fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include "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 #include "rcu-string.h"
  38 #include "math.h"
  39
  40 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  41                                 struct btrfs_root *root,
  42                                 struct btrfs_device *device);
  43 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  44 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  45 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  46
  47 static DEFINE_MUTEX(uuid_mutex);
  48 static LIST_HEAD(fs_uuids);
  49
  50 static void lock_chunks(struct btrfs_root *root)
  51 {
  52         mutex_lock(&root->fs_info->chunk_mutex);
  53 }
  54
  55 static void unlock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_unlock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  61 {
  62         struct btrfs_device *device;
  63         WARN_ON(fs_devices->opened);
  64         while (!list_empty(&fs_devices->devices)) {
  65                 device = list_entry(fs_devices->devices.next,
  66                                     struct btrfs_device, dev_list);
  67                 list_del(&device->dev_list);
  68                 rcu_string_free(device->name);
  69                 kfree(device);
  70         }
  71         kfree(fs_devices);
  72 }
  73
  74 void btrfs_cleanup_fs_uuids(void)
  75 {
  76         struct btrfs_fs_devices *fs_devices;
  77
  78         while (!list_empty(&fs_uuids)) {
  79                 fs_devices = list_entry(fs_uuids.next,
  80                                         struct btrfs_fs_devices, list);
  81                 list_del(&fs_devices->list);
  82                 free_fs_devices(fs_devices);
  83         }
  84 }
  85
  86 static noinline struct btrfs_device *__find_device(struct list_head *head,
  87                                                    u64 devid, u8 *uuid)
  88 {
  89         struct btrfs_device *dev;
  90
  91         list_for_each_entry(dev, head, dev_list) {
  92                 if (dev->devid == devid &&
  93                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  94                         return dev;
  95                 }
  96         }
  97         return NULL;
  98 }
  99
 100 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 101 {
 102         struct btrfs_fs_devices *fs_devices;
 103
 104         list_for_each_entry(fs_devices, &fs_uuids, list) {
 105                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 106                         return fs_devices;
 107         }
 108         return NULL;
 109 }
 110
 111 static int
 112 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 113                       int flush, struct block_device **bdev,
 114                       struct buffer_head **bh)
 115 {
 116         int ret;
 117
 118         *bdev = blkdev_get_by_path(device_path, flags, holder);
 119
 120         if (IS_ERR(*bdev)) {
 121                 ret = PTR_ERR(*bdev);
 122                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
 123                 goto error;
 124         }
 125
 126         if (flush)
 127                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 128         ret = set_blocksize(*bdev, 4096);
 129         if (ret) {
 130                 blkdev_put(*bdev, flags);
 131                 goto error;
 132         }
 133         invalidate_bdev(*bdev);
 134         *bh = btrfs_read_dev_super(*bdev);
 135         if (!*bh) {
 136                 ret = -EINVAL;
 137                 blkdev_put(*bdev, flags);
 138                 goto error;
 139         }
 140
 141         return 0;
 142
 143 error:
 144         *bdev = NULL;
 145         *bh = NULL;
 146         return ret;
 147 }
 148
 149 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 150                         struct bio *head, struct bio *tail)
 151 {
 152
 153         struct bio *old_head;
 154
 155         old_head = pending_bios->head;
 156         pending_bios->head = head;
 157         if (pending_bios->tail)
 158                 tail->bi_next = old_head;
 159         else
 160                 pending_bios->tail = tail;
 161 }
 162
 163 /*
 164  * we try to collect pending bios for a device so we don't get a large
 165  * number of procs sending bios down to the same device.  This greatly
 166  * improves the schedulers ability to collect and merge the bios.
 167  *
 168  * But, it also turns into a long list of bios to process and that is sure
 169  * to eventually make the worker thread block.  The solution here is to
 170  * make some progress and then put this work struct back at the end of
 171  * the list if the block device is congested.  This way, multiple devices
 172  * can make progress from a single worker thread.
 173  */
 174 static noinline void run_scheduled_bios(struct btrfs_device *device)
 175 {
 176         struct bio *pending;
 177         struct backing_dev_info *bdi;
 178         struct btrfs_fs_info *fs_info;
 179         struct btrfs_pending_bios *pending_bios;
 180         struct bio *tail;
 181         struct bio *cur;
 182         int again = 0;
 183         unsigned long num_run;
 184         unsigned long batch_run = 0;
 185         unsigned long limit;
 186         unsigned long last_waited = 0;
 187         int force_reg = 0;
 188         int sync_pending = 0;
 189         struct blk_plug plug;
 190
 191         /*
 192          * this function runs all the bios we've collected for
 193          * a particular device.  We don't want to wander off to
 194          * another device without first sending all of these down.
 195          * So, setup a plug here and finish it off before we return
 196          */
 197         blk_start_plug(&plug);
 198
 199         bdi = blk_get_backing_dev_info(device->bdev);
 200         fs_info = device->dev_root->fs_info;
 201         limit = btrfs_async_submit_limit(fs_info);
 202         limit = limit * 2 / 3;
 203
 204 loop:
 205         spin_lock(&device->io_lock);
 206
 207 loop_lock:
 208         num_run = 0;
 209
 210         /* take all the bios off the list at once and process them
 211          * later on (without the lock held).  But, remember the
 212          * tail and other pointers so the bios can be properly reinserted
 213          * into the list if we hit congestion
 214          */
 215         if (!force_reg && device->pending_sync_bios.head) {
 216                 pending_bios = &device->pending_sync_bios;
 217                 force_reg = 1;
 218         } else {
 219                 pending_bios = &device->pending_bios;
 220                 force_reg = 0;
 221         }
 222
 223         pending = pending_bios->head;
 224         tail = pending_bios->tail;
 225         WARN_ON(pending && !tail);
 226
 227         /*
 228          * if pending was null this time around, no bios need processing
 229          * at all and we can stop.  Otherwise it'll loop back up again
 230          * and do an additional check so no bios are missed.
 231          *
 232          * device->running_pending is used to synchronize with the
 233          * schedule_bio code.
 234          */
 235         if (device->pending_sync_bios.head == NULL &&
 236             device->pending_bios.head == NULL) {
 237                 again = 0;
 238                 device->running_pending = 0;
 239         } else {
 240                 again = 1;
 241                 device->running_pending = 1;
 242         }
 243
 244         pending_bios->head = NULL;
 245         pending_bios->tail = NULL;
 246
 247         spin_unlock(&device->io_lock);
 248
 249         while (pending) {
 250
 251                 rmb();
 252                 /* we want to work on both lists, but do more bios on the
 253                  * sync list than the regular list
 254                  */
 255                 if ((num_run > 32 &&
 256                     pending_bios != &device->pending_sync_bios &&
 257                     device->pending_sync_bios.head) ||
 258                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 259                     device->pending_bios.head)) {
 260                         spin_lock(&device->io_lock);
 261                         requeue_list(pending_bios, pending, tail);
 262                         goto loop_lock;
 263                 }
 264
 265                 cur = pending;
 266                 pending = pending->bi_next;
 267                 cur->bi_next = NULL;
 268
 269                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 270                     waitqueue_active(&fs_info->async_submit_wait))
 271                         wake_up(&fs_info->async_submit_wait);
 272
 273                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 274
 275                 /*
 276                  * if we're doing the sync list, record that our
 277                  * plug has some sync requests on it
 278                  *
 279                  * If we're doing the regular list and there are
 280                  * sync requests sitting around, unplug before
 281                  * we add more
 282                  */
 283                 if (pending_bios == &device->pending_sync_bios) {
 284                         sync_pending = 1;
 285                 } else if (sync_pending) {
 286                         blk_finish_plug(&plug);
 287                         blk_start_plug(&plug);
 288                         sync_pending = 0;
 289                 }
 290
 291                 btrfsic_submit_bio(cur->bi_rw, cur);
 292                 num_run++;
 293                 batch_run++;
 294                 if (need_resched())
 295                         cond_resched();
 296
 297                 /*
 298                  * we made progress, there is more work to do and the bdi
 299                  * is now congested.  Back off and let other work structs
 300                  * run instead
 301                  */
 302                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 303                     fs_info->fs_devices->open_devices > 1) {
 304                         struct io_context *ioc;
 305
 306                         ioc = current->io_context;
 307
 308                         /*
 309                          * the main goal here is that we don't want to
 310                          * block if we're going to be able to submit
 311                          * more requests without blocking.
 312                          *
 313                          * This code does two great things, it pokes into
 314                          * the elevator code from a filesystem _and_
 315                          * it makes assumptions about how batching works.
 316                          */
 317                         if (ioc && ioc->nr_batch_requests > 0 &&
 318                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 319                             (last_waited == 0 ||
 320                              ioc->last_waited == last_waited)) {
 321                                 /*
 322                                  * we want to go through our batch of
 323                                  * requests and stop.  So, we copy out
 324                                  * the ioc->last_waited time and test
 325                                  * against it before looping
 326                                  */
 327                                 last_waited = ioc->last_waited;
 328                                 if (need_resched())
 329                                         cond_resched();
 330                                 continue;
 331                         }
 332                         spin_lock(&device->io_lock);
 333                         requeue_list(pending_bios, pending, tail);
 334                         device->running_pending = 1;
 335
 336                         spin_unlock(&device->io_lock);
 337                         btrfs_requeue_work(&device->work);
 338                         goto done;
 339                 }
 340                 /* unplug every 64 requests just for good measure */
 341                 if (batch_run % 64 == 0) {
 342                         blk_finish_plug(&plug);
 343                         blk_start_plug(&plug);
 344                         sync_pending = 0;
 345                 }
 346         }
 347
 348         cond_resched();
 349         if (again)
 350                 goto loop;
 351
 352         spin_lock(&device->io_lock);
 353         if (device->pending_bios.head || device->pending_sync_bios.head)
 354                 goto loop_lock;
 355         spin_unlock(&device->io_lock);
 356
 357 done:
 358         blk_finish_plug(&plug);
 359 }
 360
 361 static void pending_bios_fn(struct btrfs_work *work)
 362 {
 363         struct btrfs_device *device;
 364
 365         device = container_of(work, struct btrfs_device, work);
 366         run_scheduled_bios(device);
 367 }
 368
 369 static noinline int device_list_add(const char *path,
 370                            struct btrfs_super_block *disk_super,
 371                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 372 {
 373         struct btrfs_device *device;
 374         struct btrfs_fs_devices *fs_devices;
 375         struct rcu_string *name;
 376         u64 found_transid = btrfs_super_generation(disk_super);
 377
 378         fs_devices = find_fsid(disk_super->fsid);
 379         if (!fs_devices) {
 380                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 381                 if (!fs_devices)
 382                         return -ENOMEM;
 383                 INIT_LIST_HEAD(&fs_devices->devices);
 384                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 385                 list_add(&fs_devices->list, &fs_uuids);
 386                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 387                 fs_devices->latest_devid = devid;
 388                 fs_devices->latest_trans = found_transid;
 389                 mutex_init(&fs_devices->device_list_mutex);
 390                 device = NULL;
 391         } else {
 392                 device = __find_device(&fs_devices->devices, devid,
 393                                        disk_super->dev_item.uuid);
 394         }
 395         if (!device) {
 396                 if (fs_devices->opened)
 397                         return -EBUSY;
 398
 399                 device = kzalloc(sizeof(*device), GFP_NOFS);
 400                 if (!device) {
 401                         /* we can safely leave the fs_devices entry around */
 402                         return -ENOMEM;
 403                 }
 404                 device->devid = devid;
 405                 device->dev_stats_valid = 0;
 406                 device->work.func = pending_bios_fn;
 407                 memcpy(device->uuid, disk_super->dev_item.uuid,
 408                        BTRFS_UUID_SIZE);
 409                 spin_lock_init(&device->io_lock);
 410
 411                 name = rcu_string_strdup(path, GFP_NOFS);
 412                 if (!name) {
 413                         kfree(device);
 414                         return -ENOMEM;
 415                 }
 416                 rcu_assign_pointer(device->name, name);
 417                 INIT_LIST_HEAD(&device->dev_alloc_list);
 418
 419                 /* init readahead state */
 420                 spin_lock_init(&device->reada_lock);
 421                 device->reada_curr_zone = NULL;
 422                 atomic_set(&device->reada_in_flight, 0);
 423                 device->reada_next = 0;
 424                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 425                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 426
 427                 mutex_lock(&fs_devices->device_list_mutex);
 428                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 429                 mutex_unlock(&fs_devices->device_list_mutex);
 430
 431                 device->fs_devices = fs_devices;
 432                 fs_devices->num_devices++;
 433         } else if (!device->name || strcmp(device->name->str, path)) {
 434                 name = rcu_string_strdup(path, GFP_NOFS);
 435                 if (!name)
 436                         return -ENOMEM;
 437                 rcu_string_free(device->name);
 438                 rcu_assign_pointer(device->name, name);
 439                 if (device->missing) {
 440                         fs_devices->missing_devices--;
 441                         device->missing = 0;
 442                 }
 443         }
 444
 445         if (found_transid > fs_devices->latest_trans) {
 446                 fs_devices->latest_devid = devid;
 447                 fs_devices->latest_trans = found_transid;
 448         }
 449         *fs_devices_ret = fs_devices;
 450         return 0;
 451 }
 452
 453 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 454 {
 455         struct btrfs_fs_devices *fs_devices;
 456         struct btrfs_device *device;
 457         struct btrfs_device *orig_dev;
 458
 459         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 460         if (!fs_devices)
 461                 return ERR_PTR(-ENOMEM);
 462
 463         INIT_LIST_HEAD(&fs_devices->devices);
 464         INIT_LIST_HEAD(&fs_devices->alloc_list);
 465         INIT_LIST_HEAD(&fs_devices->list);
 466         mutex_init(&fs_devices->device_list_mutex);
 467         fs_devices->latest_devid = orig->latest_devid;
 468         fs_devices->latest_trans = orig->latest_trans;
 469         fs_devices->total_devices = orig->total_devices;
 470         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 471
 472         /* We have held the volume lock, it is safe to get the devices. */
 473         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 474                 struct rcu_string *name;
 475
 476                 device = kzalloc(sizeof(*device), GFP_NOFS);
 477                 if (!device)
 478                         goto error;
 479
 480                 /*
 481                  * This is ok to do without rcu read locked because we hold the
 482                  * uuid mutex so nothing we touch in here is going to disappear.
