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                         goto error_close;
 610
 611                 disk_super = (struct btrfs_super_block *)bh->b_data;
 612                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 613                 if (devid != device->devid)
 614                         goto error_brelse;
 615
 616                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 617                            BTRFS_UUID_SIZE))
 618                         goto error_brelse;
 619
 620                 device->generation = btrfs_super_generation(disk_super);
 621                 if (!latest_transid || device->generation > latest_transid) {
 622                         latest_devid = devid;
 623                         latest_transid = device->generation;
 624                         latest_bdev = bdev;
 625                 }
 626
 627                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 628                         device->writeable = 0;
 629                 } else {
 630                         device->writeable = !bdev_read_only(bdev);
 631                         seeding = 0;
 632                 }
 633
 634                 q = bdev_get_queue(bdev);
 635                 if (blk_queue_discard(q)) {
 636                         device->can_discard = 1;
 637                         fs_devices->num_can_discard++;
 638                 }
 639
 640                 device->bdev = bdev;
 641                 device->in_fs_metadata = 0;
 642                 device->mode = flags;
 643
 644                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 645                         fs_devices->rotating = 1;
 646
 647                 fs_devices->open_devices++;
 648                 if (device->writeable) {
 649                         fs_devices->rw_devices++;
 650                         list_add(&device->dev_alloc_list,
 651                                  &fs_devices->alloc_list);
 652                 }
 653                 brelse(bh);
 654                 continue;
 655
 656 error_brelse:
 657                 brelse(bh);
 658 error_close:
 659                 blkdev_put(bdev, flags);
 660 error:
 661                 continue;
 662         }
 663         if (fs_devices->open_devices == 0) {
 664                 ret = -EINVAL;
 665                 goto out;
 666         }
 667         fs_devices->seeding = seeding;
 668         fs_devices->opened = 1;
 669         fs_devices->latest_bdev = latest_bdev;
 670         fs_devices->latest_devid = latest_devid;
 671         fs_devices->latest_trans = latest_transid;
 672         fs_devices->total_rw_bytes = 0;
 673 out:
 674         return ret;
 675 }
 676
 677 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 678                        fmode_t flags, void *holder)
 679 {
 680         int ret;
 681
 682         mutex_lock(&uuid_mutex);
 683         if (fs_devices->opened) {
 684                 fs_devices->opened++;
 685                 ret = 0;
 686         } else {
 687                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 688         }
 689         mutex_unlock(&uuid_mutex);
 690         return ret;
 691 }
 692
 693 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 694                           struct btrfs_fs_devices **fs_devices_ret)
 695 {
 696         struct btrfs_super_block *disk_super;
 697         struct block_device *bdev;
 698         struct buffer_head *bh;
 699         int ret;
 700         u64 devid;
 701         u64 transid;
 702
 703         mutex_lock(&uuid_mutex);
 704
 705         flags |= FMODE_EXCL;
 706         bdev = blkdev_get_by_path(path, flags, holder);
 707
 708         if (IS_ERR(bdev)) {
 709                 ret = PTR_ERR(bdev);
 710                 goto error;
 711         }
 712
 713         ret = set_blocksize(bdev, 4096);
 714         if (ret)
 715                 goto error_close;
 716         bh = btrfs_read_dev_super(bdev);
 717         if (!bh) {
 718                 ret = -EINVAL;
 719                 goto error_close;
 720         }
 721         disk_super = (struct btrfs_super_block *)bh->b_data;
 722         devid = btrfs_stack_device_id(&disk_super->dev_item);
 723         transid = btrfs_super_generation(disk_super);
 724         if (disk_super->label[0])
 725                 printk(KERN_INFO "device label %s ", disk_super->label);
 726         else
 727                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 728         printk(KERN_CONT "devid %llu transid %llu %s\n",
 729                (unsigned long long)devid, (unsigned long long)transid, path);
 730         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 731
 732         brelse(bh);
 733 error_close:
 734         blkdev_put(bdev, flags);
 735 error:
 736         mutex_unlock(&uuid_mutex);
 737         return ret;
 738 }
 739
 740 /* helper to account the used device space in the range */
 741 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 742                                    u64 end, u64 *length)
 743 {
 744         struct btrfs_key key;
 745         struct btrfs_root *root = device->dev_root;
 746         struct btrfs_dev_extent *dev_extent;
 747         struct btrfs_path *path;
 748         u64 extent_end;
 749         int ret;
 750         int slot;
 751         struct extent_buffer *l;
 752
 753         *length = 0;
 754
 755         if (start >= device->total_bytes)
 756                 return 0;
 757
 758         path = btrfs_alloc_path();
 759         if (!path)
 760                 return -ENOMEM;
 761         path->reada = 2;
 762
 763         key.objectid = device->devid;
 764         key.offset = start;
 765         key.type = BTRFS_DEV_EXTENT_KEY;
 766
 767         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 768         if (ret < 0)
 769                 goto out;
 770         if (ret > 0) {
 771                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 772                 if (ret < 0)
 773                         goto out;
 774         }
 775
 776         while (1) {
 777                 l = path->nodes[0];
 778                 slot = path->slots[0];
 779                 if (slot >= btrfs_header_nritems(l)) {
 780                         ret = btrfs_next_leaf(root, path);
 781                         if (ret == 0)
 782                                 continue;
 783                         if (ret < 0)
 784                                 goto out;
 785
 786                         break;
 787                 }
 788                 btrfs_item_key_to_cpu(l, &key, slot);
 789
 790                 if (key.objectid < device->devid)
 791                         goto next;
 792
 793                 if (key.objectid > device->devid)
 794                         break;
 795
 796                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 797                         goto next;
 798
 799                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 800                 extent_end = key.offset + btrfs_dev_extent_length(l,
 801                                                                   dev_extent);
 802                 if (key.offset <= start && extent_end > end) {
 803                         *length = end - start + 1;
 804                         break;
 805                 } else if (key.offset <= start && extent_end > start)
 806                         *length += extent_end - start;
 807                 else if (key.offset > start && extent_end <= end)
 808                         *length += extent_end - key.offset;
 809                 else if (key.offset > start && key.offset <= end) {
 810                         *length += end - key.offset + 1;
 811                         break;
 812                 } else if (key.offset > end)
 813                         break;
 814
 815 next:
 816                 path->slots[0]++;
 817         }
 818         ret = 0;
 819 out:
 820         btrfs_free_path(path);
 821         return ret;
 822 }
 823
 824 /*
 825  * find_free_dev_extent - find free space in the specified device
 826  * @trans:      transaction handler
 827  * @device:     the device which we search the free space in
 828  * @num_bytes:  the size of the free space that we need
 829  * @start:      store the start of the free space.
 830  * @len:        the size of the free space. that we find, or the size of the max
 831  *              free space if we don't find suitable free space
 832  *
 833  * this uses a pretty simple search, the expectation is that it is
 834  * called very infrequently and that a given device has a small number
 835  * of extents
 836  *
 837  * @start is used to store the start of the free space if we find. But if we
 838  * don't find suitable free space, it will be used to store the start position
 839  * of the max free space.
 840  *
 841  * @len is used to store the size of the free space that we find.
 842  * But if we don't find suitable free space, it is used to store the size of
 843  * the max free space.
 844  */
 845 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 846                          struct btrfs_device *device, u64 num_bytes,
 847                          u64 *start, u64 *len)
 848 {
 849         struct btrfs_key key;
 850         struct btrfs_root *root = device->dev_root;
 851         struct btrfs_dev_extent *dev_extent;
 852         struct btrfs_path *path;
 853         u64 hole_size;
 854         u64 max_hole_start;
 855         u64 max_hole_size;
 856         u64 extent_end;
 857         u64 search_start;
 858         u64 search_end = device->total_bytes;
 859         int ret;
 860         int slot;
 861         struct extent_buffer *l;
 862
 863         /* FIXME use last free of some kind */
 864
 865         /* we don't want to overwrite the superblock on the drive,
 866          * so we make sure to start at an offset of at least 1MB
 867          */
 868         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 869
 870         max_hole_start = search_start;
 871         max_hole_size = 0;
 872         hole_size = 0;
 873
 874         if (search_start >= search_end) {
 875                 ret = -ENOSPC;
 876                 goto error;
 877         }
 878
 879         path = btrfs_alloc_path();
 880         if (!path) {
 881                 ret = -ENOMEM;
 882                 goto error;
 883         }
 884         path->reada = 2;
 885
 886         key.objectid = device->devid;
 887         key.offset = search_start;
 888         key.type = BTRFS_DEV_EXTENT_KEY;
 889
 890         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 891         if (ret < 0)
 892                 goto out;
 893         if (ret > 0) {
 894                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 895                 if (ret < 0)
 896                         goto out;
 897         }
 898
 899         while (1) {
 900                 l = path->nodes[0];
 901                 slot = path->slots[0];
 902                 if (slot >= btrfs_header_nritems(l)) {
 903                         ret = btrfs_next_leaf(root, path);
 904                         if (ret == 0)
 905                                 continue;
 906                         if (ret < 0)
 907                                 goto out;
 908
 909                         break;
 910                 }
 911                 btrfs_item_key_to_cpu(l, &key, slot);
 912
 913                 if (key.objectid < device->devid)
 914                         goto next;
 915
 916                 if (key.objectid > device->devid)
 917                         break;
 918
 919                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 920                         goto next;
 921
 922                 if (key.offset > search_start) {
 923                         hole_size = key.offset - search_start;
 924
 925                         if (hole_size > max_hole_size) {
 926                                 max_hole_start = search_start;
 927                                 max_hole_size = hole_size;
 928                         }
 929
 930                         /*
 931                          * If this free space is greater than which we need,
 932                          * it must be the max free space that we have found
 933                          * until now, so max_hole_start must point to the start
 934                          * of this free space and the length of this free space
 935                          * is stored in max_hole_size. Thus, we return
 936                          * max_hole_start and max_hole_size and go back to the
 937                          * caller.
 938                          */
 939                         if (hole_size >= num_bytes) {
 940                                 ret = 0;
 941                                 goto out;
 942                         }
 943                 }
 944
 945                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 946                 extent_end = key.offset + btrfs_dev_extent_length(l,
 947                                                                   dev_extent);
 948                 if (extent_end > search_start)
 949                         search_start = extent_end;
 950 next:
 951                 path->slots[0]++;
 952                 cond_resched();
 953         }
 954
 955         /*
 956          * At this point, search_start should be the end of
 957          * allocated dev extents, and when shrinking the device,
 958          * search_end may be smaller than search_start.
