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 <stdio.h>
  19 #include <stdlib.h>
  20 #include <sys/types.h>
  21 #include <sys/stat.h>
  22 #include <uuid/uuid.h>
  23 #include <fcntl.h>
  24 #include <unistd.h>
  25 #include "ctree.h"
  26 #include "disk-io.h"
  27 #include "transaction.h"
  28 #include "print-tree.h"
  29 #include "volumes.h"
  30 #include "utils.h"
  31
  32 struct stripe {
  33         struct btrfs_device *dev;
  34         u64 physical;
  35 };
  36
  37 static inline int nr_parity_stripes(struct map_lookup *map)
  38 {
  39         if (map->type & BTRFS_BLOCK_GROUP_RAID5)
  40                 return 1;
  41         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
  42                 return 2;
  43         else
  44                 return 0;
  45 }
  46
  47 static inline int nr_data_stripes(struct map_lookup *map)
  48 {
  49         return map->num_stripes - nr_parity_stripes(map);
  50 }
  51
  52 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
  53
  54 static LIST_HEAD(fs_uuids);
  55
  56 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  57                                           u8 *uuid)
  58 {
  59         struct btrfs_device *dev;
  60         struct list_head *cur;
  61
  62         list_for_each(cur, head) {
  63                 dev = list_entry(cur, struct btrfs_device, dev_list);
  64                 if (dev->devid == devid &&
  65                     !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
  66                         return dev;
  67                 }
  68         }
  69         return NULL;
  70 }
  71
  72 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  73 {
  74         struct list_head *cur;
  75         struct btrfs_fs_devices *fs_devices;
  76
  77         list_for_each(cur, &fs_uuids) {
  78                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  79                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  80                         return fs_devices;
  81         }
  82         return NULL;
  83 }
  84
  85 static int device_list_add(const char *path,
  86                            struct btrfs_super_block *disk_super,
  87                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
  88 {
  89         struct btrfs_device *device;
  90         struct btrfs_fs_devices *fs_devices;
  91         u64 found_transid = btrfs_super_generation(disk_super);
  92
  93         fs_devices = find_fsid(disk_super->fsid);
  94         if (!fs_devices) {
  95                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
  96                 if (!fs_devices)
  97                         return -ENOMEM;
  98                 INIT_LIST_HEAD(&fs_devices->devices);
  99                 list_add(&fs_devices->list, &fs_uuids);
 100                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 101                 fs_devices->latest_devid = devid;
 102                 fs_devices->latest_trans = found_transid;
 103                 fs_devices->lowest_devid = (u64)-1;
 104                 device = NULL;
 105         } else {
 106                 device = __find_device(&fs_devices->devices, devid,
 107                                        disk_super->dev_item.uuid);
 108         }
 109         if (!device) {
 110                 device = kzalloc(sizeof(*device), GFP_NOFS);
 111                 if (!device) {
 112                         /* we can safely leave the fs_devices entry around */
 113                         return -ENOMEM;
 114                 }
 115                 device->fd = -1;
 116                 device->devid = devid;
 117                 device->generation = found_transid;
 118                 memcpy(device->uuid, disk_super->dev_item.uuid,
 119                        BTRFS_UUID_SIZE);
 120                 device->name = kstrdup(path, GFP_NOFS);
 121                 if (!device->name) {
 122                         kfree(device);
 123                         return -ENOMEM;
 124                 }
 125                 device->label = kstrdup(disk_super->label, GFP_NOFS);
 126                 if (!device->label) {
 127                         kfree(device->name);
 128                         kfree(device);
 129                         return -ENOMEM;
 130                 }
 131                 device->total_devs = btrfs_super_num_devices(disk_super);
 132                 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
 133                 device->total_bytes =
 134                         btrfs_stack_device_total_bytes(&disk_super->dev_item);
 135                 device->bytes_used =
 136                         btrfs_stack_device_bytes_used(&disk_super->dev_item);
 137                 list_add(&device->dev_list, &fs_devices->devices);
 138                 device->fs_devices = fs_devices;
 139         } else if (!device->name || strcmp(device->name, path)) {
 140                 char *name = strdup(path);
 141                 if (!name)
 142                         return -ENOMEM;
 143                 kfree(device->name);
 144                 device->name = name;
 145         }
 146
 147
 148         if (found_transid > fs_devices->latest_trans) {
 149                 fs_devices->latest_devid = devid;
 150                 fs_devices->latest_trans = found_transid;
 151         }
 152         if (fs_devices->lowest_devid > devid) {
 153                 fs_devices->lowest_devid = devid;
 154         }
 155         *fs_devices_ret = fs_devices;
 156         return 0;
 157 }
 158
 159 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 160 {
 161         struct btrfs_fs_devices *seed_devices;
 162         struct btrfs_device *device;
 163
 164 again:
 165         while (!list_empty(&fs_devices->devices)) {
 166                 device = list_entry(fs_devices->devices.next,
 167                                     struct btrfs_device, dev_list);
 168                 if (device->fd != -1) {
 169                         fsync(device->fd);
 170                         if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
 171                                 fprintf(stderr, "Warning, could not drop caches\n");
 172                         close(device->fd);
 173                         device->fd = -1;
 174                 }
 175                 device->writeable = 0;
 176                 list_del(&device->dev_list);
 177                 /* free the memory */
 178                 free(device->name);
 179                 free(device->label);
 180                 free(device);
 181         }
 182
 183         seed_devices = fs_devices->seed;
 184         fs_devices->seed = NULL;
 185         if (seed_devices) {
 186                 struct btrfs_fs_devices *orig;
 187
 188                 orig = fs_devices;
 189                 fs_devices = seed_devices;
 190                 list_del(&orig->list);
 191                 free(orig);
 192                 goto again;
 193         } else {
 194                 list_del(&fs_devices->list);
 195                 free(fs_devices);
 196         }
 197
 198         return 0;
 199 }
 200
 201 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
 202 {
 203         int fd;
 204         struct list_head *head = &fs_devices->devices;
 205         struct list_head *cur;
 206         struct btrfs_device *device;
 207         int ret;
 208
 209         list_for_each(cur, head) {
 210                 device = list_entry(cur, struct btrfs_device, dev_list);
 211                 if (!device->name) {
 212                         printk("no name for device %llu, skip it now\n", device->devid);
 213                         continue;
 214                 }
 215
 216                 fd = open(device->name, flags);
 217                 if (fd < 0) {
 218                         ret = -errno;
 219                         goto fail;
 220                 }
 221
 222                 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
 223                         fprintf(stderr, "Warning, could not drop caches\n");
 224
 225                 if (device->devid == fs_devices->latest_devid)
 226                         fs_devices->latest_bdev = fd;
 227                 if (device->devid == fs_devices->lowest_devid)
 228                         fs_devices->lowest_bdev = fd;
 229                 device->fd = fd;
 230                 if (flags & O_RDWR)
 231                         device->writeable = 1;
 232         }
 233         return 0;
 234 fail:
 235         btrfs_close_devices(fs_devices);
 236         return ret;
 237 }
 238
 239 int btrfs_scan_one_device(int fd, const char *path,
 240                           struct btrfs_fs_devices **fs_devices_ret,
 241                           u64 *total_devs, u64 super_offset, int super_recover)
 242 {
 243         struct btrfs_super_block *disk_super;
 244         char *buf;
 245         int ret;
 246         u64 devid;
 247
 248         buf = malloc(4096);
 249         if (!buf) {
 250                 ret = -ENOMEM;
 251                 goto error;
 252         }
 253         disk_super = (struct btrfs_super_block *)buf;
 254         ret = btrfs_read_dev_super(fd, disk_super, super_offset, super_recover);
 255         if (ret < 0) {
 256                 ret = -EIO;
 257                 goto error_brelse;
 258         }
 259         devid = btrfs_stack_device_id(&disk_super->dev_item);
 260         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
 261                 *total_devs = 1;
 262         else
 263                 *total_devs = btrfs_super_num_devices(disk_super);
 264
 265         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 266
 267 error_brelse:
 268         free(buf);
 269 error:
 270         return ret;
 271 }
 272
 273 /*
 274  * this uses a pretty simple search, the expectation is that it is
 275  * called very infrequently and that a given device has a small number
 276  * of extents
 277  */
 278 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 279                                 struct btrfs_device *device,
 280                                 struct btrfs_path *path,
 281                                 u64 num_bytes, u64 *start)
 282 {
 283         struct btrfs_key key;
 284         struct btrfs_root *root = device->dev_root;
 285         struct btrfs_dev_extent *dev_extent = NULL;
 286         u64 hole_size = 0;
 287         u64 last_byte = 0;
 288         u64 search_start = root->fs_info->alloc_start;
 289         u64 search_end = device->total_bytes;
 290         int ret;
 291         int slot = 0;
 292         int start_found;
 293         struct extent_buffer *l;
 294
 295         start_found = 0;
 296         path->reada = 2;
 297
 298         /* FIXME use last free of some kind */
 299
 300         /* we don't want to overwrite the superblock on the drive,
 301          * so we make sure to start at an offset of at least 1MB
 302          */
 303         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 304
 305         if (search_start >= search_end) {
 306                 ret = -ENOSPC;
 307                 goto error;
 308         }
 309
 310         key.objectid = device->devid;
 311         key.offset = search_start;
 312         key.type = BTRFS_DEV_EXTENT_KEY;
 313         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 314         if (ret < 0)
 315                 goto error;
 316         ret = btrfs_previous_item(root, path, 0, key.type);
 317         if (ret < 0)
 318                 goto error;
 319         l = path->nodes[0];
 320         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 321         while (1) {
 322                 l = path->nodes[0];
 323                 slot = path->slots[0];
 324                 if (slot >= btrfs_header_nritems(l)) {
 325                         ret = btrfs_next_leaf(root, path);
 326                         if (ret == 0)
 327                                 continue;
 328                         if (ret < 0)
 329                                 goto error;
 330 no_more_items:
 331                         if (!start_found) {
 332                                 if (search_start >= search_end) {
 333                                         ret = -ENOSPC;
 334                                         goto error;
 335                                 }
 336                                 *start = search_start;
 337                                 start_found = 1;
 338                                 goto check_pending;
 339                         }
 340                         *start = last_byte > search_start ?
