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 void btrfs_close_all_devices(void)
 202 {
 203         struct btrfs_fs_devices *fs_devices;
 204
 205         while (!list_empty(&fs_uuids)) {
 206                 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
 207                                         list);
 208                 btrfs_close_devices(fs_devices);
 209         }
 210 }
 211
 212 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
 213 {
 214         int fd;
 215         struct list_head *head = &fs_devices->devices;
 216         struct list_head *cur;
 217         struct btrfs_device *device;
 218         int ret;
 219
 220         list_for_each(cur, head) {
 221                 device = list_entry(cur, struct btrfs_device, dev_list);
 222                 if (!device->name) {
 223                         printk("no name for device %llu, skip it now\n", device->devid);
 224                         continue;
 225                 }
 226
 227                 fd = open(device->name, flags);
 228                 if (fd < 0) {
 229                         ret = -errno;
 230                         goto fail;
 231                 }
 232
 233                 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
 234                         fprintf(stderr, "Warning, could not drop caches\n");
 235
 236                 if (device->devid == fs_devices->latest_devid)
 237                         fs_devices->latest_bdev = fd;
 238                 if (device->devid == fs_devices->lowest_devid)
 239                         fs_devices->lowest_bdev = fd;
 240                 device->fd = fd;
 241                 if (flags & O_RDWR)
 242                         device->writeable = 1;
 243         }
 244         return 0;
 245 fail:
 246         btrfs_close_devices(fs_devices);
 247         return ret;
 248 }
 249
 250 int btrfs_scan_one_device(int fd, const char *path,
 251                           struct btrfs_fs_devices **fs_devices_ret,
 252                           u64 *total_devs, u64 super_offset, int super_recover)
 253 {
 254         struct btrfs_super_block *disk_super;
 255         char buf[BTRFS_SUPER_INFO_SIZE];
 256         int ret;
 257         u64 devid;
 258
 259         disk_super = (struct btrfs_super_block *)buf;
 260         ret = btrfs_read_dev_super(fd, disk_super, super_offset, super_recover);
 261         if (ret < 0)
 262                 return -EIO;
 263         devid = btrfs_stack_device_id(&disk_super->dev_item);
 264         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
 265                 *total_devs = 1;
 266         else
 267                 *total_devs = btrfs_super_num_devices(disk_super);
 268
 269         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 270
 271         return ret;
 272 }
 273
 274 /*
 275  * this uses a pretty simple search, the expectation is that it is
 276  * called very infrequently and that a given device has a small number
 277  * of extents
 278  */
 279 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 280                                 struct btrfs_device *device,
 281                                 struct btrfs_path *path,
 282                                 u64 num_bytes, u64 *start)
 283 {
 284         struct btrfs_key key;
 285         struct btrfs_root *root = device->dev_root;
 286         struct btrfs_dev_extent *dev_extent = NULL;
 287         u64 hole_size = 0;
 288         u64 last_byte = 0;
 289         u64 search_start = root->fs_info->alloc_start;
 290         u64 search_end = device->total_bytes;
 291         int ret;
 292         int slot = 0;
 293         int start_found;
 294         struct extent_buffer *l;
 295
 296         start_found = 0;
 297         path->reada = 2;
 298
 299         /* FIXME use last free of some kind */
 300
 301         /* we don't want to overwrite the superblock on the drive,
 302          * so we make sure to start at an offset of at least 1MB
 303          */
 304         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 305
 306         if (search_start >= search_end) {
 307                 ret = -ENOSPC;
 308                 goto error;
 309         }
 310
 311         key.objectid = device->devid;
 312         key.offset = search_start;
 313         key.type = BTRFS_DEV_EXTENT_KEY;
 314         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 315         if (ret < 0)
 316                 goto error;
 317         ret = btrfs_previous_item(root, path, 0, key.type);
 318         if (ret < 0)
 319                 goto error;
 320         l = path->nodes[0];
 321         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 322         while (1) {
 323                 l = path->nodes[0];
 324                 slot = path->slots[0];
 325                 if (slot >= btrfs_header_nritems(l)) {
 326                         ret = btrfs_next_leaf(root, path);
 327                         if (ret == 0)
 328                                 continue;
 329                         if (ret < 0)
 330                                 goto error;
 331 no_more_items:
 332                         if (!start_found) {
 333                                 if (search_start >= search_end) {
 334                                         ret = -ENOSPC;
 335                                         goto error;
 336                                 }
 337                                 *start = search_start;
 338                                 start_found = 1;
 339                                 goto check_pending;
 340                         }
 341                         *start = last_byte > search_start ?
 342                                 last_byte : search_start;
 343                         if (search_end <= *start) {
 344                                 ret = -ENOSPC;
 345                                 goto error;
 346                         }
 347                         goto check_pending;
 348                 }
 349                 btrfs_item_key_to_cpu(l, &key, slot);
 350
 351                 if (key.objectid < device->devid)
 352                         goto next;
 353
 354                 if (key.objectid > device->devid)
 355                         goto no_more_items;
 356
 357                 if (key.offset >= search_start && key.offset > last_byte &&
 358                     start_found) {
 359                         if (last_byte < search_start)
 360                                 last_byte = search_start;
 361                         hole_size = key.offset - last_byte;
 362                         if (key.offset > last_byte &&
 363                             hole_size >= num_bytes) {
 364                                 *start = last_byte;
 365                                 goto check_pending;
 366                         }
 367                 }
 368                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 369                         goto next;
 370                 }
 371
 372                 start_found = 1;
 373                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 374                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 375 next:
 376                 path->slots[0]++;
 377                 cond_resched();
 378         }
 379 check_pending:
 380         /* we have to make sure we didn't find an extent that has already
 381          * been allocated by the map tree or the original allocation
 382          */
 383         btrfs_release_path(path);
 384         BUG_ON(*start < search_start);
 385
 386         if (*start + num_bytes > search_end) {
 387                 ret = -ENOSPC;
 388                 goto error;
 389         }
 390         /* check for pending inserts here */
 391         return 0;
 392
 393 error:
 394         btrfs_release_path(path);
 395         return ret;
 396 }
 397
 398 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 399                                   struct btrfs_device *device,
 400                                   u64 chunk_tree, u64 chunk_objectid,
 401                                   u64 chunk_offset,
 402                                   u64 num_bytes, u64 *start)
 403 {
 404         int ret;
 405         struct btrfs_path *path;
 406         struct btrfs_root *root = device->dev_root;
 407         struct btrfs_dev_extent *extent;
 408         struct extent_buffer *leaf;
 409         struct btrfs_key key;
 410
 411         path = btrfs_alloc_path();
 412         if (!path)
 413                 return -ENOMEM;
 414
 415         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 416         if (ret) {
 417                 goto err;
 418         }
 419
 420         key.objectid = device->devid;
 421         key.offset = *start;
 422         key.type = BTRFS_DEV_EXTENT_KEY;
 423         ret = btrfs_insert_empty_item(trans, root, path, &key,
 424                                       sizeof(*extent));
 425         BUG_ON(ret);
 426
 427         leaf = path->nodes[0];
 428         extent = btrfs_item_ptr(leaf, path->slots[0],
 429                                 struct btrfs_dev_extent);
 430         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 431         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 432         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 433
