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         u64 cur = *logical;
1174
1175         ce = search_cache_extent(&map_tree->cache_tree, cur);
1176
1177         while (ce) {
1178                 /*
1179                  * only jump to next bg if our cur is not 0
1180                  * As the initial logical for btrfs_next_bg() is 0, and
1181                  * if we jump to next bg, we skipped a valid bg.
1182                  */
1183                 if (cur) {
1184                         ce = next_cache_extent(ce);
1185                         if (!ce)
1186                                 return -ENOENT;
1187                 }
1188
1189                 cur = ce->start;
1190                 map = container_of(ce, struct map_lookup, ce);
1191                 if (map->type & type) {
1192                         *logical = ce->start;
1193                         *size = ce->size;
1194                         return 0;
1195                 }
1196         }
1197
1198         return -ENOENT;
1199 }
1200
1201 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1202                      u64 chunk_start, u64 physical, u64 devid,
1203                      u64 **logical, int *naddrs, int *stripe_len)
1204 {
1205         struct cache_extent *ce;
1206         struct map_lookup *map;
1207         u64 *buf;
1208         u64 bytenr;
1209         u64 length;
1210         u64 stripe_nr;
1211         u64 rmap_len;
1212         int i, j, nr = 0;
1213
1214         ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1215         BUG_ON(!ce);
1216         map = container_of(ce, struct map_lookup, ce);
1217
1218         length = ce->size;
1219         rmap_len = map->stripe_len;
1220         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1221                 length = ce->size / (map->num_stripes / map->sub_stripes);
1222         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1223                 length = ce->size / map->num_stripes;
1224         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1225                               BTRFS_BLOCK_GROUP_RAID6)) {
1226                 length = ce->size / nr_data_stripes(map);
1227                 rmap_len = map->stripe_len * nr_data_stripes(map);
1228         }
1229
1230         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1231
1232         for (i = 0; i < map->num_stripes; i++) {
1233                 if (devid && map->stripes[i].dev->devid != devid)
1234                         continue;
1235                 if (map->stripes[i].physical > physical ||
1236                     map->stripes[i].physical + length <= physical)
1237                         continue;
1238
1239                 stripe_nr = (physical - map->stripes[i].physical) /
1240                             map->stripe_len;
1241
1242                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1243                         stripe_nr = (stripe_nr * map->num_stripes + i) /
1244                                     map->sub_stripes;
1245                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1246                         stripe_nr = stripe_nr * map->num_stripes + i;
1247                 } /* else if RAID[56], multiply by nr_data_stripes().
1248                    * Alternatively, just use rmap_len below instead of
1249                    * map->stripe_len */
1250
1251                 bytenr = ce->start + stripe_nr * rmap_len;
1252                 for (j = 0; j < nr; j++) {
1253                         if (buf[j] == bytenr)
1254                                 break;
1255                 }
1256                 if (j == nr)
1257                         buf[nr++] = bytenr;
1258         }
1259
1260         *logical = buf;
1261         *naddrs = nr;
1262         *stripe_len = rmap_len;
1263
1264         return 0;
1265 }
1266
1267 static inline int parity_smaller(u64 a, u64 b)
1268 {
1269         return a > b;
1270 }
1271
1272 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1273 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1274 {
1275         struct btrfs_bio_stripe s;
1276         int i;
1277         u64 l;
1278         int again = 1;
1279
1280         while (again) {
1281                 again = 0;
1282                 for (i = 0; i < bbio->num_stripes - 1; i++) {
1283                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
1284                                 s = bbio->stripes[i];
1285                                 l = raid_map[i];
1286                                 bbio->stripes[i] = bbio->stripes[i+1];
1287                                 raid_map[i] = raid_map[i+1];
1288                                 bbio->stripes[i+1] = s;
1289                                 raid_map[i+1] = l;
1290                                 again = 1;
1291                         }
1292                 }
1293         }
1294 }
1295
1296 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1297                     u64 logical, u64 *length,
1298                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1299                     u64 **raid_map_ret)
1300 {
1301         return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1302                                  multi_ret, mirror_num, raid_map_ret);
