fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <asm/div64.h>
  27 #include "compat.h"
  28 #include "ctree.h"
  29 #include "extent_map.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32 #include "print-tree.h"
  33 #include "volumes.h"
  34 #include "async-thread.h"
  35
  36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  37                                 struct btrfs_root *root,
  38                                 struct btrfs_device *device);
  39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  40
  41 static DEFINE_MUTEX(uuid_mutex);
  42 static LIST_HEAD(fs_uuids);
  43
  44 static void lock_chunks(struct btrfs_root *root)
  45 {
  46         mutex_lock(&root->fs_info->chunk_mutex);
  47 }
  48
  49 static void unlock_chunks(struct btrfs_root *root)
  50 {
  51         mutex_unlock(&root->fs_info->chunk_mutex);
  52 }
  53
  54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  55 {
  56         struct btrfs_device *device;
  57         WARN_ON(fs_devices->opened);
  58         while (!list_empty(&fs_devices->devices)) {
  59                 device = list_entry(fs_devices->devices.next,
  60                                     struct btrfs_device, dev_list);
  61                 list_del(&device->dev_list);
  62                 kfree(device->name);
  63                 kfree(device);
  64         }
  65         kfree(fs_devices);
  66 }
  67
  68 int btrfs_cleanup_fs_uuids(void)
  69 {
  70         struct btrfs_fs_devices *fs_devices;
  71
  72         while (!list_empty(&fs_uuids)) {
  73                 fs_devices = list_entry(fs_uuids.next,
  74                                         struct btrfs_fs_devices, list);
  75                 list_del(&fs_devices->list);
  76                 free_fs_devices(fs_devices);
  77         }
  78         return 0;
  79 }
  80
  81 static noinline struct btrfs_device *__find_device(struct list_head *head,
  82                                                    u64 devid, u8 *uuid)
  83 {
  84         struct btrfs_device *dev;
  85
  86         list_for_each_entry(dev, head, dev_list) {
  87                 if (dev->devid == devid &&
  88                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  89                         return dev;
  90                 }
  91         }
  92         return NULL;
  93 }
  94
  95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
  96 {
  97         struct btrfs_fs_devices *fs_devices;
  98
  99         list_for_each_entry(fs_devices, &fs_uuids, list) {
 100                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 101                         return fs_devices;
 102         }
 103         return NULL;
 104 }
 105
 106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 107                         struct bio *head, struct bio *tail)
 108 {
 109
 110         struct bio *old_head;
 111
 112         old_head = pending_bios->head;
 113         pending_bios->head = head;
 114         if (pending_bios->tail)
 115                 tail->bi_next = old_head;
 116         else
 117                 pending_bios->tail = tail;
 118 }
 119
 120 /*
 121  * we try to collect pending bios for a device so we don't get a large
 122  * number of procs sending bios down to the same device.  This greatly
 123  * improves the schedulers ability to collect and merge the bios.
 124  *
 125  * But, it also turns into a long list of bios to process and that is sure
 126  * to eventually make the worker thread block.  The solution here is to
 127  * make some progress and then put this work struct back at the end of
 128  * the list if the block device is congested.  This way, multiple devices
 129  * can make progress from a single worker thread.
 130  */
 131 static noinline int run_scheduled_bios(struct btrfs_device *device)
 132 {
 133         struct bio *pending;
 134         struct backing_dev_info *bdi;
 135         struct btrfs_fs_info *fs_info;
 136         struct btrfs_pending_bios *pending_bios;
 137         struct bio *tail;
 138         struct bio *cur;
 139         int again = 0;
 140         unsigned long num_run;
 141         unsigned long batch_run = 0;
 142         unsigned long limit;
 143         unsigned long last_waited = 0;
 144         int force_reg = 0;
 145         int sync_pending = 0;
 146         struct blk_plug plug;
 147
 148         /*
 149          * this function runs all the bios we've collected for
 150          * a particular device.  We don't want to wander off to
 151          * another device without first sending all of these down.
 152          * So, setup a plug here and finish it off before we return
 153          */
 154         blk_start_plug(&plug);
 155
 156         bdi = blk_get_backing_dev_info(device->bdev);
 157         fs_info = device->dev_root->fs_info;
 158         limit = btrfs_async_submit_limit(fs_info);
 159         limit = limit * 2 / 3;
 160
 161 loop:
 162         spin_lock(&device->io_lock);
 163
 164 loop_lock:
 165         num_run = 0;
 166
 167         /* take all the bios off the list at once and process them
 168          * later on (without the lock held).  But, remember the
 169          * tail and other pointers so the bios can be properly reinserted
 170          * into the list if we hit congestion
 171          */
 172         if (!force_reg && device->pending_sync_bios.head) {
 173                 pending_bios = &device->pending_sync_bios;
 174                 force_reg = 1;
 175         } else {
 176                 pending_bios = &device->pending_bios;
 177                 force_reg = 0;
 178         }
 179
 180         pending = pending_bios->head;
 181         tail = pending_bios->tail;
 182         WARN_ON(pending && !tail);
 183
 184         /*
 185          * if pending was null this time around, no bios need processing
 186          * at all and we can stop.  Otherwise it'll loop back up again
 187          * and do an additional check so no bios are missed.
 188          *
 189          * device->running_pending is used to synchronize with the
 190          * schedule_bio code.
 191          */
 192         if (device->pending_sync_bios.head == NULL &&
 193             device->pending_bios.head == NULL) {
 194                 again = 0;
 195                 device->running_pending = 0;
 196         } else {
 197                 again = 1;
 198                 device->running_pending = 1;
 199         }
 200
 201         pending_bios->head = NULL;
 202         pending_bios->tail = NULL;
 203
 204         spin_unlock(&device->io_lock);
 205
 206         while (pending) {
 207
 208                 rmb();
 209                 /* we want to work on both lists, but do more bios on the
 210                  * sync list than the regular list
 211                  */
 212                 if ((num_run > 32 &&
 213                     pending_bios != &device->pending_sync_bios &&
 214                     device->pending_sync_bios.head) ||
 215                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 216                     device->pending_bios.head)) {
 217                         spin_lock(&device->io_lock);
 218                         requeue_list(pending_bios, pending, tail);
 219                         goto loop_lock;
 220                 }
 221
 222                 cur = pending;
 223                 pending = pending->bi_next;
 224                 cur->bi_next = NULL;
 225                 atomic_dec(&fs_info->nr_async_bios);
 226
 227                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 228                     waitqueue_active(&fs_info->async_submit_wait))
 229                         wake_up(&fs_info->async_submit_wait);
 230
 231                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 232
 233                 /*
 234                  * if we're doing the sync list, record that our
 235                  * plug has some sync requests on it
 236                  *
 237                  * If we're doing the regular list and there are
 238                  * sync requests sitting around, unplug before
 239                  * we add more
 240                  */
 241                 if (pending_bios == &device->pending_sync_bios) {
 242                         sync_pending = 1;
 243                 } else if (sync_pending) {
 244                         blk_finish_plug(&plug);
 245                         blk_start_plug(&plug);
 246                         sync_pending = 0;
 247                 }
 248
 249                 submit_bio(cur->bi_rw, cur);
 250                 num_run++;
 251                 batch_run++;
 252                 if (need_resched())
 253                         cond_resched();
 254
 255                 /*
 256                  * we made progress, there is more work to do and the bdi
 257                  * is now congested.  Back off and let other work structs
 258                  * run instead
 259                  */
 260                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 261                     fs_info->fs_devices->open_devices > 1) {
 262                         struct io_context *ioc;
 263
 264                         ioc = current->io_context;
 265
 266                         /*
 267                          * the main goal here is that we don't want to
 268                          * block if we're going to be able to submit
 269                          * more requests without blocking.
 270                          *
 271                          * This code does two great things, it pokes into
 272                          * the elevator code from a filesystem _and_
 273                          * it makes assumptions about how batching works.
 274                          */
 275                         if (ioc && ioc->nr_batch_requests > 0 &&
 276                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 277                             (last_waited == 0 ||
 278                              ioc->last_waited == last_waited)) {
 279                                 /*
 280                                  * we want to go through our batch of
 281                                  * requests and stop.  So, we copy out
 282                                  * the ioc->last_waited time and test
 283                                  * against it before looping
 284                                  */
 285                                 last_waited = ioc->last_waited;
 286                                 if (need_resched())
 287                                         cond_resched();
 288                                 continue;
 289                         }
 290                         spin_lock(&device->io_lock);
 291                         requeue_list(pending_bios, pending, tail);
 292                         device->running_pending = 1;
 293
 294                         spin_unlock(&device->io_lock);
 295                         btrfs_requeue_work(&device->work);
 296                         goto done;
 297                 }
 298         }
 299
 300         cond_resched();
 301         if (again)
 302                 goto loop;
 303
 304         spin_lock(&device->io_lock);
 305         if (device->pending_bios.head || device->pending_sync_bios.head)
 306                 goto loop_lock;
 307         spin_unlock(&device->io_lock);
 308
 309 done:
 310         blk_finish_plug(&plug);
 311         return 0;
 312 }
 313
 314 static void pending_bios_fn(struct btrfs_work *work)
 315 {
 316         struct btrfs_device *device;
 317
 318         device = container_of(work, struct btrfs_device, work);
 319         run_scheduled_bios(device);
 320 }
 321
 322 static noinline int device_list_add(const char *path,
 323                            struct btrfs_super_block *disk_super,
 324                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 325 {
 326         struct btrfs_device *device;
 327         struct btrfs_fs_devices *fs_devices;
 328         u64 found_transid = btrfs_super_generation(disk_super);
 329         char *name;
 330
 331         fs_devices = find_fsid(disk_super->fsid);
 332         if (!fs_devices) {
 333                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 334                 if (!fs_devices)
 335                         return -ENOMEM;
 336                 INIT_LIST_HEAD(&fs_devices->devices);
 337                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 338                 list_add(&fs_devices->list, &fs_uuids);
 339                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 340                 fs_devices->latest_devid = devid;
 341                 fs_devices->latest_trans = found_transid;
 342                 mutex_init(&fs_devices->device_list_mutex);
 343                 device = NULL;
 344         } else {
 345                 device = __find_device(&fs_devices->devices, devid,
 346                                        disk_super->dev_item.uuid);
 347         }
 348         if (!device) {
 349                 if (fs_devices->opened)
 350                         return -EBUSY;
 351
 352                 device = kzalloc(sizeof(*device), GFP_NOFS);
 353                 if (!device) {
 354                         /* we can safely leave the fs_devices entry around */
 355                         return -ENOMEM;
 356                 }
 357                 device->devid = devid;
 358                 device->work.func = pending_bios_fn;
 359                 memcpy(device->uuid, disk_super->dev_item.uuid,
 360                        BTRFS_UUID_SIZE);
 361                 spin_lock_init(&device->io_lock);
 362                 device->name = kstrdup(path, GFP_NOFS);
 363                 if (!device->name) {
 364                         kfree(device);
 365                         return -ENOMEM;
 366                 }
 367                 INIT_LIST_HEAD(&device->dev_alloc_list);
 368
 369                 /* init readahead state */
 370                 spin_lock_init(&device->reada_lock);
 371                 device->reada_curr_zone = NULL;
 372                 atomic_set(&device->reada_in_flight, 0);
 373                 device->reada_next = 0;
 374                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 375                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 376
 377                 mutex_lock(&fs_devices->device_list_mutex);
 378                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 379                 mutex_unlock(&fs_devices->device_list_mutex);
 380
 381                 device->fs_devices = fs_devices;
 382                 fs_devices->num_devices++;
 383         } else if (!device->name || strcmp(device->name, path)) {
 384                 name = kstrdup(path, GFP_NOFS);
 385                 if (!name)
 386                         return -ENOMEM;
 387                 kfree(device->name);
 388                 device->name = name;
 389                 if (device->missing) {
 390                         fs_devices->missing_devices--;
 391                         device->missing = 0;
 392                 }
 393         }
 394
 395         if (found_transid > fs_devices->latest_trans) {
 396                 fs_devices->latest_devid = devid;
 397                 fs_devices->latest_trans = found_transid;
 398         }
 399         *fs_devices_ret = fs_devices;
 400         return 0;
 401 }
 402
 403 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 404 {
 405         struct btrfs_fs_devices *fs_devices;
 406         struct btrfs_device *device;
 407         struct btrfs_device *orig_dev;
 408
 409         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 410         if (!fs_devices)
 411                 return ERR_PTR(-ENOMEM);
 412
 413         INIT_LIST_HEAD(&fs_devices->devices);
 414         INIT_LIST_HEAD(&fs_devices->alloc_list);
 415         INIT_LIST_HEAD(&fs_devices->list);
 416         mutex_init(&fs_devices->device_list_mutex);
 417         fs_devices->latest_devid = orig->latest_devid;
 418         fs_devices->latest_trans = orig->latest_trans;
 419         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 420
 421         /* We have held the volume lock, it is safe to get the devices. */
 422         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 423                 device = kzalloc(sizeof(*device), GFP_NOFS);
 424                 if (!device)
 425                         goto error;
 426
 427                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 428                 if (!device->name) {
 429                         kfree(device);
 430                         goto error;
 431                 }
 432
 433                 device->devid = orig_dev->devid;
 434                 device->work.func = pending_bios_fn;
 435                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 436                 spin_lock_init(&device->io_lock);
 437                 INIT_LIST_HEAD(&device->dev_list);
 438                 INIT_LIST_HEAD(&device->dev_alloc_list);
 439
 440                 list_add(&device->dev_list, &fs_devices->devices);
 441                 device->fs_devices = fs_devices;
 442                 fs_devices->num_devices++;
 443         }
 444         return fs_devices;
 445 error:
 446         free_fs_devices(fs_devices);
 447         return ERR_PTR(-ENOMEM);
 448 }
 449
 450 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 451 {
 452         struct btrfs_device *device, *next;
 453
 454         mutex_lock(&uuid_mutex);
 455 again:
 456         /* This is the initialized path, it is safe to release the devices. */
 457         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 458                 if (device->in_fs_metadata)
 459                         continue;
 460
 461                 if (device->bdev) {
 462                         blkdev_put(device->bdev, device->mode);
