Pileus Git - ~andy/linux/blob - 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/buffer_head.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/random.h>
  23 #include <asm/div64.h>
  24 #include "ctree.h"
  25 #include "extent_map.h"
  26 #include "disk-io.h"
  27 #include "transaction.h"
  28 #include "print-tree.h"
  29 #include "volumes.h"
  30
  31 struct map_lookup {
  32         u64 type;
  33         int io_align;
  34         int io_width;
  35         int stripe_len;
  36         int sector_size;
  37         int num_stripes;
  38         int sub_stripes;
  39         struct btrfs_bio_stripe stripes[];
  40 };
  41
  42 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  43                             (sizeof(struct btrfs_bio_stripe) * (n)))
  44
  45 static DEFINE_MUTEX(uuid_mutex);
  46 static LIST_HEAD(fs_uuids);
  47
  48 void btrfs_lock_volumes(void)
  49 {
  50         mutex_lock(&uuid_mutex);
  51 }
  52
  53 void btrfs_unlock_volumes(void)
  54 {
  55         mutex_unlock(&uuid_mutex);
  56 }
  57
  58 int btrfs_cleanup_fs_uuids(void)
  59 {
  60         struct btrfs_fs_devices *fs_devices;
  61         struct list_head *uuid_cur;
  62         struct list_head *devices_cur;
  63         struct btrfs_device *dev;
  64
  65         list_for_each(uuid_cur, &fs_uuids) {
  66                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  67                                         list);
  68                 while(!list_empty(&fs_devices->devices)) {
  69                         devices_cur = fs_devices->devices.next;
  70                         dev = list_entry(devices_cur, struct btrfs_device,
  71                                          dev_list);
  72                         if (dev->bdev) {
  73                                 close_bdev_excl(dev->bdev);
  74                         }
  75                         list_del(&dev->dev_list);
  76                         kfree(dev);
  77                 }
  78         }
  79         return 0;
  80 }
  81
  82 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  83                                           u8 *uuid)
  84 {
  85         struct btrfs_device *dev;
  86         struct list_head *cur;
  87
  88         list_for_each(cur, head) {
  89                 dev = list_entry(cur, struct btrfs_device, dev_list);
  90                 if (dev->devid == devid &&
  91                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  92                         return dev;
  93                 }
  94         }
  95         return NULL;
  96 }
  97
  98 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  99 {
 100         struct list_head *cur;
 101         struct btrfs_fs_devices *fs_devices;
 102
 103         list_for_each(cur, &fs_uuids) {
 104                 fs_devices = list_entry(cur, struct btrfs_fs_devices, 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 device_list_add(const char *path,
 112                            struct btrfs_super_block *disk_super,
 113                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 114 {
 115         struct btrfs_device *device;
 116         struct btrfs_fs_devices *fs_devices;
 117         u64 found_transid = btrfs_super_generation(disk_super);
 118
 119         fs_devices = find_fsid(disk_super->fsid);
 120         if (!fs_devices) {
 121                 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
 122                 if (!fs_devices)
 123                         return -ENOMEM;
 124                 INIT_LIST_HEAD(&fs_devices->devices);
 125                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 126                 list_add(&fs_devices->list, &fs_uuids);
 127                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 128                 fs_devices->latest_devid = devid;
 129                 fs_devices->latest_trans = found_transid;
 130                 fs_devices->lowest_devid = (u64)-1;
 131                 fs_devices->num_devices = 0;
 132                 device = NULL;
 133         } else {
 134                 device = __find_device(&fs_devices->devices, devid,
 135                                        disk_super->dev_item.uuid);
 136         }
 137         if (!device) {
 138                 device = kzalloc(sizeof(*device), GFP_NOFS);
 139                 if (!device) {
 140                         /* we can safely leave the fs_devices entry around */
 141                         return -ENOMEM;
 142                 }
 143                 device->devid = devid;
 144                 memcpy(device->uuid, disk_super->dev_item.uuid,
 145                        BTRFS_UUID_SIZE);
 146                 device->barriers = 1;
 147                 spin_lock_init(&device->io_lock);
 148                 device->name = kstrdup(path, GFP_NOFS);
 149                 if (!device->name) {
 150                         kfree(device);
 151                         return -ENOMEM;
 152                 }
 153                 list_add(&device->dev_list, &fs_devices->devices);
 154                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 155                 fs_devices->num_devices++;
 156         }
 157
 158         if (found_transid > fs_devices->latest_trans) {
 159                 fs_devices->latest_devid = devid;
 160                 fs_devices->latest_trans = found_transid;
 161         }
 162         if (fs_devices->lowest_devid > devid) {
 163                 fs_devices->lowest_devid = devid;
 164         }
 165         *fs_devices_ret = fs_devices;
 166         return 0;
 167 }
 168
 169 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 170 {
 171         struct list_head *head = &fs_devices->devices;
 172         struct list_head *cur;
 173         struct btrfs_device *device;
 174
 175         mutex_lock(&uuid_mutex);
 176         list_for_each(cur, head) {
 177                 device = list_entry(cur, struct btrfs_device, dev_list);
 178                 if (device->bdev) {
 179                         close_bdev_excl(device->bdev);
 180                 }
 181                 device->bdev = NULL;
 182         }
 183         mutex_unlock(&uuid_mutex);
 184         return 0;
 185 }
 186
 187 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 188                        int flags, void *holder)
 189 {
 190         struct block_device *bdev;
 191         struct list_head *head = &fs_devices->devices;
 192         struct list_head *cur;
 193         struct btrfs_device *device;
 194         int ret;
 195
 196         mutex_lock(&uuid_mutex);
 197         list_for_each(cur, head) {
 198                 device = list_entry(cur, struct btrfs_device, dev_list);
 199                 if (device->bdev)
 200                         continue;
 201
 202                 bdev = open_bdev_excl(device->name, flags, holder);
 203
 204                 if (IS_ERR(bdev)) {
 205                         printk("open %s failed\n", device->name);
 206                         ret = PTR_ERR(bdev);
 207                         goto fail;
 208                 }
 209                 set_blocksize(bdev, 4096);
 210                 if (device->devid == fs_devices->latest_devid)
 211                         fs_devices->latest_bdev = bdev;
 212                 if (device->devid == fs_devices->lowest_devid) {
 213                         fs_devices->lowest_bdev = bdev;
 214                 }
 215                 device->bdev = bdev;
 216
 217         }
 218         mutex_unlock(&uuid_mutex);
 219         return 0;
 220 fail:
 221         mutex_unlock(&uuid_mutex);
 222         btrfs_close_devices(fs_devices);
 223         return ret;
 224 }
 225
 226 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 227                           struct btrfs_fs_devices **fs_devices_ret)
 228 {
 229         struct btrfs_super_block *disk_super;
 230         struct block_device *bdev;
 231         struct buffer_head *bh;
 232         int ret;
 233         u64 devid;
 234         u64 transid;
 235
 236         mutex_lock(&uuid_mutex);
 237
 238         bdev = open_bdev_excl(path, flags, holder);
 239
 240         if (IS_ERR(bdev)) {
 241                 ret = PTR_ERR(bdev);
 242                 goto error;
 243         }
 244
 245         ret = set_blocksize(bdev, 4096);
 246         if (ret)
 247                 goto error_close;
 248         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 249         if (!bh) {
 250                 ret = -EIO;
 251                 goto error_close;
 252         }
 253         disk_super = (struct btrfs_super_block *)bh->b_data;
 254         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 255             sizeof(disk_super->magic))) {
 256                 ret = -EINVAL;
 257                 goto error_brelse;
 258         }
 259         devid = le64_to_cpu(disk_super->dev_item.devid);
 260         transid = btrfs_super_generation(disk_super);
 261         if (disk_super->label[0])
 262                 printk("device label %s ", disk_super->label);
 263         else {
 264                 /* FIXME, make a readl uuid parser */
 265                 printk("device fsid %llx-%llx ",
 266                        *(unsigned long long *)disk_super->fsid,
 267                        *(unsigned long long *)(disk_super->fsid + 8));
 268         }
 269         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 270         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 271
 272 error_brelse:
 273         brelse(bh);
 274 error_close:
 275         close_bdev_excl(bdev);
 276 error:
 277         mutex_unlock(&uuid_mutex);
 278         return ret;
 279 }
 280
 281 /*
 282  * this uses a pretty simple search, the expectation is that it is
 283  * called very infrequently and that a given device has a small number
 284  * of extents
 285  */
 286 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 287                                 struct btrfs_device *device,
 288                                 struct btrfs_path *path,
 289                                 u64 num_bytes, u64 *start)
 290 {
 291         struct btrfs_key key;
 292         struct btrfs_root *root = device->dev_root;
 293         struct btrfs_dev_extent *dev_extent = NULL;
 294         u64 hole_size = 0;
 295         u64 last_byte = 0;
 296         u64 search_start = 0;
 297         u64 search_end = device->total_bytes;
 298         int ret;
 299         int slot = 0;
 300         int start_found;
 301         struct extent_buffer *l;
 302
 303         start_found = 0;
 304         path->reada = 2;
 305
 306         /* FIXME use last free of some kind */
 307
 308         /* we don't want to overwrite the superblock on the drive,
 309          * so we make sure to start at an offset of at least 1MB
 310          */
 311         search_start = max((u64)1024 * 1024, search_start);
 312
 313         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 314                 search_start = max(root->fs_info->alloc_start, search_start);
 315
 316         key.objectid = device->devid;
 317         key.offset = search_start;
 318         key.type = BTRFS_DEV_EXTENT_KEY;
 319         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 320         if (ret < 0)
 321                 goto error;
 322         ret = btrfs_previous_item(root, path, 0, key.type);
 323         if (ret < 0)
 324                 goto error;
 325         l = path->nodes[0];
 326         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 327         while (1) {
 328                 l = path->nodes[0];
 329                 slot = path->slots[0];
 330                 if (slot >= btrfs_header_nritems(l)) {
 331                         ret = btrfs_next_leaf(root, path);
 332                         if (ret == 0)
 333                                 continue;
 334                         if (ret < 0)
 335                                 goto error;
 336 no_more_items:
 337                         if (!start_found) {
 338                                 if (search_start >= search_end) {
 339                                         ret = -ENOSPC;
 340                                         goto error;
 341                                 }
 342                                 *start = search_start;
 343                                 start_found = 1;
 344                                 goto check_pending;
 345                         }
 346                         *start = last_byte > search_start ?
