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 <asm/div64.h>
  23 #include "ctree.h"
  24 #include "extent_map.h"
  25 #include "disk-io.h"
  26 #include "transaction.h"
  27 #include "print-tree.h"
  28 #include "volumes.h"
  29
  30 struct map_lookup {
  31         u64 type;
  32         int io_align;
  33         int io_width;
  34         int stripe_len;
  35         int sector_size;
  36         int num_stripes;
  37         int sub_stripes;
  38         struct btrfs_bio_stripe stripes[];
  39 };
  40
  41 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  42                             (sizeof(struct btrfs_bio_stripe) * (n)))
  43
  44 static DEFINE_MUTEX(uuid_mutex);
  45 static LIST_HEAD(fs_uuids);
  46
  47 int btrfs_cleanup_fs_uuids(void)
  48 {
  49         struct btrfs_fs_devices *fs_devices;
  50         struct list_head *uuid_cur;
  51         struct list_head *devices_cur;
  52         struct btrfs_device *dev;
  53
  54         list_for_each(uuid_cur, &fs_uuids) {
  55                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  56                                         list);
  57                 while(!list_empty(&fs_devices->devices)) {
  58                         devices_cur = fs_devices->devices.next;
  59                         dev = list_entry(devices_cur, struct btrfs_device,
  60                                          dev_list);
  61                         if (dev->bdev) {
  62                                 close_bdev_excl(dev->bdev);
  63                         }
  64                         list_del(&dev->dev_list);
  65                         kfree(dev);
  66                 }
  67         }
  68         return 0;
  69 }
  70
  71 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  72                                           u8 *uuid)
  73 {
  74         struct btrfs_device *dev;
  75         struct list_head *cur;
  76
  77         list_for_each(cur, head) {
  78                 dev = list_entry(cur, struct btrfs_device, dev_list);
  79                 if (dev->devid == devid &&
  80                     !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
  81                         return dev;
  82                 }
  83         }
  84         return NULL;
  85 }
  86
  87 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  88 {
  89         struct list_head *cur;
  90         struct btrfs_fs_devices *fs_devices;
  91
  92         list_for_each(cur, &fs_uuids) {
  93                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  94                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  95                         return fs_devices;
  96         }
  97         return NULL;
  98 }
  99
 100 static int device_list_add(const char *path,
 101                            struct btrfs_super_block *disk_super,
 102                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 103 {
 104         struct btrfs_device *device;
 105         struct btrfs_fs_devices *fs_devices;
 106         u64 found_transid = btrfs_super_generation(disk_super);
 107
 108         fs_devices = find_fsid(disk_super->fsid);
 109         if (!fs_devices) {
 110                 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
 111                 if (!fs_devices)
 112                         return -ENOMEM;
 113                 INIT_LIST_HEAD(&fs_devices->devices);
 114                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 115                 list_add(&fs_devices->list, &fs_uuids);
 116                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 117                 fs_devices->latest_devid = devid;
 118                 fs_devices->latest_trans = found_transid;
 119                 fs_devices->lowest_devid = (u64)-1;
 120                 fs_devices->num_devices = 0;
 121                 device = NULL;
 122         } else {
 123                 device = __find_device(&fs_devices->devices, devid,
 124                                        disk_super->dev_item.uuid);
 125         }
 126         if (!device) {
 127                 device = kzalloc(sizeof(*device), GFP_NOFS);
 128                 if (!device) {
 129                         /* we can safely leave the fs_devices entry around */
 130                         return -ENOMEM;
 131                 }
 132                 device->devid = devid;
 133                 memcpy(device->uuid, disk_super->dev_item.uuid,
 134                        BTRFS_UUID_SIZE);
 135                 device->barriers = 1;
 136                 spin_lock_init(&device->io_lock);
 137                 device->name = kstrdup(path, GFP_NOFS);
 138                 if (!device->name) {
 139                         kfree(device);
 140                         return -ENOMEM;
 141                 }
 142                 list_add(&device->dev_list, &fs_devices->devices);
 143                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 144                 fs_devices->num_devices++;
 145         }
 146
 147         if (found_transid > fs_devices->latest_trans) {
 148                 fs_devices->latest_devid = devid;
 149                 fs_devices->latest_trans = found_transid;
 150         }
 151         if (fs_devices->lowest_devid > devid) {
 152                 fs_devices->lowest_devid = devid;
 153         }
 154         *fs_devices_ret = fs_devices;
 155         return 0;
 156 }
 157
 158 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 159 {
 160         struct list_head *head = &fs_devices->devices;
 161         struct list_head *cur;
 162         struct btrfs_device *device;
 163
 164         mutex_lock(&uuid_mutex);
 165         list_for_each(cur, head) {
 166                 device = list_entry(cur, struct btrfs_device, dev_list);
 167                 if (device->bdev) {
 168                         close_bdev_excl(device->bdev);
 169                 }
 170                 device->bdev = NULL;
 171         }
 172         mutex_unlock(&uuid_mutex);
