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                 mutex_lock(&fs_devices->device_list_mutex);
 370                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 371                 mutex_unlock(&fs_devices->device_list_mutex);
 372
 373                 device->fs_devices = fs_devices;
 374                 fs_devices->num_devices++;
 375         } else if (!device->name || strcmp(device->name, path)) {
 376                 name = kstrdup(path, GFP_NOFS);
 377                 if (!name)
 378                         return -ENOMEM;
 379                 kfree(device->name);
 380                 device->name = name;
 381                 if (device->missing) {
 382                         fs_devices->missing_devices--;
 383                         device->missing = 0;
 384                 }
 385         }
 386
 387         if (found_transid > fs_devices->latest_trans) {
 388                 fs_devices->latest_devid = devid;
 389                 fs_devices->latest_trans = found_transid;
 390         }
 391         *fs_devices_ret = fs_devices;
 392         return 0;
 393 }
 394
 395 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 396 {
 397         struct btrfs_fs_devices *fs_devices;
 398         struct btrfs_device *device;
 399         struct btrfs_device *orig_dev;
 400
 401         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 402         if (!fs_devices)
 403                 return ERR_PTR(-ENOMEM);
 404
 405         INIT_LIST_HEAD(&fs_devices->devices);
 406         INIT_LIST_HEAD(&fs_devices->alloc_list);
 407         INIT_LIST_HEAD(&fs_devices->list);
 408         mutex_init(&fs_devices->device_list_mutex);
 409         fs_devices->latest_devid = orig->latest_devid;
 410         fs_devices->latest_trans = orig->latest_trans;
 411         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 412
 413         /* We have held the volume lock, it is safe to get the devices. */
 414         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 415                 device = kzalloc(sizeof(*device), GFP_NOFS);
 416                 if (!device)
 417                         goto error;
 418
 419                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 420                 if (!device->name) {
 421                         kfree(device);
 422                         goto error;
 423                 }
 424
 425                 device->devid = orig_dev->devid;
 426                 device->work.func = pending_bios_fn;
 427                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 428                 spin_lock_init(&device->io_lock);
 429                 INIT_LIST_HEAD(&device->dev_list);
 430                 INIT_LIST_HEAD(&device->dev_alloc_list);
 431
 432                 list_add(&device->dev_list, &fs_devices->devices);
 433                 device->fs_devices = fs_devices;
 434                 fs_devices->num_devices++;
 435         }
 436         return fs_devices;
 437 error:
 438         free_fs_devices(fs_devices);
 439         return ERR_PTR(-ENOMEM);
 440 }
 441
 442 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 443 {
 444         struct btrfs_device *device, *next;
 445
 446         mutex_lock(&uuid_mutex);
 447 again:
 448         /* This is the initialized path, it is safe to release the devices. */
 449         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 450                 if (device->in_fs_metadata)
 451                         continue;
 452
 453                 if (device->bdev) {
 454                         blkdev_put(device->bdev, device->mode);
 455                         device->bdev = NULL;
 456                         fs_devices->open_devices--;
 457                 }
 458                 if (device->writeable) {
 459                         list_del_init(&device->dev_alloc_list);
 460                         device->writeable = 0;
 461                         fs_devices->rw_devices--;
 462                 }
 463                 list_del_init(&device->dev_list);
 464                 fs_devices->num_devices--;
 465                 kfree(device->name);
 466                 kfree(device);
 467         }
 468
 469         if (fs_devices->seed) {
 470                 fs_devices = fs_devices->seed;
 471                 goto again;
 472         }
 473
 474         mutex_unlock(&uuid_mutex);
 475         return 0;
 476 }
 477
 478 static void __free_device(struct work_struct *work)
 479 {
 480         struct btrfs_device *device;
 481
 482         device = container_of(work, struct btrfs_device, rcu_work);
 483
 484         if (device->bdev)
 485                 blkdev_put(device->bdev, device->mode);
 486
 487         kfree(device->name);
 488         kfree(device);
 489 }
 490
 491 static void free_device(struct rcu_head *head)
 492 {
 493         struct btrfs_device *device;
 494
 495         device = container_of(head, struct btrfs_device, rcu);
 496
 497         INIT_WORK(&device->rcu_work, __free_device);
 498         schedule_work(&device->rcu_work);
 499 }
 500
 501 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 502 {
 503         struct btrfs_device *device;
 504
 505         if (--fs_devices->opened > 0)
 506                 return 0;
 507
 508         mutex_lock(&fs_devices->device_list_mutex);
 509         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 510                 struct btrfs_device *new_device;
 511
 512                 if (device->bdev)
 513                         fs_devices->open_devices--;
 514
 515                 if (device->writeable) {
 516                         list_del_init(&device->dev_alloc_list);
 517                         fs_devices->rw_devices--;
 518                 }
 519
 520                 if (device->can_discard)
 521                         fs_devices->num_can_discard--;
 522
 523                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 524                 BUG_ON(!new_device);
 525                 memcpy(new_device, device, sizeof(*new_device));
 526                 new_device->name = kstrdup(device->name, GFP_NOFS);
 527                 BUG_ON(device->name && !new_device->name);
 528                 new_device->bdev = NULL;
 529                 new_device->writeable = 0;
 530                 new_device->in_fs_metadata = 0;
 531                 new_device->can_discard = 0;
 532                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 533
 534                 call_rcu(&device->rcu, free_device);
 535         }
 536         mutex_unlock(&fs_devices->device_list_mutex);
 537
 538         WARN_ON(fs_devices->open_devices);
 539         WARN_ON(fs_devices->rw_devices);
 540         fs_devices->opened = 0;
 541         fs_devices->seeding = 0;
 542
 543         return 0;
 544 }
 545
 546 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 547 {
 548         struct btrfs_fs_devices *seed_devices = NULL;
 549         int ret;
 550
 551         mutex_lock(&uuid_mutex);
 552         ret = __btrfs_close_devices(fs_devices);
 553         if (!fs_devices->opened) {
 554                 seed_devices = fs_devices->seed;
 555                 fs_devices->seed = NULL;
 556         }
 557         mutex_unlock(&uuid_mutex);
 558
 559         while (seed_devices) {
 560                 fs_devices = seed_devices;
 561                 seed_devices = fs_devices->seed;
 562                 __btrfs_close_devices(fs_devices);
 563                 free_fs_devices(fs_devices);
 564         }
 565         return ret;
 566 }
 567
 568 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 569                                 fmode_t flags, void *holder)
 570 {
 571         struct request_queue *q;
 572         struct block_device *bdev;
 573         struct list_head *head = &fs_devices->devices;
 574         struct btrfs_device *device;
 575         struct block_device *latest_bdev = NULL;
 576         struct buffer_head *bh;
 577         struct btrfs_super_block *disk_super;
 578         u64 latest_devid = 0;
 579         u64 latest_transid = 0;
 580         u64 devid;
 581         int seeding = 1;
 582         int ret = 0;
 583
 584         flags |= FMODE_EXCL;
 585
 586         list_for_each_entry(device, head, dev_list) {
 587                 if (device->bdev)
 588                         continue;
 589                 if (!device->name)
 590                         continue;
 591
 592                 bdev = blkdev_get_by_path(device->name, flags, holder);
 593                 if (IS_ERR(bdev)) {
 594                         printk(KERN_INFO "open %s failed\n", device->name);
 595                         goto error;
 596                 }
 597                 set_blocksize(bdev, 4096);
 598
 599                 bh = btrfs_read_dev_super(bdev);
 600                 if (!bh)
 601                         goto error_close;
 602
 603                 disk_super = (struct btrfs_super_block *)bh->b_data;
 604                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 605                 if (devid != device->devid)
 606                         goto error_brelse;
 607
 608                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 609                            BTRFS_UUID_SIZE))
 610                         goto error_brelse;
 611
 612                 device->generation = btrfs_super_generation(disk_super);
 613                 if (!latest_transid || device->generation > latest_transid) {
 614                         latest_devid = devid;
 615                         latest_transid = device->generation;
 616                         latest_bdev = bdev;
 617                 }
 618
 619                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 620                         device->writeable = 0;
 621                 } else {
 622                         device->writeable = !bdev_read_only(bdev);
 623                         seeding = 0;
 624                 }
 625
 626                 q = bdev_get_queue(bdev);
 627                 if (blk_queue_discard(q)) {
 628                         device->can_discard = 1;
 629                         fs_devices->num_can_discard++;
 630                 }
 631
 632                 device->bdev = bdev;
 633                 device->in_fs_metadata = 0;
 634                 device->mode = flags;
 635
 636                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 637                         fs_devices->rotating = 1;
 638
 639                 fs_devices->open_devices++;
 640                 if (device->writeable) {
 641                         fs_devices->rw_devices++;
 642                         list_add(&device->dev_alloc_list,
 643                                  &fs_devices->alloc_list);
 644                 }
 645                 brelse(bh);
 646                 continue;
 647
 648 error_brelse:
 649                 brelse(bh);
 650 error_close:
 651                 blkdev_put(bdev, flags);
 652 error:
 653                 continue;
 654         }
 655         if (fs_devices->open_devices == 0) {
 656                 ret = -EINVAL;
 657                 goto out;
 658         }
 659         fs_devices->seeding = seeding;
 660         fs_devices->opened = 1;
 661         fs_devices->latest_bdev = latest_bdev;
 662         fs_devices->latest_devid = latest_devid;
 663         fs_devices->latest_trans = latest_transid;
 664         fs_devices->total_rw_bytes = 0;
 665 out:
 666         return ret;
 667 }
 668
 669 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 670                        fmode_t flags, void *holder)
 671 {
 672         int ret;
 673
 674         mutex_lock(&uuid_mutex);
 675         if (fs_devices->opened) {
 676                 fs_devices->opened++;
 677                 ret = 0;
 678         } else {
 679                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 680         }
 681         mutex_unlock(&uuid_mutex);
 682         return ret;
 683 }
 684
 685 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 686                           struct btrfs_fs_devices **fs_devices_ret)
 687 {
 688         struct btrfs_super_block *disk_super;
 689         struct block_device *bdev;
 690         struct buffer_head *bh;
 691         int ret;
 692         u64 devid;
 693         u64 transid;
 694
 695         mutex_lock(&uuid_mutex);
 696
 697         flags |= FMODE_EXCL;
 698         bdev = blkdev_get_by_path(path, flags, holder);
 699
 700         if (IS_ERR(bdev)) {
 701                 ret = PTR_ERR(bdev);
 702                 goto error;
 703         }
 704
 705         ret = set_blocksize(bdev, 4096);
 706         if (ret)
 707                 goto error_close;
 708         bh = btrfs_read_dev_super(bdev);
 709         if (!bh) {
 710                 ret = -EINVAL;
 711                 goto error_close;
 712         }
 713         disk_super = (struct btrfs_super_block *)bh->b_data;
 714         devid = btrfs_stack_device_id(&disk_super->dev_item);
 715         transid = btrfs_super_generation(disk_super);
 716         if (disk_super->label[0])
 717                 printk(KERN_INFO "device label %s ", disk_super->label);
 718         else
 719                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 720         printk(KERN_CONT "devid %llu transid %llu %s\n",
 721                (unsigned long long)devid, (unsigned long long)transid, path);
 722         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 723
 724         brelse(bh);
 725 error_close:
 726         blkdev_put(bdev, flags);
 727 error:
 728         mutex_unlock(&uuid_mutex);
 729         return ret;
 730 }
 731
 732 /* helper to account the used device space in the range */
 733 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 734                                    u64 end, u64 *length)
 735 {
 736         struct btrfs_key key;
 737         struct btrfs_root *root = device->dev_root;
 738         struct btrfs_dev_extent *dev_extent;
 739         struct btrfs_path *path;
 740         u64 extent_end;
 741         int ret;
 742         int slot;
 743         struct extent_buffer *l;
 744
 745         *length = 0;
 746
 747         if (start >= device->total_bytes)
 748                 return 0;
 749
 750         path = btrfs_alloc_path();
 751         if (!path)
 752                 return -ENOMEM;
 753         path->reada = 2;
 754
 755         key.objectid = device->devid;
 756         key.offset = start;
 757         key.type = BTRFS_DEV_EXTENT_KEY;
 758
 759         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 760         if (ret < 0)
 761                 goto out;
 762         if (ret > 0) {
 763                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 764                 if (ret < 0)
 765                         goto out;
 766         }
 767
 768         while (1) {
 769                 l = path->nodes[0];
 770                 slot = path->slots[0];
 771                 if (slot >= btrfs_header_nritems(l)) {
 772                         ret = btrfs_next_leaf(root, path);
 773                         if (ret == 0)
 774                                 continue;
 775                         if (ret < 0)
 776                                 goto out;
 777
 778                         break;
 779                 }
 780                 btrfs_item_key_to_cpu(l, &key, slot);
 781
 782                 if (key.objectid < device->devid)
 783                         goto next;
 784
 785                 if (key.objectid > device->devid)
 786                         break;
 787
 788                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 789                         goto next;
 790
 791                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 792                 extent_end = key.offset + btrfs_dev_extent_length(l,
 793                                                                   dev_extent);
 794                 if (key.offset <= start && extent_end > end) {
 795                         *length = end - start + 1;
 796                         break;
 797                 } else if (key.offset <= start && extent_end > start)
 798                         *length += extent_end - start;
 799                 else if (key.offset > start && extent_end <= end)
 800                         *length += extent_end - key.offset;
 801                 else if (key.offset > start && key.offset <= end) {
 802                         *length += end - key.offset + 1;
 803                         break;
 804                 } else if (key.offset > end)
 805                         break;
 806
 807 next:
 808                 path->slots[0]++;
 809         }
 810         ret = 0;
 811 out:
 812         btrfs_free_path(path);
 813         return ret;
 814 }
 815
 816 /*
 817  * find_free_dev_extent - find free space in the specified device
 818  * @trans:      transaction handler
 819  * @device:     the device which we search the free space in
 820  * @num_bytes:  the size of the free space that we need
 821  * @start:      store the start of the free space.