 347                                 last_byte : search_start;
 348                         if (search_end <= *start) {
 349                                 ret = -ENOSPC;
 350                                 goto error;
 351                         }
 352                         goto check_pending;
 353                 }
 354                 btrfs_item_key_to_cpu(l, &key, slot);
 355
 356                 if (key.objectid < device->devid)
 357                         goto next;
 358
 359                 if (key.objectid > device->devid)
 360                         goto no_more_items;
 361
 362                 if (key.offset >= search_start && key.offset > last_byte &&
 363                     start_found) {
 364                         if (last_byte < search_start)
 365                                 last_byte = search_start;
 366                         hole_size = key.offset - last_byte;
 367                         if (key.offset > last_byte &&
 368                             hole_size >= num_bytes) {
 369                                 *start = last_byte;
 370                                 goto check_pending;
 371                         }
 372                 }
 373                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 374                         goto next;
 375                 }
 376
 377                 start_found = 1;
 378                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 379                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 380 next:
 381                 path->slots[0]++;
 382                 cond_resched();
 383         }
 384 check_pending:
 385         /* we have to make sure we didn't find an extent that has already
 386          * been allocated by the map tree or the original allocation
 387          */
 388         btrfs_release_path(root, path);
 389         BUG_ON(*start < search_start);
 390
 391         if (*start + num_bytes > search_end) {
 392                 ret = -ENOSPC;
 393                 goto error;
 394         }
 395         /* check for pending inserts here */
 396         return 0;
 397
 398 error:
 399         btrfs_release_path(root, path);
 400         return ret;
 401 }
 402
 403 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 404                           struct btrfs_device *device,
 405                           u64 start)
 406 {
 407         int ret;
 408         struct btrfs_path *path;
 409         struct btrfs_root *root = device->dev_root;
 410         struct btrfs_key key;
 411         struct btrfs_key found_key;
 412         struct extent_buffer *leaf = NULL;
 413         struct btrfs_dev_extent *extent = NULL;
 414
 415         path = btrfs_alloc_path();
 416         if (!path)
 417                 return -ENOMEM;
 418
 419         key.objectid = device->devid;
 420         key.offset = start;
 421         key.type = BTRFS_DEV_EXTENT_KEY;
 422
 423         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 424         if (ret > 0) {
 425                 ret = btrfs_previous_item(root, path, key.objectid,
 426                                           BTRFS_DEV_EXTENT_KEY);
 427                 BUG_ON(ret);
 428                 leaf = path->nodes[0];
 429                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 430                 extent = btrfs_item_ptr(leaf, path->slots[0],
 431                                         struct btrfs_dev_extent);
 432                 BUG_ON(found_key.offset > start || found_key.offset +
 433                        btrfs_dev_extent_length(leaf, extent) < start);
 434                 ret = 0;
 435         } else if (ret == 0) {
 436                 leaf = path->nodes[0];
 437                 extent = btrfs_item_ptr(leaf, path->slots[0],
 438                                         struct btrfs_dev_extent);
 439         }
 440         BUG_ON(ret);
 441
 442         device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 443         ret = btrfs_del_item(trans, root, path);
 444         BUG_ON(ret);
 445
 446         btrfs_free_path(path);
 447         return ret;
 448 }
 449
 450 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 451                            struct btrfs_device *device,
 452                            u64 chunk_tree, u64 chunk_objectid,
 453                            u64 chunk_offset,
 454                            u64 num_bytes, u64 *start)
 455 {
 456         int ret;
 457         struct btrfs_path *path;
 458         struct btrfs_root *root = device->dev_root;
 459         struct btrfs_dev_extent *extent;
 460         struct extent_buffer *leaf;
 461         struct btrfs_key key;
 462
 463         path = btrfs_alloc_path();
 464         if (!path)
 465                 return -ENOMEM;
 466
 467         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 468         if (ret) {
 469                 goto err;
 470         }
 471
 472         key.objectid = device->devid;
 473         key.offset = *start;
 474         key.type = BTRFS_DEV_EXTENT_KEY;
 475         ret = btrfs_insert_empty_item(trans, root, path, &key,
 476                                       sizeof(*extent));
 477         BUG_ON(ret);
 478
 479         leaf = path->nodes[0];
 480         extent = btrfs_item_ptr(leaf, path->slots[0],
 481                                 struct btrfs_dev_extent);
 482         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 483         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 484         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 485
 486         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 487                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 488                     BTRFS_UUID_SIZE);
 489
 490         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 491         btrfs_mark_buffer_dirty(leaf);
 492 err:
 493         btrfs_free_path(path);
 494         return ret;
 495 }
 496
 497 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 498 {
 499         struct btrfs_path *path;
 500         int ret;
 501         struct btrfs_key key;
 502         struct btrfs_chunk *chunk;
 503         struct btrfs_key found_key;
 504
 505         path = btrfs_alloc_path();
 506         BUG_ON(!path);
 507
 508         key.objectid = objectid;
 509         key.offset = (u64)-1;
 510         key.type = BTRFS_CHUNK_ITEM_KEY;
 511
 512         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 513         if (ret < 0)
 514                 goto error;
 515
 516         BUG_ON(ret == 0);
 517
 518         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 519         if (ret) {
 520                 *offset = 0;
 521         } else {
 522                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 523                                       path->slots[0]);
 524                 if (found_key.objectid != objectid)
 525                         *offset = 0;
 526                 else {
 527                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 528                                                struct btrfs_chunk);
 529                         *offset = found_key.offset +
 530                                 btrfs_chunk_length(path->nodes[0], chunk);
 531                 }
 532         }
 533         ret = 0;
 534 error:
 535         btrfs_free_path(path);
 536         return ret;
 537 }
 538
 539 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 540                            u64 *objectid)
 541 {
 542         int ret;
 543         struct btrfs_key key;
 544         struct btrfs_key found_key;
 545
 546         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 547         key.type = BTRFS_DEV_ITEM_KEY;
 548         key.offset = (u64)-1;
 549
 550         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 551         if (ret < 0)
 552                 goto error;
 553
 554         BUG_ON(ret == 0);
 555
 556         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 557                                   BTRFS_DEV_ITEM_KEY);
 558         if (ret) {
 559                 *objectid = 1;
 560         } else {
 561                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 562                                       path->slots[0]);
 563                 *objectid = found_key.offset + 1;
 564         }
 565         ret = 0;
 566 error:
 567         btrfs_release_path(root, path);
 568         return ret;
 569 }
 570
 571 /*
 572  * the device information is stored in the chunk root
 573  * the btrfs_device struct should be fully filled in
 574  */
 575 int btrfs_add_device(struct btrfs_trans_handle *trans,
 576                      struct btrfs_root *root,
 577                      struct btrfs_device *device)
 578 {
 579         int ret;
 580         struct btrfs_path *path;
 581         struct btrfs_dev_item *dev_item;
 582         struct extent_buffer *leaf;
 583         struct btrfs_key key;
 584         unsigned long ptr;
 585         u64 free_devid = 0;
 586
 587         root = root->fs_info->chunk_root;
 588
 589         path = btrfs_alloc_path();
 590         if (!path)
 591                 return -ENOMEM;
 592
 593         ret = find_next_devid(root, path, &free_devid);
 594         if (ret)
 595                 goto out;
 596
 597         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 598         key.type = BTRFS_DEV_ITEM_KEY;
 599         key.offset = free_devid;
 600
 601         ret = btrfs_insert_empty_item(trans, root, path, &key,
 602                                       sizeof(*dev_item));
 603         if (ret)
 604                 goto out;
 605
 606         leaf = path->nodes[0];
 607         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 608
 609         device->devid = free_devid;
 610         btrfs_set_device_id(leaf, dev_item, device->devid);
 611         btrfs_set_device_type(leaf, dev_item, device->type);
 612         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 613         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 614         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 615         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 616         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 617         btrfs_set_device_group(leaf, dev_item, 0);
