volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #define _XOPEN_SOURCE 600
  19 #define __USE_XOPEN2K
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <uuid/uuid.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "print-tree.h"
  31 #include "volumes.h"
  32
  33 struct stripe {
  34         struct btrfs_device *dev;
  35         u64 physical;
  36 };
  37
  38 static inline int nr_parity_stripes(struct map_lookup *map)
  39 {
  40         if (map->type & BTRFS_BLOCK_GROUP_RAID5)
  41                 return 1;
  42         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
  43                 return 2;
  44         else
  45                 return 0;
  46 }
  47
  48 static inline int nr_data_stripes(struct map_lookup *map)
  49 {
  50         return map->num_stripes - nr_parity_stripes(map);
  51 }
  52
  53 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
  54
  55 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  56                             (sizeof(struct btrfs_bio_stripe) * (n)))
  57
  58 static LIST_HEAD(fs_uuids);
  59
  60 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  61                                           u8 *uuid)
  62 {
  63         struct btrfs_device *dev;
  64         struct list_head *cur;
  65
  66         list_for_each(cur, head) {
  67                 dev = list_entry(cur, struct btrfs_device, dev_list);
  68                 if (dev->devid == devid &&
  69                     !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
  70                         return dev;
  71                 }
  72         }
  73         return NULL;
  74 }
  75
  76 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  77 {
  78         struct list_head *cur;
  79         struct btrfs_fs_devices *fs_devices;
  80
  81         list_for_each(cur, &fs_uuids) {
  82                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  83                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  84                         return fs_devices;
  85         }
  86         return NULL;
  87 }
  88
  89 static int device_list_add(const char *path,
  90                            struct btrfs_super_block *disk_super,
  91                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
  92 {
  93         struct btrfs_device *device;
  94         struct btrfs_fs_devices *fs_devices;
  95         u64 found_transid = btrfs_super_generation(disk_super);
  96
  97         fs_devices = find_fsid(disk_super->fsid);
  98         if (!fs_devices) {
  99                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 100                 if (!fs_devices)
 101                         return -ENOMEM;
 102                 INIT_LIST_HEAD(&fs_devices->devices);
 103                 list_add(&fs_devices->list, &fs_uuids);
 104                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 105                 fs_devices->latest_devid = devid;
 106                 fs_devices->latest_trans = found_transid;
 107                 fs_devices->lowest_devid = (u64)-1;
 108                 device = NULL;
 109         } else {
 110                 device = __find_device(&fs_devices->devices, devid,
 111                                        disk_super->dev_item.uuid);
 112         }
 113         if (!device) {
 114                 device = kzalloc(sizeof(*device), GFP_NOFS);
 115                 if (!device) {
 116                         /* we can safely leave the fs_devices entry around */
 117                         return -ENOMEM;
 118                 }
 119                 device->devid = devid;
 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                 device->total_devs = btrfs_super_num_devices(disk_super);
 129                 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
 130                 device->total_bytes =
 131                         btrfs_stack_device_total_bytes(&disk_super->dev_item);
 132                 device->bytes_used =
 133                         btrfs_stack_device_bytes_used(&disk_super->dev_item);
 134                 list_add(&device->dev_list, &fs_devices->devices);
 135                 device->fs_devices = fs_devices;
 136         } else if (!device->name || strcmp(device->name, path)) {
 137                 char *name = strdup(path);
 138                 if (!name)
 139                         return -ENOMEM;
 140                 kfree(device->name);
 141                 device->name = name;
 142         }
 143
 144
 145         if (found_transid > fs_devices->latest_trans) {
 146                 fs_devices->latest_devid = devid;
 147                 fs_devices->latest_trans = found_transid;
 148         }
 149         if (fs_devices->lowest_devid > devid) {
 150                 fs_devices->lowest_devid = devid;
 151         }
 152         *fs_devices_ret = fs_devices;
 153         return 0;
 154 }
 155
 156 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 157 {
 158         struct btrfs_fs_devices *seed_devices;
 159         struct list_head *cur;
 160         struct btrfs_device *device;
 161 again:
 162         list_for_each(cur, &fs_devices->devices) {
 163                 device = list_entry(cur, struct btrfs_device, dev_list);
 164                 close(device->fd);
 165                 device->fd = -1;
 166                 device->writeable = 0;
 167         }
 168
 169         seed_devices = fs_devices->seed;
 170         fs_devices->seed = NULL;
 171         if (seed_devices) {
 172                 fs_devices = seed_devices;
 173                 goto again;
 174         }
 175
 176         return 0;
 177 }
 178
 179 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
 180 {
 181         int fd;
 182         struct list_head *head = &fs_devices->devices;
 183         struct list_head *cur;
 184         struct btrfs_device *device;
 185         int ret;
 186
 187         list_for_each(cur, head) {
 188                 device = list_entry(cur, struct btrfs_device, dev_list);
 189                 if (!device->name) {
 190                         printk("no name for device %llu, skip it now\n", device->devid);
 191                         continue;
 192                 }
 193
 194                 fd = open(device->name, flags);
 195                 if (fd < 0) {
 196                         ret = -errno;
 197                         goto fail;
 198                 }
 199
 200                 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
 201                         fprintf(stderr, "Warning, could not drop caches\n");
 202
 203                 if (device->devid == fs_devices->latest_devid)
 204                         fs_devices->latest_bdev = fd;
 205                 if (device->devid == fs_devices->lowest_devid)
 206                         fs_devices->lowest_bdev = fd;
 207                 device->fd = fd;
 208                 if (flags == O_RDWR)
 209                         device->writeable = 1;
 210         }
 211         return 0;
 212 fail:
 213         btrfs_close_devices(fs_devices);
 214         return ret;
 215 }
 216
 217 int btrfs_scan_one_device(int fd, const char *path,
 218                           struct btrfs_fs_devices **fs_devices_ret,
 219                           u64 *total_devs, u64 super_offset)
 220 {
 221         struct btrfs_super_block *disk_super;
 222         char *buf;
 223         int ret;
 224         u64 devid;
 225         char uuidbuf[37];
 226
 227         buf = malloc(4096);
 228         if (!buf) {
 229                 ret = -ENOMEM;
 230                 goto error;
 231         }
 232         disk_super = (struct btrfs_super_block *)buf;
 233         ret = btrfs_read_dev_super(fd, disk_super, super_offset);
 234         if (ret < 0) {
 235                 ret = -EIO;
 236                 goto error_brelse;
 237         }
 238         devid = le64_to_cpu(disk_super->dev_item.devid);
 239         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
 240                 *total_devs = 1;
 241         else
 242                 *total_devs = btrfs_super_num_devices(disk_super);
 243         uuid_unparse(disk_super->fsid, uuidbuf);
 244
 245         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 246
 247 error_brelse:
 248         free(buf);
 249 error:
 250         return ret;
 251 }
 252
 253 /*
 254  * this uses a pretty simple search, the expectation is that it is
 255  * called very infrequently and that a given device has a small number
 256  * of extents
 257  */
 258 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 259                                 struct btrfs_device *device,
 260                                 struct btrfs_path *path,
 261                                 u64 num_bytes, u64 *start)
 262 {
 263         struct btrfs_key key;
 264         struct btrfs_root *root = device->dev_root;
 265         struct btrfs_dev_extent *dev_extent = NULL;
 266         u64 hole_size = 0;
 267         u64 last_byte = 0;
 268         u64 search_start = 0;
 269         u64 search_end = device->total_bytes;
 270         int ret;
 271         int slot = 0;
 272         int start_found;
 273         struct extent_buffer *l;
 274
 275         start_found = 0;
 276         path->reada = 2;
 277
 278         /* FIXME use last free of some kind */
 279
 280         /* we don't want to overwrite the superblock on the drive,
 281          * so we make sure to start at an offset of at least 1MB
 282          */
 283         search_start = max((u64)1024 * 1024, search_start);
 284
 285         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 286                 search_start = max(root->fs_info->alloc_start, search_start);
 287
 288         key.objectid = device->devid;
 289         key.offset = search_start;
 290         key.type = BTRFS_DEV_EXTENT_KEY;
 291         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 292         if (ret < 0)
 293                 goto error;
 294         ret = btrfs_previous_item(root, path, 0, key.type);
 295         if (ret < 0)
 296                 goto error;
 297         l = path->nodes[0];
 298         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 299         while (1) {
 300                 l = path->nodes[0];
 301                 slot = path->slots[0];
 302                 if (slot >= btrfs_header_nritems(l)) {
 303                         ret = btrfs_next_leaf(root, path);
 304                         if (ret == 0)
 305                                 continue;
 306                         if (ret < 0)
 307                                 goto error;
 308 no_more_items:
 309                         if (!start_found) {
 310                                 if (search_start >= search_end) {
 311                                         ret = -ENOSPC;
 312                                         goto error;
 313                                 }
 314                                 *start = search_start;
 315                                 start_found = 1;
 316                                 goto check_pending;
 317                         }
 318                         *start = last_byte > search_start ?
