volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #define _XOPEN_SOURCE 600
  19 #define __USE_XOPEN2K
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <uuid/uuid.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "print-tree.h"
  31 #include "volumes.h"
  32 #include "utils.h"
  33
  34 struct stripe {
  35         struct btrfs_device *dev;
  36         u64 physical;
  37 };
  38
  39 static inline int nr_parity_stripes(struct map_lookup *map)
  40 {
  41         if (map->type & BTRFS_BLOCK_GROUP_RAID5)
  42                 return 1;
  43         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
  44                 return 2;
  45         else
  46                 return 0;
  47 }
  48
  49 static inline int nr_data_stripes(struct map_lookup *map)
  50 {
  51         return map->num_stripes - nr_parity_stripes(map);
  52 }
  53
  54 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
  55
  56 static LIST_HEAD(fs_uuids);
  57
  58 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  59                                           u8 *uuid)
  60 {
  61         struct btrfs_device *dev;
  62         struct list_head *cur;
  63
  64         list_for_each(cur, head) {
  65                 dev = list_entry(cur, struct btrfs_device, dev_list);
  66                 if (dev->devid == devid &&
  67                     !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
  68                         return dev;
  69                 }
  70         }
  71         return NULL;
  72 }
  73
  74 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  75 {
  76         struct list_head *cur;
  77         struct btrfs_fs_devices *fs_devices;
  78
  79         list_for_each(cur, &fs_uuids) {
  80                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  81                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  82                         return fs_devices;
  83         }
  84         return NULL;
  85 }
  86
  87 static int device_list_add(const char *path,
  88                            struct btrfs_super_block *disk_super,
  89                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
  90 {
  91         struct btrfs_device *device;
  92         struct btrfs_fs_devices *fs_devices;
  93         u64 found_transid = btrfs_super_generation(disk_super);
  94
  95         fs_devices = find_fsid(disk_super->fsid);
  96         if (!fs_devices) {
  97                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
  98                 if (!fs_devices)
  99                         return -ENOMEM;
 100                 INIT_LIST_HEAD(&fs_devices->devices);
 101                 list_add(&fs_devices->list, &fs_uuids);
 102                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 103                 fs_devices->latest_devid = devid;
 104                 fs_devices->latest_trans = found_transid;
 105                 fs_devices->lowest_devid = (u64)-1;
 106                 device = NULL;
 107         } else {
 108                 device = __find_device(&fs_devices->devices, devid,
 109                                        disk_super->dev_item.uuid);
 110         }
 111         if (!device) {
 112                 device = kzalloc(sizeof(*device), GFP_NOFS);
 113                 if (!device) {
 114                         /* we can safely leave the fs_devices entry around */
 115                         return -ENOMEM;
 116                 }
 117                 device->fd = -1;
 118                 device->devid = devid;
 119                 device->generation = found_transid;
 120                 memcpy(device->uuid, disk_super->dev_item.uuid,
 121                        BTRFS_UUID_SIZE);
 122                 device->name = kstrdup(path, GFP_NOFS);
 123                 if (!device->name) {
 124                         kfree(device);
 125                         return -ENOMEM;
 126                 }
 127                 device->label = kstrdup(disk_super->label, GFP_NOFS);
 128                 if (!device->label) {
 129                         kfree(device->name);
 130                         kfree(device);
 131                         return -ENOMEM;
 132                 }
 133                 device->total_devs = btrfs_super_num_devices(disk_super);
 134                 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
 135                 device->total_bytes =
 136                         btrfs_stack_device_total_bytes(&disk_super->dev_item);
 137                 device->bytes_used =
 138                         btrfs_stack_device_bytes_used(&disk_super->dev_item);
 139                 list_add(&device->dev_list, &fs_devices->devices);
 140                 device->fs_devices = fs_devices;
 141         } else if (!device->name || strcmp(device->name, path)) {
 142                 char *name = strdup(path);
 143                 if (!name)
 144                         return -ENOMEM;
 145                 kfree(device->name);
 146                 device->name = name;
 147         }
 148
 149
 150         if (found_transid > fs_devices->latest_trans) {
 151                 fs_devices->latest_devid = devid;
 152                 fs_devices->latest_trans = found_transid;
 153         }
 154         if (fs_devices->lowest_devid > devid) {
 155                 fs_devices->lowest_devid = devid;
 156         }
 157         *fs_devices_ret = fs_devices;
 158         return 0;
 159 }
 160
 161 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 162 {
 163         struct btrfs_fs_devices *seed_devices;
 164         struct btrfs_device *device;
 165
 166 again:
 167         while (!list_empty(&fs_devices->devices)) {
 168                 device = list_entry(fs_devices->devices.next,
 169                                     struct btrfs_device, dev_list);
 170                 if (device->fd != -1) {
 171                         fsync(device->fd);
 172                         if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
 173                                 fprintf(stderr, "Warning, could not drop caches\n");
 174                         close(device->fd);
 175                         device->fd = -1;
 176                 }
 177                 device->writeable = 0;
 178                 list_del(&device->dev_list);
 179                 /* free the memory */
 180                 free(device->name);
 181                 free(device->label);
 182                 free(device);
 183         }
 184
 185         seed_devices = fs_devices->seed;
 186         fs_devices->seed = NULL;
 187         if (seed_devices) {
 188                 struct btrfs_fs_devices *orig;
 189
 190                 orig = fs_devices;
 191                 fs_devices = seed_devices;
 192                 list_del(&orig->list);
 193                 free(orig);
 194                 goto again;
 195         } else {
 196                 list_del(&fs_devices->list);
 197                 free(fs_devices);
 198         }
 199
 200         return 0;
 201 }
 202
 203 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
 204 {
 205         int fd;
 206         struct list_head *head = &fs_devices->devices;
 207         struct list_head *cur;
 208         struct btrfs_device *device;
 209         int ret;
 210
 211         list_for_each(cur, head) {
 212                 device = list_entry(cur, struct btrfs_device, dev_list);
 213                 if (!device->name) {
 214                         printk("no name for device %llu, skip it now\n", device->devid);
 215                         continue;
 216                 }
 217
 218                 fd = open(device->name, flags);
 219                 if (fd < 0) {
 220                         ret = -errno;
 221                         goto fail;
 222                 }
 223
 224                 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
 225                         fprintf(stderr, "Warning, could not drop caches\n");
 226
 227                 if (device->devid == fs_devices->latest_devid)
 228                         fs_devices->latest_bdev = fd;
 229                 if (device->devid == fs_devices->lowest_devid)
 230                         fs_devices->lowest_bdev = fd;
 231                 device->fd = fd;
 232                 if (flags & O_RDWR)
 233                         device->writeable = 1;
 234         }
 235         return 0;
 236 fail:
 237         btrfs_close_devices(fs_devices);
 238         return ret;
 239 }
 240
 241 int btrfs_scan_one_device(int fd, const char *path,
 242                           struct btrfs_fs_devices **fs_devices_ret,
 243                           u64 *total_devs, u64 super_offset, int super_recover)
 244 {
 245         struct btrfs_super_block *disk_super;
 246         char *buf;
 247         int ret;
 248         u64 devid;
 249
 250         buf = malloc(4096);
 251         if (!buf) {
 252                 ret = -ENOMEM;
 253                 goto error;
 254         }
 255         disk_super = (struct btrfs_super_block *)buf;
 256         ret = btrfs_read_dev_super(fd, disk_super, super_offset, super_recover);
 257         if (ret < 0) {
 258                 ret = -EIO;
 259                 goto error_brelse;
 260         }
 261         devid = btrfs_stack_device_id(&disk_super->dev_item);
 262         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
 263                 *total_devs = 1;
 264         else
 265                 *total_devs = btrfs_super_num_devices(disk_super);
 266
 267         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 268
 269 error_brelse:
 270         free(buf);
 271 error:
 272         return ret;
 273 }
 274
 275 /*
 276  * this uses a pretty simple search, the expectation is that it is
 277  * called very infrequently and that a given device has a small number
 278  * of extents
 279  */