 959          */
 960         if (search_end > search_start)
 961                 hole_size = search_end - search_start;
 962
 963         if (hole_size > max_hole_size) {
 964                 max_hole_start = search_start;
 965                 max_hole_size = hole_size;
 966         }
 967
 968         /* See above. */
 969         if (hole_size < num_bytes)
 970                 ret = -ENOSPC;
 971         else
 972                 ret = 0;
 973
 974 out:
 975         btrfs_free_path(path);
 976 error:
 977         *start = max_hole_start;
 978         if (len)
 979                 *len = max_hole_size;
 980         return ret;
 981 }
 982
 983 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 984                           struct btrfs_device *device,
 985                           u64 start)
 986 {
 987         int ret;
 988         struct btrfs_path *path;
 989         struct btrfs_root *root = device->dev_root;
 990         struct btrfs_key key;
 991         struct btrfs_key found_key;
 992         struct extent_buffer *leaf = NULL;
 993         struct btrfs_dev_extent *extent = NULL;
 994
 995         path = btrfs_alloc_path();
 996         if (!path)
 997                 return -ENOMEM;
 998
 999         key.objectid = device->devid;
1000         key.offset = start;
1001         key.type = BTRFS_DEV_EXTENT_KEY;
1002
1003         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1004         if (ret > 0) {
1005                 ret = btrfs_previous_item(root, path, key.objectid,
1006                                           BTRFS_DEV_EXTENT_KEY);
1007                 if (ret)
1008                         goto out;
1009                 leaf = path->nodes[0];
1010                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1011                 extent = btrfs_item_ptr(leaf, path->slots[0],
1012                                         struct btrfs_dev_extent);
1013                 BUG_ON(found_key.offset > start || found_key.offset +
1014                        btrfs_dev_extent_length(leaf, extent) < start);
1015         } else if (ret == 0) {
1016                 leaf = path->nodes[0];
1017                 extent = btrfs_item_ptr(leaf, path->slots[0],
1018                                         struct btrfs_dev_extent);
1019         }
1020         BUG_ON(ret);
1021
1022         if (device->bytes_used > 0) {
1023                 u64 len = btrfs_dev_extent_length(leaf, extent);
1024                 device->bytes_used -= len;
1025                 spin_lock(&root->fs_info->free_chunk_lock);
1026                 root->fs_info->free_chunk_space += len;
1027                 spin_unlock(&root->fs_info->free_chunk_lock);
1028         }
1029         ret = btrfs_del_item(trans, root, path);
1030
1031 out:
1032         btrfs_free_path(path);
1033         return ret;
1034 }
1035
1036 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1037                            struct btrfs_device *device,
1038                            u64 chunk_tree, u64 chunk_objectid,
1039                            u64 chunk_offset, u64 start, u64 num_bytes)
1040 {
1041         int ret;
1042         struct btrfs_path *path;
1043         struct btrfs_root *root = device->dev_root;
1044         struct btrfs_dev_extent *extent;
1045         struct extent_buffer *leaf;
1046         struct btrfs_key key;
1047
1048         WARN_ON(!device->in_fs_metadata);
1049         path = btrfs_alloc_path();
1050         if (!path)
1051                 return -ENOMEM;
1052
1053         key.objectid = device->devid;
1054         key.offset = start;
1055         key.type = BTRFS_DEV_EXTENT_KEY;
1056         ret = btrfs_insert_empty_item(trans, root, path, &key,
1057                                       sizeof(*extent));
1058         BUG_ON(ret);
1059
1060         leaf = path->nodes[0];
1061         extent = btrfs_item_ptr(leaf, path->slots[0],
1062                                 struct btrfs_dev_extent);
1063         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1064         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1065         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1066
1067         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1068                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1069                     BTRFS_UUID_SIZE);
1070
1071         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1072         btrfs_mark_buffer_dirty(leaf);
1073         btrfs_free_path(path);
1074         return ret;
1075 }
1076
1077 static noinline int find_next_chunk(struct btrfs_root *root,
1078                                     u64 objectid, u64 *offset)
1079 {
1080         struct btrfs_path *path;
1081         int ret;
1082         struct btrfs_key key;
1083         struct btrfs_chunk *chunk;
1084         struct btrfs_key found_key;
1085
1086         path = btrfs_alloc_path();
1087         if (!path)
1088                 return -ENOMEM;
1089
1090         key.objectid = objectid;
1091         key.offset = (u64)-1;
1092         key.type = BTRFS_CHUNK_ITEM_KEY;
1093
1094         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1095         if (ret < 0)
1096                 goto error;
1097
1098         BUG_ON(ret == 0);
1099
1100         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1101         if (ret) {
1102                 *offset = 0;
1103         } else {
1104                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1105                                       path->slots[0]);
1106                 if (found_key.objectid != objectid)
1107                         *offset = 0;
1108                 else {
1109                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1110                                                struct btrfs_chunk);
1111                         *offset = found_key.offset +
1112                                 btrfs_chunk_length(path->nodes[0], chunk);
1113                 }
1114         }
1115         ret = 0;
1116 error:
1117         btrfs_free_path(path);
1118         return ret;
1119 }
1120
1121 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1122 {
1123         int ret;
1124         struct btrfs_key key;
1125         struct btrfs_key found_key;
1126         struct btrfs_path *path;
1127
1128         root = root->fs_info->chunk_root;
1129
1130         path = btrfs_alloc_path();
1131         if (!path)
1132                 return -ENOMEM;
1133
1134         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1135         key.type = BTRFS_DEV_ITEM_KEY;
1136         key.offset = (u64)-1;
1137
1138         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1139         if (ret < 0)
1140                 goto error;
1141
1142         BUG_ON(ret == 0);
1143
1144         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1145                                   BTRFS_DEV_ITEM_KEY);
1146         if (ret) {
1147                 *objectid = 1;
1148         } else {
1149                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1150                                       path->slots[0]);
1151                 *objectid = found_key.offset + 1;
1152         }
1153         ret = 0;
1154 error:
1155         btrfs_free_path(path);
1156         return ret;
1157 }
1158
1159 /*
1160  * the device information is stored in the chunk root
1161  * the btrfs_device struct should be fully filled in
1162  */
1163 int btrfs_add_device(struct btrfs_trans_handle *trans,
1164                      struct btrfs_root *root,
1165                      struct btrfs_device *device)
1166 {
1167         int ret;
1168         struct btrfs_path *path;
1169         struct btrfs_dev_item *dev_item;
1170         struct extent_buffer *leaf;
1171         struct btrfs_key key;
1172         unsigned long ptr;
1173
1174         root = root->fs_info->chunk_root;
1175
1176         path = btrfs_alloc_path();
1177         if (!path)
1178                 return -ENOMEM;
1179
1180         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1181         key.type = BTRFS_DEV_ITEM_KEY;
1182         key.offset = device->devid;
1183
1184         ret = btrfs_insert_empty_item(trans, root, path, &key,
1185                                       sizeof(*dev_item));
1186         if (ret)
1187                 goto out;
1188
1189         leaf = path->nodes[0];
1190         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1191
1192         btrfs_set_device_id(leaf, dev_item, device->devid);
1193         btrfs_set_device_generation(leaf, dev_item, 0);
1194         btrfs_set_device_type(leaf, dev_item, device->type);
1195         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1196         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1197         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1198         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1199         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1200         btrfs_set_device_group(leaf, dev_item, 0);
1201         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1202         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1203         btrfs_set_device_start_offset(leaf, dev_item, 0);
1204
1205         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1206         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1207         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1208         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1209         btrfs_mark_buffer_dirty(leaf);
1210
1211         ret = 0;
1212 out:
1213         btrfs_free_path(path);
1214         return ret;
1215 }
1216
1217 static int btrfs_rm_dev_item(struct btrfs_root *root,
1218                              struct btrfs_device *device)
1219 {
1220         int ret;
1221         struct btrfs_path *path;
1222         struct btrfs_key key;
1223         struct btrfs_trans_handle *trans;
1224
1225         root = root->fs_info->chunk_root;
1226
1227         path = btrfs_alloc_path();
1228         if (!path)
1229                 return -ENOMEM;
1230
1231         trans = btrfs_start_transaction(root, 0);
1232         if (IS_ERR(trans)) {
1233                 btrfs_free_path(path);
1234                 return PTR_ERR(trans);
1235         }
1236         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1237         key.type = BTRFS_DEV_ITEM_KEY;
1238         key.offset = device->devid;
1239         lock_chunks(root);
1240
1241         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1242         if (ret < 0)
1243                 goto out;
1244
1245         if (ret > 0) {
1246                 ret = -ENOENT;
1247                 goto out;
1248         }
1249
1250         ret = btrfs_del_item(trans, root, path);
1251         if (ret)
1252                 goto out;
1253 out:
1254         btrfs_free_path(path);
1255         unlock_chunks(root);
1256         btrfs_commit_transaction(trans, root);
1257         return ret;
1258 }
1259
1260 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1261 {
1262         struct btrfs_device *device;
1263         struct btrfs_device *next_device;
1264         struct block_device *bdev;
1265         struct buffer_head *bh = NULL;
1266         struct btrfs_super_block *disk_super;
1267         struct btrfs_fs_devices *cur_devices;
1268         u64 all_avail;
1269         u64 devid;
1270         u64 num_devices;
1271         u8 *dev_uuid;
1272         int ret = 0;
1273         bool clear_super = false;
1274
1275         mutex_lock(&uuid_mutex);
1276         mutex_lock(&root->fs_info->volume_mutex);
1277
1278         all_avail = root->fs_info->avail_data_alloc_bits |
1279                 root->fs_info->avail_system_alloc_bits |
1280                 root->fs_info->avail_metadata_alloc_bits;
1281
1282         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1283             root->fs_info->fs_devices->num_devices <= 4) {
1284                 printk(KERN_ERR "btrfs: unable to go below four devices "
1285                        "on raid10\n");
1286                 ret = -EINVAL;
1287                 goto out;
1288         }
1289
1290         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1291             root->fs_info->fs_devices->num_devices <= 2) {
1292                 printk(KERN_ERR "btrfs: unable to go below two "
1293                        "devices on raid1\n");
1294                 ret = -EINVAL;
1295                 goto out;
1296         }
1297
1298         if (strcmp(device_path, "missing") == 0) {
1299                 struct list_head *devices;
1300                 struct btrfs_device *tmp;
1301
1302                 device = NULL;
1303                 devices = &root->fs_info->fs_devices->devices;
1304                 /*
1305                  * It is safe to read the devices since the volume_mutex
1306                  * is held.
1307                  */
1308                 list_for_each_entry(tmp, devices, dev_list) {
1309                         if (tmp->in_fs_metadata && !tmp->bdev) {
1310                                 device = tmp;
1311                                 break;
1312                         }
1313                 }
1314                 bdev = NULL;
1315                 bh = NULL;
1316                 disk_super = NULL;
1317                 if (!device) {
1318                         printk(KERN_ERR "btrfs: no missing devices found to "
1319                                "remove\n");
1320                         goto out;
1321                 }
1322         } else {
1323                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1324                                           root->fs_info->bdev_holder);
1325                 if (IS_ERR(bdev)) {
1326                         ret = PTR_ERR(bdev);
1327                         goto out;
1328                 }
1329
1330                 set_blocksize(bdev, 4096);
1331                 bh = btrfs_read_dev_super(bdev);
1332                 if (!bh) {
1333                         ret = -EINVAL;
1334                         goto error_close;
1335                 }
1336                 disk_super = (struct btrfs_super_block *)bh->b_data;
1337                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1338                 dev_uuid = disk_super->dev_item.uuid;
1339                 device = btrfs_find_device(root, devid, dev_uuid,
1340                                            disk_super->fsid);
1341                 if (!device) {
1342                         ret = -ENOENT;
1343                         goto error_brelse;
1344                 }
1345         }
1346
1347         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1348                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1349                        "device\n");
1350                 ret = -EINVAL;
1351                 goto error_brelse;
1352         }
1353
1354         if (device->writeable) {
1355                 lock_chunks(root);
1356                 list_del_init(&device->dev_alloc_list);
1357                 unlock_chunks(root);
1358                 root->fs_info->fs_devices->rw_devices--;
1359                 clear_super = true;
1360         }
1361
1362         ret = btrfs_shrink_device(device, 0);
1363         if (ret)
1364                 goto error_undo;
1365
1366         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1367         if (ret)
1368                 goto error_undo;
1369
1370         spin_lock(&root->fs_info->free_chunk_lock);
1371         root->fs_info->free_chunk_space = device->total_bytes -
1372                 device->bytes_used;
1373         spin_unlock(&root->fs_info->free_chunk_lock);
1374
1375         device->in_fs_metadata = 0;
1376         btrfs_scrub_cancel_dev(root, device);
1377
1378         /*
1379          * the device list mutex makes sure that we don't change
1380          * the device list while someone else is writing out all
1381          * the device supers.
1382          */
1383
1384         cur_devices = device->fs_devices;
1385         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1386         list_del_rcu(&device->dev_list);
1387
1388         device->fs_devices->num_devices--;
1389
1390         if (device->missing)
1391                 root->fs_info->fs_devices->missing_devices--;
1392
1393         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1394                                  struct btrfs_device, dev_list);
1395         if (device->bdev == root->fs_info->sb->s_bdev)
1396                 root->fs_info->sb->s_bdev = next_device->bdev;
1397         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1398                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1399
1400         if (device->bdev)
1401                 device->fs_devices->open_devices--;
1402
1403         call_rcu(&device->rcu, free_device);
1404         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1405
1406         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1407         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1408
1409         if (cur_devices->open_devices == 0) {
1410                 struct btrfs_fs_devices *fs_devices;
1411                 fs_devices = root->fs_info->fs_devices;
1412                 while (fs_devices) {
1413                         if (fs_devices->seed == cur_devices)
1414                                 break;
1415                         fs_devices = fs_devices->seed;
1416                 }
1417                 fs_devices->seed = cur_devices->seed;
1418                 cur_devices->seed = NULL;
1419                 lock_chunks(root);
1420                 __btrfs_close_devices(cur_devices);
1421                 unlock_chunks(root);
1422                 free_fs_devices(cur_devices);
1423         }
1424
1425         /*
1426          * at this point, the device is zero sized.  We want to
1427          * remove it from the devices list and zero out the old super
1428          */
1429         if (clear_super) {
1430                 /* make sure this device isn't detected as part of
1431                  * the FS anymore
1432                  */
1433                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1434                 set_buffer_dirty(bh);
1435                 sync_dirty_buffer(bh);
1436         }
1437
1438         ret = 0;
1439
1440 error_brelse:
1441         brelse(bh);
1442 error_close:
1443         if (bdev)
1444                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1445 out:
1446         mutex_unlock(&root->fs_info->volume_mutex);
1447         mutex_unlock(&uuid_mutex);
1448         return ret;
1449 error_undo:
1450         if (device->writeable) {
1451                 lock_chunks(root);
1452                 list_add(&device->dev_alloc_list,
1453                          &root->fs_info->fs_devices->alloc_list);
1454                 unlock_chunks(root);
1455                 root->fs_info->fs_devices->rw_devices++;
1456         }
1457         goto error_brelse;
1458 }
1459
1460 /*
1461  * does all the dirty work required for changing file system's UUID.
