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         list_del(&fs_devices->list);
 192         free(fs_devices);
 193         return 0;
 194 }
 195
 196 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
 197 {
 198         int fd;
 199         struct list_head *head = &fs_devices->devices;
 200         struct list_head *cur;
 201         struct btrfs_device *device;
 202         int ret;
 203
 204         list_for_each(cur, head) {
 205                 device = list_entry(cur, struct btrfs_device, dev_list);
 206                 if (!device->name) {
 207                         printk("no name for device %llu, skip it now\n", device->devid);
 208                         continue;
 209                 }
 210
 211                 fd = open(device->name, flags);
 212                 if (fd < 0) {
 213                         ret = -errno;
 214                         goto fail;
 215                 }
 216
 217                 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
 218                         fprintf(stderr, "Warning, could not drop caches\n");
 219
 220                 if (device->devid == fs_devices->latest_devid)
 221                         fs_devices->latest_bdev = fd;
 222                 if (device->devid == fs_devices->lowest_devid)
 223                         fs_devices->lowest_bdev = fd;
 224                 device->fd = fd;
 225                 if (flags & O_RDWR)
 226                         device->writeable = 1;
 227         }
 228         return 0;
 229 fail:
 230         btrfs_close_devices(fs_devices);
 231         return ret;
 232 }
 233
 234 int btrfs_scan_one_device(int fd, const char *path,
 235                           struct btrfs_fs_devices **fs_devices_ret,
 236                           u64 *total_devs, u64 super_offset)
 237 {
 238         struct btrfs_super_block *disk_super;
 239         char *buf;
 240         int ret;
 241         u64 devid;
 242
 243         buf = malloc(4096);
 244         if (!buf) {
 245                 ret = -ENOMEM;
 246                 goto error;
 247         }
 248         disk_super = (struct btrfs_super_block *)buf;
 249         ret = btrfs_read_dev_super(fd, disk_super, super_offset);
 250         if (ret < 0) {
 251                 ret = -EIO;
 252                 goto error_brelse;
 253         }
 254         devid = btrfs_stack_device_id(&disk_super->dev_item);
 255         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
 256                 *total_devs = 1;
 257         else
 258                 *total_devs = btrfs_super_num_devices(disk_super);
 259
 260         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 261
 262 error_brelse:
 263         free(buf);
 264 error:
 265         return ret;
 266 }
 267
 268 /*
 269  * this uses a pretty simple search, the expectation is that it is
 270  * called very infrequently and that a given device has a small number
 271  * of extents
 272  */
 273 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 274                                 struct btrfs_device *device,
 275                                 struct btrfs_path *path,
 276                                 u64 num_bytes, u64 *start)
 277 {
 278         struct btrfs_key key;
 279         struct btrfs_root *root = device->dev_root;
 280         struct btrfs_dev_extent *dev_extent = NULL;
 281         u64 hole_size = 0;
 282         u64 last_byte = 0;
 283         u64 search_start = root->fs_info->alloc_start;
 284         u64 search_end = device->total_bytes;
 285         int ret;
 286         int slot = 0;
 287         int start_found;
 288         struct extent_buffer *l;
 289
 290         start_found = 0;
 291         path->reada = 2;
 292
 293         /* FIXME use last free of some kind */
 294
 295         /* we don't want to overwrite the superblock on the drive,
 296          * so we make sure to start at an offset of at least 1MB
 297          */
 298         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 299
 300         if (search_start >= search_end) {
 301                 ret = -ENOSPC;
 302                 goto error;
 303         }
 304
 305         key.objectid = device->devid;
 306         key.offset = search_start;
 307         key.type = BTRFS_DEV_EXTENT_KEY;
 308         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 309         if (ret < 0)
 310                 goto error;
 311         ret = btrfs_previous_item(root, path, 0, key.type);
 312         if (ret < 0)
 313                 goto error;
 314         l = path->nodes[0];
 315         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 316         while (1) {
 317                 l = path->nodes[0];
 318                 slot = path->slots[0];
 319                 if (slot >= btrfs_header_nritems(l)) {
 320                         ret = btrfs_next_leaf(root, path);
 321                         if (ret == 0)
 322                                 continue;
 323                         if (ret < 0)
 324                                 goto error;
 325 no_more_items:
 326                         if (!start_found) {
 327                                 if (search_start >= search_end) {
 328                                         ret = -ENOSPC;
 329                                         goto error;
 330                                 }
 331                                 *start = search_start;
 332                                 start_found = 1;
 333                                 goto check_pending;
 334                         }
 335                         *start = last_byte > search_start ?