1303 }
1304
1305 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1306                     u64 logical, u64 *length, u64 *type,
1307                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1308                     u64 **raid_map_ret)
1309 {
1310         struct cache_extent *ce;
1311         struct map_lookup *map;
1312         u64 offset;
1313         u64 stripe_offset;
1314         u64 stripe_nr;
1315         u64 *raid_map = NULL;
1316         int stripes_allocated = 8;
1317         int stripes_required = 1;
1318         int stripe_index;
1319         int i;
1320         struct btrfs_multi_bio *multi = NULL;
1321
1322         if (multi_ret && rw == READ) {
1323                 stripes_allocated = 1;
1324         }
1325 again:
1326         ce = search_cache_extent(&map_tree->cache_tree, logical);
1327         if (!ce) {
1328                 kfree(multi);
1329                 *length = (u64)-1;
1330                 return -ENOENT;
1331         }
1332         if (ce->start > logical) {
1333                 kfree(multi);
1334                 *length = ce->start - logical;
1335                 return -ENOENT;
1336         }
1337
1338         if (multi_ret) {
1339                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1340                                 GFP_NOFS);
1341                 if (!multi)
1342                         return -ENOMEM;
1343         }
1344         map = container_of(ce, struct map_lookup, ce);
1345         offset = logical - ce->start;
1346
1347         if (rw == WRITE) {
1348                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1349                                  BTRFS_BLOCK_GROUP_DUP)) {
1350                         stripes_required = map->num_stripes;
1351                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1352                         stripes_required = map->sub_stripes;
1353                 }
1354         }
1355         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1356             && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1357                     /* RAID[56] write or recovery. Return all stripes */
1358                     stripes_required = map->num_stripes;
1359
1360                     /* Only allocate the map if we've already got a large enough multi_ret */
1361                     if (stripes_allocated >= stripes_required) {
1362                             raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1363                             if (!raid_map) {
1364                                     kfree(multi);
1365                                     return -ENOMEM;
1366                             }
1367                     }
1368         }
1369
1370         /* if our multi bio struct is too small, back off and try again */
1371         if (multi_ret && stripes_allocated < stripes_required) {
1372                 stripes_allocated = stripes_required;
1373                 kfree(multi);
1374                 multi = NULL;
1375                 goto again;
1376         }
1377         stripe_nr = offset;
1378         /*
1379          * stripe_nr counts the total number of stripes we have to stride
1380          * to get to this block
1381          */
1382         stripe_nr = stripe_nr / map->stripe_len;
1383
1384         stripe_offset = stripe_nr * map->stripe_len;
1385         BUG_ON(offset < stripe_offset);
1386
1387         /* stripe_offset is the offset of this block in its stripe*/
1388         stripe_offset = offset - stripe_offset;
1389
1390         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1391                          BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1392                          BTRFS_BLOCK_GROUP_RAID10 |
1393                          BTRFS_BLOCK_GROUP_DUP)) {
1394                 /* we limit the length of each bio to what fits in a stripe */
1395                 *length = min_t(u64, ce->size - offset,
1396                               map->stripe_len - stripe_offset);
1397         } else {
1398                 *length = ce->size - offset;
1399         }
1400
1401         if (!multi_ret)
1402                 goto out;
1403
1404         multi->num_stripes = 1;
1405         stripe_index = 0;
1406         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1407                 if (rw == WRITE)
1408                         multi->num_stripes = map->num_stripes;
1409                 else if (mirror_num)
1410                         stripe_index = mirror_num - 1;
1411                 else
1412                         stripe_index = stripe_nr % map->num_stripes;
1413         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1414                 int factor = map->num_stripes / map->sub_stripes;
1415
1416                 stripe_index = stripe_nr % factor;
1417                 stripe_index *= map->sub_stripes;
1418
1419                 if (rw == WRITE)
1420                         multi->num_stripes = map->sub_stripes;
1421                 else if (mirror_num)
1422                         stripe_index += mirror_num - 1;
1423
1424                 stripe_nr = stripe_nr / factor;
1425         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1426                 if (rw == WRITE)