 173         return 0;
 174 }
 175
 176 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 177                        int flags, void *holder)
 178 {
 179         struct block_device *bdev;
 180         struct list_head *head = &fs_devices->devices;
 181         struct list_head *cur;
 182         struct btrfs_device *device;
 183         int ret;
 184
 185         mutex_lock(&uuid_mutex);
 186         list_for_each(cur, head) {
 187                 device = list_entry(cur, struct btrfs_device, dev_list);
 188                 bdev = open_bdev_excl(device->name, flags, holder);
 189
 190                 if (IS_ERR(bdev)) {
 191                         printk("open %s failed\n", device->name);
 192                         ret = PTR_ERR(bdev);
 193                         goto fail;
 194                 }
 195                 if (device->devid == fs_devices->latest_devid)
 196                         fs_devices->latest_bdev = bdev;
 197                 if (device->devid == fs_devices->lowest_devid) {
 198                         fs_devices->lowest_bdev = bdev;
 199                 }
 200                 device->bdev = bdev;
 201         }
 202         mutex_unlock(&uuid_mutex);
 203         return 0;
 204 fail:
 205         mutex_unlock(&uuid_mutex);
 206         btrfs_close_devices(fs_devices);
 207         return ret;
 208 }
 209
 210 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 211                           struct btrfs_fs_devices **fs_devices_ret)
 212 {
 213         struct btrfs_super_block *disk_super;
 214         struct block_device *bdev;
 215         struct buffer_head *bh;
 216         int ret;
 217         u64 devid;
 218         u64 transid;
 219
 220         mutex_lock(&uuid_mutex);
 221
 222         bdev = open_bdev_excl(path, flags, holder);
 223
 224         if (IS_ERR(bdev)) {
 225                 ret = PTR_ERR(bdev);
 226                 goto error;
 227         }
 228
 229         ret = set_blocksize(bdev, 4096);
 230         if (ret)
 231                 goto error_close;
 232         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 233         if (!bh) {
 234                 ret = -EIO;
 235                 goto error_close;
 236         }
 237         disk_super = (struct btrfs_super_block *)bh->b_data;
 238         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 239             sizeof(disk_super->magic))) {
 240                 ret = -EINVAL;
 241                 goto error_brelse;
 242         }
 243         devid = le64_to_cpu(disk_super->dev_item.devid);
 244         transid = btrfs_super_generation(disk_super);
 245         if (disk_super->label[0])
 246                 printk("device label %s ", disk_super->label);
 247         else {
 248                 /* FIXME, make a readl uuid parser */
 249                 printk("device fsid %llx-%llx ",
 250                        *(unsigned long long *)disk_super->fsid,
 251                        *(unsigned long long *)(disk_super->fsid + 8));
 252         }
 253         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 254         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 255
 256 error_brelse:
 257         brelse(bh);
 258 error_close:
 259         close_bdev_excl(bdev);
 260 error:
 261         mutex_unlock(&uuid_mutex);
 262         return ret;
 263 }
 264
 265 /*
 266  * this uses a pretty simple search, the expectation is that it is
 267  * called very infrequently and that a given device has a small number
 268  * of extents
 269  */
 270 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 271                                 struct btrfs_device *device,
 272                                 struct btrfs_path *path,
 273                                 u64 num_bytes, u64 *start)
 274 {
 275         struct btrfs_key key;
 276         struct btrfs_root *root = device->dev_root;
 277         struct btrfs_dev_extent *dev_extent = NULL;
 278         u64 hole_size = 0;
 279         u64 last_byte = 0;
 280         u64 search_start = 0;
 281         u64 search_end = device->total_bytes;
 282         int ret;
 283         int slot = 0;
 284         int start_found;
 285         struct extent_buffer *l;
 286
 287         start_found = 0;
 288         path->reada = 2;
 289
 290         /* FIXME use last free of some kind */
 291
 292         /* we don't want to overwrite the superblock on the drive,
 293          * so we make sure to start at an offset of at least 1MB
 294          */
 295         search_start = max((u64)1024 * 1024, search_start);
 296         key.objectid = device->devid;
 297         key.offset = search_start;
 298         key.type = BTRFS_DEV_EXTENT_KEY;
 299         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 300         if (ret < 0)
 301                 goto error;
 302         ret = btrfs_previous_item(root, path, 0, key.type);
 303         if (ret < 0)
 304                 goto error;
 305         l = path->nodes[0];
 306         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 307         while (1) {
 308                 l = path->nodes[0];
 309                 slot = path->slots[0];
 310                 if (slot >= btrfs_header_nritems(l)) {
 311                         ret = btrfs_next_leaf(root, path);
 312                         if (ret == 0)
 313                                 continue;
 314                         if (ret < 0)
 315                                 goto error;
 316 no_more_items:
 317                         if (!start_found) {
 318                                 if (search_start >= search_end) {
 319                                         ret = -ENOSPC;
 320                                         goto error;
 321                                 }
 322                                 *start = search_start;
 323                                 start_found = 1;
 324                                 goto check_pending;
 325                         }
 326                         *start = last_byte > search_start ?