 483                  */
 484                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 485                 if (!name) {
 486                         kfree(device);
 487                         goto error;
 488                 }
 489                 rcu_assign_pointer(device->name, name);
 490
 491                 device->devid = orig_dev->devid;
 492                 device->work.func = pending_bios_fn;
 493                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 494                 spin_lock_init(&device->io_lock);
 495                 INIT_LIST_HEAD(&device->dev_list);
 496                 INIT_LIST_HEAD(&device->dev_alloc_list);
 497
 498                 list_add(&device->dev_list, &fs_devices->devices);
 499                 device->fs_devices = fs_devices;
 500                 fs_devices->num_devices++;
 501         }
 502         return fs_devices;
 503 error:
 504         free_fs_devices(fs_devices);
 505         return ERR_PTR(-ENOMEM);
 506 }
 507
 508 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 509 {
 510         struct btrfs_device *device, *next;
 511
 512         struct block_device *latest_bdev = NULL;
 513         u64 latest_devid = 0;
 514         u64 latest_transid = 0;
 515
 516         mutex_lock(&uuid_mutex);
 517 again:
 518         /* This is the initialized path, it is safe to release the devices. */
 519         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 520                 if (device->in_fs_metadata) {
 521                         if (!latest_transid ||
 522                             device->generation > latest_transid) {
 523                                 latest_devid = device->devid;
 524                                 latest_transid = device->generation;
 525                                 latest_bdev = device->bdev;
 526                         }
 527                         continue;
 528                 }
 529
 530                 if (device->bdev) {
 531                         blkdev_put(device->bdev, device->mode);
 532                         device->bdev = NULL;
 533                         fs_devices->open_devices--;
 534                 }
 535                 if (device->writeable) {
 536                         list_del_init(&device->dev_alloc_list);
 537                         device->writeable = 0;
 538                         fs_devices->rw_devices--;
 539                 }
 540                 list_del_init(&device->dev_list);
 541                 fs_devices->num_devices--;
 542                 rcu_string_free(device->name);
 543                 kfree(device);
 544         }
 545
 546         if (fs_devices->seed) {
 547                 fs_devices = fs_devices->seed;
 548                 goto again;
 549         }
 550
 551         fs_devices->latest_bdev = latest_bdev;
 552         fs_devices->latest_devid = latest_devid;
 553         fs_devices->latest_trans = latest_transid;
 554
 555         mutex_unlock(&uuid_mutex);
 556 }
 557
 558 static void __free_device(struct work_struct *work)
 559 {
 560         struct btrfs_device *device;
 561
 562         device = container_of(work, struct btrfs_device, rcu_work);
 563
 564         if (device->bdev)
 565                 blkdev_put(device->bdev, device->mode);
 566
 567         rcu_string_free(device->name);
 568         kfree(device);
 569 }
 570
 571 static void free_device(struct rcu_head *head)
 572 {
 573         struct btrfs_device *device;
 574
 575         device = container_of(head, struct btrfs_device, rcu);
 576
 577         INIT_WORK(&device->rcu_work, __free_device);
 578         schedule_work(&device->rcu_work);
 579 }
 580
 581 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 582 {
 583         struct btrfs_device *device;
 584
 585         if (--fs_devices->opened > 0)
 586                 return 0;
 587
 588         mutex_lock(&fs_devices->device_list_mutex);
 589         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 590                 struct btrfs_device *new_device;
 591                 struct rcu_string *name;
 592
 593                 if (device->bdev)
 594                         fs_devices->open_devices--;
 595
 596                 if (device->writeable) {
 597                         list_del_init(&device->dev_alloc_list);
 598                         fs_devices->rw_devices--;
 599                 }
 600
 601                 if (device->can_discard)
 602                         fs_devices->num_can_discard--;
 603
 604                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 605                 BUG_ON(!new_device); /* -ENOMEM */
 606                 memcpy(new_device, device, sizeof(*new_device));
 607
 608                 /* Safe because we are under uuid_mutex */
 609                 if (device->name) {
 610                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 611                         BUG_ON(device->name && !name); /* -ENOMEM */
 612                         rcu_assign_pointer(new_device->name, name);
 613                 }
 614                 new_device->bdev = NULL;
 615                 new_device->writeable = 0;
 616                 new_device->in_fs_metadata = 0;
 617                 new_device->can_discard = 0;
 618                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 619
 620                 call_rcu(&device->rcu, free_device);
 621         }
 622         mutex_unlock(&fs_devices->device_list_mutex);
 623
 624         WARN_ON(fs_devices->open_devices);
 625         WARN_ON(fs_devices->rw_devices);
 626         fs_devices->opened = 0;
 627         fs_devices->seeding = 0;
 628
 629         return 0;
 630 }
 631
 632 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 633 {
 634         struct btrfs_fs_devices *seed_devices = NULL;
 635         int ret;
 636
 637         mutex_lock(&uuid_mutex);
 638         ret = __btrfs_close_devices(fs_devices);
 639         if (!fs_devices->opened) {
 640                 seed_devices = fs_devices->seed;
 641                 fs_devices->seed = NULL;
 642         }
 643         mutex_unlock(&uuid_mutex);
 644
 645         while (seed_devices) {
 646                 fs_devices = seed_devices;
 647                 seed_devices = fs_devices->seed;
 648                 __btrfs_close_devices(fs_devices);
 649                 free_fs_devices(fs_devices);
 650         }
 651         return ret;
 652 }
 653
 654 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 655                                 fmode_t flags, void *holder)
 656 {
 657         struct request_queue *q;
 658         struct block_device *bdev;
 659         struct list_head *head = &fs_devices->devices;
 660         struct btrfs_device *device;
 661         struct block_device *latest_bdev = NULL;
 662         struct buffer_head *bh;
 663         struct btrfs_super_block *disk_super;
 664         u64 latest_devid = 0;
 665         u64 latest_transid = 0;
 666         u64 devid;
 667         int seeding = 1;
 668         int ret = 0;
 669
 670         flags |= FMODE_EXCL;
 671
 672         list_for_each_entry(device, head, dev_list) {
 673                 if (device->bdev)
 674                         continue;
 675                 if (!device->name)
 676                         continue;
 677
 678                 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 679                                             &bdev, &bh);
 680                 if (ret)
 681                         continue;
 682
 683                 disk_super = (struct btrfs_super_block *)bh->b_data;
 684                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 685                 if (devid != device->devid)
 686                         goto error_brelse;
 687
 688                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 689                            BTRFS_UUID_SIZE))
 690                         goto error_brelse;
 691
 692                 device->generation = btrfs_super_generation(disk_super);
 693                 if (!latest_transid || device->generation > latest_transid) {
 694                         latest_devid = devid;
 695                         latest_transid = device->generation;
 696                         latest_bdev = bdev;
 697                 }
 698
 699                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 700                         device->writeable = 0;
 701                 } else {
 702                         device->writeable = !bdev_read_only(bdev);
 703                         seeding = 0;
 704                 }
 705
 706                 q = bdev_get_queue(bdev);
 707                 if (blk_queue_discard(q)) {
 708                         device->can_discard = 1;
 709                         fs_devices->num_can_discard++;
 710                 }
 711
 712                 device->bdev = bdev;
 713                 device->in_fs_metadata = 0;
 714                 device->mode = flags;
 715
 716                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 717                         fs_devices->rotating = 1;
 718
 719                 fs_devices->open_devices++;
 720                 if (device->writeable) {
 721                         fs_devices->rw_devices++;
 722                         list_add(&device->dev_alloc_list,
 723                                  &fs_devices->alloc_list);
 724                 }
 725                 brelse(bh);
 726                 continue;
 727
 728 error_brelse:
 729                 brelse(bh);
 730                 blkdev_put(bdev, flags);
 731                 continue;
 732         }
 733         if (fs_devices->open_devices == 0) {
 734                 ret = -EINVAL;
 735                 goto out;
 736         }
 737         fs_devices->seeding = seeding;
 738         fs_devices->opened = 1;
 739         fs_devices->latest_bdev = latest_bdev;
 740         fs_devices->latest_devid = latest_devid;
 741         fs_devices->latest_trans = latest_transid;
 742         fs_devices->total_rw_bytes = 0;
 743 out:
 744         return ret;
 745 }
 746
 747 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 748                        fmode_t flags, void *holder)
 749 {
 750         int ret;
 751
 752         mutex_lock(&uuid_mutex);
 753         if (fs_devices->opened) {
 754                 fs_devices->opened++;
 755                 ret = 0;
 756         } else {
 757                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 758         }
 759         mutex_unlock(&uuid_mutex);
 760         return ret;
 761 }
 762
 763 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 764                           struct btrfs_fs_devices **fs_devices_ret)
 765 {
 766         struct btrfs_super_block *disk_super;
 767         struct block_device *bdev;
 768         struct buffer_head *bh;
 769         int ret;
 770         u64 devid;
 771         u64 transid;
 772         u64 total_devices;
 773
 774         flags |= FMODE_EXCL;
 775         mutex_lock(&uuid_mutex);
 776         ret = btrfs_get_bdev_and_sb(path, flags, holder, 0, &bdev, &bh);
 777         if (ret)
 778                 goto error;
 779         disk_super = (struct btrfs_super_block *)bh->b_data;
 780         devid = btrfs_stack_device_id(&disk_super->dev_item);
 781         transid = btrfs_super_generation(disk_super);
 782         total_devices = btrfs_super_num_devices(disk_super);
 783         if (disk_super->label[0]) {
 784                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 785                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 786                 printk(KERN_INFO "device label %s ", disk_super->label);
 787         } else {
 788                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 789         }
 790         printk(KERN_CONT "devid %llu transid %llu %s\n",
 791                (unsigned long long)devid, (unsigned long long)transid, path);
 792         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 793         if (!ret && fs_devices_ret)
 794                 (*fs_devices_ret)->total_devices = total_devices;
 795         brelse(bh);
 796         blkdev_put(bdev, flags);
 797 error:
 798         mutex_unlock(&uuid_mutex);
 799         return ret;
 800 }
 801
 802 /* helper to account the used device space in the range */
 803 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 804                                    u64 end, u64 *length)
 805 {
 806         struct btrfs_key key;
 807         struct btrfs_root *root = device->dev_root;
 808         struct btrfs_dev_extent *dev_extent;
 809         struct btrfs_path *path;
 810         u64 extent_end;
 811         int ret;
 812         int slot;
 813         struct extent_buffer *l;
 814
 815         *length = 0;
 816
 817         if (start >= device->total_bytes)
 818                 return 0;
 819
 820         path = btrfs_alloc_path();
 821         if (!path)
 822                 return -ENOMEM;
 823         path->reada = 2;
 824
 825         key.objectid = device->devid;
 826         key.offset = start;
 827         key.type = BTRFS_DEV_EXTENT_KEY;
 828
 829         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 830         if (ret < 0)
 831                 goto out;
 832         if (ret > 0) {
 833                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 834                 if (ret < 0)
 835                         goto out;
 836         }
 837
 838         while (1) {
 839                 l = path->nodes[0];
 840                 slot = path->slots[0];
 841                 if (slot >= btrfs_header_nritems(l)) {
 842                         ret = btrfs_next_leaf(root, path);
 843                         if (ret == 0)
 844                                 continue;
 845                         if (ret < 0)
 846                                 goto out;
 847
 848                         break;
 849                 }
 850                 btrfs_item_key_to_cpu(l, &key, slot);
 851
 852                 if (key.objectid < device->devid)
 853                         goto next;
 854
 855                 if (key.objectid > device->devid)
 856                         break;
 857
 858                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 859                         goto next;
 860
 861                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 862                 extent_end = key.offset + btrfs_dev_extent_length(l,
 863                                                                   dev_extent);
 864                 if (key.offset <= start && extent_end > end) {
 865                         *length = end - start + 1;
 866                         break;
 867                 } else if (key.offset <= start && extent_end > start)
 868                         *length += extent_end - start;
 869                 else if (key.offset > start && extent_end <= end)
 870                         *length += extent_end - key.offset;
 871                 else if (key.offset > start && key.offset <= end) {
 872                         *length += end - key.offset + 1;
 873                         break;
 874                 } else if (key.offset > end)
 875                         break;
 876
 877 next:
 878                 path->slots[0]++;
 879         }
 880         ret = 0;
 881 out:
 882         btrfs_free_path(path);
 883         return ret;
 884 }
 885
 886 /*
 887  * find_free_dev_extent - find free space in the specified device
 888  * @device:     the device which we search the free space in
 889  * @num_bytes:  the size of the free space that we need
 890  * @start:      store the start of the free space.
 891  * @len:        the size of the free space. that we find, or the size of the max
 892  *              free space if we don't find suitable free space
 893  *
 894  * this uses a pretty simple search, the expectation is that it is
 895  * called very infrequently and that a given device has a small number
 896  * of extents
 897  *
 898  * @start is used to store the start of the free space if we find. But if we
 899  * don't find suitable free space, it will be used to store the start position
 900  * of the max free space.
 901  *
 902  * @len is used to store the size of the free space that we find.
 903  * But if we don't find suitable free space, it is used to store the size of
 904  * the max free space.
 905  */
 906 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
 907                          u64 *start, u64 *len)
 908 {
 909         struct btrfs_key key;
 910         struct btrfs_root *root = device->dev_root;
 911         struct btrfs_dev_extent *dev_extent;
 912         struct btrfs_path *path;
 913         u64 hole_size;
 914         u64 max_hole_start;
 915         u64 max_hole_size;
 916         u64 extent_end;
 917         u64 search_start;
 918         u64 search_end = device->total_bytes;
 919         int ret;
 920         int slot;
 921         struct extent_buffer *l;
 922
 923         /* FIXME use last free of some kind */
 924
 925         /* we don't want to overwrite the superblock on the drive,
 926          * so we make sure to start at an offset of at least 1MB
 927          */
 928         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 929
 930         max_hole_start = search_start;
 931         max_hole_size = 0;
 932         hole_size = 0;
 933
 934         if (search_start >= search_end) {
 935                 ret = -ENOSPC;
 936                 goto error;
 937         }
 938
 939         path = btrfs_alloc_path();
 940         if (!path) {
 941                 ret = -ENOMEM;
 942                 goto error;
 943         }
 944         path->reada = 2;
 945
 946         key.objectid = device->devid;
 947         key.offset = search_start;
 948         key.type = BTRFS_DEV_EXTENT_KEY;
 949
 950         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 951         if (ret < 0)
 952                 goto out;
 953         if (ret > 0) {
 954                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 955                 if (ret < 0)
 956                         goto out;
 957         }
 958
 959         while (1) {
 960                 l = path->nodes[0];
 961                 slot = path->slots[0];
 962                 if (slot >= btrfs_header_nritems(l)) {
 963                         ret = btrfs_next_leaf(root, path);
 964                         if (ret == 0)
 965                                 continue;
 966                         if (ret < 0)
 967                                 goto out;
 968
 969                         break;
 970                 }
 971                 btrfs_item_key_to_cpu(l, &key, slot);
 972
 973                 if (key.objectid < device->devid)
 974                         goto next;
 975
 976                 if (key.objectid > device->devid)
 977                         break;
 978
 979                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 980                         goto next;
 981
 982                 if (key.offset > search_start) {
 983                         hole_size = key.offset - search_start;
 984
 985                         if (hole_size > max_hole_size) {
 986                                 max_hole_start = search_start;
 987                                 max_hole_size = hole_size;
 988                         }
 989
 990                         /*
 991                          * If this free space is greater than which we need,
 992                          * it must be the max free space that we have found
 993                          * until now, so max_hole_start must point to the start
 994                          * of this free space and the length of this free space
 995                          * is stored in max_hole_size. Thus, we return
 996                          * max_hole_start and max_hole_size and go back to the
 997                          * caller.
 998                          */
 999                         if (hole_size >= num_bytes) {
1000                                 ret = 0;
1001                                 goto out;
1002                         }
1003                 }
1004
1005                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1006                 extent_end = key.offset + btrfs_dev_extent_length(l,
1007                                                                   dev_extent);
1008                 if (extent_end > search_start)
1009                         search_start = extent_end;
1010 next:
1011                 path->slots[0]++;
1012                 cond_resched();
1013         }
1014
1015         /*
1016          * At this point, search_start should be the end of
1017          * allocated dev extents, and when shrinking the device,
1018          * search_end may be smaller than search_start.
1019          */
1020         if (search_end > search_start)
1021                 hole_size = search_end - search_start;
1022
1023         if (hole_size > max_hole_size) {
1024                 max_hole_start = search_start;
1025                 max_hole_size = hole_size;
1026         }
1027
1028         /* See above. */
1029         if (hole_size < num_bytes)
1030                 ret = -ENOSPC;
1031         else
1032                 ret = 0;
1033
1034 out:
1035         btrfs_free_path(path);
1036 error:
1037         *start = max_hole_start;
1038         if (len)
1039                 *len = max_hole_size;
1040         return ret;
1041 }
1042
1043 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1044                           struct btrfs_device *device,
1045                           u64 start)
1046 {
1047         int ret;
1048         struct btrfs_path *path;
1049         struct btrfs_root *root = device->dev_root;
1050         struct btrfs_key key;
1051         struct btrfs_key found_key;
1052         struct extent_buffer *leaf = NULL;
1053         struct btrfs_dev_extent *extent = NULL;
1054
1055         path = btrfs_alloc_path();
1056         if (!path)
1057                 return -ENOMEM;
1058
1059         key.objectid = device->devid;
1060         key.offset = start;
1061         key.type = BTRFS_DEV_EXTENT_KEY;
1062 again:
1063         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1064         if (ret > 0) {
1065                 ret = btrfs_previous_item(root, path, key.objectid,
1066                                           BTRFS_DEV_EXTENT_KEY);
1067                 if (ret)
1068                         goto out;
1069                 leaf = path->nodes[0];
1070                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1071                 extent = btrfs_item_ptr(leaf, path->slots[0],
1072                                         struct btrfs_dev_extent);
1073                 BUG_ON(found_key.offset > start || found_key.offset +
1074                        btrfs_dev_extent_length(leaf, extent) < start);
1075                 key = found_key;
1076                 btrfs_release_path(path);
1077                 goto again;
1078         } else if (ret == 0) {
1079                 leaf = path->nodes[0];
1080                 extent = btrfs_item_ptr(leaf, path->slots[0],
1081                                         struct btrfs_dev_extent);
1082         } else {
1083                 btrfs_error(root->fs_info, ret, "Slot search failed");
1084                 goto out;
1085         }
1086
1087         if (device->bytes_used > 0) {
1088                 u64 len = btrfs_dev_extent_length(leaf, extent);
1089                 device->bytes_used -= len;
1090                 spin_lock(&root->fs_info->free_chunk_lock);
1091                 root->fs_info->free_chunk_space += len;
1092                 spin_unlock(&root->fs_info->free_chunk_lock);
1093         }
1094         ret = btrfs_del_item(trans, root, path);
1095         if (ret) {
1096                 btrfs_error(root->fs_info, ret,
1097                             "Failed to remove dev extent item");
1098         }
1099 out:
1100         btrfs_free_path(path);
1101         return ret;
1102 }
1103
1104 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1105                            struct btrfs_device *device,
1106                            u64 chunk_tree, u64 chunk_objectid,
1107                            u64 chunk_offset, u64 start, u64 num_bytes)
1108 {
1109         int ret;
1110         struct btrfs_path *path;
1111         struct btrfs_root *root = device->dev_root;
1112         struct btrfs_dev_extent *extent;
1113         struct extent_buffer *leaf;
1114         struct btrfs_key key;
1115
1116         WARN_ON(!device->in_fs_metadata);
1117         path = btrfs_alloc_path();
1118         if (!path)
1119                 return -ENOMEM;
1120
1121         key.objectid = device->devid;
1122         key.offset = start;
1123         key.type = BTRFS_DEV_EXTENT_KEY;
1124         ret = btrfs_insert_empty_item(trans, root, path, &key,
1125                                       sizeof(*extent));
1126         if (ret)
1127                 goto out;
1128
1129         leaf = path->nodes[0];
1130         extent = btrfs_item_ptr(leaf, path->slots[0],
1131                                 struct btrfs_dev_extent);
1132         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1133         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1134         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1135
1136         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1137                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1138                     BTRFS_UUID_SIZE);
1139
1140         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1141         btrfs_mark_buffer_dirty(leaf);
1142 out:
1143         btrfs_free_path(path);
1144         return ret;
1145 }
1146
1147 static noinline int find_next_chunk(struct btrfs_root *root,
1148                                     u64 objectid, u64 *offset)
1149 {
1150         struct btrfs_path *path;
1151         int ret;
1152         struct btrfs_key key;
1153         struct btrfs_chunk *chunk;
1154         struct btrfs_key found_key;
1155
1156         path = btrfs_alloc_path();
1157         if (!path)
1158                 return -ENOMEM;
1159
1160         key.objectid = objectid;
1161         key.offset = (u64)-1;
1162         key.type = BTRFS_CHUNK_ITEM_KEY;
1163
1164         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1165         if (ret < 0)
1166                 goto error;
1167
1168         BUG_ON(ret == 0); /* Corruption */
1169
1170         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1171         if (ret) {
1172                 *offset = 0;
1173         } else {
1174                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1175                                       path->slots[0]);
1176                 if (found_key.objectid != objectid)
1177                         *offset = 0;
1178                 else {
1179                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1180                                                struct btrfs_chunk);
1181                         *offset = found_key.offset +
1182                                 btrfs_chunk_length(path->nodes[0], chunk);
1183                 }
1184         }
1185         ret = 0;
1186 error:
1187         btrfs_free_path(path);
1188         return ret;
1189 }
1190
1191 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1192 {
1193         int ret;
1194         struct btrfs_key key;
1195         struct btrfs_key found_key;
1196         struct btrfs_path *path;
1197
1198         root = root->fs_info->chunk_root;
1199
1200         path = btrfs_alloc_path();
1201         if (!path)
1202                 return -ENOMEM;
1203
1204         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1205         key.type = BTRFS_DEV_ITEM_KEY;
1206         key.offset = (u64)-1;
1207
1208         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1209         if (ret < 0)
1210                 goto error;
1211
1212         BUG_ON(ret == 0); /* Corruption */
1213
1214         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1215                                   BTRFS_DEV_ITEM_KEY);
1216         if (ret) {
1217                 *objectid = 1;
1218         } else {
1219                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1220                                       path->slots[0]);
1221                 *objectid = found_key.offset + 1;
1222         }
1223         ret = 0;
1224 error:
1225         btrfs_free_path(path);
1226         return ret;
1227 }
1228
1229 /*
1230  * the device information is stored in the chunk root
1231  * the btrfs_device struct should be fully filled in
1232  */
1233 int btrfs_add_device(struct btrfs_trans_handle *trans,
1234                      struct btrfs_root *root,
1235                      struct btrfs_device *device)
1236 {
1237         int ret;
1238         struct btrfs_path *path;
1239         struct btrfs_dev_item *dev_item;
1240         struct extent_buffer *leaf;
1241         struct btrfs_key key;
1242         unsigned long ptr;
1243
1244         root = root->fs_info->chunk_root;
1245
1246         path = btrfs_alloc_path();
1247         if (!path)
1248                 return -ENOMEM;
1249
1250         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1251         key.type = BTRFS_DEV_ITEM_KEY;
1252         key.offset = device->devid;
1253
1254         ret = btrfs_insert_empty_item(trans, root, path, &key,
1255                                       sizeof(*dev_item));
1256         if (ret)
1257                 goto out;
1258
1259         leaf = path->nodes[0];
1260         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1261
1262         btrfs_set_device_id(leaf, dev_item, device->devid);
1263         btrfs_set_device_generation(leaf, dev_item, 0);
1264         btrfs_set_device_type(leaf, dev_item, device->type);
1265         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1266         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1267         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1268         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1269         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1270         btrfs_set_device_group(leaf, dev_item, 0);
1271         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1272         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1273         btrfs_set_device_start_offset(leaf, dev_item, 0);
1274
1275         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1276         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1277         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1278         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1279         btrfs_mark_buffer_dirty(leaf);
1280
1281         ret = 0;
1282 out:
1283         btrfs_free_path(path);
1284         return ret;
1285 }
1286
1287 static int btrfs_rm_dev_item(struct btrfs_root *root,
1288                              struct btrfs_device *device)
1289 {
1290         int ret;
1291         struct btrfs_path *path;
1292         struct btrfs_key key;
1293         struct btrfs_trans_handle *trans;
1294
1295         root = root->fs_info->chunk_root;
1296
1297         path = btrfs_alloc_path();
1298         if (!path)
1299                 return -ENOMEM;
1300
1301         trans = btrfs_start_transaction(root, 0);
1302         if (IS_ERR(trans)) {
1303                 btrfs_free_path(path);
1304                 return PTR_ERR(trans);
1305         }
1306         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1307         key.type = BTRFS_DEV_ITEM_KEY;
1308         key.offset = device->devid;
1309         lock_chunks(root);
1310
1311         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1312         if (ret < 0)
1313                 goto out;
1314
1315         if (ret > 0) {
1316                 ret = -ENOENT;
1317                 goto out;
1318         }
1319
1320         ret = btrfs_del_item(trans, root, path);
1321         if (ret)
1322                 goto out;
1323 out:
1324         btrfs_free_path(path);
1325         unlock_chunks(root);
1326         btrfs_commit_transaction(trans, root);
1327         return ret;
1328 }
1329
1330 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1331 {
1332         struct btrfs_device *device;
1333         struct btrfs_device *next_device;
1334         struct block_device *bdev;
1335         struct buffer_head *bh = NULL;
1336         struct btrfs_super_block *disk_super;
1337         struct btrfs_fs_devices *cur_devices;
1338         u64 all_avail;
1339         u64 devid;
1340         u64 num_devices;
1341         u8 *dev_uuid;
1342         int ret = 0;
1343         bool clear_super = false;
1344
1345         mutex_lock(&uuid_mutex);
1346
1347         all_avail = root->fs_info->avail_data_alloc_bits |
1348                 root->fs_info->avail_system_alloc_bits |
1349                 root->fs_info->avail_metadata_alloc_bits;
1350
1351         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1352             root->fs_info->fs_devices->num_devices <= 4) {
1353                 printk(KERN_ERR "btrfs: unable to go below four devices "
1354                        "on raid10\n");
1355                 ret = -EINVAL;
1356                 goto out;
1357         }
1358
1359         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1360             root->fs_info->fs_devices->num_devices <= 2) {
1361                 printk(KERN_ERR "btrfs: unable to go below two "
1362                        "devices on raid1\n");
1363                 ret = -EINVAL;
1364                 goto out;
1365         }
1366
1367         if (strcmp(device_path, "missing") == 0) {
1368                 struct list_head *devices;
1369                 struct btrfs_device *tmp;
1370
1371                 device = NULL;
1372                 devices = &root->fs_info->fs_devices->devices;
1373                 /*
1374                  * It is safe to read the devices since the volume_mutex
1375                  * is held.