 822  * @len:        the size of the free space. that we find, or the size of the max
 823  *              free space if we don't find suitable free space
 824  *
 825  * this uses a pretty simple search, the expectation is that it is
 826  * called very infrequently and that a given device has a small number
 827  * of extents
 828  *
 829  * @start is used to store the start of the free space if we find. But if we
 830  * don't find suitable free space, it will be used to store the start position
 831  * of the max free space.
 832  *
 833  * @len is used to store the size of the free space that we find.
 834  * But if we don't find suitable free space, it is used to store the size of
 835  * the max free space.
 836  */
 837 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 838                          struct btrfs_device *device, u64 num_bytes,
 839                          u64 *start, u64 *len)
 840 {
 841         struct btrfs_key key;
 842         struct btrfs_root *root = device->dev_root;
 843         struct btrfs_dev_extent *dev_extent;
 844         struct btrfs_path *path;
 845         u64 hole_size;
 846         u64 max_hole_start;
 847         u64 max_hole_size;
 848         u64 extent_end;
 849         u64 search_start;
 850         u64 search_end = device->total_bytes;
 851         int ret;
 852         int slot;
 853         struct extent_buffer *l;
 854
 855         /* FIXME use last free of some kind */
 856
 857         /* we don't want to overwrite the superblock on the drive,
 858          * so we make sure to start at an offset of at least 1MB
 859          */
 860         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 861
 862         max_hole_start = search_start;
 863         max_hole_size = 0;
 864         hole_size = 0;
 865
 866         if (search_start >= search_end) {
 867                 ret = -ENOSPC;
 868                 goto error;
 869         }
 870
 871         path = btrfs_alloc_path();
 872         if (!path) {
 873                 ret = -ENOMEM;
 874                 goto error;
 875         }
 876         path->reada = 2;
 877
 878         key.objectid = device->devid;
 879         key.offset = search_start;
 880         key.type = BTRFS_DEV_EXTENT_KEY;
 881
 882         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 883         if (ret < 0)
 884                 goto out;
 885         if (ret > 0) {
 886                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 887                 if (ret < 0)
 888                         goto out;
 889         }
 890
 891         while (1) {
 892                 l = path->nodes[0];
 893                 slot = path->slots[0];
 894                 if (slot >= btrfs_header_nritems(l)) {
 895                         ret = btrfs_next_leaf(root, path);
 896                         if (ret == 0)
 897                                 continue;
 898                         if (ret < 0)
 899                                 goto out;
 900
 901                         break;
 902                 }
 903                 btrfs_item_key_to_cpu(l, &key, slot);
 904
 905                 if (key.objectid < device->devid)
 906                         goto next;
 907
 908                 if (key.objectid > device->devid)
 909                         break;
 910
 911                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 912                         goto next;
 913
 914                 if (key.offset > search_start) {
 915                         hole_size = key.offset - search_start;
 916
 917                         if (hole_size > max_hole_size) {
 918                                 max_hole_start = search_start;
 919                                 max_hole_size = hole_size;
 920                         }
 921
 922                         /*
 923                          * If this free space is greater than which we need,
 924                          * it must be the max free space that we have found
 925                          * until now, so max_hole_start must point to the start
 926                          * of this free space and the length of this free space
 927                          * is stored in max_hole_size. Thus, we return
 928                          * max_hole_start and max_hole_size and go back to the
 929                          * caller.
 930                          */
 931                         if (hole_size >= num_bytes) {
 932                                 ret = 0;
 933                                 goto out;
 934                         }
 935                 }
 936
 937                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 938                 extent_end = key.offset + btrfs_dev_extent_length(l,
 939                                                                   dev_extent);
 940                 if (extent_end > search_start)
 941                         search_start = extent_end;
 942 next:
 943                 path->slots[0]++;
 944                 cond_resched();
 945         }
 946
 947         /*
 948          * At this point, search_start should be the end of
 949          * allocated dev extents, and when shrinking the device,
 950          * search_end may be smaller than search_start.
 951          */
 952         if (search_end > search_start)
 953                 hole_size = search_end - search_start;
 954
 955         if (hole_size > max_hole_size) {
 956                 max_hole_start = search_start;
 957                 max_hole_size = hole_size;
 958         }
 959
 960         /* See above. */
 961         if (hole_size < num_bytes)
 962                 ret = -ENOSPC;
 963         else
 964                 ret = 0;
 965
 966 out:
 967         btrfs_free_path(path);
 968 error:
 969         *start = max_hole_start;
 970         if (len)
 971                 *len = max_hole_size;
 972         return ret;
 973 }
 974
 975 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 976                           struct btrfs_device *device,
 977                           u64 start)
 978 {
 979         int ret;
 980         struct btrfs_path *path;
 981         struct btrfs_root *root = device->dev_root;
 982         struct btrfs_key key;
 983         struct btrfs_key found_key;
 984         struct extent_buffer *leaf = NULL;
 985         struct btrfs_dev_extent *extent = NULL;
 986
 987         path = btrfs_alloc_path();
 988         if (!path)
 989                 return -ENOMEM;
 990
 991         key.objectid = device->devid;
 992         key.offset = start;
 993         key.type = BTRFS_DEV_EXTENT_KEY;
 994
 995         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 996         if (ret > 0) {
 997                 ret = btrfs_previous_item(root, path, key.objectid,
 998                                           BTRFS_DEV_EXTENT_KEY);
 999                 if (ret)
1000                         goto out;
1001                 leaf = path->nodes[0];
1002                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1003                 extent = btrfs_item_ptr(leaf, path->slots[0],
1004                                         struct btrfs_dev_extent);
1005                 BUG_ON(found_key.offset > start || found_key.offset +
1006                        btrfs_dev_extent_length(leaf, extent) < start);
1007         } else if (ret == 0) {
1008                 leaf = path->nodes[0];
1009                 extent = btrfs_item_ptr(leaf, path->slots[0],
1010                                         struct btrfs_dev_extent);
1011         }
1012         BUG_ON(ret);
1013
1014         if (device->bytes_used > 0) {
1015                 u64 len = btrfs_dev_extent_length(leaf, extent);
1016                 device->bytes_used -= len;
1017                 spin_lock(&root->fs_info->free_chunk_lock);
1018                 root->fs_info->free_chunk_space += len;
1019                 spin_unlock(&root->fs_info->free_chunk_lock);
1020         }
1021         ret = btrfs_del_item(trans, root, path);
1022
1023 out:
1024         btrfs_free_path(path);
1025         return ret;
1026 }
1027
1028 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1029                            struct btrfs_device *device,
1030                            u64 chunk_tree, u64 chunk_objectid,
1031                            u64 chunk_offset, u64 start, u64 num_bytes)
1032 {
1033         int ret;
1034         struct btrfs_path *path;
1035         struct btrfs_root *root = device->dev_root;
1036         struct btrfs_dev_extent *extent;
1037         struct extent_buffer *leaf;
1038         struct btrfs_key key;
1039
1040         WARN_ON(!device->in_fs_metadata);
1041         path = btrfs_alloc_path();
1042         if (!path)
1043                 return -ENOMEM;
1044
1045         key.objectid = device->devid;
1046         key.offset = start;
1047         key.type = BTRFS_DEV_EXTENT_KEY;
1048         ret = btrfs_insert_empty_item(trans, root, path, &key,
1049                                       sizeof(*extent));
1050         BUG_ON(ret);
1051
1052         leaf = path->nodes[0];
1053         extent = btrfs_item_ptr(leaf, path->slots[0],
1054                                 struct btrfs_dev_extent);
1055         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1056         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1057         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1058
1059         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1060                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1061                     BTRFS_UUID_SIZE);
1062
1063         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1064         btrfs_mark_buffer_dirty(leaf);
1065         btrfs_free_path(path);
1066         return ret;
1067 }
1068
1069 static noinline int find_next_chunk(struct btrfs_root *root,
1070                                     u64 objectid, u64 *offset)
1071 {
1072         struct btrfs_path *path;
1073         int ret;
1074         struct btrfs_key key;
1075         struct btrfs_chunk *chunk;
1076         struct btrfs_key found_key;
1077
1078         path = btrfs_alloc_path();
1079         if (!path)
1080                 return -ENOMEM;
1081
1082         key.objectid = objectid;
1083         key.offset = (u64)-1;
1084         key.type = BTRFS_CHUNK_ITEM_KEY;
1085
1086         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1087         if (ret < 0)
1088                 goto error;
1089
1090         BUG_ON(ret == 0);
1091
1092         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1093         if (ret) {
1094                 *offset = 0;
1095         } else {
1096                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1097                                       path->slots[0]);
1098                 if (found_key.objectid != objectid)
1099                         *offset = 0;
1100                 else {
1101                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1102                                                struct btrfs_chunk);
1103                         *offset = found_key.offset +
1104                                 btrfs_chunk_length(path->nodes[0], chunk);
1105                 }
1106         }
1107         ret = 0;
1108 error:
1109         btrfs_free_path(path);
1110         return ret;
1111 }
1112
1113 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1114 {
1115         int ret;
1116         struct btrfs_key key;
1117         struct btrfs_key found_key;
1118         struct btrfs_path *path;
1119
1120         root = root->fs_info->chunk_root;
1121
1122         path = btrfs_alloc_path();
1123         if (!path)
1124                 return -ENOMEM;
1125
1126         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1127         key.type = BTRFS_DEV_ITEM_KEY;
1128         key.offset = (u64)-1;
1129
1130         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1131         if (ret < 0)
1132                 goto error;
1133
1134         BUG_ON(ret == 0);
1135
1136         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1137                                   BTRFS_DEV_ITEM_KEY);
1138         if (ret) {
1139                 *objectid = 1;
1140         } else {
1141                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1142                                       path->slots[0]);
1143                 *objectid = found_key.offset + 1;
1144         }
1145         ret = 0;
1146 error:
1147         btrfs_free_path(path);
1148         return ret;
1149 }
1150
1151 /*
1152  * the device information is stored in the chunk root
1153  * the btrfs_device struct should be fully filled in
1154  */
1155 int btrfs_add_device(struct btrfs_trans_handle *trans,
1156                      struct btrfs_root *root,
1157                      struct btrfs_device *device)
1158 {
1159         int ret;
1160         struct btrfs_path *path;
1161         struct btrfs_dev_item *dev_item;
1162         struct extent_buffer *leaf;
1163         struct btrfs_key key;
1164         unsigned long ptr;
1165
1166         root = root->fs_info->chunk_root;
1167
1168         path = btrfs_alloc_path();
1169         if (!path)
1170                 return -ENOMEM;
1171
1172         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1173         key.type = BTRFS_DEV_ITEM_KEY;
1174         key.offset = device->devid;
1175
1176         ret = btrfs_insert_empty_item(trans, root, path, &key,
1177                                       sizeof(*dev_item));
1178         if (ret)
1179                 goto out;
1180
1181         leaf = path->nodes[0];
1182         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1183
1184         btrfs_set_device_id(leaf, dev_item, device->devid);
1185         btrfs_set_device_generation(leaf, dev_item, 0);
1186         btrfs_set_device_type(leaf, dev_item, device->type);
1187         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1188         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1189         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1190         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1191         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1192         btrfs_set_device_group(leaf, dev_item, 0);
1193         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1194         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1195         btrfs_set_device_start_offset(leaf, dev_item, 0);
1196
1197         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1198         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1199         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1200         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1201         btrfs_mark_buffer_dirty(leaf);
1202
1203         ret = 0;
1204 out:
1205         btrfs_free_path(path);
1206         return ret;
1207 }
1208
1209 static int btrfs_rm_dev_item(struct btrfs_root *root,
1210                              struct btrfs_device *device)
1211 {
1212         int ret;
1213         struct btrfs_path *path;
1214         struct btrfs_key key;
1215         struct btrfs_trans_handle *trans;
1216
1217         root = root->fs_info->chunk_root;
1218
1219         path = btrfs_alloc_path();
1220         if (!path)
1221                 return -ENOMEM;
1222
1223         trans = btrfs_start_transaction(root, 0);
1224         if (IS_ERR(trans)) {
1225                 btrfs_free_path(path);
1226                 return PTR_ERR(trans);
1227         }
1228         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1229         key.type = BTRFS_DEV_ITEM_KEY;
1230         key.offset = device->devid;
1231         lock_chunks(root);
1232
1233         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1234         if (ret < 0)
1235                 goto out;
1236
1237         if (ret > 0) {
1238                 ret = -ENOENT;
1239                 goto out;
1240         }
1241
1242         ret = btrfs_del_item(trans, root, path);
1243         if (ret)
1244                 goto out;
1245 out:
1246         btrfs_free_path(path);
1247         unlock_chunks(root);
1248         btrfs_commit_transaction(trans, root);
1249         return ret;
1250 }
1251
1252 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1253 {
1254         struct btrfs_device *device;
1255         struct btrfs_device *next_device;
1256         struct block_device *bdev;
1257         struct buffer_head *bh = NULL;
1258         struct btrfs_super_block *disk_super;
1259         struct btrfs_fs_devices *cur_devices;
1260         u64 all_avail;
1261         u64 devid;
1262         u64 num_devices;
1263         u8 *dev_uuid;
1264         int ret = 0;
1265         bool clear_super = false;
1266
1267         mutex_lock(&uuid_mutex);
1268         mutex_lock(&root->fs_info->volume_mutex);
1269
1270         all_avail = root->fs_info->avail_data_alloc_bits |
1271                 root->fs_info->avail_system_alloc_bits |
1272                 root->fs_info->avail_metadata_alloc_bits;
1273
1274         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1275             root->fs_info->fs_devices->num_devices <= 4) {
1276                 printk(KERN_ERR "btrfs: unable to go below four devices "
1277                        "on raid10\n");
1278                 ret = -EINVAL;
1279                 goto out;
1280         }
1281
1282         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1283             root->fs_info->fs_devices->num_devices <= 2) {
1284                 printk(KERN_ERR "btrfs: unable to go below two "
1285                        "devices on raid1\n");
1286                 ret = -EINVAL;
1287                 goto out;
1288         }
1289
1290         if (strcmp(device_path, "missing") == 0) {
1291                 struct list_head *devices;
1292                 struct btrfs_device *tmp;
1293
1294                 device = NULL;
1295                 devices = &root->fs_info->fs_devices->devices;
1296                 /*
1297                  * It is safe to read the devices since the volume_mutex
1298                  * is held.