 341                                 last_byte : search_start;
 342                         if (search_end <= *start) {
 343                                 ret = -ENOSPC;
 344                                 goto error;
 345                         }
 346                         goto check_pending;
 347                 }
 348                 btrfs_item_key_to_cpu(l, &key, slot);
 349
 350                 if (key.objectid < device->devid)
 351                         goto next;
 352
 353                 if (key.objectid > device->devid)
 354                         goto no_more_items;
 355
 356                 if (key.offset >= search_start && key.offset > last_byte &&
 357                     start_found) {
 358                         if (last_byte < search_start)
 359                                 last_byte = search_start;
 360                         hole_size = key.offset - last_byte;
 361                         if (key.offset > last_byte &&
 362                             hole_size >= num_bytes) {
 363                                 *start = last_byte;
 364                                 goto check_pending;
 365                         }
 366                 }
 367                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 368                         goto next;
 369                 }
 370
 371                 start_found = 1;
 372                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 373                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 374 next:
 375                 path->slots[0]++;
 376                 cond_resched();
 377         }
 378 check_pending:
 379         /* we have to make sure we didn't find an extent that has already
 380          * been allocated by the map tree or the original allocation
 381          */
 382         btrfs_release_path(path);
 383         BUG_ON(*start < search_start);
 384
 385         if (*start + num_bytes > search_end) {
 386                 ret = -ENOSPC;
 387                 goto error;
 388         }
 389         /* check for pending inserts here */
 390         return 0;
 391
 392 error:
 393         btrfs_release_path(path);
 394         return ret;
 395 }
 396
 397 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 398                                   struct btrfs_device *device,
 399                                   u64 chunk_tree, u64 chunk_objectid,
 400                                   u64 chunk_offset,
 401                                   u64 num_bytes, u64 *start)
 402 {
 403         int ret;
 404         struct btrfs_path *path;
 405         struct btrfs_root *root = device->dev_root;
 406         struct btrfs_dev_extent *extent;
 407         struct extent_buffer *leaf;
 408         struct btrfs_key key;
 409
 410         path = btrfs_alloc_path();
 411         if (!path)
 412                 return -ENOMEM;
 413
 414         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 415         if (ret) {
 416                 goto err;
 417         }
 418
 419         key.objectid = device->devid;
 420         key.offset = *start;
 421         key.type = BTRFS_DEV_EXTENT_KEY;
 422         ret = btrfs_insert_empty_item(trans, root, path, &key,
 423                                       sizeof(*extent));
 424         BUG_ON(ret);
 425
 426         leaf = path->nodes[0];
 427         extent = btrfs_item_ptr(leaf, path->slots[0],
 428                                 struct btrfs_dev_extent);
 429         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 430         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 431         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 432
 433         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 434                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 435                     BTRFS_UUID_SIZE);
 436
 437         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 438         btrfs_mark_buffer_dirty(leaf);
 439 err:
 440         btrfs_free_path(path);
 441         return ret;
 442 }
 443
 444 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 445 {
 446         struct btrfs_path *path;
 447         int ret;
 448         struct btrfs_key key;
 449         struct btrfs_chunk *chunk;
 450         struct btrfs_key found_key;
 451
 452         path = btrfs_alloc_path();
 453         BUG_ON(!path);
 454
 455         key.objectid = objectid;
 456         key.offset = (u64)-1;
 457         key.type = BTRFS_CHUNK_ITEM_KEY;
 458
 459         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 460         if (ret < 0)
 461                 goto error;
 462
 463         BUG_ON(ret == 0);
 464
 465         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 466         if (ret) {
 467                 *offset = 0;
 468         } else {
 469                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 470                                       path->slots[0]);
 471                 if (found_key.objectid != objectid)
 472                         *offset = 0;
 473                 else {
 474                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 475                                                struct btrfs_chunk);
 476                         *offset = found_key.offset +
 477                                 btrfs_chunk_length(path->nodes[0], chunk);
 478                 }
 479         }
 480         ret = 0;
 481 error:
 482         btrfs_free_path(path);
 483         return ret;
 484 }
 485
 486 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 487                            u64 *objectid)
 488 {
 489         int ret;
 490         struct btrfs_key key;
 491         struct btrfs_key found_key;
 492
 493         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 494         key.type = BTRFS_DEV_ITEM_KEY;
 495         key.offset = (u64)-1;
 496
 497         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 498         if (ret < 0)
 499                 goto error;
 500
 501         BUG_ON(ret == 0);
 502
 503         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 504                                   BTRFS_DEV_ITEM_KEY);
 505         if (ret) {
 506                 *objectid = 1;
 507         } else {
 508                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 509                                       path->slots[0]);
 510                 *objectid = found_key.offset + 1;
 511         }
 512         ret = 0;
 513 error:
 514         btrfs_release_path(path);
 515         return ret;
 516 }
 517
 518 /*
 519  * the device information is stored in the chunk root
 520  * the btrfs_device struct should be fully filled in
 521  */
 522 int btrfs_add_device(struct btrfs_trans_handle *trans,
 523                      struct btrfs_root *root,
 524                      struct btrfs_device *device)
 525 {
 526         int ret;
 527         struct btrfs_path *path;
 528         struct btrfs_dev_item *dev_item;
 529         struct extent_buffer *leaf;
 530         struct btrfs_key key;
 531         unsigned long ptr;
 532         u64 free_devid = 0;
 533
 534         root = root->fs_info->chunk_root;
 535
 536         path = btrfs_alloc_path();
 537         if (!path)
 538                 return -ENOMEM;
 539
 540         ret = find_next_devid(root, path, &free_devid);
 541         if (ret)
 542                 goto out;
 543
 544         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 545         key.type = BTRFS_DEV_ITEM_KEY;
 546         key.offset = free_devid;
 547
 548         ret = btrfs_insert_empty_item(trans, root, path, &key,
 549                                       sizeof(*dev_item));
 550         if (ret)
 551                 goto out;
 552
 553         leaf = path->nodes[0];
 554         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 555
 556         device->devid = free_devid;
 557         btrfs_set_device_id(leaf, dev_item, device->devid);
 558         btrfs_set_device_generation(leaf, dev_item, 0);
 559         btrfs_set_device_type(leaf, dev_item, device->type);
 560         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 561         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 562         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 563         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 564         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 565         btrfs_set_device_group(leaf, dev_item, 0);
 566         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 567         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 568         btrfs_set_device_start_offset(leaf, dev_item, 0);
 569
 570         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 571         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 572         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 573         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 574         btrfs_mark_buffer_dirty(leaf);
 575         ret = 0;
 576
 577 out:
 578         btrfs_free_path(path);
 579         return ret;
 580 }
 581
 582 int btrfs_update_device(struct btrfs_trans_handle *trans,
 583                         struct btrfs_device *device)
 584 {
 585         int ret;
