SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/random.h>
  23 #include <asm/div64.h>
  24 #include "ctree.h"
  25 #include "extent_map.h"
  26 #include "disk-io.h"
  27 #include "transaction.h"
  28 #include "print-tree.h"
  29 #include "volumes.h"
  30
  31 struct map_lookup {
  32         u64 type;
  33         int io_align;
  34         int io_width;
  35         int stripe_len;
  36         int sector_size;
  37         int num_stripes;
  38         int sub_stripes;
  39         struct btrfs_bio_stripe stripes[];
  40 };
  41
  42 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  43                             (sizeof(struct btrfs_bio_stripe) * (n)))
  44
  45 static DEFINE_MUTEX(uuid_mutex);
  46 static LIST_HEAD(fs_uuids);
  47
  48 void btrfs_lock_volumes(void)
  49 {
  50         mutex_lock(&uuid_mutex);
  51 }
  52
  53 void btrfs_unlock_volumes(void)
  54 {
  55         mutex_unlock(&uuid_mutex);
  56 }
  57
  58 int btrfs_cleanup_fs_uuids(void)
  59 {
  60         struct btrfs_fs_devices *fs_devices;
  61         struct list_head *uuid_cur;
  62         struct list_head *devices_cur;
  63         struct btrfs_device *dev;
  64
  65         list_for_each(uuid_cur, &fs_uuids) {
  66                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  67                                         list);
  68                 while(!list_empty(&fs_devices->devices)) {
  69                         devices_cur = fs_devices->devices.next;
  70                         dev = list_entry(devices_cur, struct btrfs_device,
  71                                          dev_list);
  72                         if (dev->bdev) {
  73                                 close_bdev_excl(dev->bdev);
  74                         }
  75                         list_del(&dev->dev_list);
  76                         kfree(dev);
  77                 }
  78         }
  79         return 0;
  80 }
  81
  82 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  83                                           u8 *uuid)
  84 {
  85         struct btrfs_device *dev;
  86         struct list_head *cur;
  87
  88         list_for_each(cur, head) {
  89                 dev = list_entry(cur, struct btrfs_device, dev_list);
  90                 if (dev->devid == devid &&
  91                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  92                         return dev;
  93                 }
  94         }
  95         return NULL;
  96 }
  97
  98 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  99 {
 100         struct list_head *cur;
 101         struct btrfs_fs_devices *fs_devices;
 102
 103         list_for_each(cur, &fs_uuids) {
 104                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
 105                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 106                         return fs_devices;
 107         }
 108         return NULL;
 109 }
 110
 111 static int device_list_add(const char *path,
 112                            struct btrfs_super_block *disk_super,
 113                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 114 {
 115         struct btrfs_device *device;
 116         struct btrfs_fs_devices *fs_devices;
 117         u64 found_transid = btrfs_super_generation(disk_super);
 118
 119         fs_devices = find_fsid(disk_super->fsid);
 120         if (!fs_devices) {
 121                 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
 122                 if (!fs_devices)
 123                         return -ENOMEM;
 124                 INIT_LIST_HEAD(&fs_devices->devices);
 125                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 126                 list_add(&fs_devices->list, &fs_uuids);
 127                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 128                 fs_devices->latest_devid = devid;
 129                 fs_devices->latest_trans = found_transid;
 130                 fs_devices->lowest_devid = (u64)-1;
 131                 fs_devices->num_devices = 0;
 132                 device = NULL;
 133         } else {
 134                 device = __find_device(&fs_devices->devices, devid,
 135                                        disk_super->dev_item.uuid);
 136         }
 137         if (!device) {
 138                 device = kzalloc(sizeof(*device), GFP_NOFS);
 139                 if (!device) {
 140                         /* we can safely leave the fs_devices entry around */
 141                         return -ENOMEM;
 142                 }
 143                 device->devid = devid;
 144                 memcpy(device->uuid, disk_super->dev_item.uuid,
 145                        BTRFS_UUID_SIZE);
 146                 device->barriers = 1;
 147                 spin_lock_init(&device->io_lock);
 148                 device->name = kstrdup(path, GFP_NOFS);
 149                 if (!device->name) {
 150                         kfree(device);
 151                         return -ENOMEM;
 152                 }
 153                 list_add(&device->dev_list, &fs_devices->devices);
 154                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 155                 fs_devices->num_devices++;
 156         }
 157
 158         if (found_transid > fs_devices->latest_trans) {
 159                 fs_devices->latest_devid = devid;
 160                 fs_devices->latest_trans = found_transid;
 161         }
 162         if (fs_devices->lowest_devid > devid) {
 163                 fs_devices->lowest_devid = devid;
 164         }
 165         *fs_devices_ret = fs_devices;
 166         return 0;
 167 }
 168
 169 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 170 {
 171         struct list_head *head = &fs_devices->devices;
 172         struct list_head *cur;
 173         struct btrfs_device *device;
 174
 175         mutex_lock(&uuid_mutex);
 176         list_for_each(cur, head) {
 177                 device = list_entry(cur, struct btrfs_device, dev_list);
 178                 if (device->bdev) {
 179                         close_bdev_excl(device->bdev);
 180                 }
 181                 device->bdev = NULL;
 182         }
 183         mutex_unlock(&uuid_mutex);
 184         return 0;
 185 }
 186
 187 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 188                        int flags, void *holder)
 189 {
 190         struct block_device *bdev;
 191         struct list_head *head = &fs_devices->devices;
 192         struct list_head *cur;
 193         struct btrfs_device *device;
 194         int ret;
 195
 196         mutex_lock(&uuid_mutex);
 197         list_for_each(cur, head) {
 198                 device = list_entry(cur, struct btrfs_device, dev_list);
 199                 bdev = open_bdev_excl(device->name, flags, holder);
 200
 201                 if (IS_ERR(bdev)) {
 202                         printk("open %s failed\n", device->name);
 203                         ret = PTR_ERR(bdev);
 204                         goto fail;
 205                 }
 206                 set_blocksize(bdev, 4096);
 207                 if (device->devid == fs_devices->latest_devid)
 208                         fs_devices->latest_bdev = bdev;
 209                 if (device->devid == fs_devices->lowest_devid) {
 210                         fs_devices->lowest_bdev = bdev;
 211                 }
 212                 device->bdev = bdev;
 213
 214         }
 215         mutex_unlock(&uuid_mutex);
 216         return 0;
 217 fail:
 218         mutex_unlock(&uuid_mutex);
 219         btrfs_close_devices(fs_devices);
 220         return ret;
 221 }
 222
 223 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 224                           struct btrfs_fs_devices **fs_devices_ret)
 225 {
 226         struct btrfs_super_block *disk_super;
 227         struct block_device *bdev;
 228         struct buffer_head *bh;
 229         int ret;
 230         u64 devid;
 231         u64 transid;
 232
 233         mutex_lock(&uuid_mutex);
 234
 235         bdev = open_bdev_excl(path, flags, holder);
 236
 237         if (IS_ERR(bdev)) {
 238                 ret = PTR_ERR(bdev);
 239                 goto error;
 240         }
 241
 242         ret = set_blocksize(bdev, 4096);
 243         if (ret)
 244                 goto error_close;
 245         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 246         if (!bh) {
 247                 ret = -EIO;
 248                 goto error_close;
 249         }
 250         disk_super = (struct btrfs_super_block *)bh->b_data;
 251         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 252             sizeof(disk_super->magic))) {
 253                 ret = -EINVAL;
 254                 goto error_brelse;
 255         }
 256         devid = le64_to_cpu(disk_super->dev_item.devid);
 257         transid = btrfs_super_generation(disk_super);
 258         if (disk_super->label[0])
 259                 printk("device label %s ", disk_super->label);
 260         else {
 261                 /* FIXME, make a readl uuid parser */
 262                 printk("device fsid %llx-%llx ",
 263                        *(unsigned long long *)disk_super->fsid,
 264                        *(unsigned long long *)(disk_super->fsid + 8));
 265         }
 266         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 267         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 268
 269 error_brelse:
 270         brelse(bh);
 271 error_close:
 272         close_bdev_excl(bdev);
 273 error:
 274         mutex_unlock(&uuid_mutex);
 275         return ret;
 276 }
 277
 278 /*
 279  * this uses a pretty simple search, the expectation is that it is
 280  * called very infrequently and that a given device has a small number
 281  * of extents
 282  */
 283 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 284                                 struct btrfs_device *device,
 285                                 struct btrfs_path *path,
 286                                 u64 num_bytes, u64 *start)
 287 {
 288         struct btrfs_key key;
 289         struct btrfs_root *root = device->dev_root;
 290         struct btrfs_dev_extent *dev_extent = NULL;
 291         u64 hole_size = 0;
 292         u64 last_byte = 0;
 293         u64 search_start = 0;
 294         u64 search_end = device->total_bytes;
 295         int ret;
 296         int slot = 0;
 297         int start_found;
 298         struct extent_buffer *l;
 299
 300         start_found = 0;
 301         path->reada = 2;
 302
 303         /* FIXME use last free of some kind */
 304
 305         /* we don't want to overwrite the superblock on the drive,
 306          * so we make sure to start at an offset of at least 1MB
 307          */
 308         search_start = max((u64)1024 * 1024, search_start);
 309
 310         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 311                 search_start = max(root->fs_info->alloc_start, search_start);
 312
 313         key.objectid = device->devid;
 314         key.offset = search_start;
 315         key.type = BTRFS_DEV_EXTENT_KEY;
 316         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 317         if (ret < 0)
 318                 goto error;
 319         ret = btrfs_previous_item(root, path, 0, key.type);
 320         if (ret < 0)
 321                 goto error;
 322         l = path->nodes[0];
 323         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 324         while (1) {
 325                 l = path->nodes[0];
 326                 slot = path->slots[0];
 327                 if (slot >= btrfs_header_nritems(l)) {
 328                         ret = btrfs_next_leaf(root, path);
 329                         if (ret == 0)
 330                                 continue;
 331                         if (ret < 0)
 332                                 goto error;
 333 no_more_items:
 334                         if (!start_found) {
 335                                 if (search_start >= search_end) {
 336                                         ret = -ENOSPC;
 337                                         goto error;
 338                                 }
 339                                 *start = search_start;
 340                                 start_found = 1;
 341                                 goto check_pending;
 342                         }
 343                         *start = last_byte > search_start ?