 319                                 last_byte : search_start;
 320                         if (search_end <= *start) {
 321                                 ret = -ENOSPC;
 322                                 goto error;
 323                         }
 324                         goto check_pending;
 325                 }
 326                 btrfs_item_key_to_cpu(l, &key, slot);
 327
 328                 if (key.objectid < device->devid)
 329                         goto next;
 330
 331                 if (key.objectid > device->devid)
 332                         goto no_more_items;
 333
 334                 if (key.offset >= search_start && key.offset > last_byte &&
 335                     start_found) {
 336                         if (last_byte < search_start)
 337                                 last_byte = search_start;
 338                         hole_size = key.offset - last_byte;
 339                         if (key.offset > last_byte &&
 340                             hole_size >= num_bytes) {
 341                                 *start = last_byte;
 342                                 goto check_pending;
 343                         }
 344                 }
 345                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 346                         goto next;
 347                 }
 348
 349                 start_found = 1;
 350                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 351                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 352 next:
 353                 path->slots[0]++;
 354                 cond_resched();
 355         }
 356 check_pending:
 357         /* we have to make sure we didn't find an extent that has already
 358          * been allocated by the map tree or the original allocation
 359          */
 360         btrfs_release_path(root, path);
 361         BUG_ON(*start < search_start);
 362
 363         if (*start + num_bytes > search_end) {
 364                 ret = -ENOSPC;
 365                 goto error;
 366         }
 367         /* check for pending inserts here */
 368         return 0;
 369
 370 error:
 371         btrfs_release_path(root, path);
 372         return ret;
 373 }
 374
 375 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 376                            struct btrfs_device *device,
 377                            u64 chunk_tree, u64 chunk_objectid,
 378                            u64 chunk_offset,
 379                            u64 num_bytes, u64 *start)
 380 {
 381         int ret;
 382         struct btrfs_path *path;
 383         struct btrfs_root *root = device->dev_root;
 384         struct btrfs_dev_extent *extent;
 385         struct extent_buffer *leaf;
 386         struct btrfs_key key;
 387
 388         path = btrfs_alloc_path();
 389         if (!path)
 390                 return -ENOMEM;
 391
 392         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 393         if (ret) {
 394                 goto err;
 395         }
 396
 397         key.objectid = device->devid;
 398         key.offset = *start;
 399         key.type = BTRFS_DEV_EXTENT_KEY;
 400         ret = btrfs_insert_empty_item(trans, root, path, &key,
 401                                       sizeof(*extent));
 402         BUG_ON(ret);
 403
 404         leaf = path->nodes[0];
 405         extent = btrfs_item_ptr(leaf, path->slots[0],
 406                                 struct btrfs_dev_extent);
 407         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 408         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 409         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 410
 411         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 412                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 413                     BTRFS_UUID_SIZE);
 414
 415         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 416         btrfs_mark_buffer_dirty(leaf);
 417 err:
 418         btrfs_free_path(path);
 419         return ret;
 420 }
 421
 422 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 423 {
 424         struct btrfs_path *path;
 425         int ret;
 426         struct btrfs_key key;
 427         struct btrfs_chunk *chunk;
 428         struct btrfs_key found_key;
 429
 430         path = btrfs_alloc_path();
 431         BUG_ON(!path);
 432
 433         key.objectid = objectid;
 434         key.offset = (u64)-1;
 435         key.type = BTRFS_CHUNK_ITEM_KEY;
 436
 437         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 438         if (ret < 0)
 439                 goto error;
 440
 441         BUG_ON(ret == 0);
 442
 443         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 444         if (ret) {
 445                 *offset = 0;
 446         } else {
 447                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 448                                       path->slots[0]);
 449                 if (found_key.objectid != objectid)
 450                         *offset = 0;
 451                 else {
 452                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 453                                                struct btrfs_chunk);
 454                         *offset = found_key.offset +
 455                                 btrfs_chunk_length(path->nodes[0], chunk);
 456                 }
 457         }
 458         ret = 0;
 459 error:
 460         btrfs_free_path(path);
 461         return ret;
 462 }
 463
 464 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 465                            u64 *objectid)
 466 {
 467         int ret;
 468         struct btrfs_key key;
 469         struct btrfs_key found_key;
 470
 471         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 472         key.type = BTRFS_DEV_ITEM_KEY;
 473         key.offset = (u64)-1;
 474
 475         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 476         if (ret < 0)
 477                 goto error;
 478
 479         BUG_ON(ret == 0);
 480
 481         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 482                                   BTRFS_DEV_ITEM_KEY);
 483         if (ret) {
 484                 *objectid = 1;
 485         } else {
 486                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 487                                       path->slots[0]);
 488                 *objectid = found_key.offset + 1;
 489         }
 490         ret = 0;
 491 error:
 492         btrfs_release_path(root, path);
 493         return ret;
 494 }
 495
 496 /*
 497  * the device information is stored in the chunk root
 498  * the btrfs_device struct should be fully filled in
 499  */
 500 int btrfs_add_device(struct btrfs_trans_handle *trans,
 501                      struct btrfs_root *root,
 502                      struct btrfs_device *device)
 503 {
 504         int ret;
 505         struct btrfs_path *path;
 506         struct btrfs_dev_item *dev_item;
 507         struct extent_buffer *leaf;
 508         struct btrfs_key key;
 509         unsigned long ptr;
 510         u64 free_devid = 0;
 511
 512         root = root->fs_info->chunk_root;
 513
 514         path = btrfs_alloc_path();
 515         if (!path)
 516                 return -ENOMEM;
 517
 518         ret = find_next_devid(root, path, &free_devid);
 519         if (ret)
 520                 goto out;
 521
 522         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 523         key.type = BTRFS_DEV_ITEM_KEY;
 524         key.offset = free_devid;
 525
 526         ret = btrfs_insert_empty_item(trans, root, path, &key,
 527                                       sizeof(*dev_item));
 528         if (ret)
 529                 goto out;
 530
 531         leaf = path->nodes[0];
 532         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 533
 534         device->devid = free_devid;
 535         btrfs_set_device_id(leaf, dev_item, device->devid);
 536         btrfs_set_device_generation(leaf, dev_item, 0);
 537         btrfs_set_device_type(leaf, dev_item, device->type);
 538         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 539         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 540         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 541         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 542         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 543         btrfs_set_device_group(leaf, dev_item, 0);
 544         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 545         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 546         btrfs_set_device_start_offset(leaf, dev_item, 0);
 547
 548         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 549         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 550         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 551         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 552         btrfs_mark_buffer_dirty(leaf);
 553         ret = 0;
 554
 555 out:
 556         btrfs_free_path(path);
 557         return ret;
 558 }
 559
 560 int btrfs_update_device(struct btrfs_trans_handle *trans,
 561                         struct btrfs_device *device)
 562 {
 563         int ret;
 564         struct btrfs_path *path;
 565         struct btrfs_root *root;