 586         struct btrfs_path *path;
 587         struct btrfs_root *root;
 588         struct btrfs_dev_item *dev_item;
 589         struct extent_buffer *leaf;
 590         struct btrfs_key key;
 591
 592         root = device->dev_root->fs_info->chunk_root;
 593
 594         path = btrfs_alloc_path();
 595         if (!path)
 596                 return -ENOMEM;
 597
 598         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 599         key.type = BTRFS_DEV_ITEM_KEY;
 600         key.offset = device->devid;
 601
 602         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 603         if (ret < 0)
 604                 goto out;
 605
 606         if (ret > 0) {
 607                 ret = -ENOENT;
 608                 goto out;
 609         }
 610
 611         leaf = path->nodes[0];
 612         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 613
 614         btrfs_set_device_id(leaf, dev_item, device->devid);
 615         btrfs_set_device_type(leaf, dev_item, device->type);
 616         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 617         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 618         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 619         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 620         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 621         btrfs_mark_buffer_dirty(leaf);
 622
 623 out:
 624         btrfs_free_path(path);
 625         return ret;
 626 }
 627
 628 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 629                            struct btrfs_root *root,
 630                            struct btrfs_key *key,
 631                            struct btrfs_chunk *chunk, int item_size)
 632 {
 633         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
 634         struct btrfs_disk_key disk_key;
 635         u32 array_size;
 636         u8 *ptr;
 637
 638         array_size = btrfs_super_sys_array_size(super_copy);
 639         if (array_size + item_size + sizeof(disk_key)
 640                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 641                 return -EFBIG;
 642
 643         ptr = super_copy->sys_chunk_array + array_size;
 644         btrfs_cpu_key_to_disk(&disk_key, key);
 645         memcpy(ptr, &disk_key, sizeof(disk_key));
 646         ptr += sizeof(disk_key);
 647         memcpy(ptr, chunk, item_size);
 648         item_size += sizeof(disk_key);
 649         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 650         return 0;
 651 }
 652
 653 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 654                                int sub_stripes)
 655 {
 656         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 657                 return calc_size;
 658         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 659                 return calc_size * (num_stripes / sub_stripes);
 660         else if (type & BTRFS_BLOCK_GROUP_RAID5)
 661                 return calc_size * (num_stripes - 1);
 662         else if (type & BTRFS_BLOCK_GROUP_RAID6)
 663                 return calc_size * (num_stripes - 2);
 664         else
 665                 return calc_size * num_stripes;
 666 }
 667
 668
 669 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
 670 {
 671         /* TODO, add a way to store the preferred stripe size */
 672         return BTRFS_STRIPE_LEN;
 673 }
 674
 675 /*
 676  * btrfs_device_avail_bytes - count bytes available for alloc_chunk
 677  *
 678  * It is not equal to "device->total_bytes - device->bytes_used".
 679  * We do not allocate any chunk in 1M at beginning of device, and not
 680  * allowed to allocate any chunk before alloc_start if it is specified.
 681  * So search holes from max(1M, alloc_start) to device->total_bytes.
 682  */
 683 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
 684                                     struct btrfs_device *device,
 685                                     u64 *avail_bytes)
 686 {
 687         struct btrfs_path *path;
 688         struct btrfs_root *root = device->dev_root;
 689         struct btrfs_key key;
 690         struct btrfs_dev_extent *dev_extent = NULL;
 691         struct extent_buffer *l;
 692         u64 search_start = root->fs_info->alloc_start;
 693         u64 search_end = device->total_bytes;
 694         u64 extent_end = 0;
 695         u64 free_bytes = 0;
 696         int ret;
 697         int slot = 0;
 698
 699         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 700
 701         path = btrfs_alloc_path();
 702         if (!path)
 703                 return -ENOMEM;
 704
 705         key.objectid = device->devid;
 706         key.offset = root->fs_info->alloc_start;
 707         key.type = BTRFS_DEV_EXTENT_KEY;
 708
 709         path->reada = 2;
 710         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 711         if (ret < 0)
 712                 goto error;
 713         ret = btrfs_previous_item(root, path, 0, key.type);
 714         if (ret < 0)
 715                 goto error;
 716
 717         while (1) {
 718                 l = path->nodes[0];
 719                 slot = path->slots[0];
 720                 if (slot >= btrfs_header_nritems(l)) {
 721                         ret = btrfs_next_leaf(root, path);
 722                         if (ret == 0)
 723                                 continue;
 724                         if (ret < 0)
 725                                 goto error;
 726                         break;
 727                 }
 728                 btrfs_item_key_to_cpu(l, &key, slot);
 729
 730                 if (key.objectid < device->devid)
 731                         goto next;
 732                 if (key.objectid > device->devid)
 733                         break;
 734                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 735                         goto next;
 736                 if (key.offset > search_end)
 737                         break;
 738                 if (key.offset > search_start)
 739                         free_bytes += key.offset - search_start;
 740
 741                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 742                 extent_end = key.offset + btrfs_dev_extent_length(l,
 743                                                                   dev_extent);
 744                 if (extent_end > search_start)
 745                         search_start = extent_end;
 746                 if (search_start > search_end)
 747                         break;
 748 next:
 749                 path->slots[0]++;
 750                 cond_resched();
 751         }
 752
 753         if (search_start < search_end)
 754                 free_bytes += search_end - search_start;
 755
 756         *avail_bytes = free_bytes;
 757         ret = 0;
 758 error:
 759         btrfs_free_path(path);
 760         return ret;
 761 }
 762
 763 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
 764                         - sizeof(struct btrfs_item)             \
 765                         - sizeof(struct btrfs_chunk))           \
 766                         / sizeof(struct btrfs_stripe) + 1)
 767
 768 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
 769                                 - 2 * sizeof(struct btrfs_disk_key)     \
 770                                 - 2 * sizeof(struct btrfs_chunk))       \
 771                                 / sizeof(struct btrfs_stripe) + 1)
 772
 773 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 774                       struct btrfs_root *extent_root, u64 *start,
 775                       u64 *num_bytes, u64 type)
 776 {
 777         u64 dev_offset;
 778         struct btrfs_fs_info *info = extent_root->fs_info;
 779         struct btrfs_root *chunk_root = info->chunk_root;
 780         struct btrfs_stripe *stripes;
 781         struct btrfs_device *device = NULL;
 782         struct btrfs_chunk *chunk;
 783         struct list_head private_devs;
 784         struct list_head *dev_list = &info->fs_devices->devices;
 785         struct list_head *cur;
 786         struct map_lookup *map;
 787         int min_stripe_size = 1 * 1024 * 1024;
 788         u64 calc_size = 8 * 1024 * 1024;
 789         u64 min_free;
 790         u64 max_chunk_size = 4 * calc_size;
 791         u64 avail = 0;
 792         u64 max_avail = 0;
 793         u64 percent_max;
 794         int num_stripes = 1;
 795         int max_stripes = 0;
 796         int min_stripes = 1;
 797         int sub_stripes = 0;
 798         int looped = 0;
 799         int ret;
 800         int index;
 801         int stripe_len = BTRFS_STRIPE_LEN;
 802         struct btrfs_key key;
 803         u64 offset;
 804
 805         if (list_empty(dev_list)) {
 806                 return -ENOSPC;
 807         }
 808
 809         if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
 810                     BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
 811                     BTRFS_BLOCK_GROUP_RAID10 |
 812                     BTRFS_BLOCK_GROUP_DUP)) {
 813                 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 814                         calc_size = 8 * 1024 * 1024;
 815                         max_chunk_size = calc_size * 2;