 327                                 last_byte : search_start;
 328                         if (search_end <= *start) {
 329                                 ret = -ENOSPC;
 330                                 goto error;
 331                         }
 332                         goto check_pending;
 333                 }
 334                 btrfs_item_key_to_cpu(l, &key, slot);
 335
 336                 if (key.objectid < device->devid)
 337                         goto next;
 338
 339                 if (key.objectid > device->devid)
 340                         goto no_more_items;
 341
 342                 if (key.offset >= search_start && key.offset > last_byte &&
 343                     start_found) {
 344                         if (last_byte < search_start)
 345                                 last_byte = search_start;
 346                         hole_size = key.offset - last_byte;
 347                         if (key.offset > last_byte &&
 348                             hole_size >= num_bytes) {
 349                                 *start = last_byte;
 350                                 goto check_pending;
 351                         }
 352                 }
 353                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 354                         goto next;
 355                 }
 356
 357                 start_found = 1;
 358                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 359                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 360 next:
 361                 path->slots[0]++;
 362                 cond_resched();
 363         }
 364 check_pending:
 365         /* we have to make sure we didn't find an extent that has already
 366          * been allocated by the map tree or the original allocation
 367          */
 368         btrfs_release_path(root, path);
 369         BUG_ON(*start < search_start);
 370
 371         if (*start + num_bytes > search_end) {
 372                 ret = -ENOSPC;
 373                 goto error;
 374         }
 375         /* check for pending inserts here */
 376         return 0;
 377
 378 error:
 379         btrfs_release_path(root, path);
 380         return ret;
 381 }
 382
 383 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 384                            struct btrfs_device *device,
 385                            u64 chunk_tree, u64 chunk_objectid,
 386                            u64 chunk_offset,
 387                            u64 num_bytes, u64 *start)
 388 {
 389         int ret;
 390         struct btrfs_path *path;
 391         struct btrfs_root *root = device->dev_root;
 392         struct btrfs_dev_extent *extent;
 393         struct extent_buffer *leaf;
 394         struct btrfs_key key;
 395
 396         path = btrfs_alloc_path();
 397         if (!path)
 398                 return -ENOMEM;
 399
 400         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 401         if (ret) {
 402                 goto err;
 403         }
 404
 405         key.objectid = device->devid;
 406         key.offset = *start;
 407         key.type = BTRFS_DEV_EXTENT_KEY;
 408         ret = btrfs_insert_empty_item(trans, root, path, &key,
 409                                       sizeof(*extent));
 410         BUG_ON(ret);
 411
 412         leaf = path->nodes[0];
 413         extent = btrfs_item_ptr(leaf, path->slots[0],
 414                                 struct btrfs_dev_extent);
 415         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 416         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 417         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 418
 419         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 420                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 421                     BTRFS_UUID_SIZE);
 422
 423         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 424         btrfs_mark_buffer_dirty(leaf);
 425 err:
 426         btrfs_free_path(path);
 427         return ret;
 428 }
 429
 430 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 431 {
 432         struct btrfs_path *path;
 433         int ret;
 434         struct btrfs_key key;
 435         struct btrfs_chunk *chunk;
 436         struct btrfs_key found_key;
 437
 438         path = btrfs_alloc_path();
 439         BUG_ON(!path);
 440
 441         key.objectid = objectid;
 442         key.offset = (u64)-1;
 443         key.type = BTRFS_CHUNK_ITEM_KEY;
 444
 445         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 446         if (ret < 0)
 447                 goto error;
 448
 449         BUG_ON(ret == 0);
 450
 451         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 452         if (ret) {
 453                 *offset = 0;
 454         } else {
 455                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 456                                       path->slots[0]);
 457                 if (found_key.objectid != objectid)
 458                         *offset = 0;
 459                 else {
 460                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 461                                                struct btrfs_chunk);
 462                         *offset = found_key.offset +
 463                                 btrfs_chunk_length(path->nodes[0], chunk);
 464                 }
 465         }
 466         ret = 0;
 467 error:
 468         btrfs_free_path(path);
 469         return ret;
 470 }
 471
 472 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 473                            u64 *objectid)
 474 {
 475         int ret;
 476         struct btrfs_key key;
 477         struct btrfs_key found_key;
 478
 479         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 480         key.type = BTRFS_DEV_ITEM_KEY;
 481         key.offset = (u64)-1;
 482
 483         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 484         if (ret < 0)
 485                 goto error;
 486
 487         BUG_ON(ret == 0);
 488
 489         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 490                                   BTRFS_DEV_ITEM_KEY);
 491         if (ret) {
 492                 *objectid = 1;
 493         } else {
 494                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 495                                       path->slots[0]);
 496                 *objectid = found_key.offset + 1;
 497         }
 498         ret = 0;
 499 error:
 500         btrfs_release_path(root, path);
 501         return ret;
 502 }
 503
 504 /*
 505  * the device information is stored in the chunk root
 506  * the btrfs_device struct should be fully filled in
 507  */
 508 int btrfs_add_device(struct btrfs_trans_handle *trans,
 509                      struct btrfs_root *root,
 510                      struct btrfs_device *device)
 511 {
 512         int ret;
 513         struct btrfs_path *path;
 514         struct btrfs_dev_item *dev_item;
 515         struct extent_buffer *leaf;
 516         struct btrfs_key key;
 517         unsigned long ptr;
 518         u64 free_devid;
 519
 520         root = root->fs_info->chunk_root;
 521
 522         path = btrfs_alloc_path();
 523         if (!path)
 524                 return -ENOMEM;
 525
 526         ret = find_next_devid(root, path, &free_devid);
 527         if (ret)
 528                 goto out;
 529
 530         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 531         key.type = BTRFS_DEV_ITEM_KEY;
 532         key.offset = free_devid;
 533
 534         ret = btrfs_insert_empty_item(trans, root, path, &key,