 463                         device->bdev = NULL;
 464                         fs_devices->open_devices--;
 465                 }
 466                 if (device->writeable) {
 467                         list_del_init(&device->dev_alloc_list);
 468                         device->writeable = 0;
 469                         fs_devices->rw_devices--;
 470                 }
 471                 list_del_init(&device->dev_list);
 472                 fs_devices->num_devices--;
 473                 kfree(device->name);
 474                 kfree(device);
 475         }
 476
 477         if (fs_devices->seed) {
 478                 fs_devices = fs_devices->seed;
 479                 goto again;
 480         }
 481
 482         mutex_unlock(&uuid_mutex);
 483         return 0;
 484 }
 485
 486 static void __free_device(struct work_struct *work)
 487 {
 488         struct btrfs_device *device;
 489
 490         device = container_of(work, struct btrfs_device, rcu_work);
 491
 492         if (device->bdev)
 493                 blkdev_put(device->bdev, device->mode);
 494
 495         kfree(device->name);
 496         kfree(device);
 497 }
 498
 499 static void free_device(struct rcu_head *head)
 500 {
 501         struct btrfs_device *device;
 502
 503         device = container_of(head, struct btrfs_device, rcu);
 504
 505         INIT_WORK(&device->rcu_work, __free_device);
 506         schedule_work(&device->rcu_work);
 507 }
 508
 509 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 510 {
 511         struct btrfs_device *device;
 512
 513         if (--fs_devices->opened > 0)
 514                 return 0;
 515
 516         mutex_lock(&fs_devices->device_list_mutex);
 517         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 518                 struct btrfs_device *new_device;
 519
 520                 if (device->bdev)
 521                         fs_devices->open_devices--;
 522
 523                 if (device->writeable) {
 524                         list_del_init(&device->dev_alloc_list);
 525                         fs_devices->rw_devices--;
 526                 }
 527
 528                 if (device->can_discard)
 529                         fs_devices->num_can_discard--;
 530
 531                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 532                 BUG_ON(!new_device);
 533                 memcpy(new_device, device, sizeof(*new_device));
 534                 new_device->name = kstrdup(device->name, GFP_NOFS);
 535                 BUG_ON(device->name && !new_device->name);
 536                 new_device->bdev = NULL;
 537                 new_device->writeable = 0;
 538                 new_device->in_fs_metadata = 0;
 539                 new_device->can_discard = 0;
 540                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 541
 542                 call_rcu(&device->rcu, free_device);
 543         }
 544         mutex_unlock(&fs_devices->device_list_mutex);
 545
 546         WARN_ON(fs_devices->open_devices);
 547         WARN_ON(fs_devices->rw_devices);
 548         fs_devices->opened = 0;
 549         fs_devices->seeding = 0;
 550
 551         return 0;
 552 }
 553
 554 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 555 {
 556         struct btrfs_fs_devices *seed_devices = NULL;
 557         int ret;
 558
 559         mutex_lock(&uuid_mutex);
 560         ret = __btrfs_close_devices(fs_devices);
 561         if (!fs_devices->opened) {
 562                 seed_devices = fs_devices->seed;
 563                 fs_devices->seed = NULL;
 564         }
 565         mutex_unlock(&uuid_mutex);
 566
 567         while (seed_devices) {
 568                 fs_devices = seed_devices;
 569                 seed_devices = fs_devices->seed;
 570                 __btrfs_close_devices(fs_devices);
 571                 free_fs_devices(fs_devices);
 572         }
 573         return ret;
 574 }
 575
 576 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 577                                 fmode_t flags, void *holder)
 578 {
 579         struct request_queue *q;
 580         struct block_device *bdev;
 581         struct list_head *head = &fs_devices->devices;
 582         struct btrfs_device *device;
 583         struct block_device *latest_bdev = NULL;
 584         struct buffer_head *bh;
 585         struct btrfs_super_block *disk_super;
 586         u64 latest_devid = 0;
 587         u64 latest_transid = 0;
 588         u64 devid;
 589         int seeding = 1;
 590         int ret = 0;
 591
 592         flags |= FMODE_EXCL;
 593
 594         list_for_each_entry(device, head, dev_list) {
 595                 if (device->bdev)
 596                         continue;
 597                 if (!device->name)
 598                         continue;
 599
 600                 bdev = blkdev_get_by_path(device->name, flags, holder);
 601                 if (IS_ERR(bdev)) {
 602                         printk(KERN_INFO "open %s failed\n", device->name);
 603                         goto error;
 604                 }
 605                 set_blocksize(bdev, 4096);
 606
 607                 bh = btrfs_read_dev_super(bdev);
 608                 if (!bh) {
 609                         ret = -EINVAL;
 610                         goto error_close;
 611                 }
 612
 613                 disk_super = (struct btrfs_super_block *)bh->b_data;
 614                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 615                 if (devid != device->devid)
 616                         goto error_brelse;
 617
 618                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 619                            BTRFS_UUID_SIZE))
 620                         goto error_brelse;
 621
 622                 device->generation = btrfs_super_generation(disk_super);
 623                 if (!latest_transid || device->generation > latest_transid) {
 624                         latest_devid = devid;
 625                         latest_transid = device->generation;
 626                         latest_bdev = bdev;
 627                 }
 628
 629                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 630                         device->writeable = 0;
 631                 } else {
 632                         device->writeable = !bdev_read_only(bdev);
 633                         seeding = 0;
 634                 }
 635
 636                 q = bdev_get_queue(bdev);
 637                 if (blk_queue_discard(q)) {
 638                         device->can_discard = 1;
 639                         fs_devices->num_can_discard++;
 640                 }
 641
 642                 device->bdev = bdev;
 643                 device->in_fs_metadata = 0;
 644                 device->mode = flags;
 645
 646                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 647                         fs_devices->rotating = 1;
 648
 649                 fs_devices->open_devices++;
 650                 if (device->writeable) {
 651                         fs_devices->rw_devices++;
 652                         list_add(&device->dev_alloc_list,
 653                                  &fs_devices->alloc_list);
 654                 }
 655                 brelse(bh);
 656                 continue;
 657
 658 error_brelse:
 659                 brelse(bh);
 660 error_close:
 661                 blkdev_put(bdev, flags);
 662 error:
 663                 continue;
 664         }
 665         if (fs_devices->open_devices == 0) {
 666                 ret = -EIO;
 667                 goto out;
 668         }
 669         fs_devices->seeding = seeding;
 670         fs_devices->opened = 1;
 671         fs_devices->latest_bdev = latest_bdev;
 672         fs_devices->latest_devid = latest_devid;
 673         fs_devices->latest_trans = latest_transid;
 674         fs_devices->total_rw_bytes = 0;
 675 out:
 676         return ret;
 677 }
 678
 679 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 680                        fmode_t flags, void *holder)
 681 {
 682         int ret;
 683
 684         mutex_lock(&uuid_mutex);
 685         if (fs_devices->opened) {
 686                 fs_devices->opened++;
 687                 ret = 0;
 688         } else {
 689                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 690         }
 691         mutex_unlock(&uuid_mutex);
 692         return ret;
 693 }
 694
 695 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 696                           struct btrfs_fs_devices **fs_devices_ret)
 697 {
 698         struct btrfs_super_block *disk_super;
 699         struct block_device *bdev;
 700         struct buffer_head *bh;
 701         int ret;
 702         u64 devid;
 703         u64 transid;
 704
 705         mutex_lock(&uuid_mutex);
 706
 707         flags |= FMODE_EXCL;
 708         bdev = blkdev_get_by_path(path, flags, holder);
 709
 710         if (IS_ERR(bdev)) {
 711                 ret = PTR_ERR(bdev);
 712                 goto error;
 713         }
 714
 715         ret = set_blocksize(bdev, 4096);
 716         if (ret)
 717                 goto error_close;
 718         bh = btrfs_read_dev_super(bdev);
 719         if (!bh) {
 720                 ret = -EINVAL;
 721                 goto error_close;
 722         }
 723         disk_super = (struct btrfs_super_block *)bh->b_data;
 724         devid = btrfs_stack_device_id(&disk_super->dev_item);
 725         transid = btrfs_super_generation(disk_super);
 726         if (disk_super->label[0])
 727                 printk(KERN_INFO "device label %s ", disk_super->label);
 728         else
 729                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 730         printk(KERN_CONT "devid %llu transid %llu %s\n",
 731                (unsigned long long)devid, (unsigned long long)transid, path);
 732         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 733
 734         brelse(bh);
 735 error_close:
 736         blkdev_put(bdev, flags);
 737 error:
 738         mutex_unlock(&uuid_mutex);
 739         return ret;
 740 }
 741
 742 /* helper to account the used device space in the range */
 743 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 744                                    u64 end, u64 *length)
 745 {
 746         struct btrfs_key key;
 747         struct btrfs_root *root = device->dev_root;
 748         struct btrfs_dev_extent *dev_extent;
 749         struct btrfs_path *path;
 750         u64 extent_end;
 751         int ret;
 752         int slot;
 753         struct extent_buffer *l;
 754
 755         *length = 0;
 756
 757         if (start >= device->total_bytes)
 758                 return 0;
 759
 760         path = btrfs_alloc_path();
 761         if (!path)
 762                 return -ENOMEM;
 763         path->reada = 2;
 764
 765         key.objectid = device->devid;
 766         key.offset = start;
 767         key.type = BTRFS_DEV_EXTENT_KEY;
 768
 769         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 770         if (ret < 0)
 771                 goto out;
 772         if (ret > 0) {
 773                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 774                 if (ret < 0)
 775                         goto out;
 776         }
 777
 778         while (1) {
 779                 l = path->nodes[0];
 780                 slot = path->slots[0];
 781                 if (slot >= btrfs_header_nritems(l)) {
 782                         ret = btrfs_next_leaf(root, path);
 783                         if (ret == 0)
 784                                 continue;
 785                         if (ret < 0)
 786                                 goto out;
 787
 788                         break;
 789                 }
 790                 btrfs_item_key_to_cpu(l, &key, slot);
 791
 792                 if (key.objectid < device->devid)
 793                         goto next;
 794
 795                 if (key.objectid > device->devid)
 796                         break;
 797
 798                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 799                         goto next;
 800
 801                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 802                 extent_end = key.offset + btrfs_dev_extent_length(l,
 803                                                                   dev_extent);
 804                 if (key.offset <= start && extent_end > end) {
 805                         *length = end - start + 1;
 806                         break;
 807                 } else if (key.offset <= start && extent_end > start)
 808                         *length += extent_end - start;
 809                 else if (key.offset > start && extent_end <= end)
 810                         *length += extent_end - key.offset;
 811                 else if (key.offset > start && key.offset <= end) {
 812                         *length += end - key.offset + 1;
 813                         break;
 814                 } else if (key.offset > end)
 815                         break;
 816
 817 next:
 818                 path->slots[0]++;
 819         }
 820         ret = 0;
 821 out:
 822         btrfs_free_path(path);
 823         return ret;
 824 }
 825
 826 /*
 827  * find_free_dev_extent - find free space in the specified device
 828  * @trans:      transaction handler
 829  * @device:     the device which we search the free space in
 830  * @num_bytes:  the size of the free space that we need
 831  * @start:      store the start of the free space.
 832  * @len:        the size of the free space. that we find, or the size of the max
 833  *              free space if we don't find suitable free space
 834  *
 835  * this uses a pretty simple search, the expectation is that it is
 836  * called very infrequently and that a given device has a small number
 837  * of extents
 838  *
 839  * @start is used to store the start of the free space if we find. But if we
 840  * don't find suitable free space, it will be used to store the start position
 841  * of the max free space.
 842  *
 843  * @len is used to store the size of the free space that we find.
 844  * But if we don't find suitable free space, it is used to store the size of
 845  * the max free space.
 846  */
 847 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 848                          struct btrfs_device *device, u64 num_bytes,
 849                          u64 *start, u64 *len)
 850 {
 851         struct btrfs_key key;
 852         struct btrfs_root *root = device->dev_root;
 853         struct btrfs_dev_extent *dev_extent;
 854         struct btrfs_path *path;
 855         u64 hole_size;
 856         u64 max_hole_start;
 857         u64 max_hole_size;
 858         u64 extent_end;
 859         u64 search_start;
 860         u64 search_end = device->total_bytes;
 861         int ret;
 862         int slot;
 863         struct extent_buffer *l;
 864
 865         /* FIXME use last free of some kind */
 866
 867         /* we don't want to overwrite the superblock on the drive,
 868          * so we make sure to start at an offset of at least 1MB
 869          */
 870         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 871
 872         max_hole_start = search_start;
 873         max_hole_size = 0;
 874         hole_size = 0;
 875
 876         if (search_start >= search_end) {
 877                 ret = -ENOSPC;
 878                 goto error;
 879         }
 880
 881         path = btrfs_alloc_path();
 882         if (!path) {
 883                 ret = -ENOMEM;
 884                 goto error;
 885         }
 886         path->reada = 2;
 887
 888         key.objectid = device->devid;
 889         key.offset = search_start;
 890         key.type = BTRFS_DEV_EXTENT_KEY;
 891
 892         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 893         if (ret < 0)
 894                 goto out;
 895         if (ret > 0) {
 896                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 897                 if (ret < 0)
 898                         goto out;
 899         }
 900
 901         while (1) {
 902                 l = path->nodes[0];
 903                 slot = path->slots[0];
 904                 if (slot >= btrfs_header_nritems(l)) {
 905                         ret = btrfs_next_leaf(root, path);
 906                         if (ret == 0)
 907                                 continue;
 908                         if (ret < 0)
 909                                 goto out;
 910
 911                         break;
 912                 }
 913                 btrfs_item_key_to_cpu(l, &key, slot);
 914
 915                 if (key.objectid < device->devid)
 916                         goto next;
 917
 918                 if (key.objectid > device->devid)
 919                         break;
 920
 921                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 922                         goto next;
 923
 924                 if (key.offset > search_start) {
 925                         hole_size = key.offset - search_start;
 926
 927                         if (hole_size > max_hole_size) {
 928                                 max_hole_start = search_start;
 929                                 max_hole_size = hole_size;
 930                         }
 931
 932                         /*
 933                          * If this free space is greater than which we need,
 934                          * it must be the max free space that we have found
 935                          * until now, so max_hole_start must point to the start
 936                          * of this free space and the length of this free space
 937                          * is stored in max_hole_size. Thus, we return
 938                          * max_hole_start and max_hole_size and go back to the
 939                          * caller.