1462  */
1463 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1464                                 struct btrfs_root *root)
1465 {
1466         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1467         struct btrfs_fs_devices *old_devices;
1468         struct btrfs_fs_devices *seed_devices;
1469         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1470         struct btrfs_device *device;
1471         u64 super_flags;
1472
1473         BUG_ON(!mutex_is_locked(&uuid_mutex));
1474         if (!fs_devices->seeding)
1475                 return -EINVAL;
1476
1477         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1478         if (!seed_devices)
1479                 return -ENOMEM;
1480
1481         old_devices = clone_fs_devices(fs_devices);
1482         if (IS_ERR(old_devices)) {
1483                 kfree(seed_devices);
1484                 return PTR_ERR(old_devices);
1485         }
1486
1487         list_add(&old_devices->list, &fs_uuids);
1488
1489         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1490         seed_devices->opened = 1;
1491         INIT_LIST_HEAD(&seed_devices->devices);
1492         INIT_LIST_HEAD(&seed_devices->alloc_list);
1493         mutex_init(&seed_devices->device_list_mutex);
1494
1495         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1496         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1497                               synchronize_rcu);
1498         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1499
1500         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1501         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1502                 device->fs_devices = seed_devices;
1503         }
1504
1505         fs_devices->seeding = 0;
1506         fs_devices->num_devices = 0;
1507         fs_devices->open_devices = 0;
1508         fs_devices->seed = seed_devices;
1509
1510         generate_random_uuid(fs_devices->fsid);
1511         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1512         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1513         super_flags = btrfs_super_flags(disk_super) &
1514                       ~BTRFS_SUPER_FLAG_SEEDING;
1515         btrfs_set_super_flags(disk_super, super_flags);
1516
1517         return 0;
1518 }
1519
1520 /*
1521  * strore the expected generation for seed devices in device items.
1522  */
1523 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1524                                struct btrfs_root *root)
1525 {
1526         struct btrfs_path *path;
1527         struct extent_buffer *leaf;
1528         struct btrfs_dev_item *dev_item;
1529         struct btrfs_device *device;
1530         struct btrfs_key key;
1531         u8 fs_uuid[BTRFS_UUID_SIZE];
1532         u8 dev_uuid[BTRFS_UUID_SIZE];
1533         u64 devid;
1534         int ret;
1535
1536         path = btrfs_alloc_path();
1537         if (!path)
1538                 return -ENOMEM;
1539
1540         root = root->fs_info->chunk_root;
1541         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1542         key.offset = 0;
1543         key.type = BTRFS_DEV_ITEM_KEY;
1544
1545         while (1) {
1546                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1547                 if (ret < 0)
1548                         goto error;
1549
1550                 leaf = path->nodes[0];
1551 next_slot:
1552                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1553                         ret = btrfs_next_leaf(root, path);
1554                         if (ret > 0)
1555                                 break;
1556                         if (ret < 0)
1557                                 goto error;
1558                         leaf = path->nodes[0];
1559                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1560                         btrfs_release_path(path);
1561                         continue;
1562                 }
1563
1564                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1565                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1566                     key.type != BTRFS_DEV_ITEM_KEY)
1567                         break;
1568
1569                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1570                                           struct btrfs_dev_item);
1571                 devid = btrfs_device_id(leaf, dev_item);
1572                 read_extent_buffer(leaf, dev_uuid,
1573                                    (unsigned long)btrfs_device_uuid(dev_item),
1574                                    BTRFS_UUID_SIZE);
1575                 read_extent_buffer(leaf, fs_uuid,
1576                                    (unsigned long)btrfs_device_fsid(dev_item),
1577                                    BTRFS_UUID_SIZE);
1578                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1579                 BUG_ON(!device);
1580
1581                 if (device->fs_devices->seeding) {
1582                         btrfs_set_device_generation(leaf, dev_item,
1583                                                     device->generation);
1584                         btrfs_mark_buffer_dirty(leaf);
1585                 }
1586
1587                 path->slots[0]++;
1588                 goto next_slot;
1589         }
1590         ret = 0;
1591 error:
1592         btrfs_free_path(path);
1593         return ret;
1594 }
1595
1596 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1597 {
1598         struct request_queue *q;
1599         struct btrfs_trans_handle *trans;
1600         struct btrfs_device *device;
1601         struct block_device *bdev;
1602         struct list_head *devices;
1603         struct super_block *sb = root->fs_info->sb;
1604         u64 total_bytes;
1605         int seeding_dev = 0;
1606         int ret = 0;
1607
1608         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1609                 return -EINVAL;
1610
1611         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1612                                   root->fs_info->bdev_holder);
1613         if (IS_ERR(bdev))
1614                 return PTR_ERR(bdev);
1615
1616         if (root->fs_info->fs_devices->seeding) {
1617                 seeding_dev = 1;
1618                 down_write(&sb->s_umount);
1619                 mutex_lock(&uuid_mutex);
1620         }
1621
1622         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1623         mutex_lock(&root->fs_info->volume_mutex);
1624
1625         devices = &root->fs_info->fs_devices->devices;
1626         /*
1627          * we have the volume lock, so we don't need the extra
1628          * device list mutex while reading the list here.
1629          */
1630         list_for_each_entry(device, devices, dev_list) {
1631                 if (device->bdev == bdev) {
1632                         ret = -EEXIST;
1633                         goto error;
1634                 }
1635         }
1636
1637         device = kzalloc(sizeof(*device), GFP_NOFS);
1638         if (!device) {
1639                 /* we can safely leave the fs_devices entry around */
1640                 ret = -ENOMEM;
1641                 goto error;
1642         }
1643
1644         device->name = kstrdup(device_path, GFP_NOFS);
1645         if (!device->name) {
1646                 kfree(device);
1647                 ret = -ENOMEM;
1648                 goto error;
1649         }
1650
1651         ret = find_next_devid(root, &device->devid);
1652         if (ret) {
1653                 kfree(device->name);
1654                 kfree(device);
1655                 goto error;
1656         }
1657
1658         trans = btrfs_start_transaction(root, 0);
1659         if (IS_ERR(trans)) {
1660                 kfree(device->name);
1661                 kfree(device);
1662                 ret = PTR_ERR(trans);
1663                 goto error;
1664         }
1665
1666         lock_chunks(root);
1667
1668         q = bdev_get_queue(bdev);
1669         if (blk_queue_discard(q))
1670                 device->can_discard = 1;
1671         device->writeable = 1;
1672         device->work.func = pending_bios_fn;
1673         generate_random_uuid(device->uuid);
1674         spin_lock_init(&device->io_lock);
1675         device->generation = trans->transid;
1676         device->io_width = root->sectorsize;
1677         device->io_align = root->sectorsize;
1678         device->sector_size = root->sectorsize;
1679         device->total_bytes = i_size_read(bdev->bd_inode);
1680         device->disk_total_bytes = device->total_bytes;
1681         device->dev_root = root->fs_info->dev_root;
1682         device->bdev = bdev;
1683         device->in_fs_metadata = 1;
1684         device->mode = FMODE_EXCL;
1685         set_blocksize(device->bdev, 4096);
1686
1687         if (seeding_dev) {
1688                 sb->s_flags &= ~MS_RDONLY;
1689                 ret = btrfs_prepare_sprout(trans, root);
1690                 BUG_ON(ret);
1691         }
1692
1693         device->fs_devices = root->fs_info->fs_devices;
1694
1695         /*
1696          * we don't want write_supers to jump in here with our device
1697          * half setup
1698          */
1699         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1700         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1701         list_add(&device->dev_alloc_list,
1702                  &root->fs_info->fs_devices->alloc_list);
1703         root->fs_info->fs_devices->num_devices++;
1704         root->fs_info->fs_devices->open_devices++;
1705         root->fs_info->fs_devices->rw_devices++;
1706         if (device->can_discard)
1707                 root->fs_info->fs_devices->num_can_discard++;
1708         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1709
1710         spin_lock(&root->fs_info->free_chunk_lock);
1711         root->fs_info->free_chunk_space += device->total_bytes;
1712         spin_unlock(&root->fs_info->free_chunk_lock);
1713
1714         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1715                 root->fs_info->fs_devices->rotating = 1;
1716
1717         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1718         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1719                                     total_bytes + device->total_bytes);
1720
1721         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1722         btrfs_set_super_num_devices(root->fs_info->super_copy,
1723                                     total_bytes + 1);
1724         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1725
1726         if (seeding_dev) {
1727                 ret = init_first_rw_device(trans, root, device);
1728                 BUG_ON(ret);
1729                 ret = btrfs_finish_sprout(trans, root);
1730                 BUG_ON(ret);
1731         } else {
1732                 ret = btrfs_add_device(trans, root, device);
1733         }
1734
1735         /*
1736          * we've got more storage, clear any full flags on the space
1737          * infos
1738          */
1739         btrfs_clear_space_info_full(root->fs_info);
1740
1741         unlock_chunks(root);
1742         btrfs_commit_transaction(trans, root);
1743
1744         if (seeding_dev) {
1745                 mutex_unlock(&uuid_mutex);
1746                 up_write(&sb->s_umount);
1747
1748                 ret = btrfs_relocate_sys_chunks(root);
1749                 BUG_ON(ret);
1750         }
1751 out:
1752         mutex_unlock(&root->fs_info->volume_mutex);
1753         return ret;
1754 error:
1755         blkdev_put(bdev, FMODE_EXCL);
1756         if (seeding_dev) {
1757                 mutex_unlock(&uuid_mutex);
1758                 up_write(&sb->s_umount);
1759         }
1760         goto out;
1761 }
1762
1763 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1764                                         struct btrfs_device *device)
1765 {
1766         int ret;
1767         struct btrfs_path *path;
1768         struct btrfs_root *root;
1769         struct btrfs_dev_item *dev_item;
1770         struct extent_buffer *leaf;
1771         struct btrfs_key key;
1772
1773         root = device->dev_root->fs_info->chunk_root;
1774
1775         path = btrfs_alloc_path();
1776         if (!path)
1777                 return -ENOMEM;
1778
1779         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1780         key.type = BTRFS_DEV_ITEM_KEY;
1781         key.offset = device->devid;
1782
1783         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1784         if (ret < 0)
1785                 goto out;
1786
1787         if (ret > 0) {
1788                 ret = -ENOENT;
1789                 goto out;
1790         }
1791
1792         leaf = path->nodes[0];
1793         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1794
1795         btrfs_set_device_id(leaf, dev_item, device->devid);
1796         btrfs_set_device_type(leaf, dev_item, device->type);
1797         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1798         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1799         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1800         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1801         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1802         btrfs_mark_buffer_dirty(leaf);
1803
1804 out:
1805         btrfs_free_path(path);
1806         return ret;
1807 }
1808
1809 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1810                       struct btrfs_device *device, u64 new_size)