 344                                 last_byte : search_start;
 345                         if (search_end <= *start) {
 346                                 ret = -ENOSPC;
 347                                 goto error;
 348                         }
 349                         goto check_pending;
 350                 }
 351                 btrfs_item_key_to_cpu(l, &key, slot);
 352
 353                 if (key.objectid < device->devid)
 354                         goto next;
 355
 356                 if (key.objectid > device->devid)
 357                         goto no_more_items;
 358
 359                 if (key.offset >= search_start && key.offset > last_byte &&
 360                     start_found) {
 361                         if (last_byte < search_start)
 362                                 last_byte = search_start;
 363                         hole_size = key.offset - last_byte;
 364                         if (key.offset > last_byte &&
 365                             hole_size >= num_bytes) {
 366                                 *start = last_byte;
 367                                 goto check_pending;
 368                         }
 369                 }
 370                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 371                         goto next;
 372                 }
 373
 374                 start_found = 1;
 375                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 376                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 377 next:
 378                 path->slots[0]++;
 379                 cond_resched();
 380         }
 381 check_pending:
 382         /* we have to make sure we didn't find an extent that has already
 383          * been allocated by the map tree or the original allocation
 384          */
 385         btrfs_release_path(root, path);
 386         BUG_ON(*start < search_start);
 387
 388         if (*start + num_bytes > search_end) {
 389                 ret = -ENOSPC;
 390                 goto error;
 391         }
 392         /* check for pending inserts here */
 393         return 0;
 394
 395 error:
 396         btrfs_release_path(root, path);
 397         return ret;
 398 }
 399
 400 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 401                           struct btrfs_device *device,
 402                           u64 start)
 403 {
 404         int ret;
 405         struct btrfs_path *path;
 406         struct btrfs_root *root = device->dev_root;
 407         struct btrfs_key key;
 408         struct btrfs_key found_key;
 409         struct extent_buffer *leaf = NULL;
 410         struct btrfs_dev_extent *extent = NULL;
 411
 412         path = btrfs_alloc_path();
 413         if (!path)
 414                 return -ENOMEM;
 415
 416         key.objectid = device->devid;
 417         key.offset = start;
 418         key.type = BTRFS_DEV_EXTENT_KEY;
 419
 420         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 421         if (ret > 0) {
 422                 ret = btrfs_previous_item(root, path, key.objectid,
 423                                           BTRFS_DEV_EXTENT_KEY);
 424                 BUG_ON(ret);
 425                 leaf = path->nodes[0];
 426                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 427                 extent = btrfs_item_ptr(leaf, path->slots[0],
 428                                         struct btrfs_dev_extent);
 429                 BUG_ON(found_key.offset > start || found_key.offset +
 430                        btrfs_dev_extent_length(leaf, extent) < start);
 431                 ret = 0;
 432         } else if (ret == 0) {
 433                 leaf = path->nodes[0];
 434                 extent = btrfs_item_ptr(leaf, path->slots[0],
 435                                         struct btrfs_dev_extent);
 436         }
 437         BUG_ON(ret);
 438
 439         device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 440         ret = btrfs_del_item(trans, root, path);
 441         BUG_ON(ret);
 442
 443         btrfs_free_path(path);
 444         return ret;
 445 }
 446
 447 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 448                            struct btrfs_device *device,
 449                            u64 chunk_tree, u64 chunk_objectid,
 450                            u64 chunk_offset,
 451                            u64 num_bytes, u64 *start)
 452 {
 453         int ret;
 454         struct btrfs_path *path;
 455         struct btrfs_root *root = device->dev_root;
 456         struct btrfs_dev_extent *extent;
 457         struct extent_buffer *leaf;
 458         struct btrfs_key key;
 459
 460         path = btrfs_alloc_path();
 461         if (!path)
 462                 return -ENOMEM;
 463
 464         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 465         if (ret) {
 466                 goto err;
 467         }
 468
 469         key.objectid = device->devid;
 470         key.offset = *start;
 471         key.type = BTRFS_DEV_EXTENT_KEY;
 472         ret = btrfs_insert_empty_item(trans, root, path, &key,
 473                                       sizeof(*extent));
 474         BUG_ON(ret);
 475
 476         leaf = path->nodes[0];
 477         extent = btrfs_item_ptr(leaf, path->slots[0],
 478                                 struct btrfs_dev_extent);
 479         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 480         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 481         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 482
 483         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 484                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 485                     BTRFS_UUID_SIZE);
 486
 487         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 488         btrfs_mark_buffer_dirty(leaf);
 489 err:
 490         btrfs_free_path(path);
 491         return ret;
 492 }
 493
 494 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 495 {
 496         struct btrfs_path *path;
 497         int ret;
 498         struct btrfs_key key;
 499         struct btrfs_chunk *chunk;
 500         struct btrfs_key found_key;
 501
 502         path = btrfs_alloc_path();
 503         BUG_ON(!path);
 504
 505         key.objectid = objectid;
 506         key.offset = (u64)-1;
 507         key.type = BTRFS_CHUNK_ITEM_KEY;
 508
 509         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 510         if (ret < 0)
 511                 goto error;
 512
 513         BUG_ON(ret == 0);
 514
 515         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 516         if (ret) {
 517                 *offset = 0;
 518         } else {
 519                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 520                                       path->slots[0]);
 521                 if (found_key.objectid != objectid)
 522                         *offset = 0;
 523                 else {
 524                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 525                                                struct btrfs_chunk);
 526                         *offset = found_key.offset +
 527                                 btrfs_chunk_length(path->nodes[0], chunk);
 528                 }
 529         }
 530         ret = 0;
 531 error:
 532         btrfs_free_path(path);
 533         return ret;
 534 }
 535
 536 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 537                            u64 *objectid)
 538 {
 539         int ret;
 540         struct btrfs_key key;
 541         struct btrfs_key found_key;
 542
 543         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 544         key.type = BTRFS_DEV_ITEM_KEY;
 545         key.offset = (u64)-1;
 546
 547         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 548         if (ret < 0)
 549                 goto error;
 550
 551         BUG_ON(ret == 0);
 552
 553         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 554                                   BTRFS_DEV_ITEM_KEY);
 555         if (ret) {
 556                 *objectid = 1;
 557         } else {
 558                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 559                                       path->slots[0]);
 560                 *objectid = found_key.offset + 1;
 561         }
 562         ret = 0;
 563 error:
 564         btrfs_release_path(root, path);
 565         return ret;
 566 }
 567
 568 /*
 569  * the device information is stored in the chunk root
 570  * the btrfs_device struct should be fully filled in
 571  */
 572 int btrfs_add_device(struct btrfs_trans_handle *trans,
 573                      struct btrfs_root *root,
 574                      struct btrfs_device *device)
 575 {
 576         int ret;
 577         struct btrfs_path *path;
 578         struct btrfs_dev_item *dev_item;
 579         struct extent_buffer *leaf;
 580         struct btrfs_key key;
 581         unsigned long ptr;
 582         u64 free_devid = 0;
 583
 584         root = root->fs_info->chunk_root;
 585
 586         path = btrfs_alloc_path();
 587         if (!path)
 588                 return -ENOMEM;
 589
 590         ret = find_next_devid(root, path, &free_devid);
 591         if (ret)
 592                 goto out;
 593
 594         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 595         key.type = BTRFS_DEV_ITEM_KEY;
 596         key.offset = free_devid;
 597
 598         ret = btrfs_insert_empty_item(trans, root, path, &key,
 599                                       sizeof(*dev_item));
 600         if (ret)
 601                 goto out;
 602
 603         leaf = path->nodes[0];
 604         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 605
 606         device->devid = free_devid;
 607         btrfs_set_device_id(leaf, dev_item, device->devid);
 608         btrfs_set_device_type(leaf, dev_item, device->type);
 609         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 610         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 611         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 612         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 613         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 614         btrfs_set_device_group(leaf, dev_item, 0);