 816                         min_stripe_size = 1 * 1024 * 1024;
 817                         max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
 818                 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
 819                         calc_size = 1024 * 1024 * 1024;
 820                         max_chunk_size = 10 * calc_size;
 821                         min_stripe_size = 64 * 1024 * 1024;
 822                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 823                 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 824                         calc_size = 1024 * 1024 * 1024;
 825                         max_chunk_size = 4 * calc_size;
 826                         min_stripe_size = 32 * 1024 * 1024;
 827                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 828                 }
 829         }
 830         if (type & BTRFS_BLOCK_GROUP_RAID1) {
 831                 num_stripes = min_t(u64, 2,
 832                                   btrfs_super_num_devices(info->super_copy));
 833                 if (num_stripes < 2)
 834                         return -ENOSPC;
 835                 min_stripes = 2;
 836         }
 837         if (type & BTRFS_BLOCK_GROUP_DUP) {
 838                 num_stripes = 2;
 839                 min_stripes = 2;
 840         }
 841         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
 842                 num_stripes = btrfs_super_num_devices(info->super_copy);
 843                 if (num_stripes > max_stripes)
 844                         num_stripes = max_stripes;
 845                 min_stripes = 2;
 846         }
 847         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 848                 num_stripes = btrfs_super_num_devices(info->super_copy);
 849                 if (num_stripes > max_stripes)
 850                         num_stripes = max_stripes;
 851                 if (num_stripes < 4)
 852                         return -ENOSPC;
 853                 num_stripes &= ~(u32)1;
 854                 sub_stripes = 2;
 855                 min_stripes = 4;
 856         }
 857         if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
 858                 num_stripes = btrfs_super_num_devices(info->super_copy);
 859                 if (num_stripes > max_stripes)
 860                         num_stripes = max_stripes;
 861                 if (num_stripes < 2)
 862                         return -ENOSPC;
 863                 min_stripes = 2;
 864                 stripe_len = find_raid56_stripe_len(num_stripes - 1,
 865                                     btrfs_super_stripesize(info->super_copy));
 866         }
 867         if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
 868                 num_stripes = btrfs_super_num_devices(info->super_copy);
 869                 if (num_stripes > max_stripes)
 870                         num_stripes = max_stripes;
 871                 if (num_stripes < 3)
 872                         return -ENOSPC;
 873                 min_stripes = 3;
 874                 stripe_len = find_raid56_stripe_len(num_stripes - 2,
 875                                     btrfs_super_stripesize(info->super_copy));
 876         }
 877
 878         /* we don't want a chunk larger than 10% of the FS */
 879         percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
 880         max_chunk_size = min(percent_max, max_chunk_size);
 881
 882 again:
 883         if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
 884             max_chunk_size) {
 885                 calc_size = max_chunk_size;
 886                 calc_size /= num_stripes;
 887                 calc_size /= stripe_len;
 888                 calc_size *= stripe_len;
 889         }
 890         /* we don't want tiny stripes */
 891         calc_size = max_t(u64, calc_size, min_stripe_size);
 892
 893         calc_size /= stripe_len;
 894         calc_size *= stripe_len;
 895         INIT_LIST_HEAD(&private_devs);
 896         cur = dev_list->next;
 897         index = 0;
 898
 899         if (type & BTRFS_BLOCK_GROUP_DUP)
 900                 min_free = calc_size * 2;
 901         else
 902                 min_free = calc_size;
 903
 904         /* build a private list of devices we will allocate from */
 905         while(index < num_stripes) {
 906                 device = list_entry(cur, struct btrfs_device, dev_list);
 907                 ret = btrfs_device_avail_bytes(trans, device, &avail);
 908                 if (ret)
 909                         return ret;
 910                 cur = cur->next;
 911                 if (avail >= min_free) {
 912                         list_move_tail(&device->dev_list, &private_devs);
 913                         index++;
 914                         if (type & BTRFS_BLOCK_GROUP_DUP)
 915                                 index++;
 916                 } else if (avail > max_avail)
 917                         max_avail = avail;
 918                 if (cur == dev_list)
 919                         break;
 920         }
 921         if (index < num_stripes) {
 922                 list_splice(&private_devs, dev_list);
 923                 if (index >= min_stripes) {
 924                         num_stripes = index;
 925                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 926                                 num_stripes /= sub_stripes;
 927                                 num_stripes *= sub_stripes;
 928                         }
 929                         looped = 1;
 930                         goto again;
 931                 }
 932                 if (!looped && max_avail > 0) {
 933                         looped = 1;
 934                         calc_size = max_avail;
 935                         goto again;
 936                 }
 937                 return -ENOSPC;
 938         }
 939         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 940                               &offset);
 941         if (ret)
 942                 return ret;
 943         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 944         key.type = BTRFS_CHUNK_ITEM_KEY;
 945         key.offset = offset;
 946
 947         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 948         if (!chunk)
 949                 return -ENOMEM;
 950
 951         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
 952         if (!map) {
 953                 kfree(chunk);
 954                 return -ENOMEM;
 955         }
 956
 957         stripes = &chunk->stripe;
 958         *num_bytes = chunk_bytes_by_type(type, calc_size,
 959                                          num_stripes, sub_stripes);
 960         index = 0;
 961         while(index < num_stripes) {
 962                 struct btrfs_stripe *stripe;
 963                 BUG_ON(list_empty(&private_devs));
 964                 cur = private_devs.next;
 965                 device = list_entry(cur, struct btrfs_device, dev_list);
 966
 967                 /* loop over this device again if we're doing a dup group */
 968                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 969                     (index == num_stripes - 1))
 970                         list_move_tail(&device->dev_list, dev_list);
 971
 972                 ret = btrfs_alloc_dev_extent(trans, device,
 973                              info->chunk_root->root_key.objectid,
 974                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 975                              calc_size, &dev_offset);
 976                 BUG_ON(ret);
 977
 978                 device->bytes_used += calc_size;
 979                 ret = btrfs_update_device(trans, device);
 980                 BUG_ON(ret);
 981
 982                 map->stripes[index].dev = device;
 983                 map->stripes[index].physical = dev_offset;
 984                 stripe = stripes + index;
 985                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 986                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 987                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 988                 index++;
 989         }
 990         BUG_ON(!list_empty(&private_devs));
 991
 992         /* key was set above */
 993         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 994         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 995         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 996         btrfs_set_stack_chunk_type(chunk, type);
 997         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 998         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 999         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1000         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1001         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1002         map->sector_size = extent_root->sectorsize;
1003         map->stripe_len = stripe_len;
1004         map->io_align = stripe_len;
1005         map->io_width = stripe_len;
1006         map->type = type;
1007         map->num_stripes = num_stripes;
1008         map->sub_stripes = sub_stripes;
1009
1010         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1011                                 btrfs_chunk_item_size(num_stripes));