1811 {
1812         struct btrfs_super_block *super_copy =
1813                 device->dev_root->fs_info->super_copy;
1814         u64 old_total = btrfs_super_total_bytes(super_copy);
1815         u64 diff = new_size - device->total_bytes;
1816
1817         if (!device->writeable)
1818                 return -EACCES;
1819         if (new_size <= device->total_bytes)
1820                 return -EINVAL;
1821
1822         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1823         device->fs_devices->total_rw_bytes += diff;
1824
1825         device->total_bytes = new_size;
1826         device->disk_total_bytes = new_size;
1827         btrfs_clear_space_info_full(device->dev_root->fs_info);
1828
1829         return btrfs_update_device(trans, device);
1830 }
1831
1832 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1833                       struct btrfs_device *device, u64 new_size)
1834 {
1835         int ret;
1836         lock_chunks(device->dev_root);
1837         ret = __btrfs_grow_device(trans, device, new_size);
1838         unlock_chunks(device->dev_root);
1839         return ret;
1840 }
1841
1842 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1843                             struct btrfs_root *root,
1844                             u64 chunk_tree, u64 chunk_objectid,
1845                             u64 chunk_offset)
1846 {
1847         int ret;
1848         struct btrfs_path *path;
1849         struct btrfs_key key;
1850
1851         root = root->fs_info->chunk_root;
1852         path = btrfs_alloc_path();
1853         if (!path)
1854                 return -ENOMEM;
1855
1856         key.objectid = chunk_objectid;
1857         key.offset = chunk_offset;
1858         key.type = BTRFS_CHUNK_ITEM_KEY;
1859
1860         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1861         BUG_ON(ret);
1862
1863         ret = btrfs_del_item(trans, root, path);
1864
1865         btrfs_free_path(path);
1866         return ret;
1867 }
1868
1869 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1870                         chunk_offset)
1871 {
1872         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1873         struct btrfs_disk_key *disk_key;
1874         struct btrfs_chunk *chunk;
1875         u8 *ptr;
1876         int ret = 0;
1877         u32 num_stripes;
1878         u32 array_size;
1879         u32 len = 0;
1880         u32 cur;
1881         struct btrfs_key key;
1882
1883         array_size = btrfs_super_sys_array_size(super_copy);
1884
1885         ptr = super_copy->sys_chunk_array;
1886         cur = 0;
1887
1888         while (cur < array_size) {
1889                 disk_key = (struct btrfs_disk_key *)ptr;
1890                 btrfs_disk_key_to_cpu(&key, disk_key);
1891
1892                 len = sizeof(*disk_key);
1893
1894                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1895                         chunk = (struct btrfs_chunk *)(ptr + len);
1896                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1897                         len += btrfs_chunk_item_size(num_stripes);
1898                 } else {
1899                         ret = -EIO;
1900                         break;
1901                 }
1902                 if (key.objectid == chunk_objectid &&
1903                     key.offset == chunk_offset) {
1904                         memmove(ptr, ptr + len, array_size - (cur + len));
1905                         array_size -= len;
1906                         btrfs_set_super_sys_array_size(super_copy, array_size);
1907                 } else {
1908                         ptr += len;
1909                         cur += len;
1910                 }
1911         }
1912         return ret;
1913 }
1914
1915 static int btrfs_relocate_chunk(struct btrfs_root *root,
1916                          u64 chunk_tree, u64 chunk_objectid,
1917                          u64 chunk_offset)
1918 {
1919         struct extent_map_tree *em_tree;
1920         struct btrfs_root *extent_root;
1921         struct btrfs_trans_handle *trans;
1922         struct extent_map *em;
1923         struct map_lookup *map;
1924         int ret;
1925         int i;
1926
1927         root = root->fs_info->chunk_root;
1928         extent_root = root->fs_info->extent_root;
1929         em_tree = &root->fs_info->mapping_tree.map_tree;
1930
1931         ret = btrfs_can_relocate(extent_root, chunk_offset);
1932         if (ret)
1933                 return -ENOSPC;
1934
1935         /* step one, relocate all the extents inside this chunk */
1936         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1937         if (ret)
1938                 return ret;
1939
1940         trans = btrfs_start_transaction(root, 0);
1941         BUG_ON(IS_ERR(trans));
1942
1943         lock_chunks(root);
1944
1945         /*
1946          * step two, delete the device extents and the
1947          * chunk tree entries
1948          */
1949         read_lock(&em_tree->lock);
1950         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1951         read_unlock(&em_tree->lock);
1952
1953         BUG_ON(em->start > chunk_offset ||
1954                em->start + em->len < chunk_offset);
1955         map = (struct map_lookup *)em->bdev;
1956
1957         for (i = 0; i < map->num_stripes; i++) {
1958                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1959                                             map->stripes[i].physical);
1960                 BUG_ON(ret);
1961
1962                 if (map->stripes[i].dev) {
1963                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1964                         BUG_ON(ret);
1965                 }
1966         }
1967         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1968                                chunk_offset);
1969
1970         BUG_ON(ret);
1971
1972         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1973
1974         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1975                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1976                 BUG_ON(ret);
1977         }
1978
1979         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1980         BUG_ON(ret);
1981
1982         write_lock(&em_tree->lock);
1983         remove_extent_mapping(em_tree, em);
1984         write_unlock(&em_tree->lock);
1985
1986         kfree(map);
1987         em->bdev = NULL;
1988
1989         /* once for the tree */
1990         free_extent_map(em);
1991         /* once for us */
1992         free_extent_map(em);
1993
1994         unlock_chunks(root);
1995         btrfs_end_transaction(trans, root);
1996         return 0;
1997 }
1998
1999 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2000 {
2001         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2002         struct btrfs_path *path;
2003         struct extent_buffer *leaf;
2004         struct btrfs_chunk *chunk;
2005         struct btrfs_key key;
2006         struct btrfs_key found_key;
2007         u64 chunk_tree = chunk_root->root_key.objectid;
2008         u64 chunk_type;
2009         bool retried = false;
2010         int failed = 0;
2011         int ret;
2012
2013         path = btrfs_alloc_path();
2014         if (!path)
2015                 return -ENOMEM;
2016
2017 again:
2018         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2019         key.offset = (u64)-1;
2020         key.type = BTRFS_CHUNK_ITEM_KEY;
2021
2022         while (1) {
2023                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2024                 if (ret < 0)
2025                         goto error;
2026                 BUG_ON(ret == 0);
2027
2028                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2029                                           key.type);
2030                 if (ret < 0)
2031                         goto error;
2032                 if (ret > 0)
2033                         break;
2034
2035                 leaf = path->nodes[0];
2036                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2037
2038                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2039                                        struct btrfs_chunk);
2040                 chunk_type = btrfs_chunk_type(leaf, chunk);
2041                 btrfs_release_path(path);
2042
2043                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2044                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2045                                                    found_key.objectid,
2046                                                    found_key.offset);
2047                         if (ret == -ENOSPC)
2048                                 failed++;
2049                         else if (ret)
2050                                 BUG();
2051                 }
2052
2053                 if (found_key.offset == 0)
2054                         break;
2055                 key.offset = found_key.offset - 1;
2056         }
2057         ret = 0;
2058         if (failed && !retried) {
2059                 failed = 0;
2060                 retried = true;
2061                 goto again;
2062         } else if (failed && retried) {
2063                 WARN_ON(1);
2064                 ret = -ENOSPC;
2065         }
2066 error:
2067         btrfs_free_path(path);
2068         return ret;
2069 }
2070
2071 static u64 div_factor(u64 num, int factor)
2072 {
2073         if (factor == 10)
2074                 return num;
2075         num *= factor;
2076         do_div(num, 10);
2077         return num;
2078 }
2079
2080 int btrfs_balance(struct btrfs_root *dev_root)
2081 {
2082         int ret;
2083         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2084         struct btrfs_device *device;
2085         u64 old_size;
2086         u64 size_to_free;
2087         struct btrfs_path *path;
2088         struct btrfs_key key;
2089         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2090         struct btrfs_trans_handle *trans;
2091         struct btrfs_key found_key;
2092
2093         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2094                 return -EROFS;
2095
2096         if (!capable(CAP_SYS_ADMIN))
2097                 return -EPERM;
2098
2099         mutex_lock(&dev_root->fs_info->volume_mutex);
2100         dev_root = dev_root->fs_info->dev_root;
2101
2102         /* step one make some room on all the devices */
2103         list_for_each_entry(device, devices, dev_list) {
2104                 old_size = device->total_bytes;
2105                 size_to_free = div_factor(old_size, 1);
2106                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2107                 if (!device->writeable ||
2108                     device->total_bytes - device->bytes_used > size_to_free)
2109                         continue;
2110
2111                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2112                 if (ret == -ENOSPC)
2113                         break;
2114                 BUG_ON(ret);
2115
2116                 trans = btrfs_start_transaction(dev_root, 0);
2117                 BUG_ON(IS_ERR(trans));
2118
2119                 ret = btrfs_grow_device(trans, device, old_size);
2120                 BUG_ON(ret);
2121
2122                 btrfs_end_transaction(trans, dev_root);
2123         }
2124
2125         /* step two, relocate all the chunks */
2126         path = btrfs_alloc_path();
2127         if (!path) {
2128                 ret = -ENOMEM;
2129                 goto error;
2130         }
2131         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2132         key.offset = (u64)-1;
2133         key.type = BTRFS_CHUNK_ITEM_KEY;
2134
2135         while (1) {
2136                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2137                 if (ret < 0)
2138                         goto error;
2139
2140                 /*
2141                  * this shouldn't happen, it means the last relocate
2142                  * failed
2143                  */
2144                 if (ret == 0)
2145                         break;
2146
2147                 ret = btrfs_previous_item(chunk_root, path, 0,
2148                                           BTRFS_CHUNK_ITEM_KEY);
2149                 if (ret)
2150                         break;
2151
2152                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2153                                       path->slots[0]);
2154                 if (found_key.objectid != key.objectid)
2155                         break;
2156
2157                 /* chunk zero is special */
2158                 if (found_key.offset == 0)
2159                         break;
2160
2161                 btrfs_release_path(path);
2162                 ret = btrfs_relocate_chunk(chunk_root,
2163                                            chunk_root->root_key.objectid,
2164                                            found_key.objectid,
2165                                            found_key.offset);
2166                 if (ret && ret != -ENOSPC)
2167                         goto error;
2168                 key.offset = found_key.offset - 1;
2169         }
2170         ret = 0;
2171 error:
2172         btrfs_free_path(path);
2173         mutex_unlock(&dev_root->fs_info->volume_mutex);
2174         return ret;
2175 }
2176
2177 /*
2178  * shrinking a device means finding all of the device extents past
2179  * the new size, and then following the back refs to the chunks.
