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 <asm/div64.h>
  22 #include "ctree.h"
  23 #include "extent_map.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26 #include "print-tree.h"
  27 #include "volumes.h"
  28
  29 struct map_lookup {
  30         u64 type;
  31         int io_align;
  32         int io_width;
  33         int stripe_len;
  34         int sector_size;
  35         int num_stripes;
  36         int sub_stripes;
  37         struct btrfs_bio_stripe stripes[];
  38 };
  39
  40 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  41                             (sizeof(struct btrfs_bio_stripe) * (n)))
  42
  43 static DEFINE_MUTEX(uuid_mutex);
  44 static LIST_HEAD(fs_uuids);
  45
  46 int btrfs_cleanup_fs_uuids(void)
  47 {
  48         struct btrfs_fs_devices *fs_devices;
  49         struct list_head *uuid_cur;
  50         struct list_head *devices_cur;
  51         struct btrfs_device *dev;
  52
  53         list_for_each(uuid_cur, &fs_uuids) {
  54                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  55                                         list);
  56                 while(!list_empty(&fs_devices->devices)) {
  57                         devices_cur = fs_devices->devices.next;
  58                         dev = list_entry(devices_cur, struct btrfs_device,
  59                                          dev_list);
  60                         if (dev->bdev) {
  61                                 close_bdev_excl(dev->bdev);
  62                         }
  63                         list_del(&dev->dev_list);
  64                         kfree(dev);
  65                 }
  66         }
  67         return 0;
  68 }
  69
  70 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  71                                           u8 *uuid)
  72 {
  73         struct btrfs_device *dev;
  74         struct list_head *cur;
  75
  76         list_for_each(cur, head) {
  77                 dev = list_entry(cur, struct btrfs_device, dev_list);
  78                 if (dev->devid == devid &&
  79                     !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
  80                         return dev;
  81                 }
  82         }
  83         return NULL;
  84 }
  85
  86 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  87 {
  88         struct list_head *cur;
  89         struct btrfs_fs_devices *fs_devices;
  90
  91         list_for_each(cur, &fs_uuids) {
  92                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  93                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  94                         return fs_devices;
  95         }
  96         return NULL;
  97 }
  98
  99 static int device_list_add(const char *path,
 100                            struct btrfs_super_block *disk_super,
 101                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 102 {
 103         struct btrfs_device *device;
 104         struct btrfs_fs_devices *fs_devices;
 105         u64 found_transid = btrfs_super_generation(disk_super);
 106
 107         fs_devices = find_fsid(disk_super->fsid);
 108         if (!fs_devices) {
 109                 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
 110                 if (!fs_devices)
 111                         return -ENOMEM;
 112                 INIT_LIST_HEAD(&fs_devices->devices);
 113                 list_add(&fs_devices->list, &fs_uuids);
 114                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 115                 fs_devices->latest_devid = devid;
 116                 fs_devices->latest_trans = found_transid;
 117                 fs_devices->lowest_devid = (u64)-1;
 118                 fs_devices->num_devices = 0;
 119                 device = NULL;
 120         } else {
 121                 device = __find_device(&fs_devices->devices, devid,
 122                                        disk_super->dev_item.uuid);
 123         }
 124         if (!device) {
 125                 device = kzalloc(sizeof(*device), GFP_NOFS);
 126                 if (!device) {
 127                         /* we can safely leave the fs_devices entry around */
 128                         return -ENOMEM;
 129                 }
 130                 device->devid = devid;
 131                 memcpy(device->uuid, disk_super->dev_item.uuid,
 132                        BTRFS_UUID_SIZE);
 133                 device->barriers = 1;
 134                 spin_lock_init(&device->io_lock);
 135                 device->name = kstrdup(path, GFP_NOFS);
 136                 if (!device->name) {
 137                         kfree(device);
 138                         return -ENOMEM;
 139                 }
 140                 list_add(&device->dev_list, &fs_devices->devices);
 141                 fs_devices->num_devices++;
 142         }
 143
 144         if (found_transid > fs_devices->latest_trans) {
 145                 fs_devices->latest_devid = devid;
 146                 fs_devices->latest_trans = found_transid;
 147         }
 148         if (fs_devices->lowest_devid > devid) {
 149                 fs_devices->lowest_devid = devid;
 150         }
 151         *fs_devices_ret = fs_devices;
 152         return 0;
 153 }
 154
 155 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 156 {
 157         struct list_head *head = &fs_devices->devices;
 158         struct list_head *cur;
 159         struct btrfs_device *device;
 160
 161         mutex_lock(&uuid_mutex);
 162         list_for_each(cur, head) {
 163                 device = list_entry(cur, struct btrfs_device, dev_list);
 164                 if (device->bdev) {
 165                         close_bdev_excl(device->bdev);
 166                 }
 167                 device->bdev = NULL;
 168         }
 169         mutex_unlock(&uuid_mutex);
 170         return 0;
 171 }
 172
