fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/random.h>
  23 #include <asm/div64.h>
  24 #include "ctree.h"
  25 #include "extent_map.h"
  26 #include "disk-io.h"
  27 #include "transaction.h"
  28 #include "print-tree.h"
  29 #include "volumes.h"
  30
  31 struct map_lookup {
  32         u64 type;
  33         int io_align;
  34         int io_width;
  35         int stripe_len;
  36         int sector_size;
  37         int num_stripes;
  38         int sub_stripes;
  39         struct btrfs_bio_stripe stripes[];
  40 };
  41
  42 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  43                             (sizeof(struct btrfs_bio_stripe) * (n)))
  44
  45 static DEFINE_MUTEX(uuid_mutex);
  46 static LIST_HEAD(fs_uuids);
  47
  48 void btrfs_lock_volumes(void)
  49 {
  50         mutex_lock(&uuid_mutex);
  51 }
  52
  53 void btrfs_unlock_volumes(void)
  54 {
  55         mutex_unlock(&uuid_mutex);
  56 }
  57
  58 int btrfs_cleanup_fs_uuids(void)
  59 {
  60         struct btrfs_fs_devices *fs_devices;
  61         struct list_head *uuid_cur;
  62         struct list_head *devices_cur;
  63         struct btrfs_device *dev;
  64
  65         list_for_each(uuid_cur, &fs_uuids) {
  66                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  67                                         list);
  68                 while(!list_empty(&fs_devices->devices)) {
  69                         devices_cur = fs_devices->devices.next;
  70                         dev = list_entry(devices_cur, struct btrfs_device,
  71                                          dev_list);
  72                         if (dev->bdev) {
  73                                 close_bdev_excl(dev->bdev);
  74                         }
  75                         list_del(&dev->dev_list);
  76                         kfree(dev);
  77                 }
  78         }
  79         return 0;
  80 }
  81
  82 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  83                                           u8 *uuid)
  84 {
  85         struct btrfs_device *dev;
  86         struct list_head *cur;
  87
  88         list_for_each(cur, head) {
  89                 dev = list_entry(cur, struct btrfs_device, dev_list);
  90                 if (dev->devid == devid &&
  91                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  92                         return dev;
  93                 }
  94         }
  95         return NULL;
  96 }
  97
  98 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  99 {
 100         struct list_head *cur;
 101         struct btrfs_fs_devices *fs_devices;
 102
 103         list_for_each(cur, &fs_uuids) {
 104                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
 105                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 106                         return fs_devices;
 107         }
 108         return NULL;
 109 }
 110
 111 static int device_list_add(const char *path,
 112                            struct btrfs_super_block *disk_super,
 113                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 114 {
 115         struct btrfs_device *device;
 116         struct btrfs_fs_devices *fs_devices;
 117         u64 found_transid = btrfs_super_generation(disk_super);
 118
 119         fs_devices = find_fsid(disk_super->fsid);
 120         if (!fs_devices) {
 121                 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
 122                 if (!fs_devices)
 123                         return -ENOMEM;
 124                 INIT_LIST_HEAD(&fs_devices->devices);
 125                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 126                 list_add(&fs_devices->list, &fs_uuids);
 127                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 128                 fs_devices->latest_devid = devid;
 129                 fs_devices->latest_trans = found_transid;
 130                 fs_devices->lowest_devid = (u64)-1;
 131                 fs_devices->num_devices = 0;
 132                 device = NULL;
 133         } else {
 134                 device = __find_device(&fs_devices->devices, devid,
 135                                        disk_super->dev_item.uuid);
 136         }
 137         if (!device) {
 138                 device = kzalloc(sizeof(*device), GFP_NOFS);
 139                 if (!device) {
 140                         /* we can safely leave the fs_devices entry around */
 141                         return -ENOMEM;
 142                 }
 143                 device->devid = devid;
 144                 memcpy(device->uuid, disk_super->dev_item.uuid,
 145                        BTRFS_UUID_SIZE);
 146                 device->barriers = 1;
 147                 spin_lock_init(&device->io_lock);
 148                 device->name = kstrdup(path, GFP_NOFS);
 149                 if (!device->name) {
 150                         kfree(device);
 151                         return -ENOMEM;
 152                 }
 153                 list_add(&device->dev_list, &fs_devices->devices);
 154                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 155                 fs_devices->num_devices++;
 156         }
 157
 158         if (found_transid > fs_devices->latest_trans) {
 159                 fs_devices->latest_devid = devid;
 160                 fs_devices->latest_trans = found_transid;
 161         }
 162         if (fs_devices->lowest_devid > devid) {
 163                 fs_devices->lowest_devid = devid;
 164         }
 165         *fs_devices_ret = fs_devices;
 166         return 0;
 167 }
 168
 169 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 170 {
 171         struct list_head *head = &fs_devices->devices;
 172         struct list_head *cur;
 173         struct btrfs_device *device;
 174
 175         mutex_lock(&uuid_mutex);
 176         list_for_each(cur, head) {
 177                 device = list_entry(cur, struct btrfs_device, dev_list);
 178                 if (device->bdev) {
 179                         close_bdev_excl(device->bdev);
 180                 }
 181                 device->bdev = NULL;
 182         }
 183         mutex_unlock(&uuid_mutex);
 184         return 0;
 185 }
 186
 187 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 188                        int flags, void *holder)
 189 {
 190         struct block_device *bdev;
 191         struct list_head *head = &fs_devices->devices;
 192         struct list_head *cur;
 193         struct btrfs_device *device;
 194         int ret;
 195
 196         mutex_lock(&uuid_mutex);
 197         list_for_each(cur, head) {
 198                 device = list_entry(cur, struct btrfs_device, dev_list);
 199                 if (device->bdev)
 200                         continue;
 201
 202                 bdev = open_bdev_excl(device->name, flags, holder);
 203
 204                 if (IS_ERR(bdev)) {
 205                         printk("open %s failed\n", device->name);
 206                         ret = PTR_ERR(bdev);
 207                         goto fail;
 208                 }
 209                 set_blocksize(bdev, 4096);
 210                 if (device->devid == fs_devices->latest_devid)
 211                         fs_devices->latest_bdev = bdev;
 212                 if (device->devid == fs_devices->lowest_devid) {
 213                         fs_devices->lowest_bdev = bdev;
 214                 }
 215                 device->bdev = bdev;
 216
 217         }
 218         mutex_unlock(&uuid_mutex);
 219         return 0;
 220 fail:
 221         mutex_unlock(&uuid_mutex);
 222         btrfs_close_devices(fs_devices);
 223         return ret;
 224 }
 225
 226 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 227                           struct btrfs_fs_devices **fs_devices_ret)
 228 {
 229         struct btrfs_super_block *disk_super;
 230         struct block_device *bdev;
 231         struct buffer_head *bh;
 232         int ret;
 233         u64 devid;
 234         u64 transid;
 235
 236         mutex_lock(&uuid_mutex);
 237
 238         bdev = open_bdev_excl(path, flags, holder);
 239
 240         if (IS_ERR(bdev)) {
 241                 ret = PTR_ERR(bdev);
 242                 goto error;
 243         }
 244
 245         ret = set_blocksize(bdev, 4096);
 246         if (ret)
 247                 goto error_close;
 248         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 249         if (!bh) {
 250                 ret = -EIO;
 251                 goto error_close;
 252         }
 253         disk_super = (struct btrfs_super_block *)bh->b_data;
 254         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 255             sizeof(disk_super->magic))) {
 256                 ret = -EINVAL;
 257                 goto error_brelse;
 258         }
 259         devid = le64_to_cpu(disk_super->dev_item.devid);
 260         transid = btrfs_super_generation(disk_super);
 261         if (disk_super->label[0])
 262                 printk("device label %s ", disk_super->label);
 263         else {
 264                 /* FIXME, make a readl uuid parser */
 265                 printk("device fsid %llx-%llx ",
 266                        *(unsigned long long *)disk_super->fsid,
 267                        *(unsigned long long *)(disk_super->fsid + 8));
 268         }
 269         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 270         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 271
 272 error_brelse:
 273         brelse(bh);
 274 error_close:
 275         close_bdev_excl(bdev);
 276 error:
 277         mutex_unlock(&uuid_mutex);
 278         return ret;
 279 }
 280
 281 /*
 282  * this uses a pretty simple search, the expectation is that it is
 283  * called very infrequently and that a given device has a small number
 284  * of extents
 285  */
 286 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 287                                 struct btrfs_device *device,
 288                                 struct btrfs_path *path,
 289                                 u64 num_bytes, u64 *start)
 290 {
 291         struct btrfs_key key;
 292         struct btrfs_root *root = device->dev_root;
 293         struct btrfs_dev_extent *dev_extent = NULL;
 294         u64 hole_size = 0;
 295         u64 last_byte = 0;
 296         u64 search_start = 0;
 297         u64 search_end = device->total_bytes;
 298         int ret;
 299         int slot = 0;
 300         int start_found;
 301         struct extent_buffer *l;
 302
 303         start_found = 0;
 304         path->reada = 2;
 305
 306         /* FIXME use last free of some kind */
 307
 308         /* we don't want to overwrite the superblock on the drive,
 309          * so we make sure to start at an offset of at least 1MB
 310          */
 311         search_start = max((u64)1024 * 1024, search_start);
 312
 313         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 314                 search_start = max(root->fs_info->alloc_start, search_start);
 315
 316         key.objectid = device->devid;
 317         key.offset = search_start;
 318         key.type = BTRFS_DEV_EXTENT_KEY;
 319         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 320         if (ret < 0)
 321                 goto error;
 322         ret = btrfs_previous_item(root, path, 0, key.type);
 323         if (ret < 0)
 324                 goto error;
 325         l = path->nodes[0];
 326         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 327         while (1) {
 328                 l = path->nodes[0];
 329                 slot = path->slots[0];
 330                 if (slot >= btrfs_header_nritems(l)) {
 331                         ret = btrfs_next_leaf(root, path);
 332                         if (ret == 0)
 333                                 continue;
 334                         if (ret < 0)
 335                                 goto error;
 336 no_more_items:
 337                         if (!start_found) {
 338                                 if (search_start >= search_end) {
 339                                         ret = -ENOSPC;
 340                                         goto error;
 341                                 }
 342                                 *start = search_start;
 343                                 start_found = 1;
 344                                 goto check_pending;
 345                         }
 346                         *start = last_byte > search_start ?