2180  * The chunk relocation code actually frees the device extent
2181  */
2182 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2183 {
2184         struct btrfs_trans_handle *trans;
2185         struct btrfs_root *root = device->dev_root;
2186         struct btrfs_dev_extent *dev_extent = NULL;
2187         struct btrfs_path *path;
2188         u64 length;
2189         u64 chunk_tree;
2190         u64 chunk_objectid;
2191         u64 chunk_offset;
2192         int ret;
2193         int slot;
2194         int failed = 0;
2195         bool retried = false;
2196         struct extent_buffer *l;
2197         struct btrfs_key key;
2198         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2199         u64 old_total = btrfs_super_total_bytes(super_copy);
2200         u64 old_size = device->total_bytes;
2201         u64 diff = device->total_bytes - new_size;
2202
2203         if (new_size >= device->total_bytes)
2204                 return -EINVAL;
2205
2206         path = btrfs_alloc_path();
2207         if (!path)
2208                 return -ENOMEM;
2209
2210         path->reada = 2;
2211
2212         lock_chunks(root);
2213
2214         device->total_bytes = new_size;
2215         if (device->writeable) {
2216                 device->fs_devices->total_rw_bytes -= diff;
2217                 spin_lock(&root->fs_info->free_chunk_lock);
2218                 root->fs_info->free_chunk_space -= diff;
2219                 spin_unlock(&root->fs_info->free_chunk_lock);
2220         }
2221         unlock_chunks(root);
2222
2223 again:
2224         key.objectid = device->devid;
2225         key.offset = (u64)-1;
2226         key.type = BTRFS_DEV_EXTENT_KEY;
2227
2228         while (1) {
2229                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2230                 if (ret < 0)
2231                         goto done;
2232
2233                 ret = btrfs_previous_item(root, path, 0, key.type);
2234                 if (ret < 0)
2235                         goto done;
2236                 if (ret) {
2237                         ret = 0;
2238                         btrfs_release_path(path);
2239                         break;
2240                 }
2241
2242                 l = path->nodes[0];
2243                 slot = path->slots[0];
2244                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2245
2246                 if (key.objectid != device->devid) {
2247                         btrfs_release_path(path);
2248                         break;
2249                 }
2250
2251                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2252                 length = btrfs_dev_extent_length(l, dev_extent);
2253
2254                 if (key.offset + length <= new_size) {
2255                         btrfs_release_path(path);
2256                         break;
2257                 }
2258
2259                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2260                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2261                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2262                 btrfs_release_path(path);
2263
2264                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2265                                            chunk_offset);
2266                 if (ret && ret != -ENOSPC)
2267                         goto done;
2268                 if (ret == -ENOSPC)
2269                         failed++;
2270                 key.offset -= 1;
2271         }
2272
2273         if (failed && !retried) {
2274                 failed = 0;
2275                 retried = true;
2276                 goto again;
2277         } else if (failed && retried) {
2278                 ret = -ENOSPC;
2279                 lock_chunks(root);
2280
2281                 device->total_bytes = old_size;
2282                 if (device->writeable)
2283                         device->fs_devices->total_rw_bytes += diff;
2284                 spin_lock(&root->fs_info->free_chunk_lock);
2285                 root->fs_info->free_chunk_space += diff;
2286                 spin_unlock(&root->fs_info->free_chunk_lock);
2287                 unlock_chunks(root);
2288                 goto done;
2289         }
2290
2291         /* Shrinking succeeded, else we would be at "done". */
2292         trans = btrfs_start_transaction(root, 0);
2293         if (IS_ERR(trans)) {
2294                 ret = PTR_ERR(trans);
2295                 goto done;
2296         }
2297
2298         lock_chunks(root);
2299
2300         device->disk_total_bytes = new_size;
2301         /* Now btrfs_update_device() will change the on-disk size. */
2302         ret = btrfs_update_device(trans, device);
2303         if (ret) {
2304                 unlock_chunks(root);
2305                 btrfs_end_transaction(trans, root);
2306                 goto done;
2307         }
2308         WARN_ON(diff > old_total);
2309         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2310         unlock_chunks(root);
2311         btrfs_end_transaction(trans, root);
2312 done:
2313         btrfs_free_path(path);
2314         return ret;
2315 }
2316
2317 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2318                            struct btrfs_root *root,
2319                            struct btrfs_key *key,
2320                            struct btrfs_chunk *chunk, int item_size)
2321 {
2322         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2323         struct btrfs_disk_key disk_key;
2324         u32 array_size;
2325         u8 *ptr;
2326
2327         array_size = btrfs_super_sys_array_size(super_copy);
2328         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2329                 return -EFBIG;
2330
2331         ptr = super_copy->sys_chunk_array + array_size;
2332         btrfs_cpu_key_to_disk(&disk_key, key);
2333         memcpy(ptr, &disk_key, sizeof(disk_key));
2334         ptr += sizeof(disk_key);
2335         memcpy(ptr, chunk, item_size);
2336         item_size += sizeof(disk_key);
2337         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2338         return 0;
2339 }
2340
2341 /*
2342  * sort the devices in descending order by max_avail, total_avail
2343  */
2344 static int btrfs_cmp_device_info(const void *a, const void *b)
2345 {
2346         const struct btrfs_device_info *di_a = a;
2347         const struct btrfs_device_info *di_b = b;
2348
2349         if (di_a->max_avail > di_b->max_avail)
2350                 return -1;
2351         if (di_a->max_avail < di_b->max_avail)
2352                 return 1;
2353         if (di_a->total_avail > di_b->total_avail)
2354                 return -1;
2355         if (di_a->total_avail < di_b->total_avail)
2356                 return 1;
2357         return 0;
2358 }
2359
2360 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2361                                struct btrfs_root *extent_root,
2362                                struct map_lookup **map_ret,
2363                                u64 *num_bytes_out, u64 *stripe_size_out,
2364                                u64 start, u64 type)
2365 {
2366         struct btrfs_fs_info *info = extent_root->fs_info;
2367         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2368         struct list_head *cur;
2369         struct map_lookup *map = NULL;
2370         struct extent_map_tree *em_tree;
2371         struct extent_map *em;
2372         struct btrfs_device_info *devices_info = NULL;
2373         u64 total_avail;
2374         int num_stripes;        /* total number of stripes to allocate */
2375         int sub_stripes;        /* sub_stripes info for map */
2376         int dev_stripes;        /* stripes per dev */
2377         int devs_max;           /* max devs to use */
2378         int devs_min;           /* min devs needed */
2379         int devs_increment;     /* ndevs has to be a multiple of this */
2380         int ncopies;            /* how many copies to data has */
2381         int ret;
2382         u64 max_stripe_size;
2383         u64 max_chunk_size;
2384         u64 stripe_size;
2385         u64 num_bytes;
2386         int ndevs;
2387         int i;
2388         int j;
2389
2390         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2391             (type & BTRFS_BLOCK_GROUP_DUP)) {
2392                 WARN_ON(1);
2393                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2394         }
2395
2396         if (list_empty(&fs_devices->alloc_list))
2397                 return -ENOSPC;
2398
2399         sub_stripes = 1;
2400         dev_stripes = 1;
2401         devs_increment = 1;
2402         ncopies = 1;
2403         devs_max = 0;   /* 0 == as many as possible */
2404         devs_min = 1;
2405
2406         /*
2407          * define the properties of each RAID type.
2408          * FIXME: move this to a global table and use it in all RAID
2409          * calculation code
2410          */
2411         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2412                 dev_stripes = 2;
2413                 ncopies = 2;
2414                 devs_max = 1;
2415         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2416                 devs_min = 2;
2417         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2418                 devs_increment = 2;
2419                 ncopies = 2;
2420                 devs_max = 2;
2421                 devs_min = 2;
2422         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2423                 sub_stripes = 2;
2424                 devs_increment = 2;
2425                 ncopies = 2;
2426                 devs_min = 4;
2427         } else {
2428                 devs_max = 1;
2429         }
2430
2431         if (type & BTRFS_BLOCK_GROUP_DATA) {
2432                 max_stripe_size = 1024 * 1024 * 1024;
2433                 max_chunk_size = 10 * max_stripe_size;
2434         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2435                 max_stripe_size = 256 * 1024 * 1024;
2436                 max_chunk_size = max_stripe_size;
2437         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2438                 max_stripe_size = 8 * 1024 * 1024;
2439                 max_chunk_size = 2 * max_stripe_size;
2440         } else {
2441                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2442                        type);
2443                 BUG_ON(1);
2444         }
2445
2446         /* we don't want a chunk larger than 10% of writeable space */
2447         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2448                              max_chunk_size);
2449
2450         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2451                                GFP_NOFS);
2452         if (!devices_info)
2453                 return -ENOMEM;
2454
2455         cur = fs_devices->alloc_list.next;
2456
2457         /*
2458          * in the first pass through the devices list, we gather information
2459          * about the available holes on each device.
2460          */
2461         ndevs = 0;
2462         while (cur != &fs_devices->alloc_list) {
2463                 struct btrfs_device *device;
2464                 u64 max_avail;
2465                 u64 dev_offset;
2466
2467                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2468
2469                 cur = cur->next;
2470
2471                 if (!device->writeable) {
2472                         printk(KERN_ERR
2473                                "btrfs: read-only device in alloc_list\n");
2474                         WARN_ON(1);
2475                         continue;
2476                 }
2477
2478                 if (!device->in_fs_metadata)
2479                         continue;
2480
2481                 if (device->total_bytes > device->bytes_used)
2482                         total_avail = device->total_bytes - device->bytes_used;
2483                 else
2484                         total_avail = 0;
2485
2486                 /* If there is no space on this device, skip it. */
2487                 if (total_avail == 0)
2488                         continue;
2489
2490                 ret = find_free_dev_extent(trans, device,
2491                                            max_stripe_size * dev_stripes,
2492                                            &dev_offset, &max_avail);
2493                 if (ret && ret != -ENOSPC)
2494                         goto error;
2495
2496                 if (ret == 0)
2497                         max_avail = max_stripe_size * dev_stripes;
2498
2499                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2500                         continue;
2501
2502                 devices_info[ndevs].dev_offset = dev_offset;
2503                 devices_info[ndevs].max_avail = max_avail;
2504                 devices_info[ndevs].total_avail = total_avail;
2505                 devices_info[ndevs].dev = device;
2506                 ++ndevs;
2507         }
2508
2509         /*
2510          * now sort the devices by hole size / available space
2511          */
2512         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2513              btrfs_cmp_device_info, NULL);
2514
2515         /* round down to number of usable stripes */
2516         ndevs -= ndevs % devs_increment;
2517
2518         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2519                 ret = -ENOSPC;
2520                 goto error;
2521         }
2522
2523         if (devs_max && ndevs > devs_max)
2524                 ndevs = devs_max;
2525         /*
2526          * the primary goal is to maximize the number of stripes, so use as many
2527          * devices as possible, even if the stripes are not maximum sized.
2528          */
2529         stripe_size = devices_info[ndevs-1].max_avail;
2530         num_stripes = ndevs * dev_stripes;
2531
2532         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2533                 stripe_size = max_chunk_size * ncopies;
2534                 do_div(stripe_size, num_stripes);
2535         }
2536
2537         do_div(stripe_size, dev_stripes);
2538         do_div(stripe_size, BTRFS_STRIPE_LEN);
2539         stripe_size *= BTRFS_STRIPE_LEN;
2540
2541         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2542         if (!map) {
2543                 ret = -ENOMEM;
2544                 goto error;
2545         }
2546         map->num_stripes = num_stripes;
2547
2548         for (i = 0; i < ndevs; ++i) {
2549                 for (j = 0; j < dev_stripes; ++j) {
2550                         int s = i * dev_stripes + j;
2551                         map->stripes[s].dev = devices_info[i].dev;
2552                         map->stripes[s].physical = devices_info[i].dev_offset +
2553                                                    j * stripe_size;
2554                 }
2555         }
2556         map->sector_size = extent_root->sectorsize;
2557         map->stripe_len = BTRFS_STRIPE_LEN;
2558         map->io_align = BTRFS_STRIPE_LEN;
2559         map->io_width = BTRFS_STRIPE_LEN;
2560         map->type = type;
2561         map->sub_stripes = sub_stripes;
2562
2563         *map_ret = map;
2564         num_bytes = stripe_size * (num_stripes / ncopies);
2565
2566         *stripe_size_out = stripe_size;
2567         *num_bytes_out = num_bytes;
2568
2569         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2570
2571         em = alloc_extent_map();
2572         if (!em) {
2573                 ret = -ENOMEM;
2574                 goto error;
2575         }
2576         em->bdev = (struct block_device *)map;
2577         em->start = start;
2578         em->len = num_bytes;
2579         em->block_start = 0;
2580         em->block_len = em->len;
2581
2582         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2583         write_lock(&em_tree->lock);
2584         ret = add_extent_mapping(em_tree, em);
2585         write_unlock(&em_tree->lock);
2586         BUG_ON(ret);
2587         free_extent_map(em);
2588
2589         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2590                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2591                                      start, num_bytes);
2592         BUG_ON(ret);
2593
2594         for (i = 0; i < map->num_stripes; ++i) {
2595                 struct btrfs_device *device;
2596                 u64 dev_offset;
2597
2598                 device = map->stripes[i].dev;
2599                 dev_offset = map->stripes[i].physical;
2600
2601                 ret = btrfs_alloc_dev_extent(trans, device,
2602                                 info->chunk_root->root_key.objectid,
2603                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2604                                 start, dev_offset, stripe_size);
2605                 BUG_ON(ret);
2606         }
2607
2608         kfree(devices_info);
2609         return 0;
2610
2611 error:
2612         kfree(map);
2613         kfree(devices_info);
2614         return ret;
2615 }
2616
2617 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2618                                 struct btrfs_root *extent_root,
2619                                 struct map_lookup *map, u64 chunk_offset,
2620                                 u64 chunk_size, u64 stripe_size)
2621 {
2622         u64 dev_offset;
2623         struct btrfs_key key;
2624         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2625         struct btrfs_device *device;
2626         struct btrfs_chunk *chunk;
2627         struct btrfs_stripe *stripe;
2628         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2629         int index = 0;
2630         int ret;
2631
2632         chunk = kzalloc(item_size, GFP_NOFS);
2633         if (!chunk)
2634                 return -ENOMEM;
2635
2636         index = 0;
2637         while (index < map->num_stripes) {
2638                 device = map->stripes[index].dev;
2639                 device->bytes_used += stripe_size;
2640                 ret = btrfs_update_device(trans, device);
2641                 BUG_ON(ret);
2642                 index++;
2643         }
2644
2645         spin_lock(&extent_root->fs_info->free_chunk_lock);
2646         extent_root->fs_info->free_chunk_space -= (stripe_size *
2647                                                    map->num_stripes);
2648         spin_unlock(&extent_root->fs_info->free_chunk_lock);
2649
2650         index = 0;
2651         stripe = &chunk->stripe;
2652         while (index < map->num_stripes) {
2653                 device = map->stripes[index].dev;
2654                 dev_offset = map->stripes[index].physical;
2655
2656                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2657                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2658                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2659                 stripe++;
2660                 index++;
2661         }
2662
2663         btrfs_set_stack_chunk_length(chunk, chunk_size);
2664         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2665         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2666         btrfs_set_stack_chunk_type(chunk, map->type);
2667         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2668         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2669         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2670         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2671         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2672
2673         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2674         key.type = BTRFS_CHUNK_ITEM_KEY;
2675         key.offset = chunk_offset;
2676
2677         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2678         BUG_ON(ret);
2679
2680         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2681                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2682                                              item_size);
2683                 BUG_ON(ret);
2684         }
2685
2686         kfree(chunk);
2687         return 0;
2688 }
2689
2690 /*
2691  * Chunk allocation falls into two parts. The first part does works
2692  * that make the new allocated chunk useable, but not do any operation
2693  * that modifies the chunk tree. The second part does the works that
2694  * require modifying the chunk tree. This division is important for the
2695  * bootstrap process of adding storage to a seed btrfs.
