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 #define _XOPEN_SOURCE 600
  19 #define __USE_XOPEN2K
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <uuid/uuid.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "print-tree.h"
  31 #include "volumes.h"
  32
  33 struct stripe {
  34         struct btrfs_device *dev;
  35         u64 physical;
  36 };
  37
  38 struct map_lookup {
  39         struct cache_extent ce;
  40         u64 type;
  41         int io_align;
  42         int io_width;
  43         int stripe_len;
  44         int sector_size;
  45         int num_stripes;
  46         struct stripe stripes[];
  47 };
  48
  49 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  50                             (sizeof(struct stripe) * (n)))
  51
  52 static LIST_HEAD(fs_uuids);
  53
  54 static struct btrfs_device *__find_device(struct list_head *head, u64 devid)
  55 {
  56         struct btrfs_device *dev;
  57         struct list_head *cur;
  58
  59         list_for_each(cur, head) {
  60                 dev = list_entry(cur, struct btrfs_device, dev_list);
  61                 if (dev->devid == devid)
  62                         return dev;
  63         }
  64         return NULL;
  65 }
  66
  67 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  68 {
  69         struct list_head *cur;
  70         struct btrfs_fs_devices *fs_devices;
  71
  72         list_for_each(cur, &fs_uuids) {
  73                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  74                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  75                         return fs_devices;
  76         }
  77         return NULL;
  78 }
  79
  80 static int device_list_add(const char *path,
  81                            struct btrfs_super_block *disk_super,
  82                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
  83 {
  84         struct btrfs_device *device;
  85         struct btrfs_fs_devices *fs_devices;
  86         u64 found_transid = btrfs_super_generation(disk_super);
  87
  88         fs_devices = find_fsid(disk_super->fsid);
  89         if (!fs_devices) {
  90                 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
  91                 if (!fs_devices)
  92                         return -ENOMEM;
  93                 INIT_LIST_HEAD(&fs_devices->devices);
  94                 list_add(&fs_devices->list, &fs_uuids);
  95                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
  96                 fs_devices->latest_devid = devid;
  97                 fs_devices->latest_trans = found_transid;
  98                 fs_devices->lowest_devid = (u64)-1;
  99                 device = NULL;
 100         } else {
 101                 device = __find_device(&fs_devices->devices, devid);
 102         }
 103         if (!device) {
 104                 device = kzalloc(sizeof(*device), GFP_NOFS);
 105                 if (!device) {
 106                         /* we can safely leave the fs_devices entry around */
 107                         return -ENOMEM;
 108                 }
 109                 device->devid = devid;
 110                 device->name = kstrdup(path, GFP_NOFS);
 111                 if (!device->name) {
 112                         kfree(device);
 113                         return -ENOMEM;
 114                 }
 115                 list_add(&device->dev_list, &fs_devices->devices);
 116         }
 117
 118         if (found_transid > fs_devices->latest_trans) {
 119                 fs_devices->latest_devid = devid;
 120                 fs_devices->latest_trans = found_transid;
 121         }
 122         if (fs_devices->lowest_devid > devid) {
 123                 fs_devices->lowest_devid = devid;
 124                 printk("lowest devid now %llu\n", (unsigned long long)devid);
 125         }
 126         *fs_devices_ret = fs_devices;
 127         return 0;
 128 }
 129
 130 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 131 {
 132         struct list_head *head = &fs_devices->devices;
 133         struct list_head *cur;
 134         struct btrfs_device *device;
 135
 136         list_for_each(cur, head) {
 137                 device = list_entry(cur, struct btrfs_device, dev_list);
 138                 device->fd = 0;
 139         }
 140         return 0;
 141 }
 142
 143 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
 144 {
 145         int fd;
 146         struct list_head *head = &fs_devices->devices;
 147         struct list_head *cur;
 148         struct btrfs_device *device;
 149         int ret;
 150
 151         list_for_each(cur, head) {
 152                 device = list_entry(cur, struct btrfs_device, dev_list);
