0;
4555         u32 cur;
4556         struct btrfs_key key;
4557
4558         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4559                                           BTRFS_SUPER_INFO_SIZE);
4560         if (!sb)
4561                 return -ENOMEM;
4562         btrfs_set_buffer_uptodate(sb);
4563         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4564         /*
4565          * The sb extent buffer is artifical and just used to read the system array.
4566          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4567          * pages up-to-date when the page is larger: extent does not cover the
4568          * whole page and consequently check_page_uptodate does not find all
4569          * the page's extents up-to-date (the hole beyond sb),
4570          * write_extent_buffer then triggers a WARN_ON.
4571          *
4572          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4573          * but sb spans only this function. Add an explicit SetPageUptodate call
4574          * to silence the warning eg. on PowerPC 64.
4575          */
4576         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4577                 SetPageUptodate(sb->pages[0]);
4578
4579         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4580         array_size = btrfs_super_sys_array_size(super_copy);
4581
4582         ptr = super_copy->sys_chunk_array;
4583         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4584         cur = 0;
4585
4586         while (cur < array_size) {
4587                 disk_key = (struct btrfs_disk_key *)ptr;
4588                 btrfs_disk_key_to_cpu(&key, disk_key);
4589
4590                 len = sizeof(*disk_key); ptr += len;
4591                 sb_ptr += len;
4592                 cur += len;
4593
4594                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4595                         chunk = (struct btrfs_chunk *)sb_ptr;
4596                         ret = read_one_chunk(root, &key, sb, chunk);
4597                         if (ret)
4598                                 break;
4599                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4600                         len = btrfs_chunk_item_size(num_stripes);
4601                 } else {
4602                         ret = -EIO;
4603                         break;
4604                 }
4605                 ptr += len;
4606                 sb_ptr += len;
4607                 cur += len;
4608         }
4609         free_extent_buffer(sb);
4610         return ret;
4611 }
4612
4613 int btrfs_read_chunk_tree(struct btrfs_root *root)
4614 {
4615         struct btrfs_path *path;
4616         struct extent_buffer *leaf;
4617         struct btrfs_key key;
4618         struct btrfs_key found_key;
4619         int ret;
4620         int slot;
4621
4622         root = root->fs_info->chunk_root;
4623
4624         path = btrfs_alloc_path();
4625         if (!path)
4626                 return -ENOMEM;
4627
4628         mutex_lock(&uuid_mutex);
4629         lock_chunks(root);
4630
4631         /* first we search for all of the device items, and then we
4632          * read in all of the chunk items.  This way we can create chunk
4633          * mappings that reference all of the devices that are afound
4634          */
4635         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4636         key.offset = 0;
4637         key.type = 0;
4638 again:
4639         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4640         if (ret < 0)
4641                 goto error;
4642         while (1) {
4643                 leaf = path->nodes[0];
4644                 slot = path->slots[0];
4645                 if (slot >= btrfs_header_nritems(leaf)) {
4646                         ret = btrfs_next_leaf(root, path);
4647                         if (ret == 0)
4648                                 continue;
4649                         if (ret < 0)
4650                                 goto error;
4651                         break;
4652                 }
4653                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4654                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4655                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4656                                 break;
4657                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4658                                 struct btrfs_dev_item *dev_item;
4659                                 dev_item = btrfs_item_ptr(leaf, slot,
4660                                                   struct btrfs_dev_item);
4661                                 ret = read_one_dev(root, leaf, dev_item);
4662                                 if (ret)
4663                                         goto error;
4664                         }
4665                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4666                         struct btrfs_chunk *chunk;
4667                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4668                         ret = read_one_chunk(root, &found_key, leaf, chunk);
4669                         if (ret)
4670                                 goto error;
4671                 }
4672                 path->slots[0]++;
4673         }
4674         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4675                 key.objectid = 0;
4676                 btrfs_release_path(path);
4677                 goto again;
4678         }
4679         ret = 0;
4680 error:
4681         unlock_chunks(root);
4682         mutex_unlock(&uuid_mutex);
4683
4684         btrfs_free_path(path);
4685         return ret;
4686 }
4687
4688 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
4689 {
4690         int i;
4691
4692         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4693                 btrfs_dev_stat_reset(dev, i);
4694 }
4695
4696 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
4697 {
4698         struct btrfs_key key;
4699         struct btrfs_key found_key;
4700         struct btrfs_root *dev_root = fs_info->dev_root;
4701         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4702         struct extent_buffer *eb;
4703         int slot;
4704         int ret = 0;
4705         struct btrfs_device *device;
4706         struct btrfs_path *path = NULL;
4707         int i;
4708
4709         path = btrfs_alloc_path();
4710         if (!path) {
4711                 ret = -ENOMEM;
4712                 goto out;
4713         }
4714
4715         mutex_lock(&fs_devices->device_list_mutex);
4716         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4717                 int item_size;
4718                 struct btrfs_dev_stats_item *ptr;
4719
4720                 key.objectid = 0;
4721                 key.type = BTRFS_DEV_STATS_KEY;
4722                 key.offset = device->devid;
4723                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
4724                 if (ret) {
4725                         __btrfs_reset_dev_stats(device);
4726                         device->dev_stats_valid = 1;
4727                         btrfs_release_path(path);
4728                         continue;
4729                 }
4730                 slot = path->slots[0];
4731                 eb = path->nodes[0];
4732                 btrfs_item_key_to_cpu(eb, &found_key, slot);
4733                 item_size = btrfs_item_size_nr(eb, slot);
4734
4735                 ptr = btrfs_item_ptr(eb, slot,
4736                                      struct btrfs_dev_stats_item);
4737