 280 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 281                                 struct btrfs_device *device,
 282                                 struct btrfs_path *path,
 283                                 u64 num_bytes, u64 *start)
 284 {
 285         struct btrfs_key key;
 286         struct btrfs_root *root = device->dev_root;
 287         struct btrfs_dev_extent *dev_extent = NULL;
 288         u64 hole_size = 0;
 289         u64 last_byte = 0;
 290         u64 search_start = root->fs_info->alloc_start;
 291         u64 search_end = device->total_bytes;
 292         int ret;
 293         int slot = 0;
 294         int start_found;
 295         struct extent_buffer *l;
 296
 297         start_found = 0;
 298         path->reada = 2;
 299
 300         /* FIXME use last free of some kind */
 301
 302         /* we don't want to overwrite the superblock on the drive,
 303          * so we make sure to start at an offset of at least 1MB
 304          */
 305         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 306
 307         if (search_start >= search_end) {
 308                 ret = -ENOSPC;
 309                 goto error;
 310         }
 311
 312         key.objectid = device->devid;
 313         key.offset = search_start;
 314         key.type = BTRFS_DEV_EXTENT_KEY;
 315         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 316         if (ret < 0)
 317                 goto error;
 318         ret = btrfs_previous_item(root, path, 0, key.type);
 319         if (ret < 0)
 320                 goto error;
 321         l = path->nodes[0];
 322         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 323         while (1) {
 324                 l = path->nodes[0];
 325                 slot = path->slots[0];
 326                 if (slot >= btrfs_header_nritems(l)) {
 327                         ret = btrfs_next_leaf(root, path);
 328                         if (ret == 0)
 329                                 continue;
 330                         if (ret < 0)
 331                                 goto error;
 332 no_more_items:
 333                         if (!start_found) {
 334                                 if (search_start >= search_end) {
 335                                         ret = -ENOSPC;
 336                                         goto error;
 337                                 }
 338                                 *start = search_start;
 339                                 start_found = 1;
 340                                 goto check_pending;
 341                         }
 342                         *start = last_byte > search_start ?
 343                                 last_byte : search_start;
 344                         if (search_end <= *start) {
 345                                 ret = -ENOSPC;
 346                                 goto error;
 347                         }
 348                         goto check_pending;
 349                 }
 350                 btrfs_item_key_to_cpu(l, &key, slot);
 351
 352                 if (key.objectid < device->devid)
 353                         goto next;
 354
 355                 if (key.objectid > device->devid)
 356                         goto no_more_items;
 357
 358                 if (key.offset >= search_start && key.offset > last_byte &&
 359                     start_found) {
 360                         if (last_byte < search_start)
 361                                 last_byte = search_start;
 362                         hole_size = key.offset - last_byte;
 363                         if (key.offset > last_byte &&
 364                             hole_size >= num_bytes) {
 365                                 *start = last_byte;
 366                                 goto check_pending;
 367                         }
 368                 }
 369                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 370                         goto next;
 371                 }
 372
 373                 start_found = 1;
 374                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 375                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 376 next:
 377                 path->slots[0]++;
 378                 cond_resched();
 379         }
 380 check_pending:
 381         /* we have to make sure we didn't find an extent that has already
 382          * been allocated by the map tree or the original allocation
 383          */
 384         btrfs_release_path(path);
 385         BUG_ON(*start < search_start);
 386
 387         if (*start + num_bytes > search_end) {
 388                 ret = -ENOSPC;
 389                 goto error;
 390         }
 391         /* check for pending inserts here */
 392         return 0;
 393
 394 error:
 395         btrfs_release_path(path);
 396         return ret;
 397 }
 398
 399 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 400                                   struct btrfs_device *device,
 401                                   u64 chunk_tree, u64 chunk_objectid,
 402                                   u64 chunk_offset,
 403                                   u64 num_bytes, u64 *start)
 404 {
 405         int ret;
 406         struct btrfs_path *path;
 407         struct btrfs_root *root = device->dev_root;
 408         struct btrfs_dev_extent *extent;
 409         struct extent_buffer *leaf;
 410         struct btrfs_key key;
 411
 412         path = btrfs_alloc_path();
 413         if (!path)
 414                 return -ENOMEM;
 415
 416         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 417         if (ret) {
 418                 goto err;
 419         }
 420
 421         key.objectid = device->devid;
 422         key.offset = *start;
 423         key.type = BTRFS_DEV_EXTENT_KEY;
 424         ret = btrfs_insert_empty_item(trans, root, path, &key,
 425                                       sizeof(*extent));
 426         BUG_ON(ret);
 427
 428         leaf = path->nodes[0];
 429         extent = btrfs_item_ptr(leaf, path->slots[0],
 430                                 struct btrfs_dev_extent);
 431         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 432         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 433         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 434
 435         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 436                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 437                     BTRFS_UUID_SIZE);
 438
 439         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 440         btrfs_mark_buffer_dirty(leaf);
 441 err:
 442         btrfs_free_path(path);
 443         return ret;
 444 }
 445
 446 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 447 {
 448         struct btrfs_path *path;
 449         int ret;
 450         struct btrfs_key key;
 451         struct btrfs_chunk *chunk;
 452         struct btrfs_key found_key;
 453
 454         path = btrfs_alloc_path();
 455         BUG_ON(!path);
 456
 457         key.objectid = objectid;
 458         key.offset = (u64)-1;
 459         key.type = BTRFS_CHUNK_ITEM_KEY;
 460
 461         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 462         if (ret < 0)
 463                 goto error;
 464
 465         BUG_ON(ret == 0);
 466
 467         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 468         if (ret) {
 469                 *offset = 0;
 470         } else {
 471                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 472                                       path->slots[0]);
 473                 if (found_key.objectid != objectid)
 474                         *offset = 0;
 475                 else {
 476                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 477                                                struct btrfs_chunk);
 478                         *offset = found_key.offset +
 479                                 btrfs_chunk_length(path->nodes[0], chunk);
 480                 }
 481         }
 482         ret = 0;
 483 error:
 484         btrfs_free_path(path);
 485         return ret;
 486 }
 487
 488 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 489                            u64 *objectid)
 490 {
 491         int ret;
 492         struct btrfs_key key;
 493         struct btrfs_key found_key;
 494
 495         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 496         key.type = BTRFS_DEV_ITEM_KEY;
 497         key.offset = (u64)-1;
 498
 499         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 500         if (ret < 0)
 501                 goto error;
 502
 503         BUG_ON(ret == 0);
 504
 505         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 506                                   BTRFS_DEV_ITEM_KEY);
 507         if (ret) {
 508                 *objectid = 1;
 509         } else {
 510                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 511                                       path->slots[0]);
 512                 *objectid = found_key.offset + 1;
 513         }
 514         ret = 0;
 515 error:
 516         btrfs_release_path(path);
 517         return ret;
 518 }
 519
 520 /*
 521  * the device information is stored in the chunk root
 522  * the btrfs_device struct should be fully filled in
 523  */
 524 int btrfs_add_device(struct btrfs_trans_handle *trans,
 525                      struct btrfs_root *root,
 526                      struct btrfs_device *device)
 527 {
 528         int ret;
 529         struct btrfs_path *path;
 530         struct btrfs_dev_item *dev_item;
 531         struct extent_buffer *leaf;
 532         struct btrfs_key key;
 533         unsigned long ptr;
 534         u64 free_devid = 0;
 535
 536         root = root->fs_info->chunk_root;
 537
 538         path = btrfs_alloc_path();
 539         if (!path)
 540                 return -ENOMEM;
 541