1012         BUG_ON(ret);
1013         *start = key.offset;;
1014
1015         map->ce.start = key.offset;
1016         map->ce.size = *num_bytes;
1017
1018         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1019         BUG_ON(ret);
1020
1021         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1022                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1023                                     chunk, btrfs_chunk_item_size(num_stripes));
1024                 BUG_ON(ret);
1025         }
1026
1027         kfree(chunk);
1028         return ret;
1029 }
1030
1031 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1032                            struct btrfs_root *extent_root, u64 *start,
1033                            u64 num_bytes, u64 type)
1034 {
1035         u64 dev_offset;
1036         struct btrfs_fs_info *info = extent_root->fs_info;
1037         struct btrfs_root *chunk_root = info->chunk_root;
1038         struct btrfs_stripe *stripes;
1039         struct btrfs_device *device = NULL;
1040         struct btrfs_chunk *chunk;
1041         struct list_head *dev_list = &info->fs_devices->devices;
1042         struct list_head *cur;
1043         struct map_lookup *map;
1044         u64 calc_size = 8 * 1024 * 1024;
1045         int num_stripes = 1;
1046         int sub_stripes = 0;
1047         int ret;
1048         int index;
1049         int stripe_len = BTRFS_STRIPE_LEN;
1050         struct btrfs_key key;
1051
1052         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1053         key.type = BTRFS_CHUNK_ITEM_KEY;
1054         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1055                               &key.offset);
1056         if (ret)
1057                 return ret;
1058
1059         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1060         if (!chunk)
1061                 return -ENOMEM;
1062
1063         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1064         if (!map) {
1065                 kfree(chunk);
1066                 return -ENOMEM;
1067         }
1068
1069         stripes = &chunk->stripe;
1070         calc_size = num_bytes;
1071
1072         index = 0;
1073         cur = dev_list->next;
1074         device = list_entry(cur, struct btrfs_device, dev_list);
1075
1076         while (index < num_stripes) {
1077                 struct btrfs_stripe *stripe;
1078
1079                 ret = btrfs_alloc_dev_extent(trans, device,
1080                              info->chunk_root->root_key.objectid,
1081                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1082                              calc_size, &dev_offset);
1083                 BUG_ON(ret);
1084
1085                 device->bytes_used += calc_size;
1086                 ret = btrfs_update_device(trans, device);
1087                 BUG_ON(ret);
1088
1089                 map->stripes[index].dev = device;
1090                 map->stripes[index].physical = dev_offset;
1091                 stripe = stripes + index;
1092                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1093                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1094                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1095                 index++;
1096         }
1097
1098         /* key was set above */
1099         btrfs_set_stack_chunk_length(chunk, num_bytes);
1100         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1101         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1102         btrfs_set_stack_chunk_type(chunk, type);
1103         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1104         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1105         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1106         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1107         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1108         map->sector_size = extent_root->sectorsize;
1109         map->stripe_len = stripe_len;
1110         map->io_align = stripe_len;
1111         map->io_width = stripe_len;
1112         map->type = type;
1113         map->num_stripes = num_stripes;
1114         map->sub_stripes = sub_stripes;
1115
1116         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1117                                 btrfs_chunk_item_size(num_stripes));
1118         BUG_ON(ret);
1119         *start = key.offset;
1120
1121         map->ce.start = key.offset;
1122         map->ce.size = num_bytes;
1123
1124         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1125         BUG_ON(ret);
1126
1127         kfree(chunk);
1128         return ret;
1129 }
1130
1131 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1132 {
1133         struct cache_extent *ce;
1134         struct map_lookup *map;
1135         int ret;
1136
1137         ce = search_cache_extent(&map_tree->cache_tree, logical);
1138         if (!ce) {
1139                 fprintf(stderr, "No mapping for %llu-%llu\n",
1140                         (unsigned long long)logical,
1141                         (unsigned long long)logical+len);
1142                 return 1;
1143         }
1144         if (ce->start > logical || ce->start + ce->size < logical) {
1145                 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1146                         "%llu-%llu\n", (unsigned long long)logical,
1147                         (unsigned long long)logical+len,
1148                         (unsigned long long)ce->start,
1149                         (unsigned long long)ce->start + ce->size);
1150                 return 1;
1151         }
1152         map = container_of(ce, struct map_lookup, ce);
1153
1154         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1155                 ret = map->num_stripes;
1156         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1157                 ret = map->sub_stripes;
1158         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1159                 ret = 2;
1160         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1161                 ret = 3;
1162         else
1163                 ret = 1;
1164         return ret;
1165 }
1166
1167 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1168                         u64 *size)
1169 {
1170         struct cache_extent *ce;
1171         struct map_lookup *map;
1172
1173         ce = search_cache_extent(&map_tree->cache_tree, *logical);
1174
1175         while (ce) {
1176                 ce = next_cache_extent(ce);
1177                 if (!ce)
1178                         return -ENOENT;
1179
1180                 map = container_of(ce, struct map_lookup, ce);
1181                 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1182                         *logical = ce->start;
1183                         *size = ce->size;
1184                         return 0;
1185                 }
1186         }
1187
1188         return -ENOENT;
1189 }
1190
1191 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1192                      u64 chunk_start, u64 physical, u64 devid,
1193                      u64 **logical, int *naddrs, int *stripe_len)
1194 {
1195         struct cache_extent *ce;
1196         struct map_lookup *map;
1197         u64 *buf;
1198         u64 bytenr;
1199         u64 length;
1200         u64 stripe_nr;
1201         u64 rmap_len;
1202         int i, j, nr = 0;
1203
1204         ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1205         BUG_ON(!ce);
1206         map = container_of(ce, struct map_lookup, ce);
1207
1208         length = ce->size;
1209         rmap_len = map->stripe_len;
1210         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1211                 length = ce->size / (map->num_stripes / map->sub_stripes);
1212         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1213                 length = ce->size / map->num_stripes;
1214         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1215                               BTRFS_BLOCK_GROUP_RAID6)) {
1216                 length = ce->size / nr_data_stripes(map);
1217                 rmap_len = map->stripe_len * nr_data_stripes(map);
1218         }
1219
1220         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1221
1222         for (i = 0; i < map->num_stripes; i++) {
1223                 if (devid && map->stripes[i].dev->devid != devid)
1224                         continue;
1225                 if (map->stripes[i].physical > physical ||
1226                     map->stripes[i].physical + length <= physical)
1227                         continue;
1228
1229                 stripe_nr = (physical - map->stripes[i].physical) /
1230                             map->stripe_len;
1231
1232                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1233                         stripe_nr = (stripe_nr * map->num_stripes + i) /
1234                                     map->sub_stripes;
1235                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1236                         stripe_nr = stripe_nr * map->num_stripes + i;
1237                 } /* else if RAID[56], multiply by nr_data_stripes().