1376                  */
1377                 list_for_each_entry(tmp, devices, dev_list) {
1378                         if (tmp->in_fs_metadata && !tmp->bdev) {
1379                                 device = tmp;
1380                                 break;
1381                         }
1382                 }
1383                 bdev = NULL;
1384                 bh = NULL;
1385                 disk_super = NULL;
1386                 if (!device) {
1387                         printk(KERN_ERR "btrfs: no missing devices found to "
1388                                "remove\n");
1389                         goto out;
1390                 }
1391         } else {
1392                 ret = btrfs_get_bdev_and_sb(device_path,
1393                                             FMODE_READ | FMODE_EXCL,
1394                                             root->fs_info->bdev_holder, 0,
1395                                             &bdev, &bh);
1396                 if (ret)
1397                         goto out;
1398                 disk_super = (struct btrfs_super_block *)bh->b_data;
1399                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1400                 dev_uuid = disk_super->dev_item.uuid;
1401                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1402                                            disk_super->fsid);
1403                 if (!device) {
1404                         ret = -ENOENT;
1405                         goto error_brelse;
1406                 }
1407         }
1408
1409         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1410                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1411                        "device\n");
1412                 ret = -EINVAL;
1413                 goto error_brelse;
1414         }
1415
1416         if (device->writeable) {
1417                 lock_chunks(root);
1418                 list_del_init(&device->dev_alloc_list);
1419                 unlock_chunks(root);
1420                 root->fs_info->fs_devices->rw_devices--;
1421                 clear_super = true;
1422         }
1423
1424         ret = btrfs_shrink_device(device, 0);
1425         if (ret)
1426                 goto error_undo;
1427
1428         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1429         if (ret)
1430                 goto error_undo;
1431
1432         spin_lock(&root->fs_info->free_chunk_lock);
1433         root->fs_info->free_chunk_space = device->total_bytes -
1434                 device->bytes_used;
1435         spin_unlock(&root->fs_info->free_chunk_lock);
1436
1437         device->in_fs_metadata = 0;
1438         btrfs_scrub_cancel_dev(root->fs_info, device);
1439
1440         /*
1441          * the device list mutex makes sure that we don't change
1442          * the device list while someone else is writing out all
1443          * the device supers.
1444          */
1445
1446         cur_devices = device->fs_devices;
1447         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1448         list_del_rcu(&device->dev_list);
1449
1450         device->fs_devices->num_devices--;
1451         device->fs_devices->total_devices--;
1452
1453         if (device->missing)
1454                 root->fs_info->fs_devices->missing_devices--;
1455
1456         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1457                                  struct btrfs_device, dev_list);
1458         if (device->bdev == root->fs_info->sb->s_bdev)
1459                 root->fs_info->sb->s_bdev = next_device->bdev;
1460         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1461                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1462
1463         if (device->bdev)
1464                 device->fs_devices->open_devices--;
1465
1466         call_rcu(&device->rcu, free_device);
1467         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1468
1469         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1470         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1471
1472         if (cur_devices->open_devices == 0) {
1473                 struct btrfs_fs_devices *fs_devices;
1474                 fs_devices = root->fs_info->fs_devices;
1475                 while (fs_devices) {
1476                         if (fs_devices->seed == cur_devices)
1477                                 break;
1478                         fs_devices = fs_devices->seed;
1479                 }
1480                 fs_devices->seed = cur_devices->seed;
1481                 cur_devices->seed = NULL;
1482                 lock_chunks(root);
1483                 __btrfs_close_devices(cur_devices);
1484                 unlock_chunks(root);
1485                 free_fs_devices(cur_devices);
1486         }
1487
1488         root->fs_info->num_tolerated_disk_barrier_failures =
1489                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1490
1491         /*
1492          * at this point, the device is zero sized.  We want to
1493          * remove it from the devices list and zero out the old super
1494          */
1495         if (clear_super && disk_super) {
1496                 /* make sure this device isn't detected as part of
1497                  * the FS anymore
1498                  */
1499                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1500                 set_buffer_dirty(bh);
1501                 sync_dirty_buffer(bh);
1502         }
1503
1504         ret = 0;
1505
1506 error_brelse:
1507         brelse(bh);
1508 error_close:
1509         if (bdev)
1510                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1511 out:
1512         mutex_unlock(&uuid_mutex);
1513         return ret;
1514 error_undo:
1515         if (device->writeable) {
1516                 lock_chunks(root);
1517                 list_add(&device->dev_alloc_list,
1518                          &root->fs_info->fs_devices->alloc_list);
1519                 unlock_chunks(root);
1520                 root->fs_info->fs_devices->rw_devices++;
1521         }
1522         goto error_brelse;
1523 }
1524
1525 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1526                               struct btrfs_device **device)
1527 {
1528         int ret = 0;
1529         struct btrfs_super_block *disk_super;
1530         u64 devid;
1531         u8 *dev_uuid;
1532         struct block_device *bdev;
1533         struct buffer_head *bh;
1534
1535         *device = NULL;
1536         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1537                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1538         if (ret)
1539                 return ret;
1540         disk_super = (struct btrfs_super_block *)bh->b_data;
1541         devid = btrfs_stack_device_id(&disk_super->dev_item);
1542         dev_uuid = disk_super->dev_item.uuid;
1543         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1544                                     disk_super->fsid);
1545         brelse(bh);
1546         if (!*device)
1547                 ret = -ENOENT;
1548         blkdev_put(bdev, FMODE_READ);
1549         return ret;
1550 }
1551
1552 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1553                                          char *device_path,
1554                                          struct btrfs_device **device)
1555 {
1556         *device = NULL;
1557         if (strcmp(device_path, "missing") == 0) {
1558                 struct list_head *devices;
1559                 struct btrfs_device *tmp;
1560
1561                 devices = &root->fs_info->fs_devices->devices;
1562                 /*
1563                  * It is safe to read the devices since the volume_mutex
1564                  * is held by the caller.
1565                  */
1566                 list_for_each_entry(tmp, devices, dev_list) {
1567                         if (tmp->in_fs_metadata && !tmp->bdev) {
1568                                 *device = tmp;
1569                                 break;
1570                         }
1571                 }
1572
1573                 if (!*device) {
1574                         pr_err("btrfs: no missing device found\n");
1575                         return -ENOENT;
1576                 }
1577
1578                 return 0;
1579         } else {
1580                 return btrfs_find_device_by_path(root, device_path, device);
1581         }
1582 }
1583
1584 /*
1585  * does all the dirty work required for changing file system's UUID.
1586  */
1587 static int btrfs_prepare_sprout(struct btrfs_root *root)
1588 {
1589         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1590         struct btrfs_fs_devices *old_devices;
1591         struct btrfs_fs_devices *seed_devices;
1592         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1593         struct btrfs_device *device;
1594         u64 super_flags;
1595
1596         BUG_ON(!mutex_is_locked(&uuid_mutex));
1597         if (!fs_devices->seeding)
1598                 return -EINVAL;
1599
1600         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1601         if (!seed_devices)
1602                 return -ENOMEM;
1603
1604         old_devices = clone_fs_devices(fs_devices);
1605         if (IS_ERR(old_devices)) {
1606                 kfree(seed_devices);
1607                 return PTR_ERR(old_devices);
1608         }
1609
1610         list_add(&old_devices->list, &fs_uuids);
1611
1612         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1613         seed_devices->opened = 1;
1614         INIT_LIST_HEAD(&seed_devices->devices);
1615         INIT_LIST_HEAD(&seed_devices->alloc_list);
1616         mutex_init(&seed_devices->device_list_mutex);
1617
1618         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1619         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1620                               synchronize_rcu);
1621         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1622
1623         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1624         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1625                 device->fs_devices = seed_devices;
1626         }
1627
1628         fs_devices->seeding = 0;
1629         fs_devices->num_devices = 0;
1630         fs_devices->open_devices = 0;
1631         fs_devices->total_devices = 0;
1632         fs_devices->seed = seed_devices;
1633
1634         generate_random_uuid(fs_devices->fsid);
1635         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1636         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1637         super_flags = btrfs_super_flags(disk_super) &
1638                       ~BTRFS_SUPER_FLAG_SEEDING;
1639         btrfs_set_super_flags(disk_super, super_flags);
1640
1641         return 0;
1642 }
1643
1644 /*
1645  * strore the expected generation for seed devices in device items.
1646  */
1647 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1648                                struct btrfs_root *root)
1649 {
1650         struct btrfs_path *path;
1651         struct extent_buffer *leaf;
1652         struct btrfs_dev_item *dev_item;
1653         struct btrfs_device *device;
1654         struct btrfs_key key;
1655         u8 fs_uuid[BTRFS_UUID_SIZE];
1656         u8 dev_uuid[BTRFS_UUID_SIZE];
1657         u64 devid;
1658         int ret;
1659
1660         path = btrfs_alloc_path();
1661         if (!path)
1662                 return -ENOMEM;
1663
1664         root = root->fs_info->chunk_root;
1665         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1666         key.offset = 0;
1667         key.type = BTRFS_DEV_ITEM_KEY;
1668
1669         while (1) {
1670                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1671                 if (ret < 0)
1672                         goto error;
1673
1674                 leaf = path->nodes[0];
1675 next_slot:
1676                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1677                         ret = btrfs_next_leaf(root, path);
1678                         if (ret > 0)
1679                                 break;
1680                         if (ret < 0)
1681                                 goto error;
1682                         leaf = path->nodes[0];
1683                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1684                         btrfs_release_path(path);
1685                         continue;
1686                 }
1687
1688                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1689                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1690                     key.type != BTRFS_DEV_ITEM_KEY)
1691                         break;
1692
1693                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1694                                           struct btrfs_dev_item);
1695                 devid = btrfs_device_id(leaf, dev_item);
1696                 read_extent_buffer(leaf, dev_uuid,
1697                                    (unsigned long)btrfs_device_uuid(dev_item),
1698                                    BTRFS_UUID_SIZE);
1699                 read_extent_buffer(leaf, fs_uuid,
1700                                    (unsigned long)btrfs_device_fsid(dev_item),
1701                                    BTRFS_UUID_SIZE);
1702                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1703                                            fs_uuid);
1704                 BUG_ON(!device); /* Logic error */
1705
1706                 if (device->fs_devices->seeding) {
1707                         btrfs_set_device_generation(leaf, dev_item,
1708                                                     device->generation);
1709                         btrfs_mark_buffer_dirty(leaf);
1710                 }
1711
1712                 path->slots[0]++;
1713                 goto next_slot;
1714         }
1715         ret = 0;
1716 error:
1717         btrfs_free_path(path);
1718         return ret;
1719 }
1720
1721 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1722 {
1723         struct request_queue *q;
1724         struct btrfs_trans_handle *trans;
1725         struct btrfs_device *device;
1726         struct block_device *bdev;
1727         struct list_head *devices;
1728         struct super_block *sb = root->fs_info->sb;
1729         struct rcu_string *name;
1730         u64 total_bytes;
1731         int seeding_dev = 0;
1732         int ret = 0;
1733
1734         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1735                 return -EROFS;
1736
1737         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1738                                   root->fs_info->bdev_holder);
1739         if (IS_ERR(bdev))
1740                 return PTR_ERR(bdev);
1741
1742         if (root->fs_info->fs_devices->seeding) {
1743                 seeding_dev = 1;
1744                 down_write(&sb->s_umount);
1745                 mutex_lock(&uuid_mutex);
1746         }
1747
1748         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1749
1750         devices = &root->fs_info->fs_devices->devices;
1751
1752         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1753         list_for_each_entry(device, devices, dev_list) {
1754                 if (device->bdev == bdev) {
1755                         ret = -EEXIST;
1756                         mutex_unlock(
1757                                 &root->fs_info->fs_devices->device_list_mutex);
1758                         goto error;
1759                 }
1760         }
1761         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1762
1763         device = kzalloc(sizeof(*device), GFP_NOFS);
1764         if (!device) {
1765                 /* we can safely leave the fs_devices entry around */
1766                 ret = -ENOMEM;
1767                 goto error;
1768         }
1769
1770         name = rcu_string_strdup(device_path, GFP_NOFS);
1771         if (!name) {
1772                 kfree(device);
1773                 ret = -ENOMEM;
1774                 goto error;
1775         }
1776         rcu_assign_pointer(device->name, name);
1777
1778         ret = find_next_devid(root, &device->devid);
1779         if (ret) {
1780                 rcu_string_free(device->name);
1781                 kfree(device);
1782                 goto error;
1783         }
1784
1785         trans = btrfs_start_transaction(root, 0);
1786         if (IS_ERR(trans)) {
1787                 rcu_string_free(device->name);
1788                 kfree(device);
1789                 ret = PTR_ERR(trans);
1790                 goto error;
1791         }
1792
1793         lock_chunks(root);
1794
1795         q = bdev_get_queue(bdev);
1796         if (blk_queue_discard(q))
1797                 device->can_discard = 1;
1798         device->writeable = 1;
1799         device->work.func = pending_bios_fn;
1800         generate_random_uuid(device->uuid);
1801         spin_lock_init(&device->io_lock);
1802         device->generation = trans->transid;
1803         device->io_width = root->sectorsize;
1804         device->io_align = root->sectorsize;
1805         device->sector_size = root->sectorsize;
1806         device->total_bytes = i_size_read(bdev->bd_inode);
1807         device->disk_total_bytes = device->total_bytes;
1808         device->dev_root = root->fs_info->dev_root;
1809         device->bdev = bdev;
1810         device->in_fs_metadata = 1;
1811         device->mode = FMODE_EXCL;
1812         set_blocksize(device->bdev, 4096);
1813
1814         if (seeding_dev) {
1815                 sb->s_flags &= ~MS_RDONLY;
1816                 ret = btrfs_prepare_sprout(root);
1817                 BUG_ON(ret); /* -ENOMEM */
1818         }
1819
1820         device->fs_devices = root->fs_info->fs_devices;
1821
1822         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1823         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1824         list_add(&device->dev_alloc_list,
1825                  &root->fs_info->fs_devices->alloc_list);
1826         root->fs_info->fs_devices->num_devices++;
1827         root->fs_info->fs_devices->open_devices++;
1828         root->fs_info->fs_devices->rw_devices++;
1829         root->fs_info->fs_devices->total_devices++;
1830         if (device->can_discard)
1831                 root->fs_info->fs_devices->num_can_discard++;
1832         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1833
1834         spin_lock(&root->fs_info->free_chunk_lock);
1835         root->fs_info->free_chunk_space += device->total_bytes;
1836         spin_unlock(&root->fs_info->free_chunk_lock);
1837
1838         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1839                 root->fs_info->fs_devices->rotating = 1;
1840
1841         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1842         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1843                                     total_bytes + device->total_bytes);
1844
1845         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1846         btrfs_set_super_num_devices(root->fs_info->super_copy,
1847                                     total_bytes + 1);
1848         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1849
1850         if (seeding_dev) {
1851                 ret = init_first_rw_device(trans, root, device);
1852                 if (ret) {
1853                         btrfs_abort_transaction(trans, root, ret);
1854                         goto error_trans;
1855                 }
1856                 ret = btrfs_finish_sprout(trans, root);
1857                 if (ret) {
1858                         btrfs_abort_transaction(trans, root, ret);
1859                         goto error_trans;
1860                 }
1861         } else {
1862                 ret = btrfs_add_device(trans, root, device);
1863                 if (ret) {
1864                         btrfs_abort_transaction(trans, root, ret);
1865                         goto error_trans;
1866                 }
1867         }
1868
1869         /*
1870          * we've got more storage, clear any full flags on the space
1871          * infos
1872          */
1873         btrfs_clear_space_info_full(root->fs_info);
1874
1875         unlock_chunks(root);
1876         root->fs_info->num_tolerated_disk_barrier_failures =
1877                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1878         ret = btrfs_commit_transaction(trans, root);
1879
1880         if (seeding_dev) {
1881                 mutex_unlock(&uuid_mutex);
1882                 up_write(&sb->s_umount);
1883
1884                 if (ret) /* transaction commit */
1885                         return ret;
1886
1887                 ret = btrfs_relocate_sys_chunks(root);
1888                 if (ret < 0)
1889                         btrfs_error(root->fs_info, ret,
1890                                     "Failed to relocate sys chunks after "
1891                                     "device initialization. This can be fixed "
1892                                     "using the \"btrfs balance\" command.");
1893                 trans = btrfs_attach_transaction(root);
1894                 if (IS_ERR(trans)) {
1895                         if (PTR_ERR(trans) == -ENOENT)
1896                                 return 0;
1897                         return PTR_ERR(trans);
1898                 }
1899                 ret = btrfs_commit_transaction(trans, root);
1900         }
1901
1902         return ret;
1903
1904 error_trans:
1905         unlock_chunks(root);
1906         btrfs_end_transaction(trans, root);
1907         rcu_string_free(device->name);
1908         kfree(device);
1909 error:
1910         blkdev_put(bdev, FMODE_EXCL);
1911         if (seeding_dev) {
1912                 mutex_unlock(&uuid_mutex);
1913                 up_write(&sb->s_umount);
1914         }
1915         return ret;
1916 }
1917
1918 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1919                                         struct btrfs_device *device)
1920 {
1921         int ret;
1922         struct btrfs_path *path;
1923         struct btrfs_root *root;
1924         struct btrfs_dev_item *dev_item;
1925         struct extent_buffer *leaf;
1926         struct btrfs_key key;
1927
1928         root = device->dev_root->fs_info->chunk_root;
1929
1930         path = btrfs_alloc_path();
1931         if (!path)
1932                 return -ENOMEM;
1933
1934         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1935         key.type = BTRFS_DEV_ITEM_KEY;
1936         key.offset = device->devid;
1937
1938         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1939         if (ret < 0)
1940                 goto out;
1941
1942         if (ret > 0) {
1943                 ret = -ENOENT;
1944                 goto out;
1945         }
1946
1947         leaf = path->nodes[0];
1948         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1949
1950         btrfs_set_device_id(leaf, dev_item, device->devid);
1951         btrfs_set_device_type(leaf, dev_item, device->type);
1952         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1953         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1954         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1955         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1956         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1957         btrfs_mark_buffer_dirty(leaf);
1958
1959 out:
1960         btrfs_free_path(path);
1961         return ret;
1962 }
1963
1964 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1965                       struct btrfs_device *device, u64 new_size)
1966 {
1967         struct btrfs_super_block *super_copy =
1968                 device->dev_root->fs_info->super_copy;
1969         u64 old_total = btrfs_super_total_bytes(super_copy);
1970         u64 diff = new_size - device->total_bytes;
1971
1972         if (!device->writeable)
1973                 return -EACCES;
1974         if (new_size <= device->total_bytes)
1975                 return -EINVAL;
1976
1977         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1978         device->fs_devices->total_rw_bytes += diff;
1979
1980         device->total_bytes = new_size;
1981         device->disk_total_bytes = new_size;
1982         btrfs_clear_space_info_full(device->dev_root->fs_info);
1983
1984         return btrfs_update_device(trans, device);
1985 }
1986
1987 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1988                       struct btrfs_device *device, u64 new_size)
1989 {
1990         int ret;
1991         lock_chunks(device->dev_root);
1992         ret = __btrfs_grow_device(trans, device, new_size);
1993         unlock_chunks(device->dev_root);
1994         return ret;
1995 }
1996
1997 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1998                             struct btrfs_root *root,
1999                             u64 chunk_tree, u64 chunk_objectid,
2000                             u64 chunk_offset)
2001 {
2002         int ret;
2003         struct btrfs_path *path;
2004         struct btrfs_key key;
2005
2006         root = root->fs_info->chunk_root;
2007         path = btrfs_alloc_path();
2008         if (!path)
2009                 return -ENOMEM;
2010
2011         key.objectid = chunk_objectid;
2012         key.offset = chunk_offset;
2013         key.type = BTRFS_CHUNK_ITEM_KEY;
2014
2015         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2016         if (ret < 0)
2017                 goto out;
2018         else if (ret > 0) { /* Logic error or corruption */
2019                 btrfs_error(root->fs_info, -ENOENT,
2020                             "Failed lookup while freeing chunk.");
2021                 ret = -ENOENT;
2022                 goto out;
2023         }
2024
2025         ret = btrfs_del_item(trans, root, path);
2026         if (ret < 0)
2027                 btrfs_error(root->fs_info, ret,
2028                             "Failed to delete chunk item.");
2029 out:
2030         btrfs_free_path(path);
2031         return ret;
2032 }
2033
2034 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2035                         chunk_offset)
2036 {
2037         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2038         struct btrfs_disk_key *disk_key;
2039         struct btrfs_chunk *chunk;
2040         u8 *ptr;
2041         int ret = 0;
2042         u32 num_stripes;
2043         u32 array_size;
2044         u32 len = 0;
2045         u32 cur;
2046         struct btrfs_key key;
2047
2048         array_size = btrfs_super_sys_array_size(super_copy);
2049
2050         ptr = super_copy->sys_chunk_array;
2051         cur = 0;
2052
2053         while (cur < array_size) {
2054                 disk_key = (struct btrfs_disk_key *)ptr;
2055                 btrfs_disk_key_to_cpu(&key, disk_key);
2056
2057                 len = sizeof(*disk_key);
2058
2059                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2060                         chunk = (struct btrfs_chunk *)(ptr + len);
2061                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2062                         len += btrfs_chunk_item_size(num_stripes);
2063                 } else {