 615         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 616         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 617
 618         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 619         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 620         btrfs_mark_buffer_dirty(leaf);
 621         ret = 0;
 622
 623 out:
 624         btrfs_free_path(path);
 625         return ret;
 626 }
 627
 628 static int btrfs_rm_dev_item(struct btrfs_root *root,
 629                              struct btrfs_device *device)
 630 {
 631         int ret;
 632         struct btrfs_path *path;
 633         struct block_device *bdev = device->bdev;
 634         struct btrfs_device *next_dev;
 635         struct btrfs_key key;
 636         u64 total_bytes;
 637         struct btrfs_fs_devices *fs_devices;
 638         struct btrfs_trans_handle *trans;
 639
 640         root = root->fs_info->chunk_root;
 641
 642         path = btrfs_alloc_path();
 643         if (!path)
 644                 return -ENOMEM;
 645
 646         trans = btrfs_start_transaction(root, 1);
 647         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 648         key.type = BTRFS_DEV_ITEM_KEY;
 649         key.offset = device->devid;
 650
 651         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 652         if (ret < 0)
 653                 goto out;
 654
 655         if (ret > 0) {
 656                 ret = -ENOENT;
 657                 goto out;
 658         }
 659
 660         ret = btrfs_del_item(trans, root, path);
 661         if (ret)
 662                 goto out;
 663
 664         /*
 665          * at this point, the device is zero sized.  We want to
 666          * remove it from the devices list and zero out the old super
 667          */
 668         list_del_init(&device->dev_list);
 669         list_del_init(&device->dev_alloc_list);
 670         fs_devices = root->fs_info->fs_devices;
 671
 672         next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
 673                               dev_list);
 674         if (bdev == fs_devices->lowest_bdev)
 675                 fs_devices->lowest_bdev = next_dev->bdev;
 676         if (bdev == root->fs_info->sb->s_bdev)
 677                 root->fs_info->sb->s_bdev = next_dev->bdev;
 678         if (bdev == fs_devices->latest_bdev)
 679                 fs_devices->latest_bdev = next_dev->bdev;
 680
 681         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 682         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 683                                     total_bytes - device->total_bytes);
 684
 685         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 686         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 687                                     total_bytes - 1);
 688 out:
 689         btrfs_free_path(path);
 690         btrfs_commit_transaction(trans, root);
 691         return ret;
 692 }
 693
 694 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 695 {
 696         struct btrfs_device *device;
 697         struct block_device *bdev;
 698         struct buffer_head *bh;
 699         struct btrfs_super_block *disk_super;
 700         u64 all_avail;
 701         u64 devid;
 702         int ret = 0;
 703
 704         mutex_lock(&root->fs_info->fs_mutex);
 705         mutex_lock(&uuid_mutex);
 706
 707         all_avail = root->fs_info->avail_data_alloc_bits |
 708                 root->fs_info->avail_system_alloc_bits |
 709                 root->fs_info->avail_metadata_alloc_bits;
 710
 711         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 712             root->fs_info->fs_devices->num_devices <= 4) {
 713                 printk("btrfs: unable to go below four devices on raid10\n");
 714                 ret = -EINVAL;
 715                 goto out;
 716         }
 717
 718         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 719             root->fs_info->fs_devices->num_devices <= 2) {
 720                 printk("btrfs: unable to go below two devices on raid1\n");
 721                 ret = -EINVAL;
 722                 goto out;
 723         }
 724
 725         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
 726         if (IS_ERR(bdev)) {
 727                 ret = PTR_ERR(bdev);
 728                 goto out;
 729         }
 730
 731         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 732         if (!bh) {
 733                 ret = -EIO;
 734                 goto error_close;
 735         }
 736         disk_super = (struct btrfs_super_block *)bh->b_data;
 737         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 738             sizeof(disk_super->magic))) {
 739                 ret = -ENOENT;
 740                 goto error_brelse;
 741         }
 742         if (memcmp(disk_super->fsid, root->fs_info->fsid, BTRFS_FSID_SIZE)) {
 743                 ret = -ENOENT;
 744                 goto error_brelse;
 745         }
 746         devid = le64_to_cpu(disk_super->dev_item.devid);
 747         device = btrfs_find_device(root, devid, NULL);
 748         if (!device) {
 749                 ret = -ENOENT;
 750                 goto error_brelse;
 751         }
 752
 753         root->fs_info->fs_devices->num_devices--;
 754
 755         ret = btrfs_shrink_device(device, 0);
 756         if (ret)
 757                 goto error_brelse;
 758
 759
 760         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
 761         if (ret)
 762                 goto error_brelse;
 763
 764         /* make sure this device isn't detected as part of the FS anymore */
 765         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
 766         set_buffer_dirty(bh);
 767         sync_dirty_buffer(bh);
 768
 769         brelse(bh);
 770
 771         /* one close for the device struct or super_block */
 772         close_bdev_excl(device->bdev);
 773
 774         /* one close for us */
 775         close_bdev_excl(device->bdev);
 776
 777         kfree(device->name);
 778         kfree(device);
 779         ret = 0;
 780         goto out;
 781
 782 error_brelse:
 783         brelse(bh);
 784 error_close:
 785         close_bdev_excl(bdev);
 786 out:
 787         mutex_unlock(&uuid_mutex);
 788         mutex_unlock(&root->fs_info->fs_mutex);
 789         return ret;
 790 }
 791
 792 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 793 {
 794         struct btrfs_trans_handle *trans;
 795         struct btrfs_device *device;
 796         struct block_device *bdev;
 797         struct list_head *cur;
 798         struct list_head *devices;
 799         u64 total_bytes;
 800         int ret = 0;
 801
 802
 803         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
 804         if (!bdev) {
 805                 return -EIO;
 806         }
 807         mutex_lock(&root->fs_info->fs_mutex);
 808         trans = btrfs_start_transaction(root, 1);
 809         devices = &root->fs_info->fs_devices->devices;
 810         list_for_each(cur, devices) {
 811                 device = list_entry(cur, struct btrfs_device, dev_list);
 812                 if (device->bdev == bdev) {
 813                         ret = -EEXIST;
 814                         goto out;
 815                 }
 816         }
 817
 818         device = kzalloc(sizeof(*device), GFP_NOFS);
 819         if (!device) {
 820                 /* we can safely leave the fs_devices entry around */
 821                 ret = -ENOMEM;
 822                 goto out_close_bdev;
 823         }
 824
 825         device->barriers = 1;
 826         generate_random_uuid(device->uuid);
 827         spin_lock_init(&device->io_lock);
 828         device->name = kstrdup(device_path, GFP_NOFS);
 829         if (!device->name) {
 830                 kfree(device);
 831                 goto out_close_bdev;
 832         }
 833         device->io_width = root->sectorsize;
 834         device->io_align = root->sectorsize;
 835         device->sector_size = root->sectorsize;
 836         device->total_bytes = i_size_read(bdev->bd_inode);
 837         device->dev_root = root->fs_info->dev_root;
 838         device->bdev = bdev;
 839
 840         ret = btrfs_add_device(trans, root, device);
 841         if (ret)
 842                 goto out_close_bdev;
 843
 844         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 845         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 846                                     total_bytes + device->total_bytes);
 847
 848         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 849         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 850                                     total_bytes + 1);
 851
 852         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
 853         list_add(&device->dev_alloc_list,
 854                  &root->fs_info->fs_devices->alloc_list);
 855         root->fs_info->fs_devices->num_devices++;
 856 out:
 857         btrfs_end_transaction(trans, root);
 858         mutex_unlock(&root->fs_info->fs_mutex);
 859         return ret;
 860
 861 out_close_bdev:
 862         close_bdev_excl(bdev);
 863         goto out;
 864 }
 865
 866 int btrfs_update_device(struct btrfs_trans_handle *trans,
 867                         struct btrfs_device *device)
 868 {
 869         int ret;
 870         struct btrfs_path *path;
 871         struct btrfs_root *root;
 872         struct btrfs_dev_item *dev_item;
 873         struct extent_buffer *leaf;
 874         struct btrfs_key key;
 875
 876         root = device->dev_root->fs_info->chunk_root;
 877
 878         path = btrfs_alloc_path();
 879         if (!path)
 880                 return -ENOMEM;
 881
 882         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 883         key.type = BTRFS_DEV_ITEM_KEY;
 884         key.offset = device->devid;
 885
 886         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 887         if (ret < 0)
 888                 goto out;
 889
 890         if (ret > 0) {
 891                 ret = -ENOENT;
 892                 goto out;
 893         }
 894
 895         leaf = path->nodes[0];
 896         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 897