 535                                       sizeof(*dev_item));
 536         if (ret)
 537                 goto out;
 538
 539         leaf = path->nodes[0];
 540         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 541
 542         device->devid = free_devid;
 543         btrfs_set_device_id(leaf, dev_item, device->devid);
 544         btrfs_set_device_type(leaf, dev_item, device->type);
 545         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 546         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 547         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 548         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 549         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 550         btrfs_set_device_group(leaf, dev_item, 0);
 551         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 552         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 553
 554         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 555         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 556         btrfs_mark_buffer_dirty(leaf);
 557         ret = 0;
 558
 559 out:
 560         btrfs_free_path(path);
 561         return ret;
 562 }
 563 int btrfs_update_device(struct btrfs_trans_handle *trans,
 564                         struct btrfs_device *device)
 565 {
 566         int ret;
 567         struct btrfs_path *path;
 568         struct btrfs_root *root;
 569         struct btrfs_dev_item *dev_item;
 570         struct extent_buffer *leaf;
 571         struct btrfs_key key;
 572
 573         root = device->dev_root->fs_info->chunk_root;
 574
 575         path = btrfs_alloc_path();
 576         if (!path)
 577                 return -ENOMEM;
 578
 579         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 580         key.type = BTRFS_DEV_ITEM_KEY;
 581         key.offset = device->devid;
 582
 583         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 584         if (ret < 0)
 585                 goto out;
 586
 587         if (ret > 0) {
 588                 ret = -ENOENT;
 589                 goto out;
 590         }
 591
 592         leaf = path->nodes[0];
 593         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 594
 595         btrfs_set_device_id(leaf, dev_item, device->devid);
 596         btrfs_set_device_type(leaf, dev_item, device->type);
 597         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 598         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 599         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 600         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 601         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 602         btrfs_mark_buffer_dirty(leaf);
 603
 604 out:
 605         btrfs_free_path(path);
 606         return ret;
 607 }
 608
 609 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 610                            struct btrfs_root *root,
 611                            struct btrfs_key *key,
 612                            struct btrfs_chunk *chunk, int item_size)
 613 {
 614         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 615         struct btrfs_disk_key disk_key;
 616         u32 array_size;
 617         u8 *ptr;
 618
 619         array_size = btrfs_super_sys_array_size(super_copy);
 620         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 621                 return -EFBIG;
 622
 623         ptr = super_copy->sys_chunk_array + array_size;
 624         btrfs_cpu_key_to_disk(&disk_key, key);
 625         memcpy(ptr, &disk_key, sizeof(disk_key));
 626         ptr += sizeof(disk_key);
 627         memcpy(ptr, chunk, item_size);
 628         item_size += sizeof(disk_key);
 629         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 630         return 0;
 631 }
 632
 633 static u64 div_factor(u64 num, int factor)
 634 {
 635         if (factor == 10)
 636                 return num;
 637         num *= factor;
 638         do_div(num, 10);
 639         return num;
 640 }
 641
 642 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 643                                int sub_stripes)
 644 {
 645         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 646                 return calc_size;
 647         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 648                 return calc_size * (num_stripes / sub_stripes);
 649         else
 650                 return calc_size * num_stripes;
 651 }
 652
 653
 654 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 655                       struct btrfs_root *extent_root, u64 *start,
 656                       u64 *num_bytes, u64 type)
 657 {
 658         u64 dev_offset;
 659         struct btrfs_fs_info *info = extent_root->fs_info;
 660         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
 661         struct btrfs_stripe *stripes;
 662         struct btrfs_device *device = NULL;
 663         struct btrfs_chunk *chunk;
 664         struct list_head private_devs;
 665         struct list_head *dev_list;
 666         struct list_head *cur;
 667         struct extent_map_tree *em_tree;
 668         struct map_lookup *map;
 669         struct extent_map *em;
 670         int min_stripe_size = 1 * 1024 * 1024;
 671         u64 physical;
 672         u64 calc_size = 1024 * 1024 * 1024;
 673         u64 max_chunk_size = calc_size;
 674         u64 min_free;
 675         u64 avail;
 676         u64 max_avail = 0;
 677         u64 percent_max;
 678         int num_stripes = 1;
 679         int min_stripes = 1;
 680         int sub_stripes = 0;
 681         int looped = 0;
 682         int ret;
 683         int index;
 684         int stripe_len = 64 * 1024;
 685         struct btrfs_key key;
 686
 687         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
 688         if (list_empty(dev_list))
 689                 return -ENOSPC;
 690
 691         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
 692                 num_stripes = btrfs_super_num_devices(&info->super_copy);
 693                 min_stripes = 2;
 694         }
 695         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
 696                 num_stripes = 2;
 697                 min_stripes = 2;
 698         }
 699         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
 700                 num_stripes = min_t(u64, 2,
 701                                   btrfs_super_num_devices(&info->super_copy));
 702                 if (num_stripes < 2)
 703                         return -ENOSPC;
 704                 min_stripes = 2;
 705         }
 706         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 707                 num_stripes = btrfs_super_num_devices(&info->super_copy);
 708                 if (num_stripes < 4)
 709                         return -ENOSPC;
 710                 num_stripes &= ~(u32)1;
 711                 sub_stripes = 2;
 712                 min_stripes = 4;
 713         }
 714
 715         if (type & BTRFS_BLOCK_GROUP_DATA) {
 716                 max_chunk_size = 10 * calc_size;
 717                 min_stripe_size = 64 * 1024 * 1024;
 718         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 719                 max_chunk_size = 4 * calc_size;
 720                 min_stripe_size = 32 * 1024 * 1024;
 721         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 722                 calc_size = 8 * 1024 * 1024;
 723                 max_chunk_size = calc_size * 2;
 724                 min_stripe_size = 1 * 1024 * 1024;
 725         }
 726