2696  */
2697 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2698                       struct btrfs_root *extent_root, u64 type)
2699 {
2700         u64 chunk_offset;
2701         u64 chunk_size;
2702         u64 stripe_size;
2703         struct map_lookup *map;
2704         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2705         int ret;
2706
2707         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2708                               &chunk_offset);
2709         if (ret)
2710                 return ret;
2711
2712         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2713                                   &stripe_size, chunk_offset, type);
2714         if (ret)
2715                 return ret;
2716
2717         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2718                                    chunk_size, stripe_size);
2719         BUG_ON(ret);
2720         return 0;
2721 }
2722
2723 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2724                                          struct btrfs_root *root,
2725                                          struct btrfs_device *device)
2726 {
2727         u64 chunk_offset;
2728         u64 sys_chunk_offset;
2729         u64 chunk_size;
2730         u64 sys_chunk_size;
2731         u64 stripe_size;
2732         u64 sys_stripe_size;
2733         u64 alloc_profile;
2734         struct map_lookup *map;
2735         struct map_lookup *sys_map;
2736         struct btrfs_fs_info *fs_info = root->fs_info;
2737         struct btrfs_root *extent_root = fs_info->extent_root;
2738         int ret;
2739
2740         ret = find_next_chunk(fs_info->chunk_root,
2741                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2742         if (ret)
2743                 return ret;
2744
2745         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2746                         (fs_info->metadata_alloc_profile &
2747                          fs_info->avail_metadata_alloc_bits);
2748         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2749
2750         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2751                                   &stripe_size, chunk_offset, alloc_profile);
2752         BUG_ON(ret);
2753
2754         sys_chunk_offset = chunk_offset + chunk_size;
2755
2756         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2757                         (fs_info->system_alloc_profile &
2758                          fs_info->avail_system_alloc_bits);
2759         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2760
2761         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2762                                   &sys_chunk_size, &sys_stripe_size,
2763                                   sys_chunk_offset, alloc_profile);
2764         BUG_ON(ret);
2765
2766         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2767         BUG_ON(ret);
2768
2769         /*
2770          * Modifying chunk tree needs allocating new blocks from both
2771          * system block group and metadata block group. So we only can
2772          * do operations require modifying the chunk tree after both
2773          * block groups were created.
2774          */
2775         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2776                                    chunk_size, stripe_size);
2777         BUG_ON(ret);
2778
2779         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2780                                    sys_chunk_offset, sys_chunk_size,
2781                                    sys_stripe_size);
2782         BUG_ON(ret);
2783         return 0;
2784 }
2785
2786 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2787 {
2788         struct extent_map *em;
2789         struct map_lookup *map;
2790         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2791         int readonly = 0;
2792         int i;
2793
2794         read_lock(&map_tree->map_tree.lock);
2795         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2796         read_unlock(&map_tree->map_tree.lock);
2797         if (!em)
2798                 return 1;
2799
2800         if (btrfs_test_opt(root, DEGRADED)) {
2801                 free_extent_map(em);
2802                 return 0;
2803         }
2804
2805         map = (struct map_lookup *)em->bdev;
2806         for (i = 0; i < map->num_stripes; i++) {
2807                 if (!map->stripes[i].dev->writeable) {
2808                         readonly = 1;
2809                         break;
2810                 }
2811         }
2812         free_extent_map(em);
2813         return readonly;
2814 }
2815
2816 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2817 {
2818         extent_map_tree_init(&tree->map_tree);
2819 }
2820
2821 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2822 {
2823         struct extent_map *em;
2824
2825         while (1) {
2826                 write_lock(&tree->map_tree.lock);
2827                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2828                 if (em)
2829                         remove_extent_mapping(&tree->map_tree, em);
2830                 write_unlock(&tree->map_tree.lock);
2831                 if (!em)
2832                         break;
2833                 kfree(em->bdev);
2834                 /* once for us */
2835                 free_extent_map(em);
2836                 /* once for the tree */
2837                 free_extent_map(em);
2838         }
2839 }
2840
2841 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2842 {
2843         struct extent_map *em;
2844         struct map_lookup *map;
2845         struct extent_map_tree *em_tree = &map_tree->map_tree;
2846         int ret;
2847
2848         read_lock(&em_tree->lock);
2849         em = lookup_extent_mapping(em_tree, logical, len);
2850         read_unlock(&em_tree->lock);
2851         BUG_ON(!em);
2852
2853         BUG_ON(em->start > logical || em->start + em->len < logical);
2854         map = (struct map_lookup *)em->bdev;
2855         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2856                 ret = map->num_stripes;
2857         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2858                 ret = map->sub_stripes;
2859         else
2860                 ret = 1;
2861         free_extent_map(em);
2862         return ret;
2863 }
2864
2865 static int find_live_mirror(struct map_lookup *map, int first, int num,
2866                             int optimal)
2867 {
2868         int i;
2869         if (map->stripes[optimal].dev->bdev)
2870                 return optimal;
2871         for (i = first; i < first + num; i++) {
2872                 if (map->stripes[i].dev->bdev)
2873                         return i;
2874         }
2875         /* we couldn't find one that doesn't fail.  Just return something
2876          * and the io error handling code will clean up eventually
2877          */
2878         return optimal;
2879 }
2880
2881 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2882                              u64 logical, u64 *length,
2883                              struct btrfs_bio **bbio_ret,
2884                              int mirror_num)
2885 {
2886         struct extent_map *em;
2887         struct map_lookup *map;
2888         struct extent_map_tree *em_tree = &map_tree->map_tree;
2889         u64 offset;
2890         u64 stripe_offset;
2891         u64 stripe_end_offset;
2892         u64 stripe_nr;
2893         u64 stripe_nr_orig;
2894         u64 stripe_nr_end;
2895         int stripes_allocated = 8;
2896         int stripes_required = 1;
2897         int stripe_index;
2898         int i;
2899         int num_stripes;
2900         int max_errors = 0;
2901         struct btrfs_bio *bbio = NULL;
2902
2903         if (bbio_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2904                 stripes_allocated = 1;
2905 again:
2906         if (bbio_ret) {
2907                 bbio = kzalloc(btrfs_bio_size(stripes_allocated),
2908                                 GFP_NOFS);
2909                 if (!bbio)
2910                         return -ENOMEM;
2911
2912                 atomic_set(&bbio->error, 0);
2913         }
2914
2915         read_lock(&em_tree->lock);
2916         em = lookup_extent_mapping(em_tree, logical, *length);
2917         read_unlock(&em_tree->lock);
2918
2919         if (!em) {
2920                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2921                        (unsigned long long)logical,
2922                        (unsigned long long)*length);
2923                 BUG();
2924         }
2925
2926         BUG_ON(em->start > logical || em->start + em->len < logical);
2927         map = (struct map_lookup *)em->bdev;
2928         offset = logical - em->start;
2929
2930         if (mirror_num > map->num_stripes)
2931                 mirror_num = 0;
2932
2933         /* if our btrfs_bio struct is too small, back off and try again */
2934         if (rw & REQ_WRITE) {
2935                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2936                                  BTRFS_BLOCK_GROUP_DUP)) {
2937                         stripes_required = map->num_stripes;
2938                         max_errors = 1;
2939                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2940                         stripes_required = map->sub_stripes;
2941                         max_errors = 1;
2942                 }
2943         }
2944         if (rw & REQ_DISCARD) {
2945                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2946                                  BTRFS_BLOCK_GROUP_RAID1 |
2947                                  BTRFS_BLOCK_GROUP_DUP |
2948                                  BTRFS_BLOCK_GROUP_RAID10)) {
2949                         stripes_required = map->num_stripes;
2950                 }
2951         }
2952         if (bbio_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2953             stripes_allocated < stripes_required) {
2954                 stripes_allocated = map->num_stripes;
2955                 free_extent_map(em);
2956                 kfree(bbio);
2957                 goto again;
2958         }
2959         stripe_nr = offset;
2960         /*
2961          * stripe_nr counts the total number of stripes we have to stride
2962          * to get to this block
2963          */
2964         do_div(stripe_nr, map->stripe_len);
2965
2966         stripe_offset = stripe_nr * map->stripe_len;
2967         BUG_ON(offset < stripe_offset);
2968
2969         /* stripe_offset is the offset of this block in its stripe*/
2970         stripe_offset = offset - stripe_offset;
2971
2972         if (rw & REQ_DISCARD)
2973                 *length = min_t(u64, em->len - offset, *length);
2974         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2975                               BTRFS_BLOCK_GROUP_RAID1 |
2976                               BTRFS_BLOCK_GROUP_RAID10 |
2977                               BTRFS_BLOCK_GROUP_DUP)) {
2978                 /* we limit the length of each bio to what fits in a stripe */
2979                 *length = min_t(u64, em->len - offset,
2980                                 map->stripe_len - stripe_offset);
2981         } else {
2982                 *length = em->len - offset;
2983         }
2984
2985         if (!bbio_ret)
2986                 goto out;
2987
2988         num_stripes = 1;
2989         stripe_index = 0;
2990         stripe_nr_orig = stripe_nr;
2991         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2992                         (~(map->stripe_len - 1));
2993         do_div(stripe_nr_end, map->stripe_len);
2994         stripe_end_offset = stripe_nr_end * map->stripe_len -
2995                             (offset + *length);
2996         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2997                 if (rw & REQ_DISCARD)
2998                         num_stripes = min_t(u64, map->num_stripes,
2999                                             stripe_nr_end - stripe_nr_orig);
3000                 stripe_index = do_div(stripe_nr, map->num_stripes);
3001         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3002                 if (rw & (REQ_WRITE | REQ_DISCARD))
3003                         num_stripes = map->num_stripes;
3004                 else if (mirror_num)
3005                         stripe_index = mirror_num - 1;
3006                 else {
3007                         stripe_index = find_live_mirror(map, 0,
3008                                             map->num_stripes,
3009                                             current->pid % map->num_stripes);
3010                         mirror_num = stripe_index + 1;
3011                 }
3012
3013         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3014                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3015                         num_stripes = map->num_stripes;
3016                 } else if (mirror_num) {
3017                         stripe_index = mirror_num - 1;
3018                 } else {
3019                         mirror_num = 1;
3020                 }
3021
3022         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3023                 int factor = map->num_stripes / map->sub_stripes;
3024
3025                 stripe_index = do_div(stripe_nr, factor);
3026                 stripe_index *= map->sub_stripes;
3027
3028                 if (rw & REQ_WRITE)
3029                         num_stripes = map->sub_stripes;
3030                 else if (rw & REQ_DISCARD)
3031                         num_stripes = min_t(u64, map->sub_stripes *
3032                                             (stripe_nr_end - stripe_nr_orig),
3033                                             map->num_stripes);
3034                 else if (mirror_num)
3035                         stripe_index += mirror_num - 1;
3036                 else {
3037                         stripe_index = find_live_mirror(map, stripe_index,
3038                                               map->sub_stripes, stripe_index +
3039                                               current->pid % map->sub_stripes);
3040                         mirror_num = stripe_index + 1;
3041                 }
3042         } else {
3043                 /*
3044                  * after this do_div call, stripe_nr is the number of stripes
3045                  * on this device we have to walk to find the data, and
3046                  * stripe_index is the number of our device in the stripe array
3047                  */
3048                 stripe_index = do_div(stripe_nr, map->num_stripes);
3049                 mirror_num = stripe_index + 1;
3050         }
3051         BUG_ON(stripe_index >= map->num_stripes);
3052
3053         if (rw & REQ_DISCARD) {
3054                 for (i = 0; i < num_stripes; i++) {
3055                         bbio->stripes[i].physical =
3056                                 map->stripes[stripe_index].physical +
3057                                 stripe_offset + stripe_nr * map->stripe_len;
3058                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3059
3060                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3061                                 u64 stripes;
3062                                 u32 last_stripe = 0;
3063                                 int j;