 898         btrfs_set_device_id(leaf, dev_item, device->devid);
 899         btrfs_set_device_type(leaf, dev_item, device->type);
 900         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 901         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 902         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 903         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 904         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 905         btrfs_mark_buffer_dirty(leaf);
 906
 907 out:
 908         btrfs_free_path(path);
 909         return ret;
 910 }
 911
 912 int btrfs_grow_device(struct btrfs_trans_handle *trans,
 913                       struct btrfs_device *device, u64 new_size)
 914 {
 915         struct btrfs_super_block *super_copy =
 916                 &device->dev_root->fs_info->super_copy;
 917         u64 old_total = btrfs_super_total_bytes(super_copy);
 918         u64 diff = new_size - device->total_bytes;
 919
 920         btrfs_set_super_total_bytes(super_copy, old_total + diff);
 921         return btrfs_update_device(trans, device);
 922 }
 923
 924 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
 925                             struct btrfs_root *root,
 926                             u64 chunk_tree, u64 chunk_objectid,
 927                             u64 chunk_offset)
 928 {
 929         int ret;
 930         struct btrfs_path *path;
 931         struct btrfs_key key;
 932
 933         root = root->fs_info->chunk_root;
 934         path = btrfs_alloc_path();
 935         if (!path)
 936                 return -ENOMEM;
 937
 938         key.objectid = chunk_objectid;
 939         key.offset = chunk_offset;
 940         key.type = BTRFS_CHUNK_ITEM_KEY;
 941
 942         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 943         BUG_ON(ret);
 944
 945         ret = btrfs_del_item(trans, root, path);
 946         BUG_ON(ret);
 947
 948         btrfs_free_path(path);
 949         return 0;
 950 }
 951
 952 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
 953                         chunk_offset)
 954 {
 955         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 956         struct btrfs_disk_key *disk_key;
 957         struct btrfs_chunk *chunk;
 958         u8 *ptr;
 959         int ret = 0;
 960         u32 num_stripes;
 961         u32 array_size;
 962         u32 len = 0;
 963         u32 cur;
 964         struct btrfs_key key;
 965
 966         array_size = btrfs_super_sys_array_size(super_copy);
 967
 968         ptr = super_copy->sys_chunk_array;
 969         cur = 0;
 970
 971         while (cur < array_size) {
 972                 disk_key = (struct btrfs_disk_key *)ptr;
 973                 btrfs_disk_key_to_cpu(&key, disk_key);
 974
 975                 len = sizeof(*disk_key);
 976
 977                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
 978                         chunk = (struct btrfs_chunk *)(ptr + len);
 979                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
 980                         len += btrfs_chunk_item_size(num_stripes);
 981                 } else {
 982                         ret = -EIO;
 983                         break;
 984                 }
 985                 if (key.objectid == chunk_objectid &&
 986                     key.offset == chunk_offset) {
 987                         memmove(ptr, ptr + len, array_size - (cur + len));
 988                         array_size -= len;
 989                         btrfs_set_super_sys_array_size(super_copy, array_size);
 990                 } else {
 991                         ptr += len;
 992                         cur += len;
 993                 }
 994         }
 995         return ret;
 996 }
 997
 998
 999 int btrfs_relocate_chunk(struct btrfs_root *root,
1000                          u64 chunk_tree, u64 chunk_objectid,
1001                          u64 chunk_offset)
1002 {
1003         struct extent_map_tree *em_tree;
1004         struct btrfs_root *extent_root;
1005         struct btrfs_trans_handle *trans;
1006         struct extent_map *em;
1007         struct map_lookup *map;
1008         int ret;
1009         int i;
1010
1011         root = root->fs_info->chunk_root;
1012         extent_root = root->fs_info->extent_root;
1013         em_tree = &root->fs_info->mapping_tree.map_tree;
1014
1015         /* step one, relocate all the extents inside this chunk */
1016         ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1017         BUG_ON(ret);
1018
1019         trans = btrfs_start_transaction(root, 1);
1020         BUG_ON(!trans);
1021
1022         /*
1023          * step two, delete the device extents and the
1024          * chunk tree entries
1025          */
1026         spin_lock(&em_tree->lock);
1027         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1028         spin_unlock(&em_tree->lock);
1029
1030         BUG_ON(em->start > chunk_offset ||
1031                em->start + em->len < chunk_offset);
1032         map = (struct map_lookup *)em->bdev;
1033
1034         for (i = 0; i < map->num_stripes; i++) {
1035                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1036                                             map->stripes[i].physical);
1037                 BUG_ON(ret);
1038
1039                 ret = btrfs_update_device(trans, map->stripes[i].dev);
1040                 BUG_ON(ret);
1041         }
1042         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1043                                chunk_offset);
1044
1045         BUG_ON(ret);
1046
1047         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1048                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1049                 BUG_ON(ret);
1050         }
1051
1052         spin_lock(&em_tree->lock);
1053         remove_extent_mapping(em_tree, em);
1054         kfree(map);
1055         em->bdev = NULL;
1056
1057         /* once for the tree */
1058         free_extent_map(em);
1059         spin_unlock(&em_tree->lock);
1060
1061         /* once for us */
1062         free_extent_map(em);
1063
1064         btrfs_end_transaction(trans, root);
1065         return 0;
1066 }
1067
1068 static u64 div_factor(u64 num, int factor)
1069 {
1070         if (factor == 10)
1071                 return num;
1072         num *= factor;
1073         do_div(num, 10);
1074         return num;
1075 }
1076
1077
1078 int btrfs_balance(struct btrfs_root *dev_root)
1079 {
1080         int ret;
1081         struct list_head *cur;
1082         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1083         struct btrfs_device *device;
1084         u64 old_size;
1085         u64 size_to_free;
1086         struct btrfs_path *path;
1087         struct btrfs_key key;
1088         struct btrfs_chunk *chunk;
1089         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1090         struct btrfs_trans_handle *trans;
1091         struct btrfs_key found_key;
1092
1093
1094         dev_root = dev_root->fs_info->dev_root;
1095
1096         mutex_lock(&dev_root->fs_info->fs_mutex);
1097         /* step one make some room on all the devices */
1098         list_for_each(cur, devices) {
1099                 device = list_entry(cur, struct btrfs_device, dev_list);
1100                 old_size = device->total_bytes;
1101                 size_to_free = div_factor(old_size, 1);
1102                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1103                 if (device->total_bytes - device->bytes_used > size_to_free)
1104                         continue;
1105
1106                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1107                 BUG_ON(ret);
1108
1109                 trans = btrfs_start_transaction(dev_root, 1);
1110                 BUG_ON(!trans);
1111
1112                 ret = btrfs_grow_device(trans, device, old_size);
1113                 BUG_ON(ret);
1114
1115                 btrfs_end_transaction(trans, dev_root);
1116         }
1117
1118         /* step two, relocate all the chunks */
1119         path = btrfs_alloc_path();
1120         BUG_ON(!path);
1121
1122         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1123         key.offset = (u64)-1;
1124         key.type = BTRFS_CHUNK_ITEM_KEY;
1125
1126         while(1) {
1127                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1128                 if (ret < 0)
1129                         goto error;
1130
1131                 /*
1132                  * this shouldn't happen, it means the last relocate
1133                  * failed
1134                  */
1135                 if (ret == 0)
1136                         break;
1137
1138                 ret = btrfs_previous_item(chunk_root, path, 0,
1139                                           BTRFS_CHUNK_ITEM_KEY);
1140                 if (ret) {
1141                         break;
1142                 }
1143                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1144                                       path->slots[0]);
1145                 if (found_key.objectid != key.objectid)
1146                         break;
1147                 chunk = btrfs_item_ptr(path->nodes[0],
1148                                        path->slots[0],
1149                                        struct btrfs_chunk);
1150                 key.offset = found_key.offset;
1151                 /* chunk zero is special */
1152                 if (key.offset == 0)
1153                         break;
1154
1155                 ret = btrfs_relocate_chunk(chunk_root,
1156                                            chunk_root->root_key.objectid,
1157                                            found_key.objectid,
1158                                            found_key.offset);
1159                 BUG_ON(ret);
1160                 btrfs_release_path(chunk_root, path);
1161         }
1162         ret = 0;
1163 error:
1164         btrfs_free_path(path);
1165         mutex_unlock(&dev_root->fs_info->fs_mutex);
1166         return ret;
1167 }
1168
1169 /*
1170  * shrinking a device means finding all of the device extents past
1171  * the new size, and then following the back refs to the chunks.