2064                         ret = -EIO;
2065                         break;
2066                 }
2067                 if (key.objectid == chunk_objectid &&
2068                     key.offset == chunk_offset) {
2069                         memmove(ptr, ptr + len, array_size - (cur + len));
2070                         array_size -= len;
2071                         btrfs_set_super_sys_array_size(super_copy, array_size);
2072                 } else {
2073                         ptr += len;
2074                         cur += len;
2075                 }
2076         }
2077         return ret;
2078 }
2079
2080 static int btrfs_relocate_chunk(struct btrfs_root *root,
2081                          u64 chunk_tree, u64 chunk_objectid,
2082                          u64 chunk_offset)
2083 {
2084         struct extent_map_tree *em_tree;
2085         struct btrfs_root *extent_root;
2086         struct btrfs_trans_handle *trans;
2087         struct extent_map *em;
2088         struct map_lookup *map;
2089         int ret;
2090         int i;
2091
2092         root = root->fs_info->chunk_root;
2093         extent_root = root->fs_info->extent_root;
2094         em_tree = &root->fs_info->mapping_tree.map_tree;
2095
2096         ret = btrfs_can_relocate(extent_root, chunk_offset);
2097         if (ret)
2098                 return -ENOSPC;
2099
2100         /* step one, relocate all the extents inside this chunk */
2101         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2102         if (ret)
2103                 return ret;
2104
2105         trans = btrfs_start_transaction(root, 0);
2106         BUG_ON(IS_ERR(trans));
2107
2108         lock_chunks(root);
2109
2110         /*
2111          * step two, delete the device extents and the
2112          * chunk tree entries
2113          */
2114         read_lock(&em_tree->lock);
2115         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2116         read_unlock(&em_tree->lock);
2117
2118         BUG_ON(!em || em->start > chunk_offset ||
2119                em->start + em->len < chunk_offset);
2120         map = (struct map_lookup *)em->bdev;
2121
2122         for (i = 0; i < map->num_stripes; i++) {
2123                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2124                                             map->stripes[i].physical);
2125                 BUG_ON(ret);
2126
2127                 if (map->stripes[i].dev) {
2128                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2129                         BUG_ON(ret);
2130                 }
2131         }
2132         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2133                                chunk_offset);
2134
2135         BUG_ON(ret);
2136
2137         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2138
2139         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2140                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2141                 BUG_ON(ret);
2142         }
2143
2144         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2145         BUG_ON(ret);
2146
2147         write_lock(&em_tree->lock);
2148         remove_extent_mapping(em_tree, em);
2149         write_unlock(&em_tree->lock);
2150
2151         kfree(map);
2152         em->bdev = NULL;
2153
2154         /* once for the tree */
2155         free_extent_map(em);
2156         /* once for us */
2157         free_extent_map(em);
2158
2159         unlock_chunks(root);
2160         btrfs_end_transaction(trans, root);
2161         return 0;
2162 }
2163
2164 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2165 {
2166         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2167         struct btrfs_path *path;
2168         struct extent_buffer *leaf;
2169         struct btrfs_chunk *chunk;
2170         struct btrfs_key key;
2171         struct btrfs_key found_key;
2172         u64 chunk_tree = chunk_root->root_key.objectid;
2173         u64 chunk_type;
2174         bool retried = false;
2175         int failed = 0;
2176         int ret;
2177
2178         path = btrfs_alloc_path();
2179         if (!path)
2180                 return -ENOMEM;
2181
2182 again:
2183         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2184         key.offset = (u64)-1;
2185         key.type = BTRFS_CHUNK_ITEM_KEY;
2186
2187         while (1) {
2188                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2189                 if (ret < 0)
2190                         goto error;
2191                 BUG_ON(ret == 0); /* Corruption */
2192
2193                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2194                                           key.type);
2195                 if (ret < 0)
2196                         goto error;
2197                 if (ret > 0)
2198                         break;
2199
2200                 leaf = path->nodes[0];
2201                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2202
2203                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2204                                        struct btrfs_chunk);
2205                 chunk_type = btrfs_chunk_type(leaf, chunk);
2206                 btrfs_release_path(path);
2207
2208                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2209                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2210                                                    found_key.objectid,
2211                                                    found_key.offset);
2212                         if (ret == -ENOSPC)
2213                                 failed++;
2214                         else if (ret)
2215                                 BUG();
2216                 }
2217
2218                 if (found_key.offset == 0)
2219                         break;
2220                 key.offset = found_key.offset - 1;
2221         }
2222         ret = 0;
2223         if (failed && !retried) {
2224                 failed = 0;
2225                 retried = true;
2226                 goto again;
2227         } else if (failed && retried) {
2228                 WARN_ON(1);
2229                 ret = -ENOSPC;
2230         }
2231 error:
2232         btrfs_free_path(path);
2233         return ret;
2234 }
2235
2236 static int insert_balance_item(struct btrfs_root *root,
2237                                struct btrfs_balance_control *bctl)
2238 {
2239         struct btrfs_trans_handle *trans;
2240         struct btrfs_balance_item *item;
2241         struct btrfs_disk_balance_args disk_bargs;
2242         struct btrfs_path *path;
2243         struct extent_buffer *leaf;
2244         struct btrfs_key key;
2245         int ret, err;
2246
2247         path = btrfs_alloc_path();
2248         if (!path)
2249                 return -ENOMEM;
2250
2251         trans = btrfs_start_transaction(root, 0);
2252         if (IS_ERR(trans)) {
2253                 btrfs_free_path(path);
2254                 return PTR_ERR(trans);
2255         }
2256
2257         key.objectid = BTRFS_BALANCE_OBJECTID;
2258         key.type = BTRFS_BALANCE_ITEM_KEY;
2259         key.offset = 0;
2260
2261         ret = btrfs_insert_empty_item(trans, root, path, &key,
2262                                       sizeof(*item));
2263         if (ret)
2264                 goto out;
2265
2266         leaf = path->nodes[0];
2267         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2268
2269         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2270
2271         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2272         btrfs_set_balance_data(leaf, item, &disk_bargs);
2273         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2274         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2275         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2276         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2277
2278         btrfs_set_balance_flags(leaf, item, bctl->flags);
2279
2280         btrfs_mark_buffer_dirty(leaf);
2281 out:
2282         btrfs_free_path(path);
2283         err = btrfs_commit_transaction(trans, root);
2284         if (err && !ret)
2285                 ret = err;
2286         return ret;
2287 }
2288
2289 static int del_balance_item(struct btrfs_root *root)
2290 {
2291         struct btrfs_trans_handle *trans;
2292         struct btrfs_path *path;
2293         struct btrfs_key key;
2294         int ret, err;
2295
2296         path = btrfs_alloc_path();
2297         if (!path)
2298                 return -ENOMEM;
2299
2300         trans = btrfs_start_transaction(root, 0);
2301         if (IS_ERR(trans)) {
2302                 btrfs_free_path(path);
2303                 return PTR_ERR(trans);
2304         }
2305
2306         key.objectid = BTRFS_BALANCE_OBJECTID;
2307         key.type = BTRFS_BALANCE_ITEM_KEY;
2308         key.offset = 0;
2309
2310         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2311         if (ret < 0)
2312                 goto out;
2313         if (ret > 0) {
2314                 ret = -ENOENT;
2315                 goto out;
2316         }
2317
2318         ret = btrfs_del_item(trans, root, path);
2319 out:
2320         btrfs_free_path(path);
2321         err = btrfs_commit_transaction(trans, root);
2322         if (err && !ret)
2323                 ret = err;
2324         return ret;
2325 }
2326
2327 /*
2328  * This is a heuristic used to reduce the number of chunks balanced on
2329  * resume after balance was interrupted.
2330  */
2331 static void update_balance_args(struct btrfs_balance_control *bctl)
2332 {
2333         /*
2334          * Turn on soft mode for chunk types that were being converted.
2335          */
2336         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2337                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2338         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2339                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2340         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2341                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2342
2343         /*
2344          * Turn on usage filter if is not already used.  The idea is
2345          * that chunks that we have already balanced should be
2346          * reasonably full.  Don't do it for chunks that are being
2347          * converted - that will keep us from relocating unconverted
2348          * (albeit full) chunks.
2349          */
2350         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2351             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2352                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2353                 bctl->data.usage = 90;
2354         }
2355         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2356             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2357                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2358                 bctl->sys.usage = 90;
2359         }
2360         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2361             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2362                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2363                 bctl->meta.usage = 90;
2364         }
2365 }
2366
2367 /*
2368  * Should be called with both balance and volume mutexes held to
2369  * serialize other volume operations (add_dev/rm_dev/resize) with
2370  * restriper.  Same goes for unset_balance_control.
2371  */
2372 static void set_balance_control(struct btrfs_balance_control *bctl)
2373 {
2374         struct btrfs_fs_info *fs_info = bctl->fs_info;
2375
2376         BUG_ON(fs_info->balance_ctl);
2377
2378         spin_lock(&fs_info->balance_lock);
2379         fs_info->balance_ctl = bctl;
2380         spin_unlock(&fs_info->balance_lock);
2381 }
2382
2383 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2384 {
2385         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2386
2387         BUG_ON(!fs_info->balance_ctl);
2388
2389         spin_lock(&fs_info->balance_lock);
2390         fs_info->balance_ctl = NULL;
2391         spin_unlock(&fs_info->balance_lock);
2392
2393         kfree(bctl);
2394 }
2395
2396 /*
2397  * Balance filters.  Return 1 if chunk should be filtered out
2398  * (should not be balanced).
2399  */
2400 static int chunk_profiles_filter(u64 chunk_type,
2401                                  struct btrfs_balance_args *bargs)
2402 {
2403         chunk_type = chunk_to_extended(chunk_type) &
2404                                 BTRFS_EXTENDED_PROFILE_MASK;
2405
2406         if (bargs->profiles & chunk_type)
2407                 return 0;
2408
2409         return 1;
2410 }
2411
2412 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2413                               struct btrfs_balance_args *bargs)
2414 {
2415         struct btrfs_block_group_cache *cache;
2416         u64 chunk_used, user_thresh;
2417         int ret = 1;
2418
2419         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2420         chunk_used = btrfs_block_group_used(&cache->item);
2421
2422         user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2423         if (chunk_used < user_thresh)
2424                 ret = 0;
2425
2426         btrfs_put_block_group(cache);
2427         return ret;
2428 }
2429
2430 static int chunk_devid_filter(struct extent_buffer *leaf,
2431                               struct btrfs_chunk *chunk,
2432                               struct btrfs_balance_args *bargs)
2433 {
2434         struct btrfs_stripe *stripe;
2435         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2436         int i;
2437
2438         for (i = 0; i < num_stripes; i++) {
2439                 stripe = btrfs_stripe_nr(chunk, i);
2440                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2441                         return 0;
2442         }
2443
2444         return 1;
2445 }
2446
2447 /* [pstart, pend) */
2448 static int chunk_drange_filter(struct extent_buffer *leaf,
2449                                struct btrfs_chunk *chunk,
2450                                u64 chunk_offset,
2451                                struct btrfs_balance_args *bargs)
2452 {
2453         struct btrfs_stripe *stripe;
2454         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2455         u64 stripe_offset;
2456         u64 stripe_length;
2457         int factor;
2458         int i;
2459
2460         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2461                 return 0;
2462
2463         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2464              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2465                 factor = 2;
2466         else
2467                 factor = 1;
2468         factor = num_stripes / factor;
2469
2470         for (i = 0; i < num_stripes; i++) {
2471                 stripe = btrfs_stripe_nr(chunk, i);
2472                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2473                         continue;
2474
2475                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2476                 stripe_length = btrfs_chunk_length(leaf, chunk);
2477                 do_div(stripe_length, factor);
2478
2479                 if (stripe_offset < bargs->pend &&
2480                     stripe_offset + stripe_length > bargs->pstart)
2481                         return 0;
2482         }
2483
2484         return 1;
2485 }
2486
2487 /* [vstart, vend) */
2488 static int chunk_vrange_filter(struct extent_buffer *leaf,
2489                                struct btrfs_chunk *chunk,
2490                                u64 chunk_offset,
2491                                struct btrfs_balance_args *bargs)
2492 {
2493         if (chunk_offset < bargs->vend &&
2494             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2495                 /* at least part of the chunk is inside this vrange */
2496                 return 0;
2497
2498         return 1;
2499 }
2500
2501 static int chunk_soft_convert_filter(u64 chunk_type,
2502                                      struct btrfs_balance_args *bargs)
2503 {
2504         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2505                 return 0;
2506
2507         chunk_type = chunk_to_extended(chunk_type) &
2508                                 BTRFS_EXTENDED_PROFILE_MASK;
2509
2510         if (bargs->target == chunk_type)
2511                 return 1;
2512
2513         return 0;
2514 }
2515
2516 static int should_balance_chunk(struct btrfs_root *root,
2517                                 struct extent_buffer *leaf,
2518                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2519 {
2520         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2521         struct btrfs_balance_args *bargs = NULL;
2522         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2523
2524         /* type filter */
2525         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2526               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2527                 return 0;
2528         }
2529
2530         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2531                 bargs = &bctl->data;
2532         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2533                 bargs = &bctl->sys;
2534         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2535                 bargs = &bctl->meta;
2536
2537         /* profiles filter */
2538         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2539             chunk_profiles_filter(chunk_type, bargs)) {
2540                 return 0;
2541         }
2542
2543         /* usage filter */
2544         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2545             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2546                 return 0;
2547         }
2548
2549         /* devid filter */
2550         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2551             chunk_devid_filter(leaf, chunk, bargs)) {
2552                 return 0;
2553         }
2554
2555         /* drange filter, makes sense only with devid filter */
2556         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2557             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2558                 return 0;
2559         }
2560
2561         /* vrange filter */
2562         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2563             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2564                 return 0;
2565         }
2566
2567         /* soft profile changing mode */
2568         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2569             chunk_soft_convert_filter(chunk_type, bargs)) {
2570                 return 0;
2571         }
2572
2573         return 1;
2574 }
2575
2576 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2577 {
2578         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2579         struct btrfs_root *chunk_root = fs_info->chunk_root;
2580         struct btrfs_root *dev_root = fs_info->dev_root;
2581         struct list_head *devices;
2582         struct btrfs_device *device;
2583         u64 old_size;
2584         u64 size_to_free;
2585         struct btrfs_chunk *chunk;
2586         struct btrfs_path *path;
2587         struct btrfs_key key;
2588         struct btrfs_key found_key;
2589         struct btrfs_trans_handle *trans;
2590         struct extent_buffer *leaf;
2591         int slot;
2592         int ret;
2593         int enospc_errors = 0;
2594         bool counting = true;
2595
2596         /* step one make some room on all the devices */
2597         devices = &fs_info->fs_devices->devices;
2598         list_for_each_entry(device, devices, dev_list) {
2599                 old_size = device->total_bytes;
2600                 size_to_free = div_factor(old_size, 1);
2601                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2602                 if (!device->writeable ||
2603                     device->total_bytes - device->bytes_used > size_to_free)
2604                         continue;
2605
2606                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2607                 if (ret == -ENOSPC)
2608                         break;
2609                 BUG_ON(ret);
2610
2611                 trans = btrfs_start_transaction(dev_root, 0);
2612                 BUG_ON(IS_ERR(trans));
2613
2614                 ret = btrfs_grow_device(trans, device, old_size);
2615                 BUG_ON(ret);
2616
2617                 btrfs_end_transaction(trans, dev_root);
2618         }
2619
2620         /* step two, relocate all the chunks */
2621         path = btrfs_alloc_path();
2622         if (!path) {
2623                 ret = -ENOMEM;
2624                 goto error;
2625         }
2626
2627         /* zero out stat counters */
2628         spin_lock(&fs_info->balance_lock);
2629         memset(&bctl->stat, 0, sizeof(bctl->stat));
2630         spin_unlock(&fs_info->balance_lock);
2631 again:
2632         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2633         key.offset = (u64)-1;
2634         key.type = BTRFS_CHUNK_ITEM_KEY;
2635
2636         while (1) {
2637                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2638                     atomic_read(&fs_info->balance_cancel_req)) {
2639                         ret = -ECANCELED;
2640                         goto error;
2641                 }
2642
2643                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2644                 if (ret < 0)
2645                         goto error;
2646
2647                 /*
2648                  * this shouldn't happen, it means the last relocate
2649                  * failed
2650                  */
2651                 if (ret == 0)
2652                         BUG(); /* FIXME break ? */
2653
2654                 ret = btrfs_previous_item(chunk_root, path, 0,
2655                                           BTRFS_CHUNK_ITEM_KEY);
2656                 if (ret) {
2657                         ret = 0;
2658                         break;
2659                 }
2660
2661                 leaf = path->nodes[0];
2662                 slot = path->slots[0];
2663                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2664
2665                 if (found_key.objectid != key.objectid)
2666                         break;
2667
2668                 /* chunk zero is special */
2669                 if (found_key.offset == 0)
2670                         break;
2671
2672                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2673
2674                 if (!counting) {
2675                         spin_lock(&fs_info->balance_lock);
2676                         bctl->stat.considered++;
2677                         spin_unlock(&fs_info->balance_lock);
2678                 }
2679
2680                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2681                                            found_key.offset);
2682                 btrfs_release_path(path);
2683                 if (!ret)
2684                         goto loop;
2685
2686                 if (counting) {
2687                         spin_lock(&fs_info->balance_lock);
2688                         bctl->stat.expected++;
2689                         spin_unlock(&fs_info->balance_lock);
2690                         goto loop;
2691                 }
2692
2693                 ret = btrfs_relocate_chunk(chunk_root,
2694                                            chunk_root->root_key.objectid,
2695                                            found_key.objectid,
2696                                            found_key.offset);
2697                 if (ret && ret != -ENOSPC)
2698                         goto error;
2699                 if (ret == -ENOSPC) {
2700                         enospc_errors++;
2701                 } else {
2702                         spin_lock(&fs_info->balance_lock);
2703                         bctl->stat.completed++;
2704                         spin_unlock(&fs_info->balance_lock);
2705                 }
2706 loop:
2707                 key.offset = found_key.offset - 1;
2708         }
2709
2710         if (counting) {
2711                 btrfs_release_path(path);
2712                 counting = false;
2713                 goto again;
2714         }
2715 error:
2716         btrfs_free_path(path);
2717         if (enospc_errors) {
2718                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2719                        enospc_errors);
2720                 if (!ret)
2721                         ret = -ENOSPC;
2722         }
2723
2724         return ret;
2725 }
2726
2727 /**
2728  * alloc_profile_is_valid - see if a given profile is valid and reduced
2729  * @flags: profile to validate
2730  * @extended: if true @flags is treated as an extended profile
2731  */
2732 static int alloc_profile_is_valid(u64 flags, int extended)
2733 {
2734         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2735                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2736
2737         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2738
2739         /* 1) check that all other bits are zeroed */
2740         if (flags & ~mask)
2741                 return 0;
2742
2743         /* 2) see if profile is reduced */
2744         if (flags == 0)
2745                 return !extended; /* "0" is valid for usual profiles */
2746
2747         /* true if exactly one bit set */
2748         return (flags & (flags - 1)) == 0;
2749 }
2750
2751 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2752 {
2753         /* cancel requested || normal exit path */
2754         return atomic_read(&fs_info->balance_cancel_req) ||
2755                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2756                  atomic_read(&fs_info->balance_cancel_req) == 0);
2757 }
2758
2759 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2760 {
2761         int ret;
2762
2763         unset_balance_control(fs_info);
2764         ret = del_balance_item(fs_info->tree_root);
2765         BUG_ON(ret);
2766 }
2767
2768 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2769                                struct btrfs_ioctl_balance_args *bargs);
2770
2771 /*
2772  * Should be called with both balance and volume mutexes held
2773  */
2774 int btrfs_balance(struct btrfs_balance_control *bctl,
2775                   struct btrfs_ioctl_balance_args *bargs)
2776 {
2777         struct btrfs_fs_info *fs_info = bctl->fs_info;
2778         u64 allowed;
2779         int mixed = 0;
2780         int ret;
2781
2782         if (btrfs_fs_closing(fs_info) ||
2783             atomic_read(&fs_info->balance_pause_req) ||
2784             atomic_read(&fs_info->balance_cancel_req)) {
2785                 ret = -EINVAL;
2786                 goto out;
2787         }
2788
2789         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2790         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2791                 mixed = 1;
2792
2793         /*
2794          * In case of mixed groups both data and meta should be picked,
2795          * and identical options should be given for both of them.