 173 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 174                        int flags, void *holder)
 175 {
 176         struct block_device *bdev;
 177         struct list_head *head = &fs_devices->devices;
 178         struct list_head *cur;
 179         struct btrfs_device *device;
 180         int ret;
 181
 182         mutex_lock(&uuid_mutex);
 183         list_for_each(cur, head) {
 184                 device = list_entry(cur, struct btrfs_device, dev_list);
 185                 bdev = open_bdev_excl(device->name, flags, holder);
 186
 187                 if (IS_ERR(bdev)) {
 188                         printk("open %s failed\n", device->name);
 189                         ret = PTR_ERR(bdev);
 190                         goto fail;
 191                 }
 192                 if (device->devid == fs_devices->latest_devid)
 193                         fs_devices->latest_bdev = bdev;
 194                 if (device->devid == fs_devices->lowest_devid) {
 195                         fs_devices->lowest_bdev = bdev;
 196                 }
 197                 device->bdev = bdev;
 198         }
 199         mutex_unlock(&uuid_mutex);
 200         return 0;
 201 fail:
 202         mutex_unlock(&uuid_mutex);
 203         btrfs_close_devices(fs_devices);
 204         return ret;
 205 }
 206
 207 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 208                           struct btrfs_fs_devices **fs_devices_ret)
 209 {
 210         struct btrfs_super_block *disk_super;
 211         struct block_device *bdev;
 212         struct buffer_head *bh;
 213         int ret;
 214         u64 devid;
 215         u64 transid;
 216
 217         mutex_lock(&uuid_mutex);
 218
 219         bdev = open_bdev_excl(path, flags, holder);
 220
 221         if (IS_ERR(bdev)) {
 222                 ret = PTR_ERR(bdev);
 223                 goto error;
 224         }
 225
 226         ret = set_blocksize(bdev, 4096);
 227         if (ret)
 228                 goto error_close;
 229         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 230         if (!bh) {
 231                 ret = -EIO;
 232                 goto error_close;
 233         }
 234         disk_super = (struct btrfs_super_block *)bh->b_data;
 235         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 236             sizeof(disk_super->magic))) {
 237                 ret = -EINVAL;
 238                 goto error_brelse;
 239         }
 240         devid = le64_to_cpu(disk_super->dev_item.devid);
 241         transid = btrfs_super_generation(disk_super);
 242         if (disk_super->label[0])
 243                 printk("device label %s ", disk_super->label);
 244         else {
 245                 /* FIXME, make a readl uuid parser */
 246                 printk("device fsid %llx-%llx ",
 247                        *(unsigned long long *)disk_super->fsid,
 248                        *(unsigned long long *)(disk_super->fsid + 8));
 249         }
 250         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 251         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 252
 253 error_brelse:
 254         brelse(bh);
 255 error_close:
 256         close_bdev_excl(bdev);
 257 error:
 258         mutex_unlock(&uuid_mutex);
 259         return ret;
 260 }
 261
 262 /*
 263  * this uses a pretty simple search, the expectation is that it is
 264  * called very infrequently and that a given device has a small number
 265  * of extents
 266  */
 267 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 268                                 struct btrfs_device *device,
 269                                 struct btrfs_path *path,
 270                                 u64 num_bytes, u64 *start)
 271 {
 272         struct btrfs_key key;
 273         struct btrfs_root *root = device->dev_root;
 274         struct btrfs_dev_extent *dev_extent = NULL;
 275         u64 hole_size = 0;
 276         u64 last_byte = 0;
 277         u64 search_start = 0;
 278         u64 search_end = device->total_bytes;
 279         int ret;
 280         int slot = 0;
 281         int start_found;
 282         struct extent_buffer *l;
 283
 284         start_found = 0;
 285         path->reada = 2;
 286
 287         /* FIXME use last free of some kind */
 288
 289         /* we don't want to overwrite the superblock on the drive,
 290          * so we make sure to start at an offset of at least 1MB
 291          */
 292         search_start = max((u64)1024 * 1024, search_start);
 293         key.objectid = device->devid;
 294         key.offset = search_start;
 295         key.type = BTRFS_DEV_EXTENT_KEY;
 296         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 297         if (ret < 0)
 298                 goto error;
 299         ret = btrfs_previous_item(root, path, 0, key.type);
 300         if (ret < 0)
 301                 goto error;
 302         l = path->nodes[0];
 303         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 304         while (1) {
 305                 l = path->nodes[0];
 306                 slot = path->slots[0];
 307                 if (slot >= btrfs_header_nritems(l)) {
 308                         ret = btrfs_next_leaf(root, path);
 309                         if (ret == 0)
 310                                 continue;
 311                         if (ret < 0)
 312                                 goto error;
 313 no_more_items:
 314                         if (!start_found) {
 315                                 if (search_start >= search_end) {
 316                                         ret = -ENOSPC;
 317                                         goto error;
 318                                 }
 319                                 *start = search_start;
 320                                 start_found = 1;
 321                                 goto check_pending;
 322                         }
 323                         *start = last_byte > search_start ?