1299                  */
1300                 list_for_each_entry(tmp, devices, dev_list) {
1301                         if (tmp->in_fs_metadata && !tmp->bdev) {
1302                                 device = tmp;
1303                                 break;
1304                         }
1305                 }
1306                 bdev = NULL;
1307                 bh = NULL;
1308                 disk_super = NULL;
1309                 if (!device) {
1310                         printk(KERN_ERR "btrfs: no missing devices found to "
1311                                "remove\n");
1312                         goto out;
1313                 }
1314         } else {
1315                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1316                                           root->fs_info->bdev_holder);
1317                 if (IS_ERR(bdev)) {
1318                         ret = PTR_ERR(bdev);
1319                         goto out;
1320                 }
1321
1322                 set_blocksize(bdev, 4096);
1323                 bh = btrfs_read_dev_super(bdev);
1324                 if (!bh) {
1325                         ret = -EINVAL;
1326                         goto error_close;
1327                 }
1328                 disk_super = (struct btrfs_super_block *)bh->b_data;
1329                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1330                 dev_uuid = disk_super->dev_item.uuid;
1331                 device = btrfs_find_device(root, devid, dev_uuid,
1332                                            disk_super->fsid);
1333                 if (!device) {
1334                         ret = -ENOENT;
1335                         goto error_brelse;
1336                 }
1337         }
1338
1339         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1340                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1341                        "device\n");
1342                 ret = -EINVAL;
1343                 goto error_brelse;
1344         }
1345
1346         if (device->writeable) {
1347                 lock_chunks(root);
1348                 list_del_init(&device->dev_alloc_list);
1349                 unlock_chunks(root);
1350                 root->fs_info->fs_devices->rw_devices--;
1351                 clear_super = true;
1352         }
1353
1354         ret = btrfs_shrink_device(device, 0);
1355         if (ret)
1356                 goto error_undo;
1357
1358         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1359         if (ret)
1360                 goto error_undo;
1361
1362         spin_lock(&root->fs_info->free_chunk_lock);
1363         root->fs_info->free_chunk_space = device->total_bytes -
1364                 device->bytes_used;
1365         spin_unlock(&root->fs_info->free_chunk_lock);
1366
1367         device->in_fs_metadata = 0;
1368         btrfs_scrub_cancel_dev(root, device);
1369
1370         /*
1371          * the device list mutex makes sure that we don't change
1372          * the device list while someone else is writing out all
1373          * the device supers.
1374          */
1375
1376         cur_devices = device->fs_devices;
1377         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1378         list_del_rcu(&device->dev_list);
1379
1380         device->fs_devices->num_devices--;
1381
1382         if (device->missing)
1383                 root->fs_info->fs_devices->missing_devices--;
1384
1385         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1386                                  struct btrfs_device, dev_list);
1387         if (device->bdev == root->fs_info->sb->s_bdev)
1388                 root->fs_info->sb->s_bdev = next_device->bdev;
1389         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1390                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1391
1392         if (device->bdev)
1393                 device->fs_devices->open_devices--;
1394
1395         call_rcu(&device->rcu, free_device);
1396         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1397
1398         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1399         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1400
1401         if (cur_devices->open_devices == 0) {
1402                 struct btrfs_fs_devices *fs_devices;
1403                 fs_devices = root->fs_info->fs_devices;
1404                 while (fs_devices) {
1405                         if (fs_devices->seed == cur_devices)
1406                                 break;
1407                         fs_devices = fs_devices->seed;
1408                 }
1409                 fs_devices->seed = cur_devices->seed;
1410                 cur_devices->seed = NULL;
1411                 lock_chunks(root);
1412                 __btrfs_close_devices(cur_devices);
1413                 unlock_chunks(root);
1414                 free_fs_devices(cur_devices);
1415         }
1416
1417         /*
1418          * at this point, the device is zero sized.  We want to
1419          * remove it from the devices list and zero out the old super
1420          */
1421         if (clear_super) {
1422                 /* make sure this device isn't detected as part of
1423                  * the FS anymore
1424                  */
1425                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1426                 set_buffer_dirty(bh);
1427                 sync_dirty_buffer(bh);
1428         }
1429
1430         ret = 0;
1431
1432 error_brelse:
1433         brelse(bh);
1434 error_close:
1435         if (bdev)
1436                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1437 out:
1438         mutex_unlock(&root->fs_info->volume_mutex);
1439         mutex_unlock(&uuid_mutex);
1440         return ret;
1441 error_undo:
1442         if (device->writeable) {
1443                 lock_chunks(root);
1444                 list_add(&device->dev_alloc_list,
1445                          &root->fs_info->fs_devices->alloc_list);
1446                 unlock_chunks(root);
1447                 root->fs_info->fs_devices->rw_devices++;
1448         }
1449         goto error_brelse;
1450 }
1451
1452 /*
1453  * does all the dirty work required for changing file system's UUID.
1454  */
1455 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1456                                 struct btrfs_root *root)
1457 {
1458         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1459         struct btrfs_fs_devices *old_devices;
1460         struct btrfs_fs_devices *seed_devices;
1461         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1462         struct btrfs_device *device;
1463         u64 super_flags;
1464
1465         BUG_ON(!mutex_is_locked(&uuid_mutex));
1466         if (!fs_devices->seeding)
1467                 return -EINVAL;
1468
1469         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1470         if (!seed_devices)
1471                 return -ENOMEM;
1472
1473         old_devices = clone_fs_devices(fs_devices);
1474         if (IS_ERR(old_devices)) {
1475                 kfree(seed_devices);
1476                 return PTR_ERR(old_devices);
1477         }
1478
1479         list_add(&old_devices->list, &fs_uuids);
1480
1481         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1482         seed_devices->opened = 1;
1483         INIT_LIST_HEAD(&seed_devices->devices);
1484         INIT_LIST_HEAD(&seed_devices->alloc_list);
1485         mutex_init(&seed_devices->device_list_mutex);
1486
1487         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1488         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1489                               synchronize_rcu);
1490         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1491
1492         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1493         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1494                 device->fs_devices = seed_devices;
1495         }
1496
1497         fs_devices->seeding = 0;
1498         fs_devices->num_devices = 0;
1499         fs_devices->open_devices = 0;
1500         fs_devices->seed = seed_devices;
1501
1502         generate_random_uuid(fs_devices->fsid);
1503         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1504         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1505         super_flags = btrfs_super_flags(disk_super) &
1506                       ~BTRFS_SUPER_FLAG_SEEDING;
1507         btrfs_set_super_flags(disk_super, super_flags);
1508
1509         return 0;
1510 }
1511
1512 /*
1513  * strore the expected generation for seed devices in device items.
1514  */
1515 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1516                                struct btrfs_root *root)
1517 {
1518         struct btrfs_path *path;
1519         struct extent_buffer *leaf;
1520         struct btrfs_dev_item *dev_item;
1521         struct btrfs_device *device;
1522         struct btrfs_key key;
1523         u8 fs_uuid[BTRFS_UUID_SIZE];
1524         u8 dev_uuid[BTRFS_UUID_SIZE];
1525         u64 devid;
1526         int ret;
1527
1528         path = btrfs_alloc_path();
1529         if (!path)
1530                 return -ENOMEM;
1531
1532         root = root->fs_info->chunk_root;
1533         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1534         key.offset = 0;
1535         key.type = BTRFS_DEV_ITEM_KEY;
1536
1537         while (1) {
1538                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1539                 if (ret < 0)
1540                         goto error;
1541
1542                 leaf = path->nodes[0];
1543 next_slot:
1544                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1545                         ret = btrfs_next_leaf(root, path);
1546                         if (ret > 0)
1547                                 break;
1548                         if (ret < 0)
1549                                 goto error;
1550                         leaf = path->nodes[0];
1551                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1552                         btrfs_release_path(path);
1553                         continue;
1554                 }
1555
1556                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1557                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1558                     key.type != BTRFS_DEV_ITEM_KEY)
1559                         break;
1560
1561                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1562                                           struct btrfs_dev_item);
1563                 devid = btrfs_device_id(leaf, dev_item);
1564                 read_extent_buffer(leaf, dev_uuid,
1565                                    (unsigned long)btrfs_device_uuid(dev_item),
1566                                    BTRFS_UUID_SIZE);
1567                 read_extent_buffer(leaf, fs_uuid,
1568                                    (unsigned long)btrfs_device_fsid(dev_item),
1569                                    BTRFS_UUID_SIZE);
1570                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1571                 BUG_ON(!device);
1572
1573                 if (device->fs_devices->seeding) {
1574                         btrfs_set_device_generation(leaf, dev_item,
1575                                                     device->generation);
1576                         btrfs_mark_buffer_dirty(leaf);
1577                 }
1578
1579                 path->slots[0]++;
1580                 goto next_slot;
1581         }
1582         ret = 0;
1583 error:
1584         btrfs_free_path(path);
1585         return ret;
1586 }
1587
1588 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1589 {
1590         struct request_queue *q;
1591         struct btrfs_trans_handle *trans;
1592         struct btrfs_device *device;
1593         struct block_device *bdev;
1594         struct list_head *devices;
1595         struct super_block *sb = root->fs_info->sb;
1596         u64 total_bytes;
1597         int seeding_dev = 0;
1598         int ret = 0;
1599
1600         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1601                 return -EINVAL;
1602
1603         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1604                                   root->fs_info->bdev_holder);
1605         if (IS_ERR(bdev))
1606                 return PTR_ERR(bdev);
1607
1608         if (root->fs_info->fs_devices->seeding) {
1609                 seeding_dev = 1;
1610                 down_write(&sb->s_umount);
1611                 mutex_lock(&uuid_mutex);
1612         }
1613
1614         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1615         mutex_lock(&root->fs_info->volume_mutex);
1616
1617         devices = &root->fs_info->fs_devices->devices;
1618         /*
1619          * we have the volume lock, so we don't need the extra
1620          * device list mutex while reading the list here.