2796          */
2797         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2798         if (mixed && (bctl->flags & allowed)) {
2799                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2800                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2801                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2802                         printk(KERN_ERR "btrfs: with mixed groups data and "
2803                                "metadata balance options must be the same\n");
2804                         ret = -EINVAL;
2805                         goto out;
2806                 }
2807         }
2808
2809         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2810         if (fs_info->fs_devices->num_devices == 1)
2811                 allowed |= BTRFS_BLOCK_GROUP_DUP;
2812         else if (fs_info->fs_devices->num_devices < 4)
2813                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2814         else
2815                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2816                                 BTRFS_BLOCK_GROUP_RAID10);
2817
2818         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2819             (!alloc_profile_is_valid(bctl->data.target, 1) ||
2820              (bctl->data.target & ~allowed))) {
2821                 printk(KERN_ERR "btrfs: unable to start balance with target "
2822                        "data profile %llu\n",
2823                        (unsigned long long)bctl->data.target);
2824                 ret = -EINVAL;
2825                 goto out;
2826         }
2827         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2828             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2829              (bctl->meta.target & ~allowed))) {
2830                 printk(KERN_ERR "btrfs: unable to start balance with target "
2831                        "metadata profile %llu\n",
2832                        (unsigned long long)bctl->meta.target);
2833                 ret = -EINVAL;
2834                 goto out;
2835         }
2836         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2837             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2838              (bctl->sys.target & ~allowed))) {
2839                 printk(KERN_ERR "btrfs: unable to start balance with target "
2840                        "system profile %llu\n",
2841                        (unsigned long long)bctl->sys.target);
2842                 ret = -EINVAL;
2843                 goto out;
2844         }
2845
2846         /* allow dup'ed data chunks only in mixed mode */
2847         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2848             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2849                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2850                 ret = -EINVAL;
2851                 goto out;
2852         }
2853
2854         /* allow to reduce meta or sys integrity only if force set */
2855         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2856                         BTRFS_BLOCK_GROUP_RAID10;
2857         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2858              (fs_info->avail_system_alloc_bits & allowed) &&
2859              !(bctl->sys.target & allowed)) ||
2860             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2861              (fs_info->avail_metadata_alloc_bits & allowed) &&
2862              !(bctl->meta.target & allowed))) {
2863                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2864                         printk(KERN_INFO "btrfs: force reducing metadata "
2865                                "integrity\n");
2866                 } else {
2867                         printk(KERN_ERR "btrfs: balance will reduce metadata "
2868                                "integrity, use force if you want this\n");
2869                         ret = -EINVAL;
2870                         goto out;
2871                 }
2872         }
2873
2874         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
2875                 int num_tolerated_disk_barrier_failures;
2876                 u64 target = bctl->sys.target;
2877
2878                 num_tolerated_disk_barrier_failures =
2879                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2880                 if (num_tolerated_disk_barrier_failures > 0 &&
2881                     (target &
2882                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
2883                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
2884                         num_tolerated_disk_barrier_failures = 0;
2885                 else if (num_tolerated_disk_barrier_failures > 1 &&
2886                          (target &
2887                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
2888                         num_tolerated_disk_barrier_failures = 1;
2889
2890                 fs_info->num_tolerated_disk_barrier_failures =
2891                         num_tolerated_disk_barrier_failures;
2892         }
2893
2894         ret = insert_balance_item(fs_info->tree_root, bctl);
2895         if (ret && ret != -EEXIST)
2896                 goto out;
2897
2898         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2899                 BUG_ON(ret == -EEXIST);
2900                 set_balance_control(bctl);
2901         } else {
2902                 BUG_ON(ret != -EEXIST);
2903                 spin_lock(&fs_info->balance_lock);
2904                 update_balance_args(bctl);
2905                 spin_unlock(&fs_info->balance_lock);
2906         }
2907
2908         atomic_inc(&fs_info->balance_running);
2909         mutex_unlock(&fs_info->balance_mutex);
2910
2911         ret = __btrfs_balance(fs_info);
2912
2913         mutex_lock(&fs_info->balance_mutex);
2914         atomic_dec(&fs_info->balance_running);
2915
2916         if (bargs) {
2917                 memset(bargs, 0, sizeof(*bargs));
2918                 update_ioctl_balance_args(fs_info, 0, bargs);
2919         }
2920
2921         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2922             balance_need_close(fs_info)) {
2923                 __cancel_balance(fs_info);
2924         }
2925
2926         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
2927                 fs_info->num_tolerated_disk_barrier_failures =
2928                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2929         }
2930
2931         wake_up(&fs_info->balance_wait_q);
2932
2933         return ret;
2934 out:
2935         if (bctl->flags & BTRFS_BALANCE_RESUME)
2936                 __cancel_balance(fs_info);
2937         else
2938                 kfree(bctl);
2939         return ret;
2940 }
2941
2942 static int balance_kthread(void *data)
2943 {
2944         struct btrfs_fs_info *fs_info = data;
2945         int ret = 0;
2946
2947         mutex_lock(&fs_info->volume_mutex);
2948         mutex_lock(&fs_info->balance_mutex);
2949
2950         if (fs_info->balance_ctl) {
2951                 printk(KERN_INFO "btrfs: continuing balance\n");
2952                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
2953         }
2954
2955         mutex_unlock(&fs_info->balance_mutex);
2956         mutex_unlock(&fs_info->volume_mutex);
2957
2958         return ret;
2959 }
2960
2961 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
2962 {
2963         struct task_struct *tsk;
2964
2965         spin_lock(&fs_info->balance_lock);
2966         if (!fs_info->balance_ctl) {
2967                 spin_unlock(&fs_info->balance_lock);
2968                 return 0;
2969         }
2970         spin_unlock(&fs_info->balance_lock);
2971
2972         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2973                 printk(KERN_INFO "btrfs: force skipping balance\n");
2974                 return 0;
2975         }
2976
2977         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
2978         if (IS_ERR(tsk))
2979                 return PTR_ERR(tsk);
2980
2981         return 0;
2982 }
2983
2984 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
2985 {
2986         struct btrfs_balance_control *bctl;
2987         struct btrfs_balance_item *item;
2988         struct btrfs_disk_balance_args disk_bargs;
2989         struct btrfs_path *path;
2990         struct extent_buffer *leaf;
2991         struct btrfs_key key;
2992         int ret;
2993
2994         path = btrfs_alloc_path();
2995         if (!path)
2996                 return -ENOMEM;
2997
2998         key.objectid = BTRFS_BALANCE_OBJECTID;
2999         key.type = BTRFS_BALANCE_ITEM_KEY;
3000         key.offset = 0;
3001
3002         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3003         if (ret < 0)
3004                 goto out;
3005         if (ret > 0) { /* ret = -ENOENT; */
3006                 ret = 0;
3007                 goto out;
3008         }
3009
3010         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3011         if (!bctl) {
3012                 ret = -ENOMEM;
3013                 goto out;
3014         }
3015
3016         leaf = path->nodes[0];
3017         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3018
3019         bctl->fs_info = fs_info;
3020         bctl->flags = btrfs_balance_flags(leaf, item);
3021         bctl->flags |= BTRFS_BALANCE_RESUME;
3022
3023         btrfs_balance_data(leaf, item, &disk_bargs);
3024         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3025         btrfs_balance_meta(leaf, item, &disk_bargs);
3026         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3027         btrfs_balance_sys(leaf, item, &disk_bargs);
3028         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3029
3030         mutex_lock(&fs_info->volume_mutex);
3031         mutex_lock(&fs_info->balance_mutex);
3032
3033         set_balance_control(bctl);
3034
3035         mutex_unlock(&fs_info->balance_mutex);
3036         mutex_unlock(&fs_info->volume_mutex);
3037 out:
3038         btrfs_free_path(path);
3039         return ret;
3040 }
3041
3042 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3043 {
3044         int ret = 0;
3045
3046         mutex_lock(&fs_info->balance_mutex);
3047         if (!fs_info->balance_ctl) {
3048                 mutex_unlock(&fs_info->balance_mutex);
3049                 return -ENOTCONN;
3050         }
3051
3052         if (atomic_read(&fs_info->balance_running)) {
3053                 atomic_inc(&fs_info->balance_pause_req);
3054                 mutex_unlock(&fs_info->balance_mutex);
3055
3056                 wait_event(fs_info->balance_wait_q,
3057                            atomic_read(&fs_info->balance_running) == 0);
3058
3059                 mutex_lock(&fs_info->balance_mutex);
3060                 /* we are good with balance_ctl ripped off from under us */
3061                 BUG_ON(atomic_read(&fs_info->balance_running));
3062                 atomic_dec(&fs_info->balance_pause_req);
3063         } else {
3064                 ret = -ENOTCONN;
3065         }
3066
3067         mutex_unlock(&fs_info->balance_mutex);
3068         return ret;
3069 }
3070
3071 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3072 {
3073         mutex_lock(&fs_info->balance_mutex);
3074         if (!fs_info->balance_ctl) {
3075                 mutex_unlock(&fs_info->balance_mutex);
3076                 return -ENOTCONN;
3077         }
3078
3079         atomic_inc(&fs_info->balance_cancel_req);
3080         /*
3081          * if we are running just wait and return, balance item is
3082          * deleted in btrfs_balance in this case
3083          */
3084         if (atomic_read(&fs_info->balance_running)) {
3085                 mutex_unlock(&fs_info->balance_mutex);
3086                 wait_event(fs_info->balance_wait_q,
3087                            atomic_read(&fs_info->balance_running) == 0);
3088                 mutex_lock(&fs_info->balance_mutex);
3089         } else {
3090                 /* __cancel_balance needs volume_mutex */
3091                 mutex_unlock(&fs_info->balance_mutex);
3092                 mutex_lock(&fs_info->volume_mutex);
3093                 mutex_lock(&fs_info->balance_mutex);
3094
3095                 if (fs_info->balance_ctl)
3096                         __cancel_balance(fs_info);
3097
3098                 mutex_unlock(&fs_info->volume_mutex);
3099         }
3100
3101         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3102         atomic_dec(&fs_info->balance_cancel_req);
3103         mutex_unlock(&fs_info->balance_mutex);
3104         return 0;
3105 }
3106
3107 /*
3108  * shrinking a device means finding all of the device extents past
3109  * the new size, and then following the back refs to the chunks.
3110  * The chunk relocation code actually frees the device extent
3111  */
3112 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3113 {
3114         struct btrfs_trans_handle *trans;
3115         struct btrfs_root *root = device->dev_root;
3116         struct btrfs_dev_extent *dev_extent = NULL;
3117         struct btrfs_path *path;
3118         u64 length;
3119         u64 chunk_tree;
3120         u64 chunk_objectid;
3121         u64 chunk_offset;
3122         int ret;
3123         int slot;
3124         int failed = 0;
3125         bool retried = false;
3126         struct extent_buffer *l;
3127         struct btrfs_key key;
3128         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3129         u64 old_total = btrfs_super_total_bytes(super_copy);
3130         u64 old_size = device->total_bytes;
3131         u64 diff = device->total_bytes - new_size;
3132
3133         path = btrfs_alloc_path();
3134         if (!path)
3135                 return -ENOMEM;
3136
3137         path->reada = 2;
3138
3139         lock_chunks(root);
3140
3141         device->total_bytes = new_size;
3142         if (device->writeable) {
3143                 device->fs_devices->total_rw_bytes -= diff;
3144                 spin_lock(&root->fs_info->free_chunk_lock);
3145                 root->fs_info->free_chunk_space -= diff;
3146                 spin_unlock(&root->fs_info->free_chunk_lock);
3147         }
3148         unlock_chunks(root);
3149
3150 again:
3151         key.objectid = device->devid;
3152         key.offset = (u64)-1;
3153         key.type = BTRFS_DEV_EXTENT_KEY;
3154
3155         do {
3156                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3157                 if (ret < 0)
3158                         goto done;
3159
3160                 ret = btrfs_previous_item(root, path, 0, key.type);
3161                 if (ret < 0)
3162                         goto done;
3163                 if (ret) {
3164                         ret = 0;
3165                         btrfs_release_path(path);
3166                         break;
3167                 }
3168
3169                 l = path->nodes[0];
3170                 slot = path->slots[0];
3171                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3172
3173                 if (key.objectid != device->devid) {
3174                         btrfs_release_path(path);
3175                         break;
3176                 }
3177
3178                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3179                 length = btrfs_dev_extent_length(l, dev_extent);
3180
3181                 if (key.offset + length <= new_size) {
3182                         btrfs_release_path(path);
3183                         break;
3184                 }
3185
3186                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3187                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3188                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3189                 btrfs_release_path(path);
3190
3191                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3192                                            chunk_offset);
3193                 if (ret && ret != -ENOSPC)
3194                         goto done;
3195                 if (ret == -ENOSPC)
3196                         failed++;
3197         } while (key.offset-- > 0);
3198
3199         if (failed && !retried) {
3200                 failed = 0;
3201                 retried = true;
3202                 goto again;
3203         } else if (failed && retried) {
3204                 ret = -ENOSPC;
3205                 lock_chunks(root);
3206
3207                 device->total_bytes = old_size;
3208                 if (device->writeable)
3209                         device->fs_devices->total_rw_bytes += diff;
3210                 spin_lock(&root->fs_info->free_chunk_lock);
3211                 root->fs_info->free_chunk_space += diff;
3212                 spin_unlock(&root->fs_info->free_chunk_lock);
3213                 unlock_chunks(root);
3214                 goto done;
3215         }
3216
3217         /* Shrinking succeeded, else we would be at "done". */
3218         trans = btrfs_start_transaction(root, 0);
3219         if (IS_ERR(trans)) {
3220                 ret = PTR_ERR(trans);
3221                 goto done;
3222         }
3223
3224         lock_chunks(root);
3225
3226         device->disk_total_bytes = new_size;
3227         /* Now btrfs_update_device() will change the on-disk size. */
3228         ret = btrfs_update_device(trans, device);
3229         if (ret) {
3230                 unlock_chunks(root);
3231                 btrfs_end_transaction(trans, root);
3232                 goto done;
3233         }
3234         WARN_ON(diff > old_total);
3235         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3236         unlock_chunks(root);
3237         btrfs_end_transaction(trans, root);
3238 done:
3239         btrfs_free_path(path);
3240         return ret;
3241 }
3242
3243 static int btrfs_add_system_chunk(struct btrfs_root *root,
3244                            struct btrfs_key *key,
3245                            struct btrfs_chunk *chunk, int item_size)
3246 {
3247         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3248         struct btrfs_disk_key disk_key;
3249         u32 array_size;
3250         u8 *ptr;
3251
3252         array_size = btrfs_super_sys_array_size(super_copy);
3253         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3254                 return -EFBIG;
3255
3256         ptr = super_copy->sys_chunk_array + array_size;
3257         btrfs_cpu_key_to_disk(&disk_key, key);
3258         memcpy(ptr, &disk_key, sizeof(disk_key));
3259         ptr += sizeof(disk_key);
3260         memcpy(ptr, chunk, item_size);
3261         item_size += sizeof(disk_key);
3262         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3263         return 0;
3264 }
3265
3266 /*
3267  * sort the devices in descending order by max_avail, total_avail
3268  */
3269 static int btrfs_cmp_device_info(const void *a, const void *b)
3270 {
3271         const struct btrfs_device_info *di_a = a;
3272         const struct btrfs_device_info *di_b = b;
3273
3274         if (di_a->max_avail > di_b->max_avail)
3275                 return -1;
3276         if (di_a->max_avail < di_b->max_avail)
3277                 return 1;
3278         if (di_a->total_avail > di_b->total_avail)
3279                 return -1;
3280         if (di_a->total_avail < di_b->total_avail)
3281                 return 1;
3282         return 0;
3283 }
3284
3285 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3286                                struct btrfs_root *extent_root,
3287                                struct map_lookup **map_ret,
3288                                u64 *num_bytes_out, u64 *stripe_size_out,
3289                                u64 start, u64 type)
3290 {
3291         struct btrfs_fs_info *info = extent_root->fs_info;
3292         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3293         struct list_head *cur;
3294         struct map_lookup *map = NULL;
3295         struct extent_map_tree *em_tree;
3296         struct extent_map *em;
3297         struct btrfs_device_info *devices_info = NULL;
3298         u64 total_avail;
3299         int num_stripes;        /* total number of stripes to allocate */
3300         int sub_stripes;        /* sub_stripes info for map */
3301         int dev_stripes;        /* stripes per dev */
3302         int devs_max;           /* max devs to use */
3303         int devs_min;           /* min devs needed */
3304         int devs_increment;     /* ndevs has to be a multiple of this */
3305         int ncopies;            /* how many copies to data has */
3306         int ret;
3307         u64 max_stripe_size;
3308         u64 max_chunk_size;
3309         u64 stripe_size;
3310         u64 num_bytes;
3311         int ndevs;
3312         int i;
3313         int j;
3314
3315         BUG_ON(!alloc_profile_is_valid(type, 0));
3316
3317         if (list_empty(&fs_devices->alloc_list))
3318                 return -ENOSPC;
3319
3320         sub_stripes = 1;
3321         dev_stripes = 1;
3322         devs_increment = 1;
3323         ncopies = 1;
3324         devs_max = 0;   /* 0 == as many as possible */
3325         devs_min = 1;
3326
3327         /*
3328          * define the properties of each RAID type.