 434         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 435                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 436                     BTRFS_UUID_SIZE);
 437
 438         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 439         btrfs_mark_buffer_dirty(leaf);
 440 err:
 441         btrfs_free_path(path);
 442         return ret;
 443 }
 444
 445 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 446 {
 447         struct btrfs_path *path;
 448         int ret;
 449         struct btrfs_key key;
 450         struct btrfs_chunk *chunk;
 451         struct btrfs_key found_key;
 452
 453         path = btrfs_alloc_path();
 454         BUG_ON(!path);
 455
 456         key.objectid = objectid;
 457         key.offset = (u64)-1;
 458         key.type = BTRFS_CHUNK_ITEM_KEY;
 459
 460         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 461         if (ret < 0)
 462                 goto error;
 463
 464         BUG_ON(ret == 0);
 465
 466         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 467         if (ret) {
 468                 *offset = 0;
 469         } else {
 470                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 471                                       path->slots[0]);
 472                 if (found_key.objectid != objectid)
 473                         *offset = 0;
 474                 else {
 475                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 476                                                struct btrfs_chunk);
 477                         *offset = found_key.offset +
 478                                 btrfs_chunk_length(path->nodes[0], chunk);
 479                 }
 480         }
 481         ret = 0;
 482 error:
 483         btrfs_free_path(path);
 484         return ret;
 485 }
 486
 487 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 488                            u64 *objectid)
 489 {
 490         int ret;
 491         struct btrfs_key key;
 492         struct btrfs_key found_key;
 493
 494         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 495         key.type = BTRFS_DEV_ITEM_KEY;
 496         key.offset = (u64)-1;
 497
 498         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 499         if (ret < 0)
 500                 goto error;
 501
 502         BUG_ON(ret == 0);
 503
 504         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 505                                   BTRFS_DEV_ITEM_KEY);
 506         if (ret) {
 507                 *objectid = 1;
 508         } else {
 509                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 510                                       path->slots[0]);
 511                 *objectid = found_key.offset + 1;
 512         }
 513         ret = 0;
 514 error:
 515         btrfs_release_path(path);
 516         return ret;
 517 }
 518
 519 /*
 520  * the device information is stored in the chunk root
 521  * the btrfs_device struct should be fully filled in
 522  */
 523 int btrfs_add_device(struct btrfs_trans_handle *trans,
 524                      struct btrfs_root *root,
 525                      struct btrfs_device *device)
 526 {
 527         int ret;
 528         struct btrfs_path *path;
 529         struct btrfs_dev_item *dev_item;
 530         struct extent_buffer *leaf;
 531         struct btrfs_key key;
 532         unsigned long ptr;
 533         u64 free_devid = 0;
 534
 535         root = root->fs_info->chunk_root;
 536
 537         path = btrfs_alloc_path();
 538         if (!path)
 539                 return -ENOMEM;
 540
 541         ret = find_next_devid(root, path, &free_devid);
 542         if (ret)
 543                 goto out;
 544
 545         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 546         key.type = BTRFS_DEV_ITEM_KEY;
 547         key.offset = free_devid;
 548
 549         ret = btrfs_insert_empty_item(trans, root, path, &key,
 550                                       sizeof(*dev_item));
 551         if (ret)
 552                 goto out;
 553
 554         leaf = path->nodes[0];
 555         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 556
 557         device->devid = free_devid;
 558         btrfs_set_device_id(leaf, dev_item, device->devid);
 559         btrfs_set_device_generation(leaf, dev_item, 0);
 560         btrfs_set_device_type(leaf, dev_item, device->type);
 561         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 562         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 563         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 564         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 565         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 566         btrfs_set_device_group(leaf, dev_item, 0);
 567         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 568         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 569         btrfs_set_device_start_offset(leaf, dev_item, 0);
 570
 571         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 572         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 573         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 574         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 575         btrfs_mark_buffer_dirty(leaf);
 576         ret = 0;
 577
 578 out:
 579         btrfs_free_path(path);
 580         return ret;
 581 }
 582
 583 int btrfs_update_device(struct btrfs_trans_handle *trans,
 584                         struct btrfs_device *device)
 585 {
 586         int ret;
 587         struct btrfs_path *path;
 588         struct btrfs_root *root;
 589         struct btrfs_dev_item *dev_item;
 590         struct extent_buffer *leaf;
 591         struct btrfs_key key;
 592
 593         root = device->dev_root->fs_info->chunk_root;
 594
 595         path = btrfs_alloc_path();
 596         if (!path)
 597                 return -ENOMEM;
 598
 599         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 600         key.type = BTRFS_DEV_ITEM_KEY;
 601         key.offset = device->devid;
 602
 603         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 604         if (ret < 0)
 605                 goto out;
 606
 607         if (ret > 0) {
 608                 ret = -ENOENT;
 609                 goto out;
 610         }
 611
 612         leaf = path->nodes[0];
 613         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 614
 615         btrfs_set_device_id(leaf, dev_item, device->devid);
 616         btrfs_set_device_type(leaf, dev_item, device->type);
 617         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 618         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 619         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 620         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 621         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 622         btrfs_mark_buffer_dirty(leaf);
 623
 624 out:
 625         btrfs_free_path(path);
 626         return ret;
 627 }
 628
 629 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 630                            struct btrfs_root *root,
 631                            struct btrfs_key *key,
 632                            struct btrfs_chunk *chunk, int item_size)
 633 {
 634         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
 635         struct btrfs_disk_key disk_key;
 636         u32 array_size;
 637         u8 *ptr;
 638
 639         array_size = btrfs_super_sys_array_size(super_copy);
 640         if (array_size + item_size + sizeof(disk_key)
 641                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 642                 return -EFBIG;
 643
 644         ptr = super_copy->sys_chunk_array + array_size;
 645         btrfs_cpu_key_to_disk(&disk_key, key);
 646         memcpy(ptr, &disk_key, sizeof(disk_key));
 647         ptr += sizeof(disk_key);
 648         memcpy(ptr, chunk, item_size);
 649         item_size += sizeof(disk_key);
 650         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 651         return 0;
 652 }
 653
 654 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 655                                int sub_stripes)
 656 {
 657         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 658                 return calc_size;
 659         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 660                 return calc_size * (num_stripes / sub_stripes);
 661         else if (type & BTRFS_BLOCK_GROUP_RAID5)
 662                 return calc_size * (num_stripes - 1);
 663         else if (type & BTRFS_BLOCK_GROUP_RAID6)
 664                 return calc_size * (num_stripes - 2);
 665         else
 666                 return calc_size * num_stripes;
 667 }
 668
 669
 670 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
 671 {
 672         /* TODO, add a way to store the preferred stripe size */
 673         return BTRFS_STRIPE_LEN;
 674 }
 675
 676 /*
 677  * btrfs_device_avail_bytes - count bytes available for alloc_chunk
 678  *
 679  * It is not equal to "device->total_bytes - device->bytes_used".
 680  * We do not allocate any chunk in 1M at beginning of device, and not
 681  * allowed to allocate any chunk before alloc_start if it is specified.
 682  * So search holes from max(1M, alloc_start) to device->total_bytes.
