fs/btrfs/volumes.c

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