1172  * The chunk relocation code actually frees the device extent
1173  */
1174 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1175 {
1176         struct btrfs_trans_handle *trans;
1177         struct btrfs_root *root = device->dev_root;
1178         struct btrfs_dev_extent *dev_extent = NULL;
1179         struct btrfs_path *path;
1180         u64 length;
1181         u64 chunk_tree;
1182         u64 chunk_objectid;
1183         u64 chunk_offset;
1184         int ret;
1185         int slot;
1186         struct extent_buffer *l;
1187         struct btrfs_key key;
1188         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1189         u64 old_total = btrfs_super_total_bytes(super_copy);
1190         u64 diff = device->total_bytes - new_size;
1191
1192
1193         path = btrfs_alloc_path();
1194         if (!path)
1195                 return -ENOMEM;
1196
1197         trans = btrfs_start_transaction(root, 1);
1198         if (!trans) {
1199                 ret = -ENOMEM;
1200                 goto done;
1201         }
1202
1203         path->reada = 2;
1204
1205         device->total_bytes = new_size;
1206         ret = btrfs_update_device(trans, device);
1207         if (ret) {
1208                 btrfs_end_transaction(trans, root);
1209                 goto done;
1210         }
1211         WARN_ON(diff > old_total);
1212         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1213         btrfs_end_transaction(trans, root);
1214
1215         key.objectid = device->devid;
1216         key.offset = (u64)-1;
1217         key.type = BTRFS_DEV_EXTENT_KEY;
1218
1219         while (1) {
1220                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1221                 if (ret < 0)
1222                         goto done;
1223
1224                 ret = btrfs_previous_item(root, path, 0, key.type);
1225                 if (ret < 0)
1226                         goto done;
1227                 if (ret) {
1228                         ret = 0;
1229                         goto done;
1230                 }
1231
1232                 l = path->nodes[0];
1233                 slot = path->slots[0];
1234                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1235
1236                 if (key.objectid != device->devid)
1237                         goto done;
1238
1239                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1240                 length = btrfs_dev_extent_length(l, dev_extent);
1241
1242                 if (key.offset + length <= new_size)
1243                         goto done;
1244
1245                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1246                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1247                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1248                 btrfs_release_path(root, path);
1249
1250                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1251                                            chunk_offset);
1252                 if (ret)
1253                         goto done;
1254         }
1255
1256 done:
1257         btrfs_free_path(path);
1258         return ret;
1259 }
1260
1261 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1262                            struct btrfs_root *root,
1263                            struct btrfs_key *key,
1264                            struct btrfs_chunk *chunk, int item_size)
1265 {
1266         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1267         struct btrfs_disk_key disk_key;
1268         u32 array_size;
1269         u8 *ptr;
1270
1271         array_size = btrfs_super_sys_array_size(super_copy);
1272         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1273                 return -EFBIG;
1274
1275         ptr = super_copy->sys_chunk_array + array_size;
1276         btrfs_cpu_key_to_disk(&disk_key, key);
1277         memcpy(ptr, &disk_key, sizeof(disk_key));
1278         ptr += sizeof(disk_key);
1279         memcpy(ptr, chunk, item_size);
1280         item_size += sizeof(disk_key);
1281         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1282         return 0;
1283 }
1284
1285 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1286                                int sub_stripes)
1287 {
1288         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1289                 return calc_size;
1290         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1291                 return calc_size * (num_stripes / sub_stripes);
1292         else
1293                 return calc_size * num_stripes;
1294 }
1295
1296
1297 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1298                       struct btrfs_root *extent_root, u64 *start,
1299                       u64 *num_bytes, u64 type)
1300 {
1301         u64 dev_offset;
1302         struct btrfs_fs_info *info = extent_root->fs_info;
1303         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1304         struct btrfs_path *path;
1305         struct btrfs_stripe *stripes;
1306         struct btrfs_device *device = NULL;
1307         struct btrfs_chunk *chunk;
1308         struct list_head private_devs;
1309         struct list_head *dev_list;
1310         struct list_head *cur;
1311         struct extent_map_tree *em_tree;
1312         struct map_lookup *map;
1313         struct extent_map *em;
1314         int min_stripe_size = 1 * 1024 * 1024;
1315         u64 physical;
1316         u64 calc_size = 1024 * 1024 * 1024;
1317         u64 max_chunk_size = calc_size;
1318         u64 min_free;
1319         u64 avail;
1320         u64 max_avail = 0;
1321         u64 percent_max;
1322         int num_stripes = 1;
1323         int min_stripes = 1;
1324         int sub_stripes = 0;
1325         int looped = 0;
1326         int ret;
1327         int index;
1328         int stripe_len = 64 * 1024;
1329         struct btrfs_key key;
1330
1331         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1332             (type & BTRFS_BLOCK_GROUP_DUP)) {
1333                 WARN_ON(1);
1334                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1335         }
1336         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1337         if (list_empty(dev_list))
1338                 return -ENOSPC;
1339
1340         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1341                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1342                 min_stripes = 2;
1343         }
1344         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1345                 num_stripes = 2;
1346                 min_stripes = 2;
1347         }
1348         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1349                 num_stripes = min_t(u64, 2,
1350                                   btrfs_super_num_devices(&info->super_copy));
1351                 if (num_stripes < 2)
1352                         return -ENOSPC;
1353                 min_stripes = 2;
1354         }
1355         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1356                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1357                 if (num_stripes < 4)
1358                         return -ENOSPC;
1359                 num_stripes &= ~(u32)1;
1360                 sub_stripes = 2;
1361                 min_stripes = 4;
1362         }
1363
1364         if (type & BTRFS_BLOCK_GROUP_DATA) {
1365                 max_chunk_size = 10 * calc_size;
1366                 min_stripe_size = 64 * 1024 * 1024;
1367         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1368                 max_chunk_size = 4 * calc_size;
1369                 min_stripe_size = 32 * 1024 * 1024;
1370         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1371                 calc_size = 8 * 1024 * 1024;
1372                 max_chunk_size = calc_size * 2;
1373                 min_stripe_size = 1 * 1024 * 1024;
1374         }
1375
1376         path = btrfs_alloc_path();
1377         if (!path)
1378                 return -ENOMEM;
1379
1380         /* we don't want a chunk larger than 10% of the FS */
1381         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1382         max_chunk_size = min(percent_max, max_chunk_size);
1383
1384 again:
1385         if (calc_size * num_stripes > max_chunk_size) {
1386                 calc_size = max_chunk_size;
1387                 do_div(calc_size, num_stripes);
1388                 do_div(calc_size, stripe_len);
1389                 calc_size *= stripe_len;
1390         }
1391         /* we don't want tiny stripes */
1392         calc_size = max_t(u64, min_stripe_size, calc_size);
1393
1394         do_div(calc_size, stripe_len);
1395         calc_size *= stripe_len;
1396
1397         INIT_LIST_HEAD(&private_devs);
1398         cur = dev_list->next;
1399         index = 0;
1400
1401         if (type & BTRFS_BLOCK_GROUP_DUP)
1402                 min_free = calc_size * 2;
1403         else
1404                 min_free = calc_size;
1405
1406         /* we add 1MB because we never use the first 1MB of the device */
1407         min_free += 1024 * 1024;
1408
1409         /* build a private list of devices we will allocate from */
1410         while(index < num_stripes) {
1411                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1412
1413                 avail = device->total_bytes - device->bytes_used;
1414                 cur = cur->next;
1415
1416                 if (avail >= min_free) {
1417                         u64 ignored_start = 0;
1418                         ret = find_free_dev_extent(trans, device, path,
1419                                                    min_free,
1420                                                    &ignored_start);
1421                         if (ret == 0) {
1422                                 list_move_tail(&device->dev_alloc_list,