1427                         multi->num_stripes = map->num_stripes;
1428                 else if (mirror_num)
1429                         stripe_index = mirror_num - 1;
1430         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1431                                 BTRFS_BLOCK_GROUP_RAID6)) {
1432
1433                 if (raid_map) {
1434                         int rot;
1435                         u64 tmp;
1436                         u64 raid56_full_stripe_start;
1437                         u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1438
1439                         /*
1440                          * align the start of our data stripe in the logical
1441                          * address space
1442                          */
1443                         raid56_full_stripe_start = offset / full_stripe_len;
1444                         raid56_full_stripe_start *= full_stripe_len;
1445
1446                         /* get the data stripe number */
1447                         stripe_nr = raid56_full_stripe_start / map->stripe_len;
1448                         stripe_nr = stripe_nr / nr_data_stripes(map);
1449
1450                         /* Work out the disk rotation on this stripe-set */
1451                         rot = stripe_nr % map->num_stripes;
1452
1453                         /* Fill in the logical address of each stripe */
1454                         tmp = stripe_nr * nr_data_stripes(map);
1455
1456                         for (i = 0; i < nr_data_stripes(map); i++)
1457                                 raid_map[(i+rot) % map->num_stripes] =
1458                                         ce->start + (tmp + i) * map->stripe_len;
1459
1460                         raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1461                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1462                                 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1463
1464                         *length = map->stripe_len;
1465                         stripe_index = 0;
1466                         stripe_offset = 0;
1467                         multi->num_stripes = map->num_stripes;
1468                 } else {
1469                         stripe_index = stripe_nr % nr_data_stripes(map);
1470                         stripe_nr = stripe_nr / nr_data_stripes(map);
1471
1472                         /*
1473                          * Mirror #0 or #1 means the original data block.
1474                          * Mirror #2 is RAID5 parity block.
1475                          * Mirror #3 is RAID6 Q block.
1476                          */
1477                         if (mirror_num > 1)
1478                                 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1479
1480                         /* We distribute the parity blocks across stripes */
1481                         stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1482                 }
1483         } else {
1484                 /*
1485                  * after this do_div call, stripe_nr is the number of stripes
1486                  * on this device we have to walk to find the data, and
1487                  * stripe_index is the number of our device in the stripe array
1488                  */
1489                 stripe_index = stripe_nr % map->num_stripes;
1490                 stripe_nr = stripe_nr / map->num_stripes;
1491         }
1492         BUG_ON(stripe_index >= map->num_stripes);
1493
1494         for (i = 0; i < multi->num_stripes; i++) {
1495                 multi->stripes[i].physical =
1496                         map->stripes[stripe_index].physical + stripe_offset +
1497                         stripe_nr * map->stripe_len;
1498                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1499                 stripe_index++;
1500         }
1501         *multi_ret = multi;
1502
1503         if (type)
1504                 *type = map->type;
1505
1506         if (raid_map) {
1507                 sort_parity_stripes(multi, raid_map);
1508                 *raid_map_ret = raid_map;
1509         }
1510 out:
1511         return 0;
1512 }
1513
1514 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1515                                        u8 *uuid, u8 *fsid)
1516 {
1517         struct btrfs_device *device;
1518         struct btrfs_fs_devices *cur_devices;
1519
1520         cur_devices = root->fs_info->fs_devices;
1521         while (cur_devices) {
1522                 if (!fsid ||
1523                     (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1524                      root->fs_info->ignore_fsid_mismatch)) {
1525                         device = __find_device(&cur_devices->devices,
1526                                                devid, uuid);
1527                         if (device)
1528                                 return device;
1529                 }
1530                 cur_devices = cur_devices->seed;
1531         }
1532         return NULL;
1533 }
1534
1535 struct btrfs_device *
1536 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1537                            u64 devid, int instance)
1538 {
1539         struct list_head *head = &fs_devices->devices;
1540         struct btrfs_device *dev;