 324                                 last_byte : search_start;
 325                         if (search_end <= *start) {
 326                                 ret = -ENOSPC;
 327                                 goto error;
 328                         }
 329                         goto check_pending;
 330                 }
 331                 btrfs_item_key_to_cpu(l, &key, slot);
 332
 333                 if (key.objectid < device->devid)
 334                         goto next;
 335
 336                 if (key.objectid > device->devid)
 337                         goto no_more_items;
 338
 339                 if (key.offset >= search_start && key.offset > last_byte &&
 340                     start_found) {
 341                         if (last_byte < search_start)
 342                                 last_byte = search_start;
 343                         hole_size = key.offset - last_byte;
 344                         if (key.offset > last_byte &&
 345                             hole_size >= num_bytes) {
 346                                 *start = last_byte;
 347                                 goto check_pending;
 348                         }
 349                 }
 350                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 351                         goto next;
 352                 }
 353
 354                 start_found = 1;
 355                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 356                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 357 next:
 358                 path->slots[0]++;
 359                 cond_resched();
 360         }
 361 check_pending:
 362         /* we have to make sure we didn't find an extent that has already
 363          * been allocated by the map tree or the original allocation
 364          */
 365         btrfs_release_path(root, path);
 366         BUG_ON(*start < search_start);
 367
 368         if (*start + num_bytes > search_end) {
 369                 ret = -ENOSPC;
 370                 goto error;
 371         }
 372         /* check for pending inserts here */
 373         return 0;
 374
 375 error:
 376         btrfs_release_path(root, path);
 377         return ret;
 378 }
 379
 380 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 381                            struct btrfs_device *device,
 382                            u64 chunk_tree, u64 chunk_objectid,
 383                            u64 chunk_offset,
 384                            u64 num_bytes, u64 *start)
 385 {
 386         int ret;
 387         struct btrfs_path *path;
 388         struct btrfs_root *root = device->dev_root;
 389         struct btrfs_dev_extent *extent;
 390         struct extent_buffer *leaf;
 391         struct btrfs_key key;
 392
 393         path = btrfs_alloc_path();
 394         if (!path)
 395                 return -ENOMEM;
 396
 397         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 398         if (ret) {
 399                 goto err;
 400         }
 401
 402         key.objectid = device->devid;
 403         key.offset = *start;
 404         key.type = BTRFS_DEV_EXTENT_KEY;
 405         ret = btrfs_insert_empty_item(trans, root, path, &key,
 406                                       sizeof(*extent));
 407         BUG_ON(ret);
 408
 409         leaf = path->nodes[0];
 410         extent = btrfs_item_ptr(leaf, path->slots[0],
 411                                 struct btrfs_dev_extent);
 412         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 413         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 414         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 415
 416         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 417                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 418                     BTRFS_UUID_SIZE);
 419
 420         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 421         btrfs_mark_buffer_dirty(leaf);
 422 err:
 423         btrfs_free_path(path);
 424         return ret;
 425 }
 426
 427 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 428 {
 429         struct btrfs_path *path;
 430         int ret;
 431         struct btrfs_key key;
 432         struct btrfs_chunk *chunk;
 433         struct btrfs_key found_key;
 434
 435         path = btrfs_alloc_path();
 436         BUG_ON(!path);
 437
 438         key.objectid = objectid;
 439         key.offset = (u64)-1;
 440         key.type = BTRFS_CHUNK_ITEM_KEY;
 441
 442         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 443         if (ret < 0)
 444                 goto error;
 445
 446         BUG_ON(ret == 0);
 447
 448         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 449         if (ret) {
 450                 *offset = 0;
 451         } else {
 452                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 453                                       path->slots[0]);
 454                 if (found_key.objectid != objectid)
 455                         *offset = 0;
 456                 else {
 457                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 458                                                struct btrfs_chunk);
 459                         *offset = found_key.offset +
 460                                 btrfs_chunk_length(path->nodes[0], chunk);
 461                 }
 462         }
 463         ret = 0;
 464 error:
 465         btrfs_free_path(path);
 466         return ret;
 467 }
 468
 469 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 470                            u64 *objectid)
 471 {
 472         int ret;
 473         struct btrfs_key key;
 474         struct btrfs_key found_key;
 475
 476         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 477         key.type = BTRFS_DEV_ITEM_KEY;
 478         key.offset = (u64)-1;
 479
 480         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 481         if (ret < 0)
 482                 goto error;
 483
 484         BUG_ON(ret == 0);
 485
 486         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 487                                   BTRFS_DEV_ITEM_KEY);
 488         if (ret) {
 489                 *objectid = 1;
 490         } else {
 491                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 492                                       path->slots[0]);
 493                 *objectid = found_key.offset + 1;
 494         }
 495         ret = 0;
 496 error:
 497         btrfs_release_path(root, path);
 498         return ret;
 499 }
 500
 501 /*
 502  * the device information is stored in the chunk root
 503  * the btrfs_device struct should be fully filled in
 504  */
 505 int btrfs_add_device(struct btrfs_trans_handle *trans,
 506                      struct btrfs_root *root,
 507                      struct btrfs_device *device)
 508 {
 509         int ret;
 510         struct btrfs_path *path;
 511         struct btrfs_dev_item *dev_item;
 512         struct extent_buffer *leaf;
 513         struct btrfs_key key;
 514         unsigned long ptr;
 515         u64 free_devid;
 516
 517         root = root->fs_info->chunk_root;
 518
 519         path = btrfs_alloc_path();
 520         if (!path)
 521                 return -ENOMEM;
 522