1423                                                &private_devs);
1424                                 index++;
1425                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1426                                         index++;
1427                         }
1428                 } else if (avail > max_avail)
1429                         max_avail = avail;
1430                 if (cur == dev_list)
1431                         break;
1432         }
1433         if (index < num_stripes) {
1434                 list_splice(&private_devs, dev_list);
1435                 if (index >= min_stripes) {
1436                         num_stripes = index;
1437                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1438                                 num_stripes /= sub_stripes;
1439                                 num_stripes *= sub_stripes;
1440                         }
1441                         looped = 1;
1442                         goto again;
1443                 }
1444                 if (!looped && max_avail > 0) {
1445                         looped = 1;
1446                         calc_size = max_avail;
1447                         goto again;
1448                 }
1449                 btrfs_free_path(path);
1450                 return -ENOSPC;
1451         }
1452         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1453         key.type = BTRFS_CHUNK_ITEM_KEY;
1454         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1455                               &key.offset);
1456         if (ret) {
1457                 btrfs_free_path(path);
1458                 return ret;
1459         }
1460
1461         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1462         if (!chunk) {
1463                 btrfs_free_path(path);
1464                 return -ENOMEM;
1465         }
1466
1467         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1468         if (!map) {
1469                 kfree(chunk);
1470                 btrfs_free_path(path);
1471                 return -ENOMEM;
1472         }
1473         btrfs_free_path(path);
1474         path = NULL;
1475
1476         stripes = &chunk->stripe;
1477         *num_bytes = chunk_bytes_by_type(type, calc_size,
1478                                          num_stripes, sub_stripes);
1479
1480         index = 0;
1481         while(index < num_stripes) {
1482                 struct btrfs_stripe *stripe;
1483                 BUG_ON(list_empty(&private_devs));
1484                 cur = private_devs.next;
1485                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1486
1487                 /* loop over this device again if we're doing a dup group */
1488                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1489                     (index == num_stripes - 1))
1490                         list_move_tail(&device->dev_alloc_list, dev_list);
1491
1492                 ret = btrfs_alloc_dev_extent(trans, device,
1493                              info->chunk_root->root_key.objectid,
1494                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1495                              calc_size, &dev_offset);
1496                 BUG_ON(ret);
1497                 device->bytes_used += calc_size;
1498                 ret = btrfs_update_device(trans, device);
1499                 BUG_ON(ret);
1500
1501                 map->stripes[index].dev = device;
1502                 map->stripes[index].physical = dev_offset;
1503                 stripe = stripes + index;
1504                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1505                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1506                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1507                 physical = dev_offset;
1508                 index++;
1509         }
1510         BUG_ON(!list_empty(&private_devs));
1511
1512         /* key was set above */
1513         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1514         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1515         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1516         btrfs_set_stack_chunk_type(chunk, type);
1517         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1518         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1519         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1520         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1521         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1522         map->sector_size = extent_root->sectorsize;
1523         map->stripe_len = stripe_len;
1524         map->io_align = stripe_len;
1525         map->io_width = stripe_len;
1526         map->type = type;
1527         map->num_stripes = num_stripes;
1528         map->sub_stripes = sub_stripes;
1529
1530         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1531                                 btrfs_chunk_item_size(num_stripes));
1532         BUG_ON(ret);
1533         *start = key.offset;;
1534
1535         em = alloc_extent_map(GFP_NOFS);
1536         if (!em)
1537                 return -ENOMEM;
1538         em->bdev = (struct block_device *)map;
1539         em->start = key.offset;
1540         em->len = *num_bytes;
1541         em->block_start = 0;
1542
1543         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1544                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1545                                     chunk, btrfs_chunk_item_size(num_stripes));
1546                 BUG_ON(ret);
1547         }
1548         kfree(chunk);
1549
1550         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1551         spin_lock(&em_tree->lock);
1552         ret = add_extent_mapping(em_tree, em);
1553         spin_unlock(&em_tree->lock);
1554         BUG_ON(ret);
1555         free_extent_map(em);
1556         return ret;
1557 }
1558
1559 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1560 {
1561         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1562 }
1563
1564 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1565 {
1566         struct extent_map *em;
1567
1568         while(1) {
1569                 spin_lock(&tree->map_tree.lock);
1570                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1571                 if (em)
1572                         remove_extent_mapping(&tree->map_tree, em);
1573                 spin_unlock(&tree->map_tree.lock);
1574                 if (!em)
1575                         break;
1576                 kfree(em->bdev);
1577                 /* once for us */
1578                 free_extent_map(em);
1579                 /* once for the tree */
1580                 free_extent_map(em);
1581         }
1582 }
1583
1584 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1585 {
1586         struct extent_map *em;
1587         struct map_lookup *map;
1588         struct extent_map_tree *em_tree = &map_tree->map_tree;
1589         int ret;
1590
1591         spin_lock(&em_tree->lock);
1592         em = lookup_extent_mapping(em_tree, logical, len);
1593         spin_unlock(&em_tree->lock);
1594         BUG_ON(!em);
1595
1596         BUG_ON(em->start > logical || em->start + em->len < logical);
1597         map = (struct map_lookup *)em->bdev;
1598         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1599                 ret = map->num_stripes;
1600         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1601                 ret = map->sub_stripes;
1602         else
1603                 ret = 1;
1604         free_extent_map(em);
1605         return ret;
1606 }
1607
1608 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1609                              u64 logical, u64 *length,
1610                              struct btrfs_multi_bio **multi_ret,
1611                              int mirror_num, struct page *unplug_page)
1612 {
1613         struct extent_map *em;
1614         struct map_lookup *map;
1615         struct extent_map_tree *em_tree = &map_tree->map_tree;
1616         u64 offset;
1617         u64 stripe_offset;
1618         u64 stripe_nr;
1619         int stripes_allocated = 8;
1620         int stripes_required = 1;
1621         int stripe_index;
1622         int i;
1623         int num_stripes;
1624         int max_errors = 0;
1625         struct btrfs_multi_bio *multi = NULL;
1626
1627         if (multi_ret && !(rw & (1 << BIO_RW))) {
1628                 stripes_allocated = 1;
1629         }
1630 again:
1631         if (multi_ret) {
1632                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1633                                 GFP_NOFS);
1634                 if (!multi)
1635                         return -ENOMEM;
1636
1637                 atomic_set(&multi->error, 0);
1638         }
1639
1640         spin_lock(&em_tree->lock);
1641         em = lookup_extent_mapping(em_tree, logical, *length);
1642         spin_unlock(&em_tree->lock);
1643
1644         if (!em && unplug_page)
1645                 return 0;
1646
1647         if (!em) {
1648                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1649                 BUG();
1650         }
1651
1652         BUG_ON(em->start > logical || em->start + em->len < logical);
1653         map = (struct map_lookup *)em->bdev;
1654         offset = logical - em->start;
1655
1656         if (mirror_num > map->num_stripes)
1657                 mirror_num = 0;
1658
1659         /* if our multi bio struct is too small, back off and try again */
1660         if (rw & (1 << BIO_RW)) {
1661                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1662                                  BTRFS_BLOCK_GROUP_DUP)) {
1663                         stripes_required = map->num_stripes;
1664                         max_errors = 1;
1665                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1666                         stripes_required = map->sub_stripes;