 618         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 619         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 620
 621         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 622         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 623         btrfs_mark_buffer_dirty(leaf);
 624         ret = 0;
 625
 626 out:
 627         btrfs_free_path(path);
 628         return ret;
 629 }
 630
 631 static int btrfs_rm_dev_item(struct btrfs_root *root,
 632                              struct btrfs_device *device)
 633 {
 634         int ret;
 635         struct btrfs_path *path;
 636         struct block_device *bdev = device->bdev;
 637         struct btrfs_device *next_dev;
 638         struct btrfs_key key;
 639         u64 total_bytes;
 640         struct btrfs_fs_devices *fs_devices;
 641         struct btrfs_trans_handle *trans;
 642
 643         root = root->fs_info->chunk_root;
 644
 645         path = btrfs_alloc_path();
 646         if (!path)
 647                 return -ENOMEM;
 648
 649         trans = btrfs_start_transaction(root, 1);
 650         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 651         key.type = BTRFS_DEV_ITEM_KEY;
 652         key.offset = device->devid;
 653
 654         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 655         if (ret < 0)
 656                 goto out;
 657
 658         if (ret > 0) {
 659                 ret = -ENOENT;
 660                 goto out;
 661         }
 662
 663         ret = btrfs_del_item(trans, root, path);
 664         if (ret)
 665                 goto out;
 666
 667         /*
 668          * at this point, the device is zero sized.  We want to
 669          * remove it from the devices list and zero out the old super
 670          */
 671         list_del_init(&device->dev_list);
 672         list_del_init(&device->dev_alloc_list);
 673         fs_devices = root->fs_info->fs_devices;
 674
 675         next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
 676                               dev_list);
 677         if (bdev == fs_devices->lowest_bdev)
 678                 fs_devices->lowest_bdev = next_dev->bdev;
 679         if (bdev == root->fs_info->sb->s_bdev)
 680                 root->fs_info->sb->s_bdev = next_dev->bdev;
 681         if (bdev == fs_devices->latest_bdev)
 682                 fs_devices->latest_bdev = next_dev->bdev;
 683
 684         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 685         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 686                                     total_bytes - device->total_bytes);
 687
 688         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 689         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 690                                     total_bytes - 1);
 691 out:
 692         btrfs_free_path(path);
 693         btrfs_commit_transaction(trans, root);
 694         return ret;
 695 }
 696
 697 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 698 {
 699         struct btrfs_device *device;
 700         struct block_device *bdev;
 701         struct buffer_head *bh;
 702         struct btrfs_super_block *disk_super;
 703         u64 all_avail;
 704         u64 devid;
 705         int ret = 0;
 706
 707         mutex_lock(&root->fs_info->fs_mutex);
 708         mutex_lock(&uuid_mutex);
 709
 710         all_avail = root->fs_info->avail_data_alloc_bits |
 711                 root->fs_info->avail_system_alloc_bits |
 712                 root->fs_info->avail_metadata_alloc_bits;
 713
 714         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 715             root->fs_info->fs_devices->num_devices <= 4) {
 716                 printk("btrfs: unable to go below four devices on raid10\n");
 717                 ret = -EINVAL;
 718                 goto out;
 719         }
 720
 721         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 722             root->fs_info->fs_devices->num_devices <= 2) {
 723                 printk("btrfs: unable to go below two devices on raid1\n");
 724                 ret = -EINVAL;
 725                 goto out;
 726         }
 727
 728         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
 729         if (IS_ERR(bdev)) {
 730                 ret = PTR_ERR(bdev);
 731                 goto out;
 732         }
 733
 734         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 735         if (!bh) {
 736                 ret = -EIO;
 737                 goto error_close;
 738         }
 739         disk_super = (struct btrfs_super_block *)bh->b_data;
 740         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 741             sizeof(disk_super->magic))) {
 742                 ret = -ENOENT;
 743                 goto error_brelse;
 744         }
 745         if (memcmp(disk_super->fsid, root->fs_info->fsid, BTRFS_FSID_SIZE)) {
 746                 ret = -ENOENT;
 747                 goto error_brelse;
 748         }
 749         devid = le64_to_cpu(disk_super->dev_item.devid);
 750         device = btrfs_find_device(root, devid, NULL);
 751         if (!device) {
 752                 ret = -ENOENT;
 753                 goto error_brelse;
 754         }
 755
 756         root->fs_info->fs_devices->num_devices--;
 757
 758         ret = btrfs_shrink_device(device, 0);
 759         if (ret)
 760                 goto error_brelse;
 761
 762
 763         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
 764         if (ret)
 765                 goto error_brelse;
 766
 767         /* make sure this device isn't detected as part of the FS anymore */
 768         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
 769         set_buffer_dirty(bh);
 770         sync_dirty_buffer(bh);
 771
 772         brelse(bh);
 773
 774         /* one close for the device struct or super_block */
 775         close_bdev_excl(device->bdev);
 776
 777         /* one close for us */
 778         close_bdev_excl(device->bdev);
 779
 780         kfree(device->name);
 781         kfree(device);
 782         ret = 0;
 783         goto out;
 784
 785 error_brelse:
 786         brelse(bh);
 787 error_close:
 788         close_bdev_excl(bdev);
 789 out:
 790         mutex_unlock(&uuid_mutex);
 791         mutex_unlock(&root->fs_info->fs_mutex);
 792         return ret;
 793 }
 794
 795 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 796 {
 797         struct btrfs_trans_handle *trans;
 798         struct btrfs_device *device;
 799         struct block_device *bdev;
 800         struct list_head *cur;
 801         struct list_head *devices;
 802         u64 total_bytes;
 803         int ret = 0;
 804
 805
 806         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
 807         if (!bdev) {
 808                 return -EIO;
 809         }
 810         mutex_lock(&root->fs_info->fs_mutex);
 811         trans = btrfs_start_transaction(root, 1);
 812         devices = &root->fs_info->fs_devices->devices;
 813         list_for_each(cur, devices) {
 814                 device = list_entry(cur, struct btrfs_device, dev_list);
 815                 if (device->bdev == bdev) {
 816                         ret = -EEXIST;
 817                         goto out;
 818                 }
 819         }
 820
 821         device = kzalloc(sizeof(*device), GFP_NOFS);
 822         if (!device) {
 823                 /* we can safely leave the fs_devices entry around */
 824                 ret = -ENOMEM;
 825                 goto out_close_bdev;
 826         }
 827
 828         device->barriers = 1;
 829         generate_random_uuid(device->uuid);
 830         spin_lock_init(&device->io_lock);
 831         device->name = kstrdup(device_path, GFP_NOFS);
 832         if (!device->name) {
 833                 kfree(device);
 834                 goto out_close_bdev;
 835         }
 836         device->io_width = root->sectorsize;
 837         device->io_align = root->sectorsize;
 838         device->sector_size = root->sectorsize;
 839         device->total_bytes = i_size_read(bdev->bd_inode);
 840         device->dev_root = root->fs_info->dev_root;
 841         device->bdev = bdev;
 842
 843         ret = btrfs_add_device(trans, root, device);
 844         if (ret)
 845                 goto out_close_bdev;
 846
 847         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 848         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 849                                     total_bytes + device->total_bytes);
 850
 851         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 852         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 853                                     total_bytes + 1);
 854
 855         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
 856         list_add(&device->dev_alloc_list,
 857                  &root->fs_info->fs_devices->alloc_list);
 858         root->fs_info->fs_devices->num_devices++;
 859 out:
 860         btrfs_end_transaction(trans, root);
 861         mutex_unlock(&root->fs_info->fs_mutex);
 862         return ret;
 863
 864 out_close_bdev:
 865         close_bdev_excl(bdev);
 866         goto out;
 867 }
 868
 869 int btrfs_update_device(struct btrfs_trans_handle *trans,
 870                         struct btrfs_device *device)
 871 {
 872         int ret;
 873         struct btrfs_path *path;
 874         struct btrfs_root *root;
 875         struct btrfs_dev_item *dev_item;
 876         struct extent_buffer *leaf;
 877         struct btrfs_key key;
 878
 879         root = device->dev_root->fs_info->chunk_root;
 880
 881         path = btrfs_alloc_path();
 882         if (!path)
 883                 return -ENOMEM;
 884
 885         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 886         key.type = BTRFS_DEV_ITEM_KEY;
 887         key.offset = device->devid;
 888
 889         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 890         if (ret < 0)
 891                 goto out;
 892
 893         if (ret > 0) {
 894                 ret = -ENOENT;
 895                 goto out;
 896         }
 897
 898         leaf = path->nodes[0];
 899         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 900