 566         struct btrfs_dev_item *dev_item;
 567         struct extent_buffer *leaf;
 568         struct btrfs_key key;
 569
 570         root = device->dev_root->fs_info->chunk_root;
 571
 572         path = btrfs_alloc_path();
 573         if (!path)
 574                 return -ENOMEM;
 575
 576         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 577         key.type = BTRFS_DEV_ITEM_KEY;
 578         key.offset = device->devid;
 579
 580         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 581         if (ret < 0)
 582                 goto out;
 583
 584         if (ret > 0) {
 585                 ret = -ENOENT;
 586                 goto out;
 587         }
 588
 589         leaf = path->nodes[0];
 590         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 591
 592         btrfs_set_device_id(leaf, dev_item, device->devid);
 593         btrfs_set_device_type(leaf, dev_item, device->type);
 594         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 595         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 596         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 597         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 598         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 599         btrfs_mark_buffer_dirty(leaf);
 600
 601 out:
 602         btrfs_free_path(path);
 603         return ret;
 604 }
 605
 606 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 607                            struct btrfs_root *root,
 608                            struct btrfs_key *key,
 609                            struct btrfs_chunk *chunk, int item_size)
 610 {
 611         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
 612         struct btrfs_disk_key disk_key;
 613         u32 array_size;
 614         u8 *ptr;
 615
 616         array_size = btrfs_super_sys_array_size(super_copy);
 617         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 618                 return -EFBIG;
 619
 620         ptr = super_copy->sys_chunk_array + array_size;
 621         btrfs_cpu_key_to_disk(&disk_key, key);
 622         memcpy(ptr, &disk_key, sizeof(disk_key));
 623         ptr += sizeof(disk_key);
 624         memcpy(ptr, chunk, item_size);
 625         item_size += sizeof(disk_key);
 626         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 627         return 0;
 628 }
 629
 630 static u64 div_factor(u64 num, int factor)
 631 {
 632         if (factor == 10)
 633                 return num;
 634         num *= factor;
 635         return num / 10;
 636 }
 637
 638 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 639                                int sub_stripes)
 640 {
 641         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 642                 return calc_size;
 643         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 644                 return calc_size * (num_stripes / sub_stripes);
 645         else if (type & BTRFS_BLOCK_GROUP_RAID5)
 646                 return calc_size * (num_stripes - 1);
 647         else if (type & BTRFS_BLOCK_GROUP_RAID6)
 648                 return calc_size * (num_stripes - 2);
 649         else
 650                 return calc_size * num_stripes;
 651 }
 652
 653
 654 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
 655 {
 656         /* TODO, add a way to store the preferred stripe size */
 657         return 64 * 1024;
 658 }
 659
 660 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 661                       struct btrfs_root *extent_root, u64 *start,
 662                       u64 *num_bytes, u64 type)
 663 {
 664         u64 dev_offset;
 665         struct btrfs_fs_info *info = extent_root->fs_info;
 666         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
 667         struct btrfs_stripe *stripes;
 668         struct btrfs_device *device = NULL;
 669         struct btrfs_chunk *chunk;
 670         struct list_head private_devs;
 671         struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
 672         struct list_head *cur;
 673         struct map_lookup *map;
 674         int min_stripe_size = 1 * 1024 * 1024;
 675         u64 calc_size = 8 * 1024 * 1024;
 676         u64 min_free;
 677         u64 max_chunk_size = 4 * calc_size;
 678         u64 avail;
 679         u64 max_avail = 0;
 680         u64 percent_max;
 681         int num_stripes = 1;
 682         int min_stripes = 1;
 683         int sub_stripes = 0;
 684         int looped = 0;
 685         int ret;
 686         int index;
 687         int stripe_len = 64 * 1024;
 688         struct btrfs_key key;
 689         u64 offset;
 690
 691         if (list_empty(dev_list)) {
 692                 return -ENOSPC;
 693         }
 694
 695         if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
 696                     BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
 697                     BTRFS_BLOCK_GROUP_RAID10 |
 698                     BTRFS_BLOCK_GROUP_DUP)) {
 699                 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 700                         calc_size = 8 * 1024 * 1024;
 701                         max_chunk_size = calc_size * 2;
 702                         min_stripe_size = 1 * 1024 * 1024;
 703                 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
 704                         calc_size = 1024 * 1024 * 1024;
 705                         max_chunk_size = 10 * calc_size;
 706                         min_stripe_size = 64 * 1024 * 1024;
 707                 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 708                         calc_size = 1024 * 1024 * 1024;
 709                         max_chunk_size = 4 * calc_size;
 710                         min_stripe_size = 32 * 1024 * 1024;
 711                 }
 712         }
 713         if (type & BTRFS_BLOCK_GROUP_RAID1) {
 714                 num_stripes = min_t(u64, 2,
 715                                   btrfs_super_num_devices(info->super_copy));
 716                 if (num_stripes < 2)
 717                         return -ENOSPC;
 718                 min_stripes = 2;
 719         }
 720         if (type & BTRFS_BLOCK_GROUP_DUP) {
 721                 num_stripes = 2;
 722                 min_stripes = 2;
 723         }
 724         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
 725                 num_stripes = btrfs_super_num_devices(info->super_copy);
 726                 min_stripes = 2;
 727         }
 728         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 729                 num_stripes = btrfs_super_num_devices(info->super_copy);
 730                 if (num_stripes < 4)
 731                         return -ENOSPC;
 732                 num_stripes &= ~(u32)1;
 733                 sub_stripes = 2;
 734                 min_stripes = 4;
 735         }
 736         if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
 737                 num_stripes = btrfs_super_num_devices(info->super_copy);
 738                 if (num_stripes < 2)
 739                         return -ENOSPC;
 740                 min_stripes = 2;
 741                 stripe_len = find_raid56_stripe_len(num_stripes - 1,
 742                                     btrfs_super_stripesize(info->super_copy));
 743         }
 744         if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
 745                 num_stripes = btrfs_super_num_devices(info->super_copy);
 746                 if (num_stripes < 3)
 747                         return -ENOSPC;
 748                 min_stripes = 3;
 749                 stripe_len = find_raid56_stripe_len(num_stripes - 2,
 750                                     btrfs_super_stripesize(info->super_copy));
 751         }
 752
 753         /* we don't want a chunk larger than 10% of the FS */
 754         percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
 755         max_chunk_size = min(percent_max, max_chunk_size);
 756
 757 again:
 758         if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
 759             max_chunk_size) {
 760                 calc_size = max_chunk_size;
 761                 calc_size /= num_stripes;
 762                 calc_size /= stripe_len;
 763                 calc_size *= stripe_len;
 764         }
 765         /* we don't want tiny stripes */
 766         calc_size = max_t(u64, calc_size, min_stripe_size);
 767
 768         calc_size /= stripe_len;
 769         calc_size *= stripe_len;
 770         INIT_LIST_HEAD(&private_devs);
 771         cur = dev_list->next;
 772         index = 0;
 773
 774         if (type & BTRFS_BLOCK_GROUP_DUP)
 775                 min_free = calc_size * 2;
 776         else
 777                 min_free = calc_size;
 778
 779         /* build a private list of devices we will allocate from */
 780         while(index < num_stripes) {
 781                 device = list_entry(cur, struct btrfs_device, dev_list);
 782                 avail = device->total_bytes - device->bytes_used;
 783                 cur = cur->next;
 784                 if (avail >= min_free) {
 785                         list_move_tail(&device->dev_list, &private_devs);
 786                         index++;
 787                         if (type & BTRFS_BLOCK_GROUP_DUP)
 788                                 index++;