 347                                 last_byte : search_start;
 348                         if (search_end <= *start) {
 349                                 ret = -ENOSPC;
 350                                 goto error;
 351                         }
 352                         goto check_pending;
 353                 }
 354                 btrfs_item_key_to_cpu(l, &key, slot);
 355
 356                 if (key.objectid < device->devid)
 357                         goto next;
 358
 359                 if (key.objectid > device->devid)
 360                         goto no_more_items;
 361
 362                 if (key.offset >= search_start && key.offset > last_byte &&
 363                     start_found) {
 364                         if (last_byte < search_start)
 365                                 last_byte = search_start;
 366                         hole_size = key.offset - last_byte;
 367                         if (key.offset > last_byte &&
 368                             hole_size >= num_bytes) {
 369                                 *start = last_byte;
 370                                 goto check_pending;
 371                         }
 372                 }
 373                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 374                         goto next;
 375                 }
 376
 377                 start_found = 1;
 378                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 379                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 380 next:
 381                 path->slots[0]++;
 382                 cond_resched();
 383         }
 384 check_pending:
 385         /* we have to make sure we didn't find an extent that has already
 386          * been allocated by the map tree or the original allocation
 387          */
 388         btrfs_release_path(root, path);
 389         BUG_ON(*start < search_start);
 390
 391         if (*start + num_bytes > search_end) {
 392                 ret = -ENOSPC;
 393                 goto error;
 394         }
 395         /* check for pending inserts here */
 396         return 0;
 397
 398 error:
 399         btrfs_release_path(root, path);
 400         return ret;
 401 }
 402
 403 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 404                           struct btrfs_device *device,
 405                           u64 start)
 406 {
 407         int ret;
 408         struct btrfs_path *path;
 409         struct btrfs_root *root = device->dev_root;
 410         struct btrfs_key key;
 411         struct btrfs_key found_key;
 412         struct extent_buffer *leaf = NULL;
 413         struct btrfs_dev_extent *extent = NULL;
 414
 415         path = btrfs_alloc_path();
 416         if (!path)
 417                 return -ENOMEM;
 418
 419         key.objectid = device->devid;
 420         key.offset = start;
 421         key.type = BTRFS_DEV_EXTENT_KEY;
 422
 423         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 424         if (ret > 0) {
 425                 ret = btrfs_previous_item(root, path, key.objectid,
 426                                           BTRFS_DEV_EXTENT_KEY);
 427                 BUG_ON(ret);
 428                 leaf = path->nodes[0];
 429                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 430                 extent = btrfs_item_ptr(leaf, path->slots[0],
 431                                         struct btrfs_dev_extent);
 432                 BUG_ON(found_key.offset > start || found_key.offset +
 433                        btrfs_dev_extent_length(leaf, extent) < start);
 434                 ret = 0;
 435         } else if (ret == 0) {
 436                 leaf = path->nodes[0];
 437                 extent = btrfs_item_ptr(leaf, path->slots[0],
 438                                         struct btrfs_dev_extent);
 439         }
 440         BUG_ON(ret);
 441
 442         device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 443         ret = btrfs_del_item(trans, root, path);
 444         BUG_ON(ret);
 445
 446         btrfs_free_path(path);
 447         return ret;
 448 }
 449
 450 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 451                            struct btrfs_device *device,
 452                            u64 chunk_tree, u64 chunk_objectid,
 453                            u64 chunk_offset,
 454                            u64 num_bytes, u64 *start)
 455 {
 456         int ret;
 457         struct btrfs_path *path;
 458         struct btrfs_root *root = device->dev_root;
 459         struct btrfs_dev_extent *extent;
 460         struct extent_buffer *leaf;
 461         struct btrfs_key key;
 462
 463         path = btrfs_alloc_path();
 464         if (!path)
 465                 return -ENOMEM;
 466
 467         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 468         if (ret) {
 469                 goto err;
 470         }
 471
 472         key.objectid = device->devid;
 473         key.offset = *start;
 474         key.type = BTRFS_DEV_EXTENT_KEY;
 475         ret = btrfs_insert_empty_item(trans, root, path, &key,
 476                                       sizeof(*extent));
 477         BUG_ON(ret);
 478
 479         leaf = path->nodes[0];
 480         extent = btrfs_item_ptr(leaf, path->slots[0],
 481                                 struct btrfs_dev_extent);
 482         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 483         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 484         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 485
 486         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 487                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 488                     BTRFS_UUID_SIZE);
 489
 490         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 491         btrfs_mark_buffer_dirty(leaf);
 492 err:
 493         btrfs_free_path(path);
 494         return ret;
 495 }
 496
 497 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 498 {
 499         struct btrfs_path *path;
 500         int ret;
 501         struct btrfs_key key;
 502         struct btrfs_chunk *chunk;
 503         struct btrfs_key found_key;
 504
 505         path = btrfs_alloc_path();
 506         BUG_ON(!path);
 507
 508         key.objectid = objectid;
 509         key.offset = (u64)-1;
 510         key.type = BTRFS_CHUNK_ITEM_KEY;
 511
 512         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 513         if (ret < 0)
 514                 goto error;
 515
 516         BUG_ON(ret == 0);
 517
 518         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 519         if (ret) {
 520                 *offset = 0;
 521         } else {
 522                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 523                                       path->slots[0]);
 524                 if (found_key.objectid != objectid)
 525                         *offset = 0;
 526                 else {
 527                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 528                                                struct btrfs_chunk);
 529                         *offset = found_key.offset +
 530                                 btrfs_chunk_length(path->nodes[0], chunk);
 531                 }
 532         }
 533         ret = 0;
 534 error:
 535         btrfs_free_path(path);
 536         return ret;
 537 }
 538
 539 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 540                            u64 *objectid)
 541 {
 542         int ret;
 543         struct btrfs_key key;
 544         struct btrfs_key found_key;
 545
 546         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 547         key.type = BTRFS_DEV_ITEM_KEY;
 548         key.offset = (u64)-1;
 549
 550         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 551         if (ret < 0)
 552                 goto error;
 553
 554         BUG_ON(ret == 0);
 555
 556         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 557                                   BTRFS_DEV_ITEM_KEY);
 558         if (ret) {
 559                 *objectid = 1;
 560         } else {
 561                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 562                                       path->slots[0]);
 563                 *objectid = found_key.offset + 1;
 564         }
 565         ret = 0;
 566 error:
 567         btrfs_release_path(root, path);
 568         return ret;
 569 }
 570
 571 /*
 572  * the device information is stored in the chunk root
 573  * the btrfs_device struct should be fully filled in
 574  */
 575 int btrfs_add_device(struct btrfs_trans_handle *trans,
 576                      struct btrfs_root *root,
 577                      struct btrfs_device *device)
 578 {
 579         int ret;
 580         struct btrfs_path *path;
 581         struct btrfs_dev_item *dev_item;
 582         struct extent_buffer *leaf;
 583         struct btrfs_key key;
 584         unsigned long ptr;
 585         u64 free_devid = 0;
 586
 587         root = root->fs_info->chunk_root;
 588
 589         path = btrfs_alloc_path();
 590         if (!path)
 591                 return -ENOMEM;
 592
 593         ret = find_next_devid(root, path, &free_devid);
 594         if (ret)
 595                 goto out;
 596
 597         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 598         key.type = BTRFS_DEV_ITEM_KEY;
 599         key.offset = free_devid;
 600
 601         ret = btrfs_insert_empty_item(trans, root, path, &key,
 602                                       sizeof(*dev_item));
 603         if (ret)
 604                 goto out;
 605
 606         leaf = path->nodes[0];
 607         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 608
 609         device->devid = free_devid;
 610         btrfs_set_device_id(leaf, dev_item, device->devid);
 611         btrfs_set_device_type(leaf, dev_item, device->type);
 612         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 613         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 614         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 615         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 616         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 617         btrfs_set_device_group(leaf, dev_item, 0);