 153                 fd = open(device->name, flags);
 154 printk("opening %s devid %llu fd %d\n", device->name,
 155         (unsigned long long)device->devid, fd);
 156                 if (fd < 0) {
 157                         ret = -errno;
 158                         goto fail;
 159                 }
 160                 if (device->devid == fs_devices->latest_devid)
 161                         fs_devices->latest_bdev = fd;
 162                 if (device->devid == fs_devices->lowest_devid)
 163                         fs_devices->lowest_bdev = fd;
 164                 device->fd = fd;
 165         }
 166         return 0;
 167 fail:
 168         btrfs_close_devices(fs_devices);
 169         return ret;
 170 }
 171
 172 int btrfs_scan_one_device(int fd, const char *path,
 173                           struct btrfs_fs_devices **fs_devices_ret,
 174                           u64 *total_devs, u64 super_offset)
 175 {
 176         struct btrfs_super_block *disk_super;
 177         char *buf;
 178         int ret;
 179         u64 devid;
 180
 181         buf = malloc(4096);
 182         if (!buf) {
 183                 ret = -ENOMEM;
 184                 goto error;
 185         }
 186         ret = pread(fd, buf, 4096, super_offset);
 187         if (ret != 4096) {
 188                 ret = -EIO;
 189                 goto error;
 190         }
 191         disk_super = (struct btrfs_super_block *)buf;
 192         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 193             sizeof(disk_super->magic))) {
 194                 ret = -ENOENT;
 195                 goto error_brelse;
 196         }
 197         devid = le64_to_cpu(disk_super->dev_item.devid);
 198         *total_devs = btrfs_super_num_devices(disk_super);
 199         printk("found device %llu on %s\n", (unsigned long long)devid, path);
 200         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 201
 202 error_brelse:
 203         free(buf);
 204 error:
 205         return ret;
 206 }
 207
 208 /*
 209  * this uses a pretty simple search, the expectation is that it is
 210  * called very infrequently and that a given device has a small number
 211  * of extents
 212  */
 213 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 214                                 struct btrfs_device *device,
 215                                 struct btrfs_path *path,
 216                                 u64 num_bytes, u64 *start)
 217 {
 218         struct btrfs_key key;
 219         struct btrfs_root *root = device->dev_root;
 220         struct btrfs_dev_extent *dev_extent = NULL;
 221         u64 hole_size = 0;
 222         u64 last_byte = 0;
 223         u64 search_start = 0;
 224         u64 search_end = device->total_bytes;
 225         int ret;
 226         int slot = 0;
 227         int start_found;
 228         struct extent_buffer *l;
 229
 230         start_found = 0;
 231         path->reada = 2;
 232
 233         /* FIXME use last free of some kind */
 234
 235         /* we don't want to overwrite the superblock on the drive,
 236          * so we make sure to start at an offset of at least 1MB
 237          */
 238         search_start = max((u64)1024 * 1024, search_start);
 239         key.objectid = device->devid;
 240         key.offset = search_start;
 241         key.type = BTRFS_DEV_EXTENT_KEY;
 242         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 243         if (ret < 0)
 244                 goto error;
 245         ret = btrfs_previous_item(root, path, 0, key.type);
 246         if (ret < 0)
 247                 goto error;
 248         l = path->nodes[0];
 249         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 250         while (1) {
 251                 l = path->nodes[0];
 252                 slot = path->slots[0];
 253                 if (slot >= btrfs_header_nritems(l)) {
 254                         ret = btrfs_next_leaf(root, path);
 255                         if (ret == 0)
 256                                 continue;
 257                         if (ret < 0)
 258                                 goto error;
 259 no_more_items:
 260                         if (!start_found) {
 261                                 if (search_start >= search_end) {
 262                                         ret = -ENOSPC;
 263                                         goto error;
 264                                 }
 265                                 *start = search_start;
 266                                 start_found = 1;
 267                                 goto check_pending;
 268                         }
 269                         *start = last_byte > search_start ?