 789                 } else if (avail > max_avail)
 790                         max_avail = avail;
 791                 if (cur == dev_list)
 792                         break;
 793         }
 794         if (index < num_stripes) {
 795                 list_splice(&private_devs, dev_list);
 796                 if (index >= min_stripes) {
 797                         num_stripes = index;
 798                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 799                                 num_stripes /= sub_stripes;
 800                                 num_stripes *= sub_stripes;
 801                         }
 802                         looped = 1;
 803                         goto again;
 804                 }
 805                 if (!looped && max_avail > 0) {
 806                         looped = 1;
 807                         calc_size = max_avail;
 808                         goto again;
 809                 }
 810                 return -ENOSPC;
 811         }
 812         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 813                               &offset);
 814         if (ret)
 815                 return ret;
 816         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 817         key.type = BTRFS_CHUNK_ITEM_KEY;
 818         key.offset = offset;
 819
 820         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 821         if (!chunk)
 822                 return -ENOMEM;
 823
 824         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
 825         if (!map) {
 826                 kfree(chunk);
 827                 return -ENOMEM;
 828         }
 829
 830         stripes = &chunk->stripe;
 831         *num_bytes = chunk_bytes_by_type(type, calc_size,
 832                                          num_stripes, sub_stripes);
 833         index = 0;
 834         while(index < num_stripes) {
 835                 struct btrfs_stripe *stripe;
 836                 BUG_ON(list_empty(&private_devs));
 837                 cur = private_devs.next;
 838                 device = list_entry(cur, struct btrfs_device, dev_list);
 839
 840                 /* loop over this device again if we're doing a dup group */
 841                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 842                     (index == num_stripes - 1))
 843                         list_move_tail(&device->dev_list, dev_list);
 844
 845                 ret = btrfs_alloc_dev_extent(trans, device,
 846                              info->chunk_root->root_key.objectid,
 847                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 848                              calc_size, &dev_offset);
 849                 BUG_ON(ret);
 850
 851                 device->bytes_used += calc_size;
 852                 ret = btrfs_update_device(trans, device);
 853                 BUG_ON(ret);
 854
 855                 map->stripes[index].dev = device;
 856                 map->stripes[index].physical = dev_offset;
 857                 stripe = stripes + index;
 858                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 859                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 860                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 861                 index++;
 862         }
 863         BUG_ON(!list_empty(&private_devs));
 864
 865         /* key was set above */
 866         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 867         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 868         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 869         btrfs_set_stack_chunk_type(chunk, type);
 870         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 871         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 872         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 873         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 874         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
 875         map->sector_size = extent_root->sectorsize;
 876         map->stripe_len = stripe_len;
 877         map->io_align = stripe_len;
 878         map->io_width = stripe_len;
 879         map->type = type;
 880         map->num_stripes = num_stripes;
 881         map->sub_stripes = sub_stripes;
 882
 883         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 884                                 btrfs_chunk_item_size(num_stripes));
 885         BUG_ON(ret);
 886         *start = key.offset;;
 887
 888         map->ce.start = key.offset;
 889         map->ce.size = *num_bytes;
 890
 891         ret = insert_existing_cache_extent(
 892                            &extent_root->fs_info->mapping_tree.cache_tree,
 893                            &map->ce);
 894         BUG_ON(ret);
 895
 896         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 897                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
 898                                     chunk, btrfs_chunk_item_size(num_stripes));
 899                 BUG_ON(ret);
 900         }
 901
 902         kfree(chunk);
 903         return ret;
 904 }
 905
 906 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
 907                            struct btrfs_root *extent_root, u64 *start,
 908                            u64 num_bytes, u64 type)
 909 {
 910         u64 dev_offset;
 911         struct btrfs_fs_info *info = extent_root->fs_info;
 912         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
 913         struct btrfs_stripe *stripes;
 914         struct btrfs_device *device = NULL;
 915         struct btrfs_chunk *chunk;
 916         struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
 917         struct list_head *cur;
 918         struct map_lookup *map;
 919         u64 calc_size = 8 * 1024 * 1024;
 920         int num_stripes = 1;
 921         int sub_stripes = 0;
 922         int ret;
 923         int index;
 924         int stripe_len = 64 * 1024;
 925         struct btrfs_key key;
 926
 927         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 928         key.type = BTRFS_CHUNK_ITEM_KEY;
 929         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 930                               &key.offset);
 931         if (ret)
 932                 return ret;
 933
 934         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 935         if (!chunk)
 936                 return -ENOMEM;
 937
 938         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
 939         if (!map) {
 940                 kfree(chunk);
 941                 return -ENOMEM;
 942         }
 943
 944         stripes = &chunk->stripe;
 945         calc_size = num_bytes;
 946
 947         index = 0;
 948         cur = dev_list->next;
 949         device = list_entry(cur, struct btrfs_device, dev_list);
 950
 951         while (index < num_stripes) {
 952                 struct btrfs_stripe *stripe;
 953
 954                 ret = btrfs_alloc_dev_extent(trans, device,
 955                              info->chunk_root->root_key.objectid,
 956                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 957                              calc_size, &dev_offset);
 958                 BUG_ON(ret);
 959
 960                 device->bytes_used += calc_size;
 961                 ret = btrfs_update_device(trans, device);
 962                 BUG_ON(ret);
 963
 964                 map->stripes[index].dev = device;
 965                 map->stripes[index].physical = dev_offset;
 966                 stripe = stripes + index;
 967                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 968                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 969                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 970                 index++;
 971         }
 972
 973         /* key was set above */
 974         btrfs_set_stack_chunk_length(chunk, num_bytes);
 975         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 976         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 977         btrfs_set_stack_chunk_type(chunk, type);
 978         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 979         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 980         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 981         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 982         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
 983         map->sector_size = extent_root->sectorsize;
 984         map->stripe_len = stripe_len;
 985         map->io_align = stripe_len;
 986         map->io_width = stripe_len;
 987         map->type = type;
 988         map->num_stripes = num_stripes;
 989         map->sub_stripes = sub_stripes;
 990
 991         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 992                                 btrfs_chunk_item_size(num_stripes));
 993         BUG_ON(ret);
 994         *start = key.offset;
 995
 996         map->ce.start = key.offset;
 997         map->ce.size = num_bytes;
 998
 999         ret = insert_existing_cache_extent(
1000                            &extent_root->fs_info->mapping_tree.cache_tree,
1001                            &map->ce);
1002         BUG_ON(ret);
1003
1004         kfree(chunk);
1005         return ret;
1006 }
1007