 618         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 619         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 620
 621         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 622         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 623         btrfs_mark_buffer_dirty(leaf);
 624         ret = 0;
 625
 626 out:
 627         btrfs_free_path(path);
 628         return ret;
 629 }
 630
 631 static int btrfs_rm_dev_item(struct btrfs_root *root,
 632                              struct btrfs_device *device)
 633 {
 634         int ret;
 635         struct btrfs_path *path;
 636         struct block_device *bdev = device->bdev;
 637         struct btrfs_device *next_dev;
 638         struct btrfs_key key;
 639         u64 total_bytes;
 640         struct btrfs_fs_devices *fs_devices;
 641         struct btrfs_trans_handle *trans;
 642
 643         root = root->fs_info->chunk_root;
 644
 645         path = btrfs_alloc_path();
 646         if (!path)
 647                 return -ENOMEM;
 648
 649         trans = btrfs_start_transaction(root, 1);
 650         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 651         key.type = BTRFS_DEV_ITEM_KEY;
 652         key.offset = device->devid;
 653
 654         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 655         if (ret < 0)
 656                 goto out;
 657
 658         if (ret > 0) {
 659                 ret = -ENOENT;
 660                 goto out;
 661         }
 662
 663         ret = btrfs_del_item(trans, root, path);
 664         if (ret)
 665                 goto out;
 666
 667         /*
 668          * at this point, the device is zero sized.  We want to
 669          * remove it from the devices list and zero out the old super
 670          */
 671         list_del_init(&device->dev_list);
 672         list_del_init(&device->dev_alloc_list);
 673         fs_devices = root->fs_info->fs_devices;
 674
 675         next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
 676                               dev_list);
 677         if (bdev == fs_devices->lowest_bdev)
 678                 fs_devices->lowest_bdev = next_dev->bdev;
 679         if (bdev == root->fs_info->sb->s_bdev)
 680                 root->fs_info->sb->s_bdev = next_dev->bdev;
 681         if (bdev == fs_devices->latest_bdev)
 682                 fs_devices->latest_bdev = next_dev->bdev;
 683
 684         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 685         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 686                                     total_bytes - device->total_bytes);
 687
 688         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 689         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 690                                     total_bytes - 1);
 691 out:
 692         btrfs_free_path(path);
 693         btrfs_commit_transaction(trans, root);
 694         return ret;
 695 }
 696
 697 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 698 {
 699         struct btrfs_device *device;
 700         struct block_device *bdev;
 701         struct buffer_head *bh;
 702         struct btrfs_super_block *disk_super;
 703         u64 all_avail;
 704         u64 devid;
 705         int ret = 0;
 706
 707         mutex_lock(&root->fs_info->fs_mutex);
 708         mutex_lock(&uuid_mutex);
 709
 710         all_avail = root->fs_info->avail_data_alloc_bits |
 711                 root->fs_info->avail_system_alloc_bits |
 712                 root->fs_info->avail_metadata_alloc_bits;
 713
 714         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 715             root->fs_info->fs_devices->num_devices <= 4) {
 716                 printk("btrfs: unable to go below four devices on raid10\n");
 717                 ret = -EINVAL;
 718                 goto out;
 719         }
 720
 721         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 722             root->fs_info->fs_devices->num_devices <= 2) {
 723                 printk("btrfs: unable to go below two devices on raid1\n");
 724                 ret = -EINVAL;
 725                 goto out;
 726         }
 727
 728         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
 729         if (IS_ERR(bdev)) {
 730                 ret = PTR_ERR(bdev);
 731                 goto out;
 732         }
 733
 734         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 735         if (!bh) {
 736                 ret = -EIO;
 737                 goto error_close;
 738         }
 739         disk_super = (struct btrfs_super_block *)bh->b_data;
 740         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 741             sizeof(disk_super->magic))) {
 742                 ret = -ENOENT;
 743                 goto error_brelse;
 744         }
 745         if (memcmp(disk_super->fsid, root->fs_info->fsid, BTRFS_FSID_SIZE)) {
 746                 ret = -ENOENT;
 747                 goto error_brelse;
 748         }
 749         devid = le64_to_cpu(disk_super->dev_item.devid);
 750         device = btrfs_find_device(root, devid, NULL);
 751         if (!device) {
 752                 ret = -ENOENT;
 753                 goto error_brelse;
 754         }
 755
 756         root->fs_info->fs_devices->num_devices--;
 757
 758         ret = btrfs_shrink_device(device, 0);
 759         if (ret)
 760                 goto error_brelse;
 761
 762
 763         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
 764         if (ret)
 765                 goto error_brelse;
 766
 767         /* make sure this device isn't detected as part of the FS anymore */
 768         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
 769         set_buffer_dirty(bh);
 770         sync_dirty_buffer(bh);
 771
 772         brelse(bh);
 773
 774         /* one close for the device struct or super_block */
 775         close_bdev_excl(device->bdev);
 776
 777         /* one close for us */
 778         close_bdev_excl(device->bdev);
 779
 780         kfree(device->name);
 781         kfree(device);
 782         ret = 0;
 783         goto out;
 784
 785 error_brelse:
 786         brelse(bh);
 787 error_close:
 788         close_bdev_excl(bdev);
 789 out:
 790         mutex_unlock(&uuid_mutex);
 791         mutex_unlock(&root->fs_info->fs_mutex);
 792         return ret;
 793 }
 794
 795 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 796 {
 797         struct btrfs_trans_handle *trans;
 798         struct btrfs_device *device;
 799         struct block_device *bdev;
 800         struct list_head *cur;
 801         struct list_head *devices;
 802         u64 total_bytes;
 803         int ret = 0;
 804
 805
 806         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
 807         if (!bdev) {
 808                 return -EIO;
 809         }
 810         mutex_lock(&root->fs_info->fs_mutex);
 811         trans = btrfs_start_transaction(root, 1);
 812         devices = &root->fs_info->fs_devices->devices;
 813         list_for_each(cur, devices) {
 814                 device = list_entry(cur, struct btrfs_device, dev_list);
 815                 if (device->bdev == bdev) {
 816                         ret = -EEXIST;
 817                         goto out;
 818                 }
 819         }
 820
 821         device = kzalloc(sizeof(*device), GFP_NOFS);
 822         if (!device) {
 823                 /* we can safely leave the fs_devices entry around */
 824                 ret = -ENOMEM;
 825                 goto out_close_bdev;
 826         }
 827
 828         device->barriers = 1;
 829         generate_random_uuid(device->uuid);
 830         spin_lock_init(&device->io_lock);
 831         device->name = kstrdup(device_path, GFP_NOFS);
 832         if (!device->name) {
 833                 kfree(device);
 834                 goto out_close_bdev;
 835         }
 836         device->io_width = root->sectorsize;
 837         device->io_align = root->sectorsize;
 838         device->sector_size = root->sectorsize;
 839         device->total_bytes = i_size_read(bdev->bd_inode);
 840         device->dev_root = root->fs_info->dev_root;
 841         device->bdev = bdev;
 842
 843         ret = btrfs_add_device(trans, root, device);
 844         if (ret)
 845                 goto out_close_bdev;
 846
 847         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 848         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 849                                     total_bytes + device->total_bytes);
 850
 851         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 852         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 853                                     total_bytes + 1);
 854
 855         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
 856         list_add(&device->dev_alloc_list,
 857                  &root->fs_info->fs_devices->alloc_list);
 858         root->fs_info->fs_devices->num_devices++;
 859 out:
 860         btrfs_end_transaction(trans, root);
 861         mutex_unlock(&root->fs_info->fs_mutex);
 862         return ret;
 863
 864 out_close_bdev:
 865         close_bdev_excl(bdev);
 866         goto out;
 867 }
 868
 869 int btrfs_update_device(struct btrfs_trans_handle *trans,
 870                         struct btrfs_device *device)
 871 {
 872         int ret;
 873         struct btrfs_path *path;
 874         struct btrfs_root *root;
 875         struct btrfs_dev_item *dev_item;
 876         struct extent_buffer *leaf;
 877         struct btrfs_key key;
 878
 879         root = device->dev_root->fs_info->chunk_root;
 880
 881         path = btrfs_alloc_path();
 882         if (!path)
 883                 return -ENOMEM;
 884
 885         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 886         key.type = BTRFS_DEV_ITEM_KEY;
 887         key.offset = device->devid;
 888
 889         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 890         if (ret < 0)
 891                 goto out;
 892
 893         if (ret > 0) {
 894                 ret = -ENOENT;
 895                 goto out;
 896         }
 897
 898         leaf = path->nodes[0];
 899         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 900
 901         btrfs_set_device_id(leaf, dev_item, device->devid);