1541         int num_found = 0;
1542
1543         list_for_each_entry(dev, head, dev_list) {
1544                 if (dev->devid == devid && num_found++ == instance)
1545                         return dev;
1546         }
1547         return NULL;
1548 }
1549
1550 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1551 {
1552         struct cache_extent *ce;
1553         struct map_lookup *map;
1554         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1555         int readonly = 0;
1556         int i;
1557
1558         /*
1559          * During chunk recovering, we may fail to find block group's
1560          * corresponding chunk, we will rebuild it later
1561          */
1562         ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1563         if (!root->fs_info->is_chunk_recover)
1564                 BUG_ON(!ce);
1565         else
1566                 return 0;
1567
1568         map = container_of(ce, struct map_lookup, ce);
1569         for (i = 0; i < map->num_stripes; i++) {
1570                 if (!map->stripes[i].dev->writeable) {
1571                         readonly = 1;
1572                         break;
1573                 }
1574         }
1575
1576         return readonly;
1577 }
1578
1579 static struct btrfs_device *fill_missing_device(u64 devid)
1580 {
1581         struct btrfs_device *device;
1582
1583         device = kzalloc(sizeof(*device), GFP_NOFS);
1584         device->devid = devid;
1585         device->fd = -1;
1586         return device;
1587 }
1588
1589 /*
1590  * Slot is used to verfy the chunk item is valid
1591  *
1592  * For sys chunk in superblock, pass -1 to indicate sys chunk.
1593  */
1594 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1595                           struct extent_buffer *leaf,
1596                           struct btrfs_chunk *chunk, int slot)
1597 {
1598         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1599         struct map_lookup *map;
1600         struct cache_extent *ce;
1601         u64 logical;
1602         u64 length;
1603         u64 stripe_len;
1604         u64 devid;
1605         u8 uuid[BTRFS_UUID_SIZE];
1606         int num_stripes;
1607         int ret;
1608         int i;
1609
1610         logical = key->offset;
1611         length = btrfs_chunk_length(leaf, chunk);
1612         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1613         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1614         /* Validation check */
1615         if (!num_stripes) {
1616                 error("invalid chunk num_stripes: %u", num_stripes);
1617                 return -EIO;
1618         }
1619         if (!IS_ALIGNED(logical, root->sectorsize)) {
1620                 error("invalid chunk logical %llu", logical);
1621                 return -EIO;
1622         }
1623         if (!length || !IS_ALIGNED(length, root->sectorsize)) {
1624                 error("invalid chunk length %llu", length);
1625                 return -EIO;
1626         }
1627         if (!is_power_of_2(stripe_len)) {
1628                 error("invalid chunk stripe length: %llu", stripe_len);
1629                 return -EIO;
1630         }
1631         if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
1632             btrfs_chunk_type(leaf, chunk)) {
1633                 error("unrecognized chunk type: %llu",
1634                       ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1635                         BTRFS_BLOCK_GROUP_PROFILE_MASK) &
1636                       btrfs_chunk_type(leaf, chunk));
1637                 return -EIO;
1638         }
1639
1640         ce = search_cache_extent(&map_tree->cache_tree, logical);
1641
1642         /* already mapped? */
1643         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1644                 return 0;
1645         }
1646
1647         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1648         if (!map)
1649                 return -ENOMEM;
1650
1651         map->ce.start = logical;
1652         map->ce.size = length;
1653         map->num_stripes = num_stripes;
1654         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1655         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1656         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1657         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1658         map->type = btrfs_chunk_type(leaf, chunk);
1659         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1660
1661         /* Check on chunk item type */
1662         if (map->type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1663                           BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1664                 fprintf(stderr, "Unknown chunk type bits: %llu\n",
1665                         map->type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1666                                       BTRFS_BLOCK_GROUP_PROFILE_MASK));
1667                 ret = -EIO;
1668                 goto out;
1669         }
1670
1671         /*
1672          * Btrfs_chunk contains at least one stripe, and for sys_chunk
1673          * it can't exceed the system chunk array size
1674          * For normal chunk, it should match its chunk item size.