 542         ret = find_next_devid(root, path, &free_devid);
 543         if (ret)
 544                 goto out;
 545
 546         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 547         key.type = BTRFS_DEV_ITEM_KEY;
 548         key.offset = free_devid;
 549
 550         ret = btrfs_insert_empty_item(trans, root, path, &key,
 551                                       sizeof(*dev_item));
 552         if (ret)
 553                 goto out;
 554
 555         leaf = path->nodes[0];
 556         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 557
 558         device->devid = free_devid;
 559         btrfs_set_device_id(leaf, dev_item, device->devid);
 560         btrfs_set_device_generation(leaf, dev_item, 0);
 561         btrfs_set_device_type(leaf, dev_item, device->type);
 562         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 563         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 564         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 565         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 566         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 567         btrfs_set_device_group(leaf, dev_item, 0);
 568         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 569         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 570         btrfs_set_device_start_offset(leaf, dev_item, 0);
 571
 572         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 573         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 574         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 575         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 576         btrfs_mark_buffer_dirty(leaf);
 577         ret = 0;
 578
 579 out:
 580         btrfs_free_path(path);
 581         return ret;
 582 }
 583
 584 int btrfs_update_device(struct btrfs_trans_handle *trans,
 585                         struct btrfs_device *device)
 586 {
 587         int ret;
 588         struct btrfs_path *path;
 589         struct btrfs_root *root;
 590         struct btrfs_dev_item *dev_item;
 591         struct extent_buffer *leaf;
 592         struct btrfs_key key;
 593
 594         root = device->dev_root->fs_info->chunk_root;
 595
 596         path = btrfs_alloc_path();
 597         if (!path)
 598                 return -ENOMEM;
 599
 600         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 601         key.type = BTRFS_DEV_ITEM_KEY;
 602         key.offset = device->devid;
 603
 604         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 605         if (ret < 0)
 606                 goto out;
 607
 608         if (ret > 0) {
 609                 ret = -ENOENT;
 610                 goto out;
 611         }
 612
 613         leaf = path->nodes[0];
 614         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 615
 616         btrfs_set_device_id(leaf, dev_item, device->devid);
 617         btrfs_set_device_type(leaf, dev_item, device->type);
 618         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 619         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 620         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 621         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 622         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 623         btrfs_mark_buffer_dirty(leaf);
 624
 625 out:
 626         btrfs_free_path(path);
 627         return ret;
 628 }
 629
 630 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 631                            struct btrfs_root *root,
 632                            struct btrfs_key *key,
 633                            struct btrfs_chunk *chunk, int item_size)
 634 {
 635         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
 636         struct btrfs_disk_key disk_key;
 637         u32 array_size;
 638         u8 *ptr;
 639
 640         array_size = btrfs_super_sys_array_size(super_copy);
 641         if (array_size + item_size + sizeof(disk_key)
 642                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 643                 return -EFBIG;
 644
 645         ptr = super_copy->sys_chunk_array + array_size;
 646         btrfs_cpu_key_to_disk(&disk_key, key);
 647         memcpy(ptr, &disk_key, sizeof(disk_key));
 648         ptr += sizeof(disk_key);
 649         memcpy(ptr, chunk, item_size);
 650         item_size += sizeof(disk_key);
 651         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 652         return 0;
 653 }
 654
 655 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 656                                int sub_stripes)
 657 {
 658         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 659                 return calc_size;
 660         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 661                 return calc_size * (num_stripes / sub_stripes);
 662         else if (type & BTRFS_BLOCK_GROUP_RAID5)
 663                 return calc_size * (num_stripes - 1);
 664         else if (type & BTRFS_BLOCK_GROUP_RAID6)
 665                 return calc_size * (num_stripes - 2);
 666         else
 667                 return calc_size * num_stripes;
 668 }
 669
 670
 671 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
 672 {
 673         /* TODO, add a way to store the preferred stripe size */
 674         return BTRFS_STRIPE_LEN;
 675 }
 676
 677 /*
 678  * btrfs_device_avail_bytes - count bytes available for alloc_chunk
 679  *
 680  * It is not equal to "device->total_bytes - device->bytes_used".
 681  * We do not allocate any chunk in 1M at beginning of device, and not
 682  * allowed to allocate any chunk before alloc_start if it is specified.
 683  * So search holes from max(1M, alloc_start) to device->total_bytes.
 684  */
 685 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
 686                                     struct btrfs_device *device,
 687                                     u64 *avail_bytes)
 688 {
 689         struct btrfs_path *path;
 690         struct btrfs_root *root = device->dev_root;
 691         struct btrfs_key key;
 692         struct btrfs_dev_extent *dev_extent = NULL;
 693         struct extent_buffer *l;
 694         u64 search_start = root->fs_info->alloc_start;
 695         u64 search_end = device->total_bytes;
 696         u64 extent_end = 0;
 697         u64 free_bytes = 0;
 698         int ret;
 699         int slot = 0;
 700
 701         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 702
 703         path = btrfs_alloc_path();
 704         if (!path)
 705                 return -ENOMEM;
 706
 707         key.objectid = device->devid;
 708         key.offset = root->fs_info->alloc_start;
 709         key.type = BTRFS_DEV_EXTENT_KEY;
 710
 711         path->reada = 2;
 712         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 713         if (ret < 0)
 714                 goto error;
 715         ret = btrfs_previous_item(root, path, 0, key.type);
 716         if (ret < 0)
 717                 goto error;
 718
 719         while (1) {
 720                 l = path->nodes[0];
 721                 slot = path->slots[0];
 722                 if (slot >= btrfs_header_nritems(l)) {
 723                         ret = btrfs_next_leaf(root, path);
 724                         if (ret == 0)
 725                                 continue;
 726                         if (ret < 0)
 727                                 goto error;
 728                         break;
 729                 }
 730                 btrfs_item_key_to_cpu(l, &key, slot);
 731
 732                 if (key.objectid < device->devid)
 733                         goto next;
 734                 if (key.objectid > device->devid)
 735                         break;
 736                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 737                         goto next;
 738                 if (key.offset > search_end)
 739                         break;
 740                 if (key.offset > search_start)
 741                         free_bytes += key.offset - search_start;
 742
 743                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 744                 extent_end = key.offset + btrfs_dev_extent_length(l,
 745                                                                   dev_extent);
 746                 if (extent_end > search_start)
 747                         search_start = extent_end;
 748                 if (search_start > search_end)
 749                         break;
 750 next:
 751                 path->slots[0]++;
 752                 cond_resched();
 753         }
 754
 755         if (search_start < search_end)
 756                 free_bytes += search_end - search_start;
 757
 758         *avail_bytes = free_bytes;
 759         ret = 0;
 760 error:
 761         btrfs_free_path(path);
 762         return ret;
 763 }
 764
 765 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
 766                         - sizeof(struct btrfs_item)             \
 767                         - sizeof(struct btrfs_chunk))           \
 768                         / sizeof(struct btrfs_stripe) + 1)
 769
 770 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
 771                                 - 2 * sizeof(struct btrfs_disk_key)     \
 772                                 - 2 * sizeof(struct btrfs_chunk))       \
 773                                 / sizeof(struct btrfs_stripe) + 1)
 774
 775 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 776                       struct btrfs_root *extent_root, u64 *start,
 777                       u64 *num_bytes, u64 type)
 778 {
 779         u64 dev_offset;
 780         struct btrfs_fs_info *info = extent_root->fs_info;
 781         struct btrfs_root *chunk_root = info->chunk_root;
 782         struct btrfs_stripe *stripes;
 783         struct btrfs_device *device = NULL;
 784         struct btrfs_chunk *chunk;