 902         btrfs_set_device_type(leaf, dev_item, device->type);
 903         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 904         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 905         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 906         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 907         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 908         btrfs_mark_buffer_dirty(leaf);
 909
 910 out:
 911         btrfs_free_path(path);
 912         return ret;
 913 }
 914
 915 int btrfs_grow_device(struct btrfs_trans_handle *trans,
 916                       struct btrfs_device *device, u64 new_size)
 917 {
 918         struct btrfs_super_block *super_copy =
 919                 &device->dev_root->fs_info->super_copy;
 920         u64 old_total = btrfs_super_total_bytes(super_copy);
 921         u64 diff = new_size - device->total_bytes;
 922
 923         btrfs_set_super_total_bytes(super_copy, old_total + diff);
 924         return btrfs_update_device(trans, device);
 925 }
 926
 927 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
 928                             struct btrfs_root *root,
 929                             u64 chunk_tree, u64 chunk_objectid,
 930                             u64 chunk_offset)
 931 {
 932         int ret;
 933         struct btrfs_path *path;
 934         struct btrfs_key key;
 935
 936         root = root->fs_info->chunk_root;
 937         path = btrfs_alloc_path();
 938         if (!path)
 939                 return -ENOMEM;
 940
 941         key.objectid = chunk_objectid;
 942         key.offset = chunk_offset;
 943         key.type = BTRFS_CHUNK_ITEM_KEY;
 944
 945         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 946         BUG_ON(ret);
 947
 948         ret = btrfs_del_item(trans, root, path);
 949         BUG_ON(ret);
 950
 951         btrfs_free_path(path);
 952         return 0;
 953 }
 954
 955 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
 956                         chunk_offset)
 957 {
 958         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 959         struct btrfs_disk_key *disk_key;
 960         struct btrfs_chunk *chunk;
 961         u8 *ptr;
 962         int ret = 0;
 963         u32 num_stripes;
 964         u32 array_size;
 965         u32 len = 0;
 966         u32 cur;
 967         struct btrfs_key key;
 968
 969         array_size = btrfs_super_sys_array_size(super_copy);
 970
 971         ptr = super_copy->sys_chunk_array;
 972         cur = 0;
 973
 974         while (cur < array_size) {
 975                 disk_key = (struct btrfs_disk_key *)ptr;
 976                 btrfs_disk_key_to_cpu(&key, disk_key);
 977
 978                 len = sizeof(*disk_key);
 979
 980                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
 981                         chunk = (struct btrfs_chunk *)(ptr + len);
 982                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
 983                         len += btrfs_chunk_item_size(num_stripes);
 984                 } else {
 985                         ret = -EIO;
 986                         break;
 987                 }
 988                 if (key.objectid == chunk_objectid &&
 989                     key.offset == chunk_offset) {
 990                         memmove(ptr, ptr + len, array_size - (cur + len));
 991                         array_size -= len;
 992                         btrfs_set_super_sys_array_size(super_copy, array_size);
 993                 } else {
 994                         ptr += len;
 995                         cur += len;
 996                 }
 997         }
 998         return ret;
 999 }
1000
1001
1002 int btrfs_relocate_chunk(struct btrfs_root *root,
1003                          u64 chunk_tree, u64 chunk_objectid,
1004                          u64 chunk_offset)
1005 {
1006         struct extent_map_tree *em_tree;
1007         struct btrfs_root *extent_root;
1008         struct btrfs_trans_handle *trans;
1009         struct extent_map *em;
1010         struct map_lookup *map;
1011         int ret;
1012         int i;
1013
1014         printk("btrfs relocating chunk %llu\n",
1015                (unsigned long long)chunk_offset);
1016         root = root->fs_info->chunk_root;
1017         extent_root = root->fs_info->extent_root;
1018         em_tree = &root->fs_info->mapping_tree.map_tree;
1019
1020         /* step one, relocate all the extents inside this chunk */
1021         ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1022         BUG_ON(ret);
1023
1024         trans = btrfs_start_transaction(root, 1);
1025         BUG_ON(!trans);
1026
1027         /*
1028          * step two, delete the device extents and the
1029          * chunk tree entries
1030          */
1031         spin_lock(&em_tree->lock);
1032         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1033         spin_unlock(&em_tree->lock);
1034
1035         BUG_ON(em->start > chunk_offset ||
1036                em->start + em->len < chunk_offset);
1037         map = (struct map_lookup *)em->bdev;
1038
1039         for (i = 0; i < map->num_stripes; i++) {
1040                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1041                                             map->stripes[i].physical);
1042                 BUG_ON(ret);
1043
1044                 ret = btrfs_update_device(trans, map->stripes[i].dev);
1045                 BUG_ON(ret);
1046         }
1047         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1048                                chunk_offset);
1049
1050         BUG_ON(ret);
1051
1052         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1053                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1054                 BUG_ON(ret);
1055         }
1056
1057         spin_lock(&em_tree->lock);
1058         remove_extent_mapping(em_tree, em);
1059         kfree(map);
1060         em->bdev = NULL;
1061
1062         /* once for the tree */
1063         free_extent_map(em);
1064         spin_unlock(&em_tree->lock);
1065
1066         /* once for us */
1067         free_extent_map(em);
1068
1069         btrfs_end_transaction(trans, root);
1070         return 0;
1071 }
1072
1073 static u64 div_factor(u64 num, int factor)
1074 {
1075         if (factor == 10)
1076                 return num;
1077         num *= factor;
1078         do_div(num, 10);
1079         return num;
1080 }
1081
1082
1083 int btrfs_balance(struct btrfs_root *dev_root)
1084 {
1085         int ret;
1086         struct list_head *cur;
1087         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1088         struct btrfs_device *device;
1089         u64 old_size;
1090         u64 size_to_free;
1091         struct btrfs_path *path;
1092         struct btrfs_key key;
1093         struct btrfs_chunk *chunk;
1094         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1095         struct btrfs_trans_handle *trans;
1096         struct btrfs_key found_key;
1097
1098
1099         dev_root = dev_root->fs_info->dev_root;
1100
1101         mutex_lock(&dev_root->fs_info->fs_mutex);
1102         /* step one make some room on all the devices */
1103         list_for_each(cur, devices) {
1104                 device = list_entry(cur, struct btrfs_device, dev_list);
1105                 old_size = device->total_bytes;
1106                 size_to_free = div_factor(old_size, 1);
1107                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1108                 if (device->total_bytes - device->bytes_used > size_to_free)
1109                         continue;
1110
1111                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1112                 BUG_ON(ret);
1113
1114                 trans = btrfs_start_transaction(dev_root, 1);
1115                 BUG_ON(!trans);
1116
1117                 ret = btrfs_grow_device(trans, device, old_size);
1118                 BUG_ON(ret);
1119
1120                 btrfs_end_transaction(trans, dev_root);
1121         }
1122
1123         /* step two, relocate all the chunks */
1124         path = btrfs_alloc_path();
1125         BUG_ON(!path);
1126
1127         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1128         key.offset = (u64)-1;
1129         key.type = BTRFS_CHUNK_ITEM_KEY;
1130
1131         while(1) {
1132                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1133                 if (ret < 0)
1134                         goto error;
1135
1136                 /*
1137                  * this shouldn't happen, it means the last relocate
1138                  * failed
1139                  */
1140                 if (ret == 0)
1141                         break;
1142
1143                 ret = btrfs_previous_item(chunk_root, path, 0,
1144                                           BTRFS_CHUNK_ITEM_KEY);
1145                 if (ret) {
1146                         break;
1147                 }
1148                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1149                                       path->slots[0]);
1150                 if (found_key.objectid != key.objectid)
1151                         break;
1152                 chunk = btrfs_item_ptr(path->nodes[0],
1153                                        path->slots[0],
1154                                        struct btrfs_chunk);
1155                 key.offset = found_key.offset;
1156                 /* chunk zero is special */
1157                 if (key.offset == 0)
1158                         break;
1159
1160                 ret = btrfs_relocate_chunk(chunk_root,
1161                                            chunk_root->root_key.objectid,
1162                                            found_key.objectid,
1163                                            found_key.offset);
1164                 BUG_ON(ret);
1165                 btrfs_release_path(chunk_root, path);
1166         }
1167         ret = 0;
1168 error:
1169         btrfs_free_path(path);
1170         mutex_unlock(&dev_root->fs_info->fs_mutex);
1171         return ret;
1172 }
1173
1174 /*
1175  * shrinking a device means finding all of the device extents past
1176  * the new size, and then following the back refs to the chunks.