 727         /* we don't want a chunk larger than 10% of the FS */
 728         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
 729         max_chunk_size = min(percent_max, max_chunk_size);
 730
 731 again:
 732         if (calc_size * num_stripes > max_chunk_size) {
 733                 calc_size = max_chunk_size;
 734                 do_div(calc_size, num_stripes);
 735                 do_div(calc_size, stripe_len);
 736                 calc_size *= stripe_len;
 737         }
 738         /* we don't want tiny stripes */
 739         calc_size = max_t(u64, min_stripe_size, calc_size);
 740
 741         do_div(calc_size, stripe_len);
 742         calc_size *= stripe_len;
 743
 744         INIT_LIST_HEAD(&private_devs);
 745         cur = dev_list->next;
 746         index = 0;
 747
 748         if (type & BTRFS_BLOCK_GROUP_DUP)
 749                 min_free = calc_size * 2;
 750         else
 751                 min_free = calc_size;
 752
 753         /* we add 1MB because we never use the first 1MB of the device */
 754         min_free += 1024 * 1024;
 755
 756         /* build a private list of devices we will allocate from */
 757         while(index < num_stripes) {
 758                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
 759
 760                 avail = device->total_bytes - device->bytes_used;
 761                 cur = cur->next;
 762                 if (avail >= min_free) {
 763                         list_move_tail(&device->dev_alloc_list, &private_devs);
 764                         index++;
 765                         if (type & BTRFS_BLOCK_GROUP_DUP)
 766                                 index++;
 767                 } else if (avail > max_avail)
 768                         max_avail = avail;
 769                 if (cur == dev_list)
 770                         break;
 771         }
 772         if (index < num_stripes) {
 773                 list_splice(&private_devs, dev_list);
 774                 if (index >= min_stripes) {
 775                         num_stripes = index;
 776                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 777                                 num_stripes /= sub_stripes;
 778                                 num_stripes *= sub_stripes;
 779                         }
 780                         looped = 1;
 781                         goto again;
 782                 }
 783                 if (!looped && max_avail > 0) {
 784                         looped = 1;
 785                         calc_size = max_avail;
 786                         goto again;
 787                 }
 788                 return -ENOSPC;
 789         }
 790         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 791         key.type = BTRFS_CHUNK_ITEM_KEY;
 792         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 793                               &key.offset);
 794         if (ret)
 795                 return ret;
 796
 797         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 798         if (!chunk)
 799                 return -ENOMEM;
 800
 801         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
 802         if (!map) {
 803                 kfree(chunk);
 804                 return -ENOMEM;
 805         }
 806
 807         stripes = &chunk->stripe;
 808         *num_bytes = chunk_bytes_by_type(type, calc_size,
 809                                          num_stripes, sub_stripes);
 810
 811
 812         index = 0;
 813 printk("new chunk type %Lu start %Lu size %Lu\n", type, key.offset, *num_bytes);
 814         while(index < num_stripes) {
 815                 struct btrfs_stripe *stripe;
 816                 BUG_ON(list_empty(&private_devs));
 817                 cur = private_devs.next;
 818                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
 819
 820                 /* loop over this device again if we're doing a dup group */
 821                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 822                     (index == num_stripes - 1))
 823                         list_move_tail(&device->dev_alloc_list, dev_list);
 824
 825                 ret = btrfs_alloc_dev_extent(trans, device,
 826                              info->chunk_root->root_key.objectid,
 827                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 828                              calc_size, &dev_offset);
 829                 BUG_ON(ret);
 830 printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.offset, calc_size, device->devid, type);
 831                 device->bytes_used += calc_size;
 832                 ret = btrfs_update_device(trans, device);
 833                 BUG_ON(ret);
 834
 835                 map->stripes[index].dev = device;
 836                 map->stripes[index].physical = dev_offset;
 837                 stripe = stripes + index;
 838                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 839                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 840                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 841                 physical = dev_offset;
 842                 index++;
 843         }
 844         BUG_ON(!list_empty(&private_devs));
 845
 846         /* key was set above */
 847         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 848         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 849         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 850         btrfs_set_stack_chunk_type(chunk, type);
 851         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 852         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 853         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 854         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 855         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
 856         map->sector_size = extent_root->sectorsize;
 857         map->stripe_len = stripe_len;
 858         map->io_align = stripe_len;
 859         map->io_width = stripe_len;
 860         map->type = type;
 861         map->num_stripes = num_stripes;
 862         map->sub_stripes = sub_stripes;
 863
 864         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 865                                 btrfs_chunk_item_size(num_stripes));
 866         BUG_ON(ret);
 867         *start = key.offset;;
 868
 869         em = alloc_extent_map(GFP_NOFS);
 870         if (!em)
 871                 return -ENOMEM;
 872         em->bdev = (struct block_device *)map;
 873         em->start = key.offset;
 874         em->len = *num_bytes;
 875         em->block_start = 0;
 876
 877         kfree(chunk);
 878
 879         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
 880         spin_lock(&em_tree->lock);
 881         ret = add_extent_mapping(em_tree, em);
 882         spin_unlock(&em_tree->lock);
 883         BUG_ON(ret);
 884         free_extent_map(em);
 885         return ret;
 886 }
 887
 888 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
 889 {
 890         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
 891 }
 892
 893 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
 894 {
 895         struct extent_map *em;
 896
 897         while(1) {
 898                 spin_lock(&tree->map_tree.lock);
 899                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
 900                 if (em)
 901                         remove_extent_mapping(&tree->map_tree, em);
 902                 spin_unlock(&tree->map_tree.lock);
 903                 if (!em)
 904                         break;