 270                                 last_byte : search_start;
 271                         if (search_end <= *start) {
 272                                 ret = -ENOSPC;
 273                                 goto error;
 274                         }
 275                         goto check_pending;
 276                 }
 277                 btrfs_item_key_to_cpu(l, &key, slot);
 278
 279                 if (key.objectid < device->devid)
 280                         goto next;
 281
 282                 if (key.objectid > device->devid)
 283                         goto no_more_items;
 284
 285                 if (key.offset >= search_start && key.offset > last_byte &&
 286                     start_found) {
 287                         if (last_byte < search_start)
 288                                 last_byte = search_start;
 289                         hole_size = key.offset - last_byte;
 290                         if (key.offset > last_byte &&
 291                             hole_size >= num_bytes) {
 292                                 *start = last_byte;
 293                                 goto check_pending;
 294                         }
 295                 }
 296                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 297                         goto next;
 298                 }
 299
 300                 start_found = 1;
 301                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 302                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 303 next:
 304                 path->slots[0]++;
 305                 cond_resched();
 306         }
 307 check_pending:
 308         /* we have to make sure we didn't find an extent that has already
 309          * been allocated by the map tree or the original allocation
 310          */
 311         btrfs_release_path(root, path);
 312         BUG_ON(*start < search_start);
 313
 314         if (*start + num_bytes > search_end) {
 315                 ret = -ENOSPC;
 316                 goto error;
 317         }
 318         /* check for pending inserts here */
 319         return 0;
 320
 321 error:
 322         btrfs_release_path(root, path);
 323         return ret;
 324 }
 325
 326 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 327                            struct btrfs_device *device,
 328                            u64 owner, u64 num_bytes, u64 *start)
 329 {
 330         int ret;
 331         struct btrfs_path *path;
 332         struct btrfs_root *root = device->dev_root;
 333         struct btrfs_dev_extent *extent;
 334         struct extent_buffer *leaf;
 335         struct btrfs_key key;
 336
 337         path = btrfs_alloc_path();
 338         if (!path)
 339                 return -ENOMEM;
 340
 341         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 342         if (ret) {
 343                 goto err;
 344         }
 345
 346         key.objectid = device->devid;
 347         key.offset = *start;
 348         key.type = BTRFS_DEV_EXTENT_KEY;
 349         ret = btrfs_insert_empty_item(trans, root, path, &key,
 350                                       sizeof(*extent));
 351         BUG_ON(ret);
 352
 353         leaf = path->nodes[0];
 354         extent = btrfs_item_ptr(leaf, path->slots[0],
 355                                 struct btrfs_dev_extent);
 356         btrfs_set_dev_extent_owner(leaf, extent, owner);
 357         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 358         btrfs_mark_buffer_dirty(leaf);
 359 err:
 360         btrfs_free_path(path);
 361         return ret;
 362 }
 363
 364 static int find_next_chunk(struct btrfs_root *root, u64 *objectid)
 365 {
 366         struct btrfs_path *path;
 367         int ret;
 368         struct btrfs_key key;
 369         struct btrfs_key found_key;
 370
 371         path = btrfs_alloc_path();
 372         BUG_ON(!path);
 373
 374         key.objectid = (u64)-1;
 375         key.offset = (u64)-1;
 376         key.type = BTRFS_CHUNK_ITEM_KEY;
 377
 378         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 379         if (ret < 0)
 380                 goto error;
 381
 382         BUG_ON(ret == 0);
 383
 384         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 385         if (ret) {
 386                 *objectid = 0;
 387         } else {
 388                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 389                                       path->slots[0]);
 390                 *objectid = found_key.objectid + found_key.offset;
 391         }
 392         ret = 0;
 393 error:
 394         btrfs_free_path(path);
 395         return ret;
 396 }
 397
 398 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 399                            u64 *objectid)
 400 {
 401         int ret;
 402         struct btrfs_key key;
 403         struct btrfs_key found_key;
 404
 405         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 406         key.type = BTRFS_DEV_ITEM_KEY;
 407         key.offset = (u64)-1;
 408
 409         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 410         if (ret < 0)
 411                 goto error;
 412
 413         BUG_ON(ret == 0);
 414
 415         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 416                                   BTRFS_DEV_ITEM_KEY);
 417         if (ret) {
 418                 *objectid = 1;
 419         } else {
 420                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 421                                       path->slots[0]);
 422                 *objectid = found_key.offset + 1;
 423         }
 424         ret = 0;
 425 error:
 426         btrfs_release_path(root, path);