 940                          */
 941                         if (hole_size >= num_bytes) {
 942                                 ret = 0;
 943                                 goto out;
 944                         }
 945                 }
 946
 947                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 948                 extent_end = key.offset + btrfs_dev_extent_length(l,
 949                                                                   dev_extent);
 950                 if (extent_end > search_start)
 951                         search_start = extent_end;
 952 next:
 953                 path->slots[0]++;
 954                 cond_resched();
 955         }
 956
 957         /*
 958          * At this point, search_start should be the end of
 959          * allocated dev extents, and when shrinking the device,
 960          * search_end may be smaller than search_start.
 961          */
 962         if (search_end > search_start)
 963                 hole_size = search_end - search_start;
 964
 965         if (hole_size > max_hole_size) {
 966                 max_hole_start = search_start;
 967                 max_hole_size = hole_size;
 968         }
 969
 970         /* See above. */
 971         if (hole_size < num_bytes)
 972                 ret = -ENOSPC;
 973         else
 974                 ret = 0;
 975
 976 out:
 977         btrfs_free_path(path);
 978 error:
 979         *start = max_hole_start;
 980         if (len)
 981                 *len = max_hole_size;
 982         return ret;
 983 }
 984
 985 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 986                           struct btrfs_device *device,
 987                           u64 start)
 988 {
 989         int ret;
 990         struct btrfs_path *path;
 991         struct btrfs_root *root = device->dev_root;
 992         struct btrfs_key key;
 993         struct btrfs_key found_key;
 994         struct extent_buffer *leaf = NULL;
 995         struct btrfs_dev_extent *extent = NULL;
 996
 997         path = btrfs_alloc_path();
 998         if (!path)
 999                 return -ENOMEM;
1000
1001         key.objectid = device->devid;
1002         key.offset = start;
1003         key.type = BTRFS_DEV_EXTENT_KEY;
1004
1005         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1006         if (ret > 0) {
1007                 ret = btrfs_previous_item(root, path, key.objectid,
1008                                           BTRFS_DEV_EXTENT_KEY);
1009                 if (ret)
1010                         goto out;
1011                 leaf = path->nodes[0];
1012                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1013                 extent = btrfs_item_ptr(leaf, path->slots[0],
1014                                         struct btrfs_dev_extent);
1015                 BUG_ON(found_key.offset > start || found_key.offset +
1016                        btrfs_dev_extent_length(leaf, extent) < start);
1017         } else if (ret == 0) {
1018                 leaf = path->nodes[0];
1019                 extent = btrfs_item_ptr(leaf, path->slots[0],
1020                                         struct btrfs_dev_extent);
1021         }
1022         BUG_ON(ret);
1023
1024         if (device->bytes_used > 0)
1025                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
1026         ret = btrfs_del_item(trans, root, path);
1027
1028 out:
1029         btrfs_free_path(path);
1030         return ret;
1031 }
1032
1033 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1034                            struct btrfs_device *device,
1035                            u64 chunk_tree, u64 chunk_objectid,
1036                            u64 chunk_offset, u64 start, u64 num_bytes)
1037 {
1038         int ret;
1039         struct btrfs_path *path;
1040         struct btrfs_root *root = device->dev_root;
1041         struct btrfs_dev_extent *extent;
1042         struct extent_buffer *leaf;
1043         struct btrfs_key key;
1044
1045         WARN_ON(!device->in_fs_metadata);
1046         path = btrfs_alloc_path();
1047         if (!path)
1048                 return -ENOMEM;
1049
1050         key.objectid = device->devid;
1051         key.offset = start;
1052         key.type = BTRFS_DEV_EXTENT_KEY;
1053         ret = btrfs_insert_empty_item(trans, root, path, &key,
1054                                       sizeof(*extent));
1055         BUG_ON(ret);
1056
1057         leaf = path->nodes[0];
1058         extent = btrfs_item_ptr(leaf, path->slots[0],
1059                                 struct btrfs_dev_extent);
1060         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1061         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1062         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1063
1064         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1065                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1066                     BTRFS_UUID_SIZE);
1067
1068         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1069         btrfs_mark_buffer_dirty(leaf);
1070         btrfs_free_path(path);
1071         return ret;
1072 }
1073
1074 static noinline int find_next_chunk(struct btrfs_root *root,
1075                                     u64 objectid, u64 *offset)
1076 {
1077         struct btrfs_path *path;
1078         int ret;
1079         struct btrfs_key key;
1080         struct btrfs_chunk *chunk;
1081         struct btrfs_key found_key;
1082
1083         path = btrfs_alloc_path();
1084         if (!path)
1085                 return -ENOMEM;
1086
1087         key.objectid = objectid;
1088         key.offset = (u64)-1;
1089         key.type = BTRFS_CHUNK_ITEM_KEY;
1090
1091         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1092         if (ret < 0)
1093                 goto error;
1094
1095         BUG_ON(ret == 0);
1096
1097         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1098         if (ret) {
1099                 *offset = 0;
1100         } else {
1101                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1102                                       path->slots[0]);
1103                 if (found_key.objectid != objectid)
1104                         *offset = 0;
1105                 else {
1106                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1107                                                struct btrfs_chunk);
1108                         *offset = found_key.offset +
1109                                 btrfs_chunk_length(path->nodes[0], chunk);
1110                 }
1111         }
1112         ret = 0;
1113 error:
1114         btrfs_free_path(path);
1115         return ret;
1116 }
1117
1118 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1119 {
1120         int ret;
1121         struct btrfs_key key;
1122         struct btrfs_key found_key;
1123         struct btrfs_path *path;
1124
1125         root = root->fs_info->chunk_root;
1126
1127         path = btrfs_alloc_path();
1128         if (!path)
1129                 return -ENOMEM;
1130
1131         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1132         key.type = BTRFS_DEV_ITEM_KEY;
1133         key.offset = (u64)-1;
1134
1135         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1136         if (ret < 0)
1137                 goto error;
1138
1139         BUG_ON(ret == 0);
1140
1141         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1142                                   BTRFS_DEV_ITEM_KEY);
1143         if (ret) {
1144                 *objectid = 1;
1145         } else {
1146                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1147                                       path->slots[0]);
1148                 *objectid = found_key.offset + 1;
1149         }
1150         ret = 0;
1151 error:
1152         btrfs_free_path(path);
1153         return ret;
1154 }
1155
1156 /*
1157  * the device information is stored in the chunk root
1158  * the btrfs_device struct should be fully filled in
1159  */
1160 int btrfs_add_device(struct btrfs_trans_handle *trans,
1161                      struct btrfs_root *root,
1162                      struct btrfs_device *device)
1163 {
1164         int ret;
1165         struct btrfs_path *path;
1166         struct btrfs_dev_item *dev_item;
1167         struct extent_buffer *leaf;
1168         struct btrfs_key key;
1169         unsigned long ptr;
1170
1171         root = root->fs_info->chunk_root;
1172
1173         path = btrfs_alloc_path();
1174         if (!path)
1175                 return -ENOMEM;
1176
1177         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1178         key.type = BTRFS_DEV_ITEM_KEY;
1179         key.offset = device->devid;
1180
1181         ret = btrfs_insert_empty_item(trans, root, path, &key,
1182                                       sizeof(*dev_item));
1183         if (ret)
1184                 goto out;
1185
1186         leaf = path->nodes[0];
1187         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1188
1189         btrfs_set_device_id(leaf, dev_item, device->devid);
1190         btrfs_set_device_generation(leaf, dev_item, 0);
1191         btrfs_set_device_type(leaf, dev_item, device->type);
1192         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1193         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1194         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1195         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1196         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1197         btrfs_set_device_group(leaf, dev_item, 0);
1198         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1199         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1200         btrfs_set_device_start_offset(leaf, dev_item, 0);
1201
1202         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1203         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1204         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1205         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1206         btrfs_mark_buffer_dirty(leaf);
1207
1208         ret = 0;
1209 out:
1210         btrfs_free_path(path);
1211         return ret;
1212 }
1213
1214 static int btrfs_rm_dev_item(struct btrfs_root *root,
1215                              struct btrfs_device *device)
1216 {
1217         int ret;
1218         struct btrfs_path *path;
1219         struct btrfs_key key;
1220         struct btrfs_trans_handle *trans;
1221
1222         root = root->fs_info->chunk_root;
1223
1224         path = btrfs_alloc_path();
1225         if (!path)
1226                 return -ENOMEM;
1227
1228         trans = btrfs_start_transaction(root, 0);
1229         if (IS_ERR(trans)) {
1230                 btrfs_free_path(path);
1231                 return PTR_ERR(trans);
1232         }
1233         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1234         key.type = BTRFS_DEV_ITEM_KEY;
1235         key.offset = device->devid;
1236         lock_chunks(root);
1237
1238         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1239         if (ret < 0)
1240                 goto out;
1241
1242         if (ret > 0) {
1243                 ret = -ENOENT;
1244                 goto out;
1245         }
1246
1247         ret = btrfs_del_item(trans, root, path);
1248         if (ret)
1249                 goto out;
1250 out:
1251         btrfs_free_path(path);
1252         unlock_chunks(root);
1253         btrfs_commit_transaction(trans, root);
1254         return ret;
1255 }
1256
1257 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1258 {
1259         struct btrfs_device *device;
1260         struct btrfs_device *next_device;
1261         struct block_device *bdev;
1262         struct buffer_head *bh = NULL;
1263         struct btrfs_super_block *disk_super;
1264         struct btrfs_fs_devices *cur_devices;
1265         u64 all_avail;
1266         u64 devid;
1267         u64 num_devices;
1268         u8 *dev_uuid;
1269         int ret = 0;
1270         bool clear_super = false;
1271
1272         mutex_lock(&uuid_mutex);
1273         mutex_lock(&root->fs_info->volume_mutex);
1274
1275         all_avail = root->fs_info->avail_data_alloc_bits |
1276                 root->fs_info->avail_system_alloc_bits |
1277                 root->fs_info->avail_metadata_alloc_bits;
1278
1279         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1280             root->fs_info->fs_devices->num_devices <= 4) {
1281                 printk(KERN_ERR "btrfs: unable to go below four devices "
1282                        "on raid10\n");
1283                 ret = -EINVAL;
1284                 goto out;
1285         }
1286
1287         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1288             root->fs_info->fs_devices->num_devices <= 2) {
1289                 printk(KERN_ERR "btrfs: unable to go below two "
1290                        "devices on raid1\n");
1291                 ret = -EINVAL;
1292                 goto out;
1293         }
1294
1295         if (strcmp(device_path, "missing") == 0) {
1296                 struct list_head *devices;
1297                 struct btrfs_device *tmp;
1298
1299                 device = NULL;
1300                 devices = &root->fs_info->fs_devices->devices;
1301                 /*
1302                  * It is safe to read the devices since the volume_mutex
1303                  * is held.
1304                  */
1305                 list_for_each_entry(tmp, devices, dev_list) {
1306                         if (tmp->in_fs_metadata && !tmp->bdev) {
1307                                 device = tmp;
1308                                 break;
1309                         }
1310                 }
1311                 bdev = NULL;
1312                 bh = NULL;
1313                 disk_super = NULL;
1314                 if (!device) {
1315                         printk(KERN_ERR "btrfs: no missing devices found to "
1316                                "remove\n");
1317                         goto out;
1318                 }
1319         } else {
1320                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1321                                           root->fs_info->bdev_holder);
1322                 if (IS_ERR(bdev)) {
1323                         ret = PTR_ERR(bdev);
1324                         goto out;
1325                 }
1326
1327                 set_blocksize(bdev, 4096);
1328                 bh = btrfs_read_dev_super(bdev);
1329                 if (!bh) {
1330                         ret = -EINVAL;
1331                         goto error_close;
1332                 }
1333                 disk_super = (struct btrfs_super_block *)bh->b_data;
1334                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1335                 dev_uuid = disk_super->dev_item.uuid;
1336                 device = btrfs_find_device(root, devid, dev_uuid,
1337                                            disk_super->fsid);
1338                 if (!device) {
1339                         ret = -ENOENT;
1340                         goto error_brelse;
1341                 }
1342         }
1343
1344         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1345                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1346                        "device\n");
1347                 ret = -EINVAL;
1348                 goto error_brelse;
1349         }
1350
1351         if (device->writeable) {
1352                 lock_chunks(root);
1353                 list_del_init(&device->dev_alloc_list);
1354                 unlock_chunks(root);
1355                 root->fs_info->fs_devices->rw_devices--;
1356                 clear_super = true;
1357         }
1358
1359         ret = btrfs_shrink_device(device, 0);
1360         if (ret)
1361                 goto error_undo;
1362
1363         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1364         if (ret)
1365                 goto error_undo;
1366
1367         device->in_fs_metadata = 0;
1368         btrfs_scrub_cancel_dev(root, device);
1369
1370         /*
1371          * the device list mutex makes sure that we don't change
1372          * the device list while someone else is writing out all
1373          * the device supers.
1374          */
1375
1376         cur_devices = device->fs_devices;
1377         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1378         list_del_rcu(&device->dev_list);
1379
1380         device->fs_devices->num_devices--;
1381
1382         if (device->missing)
1383                 root->fs_info->fs_devices->missing_devices--;
1384
1385         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1386                                  struct btrfs_device, dev_list);
1387         if (device->bdev == root->fs_info->sb->s_bdev)
1388                 root->fs_info->sb->s_bdev = next_device->bdev;
1389         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1390                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1391
1392         if (device->bdev)
1393                 device->fs_devices->open_devices--;
1394
1395         call_rcu(&device->rcu, free_device);
1396         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1397
1398         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1399         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1400
1401         if (cur_devices->open_devices == 0) {
1402                 struct btrfs_fs_devices *fs_devices;
1403                 fs_devices = root->fs_info->fs_devices;
1404                 while (fs_devices) {
1405                         if (fs_devices->seed == cur_devices)
1406                                 break;
1407                         fs_devices = fs_devices->seed;
1408                 }
1409                 fs_devices->seed = cur_devices->seed;
1410                 cur_devices->seed = NULL;
1411                 lock_chunks(root);
1412                 __btrfs_close_devices(cur_devices);
1413                 unlock_chunks(root);
1414                 free_fs_devices(cur_devices);
1415         }
1416
1417         /*
1418          * at this point, the device is zero sized.  We want to
1419          * remove it from the devices list and zero out the old super
1420          */
1421         if (clear_super) {
1422                 /* make sure this device isn't detected as part of
1423                  * the FS anymore
1424                  */
1425                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1426                 set_buffer_dirty(bh);
1427                 sync_dirty_buffer(bh);
1428         }
1429
1430         ret = 0;
1431
1432 error_brelse:
1433         brelse(bh);
1434 error_close:
1435         if (bdev)
1436                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1437 out:
1438         mutex_unlock(&root->fs_info->volume_mutex);
1439         mutex_unlock(&uuid_mutex);
1440         return ret;
1441 error_undo:
1442         if (device->writeable) {
1443                 lock_chunks(root);
1444                 list_add(&device->dev_alloc_list,
1445                          &root->fs_info->fs_devices->alloc_list);
1446                 unlock_chunks(root);
1447                 root->fs_info->fs_devices->rw_devices++;
1448         }
1449         goto error_brelse;
1450 }
1451
1452 /*
1453  * does all the dirty work required for changing file system's UUID.