1621          */
1622         list_for_each_entry(device, devices, dev_list) {
1623                 if (device->bdev == bdev) {
1624                         ret = -EEXIST;
1625                         goto error;
1626                 }
1627         }
1628
1629         device = kzalloc(sizeof(*device), GFP_NOFS);
1630         if (!device) {
1631                 /* we can safely leave the fs_devices entry around */
1632                 ret = -ENOMEM;
1633                 goto error;
1634         }
1635
1636         device->name = kstrdup(device_path, GFP_NOFS);
1637         if (!device->name) {
1638                 kfree(device);
1639                 ret = -ENOMEM;
1640                 goto error;
1641         }
1642
1643         ret = find_next_devid(root, &device->devid);
1644         if (ret) {
1645                 kfree(device->name);
1646                 kfree(device);
1647                 goto error;
1648         }
1649
1650         trans = btrfs_start_transaction(root, 0);
1651         if (IS_ERR(trans)) {
1652                 kfree(device->name);
1653                 kfree(device);
1654                 ret = PTR_ERR(trans);
1655                 goto error;
1656         }
1657
1658         lock_chunks(root);
1659
1660         q = bdev_get_queue(bdev);
1661         if (blk_queue_discard(q))
1662                 device->can_discard = 1;
1663         device->writeable = 1;
1664         device->work.func = pending_bios_fn;
1665         generate_random_uuid(device->uuid);
1666         spin_lock_init(&device->io_lock);
1667         device->generation = trans->transid;
1668         device->io_width = root->sectorsize;
1669         device->io_align = root->sectorsize;
1670         device->sector_size = root->sectorsize;
1671         device->total_bytes = i_size_read(bdev->bd_inode);
1672         device->disk_total_bytes = device->total_bytes;
1673         device->dev_root = root->fs_info->dev_root;
1674         device->bdev = bdev;
1675         device->in_fs_metadata = 1;
1676         device->mode = FMODE_EXCL;
1677         set_blocksize(device->bdev, 4096);
1678
1679         if (seeding_dev) {
1680                 sb->s_flags &= ~MS_RDONLY;
1681                 ret = btrfs_prepare_sprout(trans, root);
1682                 BUG_ON(ret);
1683         }
1684
1685         device->fs_devices = root->fs_info->fs_devices;
1686
1687         /*
1688          * we don't want write_supers to jump in here with our device
1689          * half setup
1690          */
1691         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1692         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1693         list_add(&device->dev_alloc_list,
1694                  &root->fs_info->fs_devices->alloc_list);
1695         root->fs_info->fs_devices->num_devices++;
1696         root->fs_info->fs_devices->open_devices++;
1697         root->fs_info->fs_devices->rw_devices++;
1698         if (device->can_discard)
1699                 root->fs_info->fs_devices->num_can_discard++;
1700         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1701
1702         spin_lock(&root->fs_info->free_chunk_lock);
1703         root->fs_info->free_chunk_space += device->total_bytes;
1704         spin_unlock(&root->fs_info->free_chunk_lock);
1705
1706         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1707                 root->fs_info->fs_devices->rotating = 1;
1708
1709         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1710         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1711                                     total_bytes + device->total_bytes);
1712
1713         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1714         btrfs_set_super_num_devices(root->fs_info->super_copy,
1715                                     total_bytes + 1);
1716         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1717
1718         if (seeding_dev) {
1719                 ret = init_first_rw_device(trans, root, device);
1720                 BUG_ON(ret);
1721                 ret = btrfs_finish_sprout(trans, root);
1722                 BUG_ON(ret);
1723         } else {
1724                 ret = btrfs_add_device(trans, root, device);
1725         }
1726
1727         /*
1728          * we've got more storage, clear any full flags on the space
1729          * infos
1730          */
1731         btrfs_clear_space_info_full(root->fs_info);
1732
1733         unlock_chunks(root);
1734         btrfs_commit_transaction(trans, root);
1735
1736         if (seeding_dev) {
1737                 mutex_unlock(&uuid_mutex);
1738                 up_write(&sb->s_umount);
1739
1740                 ret = btrfs_relocate_sys_chunks(root);
1741                 BUG_ON(ret);
1742         }
1743 out:
1744         mutex_unlock(&root->fs_info->volume_mutex);
1745         return ret;
1746 error:
1747         blkdev_put(bdev, FMODE_EXCL);
1748         if (seeding_dev) {
1749                 mutex_unlock(&uuid_mutex);
1750                 up_write(&sb->s_umount);
1751         }
1752         goto out;
1753 }
1754
1755 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1756                                         struct btrfs_device *device)
1757 {
1758         int ret;
1759         struct btrfs_path *path;
1760         struct btrfs_root *root;
1761         struct btrfs_dev_item *dev_item;
1762         struct extent_buffer *leaf;
1763         struct btrfs_key key;
1764
1765         root = device->dev_root->fs_info->chunk_root;
1766
1767         path = btrfs_alloc_path();
1768         if (!path)
1769                 return -ENOMEM;
1770
1771         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1772         key.type = BTRFS_DEV_ITEM_KEY;
1773         key.offset = device->devid;
1774
1775         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1776         if (ret < 0)
1777                 goto out;
1778
1779         if (ret > 0) {
1780                 ret = -ENOENT;
1781                 goto out;
1782         }
1783
1784         leaf = path->nodes[0];
1785         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1786
1787         btrfs_set_device_id(leaf, dev_item, device->devid);
1788         btrfs_set_device_type(leaf, dev_item, device->type);
1789         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1790         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1791         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1792         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1793         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1794         btrfs_mark_buffer_dirty(leaf);
1795
1796 out:
1797         btrfs_free_path(path);
1798         return ret;
1799 }
1800
1801 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1802                       struct btrfs_device *device, u64 new_size)
1803 {
1804         struct btrfs_super_block *super_copy =
1805                 device->dev_root->fs_info->super_copy;
1806         u64 old_total = btrfs_super_total_bytes(super_copy);
1807         u64 diff = new_size - device->total_bytes;
1808
1809         if (!device->writeable)
1810                 return -EACCES;
1811         if (new_size <= device->total_bytes)
1812                 return -EINVAL;
1813
1814         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1815         device->fs_devices->total_rw_bytes += diff;
1816
1817         device->total_bytes = new_size;
1818         device->disk_total_bytes = new_size;
1819         btrfs_clear_space_info_full(device->dev_root->fs_info);
1820
1821         return btrfs_update_device(trans, device);
1822 }
1823
1824 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1825                       struct btrfs_device *device, u64 new_size)
1826 {
1827         int ret;
1828         lock_chunks(device->dev_root);
1829         ret = __btrfs_grow_device(trans, device, new_size);
1830         unlock_chunks(device->dev_root);
1831         return ret;
1832 }
1833
1834 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1835                             struct btrfs_root *root,
1836                             u64 chunk_tree, u64 chunk_objectid,
1837                             u64 chunk_offset)
1838 {
1839         int ret;
1840         struct btrfs_path *path;
1841         struct btrfs_key key;
1842
1843         root = root->fs_info->chunk_root;
1844         path = btrfs_alloc_path();
1845         if (!path)
1846                 return -ENOMEM;
1847
1848         key.objectid = chunk_objectid;
1849         key.offset = chunk_offset;
1850         key.type = BTRFS_CHUNK_ITEM_KEY;
1851
1852         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1853         BUG_ON(ret);
1854
1855         ret = btrfs_del_item(trans, root, path);
1856
1857         btrfs_free_path(path);
1858         return ret;
1859 }
1860
1861 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1862                         chunk_offset)
1863 {
1864         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1865         struct btrfs_disk_key *disk_key;
1866         struct btrfs_chunk *chunk;
1867         u8 *ptr;
1868         int ret = 0;
1869         u32 num_stripes;
1870         u32 array_size;
1871         u32 len = 0;
1872         u32 cur;
1873         struct btrfs_key key;
1874
1875         array_size = btrfs_super_sys_array_size(super_copy);
1876
1877         ptr = super_copy->sys_chunk_array;
1878         cur = 0;
1879
1880         while (cur < array_size) {
1881                 disk_key = (struct btrfs_disk_key *)ptr;
1882                 btrfs_disk_key_to_cpu(&key, disk_key);
1883
1884                 len = sizeof(*disk_key);
1885
1886                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1887                         chunk = (struct btrfs_chunk *)(ptr + len);
1888                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1889                         len += btrfs_chunk_item_size(num_stripes);
1890                 } else {
1891                         ret = -EIO;
1892                         break;
1893                 }
1894                 if (key.objectid == chunk_objectid &&
1895                     key.offset == chunk_offset) {
1896                         memmove(ptr, ptr + len, array_size - (cur + len));
1897                         array_size -= len;
1898                         btrfs_set_super_sys_array_size(super_copy, array_size);
1899                 } else {
1900                         ptr += len;
1901                         cur += len;
1902                 }
1903         }
1904         return ret;
1905 }
1906
1907 static int btrfs_relocate_chunk(struct btrfs_root *root,
1908                          u64 chunk_tree, u64 chunk_objectid,
1909                          u64 chunk_offset)
1910 {
1911         struct extent_map_tree *em_tree;
1912         struct btrfs_root *extent_root;
1913         struct btrfs_trans_handle *trans;
1914         struct extent_map *em;
1915         struct map_lookup *map;
1916         int ret;
1917         int i;
1918
1919         root = root->fs_info->chunk_root;
1920         extent_root = root->fs_info->extent_root;
1921         em_tree = &root->fs_info->mapping_tree.map_tree;
1922
1923         ret = btrfs_can_relocate(extent_root, chunk_offset);
1924         if (ret)
1925                 return -ENOSPC;
1926
1927         /* step one, relocate all the extents inside this chunk */
1928         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1929         if (ret)
1930                 return ret;
1931
1932         trans = btrfs_start_transaction(root, 0);
1933         BUG_ON(IS_ERR(trans));
1934
1935         lock_chunks(root);
1936
1937         /*
1938          * step two, delete the device extents and the
1939          * chunk tree entries
1940          */
1941         read_lock(&em_tree->lock);
1942         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1943         read_unlock(&em_tree->lock);
1944
1945         BUG_ON(em->start > chunk_offset ||
1946                em->start + em->len < chunk_offset);
1947         map = (struct map_lookup *)em->bdev;
1948
1949         for (i = 0; i < map->num_stripes; i++) {
1950                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1951                                             map->stripes[i].physical);
1952                 BUG_ON(ret);
1953
1954                 if (map->stripes[i].dev) {
1955                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1956                         BUG_ON(ret);
1957                 }
1958         }
1959         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1960                                chunk_offset);
1961
1962         BUG_ON(ret);
1963
1964         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1965
1966         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1967                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1968                 BUG_ON(ret);
1969         }
1970
1971         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1972         BUG_ON(ret);
1973
1974         write_lock(&em_tree->lock);
1975         remove_extent_mapping(em_tree, em);
1976         write_unlock(&em_tree->lock);
1977
1978         kfree(map);
1979         em->bdev = NULL;
1980
1981         /* once for the tree */
1982         free_extent_map(em);
1983         /* once for us */
1984         free_extent_map(em);
1985
1986         unlock_chunks(root);
1987         btrfs_end_transaction(trans, root);
1988         return 0;
1989 }
1990
1991 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1992 {
1993         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1994         struct btrfs_path *path;
1995         struct extent_buffer *leaf;
1996         struct btrfs_chunk *chunk;
1997         struct btrfs_key key;
1998         struct btrfs_key found_key;
1999         u64 chunk_tree = chunk_root->root_key.objectid;
2000         u64 chunk_type;
2001         bool retried = false;
2002         int failed = 0;
2003         int ret;
2004
2005         path = btrfs_alloc_path();
2006         if (!path)
2007                 return -ENOMEM;
2008
2009 again:
2010         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2011         key.offset = (u64)-1;
2012         key.type = BTRFS_CHUNK_ITEM_KEY;
2013
2014         while (1) {
2015                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2016                 if (ret < 0)
2017                         goto error;
2018                 BUG_ON(ret == 0);
2019
2020                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2021                                           key.type);
2022                 if (ret < 0)
2023                         goto error;
2024                 if (ret > 0)
2025                         break;
2026
2027                 leaf = path->nodes[0];
2028                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2029
2030                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2031                                        struct btrfs_chunk);
2032                 chunk_type = btrfs_chunk_type(leaf, chunk);
2033                 btrfs_release_path(path);
2034
2035                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2036                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2037                                                    found_key.objectid,
2038                                                    found_key.offset);
2039                         if (ret == -ENOSPC)
2040                                 failed++;
2041                         else if (ret)
2042                                 BUG();
2043                 }
2044
2045                 if (found_key.offset == 0)
2046                         break;
2047                 key.offset = found_key.offset - 1;
2048         }
2049         ret = 0;
2050         if (failed && !retried) {
2051                 failed = 0;
2052                 retried = true;
2053                 goto again;
2054         } else if (failed && retried) {
2055                 WARN_ON(1);
2056                 ret = -ENOSPC;
2057         }
2058 error:
2059         btrfs_free_path(path);
2060         return ret;
2061 }
2062
2063 static u64 div_factor(u64 num, int factor)
2064 {
2065         if (factor == 10)
2066                 return num;
2067         num *= factor;
2068         do_div(num, 10);
2069         return num;
2070 }
2071
2072 int btrfs_balance(struct btrfs_root *dev_root)
2073 {
2074         int ret;
2075         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2076         struct btrfs_device *device;
2077         u64 old_size;
2078         u64 size_to_free;
2079         struct btrfs_path *path;
2080         struct btrfs_key key;
2081         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2082         struct btrfs_trans_handle *trans;
2083         struct btrfs_key found_key;
2084
2085         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2086                 return -EROFS;
2087
2088         if (!capable(CAP_SYS_ADMIN))
2089                 return -EPERM;
2090
2091         mutex_lock(&dev_root->fs_info->volume_mutex);
2092         dev_root = dev_root->fs_info->dev_root;
2093
2094         /* step one make some room on all the devices */
2095         list_for_each_entry(device, devices, dev_list) {
2096                 old_size = device->total_bytes;
2097                 size_to_free = div_factor(old_size, 1);
2098                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2099                 if (!device->writeable ||
2100                     device->total_bytes - device->bytes_used > size_to_free)
2101                         continue;
2102
2103                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2104                 if (ret == -ENOSPC)
2105                         break;
2106                 BUG_ON(ret);
2107
2108                 trans = btrfs_start_transaction(dev_root, 0);
2109                 BUG_ON(IS_ERR(trans));
2110
2111                 ret = btrfs_grow_device(trans, device, old_size);
2112                 BUG_ON(ret);
2113
2114                 btrfs_end_transaction(trans, dev_root);
2115         }
2116
2117         /* step two, relocate all the chunks */
2118         path = btrfs_alloc_path();
2119         if (!path) {
2120                 ret = -ENOMEM;
2121                 goto error;
2122         }
2123         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2124         key.offset = (u64)-1;
2125         key.type = BTRFS_CHUNK_ITEM_KEY;
2126
2127         while (1) {
2128                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2129                 if (ret < 0)
2130                         goto error;
2131
2132                 /*
2133                  * this shouldn't happen, it means the last relocate
2134                  * failed
2135                  */
2136                 if (ret == 0)
2137                         break;
2138
2139                 ret = btrfs_previous_item(chunk_root, path, 0,
2140                                           BTRFS_CHUNK_ITEM_KEY);
2141                 if (ret)
2142                         break;
2143
2144                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2145                                       path->slots[0]);
2146                 if (found_key.objectid != key.objectid)
2147                         break;
2148
2149                 /* chunk zero is special */
2150                 if (found_key.offset == 0)
2151                         break;
2152
2153                 btrfs_release_path(path);
2154                 ret = btrfs_relocate_chunk(chunk_root,
2155                                            chunk_root->root_key.objectid,
2156                                            found_key.objectid,
2157                                            found_key.offset);
2158                 if (ret && ret != -ENOSPC)
2159                         goto error;
2160                 key.offset = found_key.offset - 1;
2161         }
2162         ret = 0;
2163 error:
2164         btrfs_free_path(path);
2165         mutex_unlock(&dev_root->fs_info->volume_mutex);
2166         return ret;
2167 }
2168
2169 /*
2170  * shrinking a device means finding all of the device extents past
2171  * the new size, and then following the back refs to the chunks.