 427         return ret;
 428 }
 429
 430 /*
 431  * the device information is stored in the chunk root
 432  * the btrfs_device struct should be fully filled in
 433  */
 434 int btrfs_add_device(struct btrfs_trans_handle *trans,
 435                      struct btrfs_root *root,
 436                      struct btrfs_device *device)
 437 {
 438         int ret;
 439         struct btrfs_path *path;
 440         struct btrfs_dev_item *dev_item;
 441         struct extent_buffer *leaf;
 442         struct btrfs_key key;
 443         unsigned long ptr;
 444         u64 free_devid;
 445
 446         root = root->fs_info->chunk_root;
 447
 448         path = btrfs_alloc_path();
 449         if (!path)
 450                 return -ENOMEM;
 451
 452         ret = find_next_devid(root, path, &free_devid);
 453         if (ret)
 454                 goto out;
 455
 456         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 457         key.type = BTRFS_DEV_ITEM_KEY;
 458         key.offset = free_devid;
 459
 460         ret = btrfs_insert_empty_item(trans, root, path, &key,
 461                                       sizeof(*dev_item));
 462         if (ret)
 463                 goto out;
 464
 465         leaf = path->nodes[0];
 466         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 467
 468         device->devid = free_devid;
 469         btrfs_set_device_id(leaf, dev_item, device->devid);
 470         btrfs_set_device_type(leaf, dev_item, device->type);
 471         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 472         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 473         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 474         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 475         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 476
 477         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 478         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
 479         btrfs_mark_buffer_dirty(leaf);
 480         ret = 0;
 481
 482 out:
 483         btrfs_free_path(path);
 484         return ret;
 485 }
 486
 487 int btrfs_update_device(struct btrfs_trans_handle *trans,
 488                         struct btrfs_device *device)
 489 {
 490         int ret;
 491         struct btrfs_path *path;
 492         struct btrfs_root *root;
 493         struct btrfs_dev_item *dev_item;
 494         struct extent_buffer *leaf;
 495         struct btrfs_key key;
 496
 497         root = device->dev_root->fs_info->chunk_root;
 498
 499         path = btrfs_alloc_path();
 500         if (!path)
 501                 return -ENOMEM;
 502
 503         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 504         key.type = BTRFS_DEV_ITEM_KEY;
 505         key.offset = device->devid;
 506
 507         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 508         if (ret < 0)
 509                 goto out;
 510
 511         if (ret > 0) {
 512                 ret = -ENOENT;
 513                 goto out;
 514         }
 515
 516         leaf = path->nodes[0];
 517         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 518
 519         btrfs_set_device_id(leaf, dev_item, device->devid);
 520         btrfs_set_device_type(leaf, dev_item, device->type);
 521         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 522         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 523         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 524         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 525         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 526         btrfs_mark_buffer_dirty(leaf);
 527
 528 out:
 529         btrfs_free_path(path);
 530         return ret;
 531 }
 532
 533 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 534                            struct btrfs_root *root,
 535                            struct btrfs_key *key,
 536                            struct btrfs_chunk *chunk, int item_size)
 537 {
 538         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 539         struct btrfs_disk_key disk_key;
 540         u32 array_size;
 541         u8 *ptr;
 542
 543         array_size = btrfs_super_sys_array_size(super_copy);
 544         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 545                 return -EFBIG;
 546
 547         ptr = super_copy->sys_chunk_array + array_size;
 548         btrfs_cpu_key_to_disk(&disk_key, key);
 549         memcpy(ptr, &disk_key, sizeof(disk_key));
 550         ptr += sizeof(disk_key);
 551         memcpy(ptr, chunk, item_size);
 552         item_size += sizeof(disk_key);
 553         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 554         return 0;
 555 }
 556
 557 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 558                       struct btrfs_root *extent_root, u64 *start,
 559                       u64 *num_bytes, u64 type)
 560 {
 561         u64 dev_offset;
 562         struct btrfs_fs_info *info = extent_root->fs_info;
 563         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
 564         struct btrfs_stripe *stripes;
 565         struct btrfs_device *device = NULL;
 566         struct btrfs_chunk *chunk;
 567         struct list_head private_devs;
 568         struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
 569         struct list_head *cur;
 570         struct map_lookup *map;
 571         u64 physical;
 572         u64 calc_size = 8 * 1024 * 1024;
 573         u64 avail;
 574         u64 max_avail = 0;
 575         int num_stripes = 1;
 576         int looped = 0;
 577         int ret;
 578         int index;
 579         int stripe_len = 64 * 1024;