 523         ret = find_next_devid(root, path, &free_devid);
 524         if (ret)
 525                 goto out;
 526
 527         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 528         key.type = BTRFS_DEV_ITEM_KEY;
 529         key.offset = free_devid;
 530
 531         ret = btrfs_insert_empty_item(trans, root, path, &key,
 532                                       sizeof(*dev_item));
 533         if (ret)
 534                 goto out;
 535
 536         leaf = path->nodes[0];
 537         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 538
 539         device->devid = free_devid;
 540         btrfs_set_device_id(leaf, dev_item, device->devid);
 541         btrfs_set_device_type(leaf, dev_item, device->type);
 542         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 543         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 544         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 545         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 546         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 547         btrfs_set_device_group(leaf, dev_item, 0);
 548         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 549         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 550
 551         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 552         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 553         btrfs_mark_buffer_dirty(leaf);
 554         ret = 0;
 555
 556 out:
 557         btrfs_free_path(path);
 558         return ret;
 559 }
 560 int btrfs_update_device(struct btrfs_trans_handle *trans,
 561                         struct btrfs_device *device)
 562 {
 563         int ret;
 564         struct btrfs_path *path;
 565         struct btrfs_root *root;
 566         struct btrfs_dev_item *dev_item;
 567         struct extent_buffer *leaf;
 568         struct btrfs_key key;
 569
 570         root = device->dev_root->fs_info->chunk_root;
 571
 572         path = btrfs_alloc_path();
 573         if (!path)
 574                 return -ENOMEM;
 575
 576         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 577         key.type = BTRFS_DEV_ITEM_KEY;
 578         key.offset = device->devid;
 579
 580         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 581         if (ret < 0)
 582                 goto out;
 583
 584         if (ret > 0) {
 585                 ret = -ENOENT;
 586                 goto out;
 587         }
 588
 589         leaf = path->nodes[0];
 590         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 591
 592         btrfs_set_device_id(leaf, dev_item, device->devid);
 593         btrfs_set_device_type(leaf, dev_item, device->type);
 594         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 595         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 596         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 597         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 598         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 599         btrfs_mark_buffer_dirty(leaf);
 600
 601 out:
 602         btrfs_free_path(path);
 603         return ret;
 604 }
 605
 606 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 607                            struct btrfs_root *root,
 608                            struct btrfs_key *key,
 609                            struct btrfs_chunk *chunk, int item_size)
 610 {
 611         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 612         struct btrfs_disk_key disk_key;
 613         u32 array_size;
 614         u8 *ptr;
 615
 616         array_size = btrfs_super_sys_array_size(super_copy);
 617         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 618                 return -EFBIG;
 619
 620         ptr = super_copy->sys_chunk_array + array_size;
 621         btrfs_cpu_key_to_disk(&disk_key, key);
 622         memcpy(ptr, &disk_key, sizeof(disk_key));
 623         ptr += sizeof(disk_key);
 624         memcpy(ptr, chunk, item_size);
 625         item_size += sizeof(disk_key);
 626         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 627         return 0;
 628 }
 629
 630 static u64 div_factor(u64 num, int factor)
 631 {
 632         if (factor == 10)
 633                 return num;
 634         num *= factor;
 635         do_div(num, 10);
 636         return num;
 637 }
 638
 639 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 640                                int sub_stripes)
 641 {
 642         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 643                 return calc_size;
 644         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 645                 return calc_size * (num_stripes / sub_stripes);
 646         else
 647                 return calc_size * num_stripes;
 648 }
 649
 650
 651 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 652                       struct btrfs_root *extent_root, u64 *start,
 653                       u64 *num_bytes, u64 type)
 654 {
 655         u64 dev_offset;
 656         struct btrfs_fs_info *info = extent_root->fs_info;
 657         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
 658         struct btrfs_stripe *stripes;
 659         struct btrfs_device *device = NULL;
 660         struct btrfs_chunk *chunk;
 661         struct list_head private_devs;
 662         struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
 663         struct list_head *cur;
 664         struct extent_map_tree *em_tree;
 665         struct map_lookup *map;
 666         struct extent_map *em;
 667         int min_chunk_size = 8 * 1024 * 1024;
 668         u64 physical;
 669         u64 calc_size = 1024 * 1024 * 1024;
 670         u64 max_chunk_size = calc_size;
 671         u64 min_free;
 672         u64 avail;
 673         u64 max_avail = 0;
 674         u64 percent_max;
 675         int num_stripes = 1;
 676         int sub_stripes = 0;
 677         int looped = 0;
 678         int ret;
 679         int index;
 680         int stripe_len = 64 * 1024;
 681         struct btrfs_key key;
 682
 683         if (list_empty(dev_list))
 684                 return -ENOSPC;
 685
 686         if (type & (BTRFS_BLOCK_GROUP_RAID0))
 687                 num_stripes = btrfs_super_num_devices(&info->super_copy);
 688         if (type & (BTRFS_BLOCK_GROUP_DUP))
 689                 num_stripes = 2;
 690         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
 691                 num_stripes = min_t(u64, 2,
 692                                   btrfs_super_num_devices(&info->super_copy));
 693                 if (num_stripes < 2)
 694                         return -ENOSPC;
 695         }
 696         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 697                 num_stripes = btrfs_super_num_devices(&info->super_copy);
 698                 if (num_stripes < 4)
 699                         return -ENOSPC;
 700                 num_stripes &= ~(u32)1;
 701                 sub_stripes = 2;
 702         }
 703
 704         if (type & BTRFS_BLOCK_GROUP_DATA) {
 705                 max_chunk_size = 10 * calc_size;
 706                 min_chunk_size = 256 * 1024 * 1024;
 707         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 708                 max_chunk_size = 4 * calc_size;