3329          * FIXME: move this to a global table and use it in all RAID
3330          * calculation code
3331          */
3332         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3333                 dev_stripes = 2;
3334                 ncopies = 2;
3335                 devs_max = 1;
3336         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3337                 devs_min = 2;
3338         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3339                 devs_increment = 2;
3340                 ncopies = 2;
3341                 devs_max = 2;
3342                 devs_min = 2;
3343         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3344                 sub_stripes = 2;
3345                 devs_increment = 2;
3346                 ncopies = 2;
3347                 devs_min = 4;
3348         } else {
3349                 devs_max = 1;
3350         }
3351
3352         if (type & BTRFS_BLOCK_GROUP_DATA) {
3353                 max_stripe_size = 1024 * 1024 * 1024;
3354                 max_chunk_size = 10 * max_stripe_size;
3355         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3356                 /* for larger filesystems, use larger metadata chunks */
3357                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3358                         max_stripe_size = 1024 * 1024 * 1024;
3359                 else
3360                         max_stripe_size = 256 * 1024 * 1024;
3361                 max_chunk_size = max_stripe_size;
3362         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3363                 max_stripe_size = 32 * 1024 * 1024;
3364                 max_chunk_size = 2 * max_stripe_size;
3365         } else {
3366                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3367                        type);
3368                 BUG_ON(1);
3369         }
3370
3371         /* we don't want a chunk larger than 10% of writeable space */
3372         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3373                              max_chunk_size);
3374
3375         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3376                                GFP_NOFS);
3377         if (!devices_info)
3378                 return -ENOMEM;
3379
3380         cur = fs_devices->alloc_list.next;
3381
3382         /*
3383          * in the first pass through the devices list, we gather information
3384          * about the available holes on each device.
3385          */
3386         ndevs = 0;
3387         while (cur != &fs_devices->alloc_list) {
3388                 struct btrfs_device *device;
3389                 u64 max_avail;
3390                 u64 dev_offset;
3391
3392                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3393
3394                 cur = cur->next;
3395
3396                 if (!device->writeable) {
3397                         WARN(1, KERN_ERR
3398                                "btrfs: read-only device in alloc_list\n");
3399                         continue;
3400                 }
3401
3402                 if (!device->in_fs_metadata)
3403                         continue;
3404
3405                 if (device->total_bytes > device->bytes_used)
3406                         total_avail = device->total_bytes - device->bytes_used;
3407                 else
3408                         total_avail = 0;
3409
3410                 /* If there is no space on this device, skip it. */
3411                 if (total_avail == 0)
3412                         continue;
3413
3414                 ret = find_free_dev_extent(device,
3415                                            max_stripe_size * dev_stripes,
3416                                            &dev_offset, &max_avail);
3417                 if (ret && ret != -ENOSPC)
3418                         goto error;
3419
3420                 if (ret == 0)
3421                         max_avail = max_stripe_size * dev_stripes;
3422
3423                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3424                         continue;
3425
3426                 devices_info[ndevs].dev_offset = dev_offset;
3427                 devices_info[ndevs].max_avail = max_avail;
3428                 devices_info[ndevs].total_avail = total_avail;
3429                 devices_info[ndevs].dev = device;
3430                 ++ndevs;
3431         }
3432
3433         /*
3434          * now sort the devices by hole size / available space
3435          */
3436         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3437              btrfs_cmp_device_info, NULL);
3438
3439         /* round down to number of usable stripes */
3440         ndevs -= ndevs % devs_increment;
3441
3442         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3443                 ret = -ENOSPC;
3444                 goto error;
3445         }
3446
3447         if (devs_max && ndevs > devs_max)
3448                 ndevs = devs_max;
3449         /*
3450          * the primary goal is to maximize the number of stripes, so use as many
3451          * devices as possible, even if the stripes are not maximum sized.
3452          */
3453         stripe_size = devices_info[ndevs-1].max_avail;
3454         num_stripes = ndevs * dev_stripes;
3455
3456         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3457                 stripe_size = max_chunk_size * ncopies;
3458                 do_div(stripe_size, ndevs);
3459         }
3460
3461         do_div(stripe_size, dev_stripes);
3462
3463         /* align to BTRFS_STRIPE_LEN */
3464         do_div(stripe_size, BTRFS_STRIPE_LEN);
3465         stripe_size *= BTRFS_STRIPE_LEN;
3466
3467         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3468         if (!map) {
3469                 ret = -ENOMEM;
3470                 goto error;
3471         }
3472         map->num_stripes = num_stripes;
3473
3474         for (i = 0; i < ndevs; ++i) {
3475                 for (j = 0; j < dev_stripes; ++j) {
3476                         int s = i * dev_stripes + j;
3477                         map->stripes[s].dev = devices_info[i].dev;
3478                         map->stripes[s].physical = devices_info[i].dev_offset +
3479                                                    j * stripe_size;
3480                 }
3481         }
3482         map->sector_size = extent_root->sectorsize;
3483         map->stripe_len = BTRFS_STRIPE_LEN;
3484         map->io_align = BTRFS_STRIPE_LEN;
3485         map->io_width = BTRFS_STRIPE_LEN;
3486         map->type = type;
3487         map->sub_stripes = sub_stripes;
3488
3489         *map_ret = map;
3490         num_bytes = stripe_size * (num_stripes / ncopies);
3491
3492         *stripe_size_out = stripe_size;
3493         *num_bytes_out = num_bytes;
3494
3495         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3496
3497         em = alloc_extent_map();
3498         if (!em) {
3499                 ret = -ENOMEM;
3500                 goto error;
3501         }
3502         em->bdev = (struct block_device *)map;
3503         em->start = start;
3504         em->len = num_bytes;
3505         em->block_start = 0;
3506         em->block_len = em->len;
3507
3508         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3509         write_lock(&em_tree->lock);
3510         ret = add_extent_mapping(em_tree, em);
3511         write_unlock(&em_tree->lock);
3512         free_extent_map(em);
3513         if (ret)
3514                 goto error;
3515
3516         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3517                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3518                                      start, num_bytes);
3519         if (ret)
3520                 goto error;
3521
3522         for (i = 0; i < map->num_stripes; ++i) {
3523                 struct btrfs_device *device;
3524                 u64 dev_offset;
3525
3526                 device = map->stripes[i].dev;
3527                 dev_offset = map->stripes[i].physical;
3528
3529                 ret = btrfs_alloc_dev_extent(trans, device,
3530                                 info->chunk_root->root_key.objectid,
3531                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3532                                 start, dev_offset, stripe_size);
3533                 if (ret) {
3534                         btrfs_abort_transaction(trans, extent_root, ret);
3535                         goto error;
3536                 }
3537         }
3538
3539         kfree(devices_info);
3540         return 0;
3541
3542 error:
3543         kfree(map);
3544         kfree(devices_info);
3545         return ret;
3546 }
3547
3548 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3549                                 struct btrfs_root *extent_root,
3550                                 struct map_lookup *map, u64 chunk_offset,
3551                                 u64 chunk_size, u64 stripe_size)
3552 {
3553         u64 dev_offset;
3554         struct btrfs_key key;
3555         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3556         struct btrfs_device *device;
3557         struct btrfs_chunk *chunk;
3558         struct btrfs_stripe *stripe;
3559         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3560         int index = 0;
3561         int ret;
3562
3563         chunk = kzalloc(item_size, GFP_NOFS);
3564         if (!chunk)
3565                 return -ENOMEM;
3566
3567         index = 0;
3568         while (index < map->num_stripes) {
3569                 device = map->stripes[index].dev;
3570                 device->bytes_used += stripe_size;
3571                 ret = btrfs_update_device(trans, device);
3572                 if (ret)
3573                         goto out_free;
3574                 index++;
3575         }
3576
3577         spin_lock(&extent_root->fs_info->free_chunk_lock);
3578         extent_root->fs_info->free_chunk_space -= (stripe_size *
3579                                                    map->num_stripes);
3580         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3581
3582         index = 0;
3583         stripe = &chunk->stripe;
3584         while (index < map->num_stripes) {
3585                 device = map->stripes[index].dev;
3586                 dev_offset = map->stripes[index].physical;
3587
3588                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3589                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3590                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3591                 stripe++;
3592                 index++;
3593         }
3594
3595         btrfs_set_stack_chunk_length(chunk, chunk_size);
3596         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3597         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3598         btrfs_set_stack_chunk_type(chunk, map->type);
3599         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3600         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3601         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3602         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3603         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3604
3605         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3606         key.type = BTRFS_CHUNK_ITEM_KEY;
3607         key.offset = chunk_offset;
3608
3609         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3610
3611         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3612                 /*
3613                  * TODO: Cleanup of inserted chunk root in case of
3614                  * failure.
3615                  */
3616                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3617                                              item_size);
3618         }
3619
3620 out_free:
3621         kfree(chunk);
3622         return ret;
3623 }
3624
3625 /*
3626  * Chunk allocation falls into two parts. The first part does works
3627  * that make the new allocated chunk useable, but not do any operation
3628  * that modifies the chunk tree. The second part does the works that
3629  * require modifying the chunk tree. This division is important for the
3630  * bootstrap process of adding storage to a seed btrfs.
3631  */
3632 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3633                       struct btrfs_root *extent_root, u64 type)
3634 {
3635         u64 chunk_offset;
3636         u64 chunk_size;
3637         u64 stripe_size;
3638         struct map_lookup *map;
3639         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3640         int ret;
3641
3642         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3643                               &chunk_offset);
3644         if (ret)
3645                 return ret;
3646
3647         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3648                                   &stripe_size, chunk_offset, type);
3649         if (ret)
3650                 return ret;
3651
3652         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3653                                    chunk_size, stripe_size);
3654         if (ret)
3655                 return ret;
3656         return 0;
3657 }
3658
3659 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3660                                          struct btrfs_root *root,
3661                                          struct btrfs_device *device)
3662 {
3663         u64 chunk_offset;
3664         u64 sys_chunk_offset;
3665         u64 chunk_size;
3666         u64 sys_chunk_size;
3667         u64 stripe_size;
3668         u64 sys_stripe_size;
3669         u64 alloc_profile;
3670         struct map_lookup *map;
3671         struct map_lookup *sys_map;
3672         struct btrfs_fs_info *fs_info = root->fs_info;
3673         struct btrfs_root *extent_root = fs_info->extent_root;
3674         int ret;
3675
3676         ret = find_next_chunk(fs_info->chunk_root,
3677                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3678         if (ret)
3679                 return ret;
3680
3681         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3682                                 fs_info->avail_metadata_alloc_bits;
3683         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3684
3685         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3686                                   &stripe_size, chunk_offset, alloc_profile);
3687         if (ret)
3688                 return ret;
3689
3690         sys_chunk_offset = chunk_offset + chunk_size;
3691
3692         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3693                                 fs_info->avail_system_alloc_bits;
3694         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3695
3696         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3697                                   &sys_chunk_size, &sys_stripe_size,
3698                                   sys_chunk_offset, alloc_profile);
3699         if (ret) {
3700                 btrfs_abort_transaction(trans, root, ret);
3701                 goto out;
3702         }
3703
3704         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3705         if (ret) {
3706                 btrfs_abort_transaction(trans, root, ret);
3707                 goto out;
3708         }
3709
3710         /*
3711          * Modifying chunk tree needs allocating new blocks from both
3712          * system block group and metadata block group. So we only can
3713          * do operations require modifying the chunk tree after both
3714          * block groups were created.