 785         struct list_head private_devs;
 786         struct list_head *dev_list = &info->fs_devices->devices;
 787         struct list_head *cur;
 788         struct map_lookup *map;
 789         int min_stripe_size = 1 * 1024 * 1024;
 790         u64 calc_size = 8 * 1024 * 1024;
 791         u64 min_free;
 792         u64 max_chunk_size = 4 * calc_size;
 793         u64 avail = 0;
 794         u64 max_avail = 0;
 795         u64 percent_max;
 796         int num_stripes = 1;
 797         int max_stripes = 0;
 798         int min_stripes = 1;
 799         int sub_stripes = 0;
 800         int looped = 0;
 801         int ret;
 802         int index;
 803         int stripe_len = BTRFS_STRIPE_LEN;
 804         struct btrfs_key key;
 805         u64 offset;
 806
 807         if (list_empty(dev_list)) {
 808                 return -ENOSPC;
 809         }
 810
 811         if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
 812                     BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
 813                     BTRFS_BLOCK_GROUP_RAID10 |
 814                     BTRFS_BLOCK_GROUP_DUP)) {
 815                 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 816                         calc_size = 8 * 1024 * 1024;
 817                         max_chunk_size = calc_size * 2;
 818                         min_stripe_size = 1 * 1024 * 1024;
 819                         max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
 820                 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
 821                         calc_size = 1024 * 1024 * 1024;
 822                         max_chunk_size = 10 * calc_size;
 823                         min_stripe_size = 64 * 1024 * 1024;
 824                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 825                 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 826                         calc_size = 1024 * 1024 * 1024;
 827                         max_chunk_size = 4 * calc_size;
 828                         min_stripe_size = 32 * 1024 * 1024;
 829                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 830                 }
 831         }
 832         if (type & BTRFS_BLOCK_GROUP_RAID1) {
 833                 num_stripes = min_t(u64, 2,
 834                                   btrfs_super_num_devices(info->super_copy));
 835                 if (num_stripes < 2)
 836                         return -ENOSPC;
 837                 min_stripes = 2;
 838         }
 839         if (type & BTRFS_BLOCK_GROUP_DUP) {
 840                 num_stripes = 2;
 841                 min_stripes = 2;
 842         }
 843         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
 844                 num_stripes = btrfs_super_num_devices(info->super_copy);
 845                 if (num_stripes > max_stripes)
 846                         num_stripes = max_stripes;
 847                 min_stripes = 2;
 848         }
 849         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 850                 num_stripes = btrfs_super_num_devices(info->super_copy);
 851                 if (num_stripes > max_stripes)
 852                         num_stripes = max_stripes;
 853                 if (num_stripes < 4)
 854                         return -ENOSPC;
 855                 num_stripes &= ~(u32)1;
 856                 sub_stripes = 2;
 857                 min_stripes = 4;
 858         }
 859         if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
 860                 num_stripes = btrfs_super_num_devices(info->super_copy);
 861                 if (num_stripes > max_stripes)
 862                         num_stripes = max_stripes;
 863                 if (num_stripes < 2)
 864                         return -ENOSPC;
 865                 min_stripes = 2;
 866                 stripe_len = find_raid56_stripe_len(num_stripes - 1,
 867                                     btrfs_super_stripesize(info->super_copy));
 868         }
 869         if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
 870                 num_stripes = btrfs_super_num_devices(info->super_copy);
 871                 if (num_stripes > max_stripes)
 872                         num_stripes = max_stripes;
 873                 if (num_stripes < 3)
 874                         return -ENOSPC;
 875                 min_stripes = 3;
 876                 stripe_len = find_raid56_stripe_len(num_stripes - 2,
 877                                     btrfs_super_stripesize(info->super_copy));
 878         }
 879
 880         /* we don't want a chunk larger than 10% of the FS */
 881         percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
 882         max_chunk_size = min(percent_max, max_chunk_size);
 883
 884 again:
 885         if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
 886             max_chunk_size) {
 887                 calc_size = max_chunk_size;
 888                 calc_size /= num_stripes;
 889                 calc_size /= stripe_len;
 890                 calc_size *= stripe_len;
 891         }
 892         /* we don't want tiny stripes */
 893         calc_size = max_t(u64, calc_size, min_stripe_size);
 894
 895         calc_size /= stripe_len;
 896         calc_size *= stripe_len;
 897         INIT_LIST_HEAD(&private_devs);
 898         cur = dev_list->next;
 899         index = 0;
 900
 901         if (type & BTRFS_BLOCK_GROUP_DUP)
 902                 min_free = calc_size * 2;
 903         else
 904                 min_free = calc_size;
 905
 906         /* build a private list of devices we will allocate from */
 907         while(index < num_stripes) {
 908                 device = list_entry(cur, struct btrfs_device, dev_list);
 909                 ret = btrfs_device_avail_bytes(trans, device, &avail);
 910                 if (ret)
 911                         return ret;
 912                 cur = cur->next;
 913                 if (avail >= min_free) {
 914                         list_move_tail(&device->dev_list, &private_devs);
 915                         index++;
 916                         if (type & BTRFS_BLOCK_GROUP_DUP)
 917                                 index++;
 918                 } else if (avail > max_avail)
 919                         max_avail = avail;
 920                 if (cur == dev_list)
 921                         break;
 922         }
 923         if (index < num_stripes) {
 924                 list_splice(&private_devs, dev_list);
 925                 if (index >= min_stripes) {
 926                         num_stripes = index;
 927                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 928                                 num_stripes /= sub_stripes;
 929                                 num_stripes *= sub_stripes;
 930                         }
 931                         looped = 1;
 932                         goto again;
 933                 }
 934                 if (!looped && max_avail > 0) {
 935                         looped = 1;
 936                         calc_size = max_avail;
 937                         goto again;
 938                 }
 939                 return -ENOSPC;
 940         }
 941         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 942                               &offset);
 943         if (ret)
 944                 return ret;
 945         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 946         key.type = BTRFS_CHUNK_ITEM_KEY;
 947         key.offset = offset;
 948
 949         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 950         if (!chunk)
 951                 return -ENOMEM;
 952
 953         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
 954         if (!map) {
 955                 kfree(chunk);
 956                 return -ENOMEM;
 957         }
 958
 959         stripes = &chunk->stripe;
 960         *num_bytes = chunk_bytes_by_type(type, calc_size,
 961                                          num_stripes, sub_stripes);
 962         index = 0;
 963         while(index < num_stripes) {
 964                 struct btrfs_stripe *stripe;
 965                 BUG_ON(list_empty(&private_devs));
 966                 cur = private_devs.next;
 967                 device = list_entry(cur, struct btrfs_device, dev_list);
 968
 969                 /* loop over this device again if we're doing a dup group */
 970                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 971                     (index == num_stripes - 1))
 972                         list_move_tail(&device->dev_list, dev_list);
 973
 974                 ret = btrfs_alloc_dev_extent(trans, device,
 975                              info->chunk_root->root_key.objectid,
 976                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 977                              calc_size, &dev_offset);
 978                 BUG_ON(ret);
 979
 980                 device->bytes_used += calc_size;
 981                 ret = btrfs_update_device(trans, device);
 982                 BUG_ON(ret);
 983
 984                 map->stripes[index].dev = device;
 985                 map->stripes[index].physical = dev_offset;
 986                 stripe = stripes + index;
 987                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 988                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 989                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 990                 index++;
 991         }
 992         BUG_ON(!list_empty(&private_devs));
 993
 994         /* key was set above */
 995         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 996         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 997         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 998         btrfs_set_stack_chunk_type(chunk, type);