 709                 min_chunk_size = 64 * 1024 * 1024;
 710         } else {
 711                 min_chunk_size = 32 * 1024 * 1024;
 712         }
 713
 714         /* we don't want a chunk larger than 10% of the FS */
 715         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
 716         max_chunk_size = min(percent_max, max_chunk_size);
 717
 718         if (calc_size * num_stripes > max_chunk_size) {
 719                 calc_size = max_chunk_size;
 720                 do_div(calc_size, num_stripes);
 721                 do_div(calc_size, stripe_len);
 722                 calc_size *= stripe_len;
 723         }
 724         /* we don't want tiny stripes */
 725         *num_bytes = chunk_bytes_by_type(type, calc_size,
 726                                          num_stripes, sub_stripes);
 727         calc_size = max_t(u64, chunk_bytes_by_type(type, min_chunk_size,
 728                           num_stripes, sub_stripes), calc_size);
 729
 730 again:
 731         do_div(calc_size, stripe_len);
 732         calc_size *= stripe_len;
 733
 734         INIT_LIST_HEAD(&private_devs);
 735         cur = dev_list->next;
 736         index = 0;
 737
 738         if (type & BTRFS_BLOCK_GROUP_DUP)
 739                 min_free = calc_size * 2;
 740         else
 741                 min_free = calc_size;
 742
 743         /* build a private list of devices we will allocate from */
 744         while(index < num_stripes) {
 745                 device = list_entry(cur, struct btrfs_device, dev_list);
 746
 747                 avail = device->total_bytes - device->bytes_used;
 748                 cur = cur->next;
 749                 if (avail > max_avail)
 750                         max_avail = avail;
 751                 if (avail >= min_free) {
 752                         list_move_tail(&device->dev_list, &private_devs);
 753                         index++;
 754                         if (type & BTRFS_BLOCK_GROUP_DUP)
 755                                 index++;
 756                 }
 757                 if (cur == dev_list)
 758                         break;
 759         }
 760         if (index < num_stripes) {
 761                 list_splice(&private_devs, dev_list);
 762                 if (!looped && max_avail > 0) {
 763                         looped = 1;
 764                         calc_size = max_avail;
 765                         goto again;
 766                 }
 767                 return -ENOSPC;
 768         }
 769
 770         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 771         key.type = BTRFS_CHUNK_ITEM_KEY;
 772         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 773                               &key.offset);
 774         if (ret)
 775                 return ret;
 776
 777         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 778         if (!chunk)
 779                 return -ENOMEM;
 780
 781         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
 782         if (!map) {
 783                 kfree(chunk);
 784                 return -ENOMEM;
 785         }
 786
 787         stripes = &chunk->stripe;
 788         *num_bytes = chunk_bytes_by_type(type, calc_size,
 789                                          num_stripes, sub_stripes);
 790
 791
 792         index = 0;
 793 printk("new chunk type %Lu start %Lu size %Lu\n", type, key.offset, *num_bytes);
 794         while(index < num_stripes) {
 795                 struct btrfs_stripe *stripe;
 796                 BUG_ON(list_empty(&private_devs));
 797                 cur = private_devs.next;
 798                 device = list_entry(cur, struct btrfs_device, dev_list);
 799
 800                 /* loop over this device again if we're doing a dup group */
 801                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 802                     (index == num_stripes - 1))
 803                         list_move_tail(&device->dev_list, dev_list);
 804
 805                 ret = btrfs_alloc_dev_extent(trans, device,
 806                              info->chunk_root->root_key.objectid,
 807                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 808                              calc_size, &dev_offset);
 809                 BUG_ON(ret);
 810 printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.offset, calc_size, device->devid, type);
 811                 device->bytes_used += calc_size;
 812                 ret = btrfs_update_device(trans, device);
 813                 BUG_ON(ret);
 814
 815                 map->stripes[index].dev = device;
 816                 map->stripes[index].physical = dev_offset;
 817                 stripe = stripes + index;
 818                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 819                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 820                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 821                 physical = dev_offset;
 822                 index++;
 823         }
 824         BUG_ON(!list_empty(&private_devs));
 825
 826         /* key was set above */
 827         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 828         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 829         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 830         btrfs_set_stack_chunk_type(chunk, type);
 831         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 832         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 833         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 834         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 835         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
 836         map->sector_size = extent_root->sectorsize;
 837         map->stripe_len = stripe_len;
 838         map->io_align = stripe_len;
 839         map->io_width = stripe_len;
 840         map->type = type;
 841         map->num_stripes = num_stripes;
 842         map->sub_stripes = sub_stripes;
 843
 844         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 845                                 btrfs_chunk_item_size(num_stripes));
 846         BUG_ON(ret);
 847         *start = key.offset;;
 848
 849         em = alloc_extent_map(GFP_NOFS);
 850         if (!em)
 851                 return -ENOMEM;
 852         em->bdev = (struct block_device *)map;
 853         em->start = key.offset;
 854         em->len = *num_bytes;
 855         em->block_start = 0;
 856
 857         kfree(chunk);
 858
 859         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
 860         spin_lock(&em_tree->lock);
 861         ret = add_extent_mapping(em_tree, em);
 862         spin_unlock(&em_tree->lock);
 863         BUG_ON(ret);
 864         free_extent_map(em);
 865         return ret;
 866 }
 867
 868 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
 869 {
 870         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
 871 }
 872
 873 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
 874 {
 875         struct extent_map *em;
 876
 877         while(1) {
 878                 spin_lock(&tree->map_tree.lock);
 879                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
 880                 if (em)
 881                         remove_extent_mapping(&tree->map_tree, em);