2172  * The chunk relocation code actually frees the device extent
2173  */
2174 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2175 {
2176         struct btrfs_trans_handle *trans;
2177         struct btrfs_root *root = device->dev_root;
2178         struct btrfs_dev_extent *dev_extent = NULL;
2179         struct btrfs_path *path;
2180         u64 length;
2181         u64 chunk_tree;
2182         u64 chunk_objectid;
2183         u64 chunk_offset;
2184         int ret;
2185         int slot;
2186         int failed = 0;
2187         bool retried = false;
2188         struct extent_buffer *l;
2189         struct btrfs_key key;
2190         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2191         u64 old_total = btrfs_super_total_bytes(super_copy);
2192         u64 old_size = device->total_bytes;
2193         u64 diff = device->total_bytes - new_size;
2194
2195         if (new_size >= device->total_bytes)
2196                 return -EINVAL;
2197
2198         path = btrfs_alloc_path();
2199         if (!path)
2200                 return -ENOMEM;
2201
2202         path->reada = 2;
2203
2204         lock_chunks(root);
2205
2206         device->total_bytes = new_size;
2207         if (device->writeable) {
2208                 device->fs_devices->total_rw_bytes -= diff;
2209                 spin_lock(&root->fs_info->free_chunk_lock);
2210                 root->fs_info->free_chunk_space -= diff;
2211                 spin_unlock(&root->fs_info->free_chunk_lock);
2212         }
2213         unlock_chunks(root);
2214
2215 again:
2216         key.objectid = device->devid;
2217         key.offset = (u64)-1;
2218         key.type = BTRFS_DEV_EXTENT_KEY;
2219
2220         while (1) {
2221                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2222                 if (ret < 0)
2223                         goto done;
2224
2225                 ret = btrfs_previous_item(root, path, 0, key.type);
2226                 if (ret < 0)
2227                         goto done;
2228                 if (ret) {
2229                         ret = 0;
2230                         btrfs_release_path(path);
2231                         break;
2232                 }
2233
2234                 l = path->nodes[0];
2235                 slot = path->slots[0];
2236                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2237
2238                 if (key.objectid != device->devid) {
2239                         btrfs_release_path(path);
2240                         break;
2241                 }
2242
2243                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2244                 length = btrfs_dev_extent_length(l, dev_extent);
2245
2246                 if (key.offset + length <= new_size) {
2247                         btrfs_release_path(path);
2248                         break;
2249                 }
2250
2251                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2252                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2253                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2254                 btrfs_release_path(path);
2255
2256                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2257                                            chunk_offset);
2258                 if (ret && ret != -ENOSPC)
2259                         goto done;
2260                 if (ret == -ENOSPC)
2261                         failed++;
2262                 key.offset -= 1;
2263         }
2264
2265         if (failed && !retried) {
2266                 failed = 0;
2267                 retried = true;
2268                 goto again;
2269         } else if (failed && retried) {
2270                 ret = -ENOSPC;
2271                 lock_chunks(root);
2272
2273                 device->total_bytes = old_size;
2274                 if (device->writeable)
2275                         device->fs_devices->total_rw_bytes += diff;
2276                 spin_lock(&root->fs_info->free_chunk_lock);
2277                 root->fs_info->free_chunk_space += diff;
2278                 spin_unlock(&root->fs_info->free_chunk_lock);
2279                 unlock_chunks(root);
2280                 goto done;
2281         }
2282
2283         /* Shrinking succeeded, else we would be at "done". */
2284         trans = btrfs_start_transaction(root, 0);
2285         if (IS_ERR(trans)) {
2286                 ret = PTR_ERR(trans);
2287                 goto done;
2288         }
2289
2290         lock_chunks(root);
2291
2292         device->disk_total_bytes = new_size;
2293         /* Now btrfs_update_device() will change the on-disk size. */
2294         ret = btrfs_update_device(trans, device);
2295         if (ret) {
2296                 unlock_chunks(root);
2297                 btrfs_end_transaction(trans, root);
2298                 goto done;
2299         }
2300         WARN_ON(diff > old_total);
2301         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2302         unlock_chunks(root);
2303         btrfs_end_transaction(trans, root);
2304 done:
2305         btrfs_free_path(path);
2306         return ret;
2307 }
2308
2309 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2310                            struct btrfs_root *root,
2311                            struct btrfs_key *key,
2312                            struct btrfs_chunk *chunk, int item_size)
2313 {
2314         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2315         struct btrfs_disk_key disk_key;
2316         u32 array_size;
2317         u8 *ptr;
2318
2319         array_size = btrfs_super_sys_array_size(super_copy);
2320         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2321                 return -EFBIG;
2322
2323         ptr = super_copy->sys_chunk_array + array_size;
2324         btrfs_cpu_key_to_disk(&disk_key, key);
2325         memcpy(ptr, &disk_key, sizeof(disk_key));
2326         ptr += sizeof(disk_key);
2327         memcpy(ptr, chunk, item_size);
2328         item_size += sizeof(disk_key);
2329         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2330         return 0;
2331 }
2332
2333 /*
2334  * sort the devices in descending order by max_avail, total_avail
2335  */
2336 static int btrfs_cmp_device_info(const void *a, const void *b)
2337 {
2338         const struct btrfs_device_info *di_a = a;
2339         const struct btrfs_device_info *di_b = b;
2340
2341         if (di_a->max_avail > di_b->max_avail)
2342                 return -1;
2343         if (di_a->max_avail < di_b->max_avail)
2344                 return 1;
2345         if (di_a->total_avail > di_b->total_avail)
2346                 return -1;
2347         if (di_a->total_avail < di_b->total_avail)
2348                 return 1;
2349         return 0;
2350 }
2351
2352 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2353                                struct btrfs_root *extent_root,
2354                                struct map_lookup **map_ret,
2355                                u64 *num_bytes_out, u64 *stripe_size_out,
2356                                u64 start, u64 type)
2357 {
2358         struct btrfs_fs_info *info = extent_root->fs_info;
2359         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2360         struct list_head *cur;
2361         struct map_lookup *map = NULL;
2362         struct extent_map_tree *em_tree;
2363         struct extent_map *em;
2364         struct btrfs_device_info *devices_info = NULL;
2365         u64 total_avail;
2366         int num_stripes;        /* total number of stripes to allocate */
2367         int sub_stripes;        /* sub_stripes info for map */
2368         int dev_stripes;        /* stripes per dev */
2369         int devs_max;           /* max devs to use */
2370         int devs_min;           /* min devs needed */
2371         int devs_increment;     /* ndevs has to be a multiple of this */
2372         int ncopies;            /* how many copies to data has */
2373         int ret;
2374         u64 max_stripe_size;
2375         u64 max_chunk_size;
2376         u64 stripe_size;
2377         u64 num_bytes;
2378         int ndevs;
2379         int i;
2380         int j;
2381
2382         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2383             (type & BTRFS_BLOCK_GROUP_DUP)) {
2384                 WARN_ON(1);
2385                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2386         }
2387
2388         if (list_empty(&fs_devices->alloc_list))
2389                 return -ENOSPC;
2390
2391         sub_stripes = 1;
2392         dev_stripes = 1;
2393         devs_increment = 1;
2394         ncopies = 1;
2395         devs_max = 0;   /* 0 == as many as possible */
2396         devs_min = 1;
2397
2398         /*
2399          * define the properties of each RAID type.
2400          * FIXME: move this to a global table and use it in all RAID
2401          * calculation code
2402          */
2403         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2404                 dev_stripes = 2;
2405                 ncopies = 2;
2406                 devs_max = 1;
2407         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2408                 devs_min = 2;
2409         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2410                 devs_increment = 2;
2411                 ncopies = 2;
2412                 devs_max = 2;
2413                 devs_min = 2;
2414         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2415                 sub_stripes = 2;
2416                 devs_increment = 2;
2417                 ncopies = 2;
2418                 devs_min = 4;
2419         } else {
2420                 devs_max = 1;
2421         }
2422
2423         if (type & BTRFS_BLOCK_GROUP_DATA) {
2424                 max_stripe_size = 1024 * 1024 * 1024;
2425                 max_chunk_size = 10 * max_stripe_size;
2426         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2427                 max_stripe_size = 256 * 1024 * 1024;
2428                 max_chunk_size = max_stripe_size;
2429         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2430                 max_stripe_size = 8 * 1024 * 1024;
2431                 max_chunk_size = 2 * max_stripe_size;
2432         } else {
2433                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2434                        type);
2435                 BUG_ON(1);
2436         }
2437
2438         /* we don't want a chunk larger than 10% of writeable space */
2439         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2440                              max_chunk_size);
2441
2442         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2443                                GFP_NOFS);
2444         if (!devices_info)
2445                 return -ENOMEM;
2446
2447         cur = fs_devices->alloc_list.next;
2448
2449         /*
2450          * in the first pass through the devices list, we gather information
2451          * about the available holes on each device.
2452          */
2453         ndevs = 0;
2454         while (cur != &fs_devices->alloc_list) {
2455                 struct btrfs_device *device;
2456                 u64 max_avail;
2457                 u64 dev_offset;
2458
2459                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2460
2461                 cur = cur->next;
2462
2463                 if (!device->writeable) {
2464                         printk(KERN_ERR
2465                                "btrfs: read-only device in alloc_list\n");
2466                         WARN_ON(1);
2467                         continue;
2468                 }
2469
2470                 if (!device->in_fs_metadata)
2471                         continue;
2472
2473                 if (device->total_bytes > device->bytes_used)
2474                         total_avail = device->total_bytes - device->bytes_used;
2475                 else
2476                         total_avail = 0;
2477
2478                 /* If there is no space on this device, skip it. */
2479                 if (total_avail == 0)
2480                         continue;
2481
2482                 ret = find_free_dev_extent(trans, device,
2483                                            max_stripe_size * dev_stripes,
2484                                            &dev_offset, &max_avail);
2485                 if (ret && ret != -ENOSPC)
2486                         goto error;
2487
2488                 if (ret == 0)
2489                         max_avail = max_stripe_size * dev_stripes;
2490
2491                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2492                         continue;
2493
2494                 devices_info[ndevs].dev_offset = dev_offset;
2495                 devices_info[ndevs].max_avail = max_avail;
2496                 devices_info[ndevs].total_avail = total_avail;
2497                 devices_info[ndevs].dev = device;
2498                 ++ndevs;
2499         }
2500
2501         /*
2502          * now sort the devices by hole size / available space
2503          */
2504         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2505              btrfs_cmp_device_info, NULL);
2506
2507         /* round down to number of usable stripes */
2508         ndevs -= ndevs % devs_increment;
2509
2510         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2511                 ret = -ENOSPC;
2512                 goto error;
2513         }
2514
2515         if (devs_max && ndevs > devs_max)
2516                 ndevs = devs_max;
2517         /*
2518          * the primary goal is to maximize the number of stripes, so use as many
2519          * devices as possible, even if the stripes are not maximum sized.