1008 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1009 {
1010         cache_tree_init(&tree->cache_tree);
1011 }
1012
1013 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1014 {
1015         struct cache_extent *ce;
1016         struct map_lookup *map;
1017         int ret;
1018
1019         ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1020         BUG_ON(!ce);
1021         BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1022         map = container_of(ce, struct map_lookup, ce);
1023
1024         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1025                 ret = map->num_stripes;
1026         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1027                 ret = map->sub_stripes;
1028         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1029                 ret = 2;
1030         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1031                 ret = 3;
1032         else
1033                 ret = 1;
1034         return ret;
1035 }
1036
1037 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1038                         u64 *size)
1039 {
1040         struct cache_extent *ce;
1041         struct map_lookup *map;
1042
1043         ce = find_first_cache_extent(&map_tree->cache_tree, *logical);
1044
1045         while (ce) {
1046                 ce = next_cache_extent(ce);
1047                 if (!ce)
1048                         return -ENOENT;
1049
1050                 map = container_of(ce, struct map_lookup, ce);
1051                 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1052                         *logical = ce->start;
1053                         *size = ce->size;
1054                         return 0;
1055                 }
1056         }
1057
1058         return -ENOENT;
1059 }
1060
1061 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1062                      u64 chunk_start, u64 physical, u64 devid,
1063                      u64 **logical, int *naddrs, int *stripe_len)
1064 {
1065         struct cache_extent *ce;
1066         struct map_lookup *map;
1067         u64 *buf;
1068         u64 bytenr;
1069         u64 length;
1070         u64 stripe_nr;
1071         u64 rmap_len;
1072         int i, j, nr = 0;
1073
1074         ce = find_first_cache_extent(&map_tree->cache_tree, chunk_start);
1075         BUG_ON(!ce);
1076         map = container_of(ce, struct map_lookup, ce);
1077
1078         length = ce->size;
1079         rmap_len = map->stripe_len;
1080         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1081                 length = ce->size / (map->num_stripes / map->sub_stripes);
1082         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1083                 length = ce->size / map->num_stripes;
1084         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1085                               BTRFS_BLOCK_GROUP_RAID6)) {
1086                 length = ce->size / nr_data_stripes(map);
1087                 rmap_len = map->stripe_len * nr_data_stripes(map);
1088         }
1089
1090         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1091
1092         for (i = 0; i < map->num_stripes; i++) {
1093                 if (devid && map->stripes[i].dev->devid != devid)
1094                         continue;
1095                 if (map->stripes[i].physical > physical ||
1096                     map->stripes[i].physical + length <= physical)
1097                         continue;
1098
1099                 stripe_nr = (physical - map->stripes[i].physical) /
1100                             map->stripe_len;
1101
1102                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1103                         stripe_nr = (stripe_nr * map->num_stripes + i) /
1104                                     map->sub_stripes;
1105                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1106                         stripe_nr = stripe_nr * map->num_stripes + i;
1107                 } /* else if RAID[56], multiply by nr_data_stripes().
1108                    * Alternatively, just use rmap_len below instead of
1109                    * map->stripe_len */
1110
1111                 bytenr = ce->start + stripe_nr * rmap_len;
1112                 for (j = 0; j < nr; j++) {
1113                         if (buf[j] == bytenr)
1114                                 break;
1115                 }
1116                 if (j == nr)
1117                         buf[nr++] = bytenr;
1118         }
1119
1120         *logical = buf;
1121         *naddrs = nr;
1122         *stripe_len = rmap_len;
1123
1124         return 0;
1125 }
1126
1127 static inline int parity_smaller(u64 a, u64 b)
1128 {
1129         return a > b;
1130 }
1131
1132 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1133 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1134 {
1135         struct btrfs_bio_stripe s;
1136         int i;
1137         u64 l;
1138         int again = 1;
1139
1140         while (again) {
1141                 again = 0;
1142                 for (i = 0; i < bbio->num_stripes - 1; i++) {
1143                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
1144                                 s = bbio->stripes[i];
1145                                 l = raid_map[i];
1146                                 bbio->stripes[i] = bbio->stripes[i+1];
1147                                 raid_map[i] = raid_map[i+1];
1148                                 bbio->stripes[i+1] = s;
1149                                 raid_map[i+1] = l;
1150                                 again = 1;
1151                         }
1152                 }
1153         }
1154 }
1155
1156 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1157                     u64 logical, u64 *length,
1158                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1159                     u64 **raid_map_ret)
1160 {
1161         return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1162                                  multi_ret, mirror_num, raid_map_ret);
1163 }
1164
1165 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1166                     u64 logical, u64 *length, u64 *type,
1167                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1168                     u64 **raid_map_ret)
1169 {
1170         struct cache_extent *ce;
1171         struct map_lookup *map;
1172         u64 offset;
1173         u64 stripe_offset;
1174         u64 stripe_nr;
1175         u64 *raid_map = NULL;
1176         int stripes_allocated = 8;
1177         int stripes_required = 1;
1178         int stripe_index;
1179         int i;
1180         struct btrfs_multi_bio *multi = NULL;
1181
1182         if (multi_ret && rw == READ) {
1183                 stripes_allocated = 1;
1184         }
1185 again:
1186         ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1187         if (!ce) {
1188                 if (multi)
1189                         kfree(multi);
1190                 return -ENOENT;
1191         }
1192         if (ce->start > logical || ce->start + ce->size < logical) {
1193                 if (multi)
1194                         kfree(multi);
1195                 return -ENOENT;
1196         }
1197
1198         if (multi_ret) {
1199                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1200                                 GFP_NOFS);
1201                 if (!multi)
1202                         return -ENOMEM;
1203         }
1204         map = container_of(ce, struct map_lookup, ce);
1205         offset = logical - ce->start;
1206
1207         if (rw == WRITE) {
1208                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1209                                  BTRFS_BLOCK_GROUP_DUP)) {
1210                         stripes_required = map->num_stripes;
1211                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1212                         stripes_required = map->sub_stripes;
1213                 }
1214         }
1215         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1216             && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1217                     /* RAID[56] write or recovery. Return all stripes */
1218                     stripes_required = map->num_stripes;
1219
1220                     /* Only allocate the map if we've already got a large enough multi_ret */
1221                     if (stripes_allocated >= stripes_required) {
1222                             raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1223                             if (!raid_map) {
1224                                     kfree(multi);
1225                                     return -ENOMEM;
1226                             }
1227                     }
1228         }
1229
1230         /* if our multi bio struct is too small, back off and try again */
1231         if (multi_ret && stripes_allocated < stripes_required) {
1232                 stripes_allocated = stripes_required;
1233                 kfree(multi);
1234                 multi = NULL;
1235                 goto again;
1236         }
1237         stripe_nr = offset;
1238         /*
1239          * stripe_nr counts the total number of stripes we have to stride
1240          * to get to this block
1241          */
1242         stripe_nr = stripe_nr / map->stripe_len;