1177  * The chunk relocation code actually frees the device extent
1178  */
1179 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1180 {
1181         struct btrfs_trans_handle *trans;
1182         struct btrfs_root *root = device->dev_root;
1183         struct btrfs_dev_extent *dev_extent = NULL;
1184         struct btrfs_path *path;
1185         u64 length;
1186         u64 chunk_tree;
1187         u64 chunk_objectid;
1188         u64 chunk_offset;
1189         int ret;
1190         int slot;
1191         struct extent_buffer *l;
1192         struct btrfs_key key;
1193         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1194         u64 old_total = btrfs_super_total_bytes(super_copy);
1195         u64 diff = device->total_bytes - new_size;
1196
1197
1198         path = btrfs_alloc_path();
1199         if (!path)
1200                 return -ENOMEM;
1201
1202         trans = btrfs_start_transaction(root, 1);
1203         if (!trans) {
1204                 ret = -ENOMEM;
1205                 goto done;
1206         }
1207
1208         path->reada = 2;
1209
1210         device->total_bytes = new_size;
1211         ret = btrfs_update_device(trans, device);
1212         if (ret) {
1213                 btrfs_end_transaction(trans, root);
1214                 goto done;
1215         }
1216         WARN_ON(diff > old_total);
1217         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1218         btrfs_end_transaction(trans, root);
1219
1220         key.objectid = device->devid;
1221         key.offset = (u64)-1;
1222         key.type = BTRFS_DEV_EXTENT_KEY;
1223
1224         while (1) {
1225                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1226                 if (ret < 0)
1227                         goto done;
1228
1229                 ret = btrfs_previous_item(root, path, 0, key.type);
1230                 if (ret < 0)
1231                         goto done;
1232                 if (ret) {
1233                         ret = 0;
1234                         goto done;
1235                 }
1236
1237                 l = path->nodes[0];
1238                 slot = path->slots[0];
1239                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1240
1241                 if (key.objectid != device->devid)
1242                         goto done;
1243
1244                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1245                 length = btrfs_dev_extent_length(l, dev_extent);
1246
1247                 if (key.offset + length <= new_size)
1248                         goto done;
1249
1250                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1251                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1252                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1253                 btrfs_release_path(root, path);
1254
1255                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1256                                            chunk_offset);
1257                 if (ret)
1258                         goto done;
1259         }
1260
1261 done:
1262         btrfs_free_path(path);
1263         return ret;
1264 }
1265
1266 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1267                            struct btrfs_root *root,
1268                            struct btrfs_key *key,
1269                            struct btrfs_chunk *chunk, int item_size)
1270 {
1271         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1272         struct btrfs_disk_key disk_key;
1273         u32 array_size;
1274         u8 *ptr;
1275
1276         array_size = btrfs_super_sys_array_size(super_copy);
1277         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1278                 return -EFBIG;
1279
1280         ptr = super_copy->sys_chunk_array + array_size;
1281         btrfs_cpu_key_to_disk(&disk_key, key);
1282         memcpy(ptr, &disk_key, sizeof(disk_key));
1283         ptr += sizeof(disk_key);
1284         memcpy(ptr, chunk, item_size);
1285         item_size += sizeof(disk_key);
1286         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1287         return 0;
1288 }
1289
1290 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1291                                int sub_stripes)
1292 {
1293         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1294                 return calc_size;
1295         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1296                 return calc_size * (num_stripes / sub_stripes);
1297         else
1298                 return calc_size * num_stripes;
1299 }
1300
1301
1302 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1303                       struct btrfs_root *extent_root, u64 *start,
1304                       u64 *num_bytes, u64 type)
1305 {
1306         u64 dev_offset;
1307         struct btrfs_fs_info *info = extent_root->fs_info;
1308         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1309         struct btrfs_path *path;
1310         struct btrfs_stripe *stripes;
1311         struct btrfs_device *device = NULL;
1312         struct btrfs_chunk *chunk;
1313         struct list_head private_devs;
1314         struct list_head *dev_list;
1315         struct list_head *cur;
1316         struct extent_map_tree *em_tree;
1317         struct map_lookup *map;
1318         struct extent_map *em;
1319         int min_stripe_size = 1 * 1024 * 1024;
1320         u64 physical;
1321         u64 calc_size = 1024 * 1024 * 1024;
1322         u64 max_chunk_size = calc_size;
1323         u64 min_free;
1324         u64 avail;
1325         u64 max_avail = 0;
1326         u64 percent_max;
1327         int num_stripes = 1;
1328         int min_stripes = 1;
1329         int sub_stripes = 0;
1330         int looped = 0;
1331         int ret;
1332         int index;
1333         int stripe_len = 64 * 1024;
1334         struct btrfs_key key;
1335
1336         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1337             (type & BTRFS_BLOCK_GROUP_DUP)) {
1338                 WARN_ON(1);
1339                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1340         }
1341         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1342         if (list_empty(dev_list))
1343                 return -ENOSPC;
1344
1345         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1346                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1347                 min_stripes = 2;
1348         }
1349         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1350                 num_stripes = 2;
1351                 min_stripes = 2;
1352         }
1353         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1354                 num_stripes = min_t(u64, 2,
1355                                   btrfs_super_num_devices(&info->super_copy));
1356                 if (num_stripes < 2)
1357                         return -ENOSPC;
1358                 min_stripes = 2;
1359         }
1360         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1361                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1362                 if (num_stripes < 4)
1363                         return -ENOSPC;
1364                 num_stripes &= ~(u32)1;
1365                 sub_stripes = 2;
1366                 min_stripes = 4;
1367         }
1368
1369         if (type & BTRFS_BLOCK_GROUP_DATA) {
1370                 max_chunk_size = 10 * calc_size;
1371                 min_stripe_size = 64 * 1024 * 1024;
1372         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1373                 max_chunk_size = 4 * calc_size;
1374                 min_stripe_size = 32 * 1024 * 1024;
1375         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1376                 calc_size = 8 * 1024 * 1024;
1377                 max_chunk_size = calc_size * 2;
1378                 min_stripe_size = 1 * 1024 * 1024;
1379         }
1380
1381         path = btrfs_alloc_path();
1382         if (!path)
1383                 return -ENOMEM;
1384
1385         /* we don't want a chunk larger than 10% of the FS */
1386         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1387         max_chunk_size = min(percent_max, max_chunk_size);
1388
1389 again:
1390         if (calc_size * num_stripes > max_chunk_size) {
1391                 calc_size = max_chunk_size;
1392                 do_div(calc_size, num_stripes);
1393                 do_div(calc_size, stripe_len);
1394                 calc_size *= stripe_len;
1395         }
1396         /* we don't want tiny stripes */
1397         calc_size = max_t(u64, min_stripe_size, calc_size);
1398
1399         do_div(calc_size, stripe_len);
1400         calc_size *= stripe_len;
1401
1402         INIT_LIST_HEAD(&private_devs);
1403         cur = dev_list->next;
1404         index = 0;
1405
1406         if (type & BTRFS_BLOCK_GROUP_DUP)
1407                 min_free = calc_size * 2;
1408         else
1409                 min_free = calc_size;
1410
1411         /* we add 1MB because we never use the first 1MB of the device */
1412         min_free += 1024 * 1024;
1413
1414         /* build a private list of devices we will allocate from */
1415         while(index < num_stripes) {
1416                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1417
1418                 avail = device->total_bytes - device->bytes_used;
1419                 cur = cur->next;
1420
1421                 if (avail >= min_free) {
1422                         u64 ignored_start = 0;
1423                         ret = find_free_dev_extent(trans, device, path,
1424                                                    min_free,
1425                                                    &ignored_start);
1426                         if (ret == 0) {
1427                                 list_move_tail(&device->dev_alloc_list,
1428                                                &private_devs);
1429                                 index++;