 999         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1000         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1001         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1002         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1003         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1004         map->sector_size = extent_root->sectorsize;
1005         map->stripe_len = stripe_len;
1006         map->io_align = stripe_len;
1007         map->io_width = stripe_len;
1008         map->type = type;
1009         map->num_stripes = num_stripes;
1010         map->sub_stripes = sub_stripes;
1011
1012         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1013                                 btrfs_chunk_item_size(num_stripes));
1014         BUG_ON(ret);
1015         *start = key.offset;;
1016
1017         map->ce.start = key.offset;
1018         map->ce.size = *num_bytes;
1019
1020         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1021         BUG_ON(ret);
1022
1023         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1024                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1025                                     chunk, btrfs_chunk_item_size(num_stripes));
1026                 BUG_ON(ret);
1027         }
1028
1029         kfree(chunk);
1030         return ret;
1031 }
1032
1033 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1034                            struct btrfs_root *extent_root, u64 *start,
1035                            u64 num_bytes, u64 type)
1036 {
1037         u64 dev_offset;
1038         struct btrfs_fs_info *info = extent_root->fs_info;
1039         struct btrfs_root *chunk_root = info->chunk_root;
1040         struct btrfs_stripe *stripes;
1041         struct btrfs_device *device = NULL;
1042         struct btrfs_chunk *chunk;
1043         struct list_head *dev_list = &info->fs_devices->devices;
1044         struct list_head *cur;
1045         struct map_lookup *map;
1046         u64 calc_size = 8 * 1024 * 1024;
1047         int num_stripes = 1;
1048         int sub_stripes = 0;
1049         int ret;
1050         int index;
1051         int stripe_len = BTRFS_STRIPE_LEN;
1052         struct btrfs_key key;
1053
1054         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1055         key.type = BTRFS_CHUNK_ITEM_KEY;
1056         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1057                               &key.offset);
1058         if (ret)
1059                 return ret;
1060
1061         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1062         if (!chunk)
1063                 return -ENOMEM;
1064
1065         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1066         if (!map) {
1067                 kfree(chunk);
1068                 return -ENOMEM;
1069         }
1070
1071         stripes = &chunk->stripe;
1072         calc_size = num_bytes;
1073
1074         index = 0;
1075         cur = dev_list->next;
1076         device = list_entry(cur, struct btrfs_device, dev_list);
1077
1078         while (index < num_stripes) {
1079                 struct btrfs_stripe *stripe;
1080
1081                 ret = btrfs_alloc_dev_extent(trans, device,
1082                              info->chunk_root->root_key.objectid,
1083                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1084                              calc_size, &dev_offset);
1085                 BUG_ON(ret);
1086
1087                 device->bytes_used += calc_size;
1088                 ret = btrfs_update_device(trans, device);
1089                 BUG_ON(ret);
1090
1091                 map->stripes[index].dev = device;
1092                 map->stripes[index].physical = dev_offset;
1093                 stripe = stripes + index;
1094                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1095                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1096                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1097                 index++;
1098         }
1099
1100         /* key was set above */
1101         btrfs_set_stack_chunk_length(chunk, num_bytes);
1102         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1103         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1104         btrfs_set_stack_chunk_type(chunk, type);
1105         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1106         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1107         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1108         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1109         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1110         map->sector_size = extent_root->sectorsize;
1111         map->stripe_len = stripe_len;
1112         map->io_align = stripe_len;
1113         map->io_width = stripe_len;
1114         map->type = type;
1115         map->num_stripes = num_stripes;
1116         map->sub_stripes = sub_stripes;
1117
1118         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1119                                 btrfs_chunk_item_size(num_stripes));
1120         BUG_ON(ret);
1121         *start = key.offset;
1122
1123         map->ce.start = key.offset;
1124         map->ce.size = num_bytes;
1125
1126         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1127         BUG_ON(ret);
1128
1129         kfree(chunk);
1130         return ret;
1131 }
1132
1133 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1134 {
1135         struct cache_extent *ce;
1136         struct map_lookup *map;
1137         int ret;
1138
1139         ce = search_cache_extent(&map_tree->cache_tree, logical);
1140         if (!ce) {
1141                 fprintf(stderr, "No mapping for %llu-%llu\n",
1142                         (unsigned long long)logical,
1143                         (unsigned long long)logical+len);
1144                 return 1;
1145         }
1146         if (ce->start > logical || ce->start + ce->size < logical) {
1147                 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1148                         "%llu-%llu\n", (unsigned long long)logical,
1149                         (unsigned long long)logical+len,
1150                         (unsigned long long)ce->start,
1151                         (unsigned long long)ce->start + ce->size);
1152                 return 1;
1153         }
1154         map = container_of(ce, struct map_lookup, ce);
1155
1156         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1157                 ret = map->num_stripes;
1158         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1159                 ret = map->sub_stripes;
1160         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1161                 ret = 2;
1162         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1163                 ret = 3;
1164         else
1165                 ret = 1;
1166         return ret;
1167 }
1168
1169 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1170                         u64 *size)
1171 {
1172         struct cache_extent *ce;
1173         struct map_lookup *map;
1174
1175         ce = search_cache_extent(&map_tree->cache_tree, *logical);
1176
1177         while (ce) {
1178                 ce = next_cache_extent(ce);
1179                 if (!ce)
1180                         return -ENOENT;
1181
1182                 map = container_of(ce, struct map_lookup, ce);
1183                 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1184                         *logical = ce->start;
1185                         *size = ce->size;
1186                         return 0;
1187                 }
1188         }
1189
1190         return -ENOENT;
1191 }
1192
1193 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1194                      u64 chunk_start, u64 physical, u64 devid,
1195                      u64 **logical, int *naddrs, int *stripe_len)
1196 {
1197         struct cache_extent *ce;
1198         struct map_lookup *map;
1199         u64 *buf;
1200         u64 bytenr;
1201         u64 length;
1202         u64 stripe_nr;
1203         u64 rmap_len;
1204         int i, j, nr = 0;
1205
1206         ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1207         BUG_ON(!ce);
1208         map = container_of(ce, struct map_lookup, ce);
1209
1210         length = ce->size;
1211         rmap_len = map->stripe_len;
1212         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1213                 length = ce->size / (map->num_stripes / map->sub_stripes);
1214         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1215                 length = ce->size / map->num_stripes;
1216         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1217                               BTRFS_BLOCK_GROUP_RAID6)) {
1218                 length = ce->size / nr_data_stripes(map);
1219                 rmap_len = map->stripe_len * nr_data_stripes(map);
1220         }
1221
1222         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1223
1224         for (i = 0; i < map->num_stripes; i++) {
1225                 if (devid && map->stripes[i].dev->devid != devid)
1226                         continue;
1227                 if (map->stripes[i].physical > physical ||
1228                     map->stripes[i].physical + length <= physical)
1229                         continue;
1230
1231                 stripe_nr = (physical - map->stripes[i].physical) /
1232                             map->stripe_len;
1233
1234                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1235                         stripe_nr = (stripe_nr * map->num_stripes + i) /
1236                                     map->sub_stripes;
1237                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1238                         stripe_nr = stripe_nr * map->num_stripes + i;
1239                 } /* else if RAID[56], multiply by nr_data_stripes().