1243
1244         stripe_offset = stripe_nr * map->stripe_len;
1245         BUG_ON(offset < stripe_offset);
1246
1247         /* stripe_offset is the offset of this block in its stripe*/
1248         stripe_offset = offset - stripe_offset;
1249
1250         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1251                          BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1252                          BTRFS_BLOCK_GROUP_RAID10 |
1253                          BTRFS_BLOCK_GROUP_DUP)) {
1254                 /* we limit the length of each bio to what fits in a stripe */
1255                 *length = min_t(u64, ce->size - offset,
1256                               map->stripe_len - stripe_offset);
1257         } else {
1258                 *length = ce->size - offset;
1259         }
1260
1261         if (!multi_ret)
1262                 goto out;
1263
1264         multi->num_stripes = 1;
1265         stripe_index = 0;
1266         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1267                 if (rw == WRITE)
1268                         multi->num_stripes = map->num_stripes;
1269                 else if (mirror_num)
1270                         stripe_index = mirror_num - 1;
1271                 else
1272                         stripe_index = stripe_nr % map->num_stripes;
1273         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1274                 int factor = map->num_stripes / map->sub_stripes;
1275
1276                 stripe_index = stripe_nr % factor;
1277                 stripe_index *= map->sub_stripes;
1278
1279                 if (rw == WRITE)
1280                         multi->num_stripes = map->sub_stripes;
1281                 else if (mirror_num)
1282                         stripe_index += mirror_num - 1;
1283
1284                 stripe_nr = stripe_nr / factor;
1285         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1286                 if (rw == WRITE)
1287                         multi->num_stripes = map->num_stripes;
1288                 else if (mirror_num)
1289                         stripe_index = mirror_num - 1;
1290         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1291                                 BTRFS_BLOCK_GROUP_RAID6)) {
1292
1293                 if (raid_map) {
1294                         int i, rot;
1295                         u64 tmp;
1296                         u64 raid56_full_stripe_start;
1297                         u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1298
1299                         /*
1300                          * align the start of our data stripe in the logical
1301                          * address space
1302                          */
1303                         raid56_full_stripe_start = offset / full_stripe_len;
1304                         raid56_full_stripe_start *= full_stripe_len;
1305
1306                         /* get the data stripe number */
1307                         stripe_nr = raid56_full_stripe_start / map->stripe_len;
1308                         stripe_nr = stripe_nr / nr_data_stripes(map);
1309
1310                         /* Work out the disk rotation on this stripe-set */
1311                         rot = stripe_nr % map->num_stripes;
1312
1313                         /* Fill in the logical address of each stripe */
1314                         tmp = stripe_nr * nr_data_stripes(map);
1315
1316                         for (i = 0; i < nr_data_stripes(map); i++)
1317                                 raid_map[(i+rot) % map->num_stripes] =
1318                                         ce->start + (tmp + i) * map->stripe_len;
1319
1320                         raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1321                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1322                                 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1323
1324                         *length = map->stripe_len;
1325                         stripe_index = 0;
1326                         stripe_offset = 0;
1327                         multi->num_stripes = map->num_stripes;
1328                 } else {
1329                         stripe_index = stripe_nr % nr_data_stripes(map);
1330                         stripe_nr = stripe_nr / nr_data_stripes(map);
1331
1332                         /*
1333                          * Mirror #0 or #1 means the original data block.
1334                          * Mirror #2 is RAID5 parity block.
1335                          * Mirror #3 is RAID6 Q block.
1336                          */
1337                         if (mirror_num > 1)
1338                                 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1339
1340                         /* We distribute the parity blocks across stripes */
1341                         stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1342                 }
1343         } else {
1344                 /*
1345                  * after this do_div call, stripe_nr is the number of stripes
1346                  * on this device we have to walk to find the data, and
1347                  * stripe_index is the number of our device in the stripe array
1348                  */
1349                 stripe_index = stripe_nr % map->num_stripes;
1350                 stripe_nr = stripe_nr / map->num_stripes;
1351         }
1352         BUG_ON(stripe_index >= map->num_stripes);
1353
1354         for (i = 0; i < multi->num_stripes; i++) {
1355                 multi->stripes[i].physical =
1356                         map->stripes[stripe_index].physical + stripe_offset +
1357                         stripe_nr * map->stripe_len;
1358                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1359                 stripe_index++;
1360         }
1361         *multi_ret = multi;
1362
1363         if (type)
1364                 *type = map->type;
1365
1366         if (raid_map) {
1367                 sort_parity_stripes(multi, raid_map);
1368                 *raid_map_ret = raid_map;
1369         }
1370 out:
1371         return 0;
1372 }
1373
1374 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1375                                        u8 *uuid, u8 *fsid)
1376 {
1377         struct btrfs_device *device;
1378         struct btrfs_fs_devices *cur_devices;
1379
1380         cur_devices = root->fs_info->fs_devices;
1381         while (cur_devices) {
1382                 if (!fsid ||
1383                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1384                         device = __find_device(&cur_devices->devices,
1385                                                devid, uuid);
1386                         if (device)
1387                                 return device;
1388                 }
1389                 cur_devices = cur_devices->seed;
1390         }
1391         return NULL;
1392 }
1393
1394 struct btrfs_device *btrfs_find_device_by_devid(struct btrfs_root *root,
1395                                                 u64 devid, int instance)
1396 {
1397         struct list_head *head = &root->fs_info->fs_devices->devices;
1398         struct btrfs_device *dev;
1399         struct list_head *cur;
1400         int num_found = 0;
1401
1402         list_for_each(cur, head) {
1403                 dev = list_entry(cur, struct btrfs_device, dev_list);
1404                 if (dev->devid == devid && num_found++ == instance)
1405                         return dev;
1406         }
1407         return NULL;
1408 }
1409
1410 int btrfs_bootstrap_super_map(struct btrfs_mapping_tree *map_tree,
1411                               struct btrfs_fs_devices *fs_devices)
1412 {
1413         struct map_lookup *map;
1414         u64 logical = BTRFS_SUPER_INFO_OFFSET;
1415         u64 length = BTRFS_SUPER_INFO_SIZE;
1416         int num_stripes = 0;
1417         int sub_stripes = 0;
1418         int ret;
1419         int i;
1420         struct list_head *cur;
1421
1422         list_for_each(cur, &fs_devices->devices) {
1423                 num_stripes++;
1424         }
1425         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1426         if (!map)
1427                 return -ENOMEM;
1428
1429         map->ce.start = logical;
1430         map->ce.size = length;
1431         map->num_stripes = num_stripes;
1432         map->sub_stripes = sub_stripes;
1433         map->io_width = length;
1434         map->io_align = length;
1435         map->sector_size = length;
1436         map->stripe_len = length;
1437         map->type = BTRFS_BLOCK_GROUP_RAID1;
1438
1439         i = 0;
1440         list_for_each(cur, &fs_devices->devices) {
1441                 struct btrfs_device *device = list_entry(cur,
1442                                                          struct btrfs_device,
1443                                                          dev_list);
1444                 map->stripes[i].physical = logical;
1445                 map->stripes[i].dev = device;
1446                 i++;
1447         }
1448         ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1449         if (ret == -EEXIST) {
1450                 struct cache_extent *old;
1451                 struct map_lookup *old_map;
1452                 old = find_cache_extent(&map_tree->cache_tree, logical, length);