 901         btrfs_set_device_id(leaf, dev_item, device->devid);
 902         btrfs_set_device_type(leaf, dev_item, device->type);
 903         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 904         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 905         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 906         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 907         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 908         btrfs_mark_buffer_dirty(leaf);
 909
 910 out:
 911         btrfs_free_path(path);
 912         return ret;
 913 }
 914
 915 int btrfs_grow_device(struct btrfs_trans_handle *trans,
 916                       struct btrfs_device *device, u64 new_size)
 917 {
 918         struct btrfs_super_block *super_copy =
 919                 &device->dev_root->fs_info->super_copy;
 920         u64 old_total = btrfs_super_total_bytes(super_copy);
 921         u64 diff = new_size - device->total_bytes;
 922
 923         btrfs_set_super_total_bytes(super_copy, old_total + diff);
 924         return btrfs_update_device(trans, device);
 925 }
 926
 927 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
 928                             struct btrfs_root *root,
 929                             u64 chunk_tree, u64 chunk_objectid,
 930                             u64 chunk_offset)
 931 {
 932         int ret;
 933         struct btrfs_path *path;
 934         struct btrfs_key key;
 935
 936         root = root->fs_info->chunk_root;
 937         path = btrfs_alloc_path();
 938         if (!path)
 939                 return -ENOMEM;
 940
 941         key.objectid = chunk_objectid;
 942         key.offset = chunk_offset;
 943         key.type = BTRFS_CHUNK_ITEM_KEY;
 944
 945         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 946         BUG_ON(ret);
 947
 948         ret = btrfs_del_item(trans, root, path);
 949         BUG_ON(ret);
 950
 951         btrfs_free_path(path);
 952         return 0;
 953 }
 954
 955 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
 956                         chunk_offset)
 957 {
 958         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 959         struct btrfs_disk_key *disk_key;
 960         struct btrfs_chunk *chunk;
 961         u8 *ptr;
 962         int ret = 0;
 963         u32 num_stripes;
 964         u32 array_size;
 965         u32 len = 0;
 966         u32 cur;
 967         struct btrfs_key key;
 968
 969         array_size = btrfs_super_sys_array_size(super_copy);
 970
 971         ptr = super_copy->sys_chunk_array;
 972         cur = 0;
 973
 974         while (cur < array_size) {
 975                 disk_key = (struct btrfs_disk_key *)ptr;
 976                 btrfs_disk_key_to_cpu(&key, disk_key);
 977
 978                 len = sizeof(*disk_key);
 979
 980                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
 981                         chunk = (struct btrfs_chunk *)(ptr + len);
 982                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
 983                         len += btrfs_chunk_item_size(num_stripes);
 984                 } else {
 985                         ret = -EIO;
 986                         break;
 987                 }
 988                 if (key.objectid == chunk_objectid &&
 989                     key.offset == chunk_offset) {
 990                         memmove(ptr, ptr + len, array_size - (cur + len));
 991                         array_size -= len;
 992                         btrfs_set_super_sys_array_size(super_copy, array_size);
 993                 } else {
 994                         ptr += len;
 995                         cur += len;
 996                 }
 997         }
 998         return ret;
 999 }
1000
1001
1002 int btrfs_relocate_chunk(struct btrfs_root *root,
1003                          u64 chunk_tree, u64 chunk_objectid,
1004                          u64 chunk_offset)
1005 {
1006         struct extent_map_tree *em_tree;
1007         struct btrfs_root *extent_root;
1008         struct btrfs_trans_handle *trans;
1009         struct extent_map *em;
1010         struct map_lookup *map;
1011         int ret;
1012         int i;
1013
1014         root = root->fs_info->chunk_root;
1015         extent_root = root->fs_info->extent_root;
1016         em_tree = &root->fs_info->mapping_tree.map_tree;
1017
1018         /* step one, relocate all the extents inside this chunk */
1019         ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1020         BUG_ON(ret);
1021
1022         trans = btrfs_start_transaction(root, 1);
1023         BUG_ON(!trans);
1024
1025         /*
1026          * step two, delete the device extents and the
1027          * chunk tree entries
1028          */
1029         spin_lock(&em_tree->lock);
1030         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1031         spin_unlock(&em_tree->lock);
1032
1033         BUG_ON(em->start > chunk_offset ||
1034                em->start + em->len < chunk_offset);
1035         map = (struct map_lookup *)em->bdev;
1036
1037         for (i = 0; i < map->num_stripes; i++) {
1038                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1039                                             map->stripes[i].physical);
1040                 BUG_ON(ret);
1041
1042                 ret = btrfs_update_device(trans, map->stripes[i].dev);
1043                 BUG_ON(ret);
1044         }
1045         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1046                                chunk_offset);
1047
1048         BUG_ON(ret);
1049
1050         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1051                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1052                 BUG_ON(ret);
1053         }
1054
1055         spin_lock(&em_tree->lock);
1056         remove_extent_mapping(em_tree, em);
1057         kfree(map);
1058         em->bdev = NULL;
1059
1060         /* once for the tree */
1061         free_extent_map(em);
1062         spin_unlock(&em_tree->lock);
1063
1064         /* once for us */
1065         free_extent_map(em);
1066
1067         btrfs_end_transaction(trans, root);
1068         return 0;
1069 }
1070
1071 static u64 div_factor(u64 num, int factor)
1072 {
1073         if (factor == 10)
1074                 return num;
1075         num *= factor;
1076         do_div(num, 10);
1077         return num;
1078 }
1079
1080
1081 int btrfs_balance(struct btrfs_root *dev_root)
1082 {
1083         int ret;
1084         struct list_head *cur;
1085         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1086         struct btrfs_device *device;
1087         u64 old_size;
1088         u64 size_to_free;
1089         struct btrfs_path *path;
1090         struct btrfs_key key;
1091         struct btrfs_chunk *chunk;
1092         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1093         struct btrfs_trans_handle *trans;
1094         struct btrfs_key found_key;
1095
1096
1097         dev_root = dev_root->fs_info->dev_root;
1098
1099         mutex_lock(&dev_root->fs_info->fs_mutex);
1100         /* step one make some room on all the devices */
1101         list_for_each(cur, devices) {
1102                 device = list_entry(cur, struct btrfs_device, dev_list);
1103                 old_size = device->total_bytes;
1104                 size_to_free = div_factor(old_size, 1);
1105                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1106                 if (device->total_bytes - device->bytes_used > size_to_free)
1107                         continue;
1108
1109                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1110                 BUG_ON(ret);
1111
1112                 trans = btrfs_start_transaction(dev_root, 1);
1113                 BUG_ON(!trans);
1114
1115                 ret = btrfs_grow_device(trans, device, old_size);
1116                 BUG_ON(ret);
1117
1118                 btrfs_end_transaction(trans, dev_root);
1119         }
1120
1121         /* step two, relocate all the chunks */
1122         path = btrfs_alloc_path();
1123         BUG_ON(!path);
1124
1125         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1126         key.offset = (u64)-1;
1127         key.type = BTRFS_CHUNK_ITEM_KEY;
1128
1129         while(1) {
1130                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1131                 if (ret < 0)
1132                         goto error;
1133
1134                 /*
1135                  * this shouldn't happen, it means the last relocate
1136                  * failed
1137                  */
1138                 if (ret == 0)
1139                         break;
1140
1141                 ret = btrfs_previous_item(chunk_root, path, 0,
1142                                           BTRFS_CHUNK_ITEM_KEY);
1143                 if (ret) {
1144                         break;
1145                 }
1146                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1147                                       path->slots[0]);
1148                 if (found_key.objectid != key.objectid)
1149                         break;
1150                 chunk = btrfs_item_ptr(path->nodes[0],
1151                                        path->slots[0],
1152                                        struct btrfs_chunk);
1153                 key.offset = found_key.offset;
1154                 /* chunk zero is special */
1155                 if (key.offset == 0)
1156                         break;
1157
1158                 ret = btrfs_relocate_chunk(chunk_root,
1159                                            chunk_root->root_key.objectid,
1160                                            found_key.objectid,
1161                                            found_key.offset);
1162                 BUG_ON(ret);
1163                 btrfs_release_path(chunk_root, path);
1164         }
1165         ret = 0;
1166 error:
1167         btrfs_free_path(path);
1168         mutex_unlock(&dev_root->fs_info->fs_mutex);
1169         return ret;
1170 }
1171
1172 /*
1173  * shrinking a device means finding all of the device extents past
1174  * the new size, and then following the back refs to the chunks.