1240                    * Alternatively, just use rmap_len below instead of
1241                    * map->stripe_len */
1242
1243                 bytenr = ce->start + stripe_nr * rmap_len;
1244                 for (j = 0; j < nr; j++) {
1245                         if (buf[j] == bytenr)
1246                                 break;
1247                 }
1248                 if (j == nr)
1249                         buf[nr++] = bytenr;
1250         }
1251
1252         *logical = buf;
1253         *naddrs = nr;
1254         *stripe_len = rmap_len;
1255
1256         return 0;
1257 }
1258
1259 static inline int parity_smaller(u64 a, u64 b)
1260 {
1261         return a > b;
1262 }
1263
1264 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1265 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1266 {
1267         struct btrfs_bio_stripe s;
1268         int i;
1269         u64 l;
1270         int again = 1;
1271
1272         while (again) {
1273                 again = 0;
1274                 for (i = 0; i < bbio->num_stripes - 1; i++) {
1275                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
1276                                 s = bbio->stripes[i];
1277                                 l = raid_map[i];
1278                                 bbio->stripes[i] = bbio->stripes[i+1];
1279                                 raid_map[i] = raid_map[i+1];
1280                                 bbio->stripes[i+1] = s;
1281                                 raid_map[i+1] = l;
1282                                 again = 1;
1283                         }
1284                 }
1285         }
1286 }
1287
1288 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1289                     u64 logical, u64 *length,
1290                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1291                     u64 **raid_map_ret)
1292 {
1293         return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1294                                  multi_ret, mirror_num, raid_map_ret);
1295 }
1296
1297 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1298                     u64 logical, u64 *length, u64 *type,
1299                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1300                     u64 **raid_map_ret)
1301 {
1302         struct cache_extent *ce;
1303         struct map_lookup *map;
1304         u64 offset;
1305         u64 stripe_offset;
1306         u64 stripe_nr;
1307         u64 *raid_map = NULL;
1308         int stripes_allocated = 8;
1309         int stripes_required = 1;
1310         int stripe_index;
1311         int i;
1312         struct btrfs_multi_bio *multi = NULL;
1313
1314         if (multi_ret && rw == READ) {
1315                 stripes_allocated = 1;
1316         }
1317 again:
1318         ce = search_cache_extent(&map_tree->cache_tree, logical);
1319         if (!ce) {
1320                 kfree(multi);
1321                 *length = (u64)-1;
1322                 return -ENOENT;
1323         }
1324         if (ce->start > logical) {
1325                 kfree(multi);
1326                 *length = ce->start - logical;
1327                 return -ENOENT;
1328         }
1329
1330         if (multi_ret) {
1331                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1332                                 GFP_NOFS);
1333                 if (!multi)
1334                         return -ENOMEM;
1335         }
1336         map = container_of(ce, struct map_lookup, ce);
1337         offset = logical - ce->start;
1338
1339         if (rw == WRITE) {
1340                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1341                                  BTRFS_BLOCK_GROUP_DUP)) {
1342                         stripes_required = map->num_stripes;
1343                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1344                         stripes_required = map->sub_stripes;
1345                 }
1346         }
1347         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1348             && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1349                     /* RAID[56] write or recovery. Return all stripes */
1350                     stripes_required = map->num_stripes;
1351
1352                     /* Only allocate the map if we've already got a large enough multi_ret */
1353                     if (stripes_allocated >= stripes_required) {
1354                             raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1355                             if (!raid_map) {
1356                                     kfree(multi);
1357                                     return -ENOMEM;
1358                             }
1359                     }
1360         }
1361
1362         /* if our multi bio struct is too small, back off and try again */
1363         if (multi_ret && stripes_allocated < stripes_required) {
1364                 stripes_allocated = stripes_required;
1365                 kfree(multi);
1366                 multi = NULL;
1367                 goto again;
1368         }
1369         stripe_nr = offset;
1370         /*
1371          * stripe_nr counts the total number of stripes we have to stride
1372          * to get to this block
1373          */
1374         stripe_nr = stripe_nr / map->stripe_len;
1375
1376         stripe_offset = stripe_nr * map->stripe_len;
1377         BUG_ON(offset < stripe_offset);
1378
1379         /* stripe_offset is the offset of this block in its stripe*/
1380         stripe_offset = offset - stripe_offset;
1381
1382         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1383                          BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1384                          BTRFS_BLOCK_GROUP_RAID10 |
1385                          BTRFS_BLOCK_GROUP_DUP)) {
1386                 /* we limit the length of each bio to what fits in a stripe */
1387                 *length = min_t(u64, ce->size - offset,
1388                               map->stripe_len - stripe_offset);
1389         } else {
1390                 *length = ce->size - offset;
1391         }
1392
1393         if (!multi_ret)
1394                 goto out;
1395
1396         multi->num_stripes = 1;
1397         stripe_index = 0;
1398         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1399                 if (rw == WRITE)
1400                         multi->num_stripes = map->num_stripes;
1401                 else if (mirror_num)
1402                         stripe_index = mirror_num - 1;
1403                 else
1404                         stripe_index = stripe_nr % map->num_stripes;
1405         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1406                 int factor = map->num_stripes / map->sub_stripes;
1407
1408                 stripe_index = stripe_nr % factor;
1409                 stripe_index *= map->sub_stripes;
1410
1411                 if (rw == WRITE)
1412                         multi->num_stripes = map->sub_stripes;
1413                 else if (mirror_num)
1414                         stripe_index += mirror_num - 1;
1415
1416                 stripe_nr = stripe_nr / factor;
1417         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1418                 if (rw == WRITE)
1419                         multi->num_stripes = map->num_stripes;
1420                 else if (mirror_num)
1421                         stripe_index = mirror_num - 1;
1422         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1423                                 BTRFS_BLOCK_GROUP_RAID6)) {
1424
1425                 if (raid_map) {
1426                         int rot;
1427                         u64 tmp;
1428                         u64 raid56_full_stripe_start;
1429                         u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1430
1431                         /*
1432                          * align the start of our data stripe in the logical
1433                          * address space
1434                          */
1435                         raid56_full_stripe_start = offset / full_stripe_len;
1436                         raid56_full_stripe_start *= full_stripe_len;
1437
1438                         /* get the data stripe number */
1439                         stripe_nr = raid56_full_stripe_start / map->stripe_len;
1440                         stripe_nr = stripe_nr / nr_data_stripes(map);
1441
1442                         /* Work out the disk rotation on this stripe-set */
1443                         rot = stripe_nr % map->num_stripes;
1444
1445                         /* Fill in the logical address of each stripe */
1446                         tmp = stripe_nr * nr_data_stripes(map);
1447
1448                         for (i = 0; i < nr_data_stripes(map); i++)
1449                                 raid_map[(i+rot) % map->num_stripes] =
1450                                         ce->start + (tmp + i) * map->stripe_len;
1451
1452                         raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1453                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1454                                 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1455
1456                         *length = map->stripe_len;
1457                         stripe_index = 0;
1458                         stripe_offset = 0;
1459                         multi->num_stripes = map->num_stripes;
1460                 } else {
1461                         stripe_index = stripe_nr % nr_data_stripes(map);
1462                         stripe_nr = stripe_nr / nr_data_stripes(map);
1463
1464                         /*
1465                          * Mirror #0 or #1 means the original data block.
1466                          * Mirror #2 is RAID5 parity block.
1467                          * Mirror #3 is RAID6 Q block.