2520          */
2521         stripe_size = devices_info[ndevs-1].max_avail;
2522         num_stripes = ndevs * dev_stripes;
2523
2524         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2525                 stripe_size = max_chunk_size * ncopies;
2526                 do_div(stripe_size, num_stripes);
2527         }
2528
2529         do_div(stripe_size, dev_stripes);
2530         do_div(stripe_size, BTRFS_STRIPE_LEN);
2531         stripe_size *= BTRFS_STRIPE_LEN;
2532
2533         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2534         if (!map) {
2535                 ret = -ENOMEM;
2536                 goto error;
2537         }
2538         map->num_stripes = num_stripes;
2539
2540         for (i = 0; i < ndevs; ++i) {
2541                 for (j = 0; j < dev_stripes; ++j) {
2542                         int s = i * dev_stripes + j;
2543                         map->stripes[s].dev = devices_info[i].dev;
2544                         map->stripes[s].physical = devices_info[i].dev_offset +
2545                                                    j * stripe_size;
2546                 }
2547         }
2548         map->sector_size = extent_root->sectorsize;
2549         map->stripe_len = BTRFS_STRIPE_LEN;
2550         map->io_align = BTRFS_STRIPE_LEN;
2551         map->io_width = BTRFS_STRIPE_LEN;
2552         map->type = type;
2553         map->sub_stripes = sub_stripes;
2554
2555         *map_ret = map;
2556         num_bytes = stripe_size * (num_stripes / ncopies);
2557
2558         *stripe_size_out = stripe_size;
2559         *num_bytes_out = num_bytes;
2560
2561         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2562
2563         em = alloc_extent_map();
2564         if (!em) {
2565                 ret = -ENOMEM;
2566                 goto error;
2567         }
2568         em->bdev = (struct block_device *)map;
2569         em->start = start;
2570         em->len = num_bytes;
2571         em->block_start = 0;
2572         em->block_len = em->len;
2573
2574         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2575         write_lock(&em_tree->lock);
2576         ret = add_extent_mapping(em_tree, em);
2577         write_unlock(&em_tree->lock);
2578         BUG_ON(ret);
2579         free_extent_map(em);
2580
2581         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2582                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2583                                      start, num_bytes);
2584         BUG_ON(ret);
2585
2586         for (i = 0; i < map->num_stripes; ++i) {
2587                 struct btrfs_device *device;
2588                 u64 dev_offset;
2589
2590                 device = map->stripes[i].dev;
2591                 dev_offset = map->stripes[i].physical;
2592
2593                 ret = btrfs_alloc_dev_extent(trans, device,
2594                                 info->chunk_root->root_key.objectid,
2595                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2596                                 start, dev_offset, stripe_size);
2597                 BUG_ON(ret);
2598         }
2599
2600         kfree(devices_info);
2601         return 0;
2602
2603 error:
2604         kfree(map);
2605         kfree(devices_info);
2606         return ret;
2607 }
2608
2609 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2610                                 struct btrfs_root *extent_root,
2611                                 struct map_lookup *map, u64 chunk_offset,
2612                                 u64 chunk_size, u64 stripe_size)
2613 {
2614         u64 dev_offset;
2615         struct btrfs_key key;
2616         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2617         struct btrfs_device *device;
2618         struct btrfs_chunk *chunk;
2619         struct btrfs_stripe *stripe;
2620         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2621         int index = 0;
2622         int ret;
2623
2624         chunk = kzalloc(item_size, GFP_NOFS);
2625         if (!chunk)
2626                 return -ENOMEM;
2627
2628         index = 0;
2629         while (index < map->num_stripes) {
2630                 device = map->stripes[index].dev;
2631                 device->bytes_used += stripe_size;
2632                 ret = btrfs_update_device(trans, device);
2633                 BUG_ON(ret);
2634                 index++;
2635         }
2636
2637         spin_lock(&extent_root->fs_info->free_chunk_lock);
2638         extent_root->fs_info->free_chunk_space -= (stripe_size *
2639                                                    map->num_stripes);
2640         spin_unlock(&extent_root->fs_info->free_chunk_lock);
2641
2642         index = 0;
2643         stripe = &chunk->stripe;
2644         while (index < map->num_stripes) {
2645                 device = map->stripes[index].dev;
2646                 dev_offset = map->stripes[index].physical;
2647
2648                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2649                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2650                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2651                 stripe++;
2652                 index++;
2653         }
2654
2655         btrfs_set_stack_chunk_length(chunk, chunk_size);
2656         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2657         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2658         btrfs_set_stack_chunk_type(chunk, map->type);
2659         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2660         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2661         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2662         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2663         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2664
2665         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2666         key.type = BTRFS_CHUNK_ITEM_KEY;
2667         key.offset = chunk_offset;
2668
2669         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2670         BUG_ON(ret);
2671
2672         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2673                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2674                                              item_size);
2675                 BUG_ON(ret);
2676         }
2677
2678         kfree(chunk);
2679         return 0;
2680 }
2681
2682 /*
2683  * Chunk allocation falls into two parts. The first part does works
2684  * that make the new allocated chunk useable, but not do any operation
2685  * that modifies the chunk tree. The second part does the works that
2686  * require modifying the chunk tree. This division is important for the
2687  * bootstrap process of adding storage to a seed btrfs.
2688  */
2689 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2690                       struct btrfs_root *extent_root, u64 type)
2691 {
2692         u64 chunk_offset;
2693         u64 chunk_size;
2694         u64 stripe_size;
2695         struct map_lookup *map;
2696         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2697         int ret;
2698
2699         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2700                               &chunk_offset);
2701         if (ret)
2702                 return ret;
2703
2704         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2705                                   &stripe_size, chunk_offset, type);
2706         if (ret)
2707                 return ret;
2708
2709         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2710                                    chunk_size, stripe_size);
2711         BUG_ON(ret);
2712         return 0;
2713 }
2714
2715 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2716                                          struct btrfs_root *root,
2717                                          struct btrfs_device *device)
2718 {
2719         u64 chunk_offset;
2720         u64 sys_chunk_offset;
2721         u64 chunk_size;
2722         u64 sys_chunk_size;
2723         u64 stripe_size;
2724         u64 sys_stripe_size;
2725         u64 alloc_profile;
2726         struct map_lookup *map;
2727         struct map_lookup *sys_map;
2728         struct btrfs_fs_info *fs_info = root->fs_info;
2729         struct btrfs_root *extent_root = fs_info->extent_root;
2730         int ret;
2731
2732         ret = find_next_chunk(fs_info->chunk_root,
2733                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2734         if (ret)
2735                 return ret;
2736
2737         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2738                         (fs_info->metadata_alloc_profile &
2739                          fs_info->avail_metadata_alloc_bits);
2740         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2741
2742         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2743                                   &stripe_size, chunk_offset, alloc_profile);
2744         BUG_ON(ret);
2745
2746         sys_chunk_offset = chunk_offset + chunk_size;
2747
2748         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2749                         (fs_info->system_alloc_profile &
2750                          fs_info->avail_system_alloc_bits);
2751         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2752
2753         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2754                                   &sys_chunk_size, &sys_stripe_size,
2755                                   sys_chunk_offset, alloc_profile);
2756         BUG_ON(ret);
2757
2758         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2759         BUG_ON(ret);
2760
2761         /*
2762          * Modifying chunk tree needs allocating new blocks from both
2763          * system block group and metadata block group. So we only can
2764          * do operations require modifying the chunk tree after both
2765          * block groups were created.
2766          */
2767         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2768                                    chunk_size, stripe_size);
2769         BUG_ON(ret);
2770
2771         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2772                                    sys_chunk_offset, sys_chunk_size,
2773                                    sys_stripe_size);
2774         BUG_ON(ret);
2775         return 0;
2776 }
2777
2778 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2779 {
2780         struct extent_map *em;
2781         struct map_lookup *map;
2782         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2783         int readonly = 0;
2784         int i;
2785
2786         read_lock(&map_tree->map_tree.lock);
2787         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2788         read_unlock(&map_tree->map_tree.lock);
2789         if (!em)
2790                 return 1;
2791
2792         if (btrfs_test_opt(root, DEGRADED)) {
2793                 free_extent_map(em);
2794                 return 0;
2795         }
2796
2797         map = (struct map_lookup *)em->bdev;
2798         for (i = 0; i < map->num_stripes; i++) {
2799                 if (!map->stripes[i].dev->writeable) {
2800                         readonly = 1;
2801                         break;
2802                 }
2803         }
2804         free_extent_map(em);
2805         return readonly;
2806 }
2807
2808 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2809 {
2810         extent_map_tree_init(&tree->map_tree);
2811 }
2812
2813 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2814 {
2815         struct extent_map *em;
2816
2817         while (1) {
2818                 write_lock(&tree->map_tree.lock);
2819                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2820                 if (em)
2821                         remove_extent_mapping(&tree->map_tree, em);
2822                 write_unlock(&tree->map_tree.lock);
2823                 if (!em)
2824                         break;
2825                 kfree(em->bdev);
2826                 /* once for us */
2827                 free_extent_map(em);
2828                 /* once for the tree */
2829                 free_extent_map(em);
2830         }
2831 }
2832
2833 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2834 {
2835         struct extent_map *em;
2836         struct map_lookup *map;
2837         struct extent_map_tree *em_tree = &map_tree->map_tree;
2838         int ret;
2839
2840         read_lock(&em_tree->lock);
2841         em = lookup_extent_mapping(em_tree, logical, len);
2842         read_unlock(&em_tree->lock);
2843         BUG_ON(!em);
2844
2845         BUG_ON(em->start > logical || em->start + em->len < logical);
2846         map = (struct map_lookup *)em->bdev;
2847         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2848                 ret = map->num_stripes;
2849         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2850                 ret = map->sub_stripes;
2851         else
2852                 ret = 1;
2853         free_extent_map(em);
2854         return ret;
2855 }
2856
2857 static int find_live_mirror(struct map_lookup *map, int first, int num,
2858                             int optimal)
2859 {
2860         int i;
2861         if (map->stripes[optimal].dev->bdev)
2862                 return optimal;
2863         for (i = first; i < first + num; i++) {
2864                 if (map->stripes[i].dev->bdev)
2865                         return i;
2866         }
2867         /* we couldn't find one that doesn't fail.  Just return something
2868          * and the io error handling code will clean up eventually
2869          */
2870         return optimal;
2871 }
2872
2873 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2874                              u64 logical, u64 *length,
2875                              struct btrfs_multi_bio **multi_ret,
2876                              int mirror_num)
2877 {
2878         struct extent_map *em;
2879         struct map_lookup *map;
2880         struct extent_map_tree *em_tree = &map_tree->map_tree;
2881         u64 offset;
2882         u64 stripe_offset;
2883         u64 stripe_end_offset;
2884         u64 stripe_nr;
2885         u64 stripe_nr_orig;
2886         u64 stripe_nr_end;
2887         int stripes_allocated = 8;
2888         int stripes_required = 1;
2889         int stripe_index;
2890         int i;
2891         int num_stripes;
2892         int max_errors = 0;
2893         struct btrfs_multi_bio *multi = NULL;
2894
2895         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2896                 stripes_allocated = 1;
2897 again:
2898         if (multi_ret) {
2899                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2900                                 GFP_NOFS);
2901                 if (!multi)
2902                         return -ENOMEM;
2903
2904                 atomic_set(&multi->error, 0);
2905         }
2906
2907         read_lock(&em_tree->lock);
2908         em = lookup_extent_mapping(em_tree, logical, *length);
2909         read_unlock(&em_tree->lock);
2910
2911         if (!em) {
2912                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2913                        (unsigned long long)logical,
2914                        (unsigned long long)*length);
2915                 BUG();
2916         }
2917
2918         BUG_ON(em->start > logical || em->start + em->len < logical);
2919         map = (struct map_lookup *)em->bdev;
2920         offset = logical - em->start;
2921
2922         if (mirror_num > map->num_stripes)
2923                 mirror_num = 0;
2924
2925         /* if our multi bio struct is too small, back off and try again */
2926         if (rw & REQ_WRITE) {
2927                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2928                                  BTRFS_BLOCK_GROUP_DUP)) {