1675          */
1676         if (num_stripes < 1 ||
1677             (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1678              BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1679             (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1680              btrfs_item_size_nr(leaf, slot))) {
1681                 fprintf(stderr, "Invalid num_stripes: %u\n",
1682                         num_stripes);
1683                 ret = -EIO;
1684                 goto out;
1685         }
1686
1687         /*
1688          * Device number check against profile
1689          */
1690         if ((map->type & BTRFS_BLOCK_GROUP_RAID10 && map->sub_stripes == 0) ||
1691             (map->type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1692             (map->type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1693             (map->type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1694             (map->type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1695             ((map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1696              num_stripes != 1)) {
1697                 fprintf(stderr,
1698                         "Invalid num_stripes:sub_stripes %u:%u for profile %llu\n",
1699                         num_stripes, map->sub_stripes,
1700                         map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1701                 ret = -EIO;
1702                 goto out;
1703         }
1704
1705         for (i = 0; i < num_stripes; i++) {
1706                 map->stripes[i].physical =
1707                         btrfs_stripe_offset_nr(leaf, chunk, i);
1708                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1709                 read_extent_buffer(leaf, uuid, (unsigned long)
1710                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1711                                    BTRFS_UUID_SIZE);
1712                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1713                                                         NULL);
1714                 if (!map->stripes[i].dev) {
1715                         map->stripes[i].dev = fill_missing_device(devid);
1716                         printf("warning, device %llu is missing\n",
1717                                (unsigned long long)devid);
1718                 }
1719
1720         }
1721         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1722         BUG_ON(ret);
1723
1724         return 0;
1725 out:
1726         free(map);
1727         return ret;
1728 }
1729
1730 static int fill_device_from_item(struct extent_buffer *leaf,
1731                                  struct btrfs_dev_item *dev_item,
1732                                  struct btrfs_device *device)
1733 {
1734         unsigned long ptr;
1735
1736         device->devid = btrfs_device_id(leaf, dev_item);
1737         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1738         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1739         device->type = btrfs_device_type(leaf, dev_item);
1740         device->io_align = btrfs_device_io_align(leaf, dev_item);
1741         device->io_width = btrfs_device_io_width(leaf, dev_item);
1742         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1743
1744         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1745         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1746
1747         return 0;
1748 }
1749
1750 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1751 {
1752         struct btrfs_fs_devices *fs_devices;
1753         int ret;
1754
1755         fs_devices = root->fs_info->fs_devices->seed;
1756         while (fs_devices) {
1757                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1758                         ret = 0;
1759                         goto out;
1760                 }
1761                 fs_devices = fs_devices->seed;
1762         }
1763
1764         fs_devices = find_fsid(fsid);
1765         if (!fs_devices) {
1766                 /* missing all seed devices */
1767                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1768                 if (!fs_devices) {
1769                         ret = -ENOMEM;
1770                         goto out;
1771                 }
1772                 INIT_LIST_HEAD(&fs_devices->devices);
1773                 list_add(&fs_devices->list, &fs_uuids);
1774                 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1775         }
1776
1777         ret = btrfs_open_devices(fs_devices, O_RDONLY);
1778         if (ret)
1779                 goto out;
1780
1781         fs_devices->seed = root->fs_info->fs_devices->seed;
1782         root->fs_info->fs_devices->seed = fs_devices;