3715          */
3716         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3717                                    chunk_size, stripe_size);
3718         if (ret) {
3719                 btrfs_abort_transaction(trans, root, ret);
3720                 goto out;
3721         }
3722
3723         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3724                                    sys_chunk_offset, sys_chunk_size,
3725                                    sys_stripe_size);
3726         if (ret)
3727                 btrfs_abort_transaction(trans, root, ret);
3728
3729 out:
3730
3731         return ret;
3732 }
3733
3734 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3735 {
3736         struct extent_map *em;
3737         struct map_lookup *map;
3738         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3739         int readonly = 0;
3740         int i;
3741
3742         read_lock(&map_tree->map_tree.lock);
3743         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3744         read_unlock(&map_tree->map_tree.lock);
3745         if (!em)
3746                 return 1;
3747
3748         if (btrfs_test_opt(root, DEGRADED)) {
3749                 free_extent_map(em);
3750                 return 0;
3751         }
3752
3753         map = (struct map_lookup *)em->bdev;
3754         for (i = 0; i < map->num_stripes; i++) {
3755                 if (!map->stripes[i].dev->writeable) {
3756                         readonly = 1;
3757                         break;
3758                 }
3759         }
3760         free_extent_map(em);
3761         return readonly;
3762 }
3763
3764 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3765 {
3766         extent_map_tree_init(&tree->map_tree);
3767 }
3768
3769 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3770 {
3771         struct extent_map *em;
3772
3773         while (1) {
3774                 write_lock(&tree->map_tree.lock);
3775                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3776                 if (em)
3777                         remove_extent_mapping(&tree->map_tree, em);
3778                 write_unlock(&tree->map_tree.lock);
3779                 if (!em)
3780                         break;
3781                 kfree(em->bdev);
3782                 /* once for us */
3783                 free_extent_map(em);
3784                 /* once for the tree */
3785                 free_extent_map(em);
3786         }
3787 }
3788
3789 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
3790 {
3791         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
3792         struct extent_map *em;
3793         struct map_lookup *map;
3794         struct extent_map_tree *em_tree = &map_tree->map_tree;
3795         int ret;
3796
3797         read_lock(&em_tree->lock);
3798         em = lookup_extent_mapping(em_tree, logical, len);
3799         read_unlock(&em_tree->lock);
3800         BUG_ON(!em);
3801
3802         BUG_ON(em->start > logical || em->start + em->len < logical);
3803         map = (struct map_lookup *)em->bdev;
3804         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3805                 ret = map->num_stripes;
3806         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3807                 ret = map->sub_stripes;
3808         else
3809                 ret = 1;
3810         free_extent_map(em);
3811         return ret;
3812 }
3813
3814 static int find_live_mirror(struct map_lookup *map, int first, int num,
3815                             int optimal)
3816 {
3817         int i;
3818         if (map->stripes[optimal].dev->bdev)
3819                 return optimal;
3820         for (i = first; i < first + num; i++) {
3821                 if (map->stripes[i].dev->bdev)
3822                         return i;
3823         }
3824         /* we couldn't find one that doesn't fail.  Just return something
3825          * and the io error handling code will clean up eventually
3826          */
3827         return optimal;
3828 }
3829
3830 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
3831                              u64 logical, u64 *length,
3832                              struct btrfs_bio **bbio_ret,
3833                              int mirror_num)
3834 {
3835         struct extent_map *em;
3836         struct map_lookup *map;
3837         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
3838         struct extent_map_tree *em_tree = &map_tree->map_tree;
3839         u64 offset;
3840         u64 stripe_offset;
3841         u64 stripe_end_offset;
3842         u64 stripe_nr;
3843         u64 stripe_nr_orig;
3844         u64 stripe_nr_end;
3845         int stripe_index;
3846         int i;
3847         int ret = 0;
3848         int num_stripes;
3849         int max_errors = 0;
3850         struct btrfs_bio *bbio = NULL;
3851
3852         read_lock(&em_tree->lock);
3853         em = lookup_extent_mapping(em_tree, logical, *length);
3854         read_unlock(&em_tree->lock);
3855
3856         if (!em) {
3857                 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
3858                        (unsigned long long)logical,
3859                        (unsigned long long)*length);
3860                 BUG();
3861         }
3862
3863         BUG_ON(em->start > logical || em->start + em->len < logical);
3864         map = (struct map_lookup *)em->bdev;
3865         offset = logical - em->start;
3866
3867         if (mirror_num > map->num_stripes)
3868                 mirror_num = 0;
3869
3870         stripe_nr = offset;
3871         /*
3872          * stripe_nr counts the total number of stripes we have to stride
3873          * to get to this block
3874          */
3875         do_div(stripe_nr, map->stripe_len);
3876
3877         stripe_offset = stripe_nr * map->stripe_len;
3878         BUG_ON(offset < stripe_offset);
3879
3880         /* stripe_offset is the offset of this block in its stripe*/
3881         stripe_offset = offset - stripe_offset;
3882
3883         if (rw & REQ_DISCARD)
3884                 *length = min_t(u64, em->len - offset, *length);
3885         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3886                 /* we limit the length of each bio to what fits in a stripe */
3887                 *length = min_t(u64, em->len - offset,
3888                                 map->stripe_len - stripe_offset);
3889         } else {
3890                 *length = em->len - offset;
3891         }
3892
3893         if (!bbio_ret)
3894                 goto out;
3895
3896         num_stripes = 1;
3897         stripe_index = 0;
3898         stripe_nr_orig = stripe_nr;
3899         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3900                         (~(map->stripe_len - 1));
3901         do_div(stripe_nr_end, map->stripe_len);
3902         stripe_end_offset = stripe_nr_end * map->stripe_len -
3903                             (offset + *length);
3904         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3905                 if (rw & REQ_DISCARD)
3906                         num_stripes = min_t(u64, map->num_stripes,
3907                                             stripe_nr_end - stripe_nr_orig);
3908                 stripe_index = do_div(stripe_nr, map->num_stripes);
3909         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3910                 if (rw & (REQ_WRITE | REQ_DISCARD))
3911                         num_stripes = map->num_stripes;
3912                 else if (mirror_num)
3913                         stripe_index = mirror_num - 1;
3914                 else {
3915                         stripe_index = find_live_mirror(map, 0,
3916                                             map->num_stripes,
3917                                             current->pid % map->num_stripes);
3918                         mirror_num = stripe_index + 1;
3919                 }
3920
3921         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3922                 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3923                         num_stripes = map->num_stripes;
3924                 } else if (mirror_num) {
3925                         stripe_index = mirror_num - 1;
3926                 } else {
3927                         mirror_num = 1;
3928                 }
3929
3930         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3931                 int factor = map->num_stripes / map->sub_stripes;
3932
3933                 stripe_index = do_div(stripe_nr, factor);
3934                 stripe_index *= map->sub_stripes;
3935
3936                 if (rw & REQ_WRITE)
3937                         num_stripes = map->sub_stripes;
3938                 else if (rw & REQ_DISCARD)
3939                         num_stripes = min_t(u64, map->sub_stripes *
3940                                             (stripe_nr_end - stripe_nr_orig),
3941                                             map->num_stripes);
3942                 else if (mirror_num)
3943                         stripe_index += mirror_num - 1;
3944                 else {
3945                         int old_stripe_index = stripe_index;
3946                         stripe_index = find_live_mirror(map, stripe_index,
3947                                               map->sub_stripes, stripe_index +
3948                                               current->pid % map->sub_stripes);
3949                         mirror_num = stripe_index - old_stripe_index + 1;
3950                 }
3951         } else {
3952                 /*
3953                  * after this do_div call, stripe_nr is the number of stripes
3954                  * on this device we have to walk to find the data, and
3955                  * stripe_index is the number of our device in the stripe array
3956                  */
3957                 stripe_index = do_div(stripe_nr, map->num_stripes);
3958                 mirror_num = stripe_index + 1;
3959         }
3960         BUG_ON(stripe_index >= map->num_stripes);
3961
3962         bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3963         if (!bbio) {
3964                 ret = -ENOMEM;
3965                 goto out;
3966         }
3967         atomic_set(&bbio->error, 0);
3968
3969         if (rw & REQ_DISCARD) {
3970                 int factor = 0;
3971                 int sub_stripes = 0;
3972                 u64 stripes_per_dev = 0;
3973                 u32 remaining_stripes = 0;
3974                 u32 last_stripe = 0;
3975
3976                 if (map->type &
3977                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3978                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3979                                 sub_stripes = 1;
3980                         else
3981                                 sub_stripes = map->sub_stripes;
3982
3983                         factor = map->num_stripes / sub_stripes;
3984                         stripes_per_dev = div_u64_rem(stripe_nr_end -
3985                                                       stripe_nr_orig,
3986                                                       factor,
3987                                                       &remaining_stripes);
3988                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
3989                         last_stripe *= sub_stripes;
3990                 }
3991
3992                 for (i = 0; i < num_stripes; i++) {
3993                         bbio->stripes[i].physical =
3994                                 map->stripes[stripe_index].physical +
3995                                 stripe_offset + stripe_nr * map->stripe_len;
3996                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3997
3998                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3999                                          BTRFS_BLOCK_GROUP_RAID10)) {
4000                                 bbio->stripes[i].length = stripes_per_dev *
4001                                                           map->stripe_len;
4002
4003                                 if (i / sub_stripes < remaining_stripes)
4004                                         bbio->stripes[i].length +=
4005                                                 map->stripe_len;
4006
4007                                 /*
4008                                  * Special for the first stripe and
4009                                  * the last stripe:
4010                                  *
4011                                  * |-------|...|-------|
4012                                  *     |----------|
4013                                  *    off     end_off
4014                                  */
4015                                 if (i < sub_stripes)
4016                                         bbio->stripes[i].length -=
4017                                                 stripe_offset;
4018
4019                                 if (stripe_index >= last_stripe &&
4020                                     stripe_index <= (last_stripe +
4021                                                      sub_stripes - 1))
4022                                         bbio->stripes[i].length -=
4023                                                 stripe_end_offset;
4024
4025                                 if (i == sub_stripes - 1)
4026                                         stripe_offset = 0;
4027                         } else
4028                                 bbio->stripes[i].length = *length;
4029
4030                         stripe_index++;
4031                         if (stripe_index == map->num_stripes) {
4032                                 /* This could only happen for RAID0/10 */
4033                                 stripe_index = 0;
4034                                 stripe_nr++;
4035                         }
4036                 }
4037         } else {
4038                 for (i = 0; i < num_stripes; i++) {
4039                         bbio->stripes[i].physical =
4040                                 map->stripes[stripe_index].physical +
4041                                 stripe_offset +
4042                                 stripe_nr * map->stripe_len;
4043                         bbio->stripes[i].dev =
4044                                 map->stripes[stripe_index].dev;
4045                         stripe_index++;
4046                 }
4047         }
4048
4049         if (rw & REQ_WRITE) {
4050                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4051                                  BTRFS_BLOCK_GROUP_RAID10 |
4052                                  BTRFS_BLOCK_GROUP_DUP)) {
4053                         max_errors = 1;
4054                 }
4055         }
4056
4057         *bbio_ret = bbio;
4058         bbio->num_stripes = num_stripes;
4059         bbio->max_errors = max_errors;
4060         bbio->mirror_num = mirror_num;
4061 out:
4062         free_extent_map(em);
4063         return ret;
4064 }
4065
4066 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4067                       u64 logical, u64 *length,
4068                       struct btrfs_bio **bbio_ret, int mirror_num)
4069 {
4070         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4071                                  mirror_num);
4072 }
4073
4074 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4075                      u64 chunk_start, u64 physical, u64 devid,
4076                      u64 **logical, int *naddrs, int *stripe_len)
4077 {
4078         struct extent_map_tree *em_tree = &map_tree->map_tree;
4079         struct extent_map *em;
4080         struct map_lookup *map;
4081         u64 *buf;
4082         u64 bytenr;
4083         u64 length;
4084         u64 stripe_nr;
4085         int i, j, nr = 0;
4086
4087         read_lock(&em_tree->lock);
4088         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4089         read_unlock(&em_tree->lock);
4090
4091         BUG_ON(!em || em->start != chunk_start);
4092         map = (struct map_lookup *)em->bdev;
4093
4094         length = em->len;
4095         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4096                 do_div(length, map->num_stripes / map->sub_stripes);
4097         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4098                 do_div(length, map->num_stripes);
4099
4100         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4101         BUG_ON(!buf); /* -ENOMEM */
4102
4103         for (i = 0; i < map->num_stripes; i++) {
4104                 if (devid && map->stripes[i].dev->devid != devid)
4105                         continue;
4106                 if (map->stripes[i].physical > physical ||
4107                     map->stripes[i].physical + length <= physical)
4108                         continue;
4109
4110                 stripe_nr = physical - map->stripes[i].physical;
4111                 do_div(stripe_nr, map->stripe_len);
4112
4113                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4114                         stripe_nr = stripe_nr * map->num_stripes + i;
4115                         do_div(stripe_nr, map->sub_stripes);
4116                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4117                         stripe_nr = stripe_nr * map->num_stripes + i;
4118                 }
4119                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4120                 WARN_ON(nr >= map->num_stripes);
4121                 for (j = 0; j < nr; j++) {
4122                         if (buf[j] == bytenr)
4123                                 break;
4124                 }
4125                 if (j == nr) {
4126                         WARN_ON(nr >= map->num_stripes);
4127                         buf[nr++] = bytenr;
4128                 }
4129         }
4130
4131         *logical = buf;
4132         *naddrs = nr;
4133         *stripe_len = map->stripe_len;
4134
4135         free_extent_map(em);
4136         return 0;
4137 }
4138
4139 static void *merge_stripe_index_into_bio_private(void *bi_private,
4140                                                  unsigned int stripe_index)
4141 {
4142         /*
4143          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4144          * at most 1.
4145          * The alternative solution (instead of stealing bits from the
4146          * pointer) would be to allocate an intermediate structure
4147          * that contains the old private pointer plus the stripe_index.
4148          */
4149         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4150         BUG_ON(stripe_index > 3);
4151         return (void *)(((uintptr_t)bi_private) | stripe_index);
4152 }
4153
4154 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4155 {
4156         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4157 }
4158
4159 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4160 {
4161         return (unsigned int)((uintptr_t)bi_private) & 3;
4162 }
4163
4164 static void btrfs_end_bio(struct bio *bio, int err)
4165 {
4166         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4167         int is_orig_bio = 0;
4168
4169         if (err) {
4170                 atomic_inc(&bbio->error);
4171                 if (err == -EIO || err == -EREMOTEIO) {
4172                         unsigned int stripe_index =
4173                                 extract_stripe_index_from_bio_private(
4174                                         bio->bi_private);
4175                         struct btrfs_device *dev;
4176
4177                         BUG_ON(stripe_index >= bbio->num_stripes);
4178                         dev = bbio->stripes[stripe_index].dev;
4179                         if (dev->bdev) {
4180                                 if (bio->bi_rw & WRITE)
4181                                         btrfs_dev_stat_inc(dev,
4182                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4183                                 else
4184                                         btrfs_dev_stat_inc(dev,
4185                                                 BTRFS_DEV_STAT_READ_ERRS);
4186                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4187                                         btrfs_dev_stat_inc(dev,
4188                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4189                                 btrfs_dev_stat_print_on_error(dev);
4190                         }
4191                 }
4192         }
4193
4194         if (bio == bbio->orig_bio)
4195                 is_orig_bio = 1;
4196
4197         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4198                 if (!is_orig_bio) {
4199                         bio_put(bio);
4200                         bio = bbio->orig_bio;
4201                 }
4202                 bio->bi_private = bbio->private;
4203                 bio->bi_end_io = bbio->end_io;
4204                 bio->bi_bdev = (struct block_device *)
4205                                         (unsigned long)bbio->mirror_num;
4206                 /* only send an error to the higher layers if it is
4207                  * beyond the tolerance of the multi-bio
4208                  */
4209                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4210                         err = -EIO;
4211                 } else {
4212                         /*
4213                          * this bio is actually up to date, we didn't
4214                          * go over the max number of errors
4215                          */
4216                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4217                         err = 0;
4218                 }
4219                 kfree(bbio);
4220
4221                 bio_endio(bio, err);
4222         } else if (!is_orig_bio) {
4223                 bio_put(bio);
4224         }
4225 }
4226
4227 struct async_sched {
4228         struct bio *bio;
4229         int rw;
4230         struct btrfs_fs_info *info;
4231         struct btrfs_work work;
4232 };
4233
4234 /*
4235  * see run_scheduled_bios for a description of why bios are collected for
4236  * async submit.
4237  *
4238  * This will add one bio to the pending list for a device and make sure
4239  * the work struct is scheduled.
4240  */
4241 static noinline void schedule_bio(struct btrfs_root *root,
4242                                  struct btrfs_device *device,
4243                                  int rw, struct bio *bio)
4244 {
4245         int should_queue = 1;
4246         struct btrfs_pending_bios *pending_bios;
4247
4248         /* don't bother with additional async steps for reads, right now */
4249         if (!(rw & REQ_WRITE)) {
4250                 bio_get(bio);
4251                 btrfsic_submit_bio(rw, bio);
4252                 bio_put(bio);
4253                 return;
4254         }
4255
4256         /*
4257          * nr_async_bios allows us to reliably return congestion to the
4258          * higher layers.  Otherwise, the async bio makes it appear we have
4259          * made progress against dirty pages when we've really just put it
4260          * on a queue for later
4261          */
4262         atomic_inc(&root->fs_info->nr_async_bios);
4263         WARN_ON(bio->bi_next);
4264         bio->bi_next = NULL;
4265         bio->bi_rw |= rw;
4266
4267         spin_lock(&device->io_lock);
4268         if (bio->bi_rw & REQ_SYNC)
4269                 pending_bios = &device->pending_sync_bios;
4270         else
4271                 pending_bios = &device->pending_bios;
4272
4273         if (pending_bios->tail)
4274                 pending_bios->tail->bi_next = bio;
4275
4276         pending_bios->tail = bio;
4277         if (!pending_bios->head)
4278                 pending_bios->head = bio;
4279         if (device->running_pending)
4280                 should_queue = 0;
4281
4282         spin_unlock(&device->io_lock);
4283
4284         if (should_queue)
4285                 btrfs_queue_worker(&root->fs_info->submit_workers,
4286                                    &device->work);
4287 }
4288
4289 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4290                        sector_t sector)
4291 {
4292         struct bio_vec *prev;
4293         struct request_queue *q = bdev_get_queue(bdev);
4294         unsigned short max_sectors = queue_max_sectors(q);
4295         struct bvec_merge_data bvm = {
4296                 .bi_bdev = bdev,
4297                 .bi_sector = sector,
4298                 .bi_rw = bio->bi_rw,
4299         };
4300
4301         if (bio->bi_vcnt == 0) {
4302                 WARN_ON(1);
4303                 return 1;
4304         }
4305
4306         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4307         if ((bio->bi_size >> 9) > max_sectors)
4308                 return 0;
4309
4310         if (!q->merge_bvec_fn)
4311                 return 1;
4312
4313         bvm.bi_size = bio->bi_size - prev->bv_len;
4314         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4315                 return 0;
4316         return 1;
4317 }
4318
4319 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4320                               struct bio *bio, u64 physical, int dev_nr,
4321                               int rw, int async)
4322 {
4323         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4324
4325         bio->bi_private = bbio;
4326         bio->bi_private = merge_stripe_index_into_bio_private(
4327                         bio->bi_private, (unsigned int)dev_nr);
4328         bio->bi_end_io = btrfs_end_bio;
4329         bio->bi_sector = physical >> 9;
4330 #ifdef DEBUG
4331         {
4332                 struct rcu_string *name;
4333
4334                 rcu_read_lock();
4335                 name = rcu_dereference(dev->name);
4336                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4337                          "(%s id %llu), size=%u\n", rw,
4338                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4339                          name->str, dev->devid, bio->bi_size);
4340                 rcu_read_unlock();
4341         }
4342 #endif
4343         bio->bi_bdev = dev->bdev;
4344         if (async)
4345                 schedule_bio(root, dev, rw, bio);
4346         else
4347                 btrfsic_submit_bio(rw, bio);
4348 }
4349
4350 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4351                               struct bio *first_bio, struct btrfs_device *dev,
4352                               int dev_nr, int rw, int async)
4353 {
4354         struct bio_vec *bvec = first_bio->bi_io_vec;
4355         struct bio *bio;
4356         int nr_vecs = bio_get_nr_vecs(dev->bdev);
4357         u64 physical = bbio->stripes[dev_nr].physical;
4358
4359 again:
4360         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4361         if (!bio)
4362                 return -ENOMEM;
4363
4364         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4365                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4366                                  bvec->bv_offset) < bvec->bv_len) {
4367                         u64 len = bio->bi_size;
4368
4369                         atomic_inc(&bbio->stripes_pending);
4370                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4371                                           rw, async);
4372                         physical += len;
4373                         goto again;
4374                 }
4375                 bvec++;
4376         }
4377
4378         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4379         return 0;
4380 }
4381
4382 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4383 {
4384         atomic_inc(&bbio->error);
4385         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4386                 bio->bi_private = bbio->private;
4387                 bio->bi_end_io = bbio->end_io;
4388                 bio->bi_bdev = (struct block_device *)
4389                         (unsigned long)bbio->mirror_num;
4390                 bio->bi_sector = logical >> 9;
4391                 kfree(bbio);
4392                 bio_endio(bio, -EIO);
4393         }
4394 }
4395
4396 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4397                   int mirror_num, int async_submit)
4398 {
4399         struct btrfs_device *dev;
4400         struct bio *first_bio = bio;
4401         u64 logical = (u64)bio->bi_sector << 9;
4402         u64 length = 0;
4403         u64 map_length;
4404         int ret;
4405         int dev_nr = 0;
4406         int total_devs = 1;
4407         struct btrfs_bio *bbio = NULL;
4408
4409         length = bio->bi_size;
4410         map_length = length;
4411
4412         ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4413                               mirror_num);
4414         if (ret) /* -ENOMEM */
4415                 return ret;
4416
4417         total_devs = bbio->num_stripes;
4418         if (map_length < length) {
4419                 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4420                        "len %llu\n", (unsigned long long)logical,
4421                        (unsigned long long)length,
4422                        (unsigned long long)map_length);
4423                 BUG();
4424         }
4425
4426         bbio->orig_bio = first_bio;
4427         bbio->private = first_bio->bi_private;
4428         bbio->end_io = first_bio->bi_end_io;
4429         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4430
4431         while (dev_nr < total_devs) {
4432                 dev = bbio->stripes[dev_nr].dev;
4433                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4434                         bbio_error(bbio, first_bio, logical);
4435                         dev_nr++;
4436                         continue;
4437                 }
4438
4439                 /*
4440                  * Check and see if we're ok with this bio based on it's size
4441                  * and offset with the given device.