 580         struct btrfs_key key;
 581
 582         if (list_empty(dev_list))
 583                 return -ENOSPC;
 584
 585         if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1)) {
 586                 calc_size = 1024 * 1024 * 1024;
 587         }
 588         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
 589                 num_stripes = min_t(u64, 2,
 590                                   btrfs_super_num_devices(&info->super_copy));
 591         }
 592         if (type & (BTRFS_BLOCK_GROUP_RAID0))
 593                 num_stripes = btrfs_super_num_devices(&info->super_copy);
 594 again:
 595         INIT_LIST_HEAD(&private_devs);
 596         cur = dev_list->next;
 597         index = 0;
 598         /* build a private list of devices we will allocate from */
 599         while(index < num_stripes) {
 600                 device = list_entry(cur, struct btrfs_device, dev_list);
 601                 avail = device->total_bytes - device->bytes_used;
 602                 cur = cur->next;
 603                 if (avail > max_avail)
 604                         max_avail = avail;
 605                 if (avail >= calc_size) {
 606                         list_move_tail(&device->dev_list, &private_devs);
 607                         index++;
 608                 }
 609                 if (cur == dev_list)
 610                         break;
 611         }
 612         if (index < num_stripes) {
 613                 list_splice(&private_devs, dev_list);
 614                 if (!looped && max_avail > 0) {
 615                         looped = 1;
 616                         calc_size = max_avail;
 617                         goto again;
 618                 }
 619                 return -ENOSPC;
 620         }
 621
 622         ret = find_next_chunk(chunk_root, &key.objectid);
 623         if (ret)
 624                 return ret;
 625
 626         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 627         if (!chunk)
 628                 return -ENOMEM;
 629
 630         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
 631         if (!map) {
 632                 kfree(chunk);
 633                 return -ENOMEM;
 634         }
 635
 636         stripes = &chunk->stripe;
 637
 638         if (type & BTRFS_BLOCK_GROUP_RAID1)
 639                 *num_bytes = calc_size;
 640         else
 641                 *num_bytes = calc_size * num_stripes;
 642
 643         index = 0;
 644         while(index < num_stripes) {
 645                 BUG_ON(list_empty(&private_devs));
 646                 cur = private_devs.next;
 647                 device = list_entry(cur, struct btrfs_device, dev_list);
 648                 list_move_tail(&device->dev_list, dev_list);
 649
 650                 ret = btrfs_alloc_dev_extent(trans, device,
 651                                              key.objectid,
 652                                              calc_size, &dev_offset);
 653                 BUG_ON(ret);
 654 printk("alloc chunk size %llu from dev %llu\n",
 655         (unsigned long long)calc_size,
 656         (unsigned long long)device->devid);
 657                 device->bytes_used += calc_size;
 658                 ret = btrfs_update_device(trans, device);
 659                 BUG_ON(ret);
 660
 661                 map->stripes[index].dev = device;
 662                 map->stripes[index].physical = dev_offset;
 663                 btrfs_set_stack_stripe_devid(stripes + index, device->devid);
 664                 btrfs_set_stack_stripe_offset(stripes + index, dev_offset);
 665                 physical = dev_offset;
 666                 index++;
 667         }
 668         BUG_ON(!list_empty(&private_devs));
 669
 670         /* key.objectid was set above */
 671         key.offset = *num_bytes;
 672         key.type = BTRFS_CHUNK_ITEM_KEY;
 673         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 674         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 675         btrfs_set_stack_chunk_type(chunk, type);
 676         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 677         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 678         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 679         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 680         map->sector_size = extent_root->sectorsize;
 681         map->stripe_len = stripe_len;
 682         map->io_align = stripe_len;
 683         map->io_width = stripe_len;
 684         map->type = type;
 685         map->num_stripes = num_stripes;
 686
 687         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 688                                 btrfs_chunk_item_size(num_stripes));
 689         BUG_ON(ret);
 690         *start = key.objectid;
 691
 692         map->ce.start = key.objectid;
 693         map->ce.size = key.offset;
 694
 695         ret = insert_existing_cache_extent(
 696                            &extent_root->fs_info->mapping_tree.cache_tree,
 697                            &map->ce);
 698         BUG_ON(ret);
 699
 700         kfree(chunk);
 701         return ret;
 702 }
 703
 704 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
 705 {
 706         cache_tree_init(&tree->cache_tree);
 707 }
 708
 709 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
 710                     int dev_nr, u64 logical, u64 *phys, u64 *length,
 711                     struct btrfs_device **dev, int *total_devs)
 712 {
 713         struct cache_extent *ce;
 714         struct map_lookup *map;
 715         u64 offset;
 716         u64 stripe_offset;
 717         u64 stripe_nr;
 718         int stripe_index;
 719
 720         ce = find_first_cache_extent(&map_tree->cache_tree, logical);