 683  */
 684 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
 685                                     struct btrfs_device *device,
 686                                     u64 *avail_bytes)
 687 {
 688         struct btrfs_path *path;
 689         struct btrfs_root *root = device->dev_root;
 690         struct btrfs_key key;
 691         struct btrfs_dev_extent *dev_extent = NULL;
 692         struct extent_buffer *l;
 693         u64 search_start = root->fs_info->alloc_start;
 694         u64 search_end = device->total_bytes;
 695         u64 extent_end = 0;
 696         u64 free_bytes = 0;
 697         int ret;
 698         int slot = 0;
 699
 700         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 701
 702         path = btrfs_alloc_path();
 703         if (!path)
 704                 return -ENOMEM;
 705
 706         key.objectid = device->devid;
 707         key.offset = root->fs_info->alloc_start;
 708         key.type = BTRFS_DEV_EXTENT_KEY;
 709
 710         path->reada = 2;
 711         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 712         if (ret < 0)
 713                 goto error;
 714         ret = btrfs_previous_item(root, path, 0, key.type);
 715         if (ret < 0)
 716                 goto error;
 717
 718         while (1) {
 719                 l = path->nodes[0];
 720                 slot = path->slots[0];
 721                 if (slot >= btrfs_header_nritems(l)) {
 722                         ret = btrfs_next_leaf(root, path);
 723                         if (ret == 0)
 724                                 continue;
 725                         if (ret < 0)
 726                                 goto error;
 727                         break;
 728                 }
 729                 btrfs_item_key_to_cpu(l, &key, slot);
 730
 731                 if (key.objectid < device->devid)
 732                         goto next;
 733                 if (key.objectid > device->devid)
 734                         break;
 735                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 736                         goto next;
 737                 if (key.offset > search_end)
 738                         break;
 739                 if (key.offset > search_start)
 740                         free_bytes += key.offset - search_start;
 741
 742                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 743                 extent_end = key.offset + btrfs_dev_extent_length(l,
 744                                                                   dev_extent);
 745                 if (extent_end > search_start)
 746                         search_start = extent_end;
 747                 if (search_start > search_end)
 748                         break;
 749 next:
 750                 path->slots[0]++;
 751                 cond_resched();
 752         }
 753
 754         if (search_start < search_end)
 755                 free_bytes += search_end - search_start;
 756
 757         *avail_bytes = free_bytes;
 758         ret = 0;
 759 error:
 760         btrfs_free_path(path);
 761         return ret;
 762 }
 763
 764 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
 765                         - sizeof(struct btrfs_item)             \
 766                         - sizeof(struct btrfs_chunk))           \
 767                         / sizeof(struct btrfs_stripe) + 1)
 768
 769 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
 770                                 - 2 * sizeof(struct btrfs_disk_key)     \
 771                                 - 2 * sizeof(struct btrfs_chunk))       \
 772                                 / sizeof(struct btrfs_stripe) + 1)
 773
 774 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 775                       struct btrfs_root *extent_root, u64 *start,
 776                       u64 *num_bytes, u64 type)
 777 {
 778         u64 dev_offset;
 779         struct btrfs_fs_info *info = extent_root->fs_info;
 780         struct btrfs_root *chunk_root = info->chunk_root;
 781         struct btrfs_stripe *stripes;
 782         struct btrfs_device *device = NULL;
 783         struct btrfs_chunk *chunk;
 784         struct list_head private_devs;
 785         struct list_head *dev_list = &info->fs_devices->devices;
 786         struct list_head *cur;
 787         struct map_lookup *map;
 788         int min_stripe_size = 1 * 1024 * 1024;
 789         u64 calc_size = 8 * 1024 * 1024;
 790         u64 min_free;
 791         u64 max_chunk_size = 4 * calc_size;
 792         u64 avail = 0;
 793         u64 max_avail = 0;
 794         u64 percent_max;
 795         int num_stripes = 1;
 796         int max_stripes = 0;
 797         int min_stripes = 1;
 798         int sub_stripes = 0;
 799         int looped = 0;
 800         int ret;
 801         int index;
 802         int stripe_len = BTRFS_STRIPE_LEN;
 803         struct btrfs_key key;
 804         u64 offset;
 805
 806         if (list_empty(dev_list)) {
 807                 return -ENOSPC;
 808         }
 809
 810         if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
 811                     BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
 812                     BTRFS_BLOCK_GROUP_RAID10 |
 813                     BTRFS_BLOCK_GROUP_DUP)) {
 814                 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 815                         calc_size = 8 * 1024 * 1024;
 816                         max_chunk_size = calc_size * 2;
 817                         min_stripe_size = 1 * 1024 * 1024;
 818                         max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
 819                 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
 820                         calc_size = 1024 * 1024 * 1024;
 821                         max_chunk_size = 10 * calc_size;
 822                         min_stripe_size = 64 * 1024 * 1024;
 823                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 824                 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 825                         calc_size = 1024 * 1024 * 1024;
 826                         max_chunk_size = 4 * calc_size;
 827                         min_stripe_size = 32 * 1024 * 1024;
 828                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 829                 }
 830         }
 831         if (type & BTRFS_BLOCK_GROUP_RAID1) {
 832                 num_stripes = min_t(u64, 2,
 833                                   btrfs_super_num_devices(info->super_copy));
 834                 if (num_stripes < 2)
 835                         return -ENOSPC;
 836                 min_stripes = 2;
 837         }
 838         if (type & BTRFS_BLOCK_GROUP_DUP) {
 839                 num_stripes = 2;
 840                 min_stripes = 2;
 841         }
 842         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
 843                 num_stripes = btrfs_super_num_devices(info->super_copy);
 844                 if (num_stripes > max_stripes)
 845                         num_stripes = max_stripes;
 846                 min_stripes = 2;
 847         }
 848         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 849                 num_stripes = btrfs_super_num_devices(info->super_copy);
 850                 if (num_stripes > max_stripes)
 851                         num_stripes = max_stripes;
 852                 if (num_stripes < 4)
 853                         return -ENOSPC;
 854                 num_stripes &= ~(u32)1;
 855                 sub_stripes = 2;
 856                 min_stripes = 4;
 857         }
 858         if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
 859                 num_stripes = btrfs_super_num_devices(info->super_copy);
 860                 if (num_stripes > max_stripes)
 861                         num_stripes = max_stripes;
 862                 if (num_stripes < 2)
 863                         return -ENOSPC;
 864                 min_stripes = 2;
 865                 stripe_len = find_raid56_stripe_len(num_stripes - 1,
 866                                     btrfs_super_stripesize(info->super_copy));
 867         }
 868         if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
 869                 num_stripes = btrfs_super_num_devices(info->super_copy);
 870                 if (num_stripes > max_stripes)
 871                         num_stripes = max_stripes;
 872                 if (num_stripes < 3)
 873                         return -ENOSPC;
 874                 min_stripes = 3;
 875                 stripe_len = find_raid56_stripe_len(num_stripes - 2,
 876                                     btrfs_super_stripesize(info->super_copy));
 877         }
 878
 879         /* we don't want a chunk larger than 10% of the FS */
 880         percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
 881         max_chunk_size = min(percent_max, max_chunk_size);
 882
 883 again:
 884         if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
 885             max_chunk_size) {
 886                 calc_size = max_chunk_size;
 887                 calc_size /= num_stripes;
 888                 calc_size /= stripe_len;
 889                 calc_size *= stripe_len;
 890         }
 891         /* we don't want tiny stripes */
 892         calc_size = max_t(u64, calc_size, min_stripe_size);
 893
 894         calc_size /= stripe_len;
 895         calc_size *= stripe_len;
 896         INIT_LIST_HEAD(&private_devs);
 897         cur = dev_list->next;
 898         index = 0;
 899
 900         if (type & BTRFS_BLOCK_GROUP_DUP)
 901                 min_free = calc_size * 2;
 902         else
 903                 min_free = calc_size;
 904
 905         /* build a private list of devices we will allocate from */
 906         while(index < num_stripes) {
 907                 device = list_entry(cur, struct btrfs_device, dev_list);
 908                 ret = btrfs_device_avail_bytes(trans, device, &avail);
 909                 if (ret)
 910                         return ret;
 911                 cur = cur->next;
 912                 if (avail >= min_free) {