2929                         stripes_required = map->num_stripes;
2930                         max_errors = 1;
2931                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2932                         stripes_required = map->sub_stripes;
2933                         max_errors = 1;
2934                 }
2935         }
2936         if (rw & REQ_DISCARD) {
2937                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2938                                  BTRFS_BLOCK_GROUP_RAID1 |
2939                                  BTRFS_BLOCK_GROUP_DUP |
2940                                  BTRFS_BLOCK_GROUP_RAID10)) {
2941                         stripes_required = map->num_stripes;
2942                 }
2943         }
2944         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2945             stripes_allocated < stripes_required) {
2946                 stripes_allocated = map->num_stripes;
2947                 free_extent_map(em);
2948                 kfree(multi);
2949                 goto again;
2950         }
2951         stripe_nr = offset;
2952         /*
2953          * stripe_nr counts the total number of stripes we have to stride
2954          * to get to this block
2955          */
2956         do_div(stripe_nr, map->stripe_len);
2957
2958         stripe_offset = stripe_nr * map->stripe_len;
2959         BUG_ON(offset < stripe_offset);
2960
2961         /* stripe_offset is the offset of this block in its stripe*/
2962         stripe_offset = offset - stripe_offset;
2963
2964         if (rw & REQ_DISCARD)
2965                 *length = min_t(u64, em->len - offset, *length);
2966         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2967                               BTRFS_BLOCK_GROUP_RAID1 |
2968                               BTRFS_BLOCK_GROUP_RAID10 |
2969                               BTRFS_BLOCK_GROUP_DUP)) {
2970                 /* we limit the length of each bio to what fits in a stripe */
2971                 *length = min_t(u64, em->len - offset,
2972                                 map->stripe_len - stripe_offset);
2973         } else {
2974                 *length = em->len - offset;
2975         }
2976
2977         if (!multi_ret)
2978                 goto out;
2979
2980         num_stripes = 1;
2981         stripe_index = 0;
2982         stripe_nr_orig = stripe_nr;
2983         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2984                         (~(map->stripe_len - 1));
2985         do_div(stripe_nr_end, map->stripe_len);
2986         stripe_end_offset = stripe_nr_end * map->stripe_len -
2987                             (offset + *length);
2988         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2989                 if (rw & REQ_DISCARD)
2990                         num_stripes = min_t(u64, map->num_stripes,
2991                                             stripe_nr_end - stripe_nr_orig);
2992                 stripe_index = do_div(stripe_nr, map->num_stripes);
2993         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2994                 if (rw & (REQ_WRITE | REQ_DISCARD))
2995                         num_stripes = map->num_stripes;
2996                 else if (mirror_num)
2997                         stripe_index = mirror_num - 1;
2998                 else {
2999                         stripe_index = find_live_mirror(map, 0,
3000                                             map->num_stripes,
3001                                             current->pid % map->num_stripes);
3002                 }
3003
3004         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3005                 if (rw & (REQ_WRITE | REQ_DISCARD))
3006                         num_stripes = map->num_stripes;
3007                 else if (mirror_num)
3008                         stripe_index = mirror_num - 1;
3009
3010         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3011                 int factor = map->num_stripes / map->sub_stripes;
3012
3013                 stripe_index = do_div(stripe_nr, factor);
3014                 stripe_index *= map->sub_stripes;
3015
3016                 if (rw & REQ_WRITE)
3017                         num_stripes = map->sub_stripes;
3018                 else if (rw & REQ_DISCARD)
3019                         num_stripes = min_t(u64, map->sub_stripes *
3020                                             (stripe_nr_end - stripe_nr_orig),
3021                                             map->num_stripes);
3022                 else if (mirror_num)
3023                         stripe_index += mirror_num - 1;
3024                 else {
3025                         stripe_index = find_live_mirror(map, stripe_index,
3026                                               map->sub_stripes, stripe_index +
3027                                               current->pid % map->sub_stripes);
3028                 }
3029         } else {
3030                 /*
3031                  * after this do_div call, stripe_nr is the number of stripes
3032                  * on this device we have to walk to find the data, and
3033                  * stripe_index is the number of our device in the stripe array
3034                  */
3035                 stripe_index = do_div(stripe_nr, map->num_stripes);
3036         }
3037         BUG_ON(stripe_index >= map->num_stripes);
3038
3039         if (rw & REQ_DISCARD) {
3040                 for (i = 0; i < num_stripes; i++) {
3041                         multi->stripes[i].physical =
3042                                 map->stripes[stripe_index].physical +
3043                                 stripe_offset + stripe_nr * map->stripe_len;
3044                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3045
3046                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3047                                 u64 stripes;
3048                                 u32 last_stripe = 0;
3049                                 int j;
3050
3051                                 div_u64_rem(stripe_nr_end - 1,
3052                                             map->num_stripes,
3053                                             &last_stripe);
3054
3055                                 for (j = 0; j < map->num_stripes; j++) {
3056                                         u32 test;
3057
3058                                         div_u64_rem(stripe_nr_end - 1 - j,
3059                                                     map->num_stripes, &test);
3060                                         if (test == stripe_index)
3061                                                 break;
3062                                 }
3063                                 stripes = stripe_nr_end - 1 - j;
3064                                 do_div(stripes, map->num_stripes);
3065                                 multi->stripes[i].length = map->stripe_len *
3066                                         (stripes - stripe_nr + 1);
3067
3068                                 if (i == 0) {
3069                                         multi->stripes[i].length -=
3070                                                 stripe_offset;
3071                                         stripe_offset = 0;
3072                                 }
3073                                 if (stripe_index == last_stripe)
3074                                         multi->stripes[i].length -=
3075                                                 stripe_end_offset;
3076                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3077                                 u64 stripes;
3078                                 int j;
3079                                 int factor = map->num_stripes /
3080                                              map->sub_stripes;
3081                                 u32 last_stripe = 0;
3082
3083                                 div_u64_rem(stripe_nr_end - 1,
3084                                             factor, &last_stripe);
3085                                 last_stripe *= map->sub_stripes;
3086
3087                                 for (j = 0; j < factor; j++) {
3088                                         u32 test;
3089
3090                                         div_u64_rem(stripe_nr_end - 1 - j,
3091                                                     factor, &test);
3092
3093                                         if (test ==
3094                                             stripe_index / map->sub_stripes)
3095                                                 break;
3096                                 }
3097                                 stripes = stripe_nr_end - 1 - j;
3098                                 do_div(stripes, factor);
3099                                 multi->stripes[i].length = map->stripe_len *
3100                                         (stripes - stripe_nr + 1);
3101
3102                                 if (i < map->sub_stripes) {
3103                                         multi->stripes[i].length -=
3104                                                 stripe_offset;
3105                                         if (i == map->sub_stripes - 1)
3106                                                 stripe_offset = 0;
3107                                 }
3108                                 if (stripe_index >= last_stripe &&
3109                                     stripe_index <= (last_stripe +
3110                                                      map->sub_stripes - 1)) {
3111                                         multi->stripes[i].length -=
3112                                                 stripe_end_offset;
3113                                 }
3114                         } else
3115                                 multi->stripes[i].length = *length;
3116
3117                         stripe_index++;
3118                         if (stripe_index == map->num_stripes) {
3119                                 /* This could only happen for RAID0/10 */
3120                                 stripe_index = 0;
3121                                 stripe_nr++;
3122                         }
3123                 }
3124         } else {
3125                 for (i = 0; i < num_stripes; i++) {
3126                         multi->stripes[i].physical =
3127                                 map->stripes[stripe_index].physical +
3128                                 stripe_offset +
3129                                 stripe_nr * map->stripe_len;
3130                         multi->stripes[i].dev =
3131                                 map->stripes[stripe_index].dev;
3132                         stripe_index++;
3133                 }
3134         }
3135         if (multi_ret) {
3136                 *multi_ret = multi;
3137                 multi->num_stripes = num_stripes;
3138                 multi->max_errors = max_errors;
3139         }
3140 out:
3141         free_extent_map(em);
3142         return 0;
3143 }
3144
3145 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3146                       u64 logical, u64 *length,
3147                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3148 {
3149         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3150                                  mirror_num);
3151 }
3152
3153 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3154                      u64 chunk_start, u64 physical, u64 devid,
3155                      u64 **logical, int *naddrs, int *stripe_len)
3156 {
3157         struct extent_map_tree *em_tree = &map_tree->map_tree;
3158         struct extent_map *em;
3159         struct map_lookup *map;
3160         u64 *buf;
3161         u64 bytenr;
3162         u64 length;
3163         u64 stripe_nr;
3164         int i, j, nr = 0;
3165
3166         read_lock(&em_tree->lock);
3167         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3168         read_unlock(&em_tree->lock);
3169
3170         BUG_ON(!em || em->start != chunk_start);
3171         map = (struct map_lookup *)em->bdev;
3172
3173         length = em->len;
3174         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3175                 do_div(length, map->num_stripes / map->sub_stripes);
3176         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3177                 do_div(length, map->num_stripes);
3178
3179         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3180         BUG_ON(!buf);
3181
3182         for (i = 0; i < map->num_stripes; i++) {
3183                 if (devid && map->stripes[i].dev->devid != devid)
3184                         continue;
3185                 if (map->stripes[i].physical > physical ||
3186                     map->stripes[i].physical + length <= physical)
3187                         continue;
3188
3189                 stripe_nr = physical - map->stripes[i].physical;
3190                 do_div(stripe_nr, map->stripe_len);
3191
3192                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3193                         stripe_nr = stripe_nr * map->num_stripes + i;
3194                         do_div(stripe_nr, map->sub_stripes);
3195                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3196                         stripe_nr = stripe_nr * map->num_stripes + i;
3197                 }
3198                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3199                 WARN_ON(nr >= map->num_stripes);
3200                 for (j = 0; j < nr; j++) {
3201                         if (buf[j] == bytenr)
3202                                 break;
3203                 }
3204                 if (j == nr) {
3205                         WARN_ON(nr >= map->num_stripes);
3206                         buf[nr++] = bytenr;
3207                 }
3208         }
3209
3210         *logical = buf;
3211         *naddrs = nr;
3212         *stripe_len = map->stripe_len;
3213
3214         free_extent_map(em);
3215         return 0;
3216 }
3217
3218 static void end_bio_multi_stripe(struct bio *bio, int err)
3219 {
3220         struct btrfs_multi_bio *multi = bio->bi_private;
3221         int is_orig_bio = 0;
3222
3223         if (err)
3224                 atomic_inc(&multi->error);
3225
3226         if (bio == multi->orig_bio)
3227                 is_orig_bio = 1;
3228
3229         if (atomic_dec_and_test(&multi->stripes_pending)) {
3230                 if (!is_orig_bio) {
3231                         bio_put(bio);
3232                         bio = multi->orig_bio;
3233                 }
3234                 bio->bi_private = multi->private;
3235                 bio->bi_end_io = multi->end_io;
3236                 /* only send an error to the higher layers if it is
3237                  * beyond the tolerance of the multi-bio
3238                  */
3239                 if (atomic_read(&multi->error) > multi->max_errors) {
3240                         err = -EIO;
3241                 } else if (err) {
3242                         /*
3243                          * this bio is actually up to date, we didn't
3244                          * go over the max number of errors
3245                          */
3246                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3247                         err = 0;
3248                 }
3249                 kfree(multi);
3250
3251                 bio_endio(bio, err);
3252         } else if (!is_orig_bio) {
3253                 bio_put(bio);
3254         }
3255 }
3256
3257 struct async_sched {
3258         struct bio *bio;
3259         int rw;
3260         struct btrfs_fs_info *info;
3261         struct btrfs_work work;
3262 };
3263
3264 /*
3265  * see run_scheduled_bios for a description of why bios are collected for
3266  * async submit.
3267  *
3268  * This will add one bio to the pending list for a device and make sure
3269  * the work struct is scheduled.