 721         BUG_ON(!ce);
 722         BUG_ON(ce->start > logical || ce->start + ce->size < logical);
 723         map = container_of(ce, struct map_lookup, ce);
 724         offset = logical - ce->start;
 725
 726         stripe_nr = offset;
 727         /*
 728          * stripe_nr counts the total number of stripes we have to stride
 729          * to get to this block
 730          */
 731         stripe_nr = stripe_nr / map->stripe_len;
 732
 733         stripe_offset = stripe_nr * map->stripe_len;
 734         BUG_ON(offset < stripe_offset);
 735
 736         /* stripe_offset is the offset of this block in its stripe*/
 737         stripe_offset = offset - stripe_offset;
 738
 739         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
 740                 stripe_index = dev_nr;
 741                 if (rw == WRITE)
 742                         *total_devs = map->num_stripes;
 743                 else {
 744                         stripe_index = stripe_nr % map->num_stripes;
 745                         *total_devs = 1;
 746                 }
 747         } else {
 748                 /*
 749                  * after this do_div call, stripe_nr is the number of stripes
 750                  * on this device we have to walk to find the data, and
 751                  * stripe_index is the number of our device in the stripe array
 752                  */
 753                 stripe_index = stripe_nr % map->num_stripes;
 754                 stripe_nr = stripe_nr / map->num_stripes;
 755         }
 756         BUG_ON(stripe_index >= map->num_stripes);
 757         *phys = map->stripes[stripe_index].physical + stripe_offset +
 758                 stripe_nr * map->stripe_len;
 759
 760         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1)) {
 761                 /* we limit the length of each bio to what fits in a stripe */
 762                 *length = min_t(u64, ce->size - offset,
 763                               map->stripe_len - stripe_offset);
 764         } else {
 765                 *length = ce->size - offset;
 766         }
 767         *dev = map->stripes[stripe_index].dev;
 768         return 0;
 769 }
 770
 771 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid)
 772 {
 773         struct list_head *head = &root->fs_info->fs_devices->devices;
 774
 775         return __find_device(head, devid);
 776 }
 777
 778 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
 779                           struct extent_buffer *leaf,
 780                           struct btrfs_chunk *chunk)
 781 {
 782         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
 783         struct map_lookup *map;
 784         struct cache_extent *ce;
 785         u64 logical;
 786         u64 length;
 787         u64 devid;
 788         int num_stripes;
 789         int ret;
 790         int i;
 791
 792         logical = key->objectid;
 793         length = key->offset;
 794         ce = find_first_cache_extent(&map_tree->cache_tree, logical);
 795
 796         /* already mapped? */
 797         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
 798                 return 0;
 799         }
 800
 801         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
 802         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
 803         if (!map)
 804                 return -ENOMEM;
 805
 806         map->ce.start = logical;
 807         map->ce.size = length;
 808
 809         map->num_stripes = num_stripes;
 810         map->io_width = btrfs_chunk_io_width(leaf, chunk);
 811         map->io_align = btrfs_chunk_io_align(leaf, chunk);
 812         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
 813         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
 814         map->type = btrfs_chunk_type(leaf, chunk);
 815         for (i = 0; i < num_stripes; i++) {
 816                 map->stripes[i].physical =
 817                         btrfs_stripe_offset_nr(leaf, chunk, i);
 818                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
 819                 map->stripes[i].dev = btrfs_find_device(root, devid);
 820                 if (!map->stripes[i].dev) {
 821                         kfree(map);
 822                         return -EIO;
 823                 }
 824
 825         }
 826         ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
 827         BUG_ON(ret);
 828
 829         return 0;
 830 }
 831
 832 static int fill_device_from_item(struct extent_buffer *leaf,
 833                                  struct btrfs_dev_item *dev_item,
 834                                  struct btrfs_device *device)
 835 {
 836         unsigned long ptr;
 837
 838         device->devid = btrfs_device_id(leaf, dev_item);
 839         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
 840         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
 841         device->type = btrfs_device_type(leaf, dev_item);
 842         device->io_align = btrfs_device_io_align(leaf, dev_item);
 843         device->io_width = btrfs_device_io_width(leaf, dev_item);
 844         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
 845
 846         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 847         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
 848
 849         return 0;
 850 }
 851
 852 static int read_one_dev(struct btrfs_root *root,
 853                         struct extent_buffer *leaf,
 854                         struct btrfs_dev_item *dev_item)
 855 {
 856         struct btrfs_device *device;
 857         u64 devid;
 858         int ret = 0;
 859
 860         devid = btrfs_device_id(leaf, dev_item);