1454  */
1455 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1456                                 struct btrfs_root *root)
1457 {
1458         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1459         struct btrfs_fs_devices *old_devices;
1460         struct btrfs_fs_devices *seed_devices;
1461         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1462         struct btrfs_device *device;
1463         u64 super_flags;
1464
1465         BUG_ON(!mutex_is_locked(&uuid_mutex));
1466         if (!fs_devices->seeding)
1467                 return -EINVAL;
1468
1469         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1470         if (!seed_devices)
1471                 return -ENOMEM;
1472
1473         old_devices = clone_fs_devices(fs_devices);
1474         if (IS_ERR(old_devices)) {
1475                 kfree(seed_devices);
1476                 return PTR_ERR(old_devices);
1477         }
1478
1479         list_add(&old_devices->list, &fs_uuids);
1480
1481         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1482         seed_devices->opened = 1;
1483         INIT_LIST_HEAD(&seed_devices->devices);
1484         INIT_LIST_HEAD(&seed_devices->alloc_list);
1485         mutex_init(&seed_devices->device_list_mutex);
1486
1487         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1488         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1489                               synchronize_rcu);
1490         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1491
1492         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1493         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1494                 device->fs_devices = seed_devices;
1495         }
1496
1497         fs_devices->seeding = 0;
1498         fs_devices->num_devices = 0;
1499         fs_devices->open_devices = 0;
1500         fs_devices->seed = seed_devices;
1501
1502         generate_random_uuid(fs_devices->fsid);
1503         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1504         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1505         super_flags = btrfs_super_flags(disk_super) &
1506                       ~BTRFS_SUPER_FLAG_SEEDING;
1507         btrfs_set_super_flags(disk_super, super_flags);
1508
1509         return 0;
1510 }
1511
1512 /*
1513  * strore the expected generation for seed devices in device items.
1514  */
1515 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1516                                struct btrfs_root *root)
1517 {
1518         struct btrfs_path *path;
1519         struct extent_buffer *leaf;
1520         struct btrfs_dev_item *dev_item;
1521         struct btrfs_device *device;
1522         struct btrfs_key key;
1523         u8 fs_uuid[BTRFS_UUID_SIZE];
1524         u8 dev_uuid[BTRFS_UUID_SIZE];
1525         u64 devid;
1526         int ret;
1527
1528         path = btrfs_alloc_path();
1529         if (!path)
1530                 return -ENOMEM;
1531
1532         root = root->fs_info->chunk_root;
1533         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1534         key.offset = 0;
1535         key.type = BTRFS_DEV_ITEM_KEY;
1536
1537         while (1) {
1538                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1539                 if (ret < 0)
1540                         goto error;
1541
1542                 leaf = path->nodes[0];
1543 next_slot:
1544                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1545                         ret = btrfs_next_leaf(root, path);
1546                         if (ret > 0)
1547                                 break;
1548                         if (ret < 0)
1549                                 goto error;
1550                         leaf = path->nodes[0];
1551                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1552                         btrfs_release_path(path);
1553                         continue;
1554                 }
1555
1556                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1557                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1558                     key.type != BTRFS_DEV_ITEM_KEY)
1559                         break;
1560
1561                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1562                                           struct btrfs_dev_item);
1563                 devid = btrfs_device_id(leaf, dev_item);
1564                 read_extent_buffer(leaf, dev_uuid,
1565                                    (unsigned long)btrfs_device_uuid(dev_item),
1566                                    BTRFS_UUID_SIZE);
1567                 read_extent_buffer(leaf, fs_uuid,
1568                                    (unsigned long)btrfs_device_fsid(dev_item),
1569                                    BTRFS_UUID_SIZE);
1570                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1571                 BUG_ON(!device);
1572
1573                 if (device->fs_devices->seeding) {
1574                         btrfs_set_device_generation(leaf, dev_item,
1575                                                     device->generation);
1576                         btrfs_mark_buffer_dirty(leaf);
1577                 }
1578
1579                 path->slots[0]++;
1580                 goto next_slot;
1581         }
1582         ret = 0;
1583 error:
1584         btrfs_free_path(path);
1585         return ret;
1586 }
1587
1588 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1589 {
1590         struct request_queue *q;
1591         struct btrfs_trans_handle *trans;
1592         struct btrfs_device *device;
1593         struct block_device *bdev;
1594         struct list_head *devices;
1595         struct super_block *sb = root->fs_info->sb;
1596         u64 total_bytes;
1597         int seeding_dev = 0;
1598         int ret = 0;
1599
1600         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1601                 return -EINVAL;
1602
1603         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1604                                   root->fs_info->bdev_holder);
1605         if (IS_ERR(bdev))
1606                 return PTR_ERR(bdev);
1607
1608         if (root->fs_info->fs_devices->seeding) {
1609                 seeding_dev = 1;
1610                 down_write(&sb->s_umount);
1611                 mutex_lock(&uuid_mutex);
1612         }
1613
1614         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1615         mutex_lock(&root->fs_info->volume_mutex);
1616
1617         devices = &root->fs_info->fs_devices->devices;
1618         /*
1619          * we have the volume lock, so we don't need the extra
1620          * device list mutex while reading the list here.
1621          */
1622         list_for_each_entry(device, devices, dev_list) {
1623                 if (device->bdev == bdev) {
1624                         ret = -EEXIST;
1625                         goto error;
1626                 }
1627         }
1628
1629         device = kzalloc(sizeof(*device), GFP_NOFS);
1630         if (!device) {
1631                 /* we can safely leave the fs_devices entry around */
1632                 ret = -ENOMEM;
1633                 goto error;
1634         }
1635
1636         device->name = kstrdup(device_path, GFP_NOFS);
1637         if (!device->name) {
1638                 kfree(device);
1639                 ret = -ENOMEM;
1640                 goto error;
1641         }
1642
1643         ret = find_next_devid(root, &device->devid);
1644         if (ret) {
1645                 kfree(device->name);
1646                 kfree(device);
1647                 goto error;
1648         }
1649
1650         trans = btrfs_start_transaction(root, 0);
1651         if (IS_ERR(trans)) {
1652                 kfree(device->name);
1653                 kfree(device);
1654                 ret = PTR_ERR(trans);
1655                 goto error;
1656         }
1657
1658         lock_chunks(root);
1659
1660         q = bdev_get_queue(bdev);
1661         if (blk_queue_discard(q))
1662                 device->can_discard = 1;
1663         device->writeable = 1;
1664         device->work.func = pending_bios_fn;
1665         generate_random_uuid(device->uuid);
1666         spin_lock_init(&device->io_lock);
1667         device->generation = trans->transid;
1668         device->io_width = root->sectorsize;
1669         device->io_align = root->sectorsize;
1670         device->sector_size = root->sectorsize;
1671         device->total_bytes = i_size_read(bdev->bd_inode);
1672         device->disk_total_bytes = device->total_bytes;
1673         device->dev_root = root->fs_info->dev_root;
1674         device->bdev = bdev;
1675         device->in_fs_metadata = 1;
1676         device->mode = FMODE_EXCL;
1677         set_blocksize(device->bdev, 4096);
1678
1679         if (seeding_dev) {
1680                 sb->s_flags &= ~MS_RDONLY;
1681                 ret = btrfs_prepare_sprout(trans, root);
1682                 BUG_ON(ret);
1683         }
1684
1685         device->fs_devices = root->fs_info->fs_devices;
1686
1687         /*
1688          * we don't want write_supers to jump in here with our device
1689          * half setup
1690          */
1691         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1692         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1693         list_add(&device->dev_alloc_list,
1694                  &root->fs_info->fs_devices->alloc_list);
1695         root->fs_info->fs_devices->num_devices++;
1696         root->fs_info->fs_devices->open_devices++;
1697         root->fs_info->fs_devices->rw_devices++;
1698         if (device->can_discard)
1699                 root->fs_info->fs_devices->num_can_discard++;
1700         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1701
1702         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1703                 root->fs_info->fs_devices->rotating = 1;
1704
1705         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1706         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1707                                     total_bytes + device->total_bytes);
1708
1709         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1710         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1711                                     total_bytes + 1);
1712         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1713
1714         if (seeding_dev) {
1715                 ret = init_first_rw_device(trans, root, device);
1716                 BUG_ON(ret);
1717                 ret = btrfs_finish_sprout(trans, root);
1718                 BUG_ON(ret);
1719         } else {
1720                 ret = btrfs_add_device(trans, root, device);
1721         }
1722
1723         /*
1724          * we've got more storage, clear any full flags on the space
1725          * infos
1726          */
1727         btrfs_clear_space_info_full(root->fs_info);
1728
1729         unlock_chunks(root);
1730         btrfs_commit_transaction(trans, root);
1731
1732         if (seeding_dev) {
1733                 mutex_unlock(&uuid_mutex);
1734                 up_write(&sb->s_umount);
1735
1736                 ret = btrfs_relocate_sys_chunks(root);
1737                 BUG_ON(ret);
1738         }
1739 out:
1740         mutex_unlock(&root->fs_info->volume_mutex);
1741         return ret;
1742 error:
1743         blkdev_put(bdev, FMODE_EXCL);
1744         if (seeding_dev) {
1745                 mutex_unlock(&uuid_mutex);
1746                 up_write(&sb->s_umount);
1747         }
1748         goto out;
1749 }
1750
1751 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1752                                         struct btrfs_device *device)
1753 {
1754         int ret;
1755         struct btrfs_path *path;
1756         struct btrfs_root *root;
1757         struct btrfs_dev_item *dev_item;
1758         struct extent_buffer *leaf;
1759         struct btrfs_key key;
1760
1761         root = device->dev_root->fs_info->chunk_root;
1762
1763         path = btrfs_alloc_path();
1764         if (!path)
1765                 return -ENOMEM;
1766
1767         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1768         key.type = BTRFS_DEV_ITEM_KEY;
1769         key.offset = device->devid;
1770
1771         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1772         if (ret < 0)
1773                 goto out;
1774
1775         if (ret > 0) {
1776                 ret = -ENOENT;
1777                 goto out;
1778         }
1779
1780         leaf = path->nodes[0];
1781         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1782
1783         btrfs_set_device_id(leaf, dev_item, device->devid);
1784         btrfs_set_device_type(leaf, dev_item, device->type);
1785         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1786         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1787         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1788         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1789         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1790         btrfs_mark_buffer_dirty(leaf);
1791
1792 out:
1793         btrfs_free_path(path);
1794         return ret;
1795 }
1796
1797 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1798                       struct btrfs_device *device, u64 new_size)
1799 {
1800         struct btrfs_super_block *super_copy =
1801                 &device->dev_root->fs_info->super_copy;
1802         u64 old_total = btrfs_super_total_bytes(super_copy);
1803         u64 diff = new_size - device->total_bytes;
1804
1805         if (!device->writeable)
1806                 return -EACCES;
1807         if (new_size <= device->total_bytes)
1808                 return -EINVAL;
1809
1810         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1811         device->fs_devices->total_rw_bytes += diff;
1812
1813         device->total_bytes = new_size;
1814         device->disk_total_bytes = new_size;
1815         btrfs_clear_space_info_full(device->dev_root->fs_info);
1816
1817         return btrfs_update_device(trans, device);
1818 }
1819
1820 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1821                       struct btrfs_device *device, u64 new_size)
1822 {
1823         int ret;
1824         lock_chunks(device->dev_root);
1825         ret = __btrfs_grow_device(trans, device, new_size);
1826         unlock_chunks(device->dev_root);
1827         return ret;
1828 }
1829
1830 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1831                             struct btrfs_root *root,
1832                             u64 chunk_tree, u64 chunk_objectid,
1833                             u64 chunk_offset)
1834 {
1835         int ret;
1836         struct btrfs_path *path;
1837         struct btrfs_key key;
1838
1839         root = root->fs_info->chunk_root;
1840         path = btrfs_alloc_path();
1841         if (!path)
1842                 return -ENOMEM;
1843
1844         key.objectid = chunk_objectid;
1845         key.offset = chunk_offset;
1846         key.type = BTRFS_CHUNK_ITEM_KEY;
1847
1848         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1849         BUG_ON(ret);
1850
1851         ret = btrfs_del_item(trans, root, path);
1852
1853         btrfs_free_path(path);
1854         return ret;
1855 }
1856
1857 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1858                         chunk_offset)
1859 {
1860         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1861         struct btrfs_disk_key *disk_key;
1862         struct btrfs_chunk *chunk;
1863         u8 *ptr;
1864         int ret = 0;
1865         u32 num_stripes;
1866         u32 array_size;
1867         u32 len = 0;
1868         u32 cur;
1869         struct btrfs_key key;
1870
1871         array_size = btrfs_super_sys_array_size(super_copy);
1872
1873         ptr = super_copy->sys_chunk_array;
1874         cur = 0;
1875
1876         while (cur < array_size) {
1877                 disk_key = (struct btrfs_disk_key *)ptr;
1878                 btrfs_disk_key_to_cpu(&key, disk_key);
1879
1880                 len = sizeof(*disk_key);
1881
1882                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1883                         chunk = (struct btrfs_chunk *)(ptr + len);
1884                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1885                         len += btrfs_chunk_item_size(num_stripes);
1886                 } else {
1887                         ret = -EIO;
1888                         break;
1889                 }
1890                 if (key.objectid == chunk_objectid &&
1891                     key.offset == chunk_offset) {
1892                         memmove(ptr, ptr + len, array_size - (cur + len));
1893                         array_size -= len;
1894                         btrfs_set_super_sys_array_size(super_copy, array_size);
1895                 } else {
1896                         ptr += len;
1897                         cur += len;
1898                 }
1899         }
1900         return ret;
1901 }
1902
1903 static int btrfs_relocate_chunk(struct btrfs_root *root,
1904                          u64 chunk_tree, u64 chunk_objectid,
1905                          u64 chunk_offset)
1906 {
1907         struct extent_map_tree *em_tree;
1908         struct btrfs_root *extent_root;
1909         struct btrfs_trans_handle *trans;
1910         struct extent_map *em;
1911         struct map_lookup *map;
1912         int ret;
1913         int i;
1914
1915         root = root->fs_info->chunk_root;
1916         extent_root = root->fs_info->extent_root;