1175  * The chunk relocation code actually frees the device extent
1176  */
1177 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1178 {
1179         struct btrfs_trans_handle *trans;
1180         struct btrfs_root *root = device->dev_root;
1181         struct btrfs_dev_extent *dev_extent = NULL;
1182         struct btrfs_path *path;
1183         u64 length;
1184         u64 chunk_tree;
1185         u64 chunk_objectid;
1186         u64 chunk_offset;
1187         int ret;
1188         int slot;
1189         struct extent_buffer *l;
1190         struct btrfs_key key;
1191         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1192         u64 old_total = btrfs_super_total_bytes(super_copy);
1193         u64 diff = device->total_bytes - new_size;
1194
1195
1196         path = btrfs_alloc_path();
1197         if (!path)
1198                 return -ENOMEM;
1199
1200         trans = btrfs_start_transaction(root, 1);
1201         if (!trans) {
1202                 ret = -ENOMEM;
1203                 goto done;
1204         }
1205
1206         path->reada = 2;
1207
1208         device->total_bytes = new_size;
1209         ret = btrfs_update_device(trans, device);
1210         if (ret) {
1211                 btrfs_end_transaction(trans, root);
1212                 goto done;
1213         }
1214         WARN_ON(diff > old_total);
1215         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1216         btrfs_end_transaction(trans, root);
1217
1218         key.objectid = device->devid;
1219         key.offset = (u64)-1;
1220         key.type = BTRFS_DEV_EXTENT_KEY;
1221
1222         while (1) {
1223                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1224                 if (ret < 0)
1225                         goto done;
1226
1227                 ret = btrfs_previous_item(root, path, 0, key.type);
1228                 if (ret < 0)
1229                         goto done;
1230                 if (ret) {
1231                         ret = 0;
1232                         goto done;
1233                 }
1234
1235                 l = path->nodes[0];
1236                 slot = path->slots[0];
1237                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1238
1239                 if (key.objectid != device->devid)
1240                         goto done;
1241
1242                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1243                 length = btrfs_dev_extent_length(l, dev_extent);
1244
1245                 if (key.offset + length <= new_size)
1246                         goto done;
1247
1248                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1249                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1250                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1251                 btrfs_release_path(root, path);
1252
1253                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1254                                            chunk_offset);
1255                 if (ret)
1256                         goto done;
1257         }
1258
1259 done:
1260         btrfs_free_path(path);
1261         return ret;
1262 }
1263
1264 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1265                            struct btrfs_root *root,
1266                            struct btrfs_key *key,
1267                            struct btrfs_chunk *chunk, int item_size)
1268 {
1269         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1270         struct btrfs_disk_key disk_key;
1271         u32 array_size;
1272         u8 *ptr;
1273
1274         array_size = btrfs_super_sys_array_size(super_copy);
1275         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1276                 return -EFBIG;
1277
1278         ptr = super_copy->sys_chunk_array + array_size;
1279         btrfs_cpu_key_to_disk(&disk_key, key);
1280         memcpy(ptr, &disk_key, sizeof(disk_key));
1281         ptr += sizeof(disk_key);
1282         memcpy(ptr, chunk, item_size);
1283         item_size += sizeof(disk_key);
1284         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1285         return 0;
1286 }
1287
1288 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1289                                int sub_stripes)
1290 {
1291         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1292                 return calc_size;
1293         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1294                 return calc_size * (num_stripes / sub_stripes);
1295         else
1296                 return calc_size * num_stripes;
1297 }
1298
1299
1300 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1301                       struct btrfs_root *extent_root, u64 *start,
1302                       u64 *num_bytes, u64 type)
1303 {
1304         u64 dev_offset;
1305         struct btrfs_fs_info *info = extent_root->fs_info;
1306         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1307         struct btrfs_path *path;
1308         struct btrfs_stripe *stripes;
1309         struct btrfs_device *device = NULL;
1310         struct btrfs_chunk *chunk;
1311         struct list_head private_devs;
1312         struct list_head *dev_list;
1313         struct list_head *cur;
1314         struct extent_map_tree *em_tree;
1315         struct map_lookup *map;
1316         struct extent_map *em;
1317         int min_stripe_size = 1 * 1024 * 1024;
1318         u64 physical;
1319         u64 calc_size = 1024 * 1024 * 1024;
1320         u64 max_chunk_size = calc_size;
1321         u64 min_free;
1322         u64 avail;
1323         u64 max_avail = 0;
1324         u64 percent_max;
1325         int num_stripes = 1;
1326         int min_stripes = 1;
1327         int sub_stripes = 0;
1328         int looped = 0;
1329         int ret;
1330         int index;
1331         int stripe_len = 64 * 1024;
1332         struct btrfs_key key;
1333
1334         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1335             (type & BTRFS_BLOCK_GROUP_DUP)) {
1336                 WARN_ON(1);
1337                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1338         }
1339         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1340         if (list_empty(dev_list))
1341                 return -ENOSPC;
1342
1343         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1344                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1345                 min_stripes = 2;
1346         }
1347         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1348                 num_stripes = 2;
1349                 min_stripes = 2;
1350         }
1351         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1352                 num_stripes = min_t(u64, 2,
1353                                   btrfs_super_num_devices(&info->super_copy));
1354                 if (num_stripes < 2)
1355                         return -ENOSPC;
1356                 min_stripes = 2;
1357         }
1358         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1359                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1360                 if (num_stripes < 4)
1361                         return -ENOSPC;
1362                 num_stripes &= ~(u32)1;
1363                 sub_stripes = 2;
1364                 min_stripes = 4;
1365         }
1366
1367         if (type & BTRFS_BLOCK_GROUP_DATA) {
1368                 max_chunk_size = 10 * calc_size;
1369                 min_stripe_size = 64 * 1024 * 1024;
1370         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1371                 max_chunk_size = 4 * calc_size;
1372                 min_stripe_size = 32 * 1024 * 1024;
1373         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1374                 calc_size = 8 * 1024 * 1024;
1375                 max_chunk_size = calc_size * 2;
1376                 min_stripe_size = 1 * 1024 * 1024;
1377         }
1378
1379         path = btrfs_alloc_path();
1380         if (!path)
1381                 return -ENOMEM;
1382
1383         /* we don't want a chunk larger than 10% of the FS */
1384         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1385         max_chunk_size = min(percent_max, max_chunk_size);
1386
1387 again:
1388         if (calc_size * num_stripes > max_chunk_size) {
1389                 calc_size = max_chunk_size;
1390                 do_div(calc_size, num_stripes);
1391                 do_div(calc_size, stripe_len);
1392                 calc_size *= stripe_len;
1393         }
1394         /* we don't want tiny stripes */
1395         calc_size = max_t(u64, min_stripe_size, calc_size);
1396
1397         do_div(calc_size, stripe_len);
1398         calc_size *= stripe_len;
1399
1400         INIT_LIST_HEAD(&private_devs);
1401         cur = dev_list->next;
1402         index = 0;
1403
1404         if (type & BTRFS_BLOCK_GROUP_DUP)
1405                 min_free = calc_size * 2;
1406         else
1407                 min_free = calc_size;
1408
1409         /* we add 1MB because we never use the first 1MB of the device */
1410         min_free += 1024 * 1024;
1411
1412         /* build a private list of devices we will allocate from */
1413         while(index < num_stripes) {
1414                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1415
1416                 avail = device->total_bytes - device->bytes_used;
1417                 cur = cur->next;
1418
1419                 if (avail >= min_free) {
1420                         u64 ignored_start = 0;
1421                         ret = find_free_dev_extent(trans, device, path,
1422                                                    min_free,
1423                                                    &ignored_start);
1424                         if (ret == 0) {
1425                                 list_move_tail(&device->dev_alloc_list,