1783 out:
1784         return ret;
1785 }
1786
1787 static int read_one_dev(struct btrfs_root *root,
1788                         struct extent_buffer *leaf,
1789                         struct btrfs_dev_item *dev_item)
1790 {
1791         struct btrfs_device *device;
1792         u64 devid;
1793         int ret = 0;
1794         u8 fs_uuid[BTRFS_UUID_SIZE];
1795         u8 dev_uuid[BTRFS_UUID_SIZE];
1796
1797         devid = btrfs_device_id(leaf, dev_item);
1798         read_extent_buffer(leaf, dev_uuid,
1799                            (unsigned long)btrfs_device_uuid(dev_item),
1800                            BTRFS_UUID_SIZE);
1801         read_extent_buffer(leaf, fs_uuid,
1802                            (unsigned long)btrfs_device_fsid(dev_item),
1803                            BTRFS_UUID_SIZE);
1804
1805         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1806                 ret = open_seed_devices(root, fs_uuid);
1807                 if (ret)
1808                         return ret;
1809         }
1810
1811         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1812         if (!device) {
1813                 device = kzalloc(sizeof(*device), GFP_NOFS);
1814                 if (!device)
1815                         return -ENOMEM;
1816                 device->fd = -1;
1817                 list_add(&device->dev_list,
1818                          &root->fs_info->fs_devices->devices);
1819         }
1820
1821         fill_device_from_item(leaf, dev_item, device);
1822         device->dev_root = root->fs_info->dev_root;
1823         return ret;
1824 }
1825
1826 int btrfs_read_sys_array(struct btrfs_root *root)
1827 {
1828         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1829         struct extent_buffer *sb;
1830         struct btrfs_disk_key *disk_key;
1831         struct btrfs_chunk *chunk;
1832         u8 *array_ptr;
1833         unsigned long sb_array_offset;
1834         int ret = 0;
1835         u32 num_stripes;
1836         u32 array_size;
1837         u32 len = 0;
1838         u32 cur_offset;
1839         struct btrfs_key key;
1840
1841         sb = btrfs_find_create_tree_block(root->fs_info,
1842                                           BTRFS_SUPER_INFO_OFFSET,
1843                                           BTRFS_SUPER_INFO_SIZE);
1844         if (!sb)
1845                 return -ENOMEM;
1846         btrfs_set_buffer_uptodate(sb);
1847         write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1848         array_size = btrfs_super_sys_array_size(super_copy);
1849
1850         array_ptr = super_copy->sys_chunk_array;
1851         sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1852         cur_offset = 0;
1853
1854         while (cur_offset < array_size) {
1855                 disk_key = (struct btrfs_disk_key *)array_ptr;
1856                 len = sizeof(*disk_key);
1857                 if (cur_offset + len > array_size)
1858                         goto out_short_read;
1859
1860                 btrfs_disk_key_to_cpu(&key, disk_key);
1861
1862                 array_ptr += len;
1863                 sb_array_offset += len;
1864                 cur_offset += len;
1865
1866                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1867                         chunk = (struct btrfs_chunk *)sb_array_offset;
1868                         /*
1869                          * At least one btrfs_chunk with one stripe must be
1870                          * present, exact stripe count check comes afterwards
1871                          */
1872                         len = btrfs_chunk_item_size(1);
1873                         if (cur_offset + len > array_size)
1874                                 goto out_short_read;
1875
1876                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1877                         if (!num_stripes) {
1878                                 printk(
1879             "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1880                                         num_stripes, cur_offset);
1881                                 ret = -EIO;
1882                                 break;
1883                         }
1884
1885                         len = btrfs_chunk_item_size(num_stripes);
1886                         if (cur_offset + len > array_size)
1887                                 goto out_short_read;
1888
1889                         ret = read_one_chunk(root, &key, sb, chunk, -1);
1890                         if (ret)
1891                                 break;
1892                 } else {
1893                         printk(
1894                 "ERROR: unexpected item type %u in sys_array at offset %u\n",