 905                 kfree(em->bdev);
 906                 /* once for us */
 907                 free_extent_map(em);
 908                 /* once for the tree */
 909                 free_extent_map(em);
 910         }
 911 }
 912
 913 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
 914 {
 915         struct extent_map *em;
 916         struct map_lookup *map;
 917         struct extent_map_tree *em_tree = &map_tree->map_tree;
 918         int ret;
 919
 920         spin_lock(&em_tree->lock);
 921         em = lookup_extent_mapping(em_tree, logical, len);
 922         spin_unlock(&em_tree->lock);
 923         BUG_ON(!em);
 924
 925         BUG_ON(em->start > logical || em->start + em->len < logical);
 926         map = (struct map_lookup *)em->bdev;
 927         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
 928                 ret = map->num_stripes;
 929         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
 930                 ret = map->sub_stripes;
 931         else
 932                 ret = 1;
 933         free_extent_map(em);
 934         return ret;
 935 }
 936
 937 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
 938                              u64 logical, u64 *length,
 939                              struct btrfs_multi_bio **multi_ret,
 940                              int mirror_num, struct page *unplug_page)
 941 {
 942         struct extent_map *em;
 943         struct map_lookup *map;
 944         struct extent_map_tree *em_tree = &map_tree->map_tree;
 945         u64 offset;
 946         u64 stripe_offset;
 947         u64 stripe_nr;
 948         int stripes_allocated = 8;
 949         int stripes_required = 1;
 950         int stripe_index;
 951         int i;
 952         int num_stripes;
 953         struct btrfs_multi_bio *multi = NULL;
 954
 955         if (multi_ret && !(rw & (1 << BIO_RW))) {
 956                 stripes_allocated = 1;
 957         }
 958 again:
 959         if (multi_ret) {
 960                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
 961                                 GFP_NOFS);
 962                 if (!multi)
 963                         return -ENOMEM;
 964         }
 965
 966         spin_lock(&em_tree->lock);
 967         em = lookup_extent_mapping(em_tree, logical, *length);
 968         spin_unlock(&em_tree->lock);
 969
 970         if (!em && unplug_page)
 971                 return 0;
 972
 973         if (!em) {
 974                 printk("unable to find logical %Lu\n", logical);
 975                 BUG();
 976         }
 977
 978         BUG_ON(em->start > logical || em->start + em->len < logical);
 979         map = (struct map_lookup *)em->bdev;
 980         offset = logical - em->start;
 981
 982         if (mirror_num > map->num_stripes)
 983                 mirror_num = 0;
 984
 985         /* if our multi bio struct is too small, back off and try again */
 986         if (rw & (1 << BIO_RW)) {
 987                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
 988                                  BTRFS_BLOCK_GROUP_DUP)) {
 989                         stripes_required = map->num_stripes;
 990                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
 991                         stripes_required = map->sub_stripes;
 992                 }
 993         }
 994         if (multi_ret && rw == WRITE &&
 995             stripes_allocated < stripes_required) {
 996                 stripes_allocated = map->num_stripes;
 997                 free_extent_map(em);
 998                 kfree(multi);
 999                 goto again;
1000         }
1001         stripe_nr = offset;
1002         /*
1003          * stripe_nr counts the total number of stripes we have to stride
1004          * to get to this block
1005          */
1006         do_div(stripe_nr, map->stripe_len);
1007
1008         stripe_offset = stripe_nr * map->stripe_len;
1009         BUG_ON(offset < stripe_offset);
1010
1011         /* stripe_offset is the offset of this block in its stripe*/
1012         stripe_offset = offset - stripe_offset;
1013
1014         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1015                          BTRFS_BLOCK_GROUP_RAID10 |
1016                          BTRFS_BLOCK_GROUP_DUP)) {
1017                 /* we limit the length of each bio to what fits in a stripe */
1018                 *length = min_t(u64, em->len - offset,
1019                               map->stripe_len - stripe_offset);
1020         } else {
1021                 *length = em->len - offset;
1022         }
1023
1024         if (!multi_ret && !unplug_page)
1025                 goto out;
1026
1027         num_stripes = 1;
1028         stripe_index = 0;
1029         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1030                 if (unplug_page || (rw & (1 << BIO_RW)))
1031                         num_stripes = map->num_stripes;
1032                 else if (mirror_num) {
1033                         stripe_index = mirror_num - 1;
1034                 } else {
1035                         u64 orig_stripe_nr = stripe_nr;
1036                         stripe_index = do_div(orig_stripe_nr, num_stripes);
1037                 }
1038         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1039                 if (rw & (1 << BIO_RW))
1040                         num_stripes = map->num_stripes;
1041                 else if (mirror_num)
1042                         stripe_index = mirror_num - 1;
1043         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1044                 int factor = map->num_stripes / map->sub_stripes;
1045
1046                 stripe_index = do_div(stripe_nr, factor);
1047                 stripe_index *= map->sub_stripes;
1048
1049                 if (unplug_page || (rw & (1 << BIO_RW)))
1050                         num_stripes = map->sub_stripes;
1051                 else if (mirror_num)
1052                         stripe_index += mirror_num - 1;
1053                 else {
1054                         u64 orig_stripe_nr = stripe_nr;
1055                         stripe_index += do_div(orig_stripe_nr,
1056                                                map->sub_stripes);
1057                 }
1058         } else {
1059                 /*
1060                  * after this do_div call, stripe_nr is the number of stripes
1061                  * on this device we have to walk to find the data, and
1062                  * stripe_index is the number of our device in the stripe array
1063                  */
1064                 stripe_index = do_div(stripe_nr, map->num_stripes);
1065         }
1066         BUG_ON(stripe_index >= map->num_stripes);
1067
1068         for (i = 0; i < num_stripes; i++) {
1069                 if (unplug_page) {
1070                         struct btrfs_device *device;
1071                         struct backing_dev_info *bdi;
1072
1073                         device = map->stripes[stripe_index].dev;
1074                         bdi = blk_get_backing_dev_info(device->bdev);
1075                         if (bdi->unplug_io_fn) {
1076                                 bdi->unplug_io_fn(bdi, unplug_page);
1077                         }
1078                 } else {
1079                         multi->stripes[i].physical =
1080                                 map->stripes[stripe_index].physical +
1081                                 stripe_offset + stripe_nr * map->stripe_len;
1082                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
1083                 }
1084                 stripe_index++;