1430                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1431                                         index++;
1432                         }
1433                 } else if (avail > max_avail)
1434                         max_avail = avail;
1435                 if (cur == dev_list)
1436                         break;
1437         }
1438         if (index < num_stripes) {
1439                 list_splice(&private_devs, dev_list);
1440                 if (index >= min_stripes) {
1441                         num_stripes = index;
1442                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1443                                 num_stripes /= sub_stripes;
1444                                 num_stripes *= sub_stripes;
1445                         }
1446                         looped = 1;
1447                         goto again;
1448                 }
1449                 if (!looped && max_avail > 0) {
1450                         looped = 1;
1451                         calc_size = max_avail;
1452                         goto again;
1453                 }
1454                 btrfs_free_path(path);
1455                 return -ENOSPC;
1456         }
1457         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1458         key.type = BTRFS_CHUNK_ITEM_KEY;
1459         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1460                               &key.offset);
1461         if (ret) {
1462                 btrfs_free_path(path);
1463                 return ret;
1464         }
1465
1466         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1467         if (!chunk) {
1468                 btrfs_free_path(path);
1469                 return -ENOMEM;
1470         }
1471
1472         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1473         if (!map) {
1474                 kfree(chunk);
1475                 btrfs_free_path(path);
1476                 return -ENOMEM;
1477         }
1478         btrfs_free_path(path);
1479         path = NULL;
1480
1481         stripes = &chunk->stripe;
1482         *num_bytes = chunk_bytes_by_type(type, calc_size,
1483                                          num_stripes, sub_stripes);
1484
1485         index = 0;
1486         while(index < num_stripes) {
1487                 struct btrfs_stripe *stripe;
1488                 BUG_ON(list_empty(&private_devs));
1489                 cur = private_devs.next;
1490                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1491
1492                 /* loop over this device again if we're doing a dup group */
1493                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1494                     (index == num_stripes - 1))
1495                         list_move_tail(&device->dev_alloc_list, dev_list);
1496
1497                 ret = btrfs_alloc_dev_extent(trans, device,
1498                              info->chunk_root->root_key.objectid,
1499                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1500                              calc_size, &dev_offset);
1501                 BUG_ON(ret);
1502                 device->bytes_used += calc_size;
1503                 ret = btrfs_update_device(trans, device);
1504                 BUG_ON(ret);
1505
1506                 map->stripes[index].dev = device;
1507                 map->stripes[index].physical = dev_offset;
1508                 stripe = stripes + index;
1509                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1510                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1511                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1512                 physical = dev_offset;
1513                 index++;
1514         }
1515         BUG_ON(!list_empty(&private_devs));
1516
1517         /* key was set above */
1518         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1519         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1520         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1521         btrfs_set_stack_chunk_type(chunk, type);
1522         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1523         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1524         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1525         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1526         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1527         map->sector_size = extent_root->sectorsize;
1528         map->stripe_len = stripe_len;
1529         map->io_align = stripe_len;
1530         map->io_width = stripe_len;
1531         map->type = type;
1532         map->num_stripes = num_stripes;
1533         map->sub_stripes = sub_stripes;
1534
1535         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1536                                 btrfs_chunk_item_size(num_stripes));
1537         BUG_ON(ret);
1538         *start = key.offset;;
1539
1540         em = alloc_extent_map(GFP_NOFS);
1541         if (!em)
1542                 return -ENOMEM;
1543         em->bdev = (struct block_device *)map;
1544         em->start = key.offset;
1545         em->len = *num_bytes;
1546         em->block_start = 0;
1547
1548         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1549                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1550                                     chunk, btrfs_chunk_item_size(num_stripes));
1551                 BUG_ON(ret);
1552         }
1553         kfree(chunk);
1554
1555         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1556         spin_lock(&em_tree->lock);
1557         ret = add_extent_mapping(em_tree, em);
1558         spin_unlock(&em_tree->lock);
1559         BUG_ON(ret);
1560         free_extent_map(em);
1561         return ret;
1562 }
1563
1564 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1565 {
1566         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1567 }
1568
1569 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1570 {
1571         struct extent_map *em;
1572
1573         while(1) {
1574                 spin_lock(&tree->map_tree.lock);
1575                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1576                 if (em)
1577                         remove_extent_mapping(&tree->map_tree, em);
1578                 spin_unlock(&tree->map_tree.lock);
1579                 if (!em)
1580                         break;
1581                 kfree(em->bdev);
1582                 /* once for us */
1583                 free_extent_map(em);
1584                 /* once for the tree */
1585                 free_extent_map(em);
1586         }
1587 }
1588
1589 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1590 {
1591         struct extent_map *em;
1592         struct map_lookup *map;
1593         struct extent_map_tree *em_tree = &map_tree->map_tree;
1594         int ret;
1595
1596         spin_lock(&em_tree->lock);
1597         em = lookup_extent_mapping(em_tree, logical, len);
1598         spin_unlock(&em_tree->lock);
1599         BUG_ON(!em);
1600
1601         BUG_ON(em->start > logical || em->start + em->len < logical);
1602         map = (struct map_lookup *)em->bdev;
1603         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1604                 ret = map->num_stripes;
1605         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1606                 ret = map->sub_stripes;
1607         else
1608                 ret = 1;
1609         free_extent_map(em);
1610         return ret;
1611 }
1612
1613 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1614                              u64 logical, u64 *length,
1615                              struct btrfs_multi_bio **multi_ret,
1616                              int mirror_num, struct page *unplug_page)
1617 {
1618         struct extent_map *em;
1619         struct map_lookup *map;
1620         struct extent_map_tree *em_tree = &map_tree->map_tree;
1621         u64 offset;
1622         u64 stripe_offset;
1623         u64 stripe_nr;
1624         int stripes_allocated = 8;
1625         int stripes_required = 1;
1626         int stripe_index;
1627         int i;
1628         int num_stripes;
1629         int max_errors = 0;
1630         struct btrfs_multi_bio *multi = NULL;
1631
1632         if (multi_ret && !(rw & (1 << BIO_RW))) {
1633                 stripes_allocated = 1;
1634         }
1635 again:
1636         if (multi_ret) {
1637                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1638                                 GFP_NOFS);
1639                 if (!multi)
1640                         return -ENOMEM;
1641
1642                 atomic_set(&multi->error, 0);
1643         }
1644
1645         spin_lock(&em_tree->lock);
1646         em = lookup_extent_mapping(em_tree, logical, *length);
1647         spin_unlock(&em_tree->lock);
1648
1649         if (!em && unplug_page)
1650                 return 0;
1651
1652         if (!em) {
1653                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1654                 BUG();
1655         }
1656
1657         BUG_ON(em->start > logical || em->start + em->len < logical);
1658         map = (struct map_lookup *)em->bdev;
1659         offset = logical - em->start;
1660
1661         if (mirror_num > map->num_stripes)
1662                 mirror_num = 0;
1663
1664         /* if our multi bio struct is too small, back off and try again */
1665         if (rw & (1 << BIO_RW)) {
1666                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1667                                  BTRFS_BLOCK_GROUP_DUP)) {
1668                         stripes_required = map->num_stripes;
1669                         max_errors = 1;
1670                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1671                         stripes_required = map->sub_stripes;
1672                         max_errors = 1;
1673                 }
1674         }