 882                 spin_unlock(&tree->map_tree.lock);
 883                 if (!em)
 884                         break;
 885                 kfree(em->bdev);
 886                 /* once for us */
 887                 free_extent_map(em);
 888                 /* once for the tree */
 889                 free_extent_map(em);
 890         }
 891 }
 892
 893 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
 894 {
 895         struct extent_map *em;
 896         struct map_lookup *map;
 897         struct extent_map_tree *em_tree = &map_tree->map_tree;
 898         int ret;
 899
 900         spin_lock(&em_tree->lock);
 901         em = lookup_extent_mapping(em_tree, logical, len);
 902         spin_unlock(&em_tree->lock);
 903         BUG_ON(!em);
 904
 905         BUG_ON(em->start > logical || em->start + em->len < logical);
 906         map = (struct map_lookup *)em->bdev;
 907         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
 908                 ret = map->num_stripes;
 909         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
 910                 ret = map->sub_stripes;
 911         else
 912                 ret = 1;
 913         free_extent_map(em);
 914         return ret;
 915 }
 916
 917 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
 918                     u64 logical, u64 *length,
 919                     struct btrfs_multi_bio **multi_ret, int mirror_num)
 920 {
 921         struct extent_map *em;
 922         struct map_lookup *map;
 923         struct extent_map_tree *em_tree = &map_tree->map_tree;
 924         u64 offset;
 925         u64 stripe_offset;
 926         u64 stripe_nr;
 927         int stripes_allocated = 8;
 928         int stripes_required = 1;
 929         int stripe_index;
 930         int i;
 931         struct btrfs_multi_bio *multi = NULL;
 932
 933         if (multi_ret && !(rw & (1 << BIO_RW))) {
 934                 stripes_allocated = 1;
 935         }
 936 again:
 937         if (multi_ret) {
 938                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
 939                                 GFP_NOFS);
 940                 if (!multi)
 941                         return -ENOMEM;
 942         }
 943
 944         spin_lock(&em_tree->lock);
 945         em = lookup_extent_mapping(em_tree, logical, *length);
 946         spin_unlock(&em_tree->lock);
 947         if (!em) {
 948                 printk("unable to find logical %Lu\n", logical);
 949         }
 950         BUG_ON(!em);
 951
 952         BUG_ON(em->start > logical || em->start + em->len < logical);
 953         map = (struct map_lookup *)em->bdev;
 954         offset = logical - em->start;
 955
 956         if (mirror_num > map->num_stripes)
 957                 mirror_num = 0;
 958
 959         /* if our multi bio struct is too small, back off and try again */
 960         if (rw & (1 << BIO_RW)) {
 961                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
 962                                  BTRFS_BLOCK_GROUP_DUP)) {
 963                         stripes_required = map->num_stripes;
 964                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
 965                         stripes_required = map->sub_stripes;
 966                 }
 967         }
 968         if (multi_ret && rw == WRITE &&
 969             stripes_allocated < stripes_required) {
 970                 stripes_allocated = map->num_stripes;
 971                 free_extent_map(em);
 972                 kfree(multi);
 973                 goto again;
 974         }
 975         stripe_nr = offset;
 976         /*
 977          * stripe_nr counts the total number of stripes we have to stride
 978          * to get to this block
 979          */
 980         do_div(stripe_nr, map->stripe_len);
 981
 982         stripe_offset = stripe_nr * map->stripe_len;
 983         BUG_ON(offset < stripe_offset);
 984
 985         /* stripe_offset is the offset of this block in its stripe*/
 986         stripe_offset = offset - stripe_offset;
 987
 988         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
 989                          BTRFS_BLOCK_GROUP_RAID10 |
 990                          BTRFS_BLOCK_GROUP_DUP)) {
 991                 /* we limit the length of each bio to what fits in a stripe */
 992                 *length = min_t(u64, em->len - offset,
 993                               map->stripe_len - stripe_offset);
 994         } else {
 995                 *length = em->len - offset;
 996         }
 997         if (!multi_ret)
 998                 goto out;
 999
1000         multi->num_stripes = 1;
1001         stripe_index = 0;
1002         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1003                 if (rw & (1 << BIO_RW))
1004                         multi->num_stripes = map->num_stripes;
1005                 else if (mirror_num) {
1006                         stripe_index = mirror_num - 1;
1007                 } else {
1008                         int i;
1009                         u64 least = (u64)-1;
1010                         struct btrfs_device *cur;
1011
1012                         for (i = 0; i < map->num_stripes; i++) {
1013                                 cur = map->stripes[i].dev;
1014                                 spin_lock(&cur->io_lock);
1015                                 if (cur->total_ios < least) {
1016                                         least = cur->total_ios;
1017                                         stripe_index = i;
1018                                 }
1019                                 spin_unlock(&cur->io_lock);
1020                         }
1021                 }
1022         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1023                 if (rw & (1 << BIO_RW))
1024                         multi->num_stripes = map->num_stripes;
1025                 else if (mirror_num)
1026                         stripe_index = mirror_num - 1;
1027         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1028                 int factor = map->num_stripes / map->sub_stripes;
1029                 int orig_stripe_nr = stripe_nr;
1030
1031                 stripe_index = do_div(stripe_nr, factor);
1032                 stripe_index *= map->sub_stripes;
1033
1034                 if (rw & (1 << BIO_RW))
1035                         multi->num_stripes = map->sub_stripes;
1036                 else if (mirror_num)
1037                         stripe_index += mirror_num - 1;
1038                 else
1039                         stripe_index += orig_stripe_nr % map->sub_stripes;
1040         } else {
1041                 /*
1042                  * after this do_div call, stripe_nr is the number of stripes
1043                  * on this device we have to walk to find the data, and
1044                  * stripe_index is the number of our device in the stripe array
1045                  */
1046                 stripe_index = do_div(stripe_nr, map->num_stripes);
1047         }
1048         BUG_ON(stripe_index >= map->num_stripes);
1049
1050         for (i = 0; i < multi->num_stripes; i++) {
1051                 multi->stripes[i].physical =
1052                         map->stripes[stripe_index].physical + stripe_offset +