1426                                                &private_devs);
1427                                 index++;
1428                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1429                                         index++;
1430                         }
1431                 } else if (avail > max_avail)
1432                         max_avail = avail;
1433                 if (cur == dev_list)
1434                         break;
1435         }
1436         if (index < num_stripes) {
1437                 list_splice(&private_devs, dev_list);
1438                 if (index >= min_stripes) {
1439                         num_stripes = index;
1440                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1441                                 num_stripes /= sub_stripes;
1442                                 num_stripes *= sub_stripes;
1443                         }
1444                         looped = 1;
1445                         goto again;
1446                 }
1447                 if (!looped && max_avail > 0) {
1448                         looped = 1;
1449                         calc_size = max_avail;
1450                         goto again;
1451                 }
1452                 btrfs_free_path(path);
1453                 return -ENOSPC;
1454         }
1455         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1456         key.type = BTRFS_CHUNK_ITEM_KEY;
1457         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1458                               &key.offset);
1459         if (ret) {
1460                 btrfs_free_path(path);
1461                 return ret;
1462         }
1463
1464         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1465         if (!chunk) {
1466                 btrfs_free_path(path);
1467                 return -ENOMEM;
1468         }
1469
1470         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1471         if (!map) {
1472                 kfree(chunk);
1473                 btrfs_free_path(path);
1474                 return -ENOMEM;
1475         }
1476         btrfs_free_path(path);
1477         path = NULL;
1478
1479         stripes = &chunk->stripe;
1480         *num_bytes = chunk_bytes_by_type(type, calc_size,
1481                                          num_stripes, sub_stripes);
1482
1483         index = 0;
1484         while(index < num_stripes) {
1485                 struct btrfs_stripe *stripe;
1486                 BUG_ON(list_empty(&private_devs));
1487                 cur = private_devs.next;
1488                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1489
1490                 /* loop over this device again if we're doing a dup group */
1491                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1492                     (index == num_stripes - 1))
1493                         list_move_tail(&device->dev_alloc_list, dev_list);
1494
1495                 ret = btrfs_alloc_dev_extent(trans, device,
1496                              info->chunk_root->root_key.objectid,
1497                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1498                              calc_size, &dev_offset);
1499                 BUG_ON(ret);
1500                 device->bytes_used += calc_size;
1501                 ret = btrfs_update_device(trans, device);
1502                 BUG_ON(ret);
1503
1504                 map->stripes[index].dev = device;
1505                 map->stripes[index].physical = dev_offset;
1506                 stripe = stripes + index;
1507                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1508                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1509                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1510                 physical = dev_offset;
1511                 index++;
1512         }
1513         BUG_ON(!list_empty(&private_devs));
1514
1515         /* key was set above */
1516         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1517         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1518         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1519         btrfs_set_stack_chunk_type(chunk, type);
1520         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1521         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1522         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1523         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1524         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1525         map->sector_size = extent_root->sectorsize;
1526         map->stripe_len = stripe_len;
1527         map->io_align = stripe_len;
1528         map->io_width = stripe_len;
1529         map->type = type;
1530         map->num_stripes = num_stripes;
1531         map->sub_stripes = sub_stripes;
1532
1533         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1534                                 btrfs_chunk_item_size(num_stripes));
1535         BUG_ON(ret);
1536         *start = key.offset;;
1537
1538         em = alloc_extent_map(GFP_NOFS);
1539         if (!em)
1540                 return -ENOMEM;
1541         em->bdev = (struct block_device *)map;
1542         em->start = key.offset;
1543         em->len = *num_bytes;
1544         em->block_start = 0;
1545
1546         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1547                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1548                                     chunk, btrfs_chunk_item_size(num_stripes));
1549                 BUG_ON(ret);
1550         }
1551         kfree(chunk);
1552
1553         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1554         spin_lock(&em_tree->lock);
1555         ret = add_extent_mapping(em_tree, em);
1556         spin_unlock(&em_tree->lock);
1557         BUG_ON(ret);
1558         free_extent_map(em);
1559         return ret;
1560 }
1561
1562 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1563 {
1564         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1565 }
1566
1567 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1568 {
1569         struct extent_map *em;
1570
1571         while(1) {
1572                 spin_lock(&tree->map_tree.lock);
1573                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1574                 if (em)
1575                         remove_extent_mapping(&tree->map_tree, em);
1576                 spin_unlock(&tree->map_tree.lock);
1577                 if (!em)
1578                         break;
1579                 kfree(em->bdev);
1580                 /* once for us */
1581                 free_extent_map(em);
1582                 /* once for the tree */
1583                 free_extent_map(em);
1584         }
1585 }
1586
1587 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1588 {
1589         struct extent_map *em;
1590         struct map_lookup *map;
1591         struct extent_map_tree *em_tree = &map_tree->map_tree;
1592         int ret;
1593
1594         spin_lock(&em_tree->lock);
1595         em = lookup_extent_mapping(em_tree, logical, len);
1596         spin_unlock(&em_tree->lock);
1597         BUG_ON(!em);
1598
1599         BUG_ON(em->start > logical || em->start + em->len < logical);
1600         map = (struct map_lookup *)em->bdev;
1601         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1602                 ret = map->num_stripes;
1603         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1604                 ret = map->sub_stripes;
1605         else
1606                 ret = 1;
1607         free_extent_map(em);
1608         return ret;
1609 }
1610
1611 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1612                              u64 logical, u64 *length,
1613                              struct btrfs_multi_bio **multi_ret,
1614                              int mirror_num, struct page *unplug_page)
1615 {
1616         struct extent_map *em;
1617         struct map_lookup *map;
1618         struct extent_map_tree *em_tree = &map_tree->map_tree;
1619         u64 offset;
1620         u64 stripe_offset;
1621         u64 stripe_nr;
1622         int stripes_allocated = 8;
1623         int stripes_required = 1;
1624         int stripe_index;
1625         int i;
1626         int num_stripes;
1627         int max_errors = 0;
1628         struct btrfs_multi_bio *multi = NULL;
1629
1630         if (multi_ret && !(rw & (1 << BIO_RW))) {
1631                 stripes_allocated = 1;
1632         }
1633 again:
1634         if (multi_ret) {
1635                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1636                                 GFP_NOFS);
1637                 if (!multi)
1638                         return -ENOMEM;
1639
1640                 atomic_set(&multi->error, 0);
1641         }
1642
1643         spin_lock(&em_tree->lock);
1644         em = lookup_extent_mapping(em_tree, logical, *length);
1645         spin_unlock(&em_tree->lock);
1646
1647         if (!em && unplug_page)
1648                 return 0;
1649
1650         if (!em) {
1651                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1652                 BUG();
1653         }
1654
1655         BUG_ON(em->start > logical || em->start + em->len < logical);
1656         map = (struct map_lookup *)em->bdev;
1657         offset = logical - em->start;
1658
1659         if (mirror_num > map->num_stripes)
1660                 mirror_num = 0;
1661
1662         /* if our multi bio struct is too small, back off and try again */
1663         if (rw & (1 << BIO_RW)) {
1664                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1665                                  BTRFS_BLOCK_GROUP_DUP)) {
1666                         stripes_required = map->num_stripes;
1667                         max_errors = 1;
1668                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1669                         stripes_required = map->sub_stripes;