1085         }
1086         if (multi_ret) {
1087                 *multi_ret = multi;
1088                 multi->num_stripes = num_stripes;
1089         }
1090 out:
1091         free_extent_map(em);
1092         return 0;
1093 }
1094
1095 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1096                       u64 logical, u64 *length,
1097                       struct btrfs_multi_bio **multi_ret, int mirror_num)
1098 {
1099         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1100                                  mirror_num, NULL);
1101 }
1102
1103 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1104                       u64 logical, struct page *page)
1105 {
1106         u64 length = PAGE_CACHE_SIZE;
1107         return __btrfs_map_block(map_tree, READ, logical, &length,
1108                                  NULL, 0, page);
1109 }
1110
1111
1112 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1113 static void end_bio_multi_stripe(struct bio *bio, int err)
1114 #else
1115 static int end_bio_multi_stripe(struct bio *bio,
1116                                    unsigned int bytes_done, int err)
1117 #endif
1118 {
1119         struct btrfs_multi_bio *multi = bio->bi_private;
1120
1121 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1122         if (bio->bi_size)
1123                 return 1;
1124 #endif
1125         if (err)
1126                 multi->error = err;
1127
1128         if (atomic_dec_and_test(&multi->stripes_pending)) {
1129                 bio->bi_private = multi->private;
1130                 bio->bi_end_io = multi->end_io;
1131
1132                 if (!err && multi->error)
1133                         err = multi->error;
1134                 kfree(multi);
1135
1136 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1137                 bio_endio(bio, bio->bi_size, err);
1138 #else
1139                 bio_endio(bio, err);
1140 #endif
1141         } else {
1142                 bio_put(bio);
1143         }
1144 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1145         return 0;
1146 #endif
1147 }
1148
1149 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1150                   int mirror_num)
1151 {
1152         struct btrfs_mapping_tree *map_tree;
1153         struct btrfs_device *dev;
1154         struct bio *first_bio = bio;
1155         u64 logical = bio->bi_sector << 9;
1156         u64 length = 0;
1157         u64 map_length;
1158         struct btrfs_multi_bio *multi = NULL;
1159         int ret;
1160         int dev_nr = 0;
1161         int total_devs = 1;
1162
1163         length = bio->bi_size;
1164
1165         map_tree = &root->fs_info->mapping_tree;
1166         map_length = length;
1167
1168         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1169                               mirror_num);
1170         BUG_ON(ret);
1171
1172         total_devs = multi->num_stripes;
1173         if (map_length < length) {
1174                 printk("mapping failed logical %Lu bio len %Lu "
1175                        "len %Lu\n", logical, length, map_length);
1176                 BUG();
1177         }
1178         multi->end_io = first_bio->bi_end_io;
1179         multi->private = first_bio->bi_private;
1180         atomic_set(&multi->stripes_pending, multi->num_stripes);
1181
1182         while(dev_nr < total_devs) {
1183                 if (total_devs > 1) {
1184                         if (dev_nr < total_devs - 1) {
1185                                 bio = bio_clone(first_bio, GFP_NOFS);
1186                                 BUG_ON(!bio);
1187                         } else {
1188                                 bio = first_bio;
1189                         }
1190                         bio->bi_private = multi;
1191                         bio->bi_end_io = end_bio_multi_stripe;
1192                 }
1193                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1194                 dev = multi->stripes[dev_nr].dev;
1195                 bio->bi_bdev = dev->bdev;
1196                 spin_lock(&dev->io_lock);
1197                 dev->total_ios++;
1198                 spin_unlock(&dev->io_lock);
1199                 submit_bio(rw, bio);
1200                 dev_nr++;
1201         }
1202         if (total_devs == 1)
1203                 kfree(multi);
1204         return 0;
1205 }
1206
1207 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1208                                        u8 *uuid)
1209 {
1210         struct list_head *head = &root->fs_info->fs_devices->devices;
1211
1212         return __find_device(head, devid, uuid);
1213 }
1214
1215 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1216                           struct extent_buffer *leaf,
1217                           struct btrfs_chunk *chunk)
1218 {
1219         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1220         struct map_lookup *map;
1221         struct extent_map *em;
1222         u64 logical;
1223         u64 length;
1224         u64 devid;
1225         u8 uuid[BTRFS_UUID_SIZE];
1226         int num_stripes;
1227         int ret;
1228         int i;
1229
1230         logical = key->offset;
1231         length = btrfs_chunk_length(leaf, chunk);
1232         spin_lock(&map_tree->map_tree.lock);
1233         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1234         spin_unlock(&map_tree->map_tree.lock);
1235
1236         /* already mapped? */
1237         if (em && em->start <= logical && em->start + em->len > logical) {
1238                 free_extent_map(em);
1239                 return 0;
1240         } else if (em) {
1241                 free_extent_map(em);
1242         }
1243
1244         map = kzalloc(sizeof(*map), GFP_NOFS);
1245         if (!map)
1246                 return -ENOMEM;
1247
1248         em = alloc_extent_map(GFP_NOFS);
1249         if (!em)
1250                 return -ENOMEM;
1251         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1252         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1253         if (!map) {
1254                 free_extent_map(em);
1255                 return -ENOMEM;
1256         }
1257
1258         em->bdev = (struct block_device *)map;
1259         em->start = logical;
1260         em->len = length;
1261         em->block_start = 0;
1262
1263         map->num_stripes = num_stripes;
1264         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1265         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1266         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1267         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1268         map->type = btrfs_chunk_type(leaf, chunk);
1269         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1270         for (i = 0; i < num_stripes; i++) {
1271                 map->stripes[i].physical =
1272                         btrfs_stripe_offset_nr(leaf, chunk, i);
1273                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1274                 read_extent_buffer(leaf, uuid, (unsigned long)
1275                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1276                                    BTRFS_UUID_SIZE);
1277                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1278                 if (!map->stripes[i].dev) {
1279                         kfree(map);
1280                         free_extent_map(em);
1281                         return -EIO;
1282                 }
1283         }
1284
1285         spin_lock(&map_tree->map_tree.lock);
1286         ret = add_extent_mapping(&map_tree->map_tree, em);