 861         device = btrfs_find_device(root, devid);
 862         if (!device) {
 863                 printk("warning devid %llu not found already\n",
 864                         (unsigned long long)devid);
 865                 device = kmalloc(sizeof(*device), GFP_NOFS);
 866                 if (!device)
 867                         return -ENOMEM;
 868                 device->total_ios = 0;
 869                 list_add(&device->dev_list,
 870                          &root->fs_info->fs_devices->devices);
 871         }
 872
 873         fill_device_from_item(leaf, dev_item, device);
 874         device->dev_root = root->fs_info->dev_root;
 875         return ret;
 876 }
 877
 878 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
 879 {
 880         struct btrfs_dev_item *dev_item;
 881
 882         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
 883                                                      dev_item);
 884         return read_one_dev(root, buf, dev_item);
 885 }
 886
 887 int btrfs_read_sys_array(struct btrfs_root *root)
 888 {
 889         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 890         struct extent_buffer *sb = root->fs_info->sb_buffer;
 891         struct btrfs_disk_key *disk_key;
 892         struct btrfs_chunk *chunk;
 893         struct btrfs_key key;
 894         u32 num_stripes;
 895         u32 array_size;
 896         u32 len = 0;
 897         u8 *ptr;
 898         unsigned long sb_ptr;
 899         u32 cur;
 900         int ret;
 901
 902         array_size = btrfs_super_sys_array_size(super_copy);
 903
 904         /*
 905          * we do this loop twice, once for the device items and
 906          * once for all of the chunks.  This way there are device
 907          * structs filled in for every chunk
 908          */
 909         ptr = super_copy->sys_chunk_array;
 910         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
 911         cur = 0;
 912
 913         while (cur < array_size) {
 914                 disk_key = (struct btrfs_disk_key *)ptr;
 915                 btrfs_disk_key_to_cpu(&key, disk_key);
 916
 917                 len = sizeof(*disk_key);
 918                 ptr += len;
 919                 sb_ptr += len;
 920                 cur += len;
 921
 922                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
 923                         chunk = (struct btrfs_chunk *)sb_ptr;
 924                         ret = read_one_chunk(root, &key, sb, chunk);
 925                         BUG_ON(ret);
 926                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
 927                         len = btrfs_chunk_item_size(num_stripes);
 928                 } else {
 929                         BUG();
 930                 }
 931                 ptr += len;
 932                 sb_ptr += len;
 933                 cur += len;
 934         }
 935         return 0;
 936 }
 937
 938 int btrfs_read_chunk_tree(struct btrfs_root *root)
 939 {
 940         struct btrfs_path *path;
 941         struct extent_buffer *leaf;
 942         struct btrfs_key key;
 943         struct btrfs_key found_key;
 944         int ret;
 945         int slot;
 946
 947         root = root->fs_info->chunk_root;
 948
 949         path = btrfs_alloc_path();
 950         if (!path)
 951                 return -ENOMEM;
 952
 953         /* first we search for all of the device items, and then we
 954          * read in all of the chunk items.  This way we can create chunk
 955          * mappings that reference all of the devices that are afound
 956          */
 957         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 958         key.offset = 0;
 959         key.type = 0;
 960 again:
 961         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 962         while(1) {
 963                 leaf = path->nodes[0];
 964                 slot = path->slots[0];
 965                 if (slot >= btrfs_header_nritems(leaf)) {
 966                         ret = btrfs_next_leaf(root, path);
 967                         if (ret == 0)
 968                                 continue;
 969                         if (ret < 0)
 970                                 goto error;
 971                         break;
 972                 }
 973                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 974                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
 975                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
 976                                 break;
 977                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
 978                                 struct btrfs_dev_item *dev_item;
 979                                 dev_item = btrfs_item_ptr(leaf, slot,
 980                                                   struct btrfs_dev_item);
 981                                 ret = read_one_dev(root, leaf, dev_item);
 982                                 BUG_ON(ret);
 983                         }
 984                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
 985                         struct btrfs_chunk *chunk;
 986                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
 987                         ret = read_one_chunk(root, &found_key, leaf, chunk);
 988                 }
 989                 path->slots[0]++;
 990         }
 991         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
 992                 key.objectid = 0;
 993                 btrfs_release_path(root, path);
 994                 goto again;
 995         }
 996
 997         btrfs_free_path(path);
 998         ret = 0;
 999 error:
1000         return ret;
1001 }
1002