1895                                 (u32)key.type, cur_offset);
1896                         ret = -EIO;
1897                         break;
1898                 }
1899                 array_ptr += len;
1900                 sb_array_offset += len;
1901                 cur_offset += len;
1902         }
1903         free_extent_buffer(sb);
1904         return ret;
1905
1906 out_short_read:
1907         printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
1908                         len, cur_offset);
1909         free_extent_buffer(sb);
1910         return -EIO;
1911 }
1912
1913 int btrfs_read_chunk_tree(struct btrfs_root *root)
1914 {
1915         struct btrfs_path *path;
1916         struct extent_buffer *leaf;
1917         struct btrfs_key key;
1918         struct btrfs_key found_key;
1919         int ret;
1920         int slot;
1921
1922         root = root->fs_info->chunk_root;
1923
1924         path = btrfs_alloc_path();
1925         if (!path)
1926                 return -ENOMEM;
1927
1928         /*
1929          * Read all device items, and then all the chunk items. All
1930          * device items are found before any chunk item (their object id
1931          * is smaller than the lowest possible object id for a chunk
1932          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1933          */
1934         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1935         key.offset = 0;
1936         key.type = 0;
1937         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1938         if (ret < 0)
1939                 goto error;
1940         while(1) {
1941                 leaf = path->nodes[0];
1942                 slot = path->slots[0];
1943                 if (slot >= btrfs_header_nritems(leaf)) {
1944                         ret = btrfs_next_leaf(root, path);
1945                         if (ret == 0)
1946                                 continue;
1947                         if (ret < 0)
1948                                 goto error;
1949                         break;
1950                 }
1951                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1952                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1953                         struct btrfs_dev_item *dev_item;
1954                         dev_item = btrfs_item_ptr(leaf, slot,
1955                                                   struct btrfs_dev_item);
1956                         ret = read_one_dev(root, leaf, dev_item);
1957                         BUG_ON(ret);
1958                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1959                         struct btrfs_chunk *chunk;
1960                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1961                         ret = read_one_chunk(root, &found_key, leaf, chunk,
1962                                              slot);
1963                         BUG_ON(ret);
1964                 }
1965                 path->slots[0]++;
1966         }
1967
1968         ret = 0;
1969 error:
1970         btrfs_free_path(path);
1971         return ret;
1972 }
1973
1974 struct list_head *btrfs_scanned_uuids(void)
1975 {
1976         return &fs_uuids;
1977 }
1978
1979 static int rmw_eb(struct btrfs_fs_info *info,
1980                   struct extent_buffer *eb, struct extent_buffer *orig_eb)
1981 {
1982         int ret;
1983         unsigned long orig_off = 0;
1984         unsigned long dest_off = 0;
1985         unsigned long copy_len = eb->len;
1986
1987         ret = read_whole_eb(info, eb, 0);
1988         if (ret)
1989                 return ret;
1990
1991         if (eb->start + eb->len <= orig_eb->start ||
1992             eb->start >= orig_eb->start + orig_eb->len)
1993                 return 0;
1994         /*
1995          * | ----- orig_eb ------- |
1996          *         | ----- stripe -------  |
1997          *         | ----- orig_eb ------- |
1998          *              | ----- orig_eb ------- |
1999          */
2000         if (eb->start > orig_eb->start)
2001                 orig_off = eb->start - orig_eb->start;
2002         if (orig_eb->start > eb->start)
2003                 dest_off = orig_eb->start - eb->start;
2004
2005         if (copy_len > orig_eb->len - orig_off)
2006                 copy_len = orig_eb->len - orig_off;
2007         if (copy_len > eb->len - dest_off)
2008                 copy_len = eb->len - dest_off;
2009
2010         memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2011         return 0;
2012 }
2013
2014 static void split_eb_for_raid56(struct btrfs_fs_info *info,
2015                                 struct extent_buffer *orig_eb,