 913                         list_move_tail(&device->dev_list, &private_devs);
 914                         index++;
 915                         if (type & BTRFS_BLOCK_GROUP_DUP)
 916                                 index++;
 917                 } else if (avail > max_avail)
 918                         max_avail = avail;
 919                 if (cur == dev_list)
 920                         break;
 921         }
 922         if (index < num_stripes) {
 923                 list_splice(&private_devs, dev_list);
 924                 if (index >= min_stripes) {
 925                         num_stripes = index;
 926                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 927                                 num_stripes /= sub_stripes;
 928                                 num_stripes *= sub_stripes;
 929                         }
 930                         looped = 1;
 931                         goto again;
 932                 }
 933                 if (!looped && max_avail > 0) {
 934                         looped = 1;
 935                         calc_size = max_avail;
 936                         goto again;
 937                 }
 938                 return -ENOSPC;
 939         }
 940         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 941                               &offset);
 942         if (ret)
 943                 return ret;
 944         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 945         key.type = BTRFS_CHUNK_ITEM_KEY;
 946         key.offset = offset;
 947
 948         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 949         if (!chunk)
 950                 return -ENOMEM;
 951
 952         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
 953         if (!map) {
 954                 kfree(chunk);
 955                 return -ENOMEM;
 956         }
 957
 958         stripes = &chunk->stripe;
 959         *num_bytes = chunk_bytes_by_type(type, calc_size,
 960                                          num_stripes, sub_stripes);
 961         index = 0;
 962         while(index < num_stripes) {
 963                 struct btrfs_stripe *stripe;
 964                 BUG_ON(list_empty(&private_devs));
 965                 cur = private_devs.next;
 966                 device = list_entry(cur, struct btrfs_device, dev_list);
 967
 968                 /* loop over this device again if we're doing a dup group */
 969                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 970                     (index == num_stripes - 1))
 971                         list_move_tail(&device->dev_list, dev_list);
 972
 973                 ret = btrfs_alloc_dev_extent(trans, device,
 974                              info->chunk_root->root_key.objectid,
 975                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 976                              calc_size, &dev_offset);
 977                 BUG_ON(ret);
 978
 979                 device->bytes_used += calc_size;
 980                 ret = btrfs_update_device(trans, device);
 981                 BUG_ON(ret);
 982
 983                 map->stripes[index].dev = device;
 984                 map->stripes[index].physical = dev_offset;
 985                 stripe = stripes + index;
 986                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 987                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 988                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 989                 index++;
 990         }
 991         BUG_ON(!list_empty(&private_devs));
 992
 993         /* key was set above */
 994         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 995         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 996         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 997         btrfs_set_stack_chunk_type(chunk, type);
 998         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 999         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1000         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1001         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1002         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1003         map->sector_size = extent_root->sectorsize;
1004         map->stripe_len = stripe_len;
1005         map->io_align = stripe_len;
1006         map->io_width = stripe_len;
1007         map->type = type;
1008         map->num_stripes = num_stripes;
1009         map->sub_stripes = sub_stripes;
1010
1011         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1012                                 btrfs_chunk_item_size(num_stripes));
1013         BUG_ON(ret);
1014         *start = key.offset;;
1015
1016         map->ce.start = key.offset;
1017         map->ce.size = *num_bytes;
1018
1019         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1020         BUG_ON(ret);
1021
1022         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1023                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1024                                     chunk, btrfs_chunk_item_size(num_stripes));
1025                 BUG_ON(ret);
1026         }
1027
1028         kfree(chunk);
1029         return ret;
1030 }
1031
1032 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1033                            struct btrfs_root *extent_root, u64 *start,
1034                            u64 num_bytes, u64 type)
1035 {
1036         u64 dev_offset;
1037         struct btrfs_fs_info *info = extent_root->fs_info;
1038         struct btrfs_root *chunk_root = info->chunk_root;
1039         struct btrfs_stripe *stripes;
1040         struct btrfs_device *device = NULL;
1041         struct btrfs_chunk *chunk;
1042         struct list_head *dev_list = &info->fs_devices->devices;
1043         struct list_head *cur;
1044         struct map_lookup *map;
1045         u64 calc_size = 8 * 1024 * 1024;
1046         int num_stripes = 1;
1047         int sub_stripes = 0;
1048         int ret;
1049         int index;
1050         int stripe_len = BTRFS_STRIPE_LEN;
1051         struct btrfs_key key;
1052
1053         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1054         key.type = BTRFS_CHUNK_ITEM_KEY;
1055         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1056                               &key.offset);
1057         if (ret)
1058                 return ret;
1059
1060         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1061         if (!chunk)
1062                 return -ENOMEM;
1063
1064         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1065         if (!map) {
1066                 kfree(chunk);
1067                 return -ENOMEM;
1068         }
1069
1070         stripes = &chunk->stripe;
1071         calc_size = num_bytes;
1072
1073         index = 0;
1074         cur = dev_list->next;
1075         device = list_entry(cur, struct btrfs_device, dev_list);
1076
1077         while (index < num_stripes) {
1078                 struct btrfs_stripe *stripe;
1079
1080                 ret = btrfs_alloc_dev_extent(trans, device,
1081                              info->chunk_root->root_key.objectid,
1082                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1083                              calc_size, &dev_offset);
1084                 BUG_ON(ret);
1085
1086                 device->bytes_used += calc_size;
1087                 ret = btrfs_update_device(trans, device);
1088                 BUG_ON(ret);
1089
1090                 map->stripes[index].dev = device;
1091                 map->stripes[index].physical = dev_offset;
1092                 stripe = stripes + index;
1093                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1094                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1095                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1096                 index++;
1097         }
1098
1099         /* key was set above */
1100         btrfs_set_stack_chunk_length(chunk, num_bytes);
1101         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1102         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1103         btrfs_set_stack_chunk_type(chunk, type);
1104         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1105         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1106         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1107         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1108         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1109         map->sector_size = extent_root->sectorsize;
1110         map->stripe_len = stripe_len;
1111         map->io_align = stripe_len;
1112         map->io_width = stripe_len;
1113         map->type = type;
1114         map->num_stripes = num_stripes;
1115         map->sub_stripes = sub_stripes;
1116
1117         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1118                                 btrfs_chunk_item_size(num_stripes));
1119         BUG_ON(ret);
1120         *start = key.offset;
1121
1122         map->ce.start = key.offset;
1123         map->ce.size = num_bytes;
1124
1125         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1126         BUG_ON(ret);
1127
1128         kfree(chunk);
1129         return ret;
1130 }
1131
1132 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1133 {
1134         struct cache_extent *ce;
1135         struct map_lookup *map;
1136         int ret;
1137
1138         ce = search_cache_extent(&map_tree->cache_tree, logical);
1139         if (!ce) {
1140                 fprintf(stderr, "No mapping for %llu-%llu\n",
1141                         (unsigned long long)logical,
1142                         (unsigned long long)logical+len);
1143                 return 1;
1144         }
1145         if (ce->start > logical || ce->start + ce->size < logical) {
1146                 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1147                         "%llu-%llu\n", (unsigned long long)logical,
1148                         (unsigned long long)logical+len,