1675         if (multi_ret && rw == WRITE &&
1676             stripes_allocated < stripes_required) {
1677                 stripes_allocated = map->num_stripes;
1678                 free_extent_map(em);
1679                 kfree(multi);
1680                 goto again;
1681         }
1682         stripe_nr = offset;
1683         /*
1684          * stripe_nr counts the total number of stripes we have to stride
1685          * to get to this block
1686          */
1687         do_div(stripe_nr, map->stripe_len);
1688
1689         stripe_offset = stripe_nr * map->stripe_len;
1690         BUG_ON(offset < stripe_offset);
1691
1692         /* stripe_offset is the offset of this block in its stripe*/
1693         stripe_offset = offset - stripe_offset;
1694
1695         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1696                          BTRFS_BLOCK_GROUP_RAID10 |
1697                          BTRFS_BLOCK_GROUP_DUP)) {
1698                 /* we limit the length of each bio to what fits in a stripe */
1699                 *length = min_t(u64, em->len - offset,
1700                               map->stripe_len - stripe_offset);
1701         } else {
1702                 *length = em->len - offset;
1703         }
1704
1705         if (!multi_ret && !unplug_page)
1706                 goto out;
1707
1708         num_stripes = 1;
1709         stripe_index = 0;
1710         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1711                 if (unplug_page || (rw & (1 << BIO_RW)))
1712                         num_stripes = map->num_stripes;
1713                 else if (mirror_num)
1714                         stripe_index = mirror_num - 1;
1715                 else
1716                         stripe_index = current->pid % map->num_stripes;
1717
1718         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1719                 if (rw & (1 << BIO_RW))
1720                         num_stripes = map->num_stripes;
1721                 else if (mirror_num)
1722                         stripe_index = mirror_num - 1;
1723
1724         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1725                 int factor = map->num_stripes / map->sub_stripes;
1726
1727                 stripe_index = do_div(stripe_nr, factor);
1728                 stripe_index *= map->sub_stripes;
1729
1730                 if (unplug_page || (rw & (1 << BIO_RW)))
1731                         num_stripes = map->sub_stripes;
1732                 else if (mirror_num)
1733                         stripe_index += mirror_num - 1;
1734                 else
1735                         stripe_index += current->pid % map->sub_stripes;
1736         } else {
1737                 /*
1738                  * after this do_div call, stripe_nr is the number of stripes
1739                  * on this device we have to walk to find the data, and
1740                  * stripe_index is the number of our device in the stripe array
1741                  */
1742                 stripe_index = do_div(stripe_nr, map->num_stripes);
1743         }
1744         BUG_ON(stripe_index >= map->num_stripes);
1745
1746         for (i = 0; i < num_stripes; i++) {
1747                 if (unplug_page) {
1748                         struct btrfs_device *device;
1749                         struct backing_dev_info *bdi;
1750
1751                         device = map->stripes[stripe_index].dev;
1752                         bdi = blk_get_backing_dev_info(device->bdev);
1753                         if (bdi->unplug_io_fn) {
1754                                 bdi->unplug_io_fn(bdi, unplug_page);
1755                         }
1756                 } else {
1757                         multi->stripes[i].physical =
1758                                 map->stripes[stripe_index].physical +
1759                                 stripe_offset + stripe_nr * map->stripe_len;
1760                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
1761                 }
1762                 stripe_index++;
1763         }
1764         if (multi_ret) {
1765                 *multi_ret = multi;
1766                 multi->num_stripes = num_stripes;
1767                 multi->max_errors = max_errors;
1768         }
1769 out:
1770         free_extent_map(em);
1771         return 0;
1772 }
1773
1774 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1775                       u64 logical, u64 *length,
1776                       struct btrfs_multi_bio **multi_ret, int mirror_num)
1777 {
1778         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1779                                  mirror_num, NULL);
1780 }
1781
1782 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1783                       u64 logical, struct page *page)
1784 {
1785         u64 length = PAGE_CACHE_SIZE;
1786         return __btrfs_map_block(map_tree, READ, logical, &length,
1787                                  NULL, 0, page);
1788 }
1789
1790
1791 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1792 static void end_bio_multi_stripe(struct bio *bio, int err)
1793 #else
1794 static int end_bio_multi_stripe(struct bio *bio,
1795                                    unsigned int bytes_done, int err)
1796 #endif
1797 {
1798         struct btrfs_multi_bio *multi = bio->bi_private;
1799
1800 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1801         if (bio->bi_size)
1802                 return 1;
1803 #endif
1804         if (err)
1805                 atomic_inc(&multi->error);
1806
1807         if (atomic_dec_and_test(&multi->stripes_pending)) {
1808                 bio->bi_private = multi->private;
1809                 bio->bi_end_io = multi->end_io;
1810                 /* only send an error to the higher layers if it is
1811                  * beyond the tolerance of the multi-bio
1812                  */
1813                 if (atomic_read(&multi->error) > multi->max_errors) {
1814                         err = -EIO;
1815                 } else if (err) {
1816                         /*
1817                          * this bio is actually up to date, we didn't
1818                          * go over the max number of errors
1819                          */
1820                         set_bit(BIO_UPTODATE, &bio->bi_flags);
1821                         err = 0;
1822                 }
1823                 kfree(multi);
1824
1825 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1826                 bio_endio(bio, bio->bi_size, err);
1827 #else
1828                 bio_endio(bio, err);
1829 #endif
1830         } else {
1831                 bio_put(bio);
1832         }
1833 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1834         return 0;
1835 #endif
1836 }
1837
1838 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1839                   int mirror_num)
1840 {
1841         struct btrfs_mapping_tree *map_tree;
1842         struct btrfs_device *dev;
1843         struct bio *first_bio = bio;
1844         u64 logical = bio->bi_sector << 9;
1845         u64 length = 0;
1846         u64 map_length;
1847         struct btrfs_multi_bio *multi = NULL;
1848         int ret;
1849         int dev_nr = 0;
1850         int total_devs = 1;
1851
1852         length = bio->bi_size;
1853         map_tree = &root->fs_info->mapping_tree;
1854         map_length = length;
1855
1856         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1857                               mirror_num);
1858         BUG_ON(ret);
1859
1860         total_devs = multi->num_stripes;
1861         if (map_length < length) {
1862                 printk("mapping failed logical %Lu bio len %Lu "
1863                        "len %Lu\n", logical, length, map_length);
1864                 BUG();
1865         }
1866         multi->end_io = first_bio->bi_end_io;
1867         multi->private = first_bio->bi_private;
1868         atomic_set(&multi->stripes_pending, multi->num_stripes);
1869
1870         while(dev_nr < total_devs) {
1871                 if (total_devs > 1) {
1872                         if (dev_nr < total_devs - 1) {
1873                                 bio = bio_clone(first_bio, GFP_NOFS);
1874                                 BUG_ON(!bio);
1875                         } else {
1876                                 bio = first_bio;
1877                         }
1878                         bio->bi_private = multi;
1879                         bio->bi_end_io = end_bio_multi_stripe;
1880                 }
1881                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1882                 dev = multi->stripes[dev_nr].dev;
1883
1884                 bio->bi_bdev = dev->bdev;
1885                 spin_lock(&dev->io_lock);
1886                 dev->total_ios++;
1887                 spin_unlock(&dev->io_lock);
1888                 submit_bio(rw, bio);
1889                 dev_nr++;
1890         }
1891         if (total_devs == 1)
1892                 kfree(multi);
1893         return 0;
1894 }
1895
1896 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1897                                        u8 *uuid)
1898 {
1899         struct list_head *head = &root->fs_info->fs_devices->devices;
1900
1901         return __find_device(head, devid, uuid);
1902 }
1903
1904 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1905                           struct extent_buffer *leaf,
1906                           struct btrfs_chunk *chunk)
1907 {
1908         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1909         struct map_lookup *map;
1910         struct extent_map *em;
1911         u64 logical;
1912         u64 length;
1913         u64 devid;
1914         u8 uuid[BTRFS_UUID_SIZE];
1915         int num_stripes;
1916         int ret;
1917         int i;
1918
1919         logical = key->offset;
1920         length = btrfs_chunk_length(leaf, chunk);
1921
1922         spin_lock(&map_tree->map_tree.lock);
1923         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1924         spin_unlock(&map_tree->map_tree.lock);