1053                         stripe_nr * map->stripe_len;
1054                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1055                 stripe_index++;
1056         }
1057         *multi_ret = multi;
1058 out:
1059         free_extent_map(em);
1060         return 0;
1061 }
1062
1063 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1064 static void end_bio_multi_stripe(struct bio *bio, int err)
1065 #else
1066 static int end_bio_multi_stripe(struct bio *bio,
1067                                    unsigned int bytes_done, int err)
1068 #endif
1069 {
1070         struct btrfs_multi_bio *multi = bio->bi_private;
1071
1072 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1073         if (bio->bi_size)
1074                 return 1;
1075 #endif
1076         if (err)
1077                 multi->error = err;
1078
1079         if (atomic_dec_and_test(&multi->stripes_pending)) {
1080                 bio->bi_private = multi->private;
1081                 bio->bi_end_io = multi->end_io;
1082
1083                 if (!err && multi->error)
1084                         err = multi->error;
1085                 kfree(multi);
1086
1087 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1088                 bio_endio(bio, bio->bi_size, err);
1089 #else
1090                 bio_endio(bio, err);
1091 #endif
1092         } else {
1093                 bio_put(bio);
1094         }
1095 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1096         return 0;
1097 #endif
1098 }
1099
1100 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1101                   int mirror_num)
1102 {
1103         struct btrfs_mapping_tree *map_tree;
1104         struct btrfs_device *dev;
1105         struct bio *first_bio = bio;
1106         u64 logical = bio->bi_sector << 9;
1107         u64 length = 0;
1108         u64 map_length;
1109         struct bio_vec *bvec;
1110         struct btrfs_multi_bio *multi = NULL;
1111         int i;
1112         int ret;
1113         int dev_nr = 0;
1114         int total_devs = 1;
1115
1116         bio_for_each_segment(bvec, bio, i) {
1117                 length += bvec->bv_len;
1118         }
1119
1120         map_tree = &root->fs_info->mapping_tree;
1121         map_length = length;
1122
1123         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1124                               mirror_num);
1125         BUG_ON(ret);
1126
1127         total_devs = multi->num_stripes;
1128         if (map_length < length) {
1129                 printk("mapping failed logical %Lu bio len %Lu "
1130                        "len %Lu\n", logical, length, map_length);
1131                 BUG();
1132         }
1133         multi->end_io = first_bio->bi_end_io;
1134         multi->private = first_bio->bi_private;
1135         atomic_set(&multi->stripes_pending, multi->num_stripes);
1136
1137         while(dev_nr < total_devs) {
1138                 if (total_devs > 1) {
1139                         if (dev_nr < total_devs - 1) {
1140                                 bio = bio_clone(first_bio, GFP_NOFS);
1141                                 BUG_ON(!bio);
1142                         } else {
1143                                 bio = first_bio;
1144                         }
1145                         bio->bi_private = multi;
1146                         bio->bi_end_io = end_bio_multi_stripe;
1147                 }
1148                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1149                 dev = multi->stripes[dev_nr].dev;
1150                 bio->bi_bdev = dev->bdev;
1151                 spin_lock(&dev->io_lock);
1152                 dev->total_ios++;
1153                 spin_unlock(&dev->io_lock);
1154                 submit_bio(rw, bio);
1155                 dev_nr++;
1156         }
1157         if (total_devs == 1)
1158                 kfree(multi);
1159         return 0;
1160 }
1161
1162 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1163                                        u8 *uuid)
1164 {
1165         struct list_head *head = &root->fs_info->fs_devices->devices;
1166
1167         return __find_device(head, devid, uuid);
1168 }
1169
1170 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1171                           struct extent_buffer *leaf,
1172                           struct btrfs_chunk *chunk)
1173 {
1174         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1175         struct map_lookup *map;
1176         struct extent_map *em;
1177         u64 logical;
1178         u64 length;
1179         u64 devid;
1180         u8 uuid[BTRFS_UUID_SIZE];
1181         int num_stripes;
1182         int ret;
1183         int i;
1184
1185         logical = key->offset;
1186         length = btrfs_chunk_length(leaf, chunk);
1187         spin_lock(&map_tree->map_tree.lock);
1188         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1189         spin_unlock(&map_tree->map_tree.lock);
1190
1191         /* already mapped? */
1192         if (em && em->start <= logical && em->start + em->len > logical) {
1193                 free_extent_map(em);
1194                 return 0;
1195         } else if (em) {
1196                 free_extent_map(em);
1197         }
1198
1199         map = kzalloc(sizeof(*map), GFP_NOFS);
1200         if (!map)
1201                 return -ENOMEM;
1202
1203         em = alloc_extent_map(GFP_NOFS);
1204         if (!em)
1205                 return -ENOMEM;
1206         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1207         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1208         if (!map) {
1209                 free_extent_map(em);
1210                 return -ENOMEM;
1211         }
1212
1213         em->bdev = (struct block_device *)map;
1214         em->start = logical;
1215         em->len = length;
1216         em->block_start = 0;
1217
1218         map->num_stripes = num_stripes;
1219         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1220         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1221         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1222         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1223         map->type = btrfs_chunk_type(leaf, chunk);
1224         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1225         for (i = 0; i < num_stripes; i++) {
1226                 map->stripes[i].physical =
1227                         btrfs_stripe_offset_nr(leaf, chunk, i);
1228                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1229                 read_extent_buffer(leaf, uuid, (unsigned long)
1230                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1231                                    BTRFS_UUID_SIZE);
1232                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1233                 if (!map->stripes[i].dev) {
1234                         kfree(map);
1235                         free_extent_map(em);
1236                         return -EIO;
1237                 }
1238         }
1239
1240         spin_lock(&map_tree->map_tree.lock);
1241         ret = add_extent_mapping(&map_tree->map_tree, em);
1242         spin_unlock(&map_tree->map_tree.lock);
1243         BUG_ON(ret);
1244         free_extent_map(em);
1245
1246         return 0;
1247 }
1248