1238                    * Alternatively, just use rmap_len below instead of
1239                    * map->stripe_len */
1240
1241                 bytenr = ce->start + stripe_nr * rmap_len;
1242                 for (j = 0; j < nr; j++) {
1243                         if (buf[j] == bytenr)
1244                                 break;
1245                 }
1246                 if (j == nr)
1247                         buf[nr++] = bytenr;
1248         }
1249
1250         *logical = buf;
1251         *naddrs = nr;
1252         *stripe_len = rmap_len;
1253
1254         return 0;
1255 }
1256
1257 static inline int parity_smaller(u64 a, u64 b)
1258 {
1259         return a > b;
1260 }
1261
1262 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1263 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1264 {
1265         struct btrfs_bio_stripe s;
1266         int i;
1267         u64 l;
1268         int again = 1;
1269
1270         while (again) {
1271                 again = 0;
1272                 for (i = 0; i < bbio->num_stripes - 1; i++) {
1273                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
1274                                 s = bbio->stripes[i];
1275                                 l = raid_map[i];
1276                                 bbio->stripes[i] = bbio->stripes[i+1];
1277                                 raid_map[i] = raid_map[i+1];
1278                                 bbio->stripes[i+1] = s;
1279                                 raid_map[i+1] = l;
1280                                 again = 1;
1281                         }
1282                 }
1283         }
1284 }
1285
1286 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1287                     u64 logical, u64 *length,
1288                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1289                     u64 **raid_map_ret)
1290 {
1291         return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1292                                  multi_ret, mirror_num, raid_map_ret);
1293 }
1294
1295 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1296                     u64 logical, u64 *length, u64 *type,
1297                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1298                     u64 **raid_map_ret)
1299 {
1300         struct cache_extent *ce;
1301         struct map_lookup *map;
1302         u64 offset;
1303         u64 stripe_offset;
1304         u64 stripe_nr;
1305         u64 *raid_map = NULL;
1306         int stripes_allocated = 8;
1307         int stripes_required = 1;
1308         int stripe_index;
1309         int i;
1310         struct btrfs_multi_bio *multi = NULL;
1311
1312         if (multi_ret && rw == READ) {
1313                 stripes_allocated = 1;
1314         }
1315 again:
1316         ce = search_cache_extent(&map_tree->cache_tree, logical);
1317         if (!ce) {
1318                 kfree(multi);
1319                 *length = (u64)-1;
1320                 return -ENOENT;
1321         }
1322         if (ce->start > logical) {
1323                 kfree(multi);
1324                 *length = ce->start - logical;
1325                 return -ENOENT;
1326         }
1327
1328         if (multi_ret) {
1329                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1330                                 GFP_NOFS);
1331                 if (!multi)
1332                         return -ENOMEM;
1333         }
1334         map = container_of(ce, struct map_lookup, ce);
1335         offset = logical - ce->start;
1336
1337         if (rw == WRITE) {
1338                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1339                                  BTRFS_BLOCK_GROUP_DUP)) {
1340                         stripes_required = map->num_stripes;
1341                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1342                         stripes_required = map->sub_stripes;
1343                 }
1344         }
1345         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1346             && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1347                     /* RAID[56] write or recovery. Return all stripes */
1348                     stripes_required = map->num_stripes;
1349
1350                     /* Only allocate the map if we've already got a large enough multi_ret */
1351                     if (stripes_allocated >= stripes_required) {
1352                             raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1353                             if (!raid_map) {
1354                                     kfree(multi);
1355                                     return -ENOMEM;
1356                             }
1357                     }
1358         }
1359
1360         /* if our multi bio struct is too small, back off and try again */
1361         if (multi_ret && stripes_allocated < stripes_required) {
1362                 stripes_allocated = stripes_required;
1363                 kfree(multi);
1364                 multi = NULL;
1365                 goto again;
1366         }
1367         stripe_nr = offset;
1368         /*
1369          * stripe_nr counts the total number of stripes we have to stride
1370          * to get to this block
1371          */
1372         stripe_nr = stripe_nr / map->stripe_len;
1373
1374         stripe_offset = stripe_nr * map->stripe_len;
1375         BUG_ON(offset < stripe_offset);
1376
1377         /* stripe_offset is the offset of this block in its stripe*/
1378         stripe_offset = offset - stripe_offset;
1379
1380         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1381                          BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1382                          BTRFS_BLOCK_GROUP_RAID10 |
1383                          BTRFS_BLOCK_GROUP_DUP)) {
1384                 /* we limit the length of each bio to what fits in a stripe */
1385                 *length = min_t(u64, ce->size - offset,
1386                               map->stripe_len - stripe_offset);
1387         } else {
1388                 *length = ce->size - offset;
1389         }
1390
1391         if (!multi_ret)
1392                 goto out;
1393
1394         multi->num_stripes = 1;
1395         stripe_index = 0;
1396         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1397                 if (rw == WRITE)
1398                         multi->num_stripes = map->num_stripes;
1399                 else if (mirror_num)
1400                         stripe_index = mirror_num - 1;
1401                 else
1402                         stripe_index = stripe_nr % map->num_stripes;
1403         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1404                 int factor = map->num_stripes / map->sub_stripes;
1405
1406                 stripe_index = stripe_nr % factor;
1407                 stripe_index *= map->sub_stripes;
1408
1409                 if (rw == WRITE)
1410                         multi->num_stripes = map->sub_stripes;
1411                 else if (mirror_num)
1412                         stripe_index += mirror_num - 1;
1413
1414                 stripe_nr = stripe_nr / factor;
1415         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1416                 if (rw == WRITE)
1417                         multi->num_stripes = map->num_stripes;
1418                 else if (mirror_num)
1419                         stripe_index = mirror_num - 1;
1420         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1421                                 BTRFS_BLOCK_GROUP_RAID6)) {
1422
1423                 if (raid_map) {
1424                         int rot;
1425                         u64 tmp;
1426                         u64 raid56_full_stripe_start;
1427                         u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1428
1429                         /*
1430                          * align the start of our data stripe in the logical
1431                          * address space
1432                          */
1433                         raid56_full_stripe_start = offset / full_stripe_len;
1434                         raid56_full_stripe_start *= full_stripe_len;
1435
1436                         /* get the data stripe number */
1437                         stripe_nr = raid56_full_stripe_start / map->stripe_len;
1438                         stripe_nr = stripe_nr / nr_data_stripes(map);
1439
1440                         /* Work out the disk rotation on this stripe-set */
1441                         rot = stripe_nr % map->num_stripes;
1442
1443                         /* Fill in the logical address of each stripe */
1444                         tmp = stripe_nr * nr_data_stripes(map);
1445
1446                         for (i = 0; i < nr_data_stripes(map); i++)
1447                                 raid_map[(i+rot) % map->num_stripes] =
1448                                         ce->start + (tmp + i) * map->stripe_len;
1449
1450                         raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1451                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1452                                 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1453
1454                         *length = map->stripe_len;
1455                         stripe_index = 0;
1456                         stripe_offset = 0;
1457                         multi->num_stripes = map->num_stripes;
1458                 } else {
1459                         stripe_index = stripe_nr % nr_data_stripes(map);
1460                         stripe_nr = stripe_nr / nr_data_stripes(map);
1461
1462                         /*
1463                          * Mirror #0 or #1 means the original data block.
1464                          * Mirror #2 is RAID5 parity block.
1465                          * Mirror #3 is RAID6 Q block.
1466                          */
1467                         if (mirror_num > 1)
1468                                 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1469
1470                         /* We distribute the parity blocks across stripes */
1471                         stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1472                 }
1473         } else {
1474                 /*
1475                  * after this do_div call, stripe_nr is the number of stripes
1476                  * on this device we have to walk to find the data, and
1477                  * stripe_index is the number of our device in the stripe array
1478                  */
1479                 stripe_index = stripe_nr % map->num_stripes;
1480                 stripe_nr = stripe_nr / map->num_stripes;
1481         }
1482         BUG_ON(stripe_index >= map->num_stripes);
1483
1484         for (i = 0; i < multi->num_stripes; i++) {
1485                 multi->stripes[i].physical =
1486                         map->stripes[stripe_index].physical + stripe_offset +
1487                         stripe_nr * map->stripe_len;
1488                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1489                 stripe_index++;
1490         }
1491         *multi_ret = multi;
1492
1493         if (type)
1494                 *type = map->type;
1495
1496         if (raid_map) {
1497                 sort_parity_stripes(multi, raid_map);
1498                 *raid_map_ret = raid_map;
1499         }
1500 out:
1501         return 0;
1502 }
1503
1504 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1505                                        u8 *uuid, u8 *fsid)
1506 {
1507         struct btrfs_device *device;
1508         struct btrfs_fs_devices *cur_devices;
1509
1510         cur_devices = root->fs_info->fs_devices;
1511         while (cur_devices) {
1512                 if (!fsid ||
1513                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1514                         device = __find_device(&cur_devices->devices,
1515                                                devid, uuid);
1516                         if (device)
1517                                 return device;
1518                 }
1519                 cur_devices = cur_devices->seed;
1520         }
1521         return NULL;
1522 }
1523
1524 struct btrfs_device *
1525 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1526                            u64 devid, int instance)
1527 {
1528         struct list_head *head = &fs_devices->devices;
1529         struct btrfs_device *dev;
1530         int num_found = 0;
1531
1532         list_for_each_entry(dev, head, dev_list) {
1533                 if (dev->devid == devid && num_found++ == instance)
1534                         return dev;
1535         }
1536         return NULL;
1537 }
1538
1539 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1540 {
1541         struct cache_extent *ce;
1542         struct map_lookup *map;
1543         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1544         int readonly = 0;
1545         int i;
1546
1547         /*
1548          * During chunk recovering, we may fail to find block group's
1549          * corresponding chunk, we will rebuild it later
1550          */
1551         ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1552         if (!root->fs_info->is_chunk_recover)
1553                 BUG_ON(!ce);
1554         else
1555                 return 0;
1556
1557         map = container_of(ce, struct map_lookup, ce);
1558         for (i = 0; i < map->num_stripes; i++) {
1559                 if (!map->stripes[i].dev->writeable) {
1560                         readonly = 1;
1561                         break;
1562                 }
1563         }
1564
1565         return readonly;
1566 }
1567
1568 static struct btrfs_device *fill_missing_device(u64 devid)
1569 {
1570         struct btrfs_device *device;
1571
1572         device = kzalloc(sizeof(*device), GFP_NOFS);
1573         device->devid = devid;
1574         device->fd = -1;
1575         return device;
1576 }
1577
1578 /*
1579  * Slot is used to verfy the chunk item is valid
1580  *
1581  * For sys chunk in superblock, pass -1 to indicate sys chunk.