1917         em_tree = &root->fs_info->mapping_tree.map_tree;
1918
1919         ret = btrfs_can_relocate(extent_root, chunk_offset);
1920         if (ret)
1921                 return -ENOSPC;
1922
1923         /* step one, relocate all the extents inside this chunk */
1924         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1925         if (ret)
1926                 return ret;
1927
1928         trans = btrfs_start_transaction(root, 0);
1929         BUG_ON(IS_ERR(trans));
1930
1931         lock_chunks(root);
1932
1933         /*
1934          * step two, delete the device extents and the
1935          * chunk tree entries
1936          */
1937         read_lock(&em_tree->lock);
1938         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1939         read_unlock(&em_tree->lock);
1940
1941         BUG_ON(em->start > chunk_offset ||
1942                em->start + em->len < chunk_offset);
1943         map = (struct map_lookup *)em->bdev;
1944
1945         for (i = 0; i < map->num_stripes; i++) {
1946                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1947                                             map->stripes[i].physical);
1948                 BUG_ON(ret);
1949
1950                 if (map->stripes[i].dev) {
1951                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1952                         BUG_ON(ret);
1953                 }
1954         }
1955         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1956                                chunk_offset);
1957
1958         BUG_ON(ret);
1959
1960         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1961
1962         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1963                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1964                 BUG_ON(ret);
1965         }
1966
1967         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1968         BUG_ON(ret);
1969
1970         write_lock(&em_tree->lock);
1971         remove_extent_mapping(em_tree, em);
1972         write_unlock(&em_tree->lock);
1973
1974         kfree(map);
1975         em->bdev = NULL;
1976
1977         /* once for the tree */
1978         free_extent_map(em);
1979         /* once for us */
1980         free_extent_map(em);
1981
1982         unlock_chunks(root);
1983         btrfs_end_transaction(trans, root);
1984         return 0;
1985 }
1986
1987 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1988 {
1989         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1990         struct btrfs_path *path;
1991         struct extent_buffer *leaf;
1992         struct btrfs_chunk *chunk;
1993         struct btrfs_key key;
1994         struct btrfs_key found_key;
1995         u64 chunk_tree = chunk_root->root_key.objectid;
1996         u64 chunk_type;
1997         bool retried = false;
1998         int failed = 0;
1999         int ret;
2000
2001         path = btrfs_alloc_path();
2002         if (!path)
2003                 return -ENOMEM;
2004
2005 again:
2006         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2007         key.offset = (u64)-1;
2008         key.type = BTRFS_CHUNK_ITEM_KEY;
2009
2010         while (1) {
2011                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2012                 if (ret < 0)
2013                         goto error;
2014                 BUG_ON(ret == 0);
2015
2016                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2017                                           key.type);
2018                 if (ret < 0)
2019                         goto error;
2020                 if (ret > 0)
2021                         break;
2022
2023                 leaf = path->nodes[0];
2024                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2025
2026                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2027                                        struct btrfs_chunk);
2028                 chunk_type = btrfs_chunk_type(leaf, chunk);
2029                 btrfs_release_path(path);
2030
2031                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2032                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2033                                                    found_key.objectid,
2034                                                    found_key.offset);
2035                         if (ret == -ENOSPC)
2036                                 failed++;
2037                         else if (ret)
2038                                 BUG();
2039                 }
2040
2041                 if (found_key.offset == 0)
2042                         break;
2043                 key.offset = found_key.offset - 1;
2044         }
2045         ret = 0;
2046         if (failed && !retried) {
2047                 failed = 0;
2048                 retried = true;
2049                 goto again;
2050         } else if (failed && retried) {
2051                 WARN_ON(1);
2052                 ret = -ENOSPC;
2053         }
2054 error:
2055         btrfs_free_path(path);
2056         return ret;
2057 }
2058
2059 static u64 div_factor(u64 num, int factor)
2060 {
2061         if (factor == 10)
2062                 return num;
2063         num *= factor;
2064         do_div(num, 10);
2065         return num;
2066 }
2067
2068 int btrfs_balance(struct btrfs_root *dev_root)
2069 {
2070         int ret;
2071         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2072         struct btrfs_device *device;
2073         u64 old_size;
2074         u64 size_to_free;
2075         struct btrfs_path *path;
2076         struct btrfs_key key;
2077         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2078         struct btrfs_trans_handle *trans;
2079         struct btrfs_key found_key;
2080
2081         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2082                 return -EROFS;
2083
2084         if (!capable(CAP_SYS_ADMIN))
2085                 return -EPERM;
2086
2087         mutex_lock(&dev_root->fs_info->volume_mutex);
2088         dev_root = dev_root->fs_info->dev_root;
2089
2090         /* step one make some room on all the devices */
2091         list_for_each_entry(device, devices, dev_list) {
2092                 old_size = device->total_bytes;
2093                 size_to_free = div_factor(old_size, 1);
2094                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2095                 if (!device->writeable ||
2096                     device->total_bytes - device->bytes_used > size_to_free)
2097                         continue;
2098
2099                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2100                 if (ret == -ENOSPC)
2101                         break;
2102                 BUG_ON(ret);
2103
2104                 trans = btrfs_start_transaction(dev_root, 0);
2105                 BUG_ON(IS_ERR(trans));
2106
2107                 ret = btrfs_grow_device(trans, device, old_size);
2108                 BUG_ON(ret);
2109
2110                 btrfs_end_transaction(trans, dev_root);
2111         }
2112
2113         /* step two, relocate all the chunks */
2114         path = btrfs_alloc_path();
2115         if (!path) {
2116                 ret = -ENOMEM;
2117                 goto error;
2118         }
2119         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2120         key.offset = (u64)-1;
2121         key.type = BTRFS_CHUNK_ITEM_KEY;
2122
2123         while (1) {
2124                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2125                 if (ret < 0)
2126                         goto error;
2127
2128                 /*
2129                  * this shouldn't happen, it means the last relocate
2130                  * failed
2131                  */
2132                 if (ret == 0)
2133                         break;
2134
2135                 ret = btrfs_previous_item(chunk_root, path, 0,
2136                                           BTRFS_CHUNK_ITEM_KEY);
2137                 if (ret)
2138                         break;
2139
2140                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2141                                       path->slots[0]);
2142                 if (found_key.objectid != key.objectid)
2143                         break;
2144
2145                 /* chunk zero is special */
2146                 if (found_key.offset == 0)
2147                         break;
2148
2149                 btrfs_release_path(path);
2150                 ret = btrfs_relocate_chunk(chunk_root,
2151                                            chunk_root->root_key.objectid,
2152                                            found_key.objectid,
2153                                            found_key.offset);
2154                 if (ret && ret != -ENOSPC)
2155                         goto error;
2156                 key.offset = found_key.offset - 1;
2157         }
2158         ret = 0;
2159 error:
2160         btrfs_free_path(path);
2161         mutex_unlock(&dev_root->fs_info->volume_mutex);
2162         return ret;
2163 }
2164
2165 /*
2166  * shrinking a device means finding all of the device extents past
2167  * the new size, and then following the back refs to the chunks.
2168  * The chunk relocation code actually frees the device extent
2169  */
2170 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2171 {
2172         struct btrfs_trans_handle *trans;
2173         struct btrfs_root *root = device->dev_root;
2174         struct btrfs_dev_extent *dev_extent = NULL;
2175         struct btrfs_path *path;
2176         u64 length;
2177         u64 chunk_tree;
2178         u64 chunk_objectid;
2179         u64 chunk_offset;
2180         int ret;
2181         int slot;
2182         int failed = 0;
2183         bool retried = false;
2184         struct extent_buffer *l;
2185         struct btrfs_key key;
2186         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2187         u64 old_total = btrfs_super_total_bytes(super_copy);
2188         u64 old_size = device->total_bytes;
2189         u64 diff = device->total_bytes - new_size;
2190
2191         if (new_size >= device->total_bytes)
2192                 return -EINVAL;
2193
2194         path = btrfs_alloc_path();
2195         if (!path)
2196                 return -ENOMEM;
2197
2198         path->reada = 2;
2199
2200         lock_chunks(root);
2201
2202         device->total_bytes = new_size;
2203         if (device->writeable)
2204                 device->fs_devices->total_rw_bytes -= diff;
2205         unlock_chunks(root);
2206
2207 again:
2208         key.objectid = device->devid;
2209         key.offset = (u64)-1;
2210         key.type = BTRFS_DEV_EXTENT_KEY;
2211
2212         while (1) {
2213                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2214                 if (ret < 0)
2215                         goto done;
2216
2217                 ret = btrfs_previous_item(root, path, 0, key.type);
2218                 if (ret < 0)
2219                         goto done;
2220                 if (ret) {
2221                         ret = 0;
2222                         btrfs_release_path(path);
2223                         break;
2224                 }
2225
2226                 l = path->nodes[0];
2227                 slot = path->slots[0];
2228                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2229
2230                 if (key.objectid != device->devid) {
2231                         btrfs_release_path(path);
2232                         break;
2233                 }
2234
2235                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2236                 length = btrfs_dev_extent_length(l, dev_extent);
2237
2238                 if (key.offset + length <= new_size) {
2239                         btrfs_release_path(path);
2240                         break;
2241                 }
2242
2243                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2244                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2245                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2246                 btrfs_release_path(path);
2247
2248                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2249                                            chunk_offset);
2250                 if (ret && ret != -ENOSPC)
2251                         goto done;
2252                 if (ret == -ENOSPC)
2253                         failed++;
2254                 key.offset -= 1;
2255         }
2256
2257         if (failed && !retried) {
2258                 failed = 0;
2259                 retried = true;
2260                 goto again;
2261         } else if (failed && retried) {
2262                 ret = -ENOSPC;
2263                 lock_chunks(root);
2264
2265                 device->total_bytes = old_size;
2266                 if (device->writeable)
2267                         device->fs_devices->total_rw_bytes += diff;
2268                 unlock_chunks(root);
2269                 goto done;
2270         }
2271
2272         /* Shrinking succeeded, else we would be at "done". */
2273         trans = btrfs_start_transaction(root, 0);
2274         if (IS_ERR(trans)) {
2275                 ret = PTR_ERR(trans);
2276                 goto done;
2277         }
2278
2279         lock_chunks(root);
2280
2281         device->disk_total_bytes = new_size;
2282         /* Now btrfs_update_device() will change the on-disk size. */
2283         ret = btrfs_update_device(trans, device);
2284         if (ret) {
2285                 unlock_chunks(root);
2286                 btrfs_end_transaction(trans, root);
2287                 goto done;
2288         }
2289         WARN_ON(diff > old_total);
2290         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2291         unlock_chunks(root);
2292         btrfs_end_transaction(trans, root);
2293 done:
2294         btrfs_free_path(path);
2295         return ret;
2296 }
2297
2298 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2299                            struct btrfs_root *root,
2300                            struct btrfs_key *key,
2301                            struct btrfs_chunk *chunk, int item_size)
2302 {
2303         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2304         struct btrfs_disk_key disk_key;
2305         u32 array_size;
2306         u8 *ptr;
2307
2308         array_size = btrfs_super_sys_array_size(super_copy);
2309         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2310                 return -EFBIG;
2311
2312         ptr = super_copy->sys_chunk_array + array_size;
2313         btrfs_cpu_key_to_disk(&disk_key, key);
2314         memcpy(ptr, &disk_key, sizeof(disk_key));
2315         ptr += sizeof(disk_key);
2316         memcpy(ptr, chunk, item_size);
2317         item_size += sizeof(disk_key);
2318         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2319         return 0;
2320 }
2321
2322 /*
2323  * sort the devices in descending order by max_avail, total_avail
2324  */
2325 static int btrfs_cmp_device_info(const void *a, const void *b)
2326 {
2327         const struct btrfs_device_info *di_a = a;
2328         const struct btrfs_device_info *di_b = b;
2329
2330         if (di_a->max_avail > di_b->max_avail)
2331                 return -1;
2332         if (di_a->max_avail < di_b->max_avail)
2333                 return 1;
2334         if (di_a->total_avail > di_b->total_avail)
2335                 return -1;
2336         if (di_a->total_avail < di_b->total_avail)
2337                 return 1;
2338         return 0;
2339 }
2340
2341 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2342                                struct btrfs_root *extent_root,
2343                                struct map_lookup **map_ret,
2344                                u64 *num_bytes_out, u64 *stripe_size_out,
2345                                u64 start, u64 type)
2346 {
2347         struct btrfs_fs_info *info = extent_root->fs_info;
2348         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2349         struct list_head *cur;
2350         struct map_lookup *map = NULL;
2351         struct extent_map_tree *em_tree;
2352         struct extent_map *em;
2353         struct btrfs_device_info *devices_info = NULL;
2354         u64 total_avail;
2355         int num_stripes;        /* total number of stripes to allocate */
2356         int sub_stripes;        /* sub_stripes info for map */
2357         int dev_stripes;        /* stripes per dev */
2358         int devs_max;           /* max devs to use */
2359         int devs_min;           /* min devs needed */
2360         int devs_increment;     /* ndevs has to be a multiple of this */
2361         int ncopies;            /* how many copies to data has */
2362         int ret;
2363         u64 max_stripe_size;
2364         u64 max_chunk_size;
2365         u64 stripe_size;
2366         u64 num_bytes;
2367         int ndevs;
2368         int i;
2369         int j;
2370
2371         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2372             (type & BTRFS_BLOCK_GROUP_DUP)) {
2373                 WARN_ON(1);
2374                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2375         }
2376
2377         if (list_empty(&fs_devices->alloc_list))
2378                 return -ENOSPC;
2379
2380         sub_stripes = 1;
2381         dev_stripes = 1;
2382         devs_increment = 1;
2383         ncopies = 1;
2384         devs_max = 0;   /* 0 == as many as possible */
2385         devs_min = 1;
2386
2387         /*
2388          * define the properties of each RAID type.