1287         spin_unlock(&map_tree->map_tree.lock);
1288         BUG_ON(ret);
1289         free_extent_map(em);
1290
1291         return 0;
1292 }
1293
1294 static int fill_device_from_item(struct extent_buffer *leaf,
1295                                  struct btrfs_dev_item *dev_item,
1296                                  struct btrfs_device *device)
1297 {
1298         unsigned long ptr;
1299
1300         device->devid = btrfs_device_id(leaf, dev_item);
1301         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1302         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1303         device->type = btrfs_device_type(leaf, dev_item);
1304         device->io_align = btrfs_device_io_align(leaf, dev_item);
1305         device->io_width = btrfs_device_io_width(leaf, dev_item);
1306         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1307
1308         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1309         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1310
1311         return 0;
1312 }
1313
1314 static int read_one_dev(struct btrfs_root *root,
1315                         struct extent_buffer *leaf,
1316                         struct btrfs_dev_item *dev_item)
1317 {
1318         struct btrfs_device *device;
1319         u64 devid;
1320         int ret;
1321         u8 dev_uuid[BTRFS_UUID_SIZE];
1322
1323         devid = btrfs_device_id(leaf, dev_item);
1324         read_extent_buffer(leaf, dev_uuid,
1325                            (unsigned long)btrfs_device_uuid(dev_item),
1326                            BTRFS_UUID_SIZE);
1327         device = btrfs_find_device(root, devid, dev_uuid);
1328         if (!device) {
1329                 printk("warning devid %Lu not found already\n", devid);
1330                 device = kzalloc(sizeof(*device), GFP_NOFS);
1331                 if (!device)
1332                         return -ENOMEM;
1333                 list_add(&device->dev_list,
1334                          &root->fs_info->fs_devices->devices);
1335                 list_add(&device->dev_alloc_list,
1336                          &root->fs_info->fs_devices->alloc_list);
1337                 device->barriers = 1;
1338                 spin_lock_init(&device->io_lock);
1339         }
1340
1341         fill_device_from_item(leaf, dev_item, device);
1342         device->dev_root = root->fs_info->dev_root;
1343         ret = 0;
1344 #if 0
1345         ret = btrfs_open_device(device);
1346         if (ret) {
1347                 kfree(device);
1348         }
1349 #endif
1350         return ret;
1351 }
1352
1353 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1354 {
1355         struct btrfs_dev_item *dev_item;
1356
1357         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1358                                                      dev_item);
1359         return read_one_dev(root, buf, dev_item);
1360 }
1361
1362 int btrfs_read_sys_array(struct btrfs_root *root)
1363 {
1364         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1365         struct extent_buffer *sb = root->fs_info->sb_buffer;
1366         struct btrfs_disk_key *disk_key;
1367         struct btrfs_chunk *chunk;
1368         struct btrfs_key key;
1369         u32 num_stripes;
1370         u32 array_size;
1371         u32 len = 0;
1372         u8 *ptr;
1373         unsigned long sb_ptr;
1374         u32 cur;
1375         int ret;
1376
1377         array_size = btrfs_super_sys_array_size(super_copy);
1378
1379         /*
1380          * we do this loop twice, once for the device items and
1381          * once for all of the chunks.  This way there are device
1382          * structs filled in for every chunk
1383          */
1384         ptr = super_copy->sys_chunk_array;
1385         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1386         cur = 0;
1387
1388         while (cur < array_size) {
1389                 disk_key = (struct btrfs_disk_key *)ptr;
1390                 btrfs_disk_key_to_cpu(&key, disk_key);
1391
1392                 len = sizeof(*disk_key);
1393                 ptr += len;
1394                 sb_ptr += len;
1395                 cur += len;
1396
1397                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1398                         chunk = (struct btrfs_chunk *)sb_ptr;
1399                         ret = read_one_chunk(root, &key, sb, chunk);
1400                         BUG_ON(ret);
1401                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1402                         len = btrfs_chunk_item_size(num_stripes);
1403                 } else {
1404                         BUG();
1405                 }
1406                 ptr += len;
1407                 sb_ptr += len;
1408                 cur += len;
1409         }
1410         return 0;
1411 }
1412
1413 int btrfs_read_chunk_tree(struct btrfs_root *root)
1414 {
1415         struct btrfs_path *path;
1416         struct extent_buffer *leaf;
1417         struct btrfs_key key;
1418         struct btrfs_key found_key;
1419         int ret;
1420         int slot;
1421
1422         root = root->fs_info->chunk_root;
1423
1424         path = btrfs_alloc_path();
1425         if (!path)
1426                 return -ENOMEM;
1427
1428         /* first we search for all of the device items, and then we
1429          * read in all of the chunk items.  This way we can create chunk
1430          * mappings that reference all of the devices that are afound
1431          */
1432         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1433         key.offset = 0;
1434         key.type = 0;
1435 again:
1436         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1437         while(1) {
1438                 leaf = path->nodes[0];
1439                 slot = path->slots[0];
1440                 if (slot >= btrfs_header_nritems(leaf)) {
1441                         ret = btrfs_next_leaf(root, path);
1442                         if (ret == 0)
1443                                 continue;
1444                         if (ret < 0)
1445                                 goto error;
1446                         break;
1447                 }
1448                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1449                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1450                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1451                                 break;
1452                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1453                                 struct btrfs_dev_item *dev_item;
1454                                 dev_item = btrfs_item_ptr(leaf, slot,
1455                                                   struct btrfs_dev_item);
1456                                 ret = read_one_dev(root, leaf, dev_item);
1457                                 BUG_ON(ret);
1458                         }
1459                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1460                         struct btrfs_chunk *chunk;
1461                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1462                         ret = read_one_chunk(root, &found_key, leaf, chunk);
1463                 }
1464                 path->slots[0]++;
1465         }
1466         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1467                 key.objectid = 0;
1468                 btrfs_release_path(root, path);
1469                 goto again;
1470         }
1471
1472         btrfs_free_path(path);
1473         ret = 0;
1474 error:
1475         return ret;
1476 }
1477