1468                          */
1469                         if (mirror_num > 1)
1470                                 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1471
1472                         /* We distribute the parity blocks across stripes */
1473                         stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1474                 }
1475         } else {
1476                 /*
1477                  * after this do_div call, stripe_nr is the number of stripes
1478                  * on this device we have to walk to find the data, and
1479                  * stripe_index is the number of our device in the stripe array
1480                  */
1481                 stripe_index = stripe_nr % map->num_stripes;
1482                 stripe_nr = stripe_nr / map->num_stripes;
1483         }
1484         BUG_ON(stripe_index >= map->num_stripes);
1485
1486         for (i = 0; i < multi->num_stripes; i++) {
1487                 multi->stripes[i].physical =
1488                         map->stripes[stripe_index].physical + stripe_offset +
1489                         stripe_nr * map->stripe_len;
1490                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1491                 stripe_index++;
1492         }
1493         *multi_ret = multi;
1494
1495         if (type)
1496                 *type = map->type;
1497
1498         if (raid_map) {
1499                 sort_parity_stripes(multi, raid_map);
1500                 *raid_map_ret = raid_map;
1501         }
1502 out:
1503         return 0;
1504 }
1505
1506 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1507                                        u8 *uuid, u8 *fsid)
1508 {
1509         struct btrfs_device *device;
1510         struct btrfs_fs_devices *cur_devices;
1511
1512         cur_devices = root->fs_info->fs_devices;
1513         while (cur_devices) {
1514                 if (!fsid ||
1515                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
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 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1581                           struct extent_buffer *leaf,
1582                           struct btrfs_chunk *chunk)
1583 {
1584         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1585         struct map_lookup *map;
1586         struct cache_extent *ce;
1587         u64 logical;
1588         u64 length;
1589         u64 devid;
1590         u8 uuid[BTRFS_UUID_SIZE];
1591         int num_stripes;
1592         int ret;
1593         int i;
1594
1595         logical = key->offset;
1596         length = btrfs_chunk_length(leaf, chunk);
1597
1598         ce = search_cache_extent(&map_tree->cache_tree, logical);
1599
1600         /* already mapped? */
1601         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1602                 return 0;
1603         }
1604
1605         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1606         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1607         if (!map)
1608                 return -ENOMEM;
1609
1610         map->ce.start = logical;
1611         map->ce.size = length;
1612         map->num_stripes = num_stripes;
1613         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1614         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1615         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1616         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1617         map->type = btrfs_chunk_type(leaf, chunk);
1618         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1619
1620         for (i = 0; i < num_stripes; i++) {
1621                 map->stripes[i].physical =
1622                         btrfs_stripe_offset_nr(leaf, chunk, i);
1623                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1624                 read_extent_buffer(leaf, uuid, (unsigned long)
1625                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1626                                    BTRFS_UUID_SIZE);
1627                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1628                                                         NULL);
1629                 if (!map->stripes[i].dev) {
1630                         map->stripes[i].dev = fill_missing_device(devid);
1631                         printf("warning, device %llu is missing\n",
1632                                (unsigned long long)devid);
1633                 }
1634
1635         }
1636         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1637         BUG_ON(ret);
1638
1639         return 0;
1640 }
1641
1642 static int fill_device_from_item(struct extent_buffer *leaf,
1643                                  struct btrfs_dev_item *dev_item,
1644                                  struct btrfs_device *device)
1645 {
1646         unsigned long ptr;
1647
1648         device->devid = btrfs_device_id(leaf, dev_item);
1649         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1650         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1651         device->type = btrfs_device_type(leaf, dev_item);
1652         device->io_align = btrfs_device_io_align(leaf, dev_item);
1653         device->io_width = btrfs_device_io_width(leaf, dev_item);
1654         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1655
1656         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1657         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1658
1659         return 0;
1660 }
1661
1662 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1663 {
1664         struct btrfs_fs_devices *fs_devices;
1665         int ret;
1666
1667         fs_devices = root->fs_info->fs_devices->seed;
1668         while (fs_devices) {
1669                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1670                         ret = 0;
1671                         goto out;
1672                 }
1673                 fs_devices = fs_devices->seed;
1674         }
1675
1676         fs_devices = find_fsid(fsid);
1677         if (!fs_devices) {
1678                 /* missing all seed devices */
1679                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1680                 if (!fs_devices) {
1681                         ret = -ENOMEM;
1682                         goto out;
1683                 }
1684                 INIT_LIST_HEAD(&fs_devices->devices);
1685                 list_add(&fs_devices->list, &fs_uuids);
1686                 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1687         }
1688
1689         ret = btrfs_open_devices(fs_devices, O_RDONLY);
1690         if (ret)
1691                 goto out;
1692
1693         fs_devices->seed = root->fs_info->fs_devices->seed;
1694         root->fs_info->fs_devices->seed = fs_devices;
1695 out:
1696         return ret;
1697 }
1698
1699 static int read_one_dev(struct btrfs_root *root,
1700                         struct extent_buffer *leaf,
1701                         struct btrfs_dev_item *dev_item)
1702 {
1703         struct btrfs_device *device;
1704         u64 devid;
1705         int ret = 0;
1706         u8 fs_uuid[BTRFS_UUID_SIZE];
1707         u8 dev_uuid[BTRFS_UUID_SIZE];
1708
1709         devid = btrfs_device_id(leaf, dev_item);
1710         read_extent_buffer(leaf, dev_uuid,
1711                            (unsigned long)btrfs_device_uuid(dev_item),
1712                            BTRFS_UUID_SIZE);
1713         read_extent_buffer(leaf, fs_uuid,
1714                            (unsigned long)btrfs_device_fsid(dev_item),
1715                            BTRFS_UUID_SIZE);
1716
1717         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1718                 ret = open_seed_devices(root, fs_uuid);
1719                 if (ret)
1720                         return ret;
1721         }
1722
1723         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1724         if (!device) {
1725                 printk("warning devid %llu not found already\n",
1726                         (unsigned long long)devid);
1727                 device = kzalloc(sizeof(*device), GFP_NOFS);
1728                 if (!device)
1729                         return -ENOMEM;
1730                 device->fd = -1;
1731                 list_add(&device->dev_list,
1732                          &root->fs_info->fs_devices->devices);
1733         }
1734
1735         fill_device_from_item(leaf, dev_item, device);
1736         device->dev_root = root->fs_info->dev_root;
1737         return ret;
1738 }
1739
1740 int btrfs_read_sys_array(struct btrfs_root *root)
1741 {
1742         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1743         struct extent_buffer *sb;
1744         struct btrfs_disk_key *disk_key;
1745         struct btrfs_chunk *chunk;
1746         struct btrfs_key key;
1747         u32 num_stripes;
1748         u32 len = 0;
1749         u8 *ptr;
1750         u8 *array_end;
1751         int ret = 0;
1752
1753         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1754                                           BTRFS_SUPER_INFO_SIZE);
1755         if (!sb)
1756                 return -ENOMEM;
1757         btrfs_set_buffer_uptodate(sb);
1758         write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1759         array_end = ((u8 *)super_copy->sys_chunk_array) +
1760                     btrfs_super_sys_array_size(super_copy);
1761
1762         /*
1763          * we do this loop twice, once for the device items and
1764          * once for all of the chunks.  This way there are device
1765          * structs filled in for every chunk
1766          */
1767         ptr = super_copy->sys_chunk_array;
1768
1769         while (ptr < array_end) {