1582  */
1583 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1584                           struct extent_buffer *leaf,
1585                           struct btrfs_chunk *chunk, int slot)
1586 {
1587         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1588         struct map_lookup *map;
1589         struct cache_extent *ce;
1590         u64 logical;
1591         u64 length;
1592         u64 devid;
1593         u8 uuid[BTRFS_UUID_SIZE];
1594         int num_stripes;
1595         int ret;
1596         int i;
1597
1598         logical = key->offset;
1599         length = btrfs_chunk_length(leaf, chunk);
1600
1601         ce = search_cache_extent(&map_tree->cache_tree, logical);
1602
1603         /* already mapped? */
1604         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1605                 return 0;
1606         }
1607
1608         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1609         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1610         if (!map)
1611                 return -ENOMEM;
1612
1613         map->ce.start = logical;
1614         map->ce.size = length;
1615         map->num_stripes = num_stripes;
1616         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1617         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1618         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1619         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1620         map->type = btrfs_chunk_type(leaf, chunk);
1621         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1622
1623         /* Check on chunk item type */
1624         if (map->type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1625                           BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1626                 fprintf(stderr, "Unknown chunk type bits: %llu\n",
1627                         map->type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1628                                       BTRFS_BLOCK_GROUP_PROFILE_MASK));
1629                 ret = -EIO;
1630                 goto out;
1631         }
1632
1633         /*
1634          * Btrfs_chunk contains at least one stripe, and for sys_chunk
1635          * it can't exceed the system chunk array size
1636          * For normal chunk, it should match its chunk item size.
1637          */
1638         if (num_stripes < 1 ||
1639             (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1640              BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1641             (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1642              btrfs_item_size_nr(leaf, slot))) {
1643                 fprintf(stderr, "Invalid num_stripes: %u\n",
1644                         num_stripes);
1645                 ret = -EIO;
1646                 goto out;
1647         }
1648
1649         /*
1650          * Device number check against profile
1651          */
1652         if ((map->type & BTRFS_BLOCK_GROUP_RAID10 && map->sub_stripes == 0) ||
1653             (map->type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1654             (map->type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1655             (map->type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1656             (map->type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1657             ((map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1658              num_stripes != 1)) {
1659                 fprintf(stderr,
1660                         "Invalid num_stripes:sub_stripes %u:%u for profile %llu\n",
1661                         num_stripes, map->sub_stripes,
1662                         map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1663                 ret = -EIO;
1664                 goto out;
1665         }
1666
1667         for (i = 0; i < num_stripes; i++) {
1668                 map->stripes[i].physical =
1669                         btrfs_stripe_offset_nr(leaf, chunk, i);
1670                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1671                 read_extent_buffer(leaf, uuid, (unsigned long)
1672                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1673                                    BTRFS_UUID_SIZE);
1674                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1675                                                         NULL);
1676                 if (!map->stripes[i].dev) {
1677                         map->stripes[i].dev = fill_missing_device(devid);
1678                         printf("warning, device %llu is missing\n",
1679                                (unsigned long long)devid);
1680                 }
1681
1682         }
1683         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1684         BUG_ON(ret);
1685
1686         return 0;
1687 out:
1688         free(map);
1689         return ret;
1690 }
1691
1692 static int fill_device_from_item(struct extent_buffer *leaf,
1693                                  struct btrfs_dev_item *dev_item,
1694                                  struct btrfs_device *device)
1695 {
1696         unsigned long ptr;
1697
1698         device->devid = btrfs_device_id(leaf, dev_item);
1699         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1700         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1701         device->type = btrfs_device_type(leaf, dev_item);
1702         device->io_align = btrfs_device_io_align(leaf, dev_item);
1703         device->io_width = btrfs_device_io_width(leaf, dev_item);
1704         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1705
1706         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1707         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1708
1709         return 0;
1710 }
1711
1712 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1713 {
1714         struct btrfs_fs_devices *fs_devices;
1715         int ret;
1716
1717         fs_devices = root->fs_info->fs_devices->seed;
1718         while (fs_devices) {
1719                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1720                         ret = 0;
1721                         goto out;
1722                 }
1723                 fs_devices = fs_devices->seed;
1724         }
1725
1726         fs_devices = find_fsid(fsid);
1727         if (!fs_devices) {
1728                 /* missing all seed devices */
1729                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1730                 if (!fs_devices) {
1731                         ret = -ENOMEM;
1732                         goto out;
1733                 }
1734                 INIT_LIST_HEAD(&fs_devices->devices);
1735                 list_add(&fs_devices->list, &fs_uuids);
1736                 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1737         }
1738
1739         ret = btrfs_open_devices(fs_devices, O_RDONLY);
1740         if (ret)
1741                 goto out;
1742
1743         fs_devices->seed = root->fs_info->fs_devices->seed;
1744         root->fs_info->fs_devices->seed = fs_devices;
1745 out:
1746         return ret;
1747 }
1748
1749 static int read_one_dev(struct btrfs_root *root,
1750                         struct extent_buffer *leaf,
1751                         struct btrfs_dev_item *dev_item)
1752 {
1753         struct btrfs_device *device;
1754         u64 devid;
1755         int ret = 0;
1756         u8 fs_uuid[BTRFS_UUID_SIZE];
1757         u8 dev_uuid[BTRFS_UUID_SIZE];
1758
1759         devid = btrfs_device_id(leaf, dev_item);
1760         read_extent_buffer(leaf, dev_uuid,
1761                            (unsigned long)btrfs_device_uuid(dev_item),
1762                            BTRFS_UUID_SIZE);
1763         read_extent_buffer(leaf, fs_uuid,
1764                            (unsigned long)btrfs_device_fsid(dev_item),
1765                            BTRFS_UUID_SIZE);
1766
1767         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1768                 ret = open_seed_devices(root, fs_uuid);
1769                 if (ret)
1770                         return ret;
1771         }
1772
1773         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1774         if (!device) {
1775                 printk("warning devid %llu not found already\n",
1776                         (unsigned long long)devid);
1777                 device = kzalloc(sizeof(*device), GFP_NOFS);
1778                 if (!device)
1779                         return -ENOMEM;
1780                 device->fd = -1;
1781                 list_add(&device->dev_list,
1782                          &root->fs_info->fs_devices->devices);
1783         }
1784
1785         fill_device_from_item(leaf, dev_item, device);
1786         device->dev_root = root->fs_info->dev_root;
1787         return ret;
1788 }
1789
1790 int btrfs_read_sys_array(struct btrfs_root *root)
1791 {
1792         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1793         struct extent_buffer *sb;
1794         struct btrfs_disk_key *disk_key;
1795         struct btrfs_chunk *chunk;
1796         struct btrfs_key key;
1797         u32 num_stripes;
1798         u32 len = 0;
1799         u8 *ptr;
1800         u8 *array_end;
1801         int ret = 0;
1802
1803         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1804                                           BTRFS_SUPER_INFO_SIZE);
1805         if (!sb)
1806                 return -ENOMEM;
1807         btrfs_set_buffer_uptodate(sb);
1808         write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1809         array_end = ((u8 *)super_copy->sys_chunk_array) +
1810                     btrfs_super_sys_array_size(super_copy);
1811
1812         /*
1813          * we do this loop twice, once for the device items and
1814          * once for all of the chunks.  This way there are device
1815          * structs filled in for every chunk