2389          * FIXME: move this to a global table and use it in all RAID
2390          * calculation code
2391          */
2392         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2393                 dev_stripes = 2;
2394                 ncopies = 2;
2395                 devs_max = 1;
2396         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2397                 devs_min = 2;
2398         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2399                 devs_increment = 2;
2400                 ncopies = 2;
2401                 devs_max = 2;
2402                 devs_min = 2;
2403         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2404                 sub_stripes = 2;
2405                 devs_increment = 2;
2406                 ncopies = 2;
2407                 devs_min = 4;
2408         } else {
2409                 devs_max = 1;
2410         }
2411
2412         if (type & BTRFS_BLOCK_GROUP_DATA) {
2413                 max_stripe_size = 1024 * 1024 * 1024;
2414                 max_chunk_size = 10 * max_stripe_size;
2415         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2416                 max_stripe_size = 256 * 1024 * 1024;
2417                 max_chunk_size = max_stripe_size;
2418         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2419                 max_stripe_size = 8 * 1024 * 1024;
2420                 max_chunk_size = 2 * max_stripe_size;
2421         } else {
2422                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2423                        type);
2424                 BUG_ON(1);
2425         }
2426
2427         /* we don't want a chunk larger than 10% of writeable space */
2428         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2429                              max_chunk_size);
2430
2431         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2432                                GFP_NOFS);
2433         if (!devices_info)
2434                 return -ENOMEM;
2435
2436         cur = fs_devices->alloc_list.next;
2437
2438         /*
2439          * in the first pass through the devices list, we gather information
2440          * about the available holes on each device.
2441          */
2442         ndevs = 0;
2443         while (cur != &fs_devices->alloc_list) {
2444                 struct btrfs_device *device;
2445                 u64 max_avail;
2446                 u64 dev_offset;
2447
2448                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2449
2450                 cur = cur->next;
2451
2452                 if (!device->writeable) {
2453                         printk(KERN_ERR
2454                                "btrfs: read-only device in alloc_list\n");
2455                         WARN_ON(1);
2456                         continue;
2457                 }
2458
2459                 if (!device->in_fs_metadata)
2460                         continue;
2461
2462                 if (device->total_bytes > device->bytes_used)
2463                         total_avail = device->total_bytes - device->bytes_used;
2464                 else
2465                         total_avail = 0;
2466
2467                 /* If there is no space on this device, skip it. */
2468                 if (total_avail == 0)
2469                         continue;
2470
2471                 ret = find_free_dev_extent(trans, device,
2472                                            max_stripe_size * dev_stripes,
2473                                            &dev_offset, &max_avail);
2474                 if (ret && ret != -ENOSPC)
2475                         goto error;
2476
2477                 if (ret == 0)
2478                         max_avail = max_stripe_size * dev_stripes;
2479
2480                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2481                         continue;
2482
2483                 devices_info[ndevs].dev_offset = dev_offset;
2484                 devices_info[ndevs].max_avail = max_avail;
2485                 devices_info[ndevs].total_avail = total_avail;
2486                 devices_info[ndevs].dev = device;
2487                 ++ndevs;
2488         }
2489
2490         /*
2491          * now sort the devices by hole size / available space
2492          */
2493         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2494              btrfs_cmp_device_info, NULL);
2495
2496         /* round down to number of usable stripes */
2497         ndevs -= ndevs % devs_increment;
2498
2499         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2500                 ret = -ENOSPC;
2501                 goto error;
2502         }
2503
2504         if (devs_max && ndevs > devs_max)
2505                 ndevs = devs_max;
2506         /*
2507          * the primary goal is to maximize the number of stripes, so use as many
2508          * devices as possible, even if the stripes are not maximum sized.
2509          */
2510         stripe_size = devices_info[ndevs-1].max_avail;
2511         num_stripes = ndevs * dev_stripes;
2512
2513         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2514                 stripe_size = max_chunk_size * ncopies;
2515                 do_div(stripe_size, num_stripes);
2516         }
2517
2518         do_div(stripe_size, dev_stripes);
2519         do_div(stripe_size, BTRFS_STRIPE_LEN);
2520         stripe_size *= BTRFS_STRIPE_LEN;
2521
2522         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2523         if (!map) {
2524                 ret = -ENOMEM;
2525                 goto error;
2526         }
2527         map->num_stripes = num_stripes;
2528
2529         for (i = 0; i < ndevs; ++i) {
2530                 for (j = 0; j < dev_stripes; ++j) {
2531                         int s = i * dev_stripes + j;
2532                         map->stripes[s].dev = devices_info[i].dev;
2533                         map->stripes[s].physical = devices_info[i].dev_offset +
2534                                                    j * stripe_size;
2535                 }
2536         }
2537         map->sector_size = extent_root->sectorsize;
2538         map->stripe_len = BTRFS_STRIPE_LEN;
2539         map->io_align = BTRFS_STRIPE_LEN;
2540         map->io_width = BTRFS_STRIPE_LEN;
2541         map->type = type;
2542         map->sub_stripes = sub_stripes;
2543
2544         *map_ret = map;
2545         num_bytes = stripe_size * (num_stripes / ncopies);
2546
2547         *stripe_size_out = stripe_size;
2548         *num_bytes_out = num_bytes;
2549
2550         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2551
2552         em = alloc_extent_map();
2553         if (!em) {
2554                 ret = -ENOMEM;
2555                 goto error;
2556         }
2557         em->bdev = (struct block_device *)map;
2558         em->start = start;
2559         em->len = num_bytes;
2560         em->block_start = 0;
2561         em->block_len = em->len;
2562
2563         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2564         write_lock(&em_tree->lock);
2565         ret = add_extent_mapping(em_tree, em);
2566         write_unlock(&em_tree->lock);
2567         BUG_ON(ret);
2568         free_extent_map(em);
2569
2570         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2571                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2572                                      start, num_bytes);
2573         BUG_ON(ret);
2574
2575         for (i = 0; i < map->num_stripes; ++i) {
2576                 struct btrfs_device *device;
2577                 u64 dev_offset;
2578
2579                 device = map->stripes[i].dev;
2580                 dev_offset = map->stripes[i].physical;
2581
2582                 ret = btrfs_alloc_dev_extent(trans, device,
2583                                 info->chunk_root->root_key.objectid,
2584                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2585                                 start, dev_offset, stripe_size);
2586                 BUG_ON(ret);
2587         }
2588
2589         kfree(devices_info);
2590         return 0;
2591
2592 error:
2593         kfree(map);
2594         kfree(devices_info);
2595         return ret;
2596 }
2597
2598 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2599                                 struct btrfs_root *extent_root,
2600                                 struct map_lookup *map, u64 chunk_offset,
2601                                 u64 chunk_size, u64 stripe_size)
2602 {
2603         u64 dev_offset;
2604         struct btrfs_key key;
2605         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2606         struct btrfs_device *device;
2607         struct btrfs_chunk *chunk;
2608         struct btrfs_stripe *stripe;
2609         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2610         int index = 0;
2611         int ret;
2612
2613         chunk = kzalloc(item_size, GFP_NOFS);
2614         if (!chunk)
2615                 return -ENOMEM;
2616
2617         index = 0;
2618         while (index < map->num_stripes) {
2619                 device = map->stripes[index].dev;
2620                 device->bytes_used += stripe_size;
2621                 ret = btrfs_update_device(trans, device);
2622                 BUG_ON(ret);
2623                 index++;
2624         }
2625
2626         index = 0;
2627         stripe = &chunk->stripe;
2628         while (index < map->num_stripes) {
2629                 device = map->stripes[index].dev;
2630                 dev_offset = map->stripes[index].physical;
2631
2632                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2633                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2634                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2635                 stripe++;
2636                 index++;
2637         }
2638
2639         btrfs_set_stack_chunk_length(chunk, chunk_size);
2640         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2641         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2642         btrfs_set_stack_chunk_type(chunk, map->type);
2643         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2644         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2645         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2646         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2647         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2648
2649         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2650         key.type = BTRFS_CHUNK_ITEM_KEY;
2651         key.offset = chunk_offset;
2652
2653         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2654         BUG_ON(ret);
2655
2656         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2657                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2658                                              item_size);
2659                 BUG_ON(ret);
2660         }
2661
2662         kfree(chunk);
2663         return 0;
2664 }
2665
2666 /*
2667  * Chunk allocation falls into two parts. The first part does works
2668  * that make the new allocated chunk useable, but not do any operation
2669  * that modifies the chunk tree. The second part does the works that
2670  * require modifying the chunk tree. This division is important for the
2671  * bootstrap process of adding storage to a seed btrfs.
2672  */
2673 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2674                       struct btrfs_root *extent_root, u64 type)
2675 {
2676         u64 chunk_offset;
2677         u64 chunk_size;
2678         u64 stripe_size;
2679         struct map_lookup *map;
2680         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2681         int ret;
2682
2683         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2684                               &chunk_offset);
2685         if (ret)
2686                 return ret;
2687
2688         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2689                                   &stripe_size, chunk_offset, type);
2690         if (ret)
2691                 return ret;
2692
2693         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2694                                    chunk_size, stripe_size);
2695         BUG_ON(ret);
2696         return 0;
2697 }
2698
2699 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2700                                          struct btrfs_root *root,
2701                                          struct btrfs_device *device)
2702 {
2703         u64 chunk_offset;
2704         u64 sys_chunk_offset;
2705         u64 chunk_size;
2706         u64 sys_chunk_size;
2707         u64 stripe_size;
2708         u64 sys_stripe_size;
2709         u64 alloc_profile;
2710         struct map_lookup *map;
2711         struct map_lookup *sys_map;
2712         struct btrfs_fs_info *fs_info = root->fs_info;
2713         struct btrfs_root *extent_root = fs_info->extent_root;
2714         int ret;
2715
2716         ret = find_next_chunk(fs_info->chunk_root,
2717                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2718         if (ret)
2719                 return ret;
2720
2721         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2722                         (fs_info->metadata_alloc_profile &
2723                          fs_info->avail_metadata_alloc_bits);
2724         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2725
2726         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2727                                   &stripe_size, chunk_offset, alloc_profile);
2728         BUG_ON(ret);
2729
2730         sys_chunk_offset = chunk_offset + chunk_size;
2731
2732         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2733                         (fs_info->system_alloc_profile &
2734                          fs_info->avail_system_alloc_bits);
2735         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2736
2737         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2738                                   &sys_chunk_size, &sys_stripe_size,
2739                                   sys_chunk_offset, alloc_profile);
2740         BUG_ON(ret);
2741
2742         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2743         BUG_ON(ret);
2744
2745         /*
2746          * Modifying chunk tree needs allocating new blocks from both
2747          * system block group and metadata block group. So we only can
2748          * do operations require modifying the chunk tree after both
2749          * block groups were created.