1667                         max_errors = 1;
1668                 }
1669         }
1670         if (multi_ret && rw == WRITE &&
1671             stripes_allocated < stripes_required) {
1672                 stripes_allocated = map->num_stripes;
1673                 free_extent_map(em);
1674                 kfree(multi);
1675                 goto again;
1676         }
1677         stripe_nr = offset;
1678         /*
1679          * stripe_nr counts the total number of stripes we have to stride
1680          * to get to this block
1681          */
1682         do_div(stripe_nr, map->stripe_len);
1683
1684         stripe_offset = stripe_nr * map->stripe_len;
1685         BUG_ON(offset < stripe_offset);
1686
1687         /* stripe_offset is the offset of this block in its stripe*/
1688         stripe_offset = offset - stripe_offset;
1689
1690         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1691                          BTRFS_BLOCK_GROUP_RAID10 |
1692                          BTRFS_BLOCK_GROUP_DUP)) {
1693                 /* we limit the length of each bio to what fits in a stripe */
1694                 *length = min_t(u64, em->len - offset,
1695                               map->stripe_len - stripe_offset);
1696         } else {
1697                 *length = em->len - offset;
1698         }
1699
1700         if (!multi_ret && !unplug_page)
1701                 goto out;
1702
1703         num_stripes = 1;
1704         stripe_index = 0;
1705         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1706                 if (unplug_page || (rw & (1 << BIO_RW)))
1707                         num_stripes = map->num_stripes;
1708                 else if (mirror_num)
1709                         stripe_index = mirror_num - 1;
1710                 else
1711                         stripe_index = current->pid % map->num_stripes;
1712
1713         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1714                 if (rw & (1 << BIO_RW))
1715                         num_stripes = map->num_stripes;
1716                 else if (mirror_num)
1717                         stripe_index = mirror_num - 1;
1718
1719         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1720                 int factor = map->num_stripes / map->sub_stripes;
1721
1722                 stripe_index = do_div(stripe_nr, factor);
1723                 stripe_index *= map->sub_stripes;
1724
1725                 if (unplug_page || (rw & (1 << BIO_RW)))
1726                         num_stripes = map->sub_stripes;
1727                 else if (mirror_num)
1728                         stripe_index += mirror_num - 1;
1729                 else
1730                         stripe_index += current->pid % map->sub_stripes;
1731         } else {
1732                 /*
1733                  * after this do_div call, stripe_nr is the number of stripes
1734                  * on this device we have to walk to find the data, and
1735                  * stripe_index is the number of our device in the stripe array
1736                  */
1737                 stripe_index = do_div(stripe_nr, map->num_stripes);
1738         }
1739         BUG_ON(stripe_index >= map->num_stripes);
1740
1741         for (i = 0; i < num_stripes; i++) {
1742                 if (unplug_page) {
1743                         struct btrfs_device *device;
1744                         struct backing_dev_info *bdi;
1745
1746                         device = map->stripes[stripe_index].dev;
1747                         bdi = blk_get_backing_dev_info(device->bdev);
1748                         if (bdi->unplug_io_fn) {
1749                                 bdi->unplug_io_fn(bdi, unplug_page);
1750                         }
1751                 } else {
1752                         multi->stripes[i].physical =
1753                                 map->stripes[stripe_index].physical +
1754                                 stripe_offset + stripe_nr * map->stripe_len;
1755                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
1756                 }
1757                 stripe_index++;
1758         }
1759         if (multi_ret) {
1760                 *multi_ret = multi;
1761                 multi->num_stripes = num_stripes;
1762                 multi->max_errors = max_errors;
1763         }
1764 out:
1765         free_extent_map(em);
1766         return 0;
1767 }
1768
1769 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1770                       u64 logical, u64 *length,
1771                       struct btrfs_multi_bio **multi_ret, int mirror_num)
1772 {
1773         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1774                                  mirror_num, NULL);
1775 }
1776
1777 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1778                       u64 logical, struct page *page)
1779 {
1780         u64 length = PAGE_CACHE_SIZE;
1781         return __btrfs_map_block(map_tree, READ, logical, &length,
1782                                  NULL, 0, page);
1783 }
1784
1785
1786 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1787 static void end_bio_multi_stripe(struct bio *bio, int err)
1788 #else
1789 static int end_bio_multi_stripe(struct bio *bio,
1790                                    unsigned int bytes_done, int err)
1791 #endif
1792 {
1793         struct btrfs_multi_bio *multi = bio->bi_private;
1794
1795 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1796         if (bio->bi_size)
1797                 return 1;
1798 #endif
1799         if (err)
1800                 atomic_inc(&multi->error);
1801
1802         if (atomic_dec_and_test(&multi->stripes_pending)) {
1803                 bio->bi_private = multi->private;
1804                 bio->bi_end_io = multi->end_io;
1805
1806                 /* only send an error to the higher layers if it is
1807                  * beyond the tolerance of the multi-bio
1808                  */
1809                 if (atomic_read(&multi->error) > multi->max_errors)
1810                         err = -EIO;
1811                 else
1812                         err = 0;
1813                 kfree(multi);
1814
1815 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1816                 bio_endio(bio, bio->bi_size, err);
1817 #else
1818                 bio_endio(bio, err);
1819 #endif
1820         } else {
1821                 bio_put(bio);
1822         }
1823 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1824         return 0;
1825 #endif
1826 }
1827
1828 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1829                   int mirror_num)
1830 {
1831         struct btrfs_mapping_tree *map_tree;
1832         struct btrfs_device *dev;
1833         struct bio *first_bio = bio;
1834         u64 logical = bio->bi_sector << 9;
1835         u64 length = 0;
1836         u64 map_length;
1837         struct btrfs_multi_bio *multi = NULL;
1838         int ret;
1839         int dev_nr = 0;
1840         int total_devs = 1;
1841
1842         length = bio->bi_size;
1843         map_tree = &root->fs_info->mapping_tree;
1844         map_length = length;
1845
1846         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1847                               mirror_num);
1848         BUG_ON(ret);
1849
1850         total_devs = multi->num_stripes;
1851         if (map_length < length) {
1852                 printk("mapping failed logical %Lu bio len %Lu "
1853                        "len %Lu\n", logical, length, map_length);
1854                 BUG();
1855         }
1856         multi->end_io = first_bio->bi_end_io;
1857         multi->private = first_bio->bi_private;
1858         atomic_set(&multi->stripes_pending, multi->num_stripes);
1859
1860         while(dev_nr < total_devs) {
1861                 if (total_devs > 1) {
1862                         if (dev_nr < total_devs - 1) {
1863                                 bio = bio_clone(first_bio, GFP_NOFS);
1864                                 BUG_ON(!bio);
1865                         } else {
1866                                 bio = first_bio;
1867                         }
1868                         bio->bi_private = multi;
1869                         bio->bi_end_io = end_bio_multi_stripe;
1870                 }
1871                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1872                 dev = multi->stripes[dev_nr].dev;
1873
1874                 bio->bi_bdev = dev->bdev;
1875                 spin_lock(&dev->io_lock);
1876                 dev->total_ios++;
1877                 spin_unlock(&dev->io_lock);
1878                 submit_bio(rw, bio);
1879                 dev_nr++;
1880         }
1881         if (total_devs == 1)
1882                 kfree(multi);
1883         return 0;
1884 }
1885
1886 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1887                                        u8 *uuid)
1888 {
1889         struct list_head *head = &root->fs_info->fs_devices->devices;
1890
1891         return __find_device(head, devid, uuid);
1892 }
1893
1894 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1895                           struct extent_buffer *leaf,
1896                           struct btrfs_chunk *chunk)
1897 {
1898         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1899         struct map_lookup *map;
1900         struct extent_map *em;
1901         u64 logical;
1902         u64 length;
1903         u64 devid;
1904         u8 uuid[BTRFS_UUID_SIZE];
1905         int num_stripes;
1906         int ret;
1907         int i;
1908
1909         logical = key->offset;
1910         length = btrfs_chunk_length(leaf, chunk);
1911
1912         spin_lock(&map_tree->map_tree.lock);
1913         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1914         spin_unlock(&map_tree->map_tree.lock);
1915
1916         /* already mapped? */