1149                         (unsigned long long)ce->start,
1150                         (unsigned long long)ce->start + ce->size);
1151                 return 1;
1152         }
1153         map = container_of(ce, struct map_lookup, ce);
1154
1155         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1156                 ret = map->num_stripes;
1157         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1158                 ret = map->sub_stripes;
1159         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1160                 ret = 2;
1161         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1162                 ret = 3;
1163         else
1164                 ret = 1;
1165         return ret;
1166 }
1167
1168 int btrfs_next_bg(struct btrfs_mapping_tree *map_tree, u64 *logical,
1169                      u64 *size, u64 type)
1170 {
1171         struct cache_extent *ce;
1172         struct map_lookup *map;
1173
1174         ce = search_cache_extent(&map_tree->cache_tree, *logical);
1175
1176         while (ce) {
1177                 ce = next_cache_extent(ce);
1178                 if (!ce)
1179                         return -ENOENT;
1180
1181                 map = container_of(ce, struct map_lookup, ce);
1182                 if (map->type & type) {
1183                         *logical = ce->start;
1184                         *size = ce->size;
1185                         return 0;
1186                 }
1187         }
1188
1189         return -ENOENT;
1190 }
1191
1192 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1193                      u64 chunk_start, u64 physical, u64 devid,
1194                      u64 **logical, int *naddrs, int *stripe_len)
1195 {
1196         struct cache_extent *ce;
1197         struct map_lookup *map;
1198         u64 *buf;
1199         u64 bytenr;
1200         u64 length;
1201         u64 stripe_nr;
1202         u64 rmap_len;
1203         int i, j, nr = 0;
1204
1205         ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1206         BUG_ON(!ce);
1207         map = container_of(ce, struct map_lookup, ce);
1208
1209         length = ce->size;
1210         rmap_len = map->stripe_len;
1211         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1212                 length = ce->size / (map->num_stripes / map->sub_stripes);
1213         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1214                 length = ce->size / map->num_stripes;
1215         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1216                               BTRFS_BLOCK_GROUP_RAID6)) {
1217                 length = ce->size / nr_data_stripes(map);
1218                 rmap_len = map->stripe_len * nr_data_stripes(map);
1219         }
1220
1221         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1222
1223         for (i = 0; i < map->num_stripes; i++) {
1224                 if (devid && map->stripes[i].dev->devid != devid)
1225                         continue;
1226                 if (map->stripes[i].physical > physical ||
1227                     map->stripes[i].physical + length <= physical)
1228                         continue;
1229
1230                 stripe_nr = (physical - map->stripes[i].physical) /
1231                             map->stripe_len;
1232
1233                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1234                         stripe_nr = (stripe_nr * map->num_stripes + i) /
1235                                     map->sub_stripes;
1236                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1237                         stripe_nr = stripe_nr * map->num_stripes + i;
1238                 } /* else if RAID[56], multiply by nr_data_stripes().
1239                    * Alternatively, just use rmap_len below instead of
1240                    * map->stripe_len */
1241
1242                 bytenr = ce->start + stripe_nr * rmap_len;
1243                 for (j = 0; j < nr; j++) {
1244                         if (buf[j] == bytenr)
1245                                 break;
1246                 }
1247                 if (j == nr)
1248                         buf[nr++] = bytenr;
1249         }
1250
1251         *logical = buf;
1252         *naddrs = nr;
1253         *stripe_len = rmap_len;
1254
1255         return 0;
1256 }
1257
1258 static inline int parity_smaller(u64 a, u64 b)
1259 {
1260         return a > b;
1261 }
1262
1263 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1264 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1265 {
1266         struct btrfs_bio_stripe s;
1267         int i;
1268         u64 l;
1269         int again = 1;
1270
1271         while (again) {
1272                 again = 0;
1273                 for (i = 0; i < bbio->num_stripes - 1; i++) {
1274                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
1275                                 s = bbio->stripes[i];
1276                                 l = raid_map[i];
1277                                 bbio->stripes[i] = bbio->stripes[i+1];
1278                                 raid_map[i] = raid_map[i+1];
1279                                 bbio->stripes[i+1] = s;
1280                                 raid_map[i+1] = l;
1281                                 again = 1;
1282                         }
1283                 }
1284         }
1285 }
1286
1287 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1288                     u64 logical, u64 *length,
1289                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1290                     u64 **raid_map_ret)
1291 {
1292         return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1293                                  multi_ret, mirror_num, raid_map_ret);
1294 }
1295
1296 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1297                     u64 logical, u64 *length, u64 *type,
1298                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1299                     u64 **raid_map_ret)
1300 {
1301         struct cache_extent *ce;
1302         struct map_lookup *map;
1303         u64 offset;
1304         u64 stripe_offset;
1305         u64 stripe_nr;
1306         u64 *raid_map = NULL;
1307         int stripes_allocated = 8;
1308         int stripes_required = 1;
1309         int stripe_index;
1310         int i;
1311         struct btrfs_multi_bio *multi = NULL;
1312
1313         if (multi_ret && rw == READ) {
1314                 stripes_allocated = 1;
1315         }
1316 again:
1317         ce = search_cache_extent(&map_tree->cache_tree, logical);
1318         if (!ce) {
1319                 kfree(multi);
1320                 *length = (u64)-1;
1321                 return -ENOENT;
1322         }
1323         if (ce->start > logical) {
1324                 kfree(multi);
1325                 *length = ce->start - logical;
1326                 return -ENOENT;
1327         }
1328
1329         if (multi_ret) {
1330                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1331                                 GFP_NOFS);
1332                 if (!multi)
1333                         return -ENOMEM;
1334         }
1335         map = container_of(ce, struct map_lookup, ce);
1336         offset = logical - ce->start;
1337
1338         if (rw == WRITE) {
1339                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1340                                  BTRFS_BLOCK_GROUP_DUP)) {
1341                         stripes_required = map->num_stripes;
1342                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1343                         stripes_required = map->sub_stripes;
1344                 }
1345         }
1346         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1347             && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1348                     /* RAID[56] write or recovery. Return all stripes */
1349                     stripes_required = map->num_stripes;
1350
1351                     /* Only allocate the map if we've already got a large enough multi_ret */
1352                     if (stripes_allocated >= stripes_required) {
1353                             raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1354                             if (!raid_map) {
1355                                     kfree(multi);
1356                                     return -ENOMEM;
1357                             }
1358                     }
1359         }
1360
1361         /* if our multi bio struct is too small, back off and try again */
1362         if (multi_ret && stripes_allocated < stripes_required) {
1363                 stripes_allocated = stripes_required;
1364                 kfree(multi);
1365                 multi = NULL;
1366                 goto again;
1367         }
1368         stripe_nr = offset;
1369         /*
1370          * stripe_nr counts the total number of stripes we have to stride
1371          * to get to this block
1372          */
1373         stripe_nr = stripe_nr / map->stripe_len;
1374
1375         stripe_offset = stripe_nr * map->stripe_len;
1376         BUG_ON(offset < stripe_offset);
1377
1378         /* stripe_offset is the offset of this block in its stripe*/
1379         stripe_offset = offset - stripe_offset;
1380
1381         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1382                          BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1383                          BTRFS_BLOCK_GROUP_RAID10 |
1384                          BTRFS_BLOCK_GROUP_DUP)) {
1385                 /* we limit the length of each bio to what fits in a stripe */
1386                 *length = min_t(u64, ce->size - offset,
1387                               map->stripe_len - stripe_offset);
1388         } else {
1389                 *length = ce->size - offset;
1390         }
1391
1392         if (!multi_ret)
1393                 goto out;
1394
1395         multi->num_stripes = 1;
1396         stripe_index = 0;
1397         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1398                 if (rw == WRITE)
1399                         multi->num_stripes = map->num_stripes;