1670                         max_errors = 1;
1671                 }
1672         }
1673         if (multi_ret && rw == WRITE &&
1674             stripes_allocated < stripes_required) {
1675                 stripes_allocated = map->num_stripes;
1676                 free_extent_map(em);
1677                 kfree(multi);
1678                 goto again;
1679         }
1680         stripe_nr = offset;
1681         /*
1682          * stripe_nr counts the total number of stripes we have to stride
1683          * to get to this block
1684          */
1685         do_div(stripe_nr, map->stripe_len);
1686
1687         stripe_offset = stripe_nr * map->stripe_len;
1688         BUG_ON(offset < stripe_offset);
1689
1690         /* stripe_offset is the offset of this block in its stripe*/
1691         stripe_offset = offset - stripe_offset;
1692
1693         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1694                          BTRFS_BLOCK_GROUP_RAID10 |
1695                          BTRFS_BLOCK_GROUP_DUP)) {
1696                 /* we limit the length of each bio to what fits in a stripe */
1697                 *length = min_t(u64, em->len - offset,
1698                               map->stripe_len - stripe_offset);
1699         } else {
1700                 *length = em->len - offset;
1701         }
1702
1703         if (!multi_ret && !unplug_page)
1704                 goto out;
1705
1706         num_stripes = 1;
1707         stripe_index = 0;
1708         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1709                 if (unplug_page || (rw & (1 << BIO_RW)))
1710                         num_stripes = map->num_stripes;
1711                 else if (mirror_num)
1712                         stripe_index = mirror_num - 1;
1713                 else
1714                         stripe_index = current->pid % map->num_stripes;
1715
1716         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1717                 if (rw & (1 << BIO_RW))
1718                         num_stripes = map->num_stripes;
1719                 else if (mirror_num)
1720                         stripe_index = mirror_num - 1;
1721
1722         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1723                 int factor = map->num_stripes / map->sub_stripes;
1724
1725                 stripe_index = do_div(stripe_nr, factor);
1726                 stripe_index *= map->sub_stripes;
1727
1728                 if (unplug_page || (rw & (1 << BIO_RW)))
1729                         num_stripes = map->sub_stripes;
1730                 else if (mirror_num)
1731                         stripe_index += mirror_num - 1;
1732                 else
1733                         stripe_index += current->pid % map->sub_stripes;
1734         } else {
1735                 /*
1736                  * after this do_div call, stripe_nr is the number of stripes
1737                  * on this device we have to walk to find the data, and
1738                  * stripe_index is the number of our device in the stripe array
1739                  */
1740                 stripe_index = do_div(stripe_nr, map->num_stripes);
1741         }
1742         BUG_ON(stripe_index >= map->num_stripes);
1743
1744         for (i = 0; i < num_stripes; i++) {
1745                 if (unplug_page) {
1746                         struct btrfs_device *device;
1747                         struct backing_dev_info *bdi;
1748
1749                         device = map->stripes[stripe_index].dev;
1750                         bdi = blk_get_backing_dev_info(device->bdev);
1751                         if (bdi->unplug_io_fn) {
1752                                 bdi->unplug_io_fn(bdi, unplug_page);
1753                         }
1754                 } else {
1755                         multi->stripes[i].physical =
1756                                 map->stripes[stripe_index].physical +
1757                                 stripe_offset + stripe_nr * map->stripe_len;
1758                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
1759                 }
1760                 stripe_index++;
1761         }
1762         if (multi_ret) {
1763                 *multi_ret = multi;
1764                 multi->num_stripes = num_stripes;
1765                 multi->max_errors = max_errors;
1766         }
1767 out:
1768         free_extent_map(em);
1769         return 0;
1770 }
1771
1772 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1773                       u64 logical, u64 *length,
1774                       struct btrfs_multi_bio **multi_ret, int mirror_num)
1775 {
1776         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1777                                  mirror_num, NULL);
1778 }
1779
1780 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1781                       u64 logical, struct page *page)
1782 {
1783         u64 length = PAGE_CACHE_SIZE;
1784         return __btrfs_map_block(map_tree, READ, logical, &length,
1785                                  NULL, 0, page);
1786 }
1787
1788
1789 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1790 static void end_bio_multi_stripe(struct bio *bio, int err)
1791 #else
1792 static int end_bio_multi_stripe(struct bio *bio,
1793                                    unsigned int bytes_done, int err)
1794 #endif
1795 {
1796         struct btrfs_multi_bio *multi = bio->bi_private;
1797
1798 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1799         if (bio->bi_size)
1800                 return 1;
1801 #endif
1802         if (err)
1803                 atomic_inc(&multi->error);
1804
1805         if (atomic_dec_and_test(&multi->stripes_pending)) {
1806                 bio->bi_private = multi->private;
1807                 bio->bi_end_io = multi->end_io;
1808
1809                 /* only send an error to the higher layers if it is
1810                  * beyond the tolerance of the multi-bio
1811                  */
1812                 if (atomic_read(&multi->error) > multi->max_errors)
1813                         err = -EIO;
1814                 else
1815                         err = 0;
1816                 kfree(multi);
1817
1818 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1819                 bio_endio(bio, bio->bi_size, err);
1820 #else
1821                 bio_endio(bio, err);
1822 #endif
1823         } else {
1824                 bio_put(bio);
1825         }
1826 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1827         return 0;
1828 #endif
1829 }
1830
1831 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1832                   int mirror_num)
1833 {
1834         struct btrfs_mapping_tree *map_tree;
1835         struct btrfs_device *dev;
1836         struct bio *first_bio = bio;
1837         u64 logical = bio->bi_sector << 9;
1838         u64 length = 0;
1839         u64 map_length;
1840         struct btrfs_multi_bio *multi = NULL;
1841         int ret;
1842         int dev_nr = 0;
1843         int total_devs = 1;
1844
1845         length = bio->bi_size;
1846         map_tree = &root->fs_info->mapping_tree;
1847         map_length = length;
1848
1849         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1850                               mirror_num);
1851         BUG_ON(ret);
1852
1853         total_devs = multi->num_stripes;
1854         if (map_length < length) {
1855                 printk("mapping failed logical %Lu bio len %Lu "
1856                        "len %Lu\n", logical, length, map_length);
1857                 BUG();
1858         }
1859         multi->end_io = first_bio->bi_end_io;
1860         multi->private = first_bio->bi_private;
1861         atomic_set(&multi->stripes_pending, multi->num_stripes);
1862
1863         while(dev_nr < total_devs) {
1864                 if (total_devs > 1) {
1865                         if (dev_nr < total_devs - 1) {
1866                                 bio = bio_clone(first_bio, GFP_NOFS);
1867                                 BUG_ON(!bio);
1868                         } else {
1869                                 bio = first_bio;
1870                         }
1871                         bio->bi_private = multi;
1872                         bio->bi_end_io = end_bio_multi_stripe;
1873                 }
1874                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1875                 dev = multi->stripes[dev_nr].dev;
1876
1877                 bio->bi_bdev = dev->bdev;
1878                 spin_lock(&dev->io_lock);
1879                 dev->total_ios++;
1880                 spin_unlock(&dev->io_lock);
1881                 submit_bio(rw, bio);
1882                 dev_nr++;
1883         }
1884         if (total_devs == 1)
1885                 kfree(multi);
1886         return 0;
1887 }
1888
1889 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1890                                        u8 *uuid)
1891 {
1892         struct list_head *head = &root->fs_info->fs_devices->devices;
1893
1894         return __find_device(head, devid, uuid);
1895 }
1896
1897 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1898                           struct extent_buffer *leaf,
1899                           struct btrfs_chunk *chunk)
1900 {
1901         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1902         struct map_lookup *map;
1903         struct extent_map *em;
1904         u64 logical;
1905         u64 length;
1906         u64 devid;
1907         u8 uuid[BTRFS_UUID_SIZE];
1908         int num_stripes;
1909         int ret;
1910         int i;
1911
1912         logical = key->offset;
1913         length = btrfs_chunk_length(leaf, chunk);
1914
1915         spin_lock(&map_tree->map_tree.lock);
1916         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1917         spin_unlock(&map_tree->map_tree.lock);
1918
1919         /* already mapped? */