1917         if (em && em->start <= logical && em->start + em->len > logical) {
1918                 free_extent_map(em);
1919                 return 0;
1920         } else if (em) {
1921                 free_extent_map(em);
1922         }
1923
1924         map = kzalloc(sizeof(*map), GFP_NOFS);
1925         if (!map)
1926                 return -ENOMEM;
1927
1928         em = alloc_extent_map(GFP_NOFS);
1929         if (!em)
1930                 return -ENOMEM;
1931         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1932         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1933         if (!map) {
1934                 free_extent_map(em);
1935                 return -ENOMEM;
1936         }
1937
1938         em->bdev = (struct block_device *)map;
1939         em->start = logical;
1940         em->len = length;
1941         em->block_start = 0;
1942
1943         map->num_stripes = num_stripes;
1944         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1945         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1946         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1947         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1948         map->type = btrfs_chunk_type(leaf, chunk);
1949         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1950         for (i = 0; i < num_stripes; i++) {
1951                 map->stripes[i].physical =
1952                         btrfs_stripe_offset_nr(leaf, chunk, i);
1953                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1954                 read_extent_buffer(leaf, uuid, (unsigned long)
1955                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1956                                    BTRFS_UUID_SIZE);
1957                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1958                 if (!map->stripes[i].dev) {
1959                         kfree(map);
1960                         free_extent_map(em);
1961                         return -EIO;
1962                 }
1963         }
1964
1965         spin_lock(&map_tree->map_tree.lock);
1966         ret = add_extent_mapping(&map_tree->map_tree, em);
1967         spin_unlock(&map_tree->map_tree.lock);
1968         BUG_ON(ret);
1969         free_extent_map(em);
1970
1971         return 0;
1972 }
1973
1974 static int fill_device_from_item(struct extent_buffer *leaf,
1975                                  struct btrfs_dev_item *dev_item,
1976                                  struct btrfs_device *device)
1977 {
1978         unsigned long ptr;
1979
1980         device->devid = btrfs_device_id(leaf, dev_item);
1981         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1982         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1983         device->type = btrfs_device_type(leaf, dev_item);
1984         device->io_align = btrfs_device_io_align(leaf, dev_item);
1985         device->io_width = btrfs_device_io_width(leaf, dev_item);
1986         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1987
1988         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1989         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1990
1991         return 0;
1992 }
1993
1994 static int read_one_dev(struct btrfs_root *root,
1995                         struct extent_buffer *leaf,
1996                         struct btrfs_dev_item *dev_item)
1997 {
1998         struct btrfs_device *device;
1999         u64 devid;
2000         int ret;
2001         u8 dev_uuid[BTRFS_UUID_SIZE];
2002
2003         devid = btrfs_device_id(leaf, dev_item);
2004         read_extent_buffer(leaf, dev_uuid,
2005                            (unsigned long)btrfs_device_uuid(dev_item),
2006                            BTRFS_UUID_SIZE);
2007         device = btrfs_find_device(root, devid, dev_uuid);
2008         if (!device) {
2009                 printk("warning devid %Lu not found already\n", devid);
2010                 device = kzalloc(sizeof(*device), GFP_NOFS);
2011                 if (!device)
2012                         return -ENOMEM;
2013                 list_add(&device->dev_list,
2014                          &root->fs_info->fs_devices->devices);
2015                 list_add(&device->dev_alloc_list,
2016                          &root->fs_info->fs_devices->alloc_list);
2017                 device->barriers = 1;
2018                 spin_lock_init(&device->io_lock);
2019         }
2020
2021         fill_device_from_item(leaf, dev_item, device);
2022         device->dev_root = root->fs_info->dev_root;
2023         ret = 0;
2024 #if 0
2025         ret = btrfs_open_device(device);
2026         if (ret) {
2027                 kfree(device);
2028         }
2029 #endif
2030         return ret;
2031 }
2032
2033 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2034 {
2035         struct btrfs_dev_item *dev_item;
2036
2037         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2038                                                      dev_item);
2039         return read_one_dev(root, buf, dev_item);
2040 }
2041
2042 int btrfs_read_sys_array(struct btrfs_root *root)
2043 {
2044         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2045         struct extent_buffer *sb;
2046         struct btrfs_disk_key *disk_key;
2047         struct btrfs_chunk *chunk;
2048         u8 *ptr;
2049         unsigned long sb_ptr;
2050         int ret = 0;
2051         u32 num_stripes;
2052         u32 array_size;
2053         u32 len = 0;
2054         u32 cur;
2055         struct btrfs_key key;
2056
2057         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2058                                           BTRFS_SUPER_INFO_SIZE);
2059         if (!sb)
2060                 return -ENOMEM;
2061         btrfs_set_buffer_uptodate(sb);
2062         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2063         array_size = btrfs_super_sys_array_size(super_copy);
2064
2065         ptr = super_copy->sys_chunk_array;
2066         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2067         cur = 0;
2068
2069         while (cur < array_size) {
2070                 disk_key = (struct btrfs_disk_key *)ptr;
2071                 btrfs_disk_key_to_cpu(&key, disk_key);
2072
2073                 len = sizeof(*disk_key); ptr += len;
2074                 sb_ptr += len;
2075                 cur += len;
2076
2077                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2078                         chunk = (struct btrfs_chunk *)sb_ptr;
2079                         ret = read_one_chunk(root, &key, sb, chunk);
2080                         if (ret)
2081                                 break;
2082                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2083                         len = btrfs_chunk_item_size(num_stripes);
2084                 } else {
2085                         ret = -EIO;
2086                         break;
2087                 }
2088                 ptr += len;
2089                 sb_ptr += len;
2090                 cur += len;
2091         }
2092         free_extent_buffer(sb);
2093         return ret;
2094 }
2095
2096 int btrfs_read_chunk_tree(struct btrfs_root *root)
2097 {
2098         struct btrfs_path *path;
2099         struct extent_buffer *leaf;
2100         struct btrfs_key key;
2101         struct btrfs_key found_key;
2102         int ret;
2103         int slot;
2104
2105         root = root->fs_info->chunk_root;
2106
2107         path = btrfs_alloc_path();
2108         if (!path)
2109                 return -ENOMEM;
2110
2111         /* first we search for all of the device items, and then we
2112          * read in all of the chunk items.  This way we can create chunk
2113          * mappings that reference all of the devices that are afound
2114          */
2115         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2116         key.offset = 0;
2117         key.type = 0;
2118 again:
2119         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2120         while(1) {
2121                 leaf = path->nodes[0];
2122                 slot = path->slots[0];
2123                 if (slot >= btrfs_header_nritems(leaf)) {
2124                         ret = btrfs_next_leaf(root, path);
2125                         if (ret == 0)
2126                                 continue;
2127                         if (ret < 0)
2128                                 goto error;
2129                         break;
2130                 }
2131                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2132                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2133                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2134                                 break;
2135                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2136                                 struct btrfs_dev_item *dev_item;
2137                                 dev_item = btrfs_item_ptr(leaf, slot,
2138                                                   struct btrfs_dev_item);
2139                                 ret = read_one_dev(root, leaf, dev_item);
2140                                 BUG_ON(ret);
2141                         }
2142                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2143                         struct btrfs_chunk *chunk;
2144                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2145                         ret = read_one_chunk(root, &found_key, leaf, chunk);
2146                 }
2147                 path->slots[0]++;
2148         }
2149         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2150                 key.objectid = 0;
2151                 btrfs_release_path(root, path);
2152                 goto again;
2153         }
2154
2155         btrfs_free_path(path);
2156         ret = 0;
2157 error:
2158         return ret;
2159 }
2160