1400                 else if (mirror_num)
1401                         stripe_index = mirror_num - 1;
1402                 else
1403                         stripe_index = stripe_nr % map->num_stripes;
1404         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1405                 int factor = map->num_stripes / map->sub_stripes;
1406
1407                 stripe_index = stripe_nr % factor;
1408                 stripe_index *= map->sub_stripes;
1409
1410                 if (rw == WRITE)
1411                         multi->num_stripes = map->sub_stripes;
1412                 else if (mirror_num)
1413                         stripe_index += mirror_num - 1;
1414
1415                 stripe_nr = stripe_nr / factor;
1416         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1417                 if (rw == WRITE)
1418                         multi->num_stripes = map->num_stripes;
1419                 else if (mirror_num)
1420                         stripe_index = mirror_num - 1;
1421         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1422                                 BTRFS_BLOCK_GROUP_RAID6)) {
1423
1424                 if (raid_map) {
1425                         int rot;
1426                         u64 tmp;
1427                         u64 raid56_full_stripe_start;
1428                         u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1429
1430                         /*
1431                          * align the start of our data stripe in the logical
1432                          * address space
1433                          */
1434                         raid56_full_stripe_start = offset / full_stripe_len;
1435                         raid56_full_stripe_start *= full_stripe_len;
1436
1437                         /* get the data stripe number */
1438                         stripe_nr = raid56_full_stripe_start / map->stripe_len;
1439                         stripe_nr = stripe_nr / nr_data_stripes(map);
1440
1441                         /* Work out the disk rotation on this stripe-set */
1442                         rot = stripe_nr % map->num_stripes;
1443
1444                         /* Fill in the logical address of each stripe */
1445                         tmp = stripe_nr * nr_data_stripes(map);
1446
1447                         for (i = 0; i < nr_data_stripes(map); i++)
1448                                 raid_map[(i+rot) % map->num_stripes] =
1449                                         ce->start + (tmp + i) * map->stripe_len;
1450
1451                         raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1452                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1453                                 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1454
1455                         *length = map->stripe_len;
1456                         stripe_index = 0;
1457                         stripe_offset = 0;
1458                         multi->num_stripes = map->num_stripes;
1459                 } else {
1460                         stripe_index = stripe_nr % nr_data_stripes(map);
1461                         stripe_nr = stripe_nr / nr_data_stripes(map);
1462
1463                         /*
1464                          * Mirror #0 or #1 means the original data block.
1465                          * Mirror #2 is RAID5 parity block.
1466                          * Mirror #3 is RAID6 Q block.
1467                          */
1468                         if (mirror_num > 1)
1469                                 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1470
1471                         /* We distribute the parity blocks across stripes */
1472                         stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1473                 }
1474         } else {
1475                 /*
1476                  * after this do_div call, stripe_nr is the number of stripes
1477                  * on this device we have to walk to find the data, and
1478                  * stripe_index is the number of our device in the stripe array
1479                  */
1480                 stripe_index = stripe_nr % map->num_stripes;
1481                 stripe_nr = stripe_nr / map->num_stripes;
1482         }
1483         BUG_ON(stripe_index >= map->num_stripes);
1484
1485         for (i = 0; i < multi->num_stripes; i++) {
1486                 multi->stripes[i].physical =
1487                         map->stripes[stripe_index].physical + stripe_offset +
1488                         stripe_nr * map->stripe_len;
1489                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1490                 stripe_index++;
1491         }
1492         *multi_ret = multi;
1493
1494         if (type)
1495                 *type = map->type;
1496
1497         if (raid_map) {
1498                 sort_parity_stripes(multi, raid_map);
1499                 *raid_map_ret = raid_map;
1500         }
1501 out:
1502         return 0;
1503 }
1504
1505 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1506                                        u8 *uuid, u8 *fsid)
1507 {
1508         struct btrfs_device *device;
1509         struct btrfs_fs_devices *cur_devices;
1510
1511         cur_devices = root->fs_info->fs_devices;
1512         while (cur_devices) {
1513                 if (!fsid ||
1514                     (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1515                      root->fs_info->ignore_fsid_mismatch)) {
1516                         device = __find_device(&cur_devices->devices,
1517                                                devid, uuid);
1518                         if (device)
1519                                 return device;
1520                 }
1521                 cur_devices = cur_devices->seed;
1522         }
1523         return NULL;
1524 }
1525
1526 struct btrfs_device *
1527 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1528                            u64 devid, int instance)
1529 {
1530         struct list_head *head = &fs_devices->devices;
1531         struct btrfs_device *dev;
1532         int num_found = 0;
1533
1534         list_for_each_entry(dev, head, dev_list) {
1535                 if (dev->devid == devid && num_found++ == instance)
1536                         return dev;
1537         }
1538         return NULL;
1539 }
1540
1541 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1542 {
1543         struct cache_extent *ce;
1544         struct map_lookup *map;
1545         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1546         int readonly = 0;
1547         int i;
1548
1549         /*
1550          * During chunk recovering, we may fail to find block group's
1551          * corresponding chunk, we will rebuild it later
1552          */
1553         ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1554         if (!root->fs_info->is_chunk_recover)
1555                 BUG_ON(!ce);
1556         else
1557                 return 0;
1558
1559         map = container_of(ce, struct map_lookup, ce);
1560         for (i = 0; i < map->num_stripes; i++) {
1561                 if (!map->stripes[i].dev->writeable) {
1562                         readonly = 1;
1563                         break;
1564                 }
1565         }
1566
1567         return readonly;
1568 }
1569
1570 static struct btrfs_device *fill_missing_device(u64 devid)
1571 {
1572         struct btrfs_device *device;
1573
1574         device = kzalloc(sizeof(*device), GFP_NOFS);
1575         device->devid = devid;
1576         device->fd = -1;
1577         return device;
1578 }
1579
1580 /*
1581  * Slot is used to verfy the chunk item is valid
1582  *
1583  * For sys chunk in superblock, pass -1 to indicate sys chunk.
1584  */
1585 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1586                           struct extent_buffer *leaf,
1587                           struct btrfs_chunk *chunk, int slot)
1588 {
1589         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1590         struct map_lookup *map;
1591         struct cache_extent *ce;
1592         u64 logical;
1593         u64 length;
1594         u64 devid;
1595         u8 uuid[BTRFS_UUID_SIZE];
1596         int num_stripes;
1597         int ret;
1598         int i;
1599
1600         logical = key->offset;
1601         length = btrfs_chunk_length(leaf, chunk);
1602
1603         ce = search_cache_extent(&map_tree->cache_tree, logical);
1604
1605         /* already mapped? */
1606         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1607                 return 0;
1608         }
1609
1610         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1611         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1612         if (!map)
1613                 return -ENOMEM;
1614
1615         map->ce.start = logical;
1616         map->ce.size = length;
1617         map->num_stripes = num_stripes;
1618         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1619         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1620         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1621         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1622         map->type = btrfs_chunk_type(leaf, chunk);
1623         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1624
1625         /* Check on chunk item type */
1626         if (map->type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1627                           BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1628                 fprintf(stderr, "Unknown chunk type bits: %llu\n",
1629                         map->type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1630                                       BTRFS_BLOCK_GROUP_PROFILE_MASK));
1631                 ret = -EIO;
1632                 goto out;
1633         }
1634
1635         /*
1636          * Btrfs_chunk contains at least one stripe, and for sys_chunk
1637          * it can't exceed the system chunk array size
1638          * For normal chunk, it should match its chunk item size.