1770                 disk_key = (struct btrfs_disk_key *)ptr;
1771                 btrfs_disk_key_to_cpu(&key, disk_key);
1772
1773                 len = sizeof(*disk_key);
1774                 ptr += len;
1775
1776                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1777                         chunk = (struct btrfs_chunk *)(ptr - (u8 *)super_copy);
1778                         ret = read_one_chunk(root, &key, sb, chunk);
1779                         if (ret)
1780                                 break;
1781                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1782                         len = btrfs_chunk_item_size(num_stripes);
1783                 } else {
1784                         BUG();
1785                 }
1786                 ptr += len;
1787         }
1788         free_extent_buffer(sb);
1789         return ret;
1790 }
1791
1792 int btrfs_read_chunk_tree(struct btrfs_root *root)
1793 {
1794         struct btrfs_path *path;
1795         struct extent_buffer *leaf;
1796         struct btrfs_key key;
1797         struct btrfs_key found_key;
1798         int ret;
1799         int slot;
1800
1801         root = root->fs_info->chunk_root;
1802
1803         path = btrfs_alloc_path();
1804         if (!path)
1805                 return -ENOMEM;
1806
1807         /*
1808          * Read all device items, and then all the chunk items. All
1809          * device items are found before any chunk item (their object id
1810          * is smaller than the lowest possible object id for a chunk
1811          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1812          */
1813         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1814         key.offset = 0;
1815         key.type = 0;
1816         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1817         if (ret < 0)
1818                 goto error;
1819         while(1) {
1820                 leaf = path->nodes[0];
1821                 slot = path->slots[0];
1822                 if (slot >= btrfs_header_nritems(leaf)) {
1823                         ret = btrfs_next_leaf(root, path);
1824                         if (ret == 0)
1825                                 continue;
1826                         if (ret < 0)
1827                                 goto error;
1828                         break;
1829                 }
1830                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1831                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1832                         struct btrfs_dev_item *dev_item;
1833                         dev_item = btrfs_item_ptr(leaf, slot,
1834                                                   struct btrfs_dev_item);
1835                         ret = read_one_dev(root, leaf, dev_item);
1836                         BUG_ON(ret);
1837                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1838                         struct btrfs_chunk *chunk;
1839                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1840                         ret = read_one_chunk(root, &found_key, leaf, chunk);
1841                         BUG_ON(ret);
1842                 }
1843                 path->slots[0]++;
1844         }
1845
1846         ret = 0;
1847 error:
1848         btrfs_free_path(path);
1849         return ret;
1850 }
1851
1852 struct list_head *btrfs_scanned_uuids(void)
1853 {
1854         return &fs_uuids;
1855 }
1856
1857 static int rmw_eb(struct btrfs_fs_info *info,
1858                   struct extent_buffer *eb, struct extent_buffer *orig_eb)
1859 {
1860         int ret;
1861         unsigned long orig_off = 0;
1862         unsigned long dest_off = 0;
1863         unsigned long copy_len = eb->len;
1864
1865         ret = read_whole_eb(info, eb, 0);
1866         if (ret)
1867                 return ret;
1868
1869         if (eb->start + eb->len <= orig_eb->start ||
1870             eb->start >= orig_eb->start + orig_eb->len)
1871                 return 0;
1872         /*
1873          * | ----- orig_eb ------- |
1874          *         | ----- stripe -------  |
1875          *         | ----- orig_eb ------- |
1876          *              | ----- orig_eb ------- |
1877          */
1878         if (eb->start > orig_eb->start)
1879                 orig_off = eb->start - orig_eb->start;
1880         if (orig_eb->start > eb->start)
1881                 dest_off = orig_eb->start - eb->start;
1882
1883         if (copy_len > orig_eb->len - orig_off)
1884                 copy_len = orig_eb->len - orig_off;
1885         if (copy_len > eb->len - dest_off)
1886                 copy_len = eb->len - dest_off;
1887
1888         memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1889         return 0;
1890 }
1891
1892 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1893                                 struct extent_buffer *orig_eb,
1894                                struct extent_buffer **ebs,
1895                                u64 stripe_len, u64 *raid_map,
1896                                int num_stripes)
1897 {
1898         struct extent_buffer *eb;
1899         u64 start = orig_eb->start;
1900         u64 this_eb_start;
1901         int i;
1902         int ret;
1903
1904         for (i = 0; i < num_stripes; i++) {
1905                 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1906                         break;
1907
1908                 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1909                 if (!eb)
1910                         BUG();
1911                 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1912
1913                 eb->start = raid_map[i];
1914                 eb->len = stripe_len;
1915                 eb->refs = 1;
1916                 eb->flags = 0;
1917                 eb->fd = -1;
1918                 eb->dev_bytenr = (u64)-1;
1919
1920                 this_eb_start = raid_map[i];
1921
1922                 if (start > this_eb_start ||
1923                     start + orig_eb->len < this_eb_start + stripe_len) {
1924                         ret = rmw_eb(info, eb, orig_eb);
1925                         BUG_ON(ret);
1926                 } else {
1927                         memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1928                 }
1929                 ebs[i] = eb;
1930         }
1931 }
1932
1933 int write_raid56_with_parity(struct btrfs_fs_info *info,
1934                              struct extent_buffer *eb,
1935                              struct btrfs_multi_bio *multi,
1936                              u64 stripe_len, u64 *raid_map)
1937 {
1938         struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
1939         int i;
1940         int j;
1941         int ret;
1942         int alloc_size = eb->len;
1943
1944         ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
1945         BUG_ON(!ebs);
1946
1947         if (stripe_len > alloc_size)
1948                 alloc_size = stripe_len;
1949
1950         split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1951                             multi->num_stripes);
1952
1953         for (i = 0; i < multi->num_stripes; i++) {
1954                 struct extent_buffer *new_eb;
1955                 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1956                         ebs[i]->dev_bytenr = multi->stripes[i].physical;
1957                         ebs[i]->fd = multi->stripes[i].dev->fd;
1958                         multi->stripes[i].dev->total_ios++;
1959                         BUG_ON(ebs[i]->start != raid_map[i]);
1960                         continue;
1961                 }
1962                 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1963                 BUG_ON(!new_eb);
1964                 new_eb->dev_bytenr = multi->stripes[i].physical;
1965                 new_eb->fd = multi->stripes[i].dev->fd;
1966                 multi->stripes[i].dev->total_ios++;
1967                 new_eb->len = stripe_len;
1968
1969                 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1970                         p_eb = new_eb;
1971                 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1972                         q_eb = new_eb;
1973         }
1974         if (q_eb) {
1975                 void **pointers;
1976
1977                 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
1978                                    GFP_NOFS);
1979                 BUG_ON(!pointers);
1980
1981                 ebs[multi->num_stripes - 2] = p_eb;
1982                 ebs[multi->num_stripes - 1] = q_eb;
1983
1984                 for (i = 0; i < multi->num_stripes; i++)
1985                         pointers[i] = ebs[i]->data;
1986
1987                 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1988                 kfree(pointers);
1989         } else {
1990                 ebs[multi->num_stripes - 1] = p_eb;
1991                 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1992                 for (j = 1; j < multi->num_stripes - 1; j++) {
1993                         for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1994                                 *(unsigned long *)(p_eb->data + i) ^=
1995                                         *(unsigned long *)(ebs[j]->data + i);
1996                         }
1997                 }
1998         }
1999
2000         for (i = 0; i < multi->num_stripes; i++) {
2001                 ret = write_extent_to_disk(ebs[i]);
2002                 BUG_ON(ret);
2003                 if (ebs[i] != eb)
2004                         kfree(ebs[i]);
2005         }
2006
2007         kfree(ebs);
2008
2009         return 0;
2010 }