1816          */
1817         ptr = super_copy->sys_chunk_array;
1818
1819         while (ptr < array_end) {
1820                 disk_key = (struct btrfs_disk_key *)ptr;
1821                 btrfs_disk_key_to_cpu(&key, disk_key);
1822
1823                 len = sizeof(*disk_key);
1824                 ptr += len;
1825
1826                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1827                         chunk = (struct btrfs_chunk *)(ptr - (u8 *)super_copy);
1828                         ret = read_one_chunk(root, &key, sb, chunk, -1);
1829                         if (ret)
1830                                 break;
1831                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1832                         len = btrfs_chunk_item_size(num_stripes);
1833                 } else {
1834                         BUG();
1835                 }
1836                 ptr += len;
1837         }
1838         free_extent_buffer(sb);
1839         return ret;
1840 }
1841
1842 int btrfs_read_chunk_tree(struct btrfs_root *root)
1843 {
1844         struct btrfs_path *path;
1845         struct extent_buffer *leaf;
1846         struct btrfs_key key;
1847         struct btrfs_key found_key;
1848         int ret;
1849         int slot;
1850
1851         root = root->fs_info->chunk_root;
1852
1853         path = btrfs_alloc_path();
1854         if (!path)
1855                 return -ENOMEM;
1856
1857         /*
1858          * Read all device items, and then all the chunk items. All
1859          * device items are found before any chunk item (their object id
1860          * is smaller than the lowest possible object id for a chunk
1861          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1862          */
1863         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1864         key.offset = 0;
1865         key.type = 0;
1866         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1867         if (ret < 0)
1868                 goto error;
1869         while(1) {
1870                 leaf = path->nodes[0];
1871                 slot = path->slots[0];
1872                 if (slot >= btrfs_header_nritems(leaf)) {
1873                         ret = btrfs_next_leaf(root, path);
1874                         if (ret == 0)
1875                                 continue;
1876                         if (ret < 0)
1877                                 goto error;
1878                         break;
1879                 }
1880                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1881                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1882                         struct btrfs_dev_item *dev_item;
1883                         dev_item = btrfs_item_ptr(leaf, slot,
1884                                                   struct btrfs_dev_item);
1885                         ret = read_one_dev(root, leaf, dev_item);
1886                         BUG_ON(ret);
1887                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1888                         struct btrfs_chunk *chunk;
1889                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1890                         ret = read_one_chunk(root, &found_key, leaf, chunk,
1891                                              slot);
1892                         BUG_ON(ret);
1893                 }
1894                 path->slots[0]++;
1895         }
1896
1897         ret = 0;
1898 error:
1899         btrfs_free_path(path);
1900         return ret;
1901 }
1902
1903 struct list_head *btrfs_scanned_uuids(void)
1904 {
1905         return &fs_uuids;
1906 }
1907
1908 static int rmw_eb(struct btrfs_fs_info *info,
1909                   struct extent_buffer *eb, struct extent_buffer *orig_eb)
1910 {
1911         int ret;
1912         unsigned long orig_off = 0;
1913         unsigned long dest_off = 0;
1914         unsigned long copy_len = eb->len;
1915
1916         ret = read_whole_eb(info, eb, 0);
1917         if (ret)
1918                 return ret;
1919
1920         if (eb->start + eb->len <= orig_eb->start ||
1921             eb->start >= orig_eb->start + orig_eb->len)
1922                 return 0;
1923         /*
1924          * | ----- orig_eb ------- |
1925          *         | ----- stripe -------  |
1926          *         | ----- orig_eb ------- |
1927          *              | ----- orig_eb ------- |
1928          */
1929         if (eb->start > orig_eb->start)
1930                 orig_off = eb->start - orig_eb->start;
1931         if (orig_eb->start > eb->start)
1932                 dest_off = orig_eb->start - eb->start;
1933
1934         if (copy_len > orig_eb->len - orig_off)
1935                 copy_len = orig_eb->len - orig_off;
1936         if (copy_len > eb->len - dest_off)
1937                 copy_len = eb->len - dest_off;
1938
1939         memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1940         return 0;
1941 }
1942
1943 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1944                                 struct extent_buffer *orig_eb,
1945                                struct extent_buffer **ebs,
1946                                u64 stripe_len, u64 *raid_map,
1947                                int num_stripes)
1948 {
1949         struct extent_buffer *eb;
1950         u64 start = orig_eb->start;
1951         u64 this_eb_start;
1952         int i;
1953         int ret;
1954
1955         for (i = 0; i < num_stripes; i++) {
1956                 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1957                         break;
1958
1959                 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1960                 if (!eb)
1961                         BUG();
1962                 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1963
1964                 eb->start = raid_map[i];
1965                 eb->len = stripe_len;
1966                 eb->refs = 1;
1967                 eb->flags = 0;
1968                 eb->fd = -1;
1969                 eb->dev_bytenr = (u64)-1;
1970
1971                 this_eb_start = raid_map[i];
1972
1973                 if (start > this_eb_start ||
1974                     start + orig_eb->len < this_eb_start + stripe_len) {
1975                         ret = rmw_eb(info, eb, orig_eb);
1976                         BUG_ON(ret);
1977                 } else {
1978                         memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1979                 }
1980                 ebs[i] = eb;
1981         }
1982 }
1983
1984 int write_raid56_with_parity(struct btrfs_fs_info *info,
1985                              struct extent_buffer *eb,
1986                              struct btrfs_multi_bio *multi,
1987                              u64 stripe_len, u64 *raid_map)
1988 {
1989         struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
1990         int i;
1991         int j;
1992         int ret;
1993         int alloc_size = eb->len;
1994
1995         ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
1996         BUG_ON(!ebs);
1997
1998         if (stripe_len > alloc_size)
1999                 alloc_size = stripe_len;
2000
2001         split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2002                             multi->num_stripes);
2003
2004         for (i = 0; i < multi->num_stripes; i++) {
2005                 struct extent_buffer *new_eb;
2006                 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2007                         ebs[i]->dev_bytenr = multi->stripes[i].physical;
2008                         ebs[i]->fd = multi->stripes[i].dev->fd;
2009                         multi->stripes[i].dev->total_ios++;
2010                         BUG_ON(ebs[i]->start != raid_map[i]);
2011                         continue;
2012                 }
2013                 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
2014                 BUG_ON(!new_eb);
2015                 new_eb->dev_bytenr = multi->stripes[i].physical;
2016                 new_eb->fd = multi->stripes[i].dev->fd;
2017                 multi->stripes[i].dev->total_ios++;
2018                 new_eb->len = stripe_len;
2019
2020                 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2021                         p_eb = new_eb;
2022                 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2023                         q_eb = new_eb;
2024         }
2025         if (q_eb) {
2026                 void **pointers;
2027
2028                 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
2029                                    GFP_NOFS);
2030                 BUG_ON(!pointers);
2031
2032                 ebs[multi->num_stripes - 2] = p_eb;
2033                 ebs[multi->num_stripes - 1] = q_eb;
2034
2035                 for (i = 0; i < multi->num_stripes; i++)
2036                         pointers[i] = ebs[i]->data;
2037
2038                 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2039                 kfree(pointers);
2040         } else {
2041                 ebs[multi->num_stripes - 1] = p_eb;
2042                 memcpy(p_eb->data, ebs[0]->data, stripe_len);
2043                 for (j = 1; j < multi->num_stripes - 1; j++) {
2044                         for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
2045                                 *(unsigned long *)(p_eb->data + i) ^=
2046                                         *(unsigned long *)(ebs[j]->data + i);
2047                         }
2048                 }
2049         }
2050
2051         for (i = 0; i < multi->num_stripes; i++) {
2052                 ret = write_extent_to_disk(ebs[i]);
2053                 BUG_ON(ret);
2054                 if (ebs[i] != eb)
2055                         kfree(ebs[i]);
2056         }
2057
2058         kfree(ebs);
2059
2060         return 0;
2061 }