4442                  */
4443                 if (!bio_size_ok(dev->bdev, first_bio,
4444                                  bbio->stripes[dev_nr].physical >> 9)) {
4445                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4446                                                  dev_nr, rw, async_submit);
4447                         BUG_ON(ret);
4448                         dev_nr++;
4449                         continue;
4450                 }
4451
4452                 if (dev_nr < total_devs - 1) {
4453                         bio = bio_clone(first_bio, GFP_NOFS);
4454                         BUG_ON(!bio); /* -ENOMEM */
4455                 } else {
4456                         bio = first_bio;
4457                 }
4458
4459                 submit_stripe_bio(root, bbio, bio,
4460                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
4461                                   async_submit);
4462                 dev_nr++;
4463         }
4464         return 0;
4465 }
4466
4467 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
4468                                        u8 *uuid, u8 *fsid)
4469 {
4470         struct btrfs_device *device;
4471         struct btrfs_fs_devices *cur_devices;
4472
4473         cur_devices = fs_info->fs_devices;
4474         while (cur_devices) {
4475                 if (!fsid ||
4476                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4477                         device = __find_device(&cur_devices->devices,
4478                                                devid, uuid);
4479                         if (device)
4480                                 return device;
4481                 }
4482                 cur_devices = cur_devices->seed;
4483         }
4484         return NULL;
4485 }
4486
4487 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4488                                             u64 devid, u8 *dev_uuid)
4489 {
4490         struct btrfs_device *device;
4491         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4492
4493         device = kzalloc(sizeof(*device), GFP_NOFS);
4494         if (!device)
4495                 return NULL;
4496         list_add(&device->dev_list,
4497                  &fs_devices->devices);
4498         device->dev_root = root->fs_info->dev_root;
4499         device->devid = devid;
4500         device->work.func = pending_bios_fn;
4501         device->fs_devices = fs_devices;
4502         device->missing = 1;
4503         fs_devices->num_devices++;
4504         fs_devices->missing_devices++;
4505         spin_lock_init(&device->io_lock);
4506         INIT_LIST_HEAD(&device->dev_alloc_list);
4507         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4508         return device;
4509 }
4510
4511 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4512                           struct extent_buffer *leaf,
4513                           struct btrfs_chunk *chunk)
4514 {
4515         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4516         struct map_lookup *map;
4517         struct extent_map *em;
4518         u64 logical;
4519         u64 length;
4520         u64 devid;
4521         u8 uuid[BTRFS_UUID_SIZE];
4522         int num_stripes;
4523         int ret;
4524         int i;
4525
4526         logical = key->offset;
4527         length = btrfs_chunk_length(leaf, chunk);
4528
4529         read_lock(&map_tree->map_tree.lock);
4530         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4531         read_unlock(&map_tree->map_tree.lock);
4532
4533         /* already mapped? */
4534         if (em && em->start <= logical && em->start + em->len > logical) {
4535                 free_extent_map(em);
4536                 return 0;
4537         } else if (em) {
4538                 free_extent_map(em);
4539         }
4540
4541         em = alloc_extent_map();
4542         if (!em)
4543                 return -ENOMEM;
4544         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4545         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4546         if (!map) {
4547                 free_extent_map(em);
4548                 return -ENOMEM;
4549         }
4550
4551         em->bdev = (struct block_device *)map;
4552         em->start = logical;
4553         em->len = length;
4554         em->block_start = 0;
4555         em->block_len = em->len;
4556
4557         map->num_stripes = num_stripes;
4558         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4559         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4560         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4561         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4562         map->type = btrfs_chunk_type(leaf, chunk);
4563         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4564         for (i = 0; i < num_stripes; i++) {
4565                 map->stripes[i].physical =
4566                         btrfs_stripe_offset_nr(leaf, chunk, i);
4567                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4568                 read_extent_buffer(leaf, uuid, (unsigned long)
4569                                    btrfs_stripe_dev_uuid_nr(chunk, i),
4570                                    BTRFS_UUID_SIZE);
4571                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
4572                                                         uuid, NULL);
4573                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4574                         kfree(map);
4575                         free_extent_map(em);
4576                         return -EIO;
4577                 }
4578                 if (!map->stripes[i].dev) {
4579                         map->stripes[i].dev =
4580                                 add_missing_dev(root, devid, uuid);
4581                         if (!map->stripes[i].dev) {
4582                                 kfree(map);
4583                                 free_extent_map(em);
4584                                 return -EIO;
4585                         }
4586                 }
4587                 map->stripes[i].dev->in_fs_metadata = 1;
4588         }
4589
4590         write_lock(&map_tree->map_tree.lock);
4591         ret = add_extent_mapping(&map_tree->map_tree, em);
4592         write_unlock(&map_tree->map_tree.lock);
4593         BUG_ON(ret); /* Tree corruption */
4594         free_extent_map(em);
4595
4596         return 0;
4597 }
4598
4599 static void fill_device_from_item(struct extent_buffer *leaf,
4600                                  struct btrfs_dev_item *dev_item,
4601                                  struct btrfs_device *device)
4602 {
4603         unsigned long ptr;
4604
4605         device->devid = btrfs_device_id(leaf, dev_item);
4606         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4607         device->total_bytes = device->disk_total_bytes;
4608         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4609         device->type = btrfs_device_type(leaf, dev_item);
4610         device->io_align = btrfs_device_io_align(leaf, dev_item);
4611         device->io_width = btrfs_device_io_width(leaf, dev_item);
4612         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4613
4614         ptr = (unsigned long)btrfs_device_uuid(dev_item);
4615         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4616 }
4617
4618 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4619 {
4620         struct btrfs_fs_devices *fs_devices;
4621         int ret;
4622
4623         BUG_ON(!mutex_is_locked(&uuid_mutex));
4624
4625         fs_devices = root->fs_info->fs_devices->seed;
4626         while (fs_devices) {
4627                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4628                         ret = 0;
4629                         goto out;
4630                 }
4631                 fs_devices = fs_devices->seed;
4632         }
4633
4634         fs_devices = find_fsid(fsid);
4635         if (!fs_devices) {
4636                 ret = -ENOENT;
4637                 goto out;
4638         }
4639
4640         fs_devices = clone_fs_devices(fs_devices);
4641         if (IS_ERR(fs_devices)) {
4642                 ret = PTR_ERR(fs_devices);
4643                 goto out;
4644         }
4645
4646         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4647                                    root->fs_info->bdev_holder);
4648         if (ret) {
4649                 free_fs_devices(fs_devices);
4650                 goto out;
4651         }
4652
4653         if (!fs_devices->seeding) {
4654                 __btrfs_close_devices(fs_devices);
4655                 free_fs_devices(fs_devices);
4656                 ret = -EINVAL;
4657                 goto out;
4658         }
4659
4660         fs_devices->seed = root->fs_info->fs_devices->seed;
4661         root->fs_info->fs_devices->seed = fs_devices;
4662 out:
4663         return ret;
4664 }
4665
4666 static int read_one_dev(struct btrfs_root *root,
4667                         struct extent_buffer *leaf,
4668                         struct btrfs_dev_item *dev_item)
4669 {
4670         struct btrfs_device *device;
4671         u64 devid;
4672         int ret;
4673         u8 fs_uuid[BTRFS_UUID_SIZE];
4674         u8 dev_uuid[BTRFS_UUID_SIZE];
4675
4676         devid = btrfs_device_id(leaf, dev_item);
4677         read_extent_buffer(leaf, dev_uuid,
4678                            (unsigned long)btrfs_device_uuid(dev_item),
4679                            BTRFS_UUID_SIZE);
4680         read_extent_buffer(leaf, fs_uuid,
4681                            (unsigned long)btrfs_device_fsid(dev_item),
4682                            BTRFS_UUID_SIZE);
4683
4684         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4685                 ret = open_seed_devices(root, fs_uuid);
4686                 if (ret && !btrfs_test_opt(root, DEGRADED))
4687                         return ret;
4688         }
4689
4690         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
4691         if (!device || !device->bdev) {
4692                 if (!btrfs_test_opt(root, DEGRADED))
4693                         return -EIO;
4694
4695                 if (!device) {
4696                         printk(KERN_WARNING "warning devid %llu missing\n",
4697                                (unsigned long long)devid);
4698                         device = add_missing_dev(root, devid, dev_uuid);
4699                         if (!device)
4700                                 return -ENOMEM;
4701                 } else if (!device->missing) {
4702                         /*
4703                          * this happens when a device that was properly setup
4704                          * in the device info lists suddenly goes bad.
4705                          * device->bdev is NULL, and so we have to set
4706                          * device->missing to one here
4707                          */
4708                         root->fs_info->fs_devices->missing_devices++;
4709                         device->missing = 1;
4710                 }
4711         }
4712
4713         if (device->fs_devices != root->fs_info->fs_devices) {
4714                 BUG_ON(device->writeable);
4715                 if (device->generation !=
4716                     btrfs_device_generation(leaf, dev_item))
4717                         return -EINVAL;
4718         }
4719
4720         fill_device_from_item(leaf, dev_item, device);
4721         device->dev_root = root->fs_info->dev_root;
4722         device->in_fs_metadata = 1;
4723         if (device->writeable) {
4724                 device->fs_devices->total_rw_bytes += device->total_bytes;
4725                 spin_lock(&root->fs_info->free_chunk_lock);
4726                 root->fs_info->free_chunk_space += device->total_bytes -
4727                         device->bytes_used;
4728                 spin_unlock(&root->fs_info->free_chunk_lock);
4729         }
4730         ret = 0;
4731         return ret;
4732 }
4733
4734 int btrfs_read_sys_array(struct btrfs_root *root)
4735 {
4736         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4737         struct extent_buffer *sb;
4738         struct btrfs_disk_key *disk_key;
4739         struct btrfs_chunk *chunk;
4740         u8 *ptr;
4741         unsigned long sb_ptr;
4742         int ret = 0;
4743         u32 num_stripes;
4744         u32 array_size;
4745         u32 len = 0;
4746         u32 cur;
4747         struct btrfs_key key;
4748
4749         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4750                                           BTRFS_SUPER_INFO_SIZE);
4751         if (!sb)
4752                 return -ENOMEM;
4753         btrfs_set_buffer_uptodate(sb);
4754         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4755         /*
4756          * The sb extent buffer is artifical and just used to read the system array.
4757          * btrfs_set_buffer_uptodate() call does not properly mark all it's
4758          * pages up-to-date when the page is larger: extent does not cover the
4759          * whole page and consequently check_page_uptodate does not find all
4760          * the page's extents up-to-date (the hole beyond sb),
4761          * write_extent_buffer then triggers a WARN_ON.
4762          *
4763          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4764          * but sb spans only this function. Add an explicit SetPageUptodate call
4765          * to silence the warning eg. on PowerPC 64.
4766          */
4767         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4768                 SetPageUptodate(sb->pages[0]);
4769
4770         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4771         array_size = btrfs_super_sys_array_size(super_copy);
4772
4773         ptr = super_copy->sys_chunk_array;
4774         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4775         cur = 0;
4776
4777         while (cur < array_size) {
4778                 disk_key = (struct btrfs_disk_key *)ptr;
4779                 btrfs_disk_key_to_cpu(&key, disk_key);
4780
4781                 len = sizeof(*disk_key); ptr += len;
4782                 sb_ptr += len;
4783                 cur += len;
4784
4785                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4786                         chunk = (struct btrfs_chunk *)sb_ptr;
4787                         ret = read_one_chunk(root, &key, sb, chunk);
4788                         if (ret)
4789                                 break;
4790                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4791                         len = btrfs_chunk_item_size(num_stripes);
4792                 } else {
4793                         ret = -EIO;
4794                         break;
4795                 }
4796                 ptr += len;
4797                 sb_ptr += len;
4798                 cur += len;
4799         }
4800         free_extent_buffer(sb);
4801         return ret;
4802 }
4803
4804 int btrfs_read_chunk_tree(struct btrfs_root *root)
4805 {
4806         struct btrfs_path *path;
4807         struct extent_buffer *leaf;
4808         struct btrfs_key key;
4809         struct btrfs_key found_key;
4810         int ret;
4811         int slot;
4812
4813         root = root->fs_info->chunk_root;
4814
4815         path = btrfs_alloc_path();
4816         if (!path)
4817                 return -ENOMEM;
4818
4819         mutex_lock(&uuid_mutex);
4820         lock_chunks(root);
4821
4822         /* first we search for all of the device items, and then we
4823          * read in all of the chunk items.  This way we can create chunk
4824          * mappings that reference all of the devices that are afound
4825          */
4826         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4827         key.offset = 0;
4828         key.type = 0;
4829 again:
4830         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4831         if (ret < 0)
4832                 goto error;
4833         while (1) {
4834                 leaf = path->nodes[0];
4835                 slot = path->slots[0];
4836                 if (slot >= btrfs_header_nritems(leaf)) {
4837                         ret = btrfs_next_leaf(root, path);
4838                         if (ret == 0)
4839                                 continue;
4840                         if (ret < 0)
4841                                 goto error;
4842                         break;
4843                 }
4844                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4845                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4846                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4847                                 break;
4848                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4849                                 struct btrfs_dev_item *dev_item;
4850                                 dev_item = btrfs_item_ptr(leaf, slot,
4851                                                   struct btrfs_dev_item);
4852                                 ret = read_one_dev(root, leaf, dev_item);
4853                                 if (ret)
4854                                         goto error;
4855                         }
4856                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4857                         struct btrfs_chunk *chunk;
4858                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4859                         ret = read_one_chunk(root, &found_key, leaf, chunk);
4860                         if (ret)
4861                                 goto error;
4862                 }
4863                 path->slots[0]++;
4864         }
4865         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4866                 key.objectid = 0;
4867                 btrfs_release_path(path);
4868                 goto again;
4869         }
4870         ret = 0;
4871 error:
4872         unlock_chunks(root);
4873         mutex_unlock(&uuid_mutex);
4874
4875         btrfs_free_path(path);
4876         return ret;
4877 }
4878
4879 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
4880 {
4881         int i;
4882
4883         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4884                 btrfs_dev_stat_reset(dev, i);
4885 }
4886
4887 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
4888 {
4889         struct btrfs_key key;
4890         struct btrfs_key found_key;
4891         struct btrfs_root *dev_root = fs_info->dev_root;
4892         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4893         struct extent_buffer *eb;
4894         int slot;
4895         int ret = 0;
4896         struct btrfs_device *device;
4897         struct btrfs_path *path = NULL;
4898         int i;
4899
4900         path = btrfs_alloc_path();
4901         if (!path) {
4902                 ret = -ENOMEM;
4903                 goto out;
4904         }
4905
4906         mutex_lock(&fs_devices->device_list_mutex);
4907         list_for_each_entry(device, &fs_devices->devices, dev_list) {
4908                 int item_size;
4909                 struct btrfs_dev_stats_item *ptr;
4910
4911                 key.objectid = 0;
4912                 key.type = BTRFS_DEV_STATS_KEY;
4913                 key.offset = device->devid;
4914                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
4915                 if (ret) {
4916                         __btrfs_reset_dev_stats(device);
4917                         device->dev_stats_valid = 1;
4918                         btrfs_release_path(path);
4919                         continue;
4920                 }
4921                 slot = path->slots[0];
4922                 eb = path->nodes[0];
4923                 btrfs_item_key_to_cpu(eb, &found_key, slot);
4924                 item_size = btrfs_item_size_nr(eb, slot);
4925
4926                 ptr = btrfs_item_ptr(eb, slot,
4927                                      struct btrfs_dev_stats_item);
4928
4929                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4930                         if (item_size >= (1 + i) * sizeof(__le64))
4931                                 btrfs_dev_stat_set(device, i,
4932                                         btrfs_dev_stats_value(eb, ptr, i));
4933                         else
4934                                 btrfs_dev_stat_reset(device, i);
4935                 }
4936
4937                 device->dev_stats_valid = 1;
4938                 btrfs_dev_stat_print_on_load(device);
4939                 btrfs_release_path(path);
4940         }
4941         mutex_unlock(&fs_devices->device_list_mutex);
4942
4943 out:
4944         btrfs_free_path(path);
4945         return ret < 0 ? ret : 0;
4946 }
4947
4948 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
4949                                 struct btrfs_root *dev_root,
4950                                 struct btrfs_device *device)
4951 {
4952         struct btrfs_path *path;
4953         struct btrfs_key key;
4954         struct extent_buffer *eb;
4955         struct btrfs_dev_stats_item *ptr;
4956         int ret;
4957         int i;
4958
4959         key.objectid = 0;
4960         key.type = BTRFS_DEV_STATS_KEY;
4961         key.offset = device->devid;
4962
4963         path = btrfs_alloc_path();
4964         BUG_ON(!path);
4965         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
4966         if (ret < 0) {
4967                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
4968                               ret, rcu_str_deref(device->name));
4969                 goto out;
4970         }
4971
4972         if (ret == 0 &&
4973             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
4974                 /* need to delete old one and insert a new one */
4975                 ret = btrfs_del_item(trans, dev_root, path);
4976                 if (ret != 0) {
4977                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
4978                                       rcu_str_deref(device->name), ret);
4979                         goto out;
4980                 }
4981                 ret = 1;
4982         }
4983
4984         if (ret == 1) {
4985                 /* need to insert a new item */
4986                 btrfs_release_path(path);
4987                 ret = btrfs_insert_empty_item(trans, dev_root, path,
4988                                               &key, sizeof(*ptr));
4989                 if (ret < 0) {
4990                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
4991                                       rcu_str_deref(device->name), ret);
4992                         goto out;
4993                 }
4994         }
4995
4996         eb = path->nodes[0];
4997         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
4998         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4999                 btrfs_set_dev_stats_value(eb, ptr, i,
5000                                           btrfs_dev_stat_read(device, i));
5001         btrfs_mark_buffer_dirty(eb);
5002
5003 out:
5004         btrfs_free_path(path);
5005         return ret;
5006 }
5007
5008 /*
5009  * called from commit_transaction. Writes all changed device stats to disk.
5010  */
5011 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5012                         struct btrfs_fs_info *fs_info)
5013 {
5014         struct btrfs_root *dev_root = fs_info->dev_root;
5015         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5016         struct btrfs_device *device;
5017         int ret = 0;
5018
5019         mutex_lock(&fs_devices->device_list_mutex);
5020         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5021                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5022                         continue;
5023
5024                 ret = update_dev_stat_item(trans, dev_root, device);
5025                 if (!ret)
5026                         device->dev_stats_dirty = 0;
5027         }
5028         mutex_unlock(&fs_devices->device_list_mutex);
5029
5030         return ret;
5031 }
5032
5033 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5034 {
5035         btrfs_dev_stat_inc(dev, index);
5036         btrfs_dev_stat_print_on_error(dev);
5037 }
5038
5039 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5040 {
5041         if (!dev->dev_stats_valid)
5042                 return;
5043         printk_ratelimited_in_rcu(KERN_ERR
5044                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5045                            rcu_str_deref(dev->name),
5046                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5047                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5048                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5049                            btrfs_dev_stat_read(dev,
5050                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5051                            btrfs_dev_stat_read(dev,
5052                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5053 }
5054
5055 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5056 {
5057         int i;
5058
5059         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5060                 if (btrfs_dev_stat_read(dev, i) != 0)
5061                         break;
5062         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5063                 return; /* all values == 0, suppress message */
5064
5065         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5066                rcu_str_deref(dev->name),
5067                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5068                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5069                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5070                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5071                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5072 }
5073
5074 int btrfs_get_dev_stats(struct btrfs_root *root,
5075                         struct btrfs_ioctl_get_dev_stats *stats)
5076 {
5077         struct btrfs_device *dev;
5078         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5079         int i;
5080
5081         mutex_lock(&fs_devices->device_list_mutex);
5082         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5083         mutex_unlock(&fs_devices->device_list_mutex);
5084
5085         if (!dev) {
5086                 printk(KERN_WARNING
5087                        "btrfs: get dev_stats failed, device not found\n");
5088                 return -ENODEV;
5089         } else if (!dev->dev_stats_valid) {
5090                 printk(KERN_WARNING
5091                        "btrfs: get dev_stats failed, not yet valid\n");
5092                 return -ENODEV;
5093         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5094                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5095                         if (stats->nr_items > i)
5096                                 stats->values[i] =
5097                                         btrfs_dev_stat_read_and_reset(dev, i);
5098                         else
5099                                 btrfs_dev_stat_reset(dev, i);
5100                 }
5101         } else {
5102                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5103                         if (stats->nr_items > i)
5104                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
5105         }
5106         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5107                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5108         return 0;
5109 }
5110
5111 int btrfs_scratch_superblock(struct btrfs_device *device)
5112 {
5113         struct buffer_head *bh;
5114         struct btrfs_super_block *disk_super;
5115
5116         bh = btrfs_read_dev_super(device->bdev);
5117         if (!bh)
5118                 return -EINVAL;
5119         disk_super = (struct btrfs_super_block *)bh->b_data;
5120
5121         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5122         set_buffer_dirty(bh);
5123         sync_dirty_buffer(bh);
5124         brelse(bh);
5125
5126         return 0;
5127 }