1920         if (em && em->start <= logical && em->start + em->len > logical) {
1921                 free_extent_map(em);
1922                 return 0;
1923         } else if (em) {
1924                 free_extent_map(em);
1925         }
1926
1927         map = kzalloc(sizeof(*map), GFP_NOFS);
1928         if (!map)
1929                 return -ENOMEM;
1930
1931         em = alloc_extent_map(GFP_NOFS);
1932         if (!em)
1933                 return -ENOMEM;
1934         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1935         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1936         if (!map) {
1937                 free_extent_map(em);
1938                 return -ENOMEM;
1939         }
1940
1941         em->bdev = (struct block_device *)map;
1942         em->start = logical;
1943         em->len = length;
1944         em->block_start = 0;
1945
1946         map->num_stripes = num_stripes;
1947         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1948         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1949         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1950         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1951         map->type = btrfs_chunk_type(leaf, chunk);
1952         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1953         for (i = 0; i < num_stripes; i++) {
1954                 map->stripes[i].physical =
1955                         btrfs_stripe_offset_nr(leaf, chunk, i);
1956                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1957                 read_extent_buffer(leaf, uuid, (unsigned long)
1958                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1959                                    BTRFS_UUID_SIZE);
1960                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1961                 if (!map->stripes[i].dev) {
1962                         kfree(map);
1963                         free_extent_map(em);
1964                         return -EIO;
1965                 }
1966         }
1967
1968         spin_lock(&map_tree->map_tree.lock);
1969         ret = add_extent_mapping(&map_tree->map_tree, em);
1970         spin_unlock(&map_tree->map_tree.lock);
1971         BUG_ON(ret);
1972         free_extent_map(em);
1973
1974         return 0;
1975 }
1976
1977 static int fill_device_from_item(struct extent_buffer *leaf,
1978                                  struct btrfs_dev_item *dev_item,
1979                                  struct btrfs_device *device)
1980 {
1981         unsigned long ptr;
1982
1983         device->devid = btrfs_device_id(leaf, dev_item);
1984         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1985         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1986         device->type = btrfs_device_type(leaf, dev_item);
1987         device->io_align = btrfs_device_io_align(leaf, dev_item);
1988         device->io_width = btrfs_device_io_width(leaf, dev_item);
1989         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1990
1991         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1992         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1993
1994         return 0;
1995 }
1996
1997 static int read_one_dev(struct btrfs_root *root,
1998                         struct extent_buffer *leaf,
1999                         struct btrfs_dev_item *dev_item)
2000 {
2001         struct btrfs_device *device;
2002         u64 devid;
2003         int ret;
2004         u8 dev_uuid[BTRFS_UUID_SIZE];
2005
2006         devid = btrfs_device_id(leaf, dev_item);
2007         read_extent_buffer(leaf, dev_uuid,
2008                            (unsigned long)btrfs_device_uuid(dev_item),
2009                            BTRFS_UUID_SIZE);
2010         device = btrfs_find_device(root, devid, dev_uuid);
2011         if (!device) {
2012                 printk("warning devid %Lu not found already\n", devid);
2013                 device = kzalloc(sizeof(*device), GFP_NOFS);
2014                 if (!device)
2015                         return -ENOMEM;
2016                 list_add(&device->dev_list,
2017                          &root->fs_info->fs_devices->devices);
2018                 list_add(&device->dev_alloc_list,
2019                          &root->fs_info->fs_devices->alloc_list);
2020                 device->barriers = 1;
2021                 spin_lock_init(&device->io_lock);
2022         }
2023
2024         fill_device_from_item(leaf, dev_item, device);
2025         device->dev_root = root->fs_info->dev_root;
2026         ret = 0;
2027 #if 0
2028         ret = btrfs_open_device(device);
2029         if (ret) {
2030                 kfree(device);
2031         }
2032 #endif
2033         return ret;
2034 }
2035
2036 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2037 {
2038         struct btrfs_dev_item *dev_item;
2039
2040         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2041                                                      dev_item);
2042         return read_one_dev(root, buf, dev_item);
2043 }
2044
2045 int btrfs_read_sys_array(struct btrfs_root *root)
2046 {
2047         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2048         struct extent_buffer *sb;
2049         struct btrfs_disk_key *disk_key;
2050         struct btrfs_chunk *chunk;
2051         u8 *ptr;
2052         unsigned long sb_ptr;
2053         int ret = 0;
2054         u32 num_stripes;
2055         u32 array_size;
2056         u32 len = 0;
2057         u32 cur;
2058         struct btrfs_key key;
2059
2060         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2061                                           BTRFS_SUPER_INFO_SIZE);
2062         if (!sb)
2063                 return -ENOMEM;
2064         btrfs_set_buffer_uptodate(sb);
2065         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2066         array_size = btrfs_super_sys_array_size(super_copy);
2067
2068         ptr = super_copy->sys_chunk_array;
2069         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2070         cur = 0;
2071
2072         while (cur < array_size) {
2073                 disk_key = (struct btrfs_disk_key *)ptr;
2074                 btrfs_disk_key_to_cpu(&key, disk_key);
2075
2076                 len = sizeof(*disk_key); ptr += len;
2077                 sb_ptr += len;
2078                 cur += len;
2079
2080                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2081                         chunk = (struct btrfs_chunk *)sb_ptr;
2082                         ret = read_one_chunk(root, &key, sb, chunk);
2083                         if (ret)
2084                                 break;
2085                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2086                         len = btrfs_chunk_item_size(num_stripes);
2087                 } else {
2088                         ret = -EIO;
2089                         break;
2090                 }
2091                 ptr += len;
2092                 sb_ptr += len;
2093                 cur += len;
2094         }
2095         free_extent_buffer(sb);
2096         return ret;
2097 }
2098
2099 int btrfs_read_chunk_tree(struct btrfs_root *root)
2100 {
2101         struct btrfs_path *path;
2102         struct extent_buffer *leaf;
2103         struct btrfs_key key;
2104         struct btrfs_key found_key;
2105         int ret;
2106         int slot;
2107
2108         root = root->fs_info->chunk_root;
2109
2110         path = btrfs_alloc_path();
2111         if (!path)
2112                 return -ENOMEM;
2113
2114         /* first we search for all of the device items, and then we
2115          * read in all of the chunk items.  This way we can create chunk
2116          * mappings that reference all of the devices that are afound
2117          */
2118         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2119         key.offset = 0;
2120         key.type = 0;
2121 again:
2122         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2123         while(1) {
2124                 leaf = path->nodes[0];
2125                 slot = path->slots[0];
2126                 if (slot >= btrfs_header_nritems(leaf)) {
2127                         ret = btrfs_next_leaf(root, path);
2128                         if (ret == 0)
2129                                 continue;
2130                         if (ret < 0)
2131                                 goto error;
2132                         break;
2133                 }
2134                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2135                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2136                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2137                                 break;
2138                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2139                                 struct btrfs_dev_item *dev_item;
2140                                 dev_item = btrfs_item_ptr(leaf, slot,
2141                                                   struct btrfs_dev_item);
2142                                 ret = read_one_dev(root, leaf, dev_item);
2143                                 BUG_ON(ret);
2144                         }
2145                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2146                         struct btrfs_chunk *chunk;
2147                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2148                         ret = read_one_chunk(root, &found_key, leaf, chunk);
2149                 }
2150                 path->slots[0]++;
2151         }
2152         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2153                 key.objectid = 0;
2154                 btrfs_release_path(root, path);
2155                 goto again;
2156         }
2157
2158         btrfs_free_path(path);
2159         ret = 0;
2160 error:
2161         return ret;
2162 }
2163