2750          */
2751         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2752                                    chunk_size, stripe_size);
2753         BUG_ON(ret);
2754
2755         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2756                                    sys_chunk_offset, sys_chunk_size,
2757                                    sys_stripe_size);
2758         BUG_ON(ret);
2759         return 0;
2760 }
2761
2762 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2763 {
2764         struct extent_map *em;
2765         struct map_lookup *map;
2766         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2767         int readonly = 0;
2768         int i;
2769
2770         read_lock(&map_tree->map_tree.lock);
2771         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2772         read_unlock(&map_tree->map_tree.lock);
2773         if (!em)
2774                 return 1;
2775
2776         if (btrfs_test_opt(root, DEGRADED)) {
2777                 free_extent_map(em);
2778                 return 0;
2779         }
2780
2781         map = (struct map_lookup *)em->bdev;
2782         for (i = 0; i < map->num_stripes; i++) {
2783                 if (!map->stripes[i].dev->writeable) {
2784                         readonly = 1;
2785                         break;
2786                 }
2787         }
2788         free_extent_map(em);
2789         return readonly;
2790 }
2791
2792 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2793 {
2794         extent_map_tree_init(&tree->map_tree);
2795 }
2796
2797 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2798 {
2799         struct extent_map *em;
2800
2801         while (1) {
2802                 write_lock(&tree->map_tree.lock);
2803                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2804                 if (em)
2805                         remove_extent_mapping(&tree->map_tree, em);
2806                 write_unlock(&tree->map_tree.lock);
2807                 if (!em)
2808                         break;
2809                 kfree(em->bdev);
2810                 /* once for us */
2811                 free_extent_map(em);
2812                 /* once for the tree */
2813                 free_extent_map(em);
2814         }
2815 }
2816
2817 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2818 {
2819         struct extent_map *em;
2820         struct map_lookup *map;
2821         struct extent_map_tree *em_tree = &map_tree->map_tree;
2822         int ret;
2823
2824         read_lock(&em_tree->lock);
2825         em = lookup_extent_mapping(em_tree, logical, len);
2826         read_unlock(&em_tree->lock);
2827         BUG_ON(!em);
2828
2829         BUG_ON(em->start > logical || em->start + em->len < logical);
2830         map = (struct map_lookup *)em->bdev;
2831         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2832                 ret = map->num_stripes;
2833         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2834                 ret = map->sub_stripes;
2835         else
2836                 ret = 1;
2837         free_extent_map(em);
2838         return ret;
2839 }
2840
2841 static int find_live_mirror(struct map_lookup *map, int first, int num,
2842                             int optimal)
2843 {
2844         int i;
2845         if (map->stripes[optimal].dev->bdev)
2846                 return optimal;
2847         for (i = first; i < first + num; i++) {
2848                 if (map->stripes[i].dev->bdev)
2849                         return i;
2850         }
2851         /* we couldn't find one that doesn't fail.  Just return something
2852          * and the io error handling code will clean up eventually
2853          */
2854         return optimal;
2855 }
2856
2857 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2858                              u64 logical, u64 *length,
2859                              struct btrfs_multi_bio **multi_ret,
2860                              int mirror_num)
2861 {
2862         struct extent_map *em;
2863         struct map_lookup *map;
2864         struct extent_map_tree *em_tree = &map_tree->map_tree;
2865         u64 offset;
2866         u64 stripe_offset;
2867         u64 stripe_end_offset;
2868         u64 stripe_nr;
2869         u64 stripe_nr_orig;
2870         u64 stripe_nr_end;
2871         int stripes_allocated = 8;
2872         int stripes_required = 1;
2873         int stripe_index;
2874         int i;
2875         int num_stripes;
2876         int max_errors = 0;
2877         struct btrfs_multi_bio *multi = NULL;
2878
2879         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2880                 stripes_allocated = 1;
2881 again:
2882         if (multi_ret) {
2883                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2884                                 GFP_NOFS);
2885                 if (!multi)
2886                         return -ENOMEM;
2887
2888                 atomic_set(&multi->error, 0);
2889         }
2890
2891         read_lock(&em_tree->lock);
2892         em = lookup_extent_mapping(em_tree, logical, *length);
2893         read_unlock(&em_tree->lock);
2894
2895         if (!em) {
2896                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2897                        (unsigned long long)logical,
2898                        (unsigned long long)*length);
2899                 BUG();
2900         }
2901
2902         BUG_ON(em->start > logical || em->start + em->len < logical);
2903         map = (struct map_lookup *)em->bdev;
2904         offset = logical - em->start;
2905
2906         if (mirror_num > map->num_stripes)
2907                 mirror_num = 0;
2908
2909         /* if our multi bio struct is too small, back off and try again */
2910         if (rw & REQ_WRITE) {
2911                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2912                                  BTRFS_BLOCK_GROUP_DUP)) {
2913                         stripes_required = map->num_stripes;
2914                         max_errors = 1;
2915                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2916                         stripes_required = map->sub_stripes;
2917                         max_errors = 1;
2918                 }
2919         }
2920         if (rw & REQ_DISCARD) {
2921                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2922                                  BTRFS_BLOCK_GROUP_RAID1 |
2923                                  BTRFS_BLOCK_GROUP_DUP |
2924                                  BTRFS_BLOCK_GROUP_RAID10)) {
2925                         stripes_required = map->num_stripes;
2926                 }
2927         }
2928         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2929             stripes_allocated < stripes_required) {
2930                 stripes_allocated = map->num_stripes;
2931                 free_extent_map(em);
2932                 kfree(multi);
2933                 goto again;
2934         }
2935         stripe_nr = offset;
2936         /*
2937          * stripe_nr counts the total number of stripes we have to stride
2938          * to get to this block
2939          */
2940         do_div(stripe_nr, map->stripe_len);
2941
2942         stripe_offset = stripe_nr * map->stripe_len;
2943         BUG_ON(offset < stripe_offset);
2944
2945         /* stripe_offset is the offset of this block in its stripe*/
2946         stripe_offset = offset - stripe_offset;
2947
2948         if (rw & REQ_DISCARD)
2949                 *length = min_t(u64, em->len - offset, *length);
2950         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2951                               BTRFS_BLOCK_GROUP_RAID1 |
2952                               BTRFS_BLOCK_GROUP_RAID10 |
2953                               BTRFS_BLOCK_GROUP_DUP)) {
2954                 /* we limit the length of each bio to what fits in a stripe */
2955                 *length = min_t(u64, em->len - offset,
2956                                 map->stripe_len - stripe_offset);
2957         } else {
2958                 *length = em->len - offset;
2959         }
2960
2961         if (!multi_ret)
2962                 goto out;
2963
2964         num_stripes = 1;
2965         stripe_index = 0;
2966         stripe_nr_orig = stripe_nr;
2967         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2968                         (~(map->stripe_len - 1));
2969         do_div(stripe_nr_end, map->stripe_len);
2970         stripe_end_offset = stripe_nr_end * map->stripe_len -
2971                             (offset + *length);
2972         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2973                 if (rw & REQ_DISCARD)
2974                         num_stripes = min_t(u64, map->num_stripes,
2975                                             stripe_nr_end - stripe_nr_orig);
2976                 stripe_index = do_div(stripe_nr, map->num_stripes);
2977         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2978                 if (rw & (REQ_WRITE | REQ_DISCARD))
2979                         num_stripes = map->num_stripes;
2980                 else if (mirror_num)
2981                         stripe_index = mirror_num - 1;
2982                 else {
2983                         stripe_index = find_live_mirror(map, 0,
2984                                             map->num_stripes,
2985                                             current->pid % map->num_stripes);
2986                 }
2987
2988         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2989                 if (rw & (REQ_WRITE | REQ_DISCARD))
2990                         num_stripes = map->num_stripes;
2991                 else if (mirror_num)
2992                         stripe_index = mirror_num - 1;
2993
2994         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2995                 int factor = map->num_stripes / map->sub_stripes;
2996
2997                 stripe_index = do_div(stripe_nr, factor);
2998                 stripe_index *= map->sub_stripes;
2999
3000                 if (rw & REQ_WRITE)
3001                         num_stripes = map->sub_stripes;
3002                 else if (rw & REQ_DISCARD)
3003                         num_stripes = min_t(u64, map->sub_stripes *
3004                                             (stripe_nr_end - stripe_nr_orig),
3005                                             map->num_stripes);
3006                 else if (mirror_num)
3007                         stripe_index += mirror_num - 1;
3008                 else {
3009                         stripe_index = find_live_mirror(map, stripe_index,
3010                                               map->sub_stripes, stripe_index +
3011                                               current->pid % map->sub_stripes);
3012                 }
3013         } else {
3014                 /*
3015                  * after this do_div call, stripe_nr is the number of stripes
3016                  * on this device we have to walk to find the data, and
3017                  * stripe_index is the number of our device in the stripe array
3018                  */
3019                 stripe_index = do_div(stripe_nr, map->num_stripes);
3020         }
3021         BUG_ON(stripe_index >= map->num_stripes);
3022
3023         if (rw & REQ_DISCARD) {
3024                 for (i = 0; i < num_stripes; i++) {
3025                         multi->stripes[i].physical =
3026                                 map->stripes[stripe_index].physical +
3027                                 stripe_offset + stripe_nr * map->stripe_len;
3028                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3029
3030                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3031                                 u64 stripes;
3032                                 u32 last_stripe = 0;
3033                                 int j;
3034
3035                                 div_u64_rem(stripe_nr_end - 1,
3036                                             map->num_stripes,
3037                                             &last_stripe);
3038
3039                                 for (j = 0; j < map->num_stripes; j++) {
3040                                         u32 test;
3041
3042                                         div_u64_rem(stripe_nr_end - 1 - j,
3043                                                     map->num_stripes, &test);
3044                                         if (test == stripe_index)
3045                                                 break;
3046                                 }
3047                                 stripes = stripe_nr_end - 1 - j;
3048                                 do_div(stripes, map->num_stripes);
3049                                 multi->stripes[i].length = map->stripe_len *
3050                                         (stripes - stripe_nr + 1);
3051
3052                                 if (i == 0) {
3053                                         multi->stripes[i].length -=
3054                                                 stripe_offset;
3055                                         stripe_offset = 0;
3056                                 }
3057                                 if (stripe_index == last_stripe)
3058                                         multi->stripes[i].length -=
3059                                                 stripe_end_offset;
3060                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3061                                 u64 stripes;
3062                                 int j;
3063                                 int factor = map->num_stripes /
3064                                              map->sub_stripes;
3065                                 u32 last_stripe = 0;
3066
3067                                 div_u64_rem(stripe_nr_end - 1,
3068                                             factor, &last_stripe);
3069                                 last_stripe *= map->sub_stripes;
3070
3071                                 for (j = 0; j < factor; j++) {
3072                                         u32 test;
3073
3074                                         div_u64_rem(stripe_nr_end - 1 - j,
3075                                                     factor, &test);
3076
3077                                         if (test ==
3078                                             stripe_index / map->sub_stripes)
3079                                                 break;
3080                                 }
3081                                 stripes = stripe_nr_end - 1 - j;
3082                                 do_div(stripes, factor);
3083                                 multi->stripes[i].length = map->stripe_len *
3084                                         (stripes - stripe_nr + 1);
3085
3086                                 if (i < map->sub_stripes) {
3087                                         multi->stripes[i].length -=
3088                                                 stripe_offset;
3089                                         if (i == map->sub_stripes - 1)
3090                                                 stripe_offset = 0;
3091                                 }
3092                                 if (stripe_index >= last_stripe &&
3093                                     stripe_index <= (last_stripe +
3094                                                      map->sub_stripes - 1)) {
3095                                         multi->stripes[i].length -=
3096                                                 stripe_end_offset;
3097                                 }
3098                         } else
3099                                 multi->stripes[i].length = *length;
3100
3101                         stripe_index++;
3102                         if (stripe_index == map->num_stripes) {
3103                                 /* This could only happen for RAID0/10 */
3104                                 stripe_index = 0;
3105                                 stripe_nr++;
3106                         }
3107                 }
3108         } else {
3109                 for (i = 0; i < num_stripes; i++) {
3110                         multi->stripes[i].physical =
3111                                 map->stripes[stripe_index].physical +
3112                                 stripe_offset +
3113                                 stripe_nr * map->stripe_len;
3114                         multi->stripes[i].dev =
3115                                 map->stripes[stripe_index].dev;
3116                         stripe_index++;
3117                 }
3118         }
3119         if (multi_ret) {
3120                 *multi_ret = multi;
3121                 multi->num_stripes = num_stripes;
3122                 multi->max_errors = max_errors;
3123         }
3124 out:
3125         free_extent_map(em);
3126         return 0;
3127 }
3128
3129 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3130                       u64 logical, u64 *length,
3131                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3132 {
3133         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3134                                  mirror_num);
3135 }
3136
3137 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3138                      u64 chunk_start, u64 physical, u64 devid,
3139                      u64 **logical, int *naddrs, int *stripe_len)
3140 {
3141         struct extent_map_tree *em_tree = &map_tree->map_tree;
3142         struct extent_map *em;
3143         struct map_lookup *map;
3144         u64 *buf;
3145         u64 bytenr;
3146         u64 length;
3147         u64 stripe_nr;
3148         int i, j, nr = 0;
3149
3150         read_lock(&em_tree->lock);
3151         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3152         read_unlock(&em_tree->lock);
3153
3154         BUG_ON(!em || em->start != chunk_start);
3155         map = (struct map_lookup *)em->bdev;
3156
3157         length = em->len;
3158         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3159                 do_div(length, map->num_stripes / map->sub_stripes);
3160         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3161                 do_div(length, map->num_stripes);
3162
3163         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3164         BUG_ON(!buf);
3165
3166         for (i = 0; i < map->num_stripes; i++) {
3167                 if (devid && map->stripes[i].dev->devid != devid)
3168                         continue;
3169                 if (map->stripes[i].physical > physical ||
3170                     map->stripes[i].physical + length <= physical)
3171                         continue;
3172
3173                 stripe_nr = physical - map->stripes[i].physical;
3174                 do_div(stripe_nr, map->stripe_len);
3175
3176                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3177                         stripe_nr = stripe_nr * map->num_stripes + i;
3178                         do_div(stripe_nr, map->sub_stripes);
3179                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3180                         stripe_nr = stripe_nr * map->num_stripes + i;
3181                 }
3182                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3183                 WARN_ON(nr >= map->num_stripes);
3184                 for (j = 0; j < nr; j++) {
3185                         if (buf[j] == bytenr)
3186                                 break;
3187                 }
3188                 if (j == nr) {
3189                         WARN_ON(nr >= map->num_stripes);
3190                         buf[nr++] = bytenr;
3191                 }
3192         }
3193
3194         *logical = buf;
3195         *naddrs = nr;
3196         *stripe_len = map->stripe_len;
3197
3198         free_extent_map(em);
3199         return 0;
3200 }
3201
3202 static void end_bio_multi_stripe(struct bio *bio, int err)
3203 {
3204         struct btrfs_multi_bio *multi = bio->bi_private;
3205         int is_orig_bio = 0;
3206
3207         if (err)
3208                 atomic_inc(&multi->error);
3209
3210         if (bio == multi->orig_bio)
3211                 is_orig_bio = 1;
3212
3213         if (atomic_dec_and_test(&multi->stripes_pending)) {
3214                 if (!is_orig_bio) {
3215                         bio_put(bio);
3216                         bio = multi->orig_bio;
3217                 }
3218                 bio->bi_private = multi->private;
3219                 bio->bi_end_io = multi->end_io;
3220                 /* only send an error to the higher layers if it is
3221                  * beyond the tolerance of the multi-bio
3222                  */
3223                 if (atomic_read(&multi->error) > multi->max_errors) {
3224                         err = -EIO;
3225                 } else if (err) {
3226                         /*
3227                          * this bio is actually up to date, we didn't
3228                          * go over the max number of errors
3229                          */
3230                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3231                         err = 0;
3232                 }
3233                 kfree(multi);
3234
3235                 bio_endio(bio, err);
3236         } else if (!is_orig_bio) {
3237                 bio_put(bio);
3238         }
3239 }
3240
3241 struct async_sched {
3242         struct bio *bio;
3243         int rw;
3244         struct btrfs_fs_info *info;
3245         struct btrfs_work work;
3246 };
3247
3248 /*
3249  * see run_scheduled_bios for a description of why bios are collected for
3250  * async submit.
3251  *
3252  * This will add one bio to the pending list for a device and make sure
3253  * the work struct is scheduled.