1249 static int fill_device_from_item(struct extent_buffer *leaf,
1250                                  struct btrfs_dev_item *dev_item,
1251                                  struct btrfs_device *device)
1252 {
1253         unsigned long ptr;
1254
1255         device->devid = btrfs_device_id(leaf, dev_item);
1256         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1257         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1258         device->type = btrfs_device_type(leaf, dev_item);
1259         device->io_align = btrfs_device_io_align(leaf, dev_item);
1260         device->io_width = btrfs_device_io_width(leaf, dev_item);
1261         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1262
1263         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1264         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1265
1266         return 0;
1267 }
1268
1269 static int read_one_dev(struct btrfs_root *root,
1270                         struct extent_buffer *leaf,
1271                         struct btrfs_dev_item *dev_item)
1272 {
1273         struct btrfs_device *device;
1274         u64 devid;
1275         int ret;
1276         u8 dev_uuid[BTRFS_UUID_SIZE];
1277
1278         devid = btrfs_device_id(leaf, dev_item);
1279         read_extent_buffer(leaf, dev_uuid,
1280                            (unsigned long)btrfs_device_uuid(dev_item),
1281                            BTRFS_UUID_SIZE);
1282         device = btrfs_find_device(root, devid, dev_uuid);
1283         if (!device) {
1284                 printk("warning devid %Lu not found already\n", devid);
1285                 device = kzalloc(sizeof(*device), GFP_NOFS);
1286                 if (!device)
1287                         return -ENOMEM;
1288                 list_add(&device->dev_list,
1289                          &root->fs_info->fs_devices->devices);
1290                 device->barriers = 1;
1291                 spin_lock_init(&device->io_lock);
1292         }
1293
1294         fill_device_from_item(leaf, dev_item, device);
1295         device->dev_root = root->fs_info->dev_root;
1296         ret = 0;
1297 #if 0
1298         ret = btrfs_open_device(device);
1299         if (ret) {
1300                 kfree(device);
1301         }
1302 #endif
1303         return ret;
1304 }
1305
1306 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1307 {
1308         struct btrfs_dev_item *dev_item;
1309
1310         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1311                                                      dev_item);
1312         return read_one_dev(root, buf, dev_item);
1313 }
1314
1315 int btrfs_read_sys_array(struct btrfs_root *root)
1316 {
1317         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1318         struct extent_buffer *sb = root->fs_info->sb_buffer;
1319         struct btrfs_disk_key *disk_key;
1320         struct btrfs_chunk *chunk;
1321         struct btrfs_key key;
1322         u32 num_stripes;
1323         u32 array_size;
1324         u32 len = 0;
1325         u8 *ptr;
1326         unsigned long sb_ptr;
1327         u32 cur;
1328         int ret;
1329
1330         array_size = btrfs_super_sys_array_size(super_copy);
1331
1332         /*
1333          * we do this loop twice, once for the device items and
1334          * once for all of the chunks.  This way there are device
1335          * structs filled in for every chunk
1336          */
1337         ptr = super_copy->sys_chunk_array;
1338         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1339         cur = 0;
1340
1341         while (cur < array_size) {
1342                 disk_key = (struct btrfs_disk_key *)ptr;
1343                 btrfs_disk_key_to_cpu(&key, disk_key);
1344
1345                 len = sizeof(*disk_key);
1346                 ptr += len;
1347                 sb_ptr += len;
1348                 cur += len;
1349
1350                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1351                         chunk = (struct btrfs_chunk *)sb_ptr;
1352                         ret = read_one_chunk(root, &key, sb, chunk);
1353                         BUG_ON(ret);
1354                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1355                         len = btrfs_chunk_item_size(num_stripes);
1356                 } else {
1357                         BUG();
1358                 }
1359                 ptr += len;
1360                 sb_ptr += len;
1361                 cur += len;
1362         }
1363         return 0;
1364 }
1365
1366 int btrfs_read_chunk_tree(struct btrfs_root *root)
1367 {
1368         struct btrfs_path *path;
1369         struct extent_buffer *leaf;
1370         struct btrfs_key key;
1371         struct btrfs_key found_key;
1372         int ret;
1373         int slot;
1374
1375         root = root->fs_info->chunk_root;
1376
1377         path = btrfs_alloc_path();
1378         if (!path)
1379                 return -ENOMEM;
1380
1381         /* first we search for all of the device items, and then we
1382          * read in all of the chunk items.  This way we can create chunk
1383          * mappings that reference all of the devices that are afound
1384          */
1385         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1386         key.offset = 0;
1387         key.type = 0;
1388 again:
1389         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1390         while(1) {
1391                 leaf = path->nodes[0];
1392                 slot = path->slots[0];
1393                 if (slot >= btrfs_header_nritems(leaf)) {
1394                         ret = btrfs_next_leaf(root, path);
1395                         if (ret == 0)
1396                                 continue;
1397                         if (ret < 0)
1398                                 goto error;
1399                         break;
1400                 }
1401                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1402                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1403                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1404                                 break;
1405                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1406                                 struct btrfs_dev_item *dev_item;
1407                                 dev_item = btrfs_item_ptr(leaf, slot,
1408                                                   struct btrfs_dev_item);
1409                                 ret = read_one_dev(root, leaf, dev_item);
1410                                 BUG_ON(ret);
1411                         }
1412                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1413                         struct btrfs_chunk *chunk;
1414                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1415                         ret = read_one_chunk(root, &found_key, leaf, chunk);
1416                 }
1417                 path->slots[0]++;
1418         }
1419         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1420                 key.objectid = 0;
1421                 btrfs_release_path(root, path);
1422                 goto again;
1423         }
1424
1425         btrfs_free_path(path);
1426         ret = 0;
1427 error:
1428         return ret;
1429 }
1430