3270  */
3271 static noinline int schedule_bio(struct btrfs_root *root,
3272                                  struct btrfs_device *device,
3273                                  int rw, struct bio *bio)
3274 {
3275         int should_queue = 1;
3276         struct btrfs_pending_bios *pending_bios;
3277
3278         /* don't bother with additional async steps for reads, right now */
3279         if (!(rw & REQ_WRITE)) {
3280                 bio_get(bio);
3281                 submit_bio(rw, bio);
3282                 bio_put(bio);
3283                 return 0;
3284         }
3285
3286         /*
3287          * nr_async_bios allows us to reliably return congestion to the
3288          * higher layers.  Otherwise, the async bio makes it appear we have
3289          * made progress against dirty pages when we've really just put it
3290          * on a queue for later
3291          */
3292         atomic_inc(&root->fs_info->nr_async_bios);
3293         WARN_ON(bio->bi_next);
3294         bio->bi_next = NULL;
3295         bio->bi_rw |= rw;
3296
3297         spin_lock(&device->io_lock);
3298         if (bio->bi_rw & REQ_SYNC)
3299                 pending_bios = &device->pending_sync_bios;
3300         else
3301                 pending_bios = &device->pending_bios;
3302
3303         if (pending_bios->tail)
3304                 pending_bios->tail->bi_next = bio;
3305
3306         pending_bios->tail = bio;
3307         if (!pending_bios->head)
3308                 pending_bios->head = bio;
3309         if (device->running_pending)
3310                 should_queue = 0;
3311
3312         spin_unlock(&device->io_lock);
3313
3314         if (should_queue)
3315                 btrfs_queue_worker(&root->fs_info->submit_workers,
3316                                    &device->work);
3317         return 0;
3318 }
3319
3320 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3321                   int mirror_num, int async_submit)
3322 {
3323         struct btrfs_mapping_tree *map_tree;
3324         struct btrfs_device *dev;
3325         struct bio *first_bio = bio;
3326         u64 logical = (u64)bio->bi_sector << 9;
3327         u64 length = 0;
3328         u64 map_length;
3329         struct btrfs_multi_bio *multi = NULL;
3330         int ret;
3331         int dev_nr = 0;
3332         int total_devs = 1;
3333
3334         length = bio->bi_size;
3335         map_tree = &root->fs_info->mapping_tree;
3336         map_length = length;
3337
3338         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3339                               mirror_num);
3340         BUG_ON(ret);
3341
3342         total_devs = multi->num_stripes;
3343         if (map_length < length) {
3344                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3345                        "len %llu\n", (unsigned long long)logical,
3346                        (unsigned long long)length,
3347                        (unsigned long long)map_length);
3348                 BUG();
3349         }
3350         multi->end_io = first_bio->bi_end_io;
3351         multi->private = first_bio->bi_private;
3352         multi->orig_bio = first_bio;
3353         atomic_set(&multi->stripes_pending, multi->num_stripes);
3354
3355         while (dev_nr < total_devs) {
3356                 if (total_devs > 1) {
3357                         if (dev_nr < total_devs - 1) {
3358                                 bio = bio_clone(first_bio, GFP_NOFS);
3359                                 BUG_ON(!bio);
3360                         } else {
3361                                 bio = first_bio;
3362                         }
3363                         bio->bi_private = multi;
3364                         bio->bi_end_io = end_bio_multi_stripe;
3365                 }
3366                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3367                 dev = multi->stripes[dev_nr].dev;
3368                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3369                         bio->bi_bdev = dev->bdev;
3370                         if (async_submit)
3371                                 schedule_bio(root, dev, rw, bio);
3372                         else
3373                                 submit_bio(rw, bio);
3374                 } else {
3375                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3376                         bio->bi_sector = logical >> 9;
3377                         bio_endio(bio, -EIO);
3378                 }
3379                 dev_nr++;
3380         }
3381         if (total_devs == 1)
3382                 kfree(multi);
3383         return 0;
3384 }
3385
3386 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3387                                        u8 *uuid, u8 *fsid)
3388 {
3389         struct btrfs_device *device;
3390         struct btrfs_fs_devices *cur_devices;
3391
3392         cur_devices = root->fs_info->fs_devices;
3393         while (cur_devices) {
3394                 if (!fsid ||
3395                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3396                         device = __find_device(&cur_devices->devices,
3397                                                devid, uuid);
3398                         if (device)
3399                                 return device;
3400                 }
3401                 cur_devices = cur_devices->seed;
3402         }
3403         return NULL;
3404 }
3405
3406 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3407                                             u64 devid, u8 *dev_uuid)
3408 {
3409         struct btrfs_device *device;
3410         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3411
3412         device = kzalloc(sizeof(*device), GFP_NOFS);
3413         if (!device)
3414                 return NULL;
3415         list_add(&device->dev_list,
3416                  &fs_devices->devices);
3417         device->dev_root = root->fs_info->dev_root;
3418         device->devid = devid;
3419         device->work.func = pending_bios_fn;
3420         device->fs_devices = fs_devices;
3421         device->missing = 1;
3422         fs_devices->num_devices++;
3423         fs_devices->missing_devices++;
3424         spin_lock_init(&device->io_lock);
3425         INIT_LIST_HEAD(&device->dev_alloc_list);
3426         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3427         return device;
3428 }
3429
3430 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3431                           struct extent_buffer *leaf,
3432                           struct btrfs_chunk *chunk)
3433 {
3434         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3435         struct map_lookup *map;
3436         struct extent_map *em;
3437         u64 logical;
3438         u64 length;
3439         u64 devid;
3440         u8 uuid[BTRFS_UUID_SIZE];
3441         int num_stripes;
3442         int ret;
3443         int i;
3444
3445         logical = key->offset;
3446         length = btrfs_chunk_length(leaf, chunk);
3447
3448         read_lock(&map_tree->map_tree.lock);
3449         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3450         read_unlock(&map_tree->map_tree.lock);
3451
3452         /* already mapped? */
3453         if (em && em->start <= logical && em->start + em->len > logical) {
3454                 free_extent_map(em);
3455                 return 0;
3456         } else if (em) {
3457                 free_extent_map(em);
3458         }
3459
3460         em = alloc_extent_map();
3461         if (!em)
3462                 return -ENOMEM;
3463         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3464         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3465         if (!map) {
3466                 free_extent_map(em);
3467                 return -ENOMEM;
3468         }
3469
3470         em->bdev = (struct block_device *)map;
3471         em->start = logical;
3472         em->len = length;
3473         em->block_start = 0;
3474         em->block_len = em->len;
3475
3476         map->num_stripes = num_stripes;
3477         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3478         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3479         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3480         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3481         map->type = btrfs_chunk_type(leaf, chunk);
3482         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3483         for (i = 0; i < num_stripes; i++) {
3484                 map->stripes[i].physical =
3485                         btrfs_stripe_offset_nr(leaf, chunk, i);
3486                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3487                 read_extent_buffer(leaf, uuid, (unsigned long)
3488                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3489                                    BTRFS_UUID_SIZE);
3490                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3491                                                         NULL);
3492                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3493                         kfree(map);
3494                         free_extent_map(em);
3495                         return -EIO;
3496                 }
3497                 if (!map->stripes[i].dev) {
3498                         map->stripes[i].dev =
3499                                 add_missing_dev(root, devid, uuid);
3500                         if (!map->stripes[i].dev) {
3501                                 kfree(map);
3502                                 free_extent_map(em);
3503                                 return -EIO;
3504                         }
3505                 }
3506                 map->stripes[i].dev->in_fs_metadata = 1;
3507         }
3508
3509         write_lock(&map_tree->map_tree.lock);
3510         ret = add_extent_mapping(&map_tree->map_tree, em);
3511         write_unlock(&map_tree->map_tree.lock);
3512         BUG_ON(ret);
3513         free_extent_map(em);
3514
3515         return 0;
3516 }
3517
3518 static int fill_device_from_item(struct extent_buffer *leaf,
3519                                  struct btrfs_dev_item *dev_item,
3520                                  struct btrfs_device *device)
3521 {
3522         unsigned long ptr;
3523
3524         device->devid = btrfs_device_id(leaf, dev_item);
3525         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3526         device->total_bytes = device->disk_total_bytes;
3527         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3528         device->type = btrfs_device_type(leaf, dev_item);
3529         device->io_align = btrfs_device_io_align(leaf, dev_item);
3530         device->io_width = btrfs_device_io_width(leaf, dev_item);
3531         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3532
3533         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3534         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3535
3536         return 0;
3537 }
3538
3539 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3540 {
3541         struct btrfs_fs_devices *fs_devices;
3542         int ret;
3543
3544         mutex_lock(&uuid_mutex);
3545
3546         fs_devices = root->fs_info->fs_devices->seed;
3547         while (fs_devices) {
3548                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3549                         ret = 0;
3550                         goto out;
3551                 }
3552                 fs_devices = fs_devices->seed;
3553         }
3554
3555         fs_devices = find_fsid(fsid);
3556         if (!fs_devices) {
3557                 ret = -ENOENT;
3558                 goto out;
3559         }
3560
3561         fs_devices = clone_fs_devices(fs_devices);
3562         if (IS_ERR(fs_devices)) {
3563                 ret = PTR_ERR(fs_devices);
3564                 goto out;
3565         }
3566
3567         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3568                                    root->fs_info->bdev_holder);
3569         if (ret)
3570                 goto out;
3571
3572         if (!fs_devices->seeding) {
3573                 __btrfs_close_devices(fs_devices);
3574                 free_fs_devices(fs_devices);
3575                 ret = -EINVAL;
3576                 goto out;
3577         }
3578
3579         fs_devices->seed = root->fs_info->fs_devices->seed;
3580         root->fs_info->fs_devices->seed = fs_devices;
3581 out:
3582         mutex_unlock(&uuid_mutex);
3583         return ret;
3584 }
3585
3586 static int read_one_dev(struct btrfs_root *root,
3587                         struct extent_buffer *leaf,
3588                         struct btrfs_dev_item *dev_item)
3589 {
3590         struct btrfs_device *device;
3591         u64 devid;
3592         int ret;
3593         u8 fs_uuid[BTRFS_UUID_SIZE];
3594         u8 dev_uuid[BTRFS_UUID_SIZE];
3595
3596         devid = btrfs_device_id(leaf, dev_item);
3597         read_extent_buffer(leaf, dev_uuid,
3598                            (unsigned long)btrfs_device_uuid(dev_item),
3599                            BTRFS_UUID_SIZE);
3600         read_extent_buffer(leaf, fs_uuid,
3601                            (unsigned long)btrfs_device_fsid(dev_item),
3602                            BTRFS_UUID_SIZE);
3603
3604         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3605                 ret = open_seed_devices(root, fs_uuid);
3606                 if (ret && !btrfs_test_opt(root, DEGRADED))
3607                         return ret;
3608         }
3609
3610         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3611         if (!device || !device->bdev) {
3612                 if (!btrfs_test_opt(root, DEGRADED))
3613                         return -EIO;
3614
3615                 if (!device) {
3616                         printk(KERN_WARNING "warning devid %llu missing\n",
3617                                (unsigned long long)devid);
3618                         device = add_missing_dev(root, devid, dev_uuid);
3619                         if (!device)
3620                                 return -ENOMEM;
3621                 } else if (!device->missing) {
3622                         /*
3623                          * this happens when a device that was properly setup
3624                          * in the device info lists suddenly goes bad.
3625                          * device->bdev is NULL, and so we have to set
3626                          * device->missing to one here
3627                          */
3628                         root->fs_info->fs_devices->missing_devices++;
3629                         device->missing = 1;
3630                 }
3631         }
3632
3633         if (device->fs_devices != root->fs_info->fs_devices) {
3634                 BUG_ON(device->writeable);
3635                 if (device->generation !=
3636                     btrfs_device_generation(leaf, dev_item))
3637                         return -EINVAL;
3638         }
3639
3640         fill_device_from_item(leaf, dev_item, device);
3641         device->dev_root = root->fs_info->dev_root;
3642         device->in_fs_metadata = 1;
3643         if (device->writeable) {
3644                 device->fs_devices->total_rw_bytes += device->total_bytes;
3645                 spin_lock(&root->fs_info->free_chunk_lock);
3646                 root->fs_info->free_chunk_space += device->total_bytes -
3647                         device->bytes_used;
3648                 spin_unlock(&root->fs_info->free_chunk_lock);
3649         }
3650         ret = 0;
3651         return ret;
3652 }
3653
3654 int btrfs_read_sys_array(struct btrfs_root *root)
3655 {
3656         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3657         struct extent_buffer *sb;
3658         struct btrfs_disk_key *disk_key;
3659         struct btrfs_chunk *chunk;
3660         u8 *ptr;
3661         unsigned long sb_ptr;
3662         int ret = 0;
3663         u32 num_stripes;
3664         u32 array_size;
3665         u32 len = 0;
3666         u32 cur;
3667         struct btrfs_key key;
3668
3669         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3670                                           BTRFS_SUPER_INFO_SIZE);
3671         if (!sb)
3672                 return -ENOMEM;
3673         btrfs_set_buffer_uptodate(sb);
3674         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
3675
3676         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3677         array_size = btrfs_super_sys_array_size(super_copy);
3678
3679         ptr = super_copy->sys_chunk_array;
3680         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3681         cur = 0;
3682
3683         while (cur < array_size) {
3684                 disk_key = (struct btrfs_disk_key *)ptr;
3685                 btrfs_disk_key_to_cpu(&key, disk_key);
3686
3687                 len = sizeof(*disk_key); ptr += len;
3688                 sb_ptr += len;
3689                 cur += len;
3690
3691                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3692                         chunk = (struct btrfs_chunk *)sb_ptr;
3693                         ret = read_one_chunk(root, &key, sb, chunk);
3694                         if (ret)
3695                                 break;
3696                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3697                         len = btrfs_chunk_item_size(num_stripes);
3698                 } else {
3699                         ret = -EIO;
3700                         break;
3701                 }
3702                 ptr += len;
3703                 sb_ptr += len;
3704                 cur += len;
3705         }
3706         free_extent_buffer(sb);
3707         return ret;
3708 }
3709
3710 int btrfs_read_chunk_tree(struct btrfs_root *root)
3711 {
3712         struct btrfs_path *path;
3713         struct extent_buffer *leaf;
3714         struct btrfs_key key;
3715         struct btrfs_key found_key;
3716         int ret;
3717         int slot;
3718
3719         root = root->fs_info->chunk_root;
3720
3721         path = btrfs_alloc_path();
3722         if (!path)
3723                 return -ENOMEM;
3724
3725         /* first we search for all of the device items, and then we
3726          * read in all of the chunk items.  This way we can create chunk
3727          * mappings that reference all of the devices that are afound
3728          */
3729         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3730         key.offset = 0;
3731         key.type = 0;
3732 again:
3733         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3734         if (ret < 0)
3735                 goto error;
3736         while (1) {
3737                 leaf = path->nodes[0];
3738                 slot = path->slots[0];
3739                 if (slot >= btrfs_header_nritems(leaf)) {
3740                         ret = btrfs_next_leaf(root, path);
3741                         if (ret == 0)
3742                                 continue;
3743                         if (ret < 0)
3744                                 goto error;
3745                         break;
3746                 }
3747                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3748                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3749                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3750                                 break;
3751                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3752                                 struct btrfs_dev_item *dev_item;
3753                                 dev_item = btrfs_item_ptr(leaf, slot,
3754                                                   struct btrfs_dev_item);
3755                                 ret = read_one_dev(root, leaf, dev_item);
3756                                 if (ret)
3757                                         goto error;
3758                         }
3759                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3760                         struct btrfs_chunk *chunk;
3761                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3762                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3763                         if (ret)
3764                                 goto error;
3765                 }
3766                 path->slots[0]++;
3767         }
3768         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3769                 key.objectid = 0;
3770                 btrfs_release_path(path);
3771                 goto again;
3772         }
3773         ret = 0;
3774 error:
3775         btrfs_free_path(path);
3776         return ret;
3777 }