2016                                struct extent_buffer **ebs,
2017                                u64 stripe_len, u64 *raid_map,
2018                                int num_stripes)
2019 {
2020         struct extent_buffer *eb;
2021         u64 start = orig_eb->start;
2022         u64 this_eb_start;
2023         int i;
2024         int ret;
2025
2026         for (i = 0; i < num_stripes; i++) {
2027                 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2028                         break;
2029
2030                 eb = calloc(1, sizeof(struct extent_buffer) + stripe_len);
2031                 if (!eb)
2032                         BUG();
2033
2034                 eb->start = raid_map[i];
2035                 eb->len = stripe_len;
2036                 eb->refs = 1;
2037                 eb->flags = 0;
2038                 eb->fd = -1;
2039                 eb->dev_bytenr = (u64)-1;
2040
2041                 this_eb_start = raid_map[i];
2042
2043                 if (start > this_eb_start ||
2044                     start + orig_eb->len < this_eb_start + stripe_len) {
2045                         ret = rmw_eb(info, eb, orig_eb);
2046                         BUG_ON(ret);
2047                 } else {
2048                         memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
2049                 }
2050                 ebs[i] = eb;
2051         }
2052 }
2053
2054 int write_raid56_with_parity(struct btrfs_fs_info *info,
2055                              struct extent_buffer *eb,
2056                              struct btrfs_multi_bio *multi,
2057                              u64 stripe_len, u64 *raid_map)
2058 {
2059         struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2060         int i;
2061         int j;
2062         int ret;
2063         int alloc_size = eb->len;
2064
2065         ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
2066         BUG_ON(!ebs);
2067
2068         if (stripe_len > alloc_size)
2069                 alloc_size = stripe_len;
2070
2071         split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2072                             multi->num_stripes);
2073
2074         for (i = 0; i < multi->num_stripes; i++) {
2075                 struct extent_buffer *new_eb;
2076                 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2077                         ebs[i]->dev_bytenr = multi->stripes[i].physical;
2078                         ebs[i]->fd = multi->stripes[i].dev->fd;
2079                         multi->stripes[i].dev->total_ios++;
2080                         BUG_ON(ebs[i]->start != raid_map[i]);
2081                         continue;
2082                 }
2083                 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
2084                 BUG_ON(!new_eb);
2085                 new_eb->dev_bytenr = multi->stripes[i].physical;
2086                 new_eb->fd = multi->stripes[i].dev->fd;
2087                 multi->stripes[i].dev->total_ios++;
2088                 new_eb->len = stripe_len;
2089
2090                 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2091                         p_eb = new_eb;
2092                 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2093                         q_eb = new_eb;
2094         }
2095         if (q_eb) {
2096                 void **pointers;
2097
2098                 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
2099                                    GFP_NOFS);
2100                 BUG_ON(!pointers);
2101
2102                 ebs[multi->num_stripes - 2] = p_eb;
2103                 ebs[multi->num_stripes - 1] = q_eb;
2104
2105                 for (i = 0; i < multi->num_stripes; i++)
2106                         pointers[i] = ebs[i]->data;
2107
2108                 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2109                 kfree(pointers);
2110         } else {
2111                 ebs[multi->num_stripes - 1] = p_eb;
2112                 memcpy(p_eb->data, ebs[0]->data, stripe_len);
2113                 for (j = 1; j < multi->num_stripes - 1; j++) {
2114                         for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
2115                                 *(unsigned long *)(p_eb->data + i) ^=
2116                                         *(unsigned long *)(ebs[j]->data + i);
2117                         }
2118                 }
2119         }
2120
2121         for (i = 0; i < multi->num_stripes; i++) {
2122                 ret = write_extent_to_disk(ebs[i]);
2123                 BUG_ON(ret);
2124                 if (ebs[i] != eb)
2125                         kfree(ebs[i]);
2126         }
2127
2128         kfree(ebs);
2129
2130         return 0;
2131 }