1453                 old_map = container_of(old, struct map_lookup, ce);
1454                 remove_cache_extent(&map_tree->cache_tree, old);
1455                 kfree(old_map);
1456                 ret = insert_existing_cache_extent(&map_tree->cache_tree,
1457                                                    &map->ce);
1458         }
1459         BUG_ON(ret);
1460         return 0;
1461 }
1462
1463 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1464 {
1465         struct cache_extent *ce;
1466         struct map_lookup *map;
1467         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1468         int readonly = 0;
1469         int i;
1470
1471         ce = find_first_cache_extent(&map_tree->cache_tree, chunk_offset);
1472         BUG_ON(!ce);
1473
1474         map = container_of(ce, struct map_lookup, ce);
1475         for (i = 0; i < map->num_stripes; i++) {
1476                 if (!map->stripes[i].dev->writeable) {
1477                         readonly = 1;
1478                         break;
1479                 }
1480         }
1481
1482         return readonly;
1483 }
1484
1485 static struct btrfs_device *fill_missing_device(u64 devid)
1486 {
1487         struct btrfs_device *device;
1488
1489         device = kzalloc(sizeof(*device), GFP_NOFS);
1490         device->devid = devid;
1491         device->fd = -1;
1492         return device;
1493 }
1494
1495 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1496                           struct extent_buffer *leaf,
1497                           struct btrfs_chunk *chunk)
1498 {
1499         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1500         struct map_lookup *map;
1501         struct cache_extent *ce;
1502         u64 logical;
1503         u64 length;
1504         u64 devid;
1505         u8 uuid[BTRFS_UUID_SIZE];
1506         int num_stripes;
1507         int ret;
1508         int i;
1509
1510         logical = key->offset;
1511         length = btrfs_chunk_length(leaf, chunk);
1512
1513         ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1514
1515         /* already mapped? */
1516         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1517                 return 0;
1518         }
1519
1520         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1521         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1522         if (!map)
1523                 return -ENOMEM;
1524
1525         map->ce.start = logical;
1526         map->ce.size = length;
1527         map->num_stripes = num_stripes;
1528         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1529         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1530         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1531         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1532         map->type = btrfs_chunk_type(leaf, chunk);
1533         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1534
1535         for (i = 0; i < num_stripes; i++) {
1536                 map->stripes[i].physical =
1537                         btrfs_stripe_offset_nr(leaf, chunk, i);
1538                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1539                 read_extent_buffer(leaf, uuid, (unsigned long)
1540                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1541                                    BTRFS_UUID_SIZE);
1542                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1543                                                         NULL);
1544                 if (!map->stripes[i].dev) {
1545                         map->stripes[i].dev = fill_missing_device(devid);
1546                         printf("warning, device %llu is missing\n",
1547                                (unsigned long long)devid);
1548                 }
1549
1550         }
1551         ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1552         BUG_ON(ret);
1553
1554         return 0;
1555 }
1556
1557 static int fill_device_from_item(struct extent_buffer *leaf,
1558                                  struct btrfs_dev_item *dev_item,
1559                                  struct btrfs_device *device)
1560 {
1561         unsigned long ptr;
1562
1563         device->devid = btrfs_device_id(leaf, dev_item);
1564         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1565         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1566         device->type = btrfs_device_type(leaf, dev_item);
1567         device->io_align = btrfs_device_io_align(leaf, dev_item);
1568         device->io_width = btrfs_device_io_width(leaf, dev_item);
1569         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1570
1571         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1572         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1573
1574         return 0;
1575 }
1576
1577 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1578 {
1579         struct btrfs_fs_devices *fs_devices;
1580         int ret;
1581
1582         fs_devices = root->fs_info->fs_devices->seed;
1583         while (fs_devices) {
1584                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1585                         ret = 0;
1586                         goto out;
1587                 }
1588                 fs_devices = fs_devices->seed;
1589         }
1590
1591         fs_devices = find_fsid(fsid);
1592         if (!fs_devices) {
1593                 ret = -ENOENT;
1594                 goto out;
1595         }
1596
1597         ret = btrfs_open_devices(fs_devices, O_RDONLY);
1598         if (ret)
1599                 goto out;
1600
1601         fs_devices->seed = root->fs_info->fs_devices->seed;
1602         root->fs_info->fs_devices->seed = fs_devices;
1603 out:
1604         return ret;
1605 }
1606
1607 static int read_one_dev(struct btrfs_root *root,
1608                         struct extent_buffer *leaf,
1609                         struct btrfs_dev_item *dev_item)
1610 {
1611         struct btrfs_device *device;
1612         u64 devid;
1613         int ret = 0;
1614         u8 fs_uuid[BTRFS_UUID_SIZE];
1615         u8 dev_uuid[BTRFS_UUID_SIZE];
1616
1617         devid = btrfs_device_id(leaf, dev_item);
1618         read_extent_buffer(leaf, dev_uuid,
1619                            (unsigned long)btrfs_device_uuid(dev_item),
1620                            BTRFS_UUID_SIZE);
1621         read_extent_buffer(leaf, fs_uuid,
1622                            (unsigned long)btrfs_device_fsid(dev_item),
1623                            BTRFS_UUID_SIZE);
1624
1625         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1626                 ret = open_seed_devices(root, fs_uuid);
1627                 if (ret)
1628                         return ret;
1629         }
1630
1631         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1632         if (!device) {
1633                 printk("warning devid %llu not found already\n",
1634                         (unsigned long long)devid);
1635                 device = kmalloc(sizeof(*device), GFP_NOFS);
1636                 if (!device)
1637                         return -ENOMEM;
1638                 device->total_ios = 0;
1639                 list_add(&device->dev_list,
1640                          &root->fs_info->fs_devices->devices);
1641         }
1642
1643         fill_device_from_item(leaf, dev_item, device);
1644         device->dev_root = root->fs_info->dev_root;
1645         return ret;
1646 }
1647
1648 int btrfs_read_sys_array(struct btrfs_root *root)
1649 {
1650         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1651         struct extent_buffer *sb;
1652         struct btrfs_disk_key *disk_key;
1653         struct btrfs_chunk *chunk;
1654         struct btrfs_key key;
1655         u32 num_stripes;
1656         u32 array_size;
1657         u32 len = 0;
1658         u8 *ptr;
1659         unsigned long sb_ptr;
1660         u32 cur;
1661         int ret = 0;
1662
1663         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1664                                           BTRFS_SUPER_INFO_SIZE);
1665         if (!sb)
1666                 return -ENOMEM;
1667         btrfs_set_buffer_uptodate(sb);
1668         write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1669         array_size = btrfs_super_sys_array_size(super_copy);
1670
1671         /*
1672          * we do this loop twice, once for the device items and
1673          * once for all of the chunks.  This way there are device
1674          * structs filled in for every chunk
1675          */
1676         ptr = super_copy->sys_chunk_array;
1677         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1678         cur = 0;
1679
1680         while (cur < array_size) {
1681                 disk_key = (struct btrfs_disk_key *)ptr;
1682                 btrfs_disk_key_to_cpu(&key, disk_key);
1683
1684                 len = sizeof(*disk_key);
1685                 ptr += len;
1686                 sb_ptr += len;
1687                 cur += len;
1688
1689                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1690                         chunk = (struct btrfs_chunk *)sb_ptr;
1691                         ret = read_one_chunk(root, &key, sb, chunk);
1692                         if (ret)