1925
1926         /* already mapped? */
1927         if (em && em->start <= logical && em->start + em->len > logical) {
1928                 free_extent_map(em);
1929                 return 0;
1930         } else if (em) {
1931                 free_extent_map(em);
1932         }
1933
1934         map = kzalloc(sizeof(*map), GFP_NOFS);
1935         if (!map)
1936                 return -ENOMEM;
1937
1938         em = alloc_extent_map(GFP_NOFS);
1939         if (!em)
1940                 return -ENOMEM;
1941         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1942         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1943         if (!map) {
1944                 free_extent_map(em);
1945                 return -ENOMEM;
1946         }
1947
1948         em->bdev = (struct block_device *)map;
1949         em->start = logical;
1950         em->len = length;
1951         em->block_start = 0;
1952
1953         map->num_stripes = num_stripes;
1954         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1955         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1956         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1957         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1958         map->type = btrfs_chunk_type(leaf, chunk);
1959         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1960         for (i = 0; i < num_stripes; i++) {
1961                 map->stripes[i].physical =
1962                         btrfs_stripe_offset_nr(leaf, chunk, i);
1963                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1964                 read_extent_buffer(leaf, uuid, (unsigned long)
1965                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1966                                    BTRFS_UUID_SIZE);
1967                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1968                 if (!map->stripes[i].dev) {
1969                         kfree(map);
1970                         free_extent_map(em);
1971                         return -EIO;
1972                 }
1973         }
1974
1975         spin_lock(&map_tree->map_tree.lock);
1976         ret = add_extent_mapping(&map_tree->map_tree, em);
1977         spin_unlock(&map_tree->map_tree.lock);
1978         BUG_ON(ret);
1979         free_extent_map(em);
1980
1981         return 0;
1982 }
1983
1984 static int fill_device_from_item(struct extent_buffer *leaf,
1985                                  struct btrfs_dev_item *dev_item,
1986                                  struct btrfs_device *device)
1987 {
1988         unsigned long ptr;
1989
1990         device->devid = btrfs_device_id(leaf, dev_item);
1991         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1992         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1993         device->type = btrfs_device_type(leaf, dev_item);
1994         device->io_align = btrfs_device_io_align(leaf, dev_item);
1995         device->io_width = btrfs_device_io_width(leaf, dev_item);
1996         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1997
1998         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1999         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2000
2001         return 0;
2002 }
2003
2004 static int read_one_dev(struct btrfs_root *root,
2005                         struct extent_buffer *leaf,
2006                         struct btrfs_dev_item *dev_item)
2007 {
2008         struct btrfs_device *device;
2009         u64 devid;
2010         int ret;
2011         u8 dev_uuid[BTRFS_UUID_SIZE];
2012
2013         devid = btrfs_device_id(leaf, dev_item);
2014         read_extent_buffer(leaf, dev_uuid,
2015                            (unsigned long)btrfs_device_uuid(dev_item),
2016                            BTRFS_UUID_SIZE);
2017         device = btrfs_find_device(root, devid, dev_uuid);
2018         if (!device) {
2019                 printk("warning devid %Lu not found already\n", devid);
2020                 device = kzalloc(sizeof(*device), GFP_NOFS);
2021                 if (!device)
2022                         return -ENOMEM;
2023                 list_add(&device->dev_list,
2024                          &root->fs_info->fs_devices->devices);
2025                 list_add(&device->dev_alloc_list,
2026                          &root->fs_info->fs_devices->alloc_list);
2027                 device->barriers = 1;
2028                 spin_lock_init(&device->io_lock);
2029         }
2030
2031         fill_device_from_item(leaf, dev_item, device);
2032         device->dev_root = root->fs_info->dev_root;
2033         ret = 0;
2034 #if 0
2035         ret = btrfs_open_device(device);
2036         if (ret) {
2037                 kfree(device);
2038         }
2039 #endif
2040         return ret;
2041 }
2042
2043 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2044 {
2045         struct btrfs_dev_item *dev_item;
2046
2047         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2048                                                      dev_item);
2049         return read_one_dev(root, buf, dev_item);
2050 }
2051
2052 int btrfs_read_sys_array(struct btrfs_root *root)
2053 {
2054         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2055         struct extent_buffer *sb;
2056         struct btrfs_disk_key *disk_key;
2057         struct btrfs_chunk *chunk;
2058         u8 *ptr;
2059         unsigned long sb_ptr;
2060         int ret = 0;
2061         u32 num_stripes;
2062         u32 array_size;
2063         u32 len = 0;
2064         u32 cur;
2065         struct btrfs_key key;
2066
2067         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2068                                           BTRFS_SUPER_INFO_SIZE);
2069         if (!sb)
2070                 return -ENOMEM;
2071         btrfs_set_buffer_uptodate(sb);
2072         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2073         array_size = btrfs_super_sys_array_size(super_copy);
2074
2075         ptr = super_copy->sys_chunk_array;
2076         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2077         cur = 0;
2078
2079         while (cur < array_size) {
2080                 disk_key = (struct btrfs_disk_key *)ptr;
2081                 btrfs_disk_key_to_cpu(&key, disk_key);
2082
2083                 len = sizeof(*disk_key); ptr += len;
2084                 sb_ptr += len;
2085                 cur += len;
2086
2087                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2088                         chunk = (struct btrfs_chunk *)sb_ptr;
2089                         ret = read_one_chunk(root, &key, sb, chunk);
2090                         if (ret)
2091                                 break;
2092                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2093                         len = btrfs_chunk_item_size(num_stripes);
2094                 } else {
2095                         ret = -EIO;
2096                         break;
2097                 }
2098                 ptr += len;
2099                 sb_ptr += len;
2100                 cur += len;
2101         }
2102         free_extent_buffer(sb);
2103         return ret;
2104 }
2105
2106 int btrfs_read_chunk_tree(struct btrfs_root *root)
2107 {
2108         struct btrfs_path *path;
2109         struct extent_buffer *leaf;
2110         struct btrfs_key key;
2111         struct btrfs_key found_key;
2112         int ret;
2113         int slot;
2114
2115         root = root->fs_info->chunk_root;
2116
2117         path = btrfs_alloc_path();
2118         if (!path)
2119                 return -ENOMEM;
2120
2121         /* first we search for all of the device items, and then we
2122          * read in all of the chunk items.  This way we can create chunk
2123          * mappings that reference all of the devices that are afound
2124          */
2125         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2126         key.offset = 0;
2127         key.type = 0;
2128 again:
2129         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2130         while(1) {
2131                 leaf = path->nodes[0];
2132                 slot = path->slots[0];
2133                 if (slot >= btrfs_header_nritems(leaf)) {
2134                         ret = btrfs_next_leaf(root, path);
2135                         if (ret == 0)
2136                                 continue;
2137                         if (ret < 0)
2138                                 goto error;
2139                         break;
2140                 }
2141                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2142                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2143                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2144                                 break;
2145                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2146                                 struct btrfs_dev_item *dev_item;
2147                                 dev_item = btrfs_item_ptr(leaf, slot,
2148                                                   struct btrfs_dev_item);
2149                                 ret = read_one_dev(root, leaf, dev_item);
2150                                 BUG_ON(ret);
2151                         }
2152                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2153                         struct btrfs_chunk *chunk;
2154                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2155                         ret = read_one_chunk(root, &found_key, leaf, chunk);
2156                 }
2157                 path->slots[0]++;
2158         }
2159         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2160                 key.objectid = 0;
2161                 btrfs_release_path(root, path);
2162                 goto again;
2163         }
2164
2165         btrfs_free_path(path);
2166         ret = 0;
2167 error:
2168         return ret;
2169 }
2170