1639          */
1640         if (num_stripes < 1 ||
1641             (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1642              BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1643             (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1644              btrfs_item_size_nr(leaf, slot))) {
1645                 fprintf(stderr, "Invalid num_stripes: %u\n",
1646                         num_stripes);
1647                 ret = -EIO;
1648                 goto out;
1649         }
1650
1651         /*
1652          * Device number check against profile
1653          */
1654         if ((map->type & BTRFS_BLOCK_GROUP_RAID10 && map->sub_stripes == 0) ||
1655             (map->type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1656             (map->type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1657             (map->type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1658             (map->type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1659             ((map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1660              num_stripes != 1)) {
1661                 fprintf(stderr,
1662                         "Invalid num_stripes:sub_stripes %u:%u for profile %llu\n",
1663                         num_stripes, map->sub_stripes,
1664                         map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1665                 ret = -EIO;
1666                 goto out;
1667         }
1668
1669         for (i = 0; i < num_stripes; i++) {
1670                 map->stripes[i].physical =
1671                         btrfs_stripe_offset_nr(leaf, chunk, i);
1672                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1673                 read_extent_buffer(leaf, uuid, (unsigned long)
1674                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1675                                    BTRFS_UUID_SIZE);
1676                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1677                                                         NULL);
1678                 if (!map->stripes[i].dev) {
1679                         map->stripes[i].dev = fill_missing_device(devid);
1680                         printf("warning, device %llu is missing\n",
1681                                (unsigned long long)devid);
1682                 }
1683
1684         }
1685         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1686         BUG_ON(ret);
1687
1688         return 0;
1689 out:
1690         free(map);
1691         return ret;
1692 }
1693
1694 static int fill_device_from_item(struct extent_buffer *leaf,
1695                                  struct btrfs_dev_item *dev_item,
1696                                  struct btrfs_device *device)
1697 {
1698         unsigned long ptr;
1699
1700         device->devid = btrfs_device_id(leaf, dev_item);
1701         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1702         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1703         device->type = btrfs_device_type(leaf, dev_item);
1704         device->io_align = btrfs_device_io_align(leaf, dev_item);
1705         device->io_width = btrfs_device_io_width(leaf, dev_item);
1706         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1707
1708         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1709         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1710
1711         return 0;
1712 }
1713
1714 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1715 {
1716         struct btrfs_fs_devices *fs_devices;
1717         int ret;
1718
1719         fs_devices = root->fs_info->fs_devices->seed;
1720         while (fs_devices) {
1721                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1722                         ret = 0;
1723                         goto out;
1724                 }
1725                 fs_devices = fs_devices->seed;
1726         }
1727
1728         fs_devices = find_fsid(fsid);
1729         if (!fs_devices) {
1730                 /* missing all seed devices */
1731                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1732                 if (!fs_devices) {
1733                         ret = -ENOMEM;
1734                         goto out;
1735                 }
1736                 INIT_LIST_HEAD(&fs_devices->devices);
1737                 list_add(&fs_devices->list, &fs_uuids);
1738                 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1739         }
1740
1741         ret = btrfs_open_devices(fs_devices, O_RDONLY);
1742         if (ret)
1743                 goto out;
1744
1745         fs_devices->seed = root->fs_info->fs_devices->seed;
1746         root->fs_info->fs_devices->seed = fs_devices;
1747 out:
1748         return ret;
1749 }
1750
1751 static int read_one_dev(struct btrfs_root *root,
1752                         struct extent_buffer *leaf,
1753                         struct btrfs_dev_item *dev_item)
1754 {
1755         struct btrfs_device *device;
1756         u64 devid;
1757         int ret = 0;
1758         u8 fs_uuid[BTRFS_UUID_SIZE];
1759         u8 dev_uuid[BTRFS_UUID_SIZE];
1760
1761         devid = btrfs_device_id(leaf, dev_item);
1762         read_extent_buffer(leaf, dev_uuid,
1763                            (unsigned long)btrfs_device_uuid(dev_item),
1764                            BTRFS_UUID_SIZE);
1765         read_extent_buffer(leaf, fs_uuid,
1766                            (unsigned long)btrfs_device_fsid(dev_item),
1767                            BTRFS_UUID_SIZE);
1768
1769         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1770                 ret = open_seed_devices(root, fs_uuid);
1771                 if (ret)
1772                         return ret;
1773         }
1774
1775         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1776         if (!device) {
1777                 printk("warning devid %llu not found already\n",
1778                         (unsigned long long)devid);
1779                 device = kzalloc(sizeof(*device), GFP_NOFS);
1780                 if (!device)
1781                         return -ENOMEM;
1782                 device->fd = -1;
1783                 list_add(&device->dev_list,
1784                          &root->fs_info->fs_devices->devices);
1785         }
1786
1787         fill_device_from_item(leaf, dev_item, device);
1788         device->dev_root = root->fs_info->dev_root;
1789         return ret;
1790 }
1791
1792 int btrfs_read_sys_array(struct btrfs_root *root)
1793 {
1794         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1795         struct extent_buffer *sb;
1796         struct btrfs_disk_key *disk_key;
1797         struct btrfs_chunk *chunk;
1798         u8 *array_ptr;
1799         unsigned long sb_array_offset;
1800         int ret = 0;
1801         u32 num_stripes;
1802         u32 array_size;
1803         u32 len = 0;
1804         u32 cur_offset;
1805         struct btrfs_key key;
1806
1807         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1808                                           BTRFS_SUPER_INFO_SIZE);
1809         if (!sb)
1810                 return -ENOMEM;
1811         btrfs_set_buffer_uptodate(sb);
1812         write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1813         array_size = btrfs_super_sys_array_size(super_copy);
1814
1815         array_ptr = super_copy->sys_chunk_array;
1816         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1817         cur_offset = 0;
1818
1819         while (cur_offset < array_size) {
1820                 disk_key = (struct btrfs_disk_key *)array_ptr;
1821                 len = sizeof(*disk_key);
1822                 if (cur_offset + len > array_size)
1823                         goto out_short_read;
1824
1825                 btrfs_disk_key_to_cpu(&key, disk_key);
1826
1827                 array_ptr += len;
1828                 sb_array_offset += len;
1829                 cur_offset += len;
1830
1831                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1832                         chunk = (struct btrfs_chunk *)sb_array_offset;
1833                         /*
1834                          * At least one btrfs_chunk with one stripe must be
1835                          * present, exact stripe count check comes afterwards
1836                          */
1837                         len = btrfs_chunk_item_size(1);
1838                         if (cur_offset + len > array_size)
1839                                 goto out_short_read;
1840
1841                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1842                         if (!num_stripes) {
1843                                 printk(
1844             "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1845                                         num_stripes, cur_offset);
1846                                 ret = -EIO;
1847                                 break;
1848                         }
1849
1850                         len = btrfs_chunk_item_size(num_stripes);
1851                         if (cur_offset + len > array_size)
1852                                 goto out_short_read;
1853
1854                         ret = read_one_chunk(root, &key, sb, chunk, -1);
1855                         if (ret)
1856                                 break;