3254  */
3255 static noinline int schedule_bio(struct btrfs_root *root,
3256                                  struct btrfs_device *device,
3257                                  int rw, struct bio *bio)
3258 {
3259         int should_queue = 1;
3260         struct btrfs_pending_bios *pending_bios;
3261
3262         /* don't bother with additional async steps for reads, right now */
3263         if (!(rw & REQ_WRITE)) {
3264                 bio_get(bio);
3265                 submit_bio(rw, bio);
3266                 bio_put(bio);
3267                 return 0;
3268         }
3269
3270         /*
3271          * nr_async_bios allows us to reliably return congestion to the
3272          * higher layers.  Otherwise, the async bio makes it appear we have
3273          * made progress against dirty pages when we've really just put it
3274          * on a queue for later
3275          */
3276         atomic_inc(&root->fs_info->nr_async_bios);
3277         WARN_ON(bio->bi_next);
3278         bio->bi_next = NULL;
3279         bio->bi_rw |= rw;
3280
3281         spin_lock(&device->io_lock);
3282         if (bio->bi_rw & REQ_SYNC)
3283                 pending_bios = &device->pending_sync_bios;
3284         else
3285                 pending_bios = &device->pending_bios;
3286
3287         if (pending_bios->tail)
3288                 pending_bios->tail->bi_next = bio;
3289
3290         pending_bios->tail = bio;
3291         if (!pending_bios->head)
3292                 pending_bios->head = bio;
3293         if (device->running_pending)
3294                 should_queue = 0;
3295
3296         spin_unlock(&device->io_lock);
3297
3298         if (should_queue)
3299                 btrfs_queue_worker(&root->fs_info->submit_workers,
3300                                    &device->work);
3301         return 0;
3302 }
3303
3304 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3305                   int mirror_num, int async_submit)
3306 {
3307         struct btrfs_mapping_tree *map_tree;
3308         struct btrfs_device *dev;
3309         struct bio *first_bio = bio;
3310         u64 logical = (u64)bio->bi_sector << 9;
3311         u64 length = 0;
3312         u64 map_length;
3313         struct btrfs_multi_bio *multi = NULL;
3314         int ret;
3315         int dev_nr = 0;
3316         int total_devs = 1;
3317
3318         length = bio->bi_size;
3319         map_tree = &root->fs_info->mapping_tree;
3320         map_length = length;
3321
3322         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3323                               mirror_num);
3324         BUG_ON(ret);
3325
3326         total_devs = multi->num_stripes;
3327         if (map_length < length) {
3328                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3329                        "len %llu\n", (unsigned long long)logical,
3330                        (unsigned long long)length,
3331                        (unsigned long long)map_length);
3332                 BUG();
3333         }
3334         multi->end_io = first_bio->bi_end_io;
3335         multi->private = first_bio->bi_private;
3336         multi->orig_bio = first_bio;
3337         atomic_set(&multi->stripes_pending, multi->num_stripes);
3338
3339         while (dev_nr < total_devs) {
3340                 if (total_devs > 1) {
3341                         if (dev_nr < total_devs - 1) {
3342                                 bio = bio_clone(first_bio, GFP_NOFS);
3343                                 BUG_ON(!bio);
3344                         } else {
3345                                 bio = first_bio;
3346                         }
3347                         bio->bi_private = multi;
3348                         bio->bi_end_io = end_bio_multi_stripe;
3349                 }
3350                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3351                 dev = multi->stripes[dev_nr].dev;
3352                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3353                         bio->bi_bdev = dev->bdev;
3354                         if (async_submit)
3355                                 schedule_bio(root, dev, rw, bio);
3356                         else
3357                                 submit_bio(rw, bio);
3358                 } else {
3359                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3360                         bio->bi_sector = logical >> 9;
3361                         bio_endio(bio, -EIO);
3362                 }
3363                 dev_nr++;
3364         }
3365         if (total_devs == 1)
3366                 kfree(multi);
3367         return 0;
3368 }
3369
3370 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3371                                        u8 *uuid, u8 *fsid)
3372 {
3373         struct btrfs_device *device;
3374         struct btrfs_fs_devices *cur_devices;
3375
3376         cur_devices = root->fs_info->fs_devices;
3377         while (cur_devices) {
3378                 if (!fsid ||
3379                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3380                         device = __find_device(&cur_devices->devices,
3381                                                devid, uuid);
3382                         if (device)
3383                                 return device;
3384                 }
3385                 cur_devices = cur_devices->seed;
3386         }
3387         return NULL;
3388 }
3389
3390 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3391                                             u64 devid, u8 *dev_uuid)
3392 {
3393         struct btrfs_device *device;
3394         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3395
3396         device = kzalloc(sizeof(*device), GFP_NOFS);
3397         if (!device)
3398                 return NULL;
3399         list_add(&device->dev_list,
3400                  &fs_devices->devices);
3401         device->dev_root = root->fs_info->dev_root;
3402         device->devid = devid;
3403         device->work.func = pending_bios_fn;
3404         device->fs_devices = fs_devices;
3405         device->missing = 1;
3406         fs_devices->num_devices++;
3407         fs_devices->missing_devices++;
3408         spin_lock_init(&device->io_lock);
3409         INIT_LIST_HEAD(&device->dev_alloc_list);
3410         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3411         return device;
3412 }
3413
3414 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3415                           struct extent_buffer *leaf,
3416                           struct btrfs_chunk *chunk)
3417 {
3418         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3419         struct map_lookup *map;
3420         struct extent_map *em;
3421         u64 logical;
3422         u64 length;
3423         u64 devid;
3424         u8 uuid[BTRFS_UUID_SIZE];
3425         int num_stripes;
3426         int ret;
3427         int i;
3428
3429         logical = key->offset;
3430         length = btrfs_chunk_length(leaf, chunk);
3431
3432         read_lock(&map_tree->map_tree.lock);
3433         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3434         read_unlock(&map_tree->map_tree.lock);
3435
3436         /* already mapped? */
3437         if (em && em->start <= logical && em->start + em->len > logical) {
3438                 free_extent_map(em);
3439                 return 0;
3440         } else if (em) {
3441                 free_extent_map(em);
3442         }
3443
3444         em = alloc_extent_map();
3445         if (!em)
3446                 return -ENOMEM;
3447         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3448         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3449         if (!map) {
3450                 free_extent_map(em);
3451                 return -ENOMEM;
3452         }
3453
3454         em->bdev = (struct block_device *)map;
3455         em->start = logical;
3456         em->len = length;
3457         em->block_start = 0;
3458         em->block_len = em->len;
3459
3460         map->num_stripes = num_stripes;
3461         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3462         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3463         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3464         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3465         map->type = btrfs_chunk_type(leaf, chunk);
3466         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3467         for (i = 0; i < num_stripes; i++) {
3468                 map->stripes[i].physical =
3469                         btrfs_stripe_offset_nr(leaf, chunk, i);
3470                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3471                 read_extent_buffer(leaf, uuid, (unsigned long)
3472                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3473                                    BTRFS_UUID_SIZE);
3474                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3475                                                         NULL);
3476                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3477                         kfree(map);
3478                         free_extent_map(em);
3479                         return -EIO;
3480                 }
3481                 if (!map->stripes[i].dev) {
3482                         map->stripes[i].dev =
3483                                 add_missing_dev(root, devid, uuid);
3484                         if (!map->stripes[i].dev) {
3485                                 kfree(map);
3486                                 free_extent_map(em);
3487                                 return -EIO;
3488                         }
3489                 }
3490                 map->stripes[i].dev->in_fs_metadata = 1;
3491         }
3492
3493         write_lock(&map_tree->map_tree.lock);
3494         ret = add_extent_mapping(&map_tree->map_tree, em);
3495         write_unlock(&map_tree->map_tree.lock);
3496         BUG_ON(ret);
3497         free_extent_map(em);
3498
3499         return 0;
3500 }
3501
3502 static int fill_device_from_item(struct extent_buffer *leaf,
3503                                  struct btrfs_dev_item *dev_item,
3504                                  struct btrfs_device *device)
3505 {
3506         unsigned long ptr;
3507
3508         device->devid = btrfs_device_id(leaf, dev_item);
3509         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3510         device->total_bytes = device->disk_total_bytes;
3511         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3512         device->type = btrfs_device_type(leaf, dev_item);
3513         device->io_align = btrfs_device_io_align(leaf, dev_item);
3514         device->io_width = btrfs_device_io_width(leaf, dev_item);
3515         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3516
3517         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3518         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3519
3520         return 0;
3521 }
3522
3523 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3524 {
3525         struct btrfs_fs_devices *fs_devices;
3526         int ret;
3527
3528         mutex_lock(&uuid_mutex);
3529
3530         fs_devices = root->fs_info->fs_devices->seed;
3531         while (fs_devices) {
3532                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3533                         ret = 0;
3534                         goto out;
3535                 }
3536                 fs_devices = fs_devices->seed;
3537         }
3538
3539         fs_devices = find_fsid(fsid);
3540         if (!fs_devices) {
3541                 ret = -ENOENT;
3542                 goto out;
3543         }
3544
3545         fs_devices = clone_fs_devices(fs_devices);
3546         if (IS_ERR(fs_devices)) {
3547                 ret = PTR_ERR(fs_devices);
3548                 goto out;
3549         }
3550
3551         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3552                                    root->fs_info->bdev_holder);
3553         if (ret)
3554                 goto out;
3555
3556         if (!fs_devices->seeding) {
3557                 __btrfs_close_devices(fs_devices);
3558                 free_fs_devices(fs_devices);
3559                 ret = -EINVAL;
3560                 goto out;
3561         }
3562
3563         fs_devices->seed = root->fs_info->fs_devices->seed;
3564         root->fs_info->fs_devices->seed = fs_devices;
3565 out:
3566         mutex_unlock(&uuid_mutex);
3567         return ret;
3568 }
3569
3570 static int read_one_dev(struct btrfs_root *root,
3571                         struct extent_buffer *leaf,
3572                         struct btrfs_dev_item *dev_item)
3573 {
3574         struct btrfs_device *device;
3575         u64 devid;
3576         int ret;
3577         u8 fs_uuid[BTRFS_UUID_SIZE];
3578         u8 dev_uuid[BTRFS_UUID_SIZE];
3579
3580         devid = btrfs_device_id(leaf, dev_item);
3581         read_extent_buffer(leaf, dev_uuid,
3582                            (unsigned long)btrfs_device_uuid(dev_item),
3583                            BTRFS_UUID_SIZE);
3584         read_extent_buffer(leaf, fs_uuid,
3585                            (unsigned long)btrfs_device_fsid(dev_item),
3586                            BTRFS_UUID_SIZE);
3587
3588         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3589                 ret = open_seed_devices(root, fs_uuid);
3590                 if (ret && !btrfs_test_opt(root, DEGRADED))
3591                         return ret;
3592         }
3593
3594         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3595         if (!device || !device->bdev) {
3596                 if (!btrfs_test_opt(root, DEGRADED))
3597                         return -EIO;
3598
3599                 if (!device) {
3600                         printk(KERN_WARNING "warning devid %llu missing\n",
3601                                (unsigned long long)devid);
3602                         device = add_missing_dev(root, devid, dev_uuid);
3603                         if (!device)
3604                                 return -ENOMEM;
3605                 } else if (!device->missing) {
3606                         /*
3607                          * this happens when a device that was properly setup
3608                          * in the device info lists suddenly goes bad.
3609                          * device->bdev is NULL, and so we have to set
3610                          * device->missing to one here
3611                          */
3612                         root->fs_info->fs_devices->missing_devices++;
3613                         device->missing = 1;
3614                 }
3615         }
3616
3617         if (device->fs_devices != root->fs_info->fs_devices) {
3618                 BUG_ON(device->writeable);
3619                 if (device->generation !=
3620                     btrfs_device_generation(leaf, dev_item))
3621                         return -EINVAL;
3622         }
3623
3624         fill_device_from_item(leaf, dev_item, device);
3625         device->dev_root = root->fs_info->dev_root;
3626         device->in_fs_metadata = 1;
3627         if (device->writeable)
3628                 device->fs_devices->total_rw_bytes += device->total_bytes;
3629         ret = 0;
3630         return ret;
3631 }
3632
3633 int btrfs_read_sys_array(struct btrfs_root *root)
3634 {
3635         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3636         struct extent_buffer *sb;
3637         struct btrfs_disk_key *disk_key;
3638         struct btrfs_chunk *chunk;
3639         u8 *ptr;
3640         unsigned long sb_ptr;
3641         int ret = 0;
3642         u32 num_stripes;
3643         u32 array_size;
3644         u32 len = 0;
3645         u32 cur;
3646         struct btrfs_key key;
3647
3648         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3649                                           BTRFS_SUPER_INFO_SIZE);
3650         if (!sb)
3651                 return -ENOMEM;
3652         btrfs_set_buffer_uptodate(sb);
3653         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
3654
3655         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3656         array_size = btrfs_super_sys_array_size(super_copy);
3657
3658         ptr = super_copy->sys_chunk_array;
3659         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3660         cur = 0;
3661
3662         while (cur < array_size) {
3663                 disk_key = (struct btrfs_disk_key *)ptr;
3664                 btrfs_disk_key_to_cpu(&key, disk_key);
3665
3666                 len = sizeof(*disk_key); ptr += len;
3667                 sb_ptr += len;
3668                 cur += len;
3669
3670                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3671                         chunk = (struct btrfs_chunk *)sb_ptr;
3672                         ret = read_one_chunk(root, &key, sb, chunk);
3673                         if (ret)
3674                                 break;
3675                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3676                         len = btrfs_chunk_item_size(num_stripes);
3677                 } else {
3678                         ret = -EIO;
3679                         break;
3680                 }
3681                 ptr += len;
3682                 sb_ptr += len;
3683                 cur += len;
3684         }
3685         free_extent_buffer(sb);
3686         return ret;
3687 }
3688
3689 int btrfs_read_chunk_tree(struct btrfs_root *root)
3690 {
3691         struct btrfs_path *path;
3692         struct extent_buffer *leaf;
3693         struct btrfs_key key;
3694         struct btrfs_key found_key;
3695         int ret;
3696         int slot;
3697
3698         root = root->fs_info->chunk_root;
3699
3700         path = btrfs_alloc_path();
3701         if (!path)
3702                 return -ENOMEM;
3703
3704         /* first we search for all of the device items, and then we
3705          * read in all of the chunk items.  This way we can create chunk
3706          * mappings that reference all of the devices that are afound
3707          */
3708         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3709         key.offset = 0;
3710         key.type = 0;
3711 again:
3712         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3713         if (ret < 0)
3714                 goto error;
3715         while (1) {
3716                 leaf = path->nodes[0];
3717                 slot = path->slots[0];
3718                 if (slot >= btrfs_header_nritems(leaf)) {
3719                         ret = btrfs_next_leaf(root, path);
3720                         if (ret == 0)
3721                                 continue;
3722                         if (ret < 0)
3723                                 goto error;
3724                         break;
3725                 }
3726                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3727                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3728                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3729                                 break;
3730                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3731                                 struct btrfs_dev_item *dev_item;
3732                                 dev_item = btrfs_item_ptr(leaf, slot,
3733                                                   struct btrfs_dev_item);
3734                                 ret = read_one_dev(root, leaf, dev_item);
3735                                 if (ret)
3736                                         goto error;
3737                         }
3738                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3739                         struct btrfs_chunk *chunk;
3740                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3741                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3742                         if (ret)
3743                                 goto error;
3744                 }
3745                 path->slots[0]++;
3746         }
3747         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3748                 key.objectid = 0;
3749                 btrfs_release_path(path);
3750                 goto again;
3751         }
3752         ret = 0;
3753 error:
3754         btrfs_free_path(path);
3755         return ret;
3756 }