4738                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4739                         if (item_size >= (1 + i) * sizeof(__le64))
4740                                 btrfs_dev_stat_set(device, i,
4741                                         btrfs_dev_stats_value(eb, ptr, i));
4742                         else
4743                                 btrfs_dev_stat_reset(device, i);
4744                 }
4745
4746                 device->dev_stats_valid = 1;
4747                 btrfs_dev_stat_print_on_load(device);
4748                 btrfs_release_path(path);
4749         }
4750         mutex_unlock(&fs_devices->device_list_mutex);
4751
4752 out:
4753         btrfs_free_path(path);
4754         return ret < 0 ? ret : 0;
4755 }
4756
4757 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
4758                                 struct btrfs_root *dev_root,
4759                                 struct btrfs_device *device)
4760 {
4761         struct btrfs_path *path;
4762         struct btrfs_key key;
4763         struct extent_buffer *eb;
4764         struct btrfs_dev_stats_item *ptr;
4765         int ret;
4766         int i;
4767
4768         key.objectid = 0;
4769         key.type = BTRFS_DEV_STATS_KEY;
4770         key.offset = device->devid;
4771
4772         path = btrfs_alloc_path();
4773         BUG_ON(!path);
4774         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
4775         if (ret < 0) {
4776                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
4777                               ret, rcu_str_deref(device->name));
4778                 goto out;
4779         }
4780
4781         if (ret == 0 &&
4782             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
4783                 /* need to delete old one and insert a new one */
4784                 ret = btrfs_del_item(trans, dev_root, path);
4785                 if (ret != 0) {
4786                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
4787                                       rcu_str_deref(device->name), ret);
4788                         goto out;
4789                 }
4790                 ret = 1;
4791         }
4792
4793         if (ret == 1) {
4794                 /* need to insert a new item */
4795                 btrfs_release_path(path);
4796                 ret = btrfs_insert_empty_item(trans, dev_root, path,
4797                                               &key, sizeof(*ptr));
4798                 if (ret < 0) {
4799                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
4800                                       rcu_str_deref(device->name), ret);
4801                         goto out;
4802                 }
4803         }
4804
4805         eb = path->nodes[0];
4806         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
4807         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4808                 btrfs_set_dev_stats_value(eb, ptr, i,
4809                                           btrfs_dev_stat_read(device, i));
4810         btrfs_mark_buffer_dirty(eb);
4811
4812 out:
4813         btrfs_free_path(path);
4814         return ret;
4815 }
4816
4817 /*
4818  * called from commit_transaction. Writes all changed device stats to disk.
4819  */
4820 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
4821                         struct btrfs_fs_info *fs_info)
4822 {
4823         struct btrfs_root *dev_root = fs_info->dev_root;
4824         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4825         struct btrfs_device *device;
4826         int ret = 0;
4827
4828         mutex_lock(&fs_devices->device_list_mutex);
4829         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4830                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
4831                         continue;
4832
4833                 ret = update_dev_stat_item(trans, dev_root, device);
4834                 if (!ret)
4835                         device->dev_stats_dirty = 0;
4836         }
4837         mutex_unlock(&fs_devices->device_list_mutex);
4838
4839         return ret;
4840 }
4841
4842 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
4843 {
4844         btrfs_dev_stat_inc(dev, index);
4845         btrfs_dev_stat_print_on_error(dev);
4846 }
4847
4848 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
4849 {
4850         if (!dev->dev_stats_valid)
4851                 return;
4852         printk_ratelimited_in_rcu(KERN_ERR
4853                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4854                            rcu_str_deref(dev->name),
4855                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4856                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4857                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4858                            btrfs_dev_stat_read(dev,
4859                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
4860                            btrfs_dev_stat_read(dev,
4861                                                BTRFS_DEV_STAT_GENERATION_ERRS));
4862 }
4863
4864 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
4865 {
4866         int i;
4867
4868         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4869                 if (btrfs_dev_stat_read(dev, i) != 0)
4870                         break;
4871         if (i == BTRFS_DEV_STAT_VALUES_MAX)
4872                 return; /* all values == 0, suppress message */
4873
4874         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4875                rcu_str_deref(dev->name),
4876                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4877                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4878                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4879                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
4880                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
4881 }
4882
4883 int btrfs_get_dev_stats(struct btrfs_root *root,
4884                         struct btrfs_ioctl_get_dev_stats *stats)
4885 {
4886         struct btrfs_device *dev;
4887         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4888         int i;
4889
4890         mutex_lock(&fs_devices->device_list_mutex);
4891         dev = btrfs_find_device(root, stats->devid, NULL, NULL);
4892         mutex_unlock(&fs_devices->device_list_mutex);
4893
4894         if (!dev) {
4895                 printk(KERN_WARNING
4896                        "btrfs: get dev_stats failed, device not found\n");
4897                 return -ENODEV;
4898         } else if (!dev->dev_stats_valid) {
4899                 printk(KERN_WARNING
4900                        "btrfs: get dev_stats failed, not yet valid\n");
4901                 return -ENODEV;
4902         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
4903                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4904                         if (stats->nr_items > i)
4905                                 stats->values[i] =
4906                                         btrfs_dev_stat_read_and_reset(dev, i);
4907                         else
4908                                 btrfs_dev_stat_reset(dev, i);
4909                 }
4910         } else {
4911                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4912                         if (stats->nr_items > i)
4913                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
4914         }
4915         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
4916                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
4917         return 0;
4918 }