3064
3065                                 div_u64_rem(stripe_nr_end - 1,
3066                                             map->num_stripes,
3067                                             &last_stripe);
3068
3069                                 for (j = 0; j < map->num_stripes; j++) {
3070                                         u32 test;
3071
3072                                         div_u64_rem(stripe_nr_end - 1 - j,
3073                                                     map->num_stripes, &test);
3074                                         if (test == stripe_index)
3075                                                 break;
3076                                 }
3077                                 stripes = stripe_nr_end - 1 - j;
3078                                 do_div(stripes, map->num_stripes);
3079                                 bbio->stripes[i].length = map->stripe_len *
3080                                         (stripes - stripe_nr + 1);
3081
3082                                 if (i == 0) {
3083                                         bbio->stripes[i].length -=
3084                                                 stripe_offset;
3085                                         stripe_offset = 0;
3086                                 }
3087                                 if (stripe_index == last_stripe)
3088                                         bbio->stripes[i].length -=
3089                                                 stripe_end_offset;
3090                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3091                                 u64 stripes;
3092                                 int j;
3093                                 int factor = map->num_stripes /
3094                                              map->sub_stripes;
3095                                 u32 last_stripe = 0;
3096
3097                                 div_u64_rem(stripe_nr_end - 1,
3098                                             factor, &last_stripe);
3099                                 last_stripe *= map->sub_stripes;
3100
3101                                 for (j = 0; j < factor; j++) {
3102                                         u32 test;
3103
3104                                         div_u64_rem(stripe_nr_end - 1 - j,
3105                                                     factor, &test);
3106
3107                                         if (test ==
3108                                             stripe_index / map->sub_stripes)
3109                                                 break;
3110                                 }
3111                                 stripes = stripe_nr_end - 1 - j;
3112                                 do_div(stripes, factor);
3113                                 bbio->stripes[i].length = map->stripe_len *
3114                                         (stripes - stripe_nr + 1);
3115
3116                                 if (i < map->sub_stripes) {
3117                                         bbio->stripes[i].length -=
3118                                                 stripe_offset;
3119                                         if (i == map->sub_stripes - 1)
3120                                                 stripe_offset = 0;
3121                                 }
3122                                 if (stripe_index >= last_stripe &&
3123                                     stripe_index <= (last_stripe +
3124                                                      map->sub_stripes - 1)) {
3125                                         bbio->stripes[i].length -=
3126                                                 stripe_end_offset;
3127                                 }
3128                         } else
3129                                 bbio->stripes[i].length = *length;
3130
3131                         stripe_index++;
3132                         if (stripe_index == map->num_stripes) {
3133                                 /* This could only happen for RAID0/10 */
3134                                 stripe_index = 0;
3135                                 stripe_nr++;
3136                         }
3137                 }
3138         } else {
3139                 for (i = 0; i < num_stripes; i++) {
3140                         bbio->stripes[i].physical =
3141                                 map->stripes[stripe_index].physical +
3142                                 stripe_offset +
3143                                 stripe_nr * map->stripe_len;
3144                         bbio->stripes[i].dev =
3145                                 map->stripes[stripe_index].dev;
3146                         stripe_index++;
3147                 }
3148         }
3149         if (bbio_ret) {
3150                 *bbio_ret = bbio;
3151                 bbio->num_stripes = num_stripes;
3152                 bbio->max_errors = max_errors;
3153                 bbio->mirror_num = mirror_num;
3154         }
3155 out:
3156         free_extent_map(em);
3157         return 0;
3158 }
3159
3160 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3161                       u64 logical, u64 *length,
3162                       struct btrfs_bio **bbio_ret, int mirror_num)
3163 {
3164         return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3165                                  mirror_num);
3166 }
3167
3168 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3169                      u64 chunk_start, u64 physical, u64 devid,
3170                      u64 **logical, int *naddrs, int *stripe_len)
3171 {
3172         struct extent_map_tree *em_tree = &map_tree->map_tree;
3173         struct extent_map *em;
3174         struct map_lookup *map;
3175         u64 *buf;
3176         u64 bytenr;
3177         u64 length;
3178         u64 stripe_nr;
3179         int i, j, nr = 0;
3180
3181         read_lock(&em_tree->lock);
3182         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3183         read_unlock(&em_tree->lock);
3184
3185         BUG_ON(!em || em->start != chunk_start);
3186         map = (struct map_lookup *)em->bdev;
3187
3188         length = em->len;
3189         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3190                 do_div(length, map->num_stripes / map->sub_stripes);
3191         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3192                 do_div(length, map->num_stripes);
3193
3194         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3195         BUG_ON(!buf);
3196
3197         for (i = 0; i < map->num_stripes; i++) {
3198                 if (devid && map->stripes[i].dev->devid != devid)
3199                         continue;
3200                 if (map->stripes[i].physical > physical ||
3201                     map->stripes[i].physical + length <= physical)
3202                         continue;
3203
3204                 stripe_nr = physical - map->stripes[i].physical;
3205                 do_div(stripe_nr, map->stripe_len);
3206
3207                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3208                         stripe_nr = stripe_nr * map->num_stripes + i;
3209                         do_div(stripe_nr, map->sub_stripes);
3210                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3211                         stripe_nr = stripe_nr * map->num_stripes + i;
3212                 }
3213                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3214                 WARN_ON(nr >= map->num_stripes);
3215                 for (j = 0; j < nr; j++) {
3216                         if (buf[j] == bytenr)
3217                                 break;
3218                 }
3219                 if (j == nr) {
3220                         WARN_ON(nr >= map->num_stripes);
3221                         buf[nr++] = bytenr;
3222                 }
3223         }
3224
3225         *logical = buf;
3226         *naddrs = nr;
3227         *stripe_len = map->stripe_len;
3228
3229         free_extent_map(em);
3230         return 0;
3231 }
3232
3233 static void btrfs_end_bio(struct bio *bio, int err)
3234 {
3235         struct btrfs_bio *bbio = bio->bi_private;
3236         int is_orig_bio = 0;
3237
3238         if (err)
3239                 atomic_inc(&bbio->error);
3240
3241         if (bio == bbio->orig_bio)
3242                 is_orig_bio = 1;
3243
3244         if (atomic_dec_and_test(&bbio->stripes_pending)) {
3245                 if (!is_orig_bio) {
3246                         bio_put(bio);
3247                         bio = bbio->orig_bio;
3248                 }
3249                 bio->bi_private = bbio->private;
3250                 bio->bi_end_io = bbio->end_io;
3251                 bio->bi_bdev = (struct block_device *)
3252                                         (unsigned long)bbio->mirror_num;
3253                 /* only send an error to the higher layers if it is
3254                  * beyond the tolerance of the multi-bio
3255                  */
3256                 if (atomic_read(&bbio->error) > bbio->max_errors) {
3257                         err = -EIO;
3258                 } else if (err) {
3259                         /*
3260                          * this bio is actually up to date, we didn't
3261                          * go over the max number of errors
3262                          */
3263                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3264                         err = 0;
3265                 }
3266                 kfree(bbio);
3267
3268                 bio_endio(bio, err);
3269         } else if (!is_orig_bio) {
3270                 bio_put(bio);
3271         }
3272 }
3273
3274 struct async_sched {
3275         struct bio *bio;
3276         int rw;
3277         struct btrfs_fs_info *info;
3278         struct btrfs_work work;
3279 };
3280
3281 /*
3282  * see run_scheduled_bios for a description of why bios are collected for
3283  * async submit.
3284  *
3285  * This will add one bio to the pending list for a device and make sure
3286  * the work struct is scheduled.
3287  */
3288 static noinline int schedule_bio(struct btrfs_root *root,
3289                                  struct btrfs_device *device,
3290                                  int rw, struct bio *bio)
3291 {
3292         int should_queue = 1;
3293         struct btrfs_pending_bios *pending_bios;
3294
3295         /* don't bother with additional async steps for reads, right now */
3296         if (!(rw & REQ_WRITE)) {
3297                 bio_get(bio);
3298                 submit_bio(rw, bio);
3299                 bio_put(bio);
3300                 return 0;
3301         }
3302
3303         /*
3304          * nr_async_bios allows us to reliably return congestion to the
3305          * higher layers.  Otherwise, the async bio makes it appear we have
3306          * made progress against dirty pages when we've really just put it
3307          * on a queue for later
3308          */
3309         atomic_inc(&root->fs_info->nr_async_bios);
3310         WARN_ON(bio->bi_next);
3311         bio->bi_next = NULL;
3312         bio->bi_rw |= rw;
3313
3314         spin_lock(&device->io_lock);
3315         if (bio->bi_rw & REQ_SYNC)
3316                 pending_bios = &device->pending_sync_bios;
3317         else
3318                 pending_bios = &device->pending_bios;
3319
3320         if (pending_bios->tail)
3321                 pending_bios->tail->bi_next = bio;
3322
3323         pending_bios->tail = bio;
3324         if (!pending_bios->head)
3325                 pending_bios->head = bio;
3326         if (device->running_pending)
3327                 should_queue = 0;
3328
3329         spin_unlock(&device->io_lock);
3330
3331         if (should_queue)
3332                 btrfs_queue_worker(&root->fs_info->submit_workers,
3333                                    &device->work);
3334         return 0;
3335 }
3336
3337 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3338                   int mirror_num, int async_submit)
3339 {
3340         struct btrfs_mapping_tree *map_tree;
3341         struct btrfs_device *dev;
3342         struct bio *first_bio = bio;
3343         u64 logical = (u64)bio->bi_sector << 9;
3344         u64 length = 0;
3345         u64 map_length;
3346         int ret;
3347         int dev_nr = 0;
3348         int total_devs = 1;
3349         struct btrfs_bio *bbio = NULL;
3350
3351         length = bio->bi_size;
3352         map_tree = &root->fs_info->mapping_tree;
3353         map_length = length;
3354
3355         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
3356                               mirror_num);
3357         BUG_ON(ret);
3358
3359         total_devs = bbio->num_stripes;
3360         if (map_length < length) {
3361                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3362                        "len %llu\n", (unsigned long long)logical,
3363                        (unsigned long long)length,
3364                        (unsigned long long)map_length);
3365                 BUG();
3366         }
3367
3368         bbio->orig_bio = first_bio;
3369         bbio->private = first_bio->bi_private;
3370         bbio->end_io = first_bio->bi_end_io;
3371         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
3372
3373         while (dev_nr < total_devs) {
3374                 if (dev_nr < total_devs - 1) {
3375                         bio = bio_clone(first_bio, GFP_NOFS);
3376                         BUG_ON(!bio);
3377                 } else {
3378                         bio = first_bio;
3379                 }
3380                 bio->bi_private = bbio;
3381                 bio->bi_end_io = btrfs_end_bio;
3382                 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
3383                 dev = bbio->stripes[dev_nr].dev;
3384                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3385                         pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
3386                                  "(%s id %llu), size=%u\n", rw,
3387                                  (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
3388                                  dev->name, dev->devid, bio->bi_size);
3389                         bio->bi_bdev = dev->bdev;
3390                         if (async_submit)
3391                                 schedule_bio(root, dev, rw, bio);
3392                         else
3393                                 submit_bio(rw, bio);
3394                 } else {
3395                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3396                         bio->bi_sector = logical >> 9;
3397                         bio_endio(bio, -EIO);
3398                 }
3399                 dev_nr++;
3400         }
3401         return 0;
3402 }
3403
3404 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3405                                        u8 *uuid, u8 *fsid)
3406 {
3407         struct btrfs_device *device;
3408         struct btrfs_fs_devices *cur_devices;
3409
3410         cur_devices = root->fs_info->fs_devices;
3411         while (cur_devices) {
3412                 if (!fsid ||
3413                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3414                         device = __find_device(&cur_devices->devices,
3415                                                devid, uuid);
3416                         if (device)