1693                                 break;
1694                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1695                         len = btrfs_chunk_item_size(num_stripes);
1696                 } else {
1697                         BUG();
1698                 }
1699                 ptr += len;
1700                 sb_ptr += len;
1701                 cur += len;
1702         }
1703         free_extent_buffer(sb);
1704         return ret;
1705 }
1706
1707 int btrfs_read_chunk_tree(struct btrfs_root *root)
1708 {
1709         struct btrfs_path *path;
1710         struct extent_buffer *leaf;
1711         struct btrfs_key key;
1712         struct btrfs_key found_key;
1713         int ret;
1714         int slot;
1715
1716         root = root->fs_info->chunk_root;
1717
1718         path = btrfs_alloc_path();
1719         if (!path)
1720                 return -ENOMEM;
1721
1722         /* first we search for all of the device items, and then we
1723          * read in all of the chunk items.  This way we can create chunk
1724          * mappings that reference all of the devices that are afound
1725          */
1726         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1727         key.offset = 0;
1728         key.type = 0;
1729 again:
1730         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1731         while(1) {
1732                 leaf = path->nodes[0];
1733                 slot = path->slots[0];
1734                 if (slot >= btrfs_header_nritems(leaf)) {
1735                         ret = btrfs_next_leaf(root, path);
1736                         if (ret == 0)
1737                                 continue;
1738                         if (ret < 0)
1739                                 goto error;
1740                         break;
1741                 }
1742                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1743                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1744                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1745                                 break;
1746                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1747                                 struct btrfs_dev_item *dev_item;
1748                                 dev_item = btrfs_item_ptr(leaf, slot,
1749                                                   struct btrfs_dev_item);
1750                                 ret = read_one_dev(root, leaf, dev_item);
1751                                 BUG_ON(ret);
1752                         }
1753                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1754                         struct btrfs_chunk *chunk;
1755                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1756                         ret = read_one_chunk(root, &found_key, leaf, chunk);
1757                         BUG_ON(ret);
1758                 }
1759                 path->slots[0]++;
1760         }
1761         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1762                 key.objectid = 0;
1763                 btrfs_release_path(root, path);
1764                 goto again;
1765         }
1766
1767         ret = 0;
1768 error:
1769         btrfs_free_path(path);
1770         return ret;
1771 }
1772
1773 struct list_head *btrfs_scanned_uuids(void)
1774 {
1775         return &fs_uuids;
1776 }
1777
1778 static int rmw_eb(struct btrfs_fs_info *info,
1779                   struct extent_buffer *eb, struct extent_buffer *orig_eb)
1780 {
1781         int ret;
1782         unsigned long orig_off = 0;
1783         unsigned long dest_off = 0;
1784         unsigned long copy_len = eb->len;
1785
1786         ret = read_whole_eb(info, eb, 0);
1787         if (ret)
1788                 return ret;
1789
1790         if (eb->start + eb->len <= orig_eb->start ||
1791             eb->start >= orig_eb->start + orig_eb->len)
1792                 return 0;
1793         /*
1794          * | ----- orig_eb ------- |
1795          *         | ----- stripe -------  |
1796          *         | ----- orig_eb ------- |
1797          *              | ----- orig_eb ------- |
1798          */
1799         if (eb->start > orig_eb->start)
1800                 orig_off = eb->start - orig_eb->start;
1801         if (orig_eb->start > eb->start)
1802                 dest_off = orig_eb->start - eb->start;
1803
1804         if (copy_len > orig_eb->len - orig_off)
1805                 copy_len = orig_eb->len - orig_off;
1806         if (copy_len > eb->len - dest_off)
1807                 copy_len = eb->len - dest_off;
1808
1809         memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1810         return 0;
1811 }
1812
1813 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1814                                 struct extent_buffer *orig_eb,
1815                                struct extent_buffer **ebs,
1816                                u64 stripe_len, u64 *raid_map,
1817                                int num_stripes)
1818 {
1819         struct extent_buffer *eb;
1820         u64 start = orig_eb->start;
1821         u64 this_eb_start;
1822         int i;
1823         int ret;
1824
1825         for (i = 0; i < num_stripes; i++) {
1826                 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1827                         break;
1828
1829                 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1830                 if (!eb)
1831                         BUG();
1832                 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1833
1834                 eb->start = raid_map[i];
1835                 eb->len = stripe_len;
1836                 eb->refs = 1;
1837                 eb->flags = 0;
1838                 eb->fd = -1;
1839                 eb->dev_bytenr = (u64)-1;
1840
1841                 this_eb_start = raid_map[i];
1842
1843                 if (start > this_eb_start ||
1844                     start + orig_eb->len < this_eb_start + stripe_len) {
1845                         ret = rmw_eb(info, eb, orig_eb);
1846                         BUG_ON(ret);
1847                 } else {
1848                         memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1849                 }
1850                 ebs[i] = eb;
1851         }
1852 }
1853
1854 int write_raid56_with_parity(struct btrfs_fs_info *info,
1855                              struct extent_buffer *eb,
1856                              struct btrfs_multi_bio *multi,
1857                              u64 stripe_len, u64 *raid_map)
1858 {
1859         struct extent_buffer *ebs[multi->num_stripes], *p_eb = NULL, *q_eb = NULL;
1860         int i;
1861         int j;
1862         int ret;
1863         int alloc_size = eb->len;
1864
1865         if (stripe_len > alloc_size)
1866                 alloc_size = stripe_len;
1867
1868         split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1869                             multi->num_stripes);
1870
1871         for (i = 0; i < multi->num_stripes; i++) {
1872                 struct extent_buffer *new_eb;
1873                 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1874                         ebs[i]->dev_bytenr = multi->stripes[i].physical;
1875                         ebs[i]->fd = multi->stripes[i].dev->fd;
1876                         multi->stripes[i].dev->total_ios++;
1877                         BUG_ON(ebs[i]->start != raid_map[i]);
1878                         continue;
1879                 }
1880                 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1881                 BUG_ON(!new_eb);
1882                 new_eb->dev_bytenr = multi->stripes[i].physical;
1883                 new_eb->fd = multi->stripes[i].dev->fd;
1884                 multi->stripes[i].dev->total_ios++;
1885                 new_eb->len = stripe_len;
1886
1887                 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1888                         p_eb = new_eb;
1889                 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1890                         q_eb = new_eb;
1891         }
1892         if (q_eb) {
1893                 void *pointers[multi->num_stripes];
1894                 ebs[multi->num_stripes - 2] = p_eb;
1895                 ebs[multi->num_stripes - 1] = q_eb;
1896
1897                 for (i = 0; i < multi->num_stripes; i++)
1898                         pointers[i] = ebs[i]->data;
1899
1900                 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1901         } else {
1902                 ebs[multi->num_stripes - 1] = p_eb;
1903                 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1904                 for (j = 1; j < multi->num_stripes - 1; j++) {
1905                         for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1906                                 *(unsigned long *)(p_eb->data + i) ^=
1907                                         *(unsigned long *)(ebs[j]->data + i);
1908                         }
1909                 }
1910         }
1911
1912         for (i = 0; i < multi->num_stripes; i++) {
1913                 ret = write_extent_to_disk(ebs[i]);
1914                 BUG_ON(ret);
1915                 if (ebs[i] != eb)
1916                         kfree(ebs[i]);
1917         }
1918         return 0;
1919 }