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