3417                                 return device;
3418                 }
3419                 cur_devices = cur_devices->seed;
3420         }
3421         return NULL;
3422 }
3423
3424 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3425                                             u64 devid, u8 *dev_uuid)
3426 {
3427         struct btrfs_device *device;
3428         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3429
3430         device = kzalloc(sizeof(*device), GFP_NOFS);
3431         if (!device)
3432                 return NULL;
3433         list_add(&device->dev_list,
3434                  &fs_devices->devices);
3435         device->dev_root = root->fs_info->dev_root;
3436         device->devid = devid;
3437         device->work.func = pending_bios_fn;
3438         device->fs_devices = fs_devices;
3439         device->missing = 1;
3440         fs_devices->num_devices++;
3441         fs_devices->missing_devices++;
3442         spin_lock_init(&device->io_lock);
3443         INIT_LIST_HEAD(&device->dev_alloc_list);
3444         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3445         return device;
3446 }
3447
3448 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3449                           struct extent_buffer *leaf,
3450                           struct btrfs_chunk *chunk)
3451 {
3452         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3453         struct map_lookup *map;
3454         struct extent_map *em;
3455         u64 logical;
3456         u64 length;
3457         u64 devid;
3458         u8 uuid[BTRFS_UUID_SIZE];
3459         int num_stripes;
3460         int ret;
3461         int i;
3462
3463         logical = key->offset;
3464         length = btrfs_chunk_length(leaf, chunk);
3465
3466         read_lock(&map_tree->map_tree.lock);
3467         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3468         read_unlock(&map_tree->map_tree.lock);
3469
3470         /* already mapped? */
3471         if (em && em->start <= logical && em->start + em->len > logical) {
3472                 free_extent_map(em);
3473                 return 0;
3474         } else if (em) {
3475                 free_extent_map(em);
3476         }
3477
3478         em = alloc_extent_map();
3479         if (!em)
3480                 return -ENOMEM;
3481         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3482         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3483         if (!map) {
3484                 free_extent_map(em);
3485                 return -ENOMEM;
3486         }
3487
3488         em->bdev = (struct block_device *)map;
3489         em->start = logical;
3490         em->len = length;
3491         em->block_start = 0;
3492         em->block_len = em->len;
3493
3494         map->num_stripes = num_stripes;
3495         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3496         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3497         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3498         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3499         map->type = btrfs_chunk_type(leaf, chunk);
3500         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3501         for (i = 0; i < num_stripes; i++) {
3502                 map->stripes[i].physical =
3503                         btrfs_stripe_offset_nr(leaf, chunk, i);
3504                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3505                 read_extent_buffer(leaf, uuid, (unsigned long)
3506                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3507                                    BTRFS_UUID_SIZE);
3508                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3509                                                         NULL);
3510                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3511                         kfree(map);
3512                         free_extent_map(em);
3513                         return -EIO;
3514                 }
3515                 if (!map->stripes[i].dev) {
3516                         map->stripes[i].dev =
3517                                 add_missing_dev(root, devid, uuid);
3518                         if (!map->stripes[i].dev) {
3519                                 kfree(map);
3520                                 free_extent_map(em);
3521                                 return -EIO;
3522                         }
3523                 }
3524                 map->stripes[i].dev->in_fs_metadata = 1;
3525         }
3526
3527         write_lock(&map_tree->map_tree.lock);
3528         ret = add_extent_mapping(&map_tree->map_tree, em);
3529         write_unlock(&map_tree->map_tree.lock);
3530         BUG_ON(ret);
3531         free_extent_map(em);
3532
3533         return 0;
3534 }
3535
3536 static int fill_device_from_item(struct extent_buffer *leaf,
3537                                  struct btrfs_dev_item *dev_item,
3538                                  struct btrfs_device *device)
3539 {
3540         unsigned long ptr;
3541
3542         device->devid = btrfs_device_id(leaf, dev_item);
3543         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3544         device->total_bytes = device->disk_total_bytes;
3545         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3546         device->type = btrfs_device_type(leaf, dev_item);
3547         device->io_align = btrfs_device_io_align(leaf, dev_item);
3548         device->io_width = btrfs_device_io_width(leaf, dev_item);
3549         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3550
3551         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3552         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3553
3554         return 0;
3555 }
3556
3557 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3558 {
3559         struct btrfs_fs_devices *fs_devices;
3560         int ret;
3561
3562         mutex_lock(&uuid_mutex);
3563
3564         fs_devices = root->fs_info->fs_devices->seed;
3565         while (fs_devices) {
3566                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3567                         ret = 0;
3568                         goto out;
3569                 }
3570                 fs_devices = fs_devices->seed;
3571         }
3572
3573         fs_devices = find_fsid(fsid);
3574         if (!fs_devices) {
3575                 ret = -ENOENT;
3576                 goto out;
3577         }
3578
3579         fs_devices = clone_fs_devices(fs_devices);
3580         if (IS_ERR(fs_devices)) {
3581                 ret = PTR_ERR(fs_devices);
3582                 goto out;
3583         }
3584
3585         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3586                                    root->fs_info->bdev_holder);
3587         if (ret)
3588                 goto out;
3589
3590         if (!fs_devices->seeding) {
3591                 __btrfs_close_devices(fs_devices);
3592                 free_fs_devices(fs_devices);
3593                 ret = -EINVAL;
3594                 goto out;
3595         }
3596
3597         fs_devices->seed = root->fs_info->fs_devices->seed;
3598         root->fs_info->fs_devices->seed = fs_devices;
3599 out:
3600         mutex_unlock(&uuid_mutex);
3601         return ret;
3602 }
3603
3604 static int read_one_dev(struct btrfs_root *root,
3605                         struct extent_buffer *leaf,
3606                         struct btrfs_dev_item *dev_item)
3607 {
3608         struct btrfs_device *device;
3609         u64 devid;
3610         int ret;
3611         u8 fs_uuid[BTRFS_UUID_SIZE];
3612         u8 dev_uuid[BTRFS_UUID_SIZE];
3613
3614         devid = btrfs_device_id(leaf, dev_item);
3615         read_extent_buffer(leaf, dev_uuid,
3616                            (unsigned long)btrfs_device_uuid(dev_item),
3617                            BTRFS_UUID_SIZE);
3618         read_extent_buffer(leaf, fs_uuid,
3619                            (unsigned long)btrfs_device_fsid(dev_item),
3620                            BTRFS_UUID_SIZE);
3621
3622         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3623                 ret = open_seed_devices(root, fs_uuid);
3624                 if (ret && !btrfs_test_opt(root, DEGRADED))
3625                         return ret;
3626         }
3627
3628         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3629         if (!device || !device->bdev) {
3630                 if (!btrfs_test_opt(root, DEGRADED))
3631                         return -EIO;
3632
3633                 if (!device) {
3634                         printk(KERN_WARNING "warning devid %llu missing\n",
3635                                (unsigned long long)devid);
3636                         device = add_missing_dev(root, devid, dev_uuid);
3637                         if (!device)
3638                                 return -ENOMEM;
3639                 } else if (!device->missing) {
3640                         /*
3641                          * this happens when a device that was properly setup
3642                          * in the device info lists suddenly goes bad.
3643                          * device->bdev is NULL, and so we have to set
3644                          * device->missing to one here
3645                          */
3646                         root->fs_info->fs_devices->missing_devices++;
3647                         device->missing = 1;
3648                 }
3649         }
3650
3651         if (device->fs_devices != root->fs_info->fs_devices) {
3652                 BUG_ON(device->writeable);
3653                 if (device->generation !=
3654                     btrfs_device_generation(leaf, dev_item))
3655                         return -EINVAL;
3656         }
3657
3658         fill_device_from_item(leaf, dev_item, device);
3659         device->dev_root = root->fs_info->dev_root;
3660         device->in_fs_metadata = 1;
3661         if (device->writeable) {
3662                 device->fs_devices->total_rw_bytes += device->total_bytes;
3663                 spin_lock(&root->fs_info->free_chunk_lock);
3664                 root->fs_info->free_chunk_space += device->total_bytes -
3665                         device->bytes_used;
3666                 spin_unlock(&root->fs_info->free_chunk_lock);
3667         }
3668         ret = 0;
3669         return ret;
3670 }
3671
3672 int btrfs_read_sys_array(struct btrfs_root *root)
3673 {
3674         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3675         struct extent_buffer *sb;
3676         struct btrfs_disk_key *disk_key;
3677         struct btrfs_chunk *chunk;
3678         u8 *ptr;
3679         unsigned long sb_ptr;
3680         int ret = 0;
3681         u32 num_stripes;
3682         u32 array_size;
3683         u32 len = 0;
3684         u32 cur;
3685         struct btrfs_key key;
3686
3687         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3688                                           BTRFS_SUPER_INFO_SIZE);
3689         if (!sb)
3690                 return -ENOMEM;
3691         btrfs_set_buffer_uptodate(sb);
3692         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
3693
3694         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3695         array_size = btrfs_super_sys_array_size(super_copy);
3696
3697         ptr = super_copy->sys_chunk_array;
3698         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3699         cur = 0;
3700
3701         while (cur < array_size) {
3702                 disk_key = (struct btrfs_disk_key *)ptr;
3703                 btrfs_disk_key_to_cpu(&key, disk_key);
3704
3705                 len = sizeof(*disk_key); ptr += len;
3706                 sb_ptr += len;
3707                 cur += len;
3708
3709                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3710                         chunk = (struct btrfs_chunk *)sb_ptr;
3711                         ret = read_one_chunk(root, &key, sb, chunk);
3712                         if (ret)
3713                                 break;
3714                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3715                         len = btrfs_chunk_item_size(num_stripes);
3716                 } else {
3717                         ret = -EIO;
3718                         break;
3719                 }
3720                 ptr += len;
3721                 sb_ptr += len;
3722                 cur += len;
3723         }
3724         free_extent_buffer(sb);
3725         return ret;
3726 }
3727
3728 int btrfs_read_chunk_tree(struct btrfs_root *root)
3729 {
3730         struct btrfs_path *path;
3731         struct extent_buffer *leaf;
3732         struct btrfs_key key;
3733         struct btrfs_key found_key;
3734         int ret;
3735         int slot;
3736
3737         root = root->fs_info->chunk_root;
3738
3739         path = btrfs_alloc_path();
3740         if (!path)
3741                 return -ENOMEM;
3742
3743         /* first we search for all of the device items, and then we
3744          * read in all of the chunk items.  This way we can create chunk
3745          * mappings that reference all of the devices that are afound
3746          */
3747         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3748         key.offset = 0;
3749         key.type = 0;
3750 again:
3751         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3752         if (ret < 0)
3753                 goto error;
3754         while (1) {
3755                 leaf = path->nodes[0];
3756                 slot = path->slots[0];
3757                 if (slot >= btrfs_header_nritems(leaf)) {
3758                         ret = btrfs_next_leaf(root, path);
3759                         if (ret == 0)
3760                                 continue;
3761                         if (ret < 0)
3762                                 goto error;
3763                         break;
3764                 }
3765                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3766                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3767                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3768                                 break;
3769                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3770                                 struct btrfs_dev_item *dev_item;
3771                                 dev_item = btrfs_item_ptr(leaf, slot,
3772                                                   struct btrfs_dev_item);
3773                                 ret = read_one_dev(root, leaf, dev_item);
3774                                 if (ret)
3775                                         goto error;
3776                         }
3777                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3778                         struct btrfs_chunk *chunk;
3779                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3780                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3781                         if (ret)
3782                                 goto error;
3783                 }
3784                 path->slots[0]++;
3785         }
3786         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3787                 key.objectid = 0;
3788                 btrfs_release_path(path);
3789                 goto again;
3790         }
3791         ret = 0;
3792 error:
3793         btrfs_free_path(path);
3794         return ret;
3795 }