1857                 } else {
1858                         BUG();
1859                 }
1860                 array_ptr += len;
1861                 sb_array_offset += len;
1862                 cur_offset += len;
1863         }
1864         free_extent_buffer(sb);
1865         return ret;
1866
1867 out_short_read:
1868         printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
1869                         len, cur_offset);
1870         free_extent_buffer(sb);
1871         return -EIO;
1872 }
1873
1874 int btrfs_read_chunk_tree(struct btrfs_root *root)
1875 {
1876         struct btrfs_path *path;
1877         struct extent_buffer *leaf;
1878         struct btrfs_key key;
1879         struct btrfs_key found_key;
1880         int ret;
1881         int slot;
1882
1883         root = root->fs_info->chunk_root;
1884
1885         path = btrfs_alloc_path();
1886         if (!path)
1887                 return -ENOMEM;
1888
1889         /*
1890          * Read all device items, and then all the chunk items. All
1891          * device items are found before any chunk item (their object id
1892          * is smaller than the lowest possible object id for a chunk
1893          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1894          */
1895         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1896         key.offset = 0;
1897         key.type = 0;
1898         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1899         if (ret < 0)
1900                 goto error;
1901         while(1) {
1902                 leaf = path->nodes[0];
1903                 slot = path->slots[0];
1904                 if (slot >= btrfs_header_nritems(leaf)) {
1905                         ret = btrfs_next_leaf(root, path);
1906                         if (ret == 0)
1907                                 continue;
1908                         if (ret < 0)
1909                                 goto error;
1910                         break;
1911                 }
1912                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1913                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1914                         struct btrfs_dev_item *dev_item;
1915                         dev_item = btrfs_item_ptr(leaf, slot,
1916                                                   struct btrfs_dev_item);
1917                         ret = read_one_dev(root, leaf, dev_item);
1918                         BUG_ON(ret);
1919                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1920                         struct btrfs_chunk *chunk;
1921                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1922                         ret = read_one_chunk(root, &found_key, leaf, chunk,
1923                                              slot);
1924                         BUG_ON(ret);
1925                 }
1926                 path->slots[0]++;
1927         }
1928
1929         ret = 0;
1930 error:
1931         btrfs_free_path(path);
1932         return ret;
1933 }
1934
1935 struct list_head *btrfs_scanned_uuids(void)
1936 {
1937         return &fs_uuids;
1938 }
1939
1940 static int rmw_eb(struct btrfs_fs_info *info,
1941                   struct extent_buffer *eb, struct extent_buffer *orig_eb)
1942 {
1943         int ret;
1944         unsigned long orig_off = 0;
1945         unsigned long dest_off = 0;
1946         unsigned long copy_len = eb->len;
1947
1948         ret = read_whole_eb(info, eb, 0);
1949         if (ret)
1950                 return ret;
1951
1952         if (eb->start + eb->len <= orig_eb->start ||
1953             eb->start >= orig_eb->start + orig_eb->len)
1954                 return 0;
1955         /*
1956          * | ----- orig_eb ------- |
1957          *         | ----- stripe -------  |
1958          *         | ----- orig_eb ------- |
1959          *              | ----- orig_eb ------- |
1960          */
1961         if (eb->start > orig_eb->start)
1962                 orig_off = eb->start - orig_eb->start;
1963         if (orig_eb->start > eb->start)
1964                 dest_off = orig_eb->start - eb->start;
1965
1966         if (copy_len > orig_eb->len - orig_off)
1967                 copy_len = orig_eb->len - orig_off;
1968         if (copy_len > eb->len - dest_off)
1969                 copy_len = eb->len - dest_off;
1970
1971         memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1972         return 0;
1973 }
1974
1975 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1976                                 struct extent_buffer *orig_eb,
1977                                struct extent_buffer **ebs,
1978                                u64 stripe_len, u64 *raid_map,
1979                                int num_stripes)
1980 {
1981         struct extent_buffer *eb;
1982         u64 start = orig_eb->start;
1983         u64 this_eb_start;
1984         int i;
1985         int ret;
1986
1987         for (i = 0; i < num_stripes; i++) {
1988                 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1989                         break;
1990
1991                 eb = calloc(1, sizeof(struct extent_buffer) + stripe_len);
1992                 if (!eb)
1993                         BUG();
1994
1995                 eb->start = raid_map[i];
1996                 eb->len = stripe_len;
1997                 eb->refs = 1;
1998                 eb->flags = 0;
1999                 eb->fd = -1;
2000                 eb->dev_bytenr = (u64)-1;
2001
2002                 this_eb_start = raid_map[i];
2003
2004                 if (start > this_eb_start ||
2005                     start + orig_eb->len < this_eb_start + stripe_len) {
2006                         ret = rmw_eb(info, eb, orig_eb);
2007                         BUG_ON(ret);
2008                 } else {
2009                         memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
2010                 }
2011                 ebs[i] = eb;
2012         }
2013 }
2014
2015 int write_raid56_with_parity(struct btrfs_fs_info *info,
2016                              struct extent_buffer *eb,
2017                              struct btrfs_multi_bio *multi,
2018                              u64 stripe_len, u64 *raid_map)
2019 {
2020         struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2021         int i;
2022         int j;
2023         int ret;
2024         int alloc_size = eb->len;
2025
2026         ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
2027         BUG_ON(!ebs);
2028
2029         if (stripe_len > alloc_size)
2030                 alloc_size = stripe_len;
2031
2032         split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2033                             multi->num_stripes);
2034
2035         for (i = 0; i < multi->num_stripes; i++) {
2036                 struct extent_buffer *new_eb;
2037                 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2038                         ebs[i]->dev_bytenr = multi->stripes[i].physical;
2039                         ebs[i]->fd = multi->stripes[i].dev->fd;
2040                         multi->stripes[i].dev->total_ios++;
2041                         BUG_ON(ebs[i]->start != raid_map[i]);
2042                         continue;
2043                 }
2044                 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
2045                 BUG_ON(!new_eb);
2046                 new_eb->dev_bytenr = multi->stripes[i].physical;
2047                 new_eb->fd = multi->stripes[i].dev->fd;
2048                 multi->stripes[i].dev->total_ios++;
2049                 new_eb->len = stripe_len;
2050
2051                 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2052                         p_eb = new_eb;
2053                 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2054                         q_eb = new_eb;
2055         }
2056         if (q_eb) {
2057                 void **pointers;
2058
2059                 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
2060                                    GFP_NOFS);
2061                 BUG_ON(!pointers);
2062
2063                 ebs[multi->num_stripes - 2] = p_eb;
2064                 ebs[multi->num_stripes - 1] = q_eb;
2065
2066                 for (i = 0; i < multi->num_stripes; i++)
2067                         pointers[i] = ebs[i]->data;
2068
2069                 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2070                 kfree(pointers);
2071         } else {
2072                 ebs[multi->num_stripes - 1] = p_eb;
2073                 memcpy(p_eb->data, ebs[0]->data, stripe_len);
2074                 for (j = 1; j < multi->num_stripes - 1; j++) {
2075                         for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
2076                                 *(unsigned long *)(p_eb->data + i) ^=
2077                                         *(unsigned long *)(ebs[j]->data + i);
2078                         }
2079                 }
2080         }
2081
2082         for (i = 0; i < multi->num_stripes; i++) {
2083                 ret = write_extent_to_disk(ebs[i]);
2084                 BUG_ON(ret);
2085                 if (ebs[i] != eb)
2086                         kfree(ebs[i]);
2087         }
2088
2089         kfree(ebs);
2090
2091         return 0;
2092 }