 336                                 last_byte : search_start;
 337                         if (search_end <= *start) {
 338                                 ret = -ENOSPC;
 339                                 goto error;
 340                         }
 341                         goto check_pending;
 342                 }
 343                 btrfs_item_key_to_cpu(l, &key, slot);
 344
 345                 if (key.objectid < device->devid)
 346                         goto next;
 347
 348                 if (key.objectid > device->devid)
 349                         goto no_more_items;
 350
 351                 if (key.offset >= search_start && key.offset > last_byte &&
 352                     start_found) {
 353                         if (last_byte < search_start)
 354                                 last_byte = search_start;
 355                         hole_size = key.offset - last_byte;
 356                         if (key.offset > last_byte &&
 357                             hole_size >= num_bytes) {
 358                                 *start = last_byte;
 359                                 goto check_pending;
 360                         }
 361                 }
 362                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 363                         goto next;
 364                 }
 365
 366                 start_found = 1;
 367                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 368                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 369 next:
 370                 path->slots[0]++;
 371                 cond_resched();
 372         }
 373 check_pending:
 374         /* we have to make sure we didn't find an extent that has already
 375          * been allocated by the map tree or the original allocation
 376          */
 377         btrfs_release_path(path);
 378         BUG_ON(*start < search_start);
 379
 380         if (*start + num_bytes > search_end) {
 381                 ret = -ENOSPC;
 382                 goto error;
 383         }
 384         /* check for pending inserts here */
 385         return 0;
 386
 387 error:
 388         btrfs_release_path(path);
 389         return ret;
 390 }
 391
 392 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 393                                   struct btrfs_device *device,
 394                                   u64 chunk_tree, u64 chunk_objectid,
 395                                   u64 chunk_offset,
 396                                   u64 num_bytes, u64 *start)
 397 {
 398         int ret;
 399         struct btrfs_path *path;
 400         struct btrfs_root *root = device->dev_root;
 401         struct btrfs_dev_extent *extent;
 402         struct extent_buffer *leaf;
 403         struct btrfs_key key;
 404
 405         path = btrfs_alloc_path();
 406         if (!path)
 407                 return -ENOMEM;
 408
 409         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 410         if (ret) {
 411                 goto err;
 412         }
 413
 414         key.objectid = device->devid;
 415         key.offset = *start;
 416         key.type = BTRFS_DEV_EXTENT_KEY;
 417         ret = btrfs_insert_empty_item(trans, root, path, &key,
 418                                       sizeof(*extent));
 419         BUG_ON(ret);
 420
 421         leaf = path->nodes[0];
 422         extent = btrfs_item_ptr(leaf, path->slots[0],
 423                                 struct btrfs_dev_extent);
 424         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 425         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 426         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 427
 428         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 429                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 430                     BTRFS_UUID_SIZE);
 431
 432         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 433         btrfs_mark_buffer_dirty(leaf);
 434 err:
 435         btrfs_free_path(path);
 436         return ret;
 437 }
 438
 439 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 440 {
 441         struct btrfs_path *path;
 442         int ret;
 443         struct btrfs_key key;
 444         struct btrfs_chunk *chunk;
 445         struct btrfs_key found_key;
 446
 447         path = btrfs_alloc_path();
 448         BUG_ON(!path);
 449
 450         key.objectid = objectid;
 451         key.offset = (u64)-1;
 452         key.type = BTRFS_CHUNK_ITEM_KEY;
 453
 454         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 455         if (ret < 0)
 456                 goto error;
 457
 458         BUG_ON(ret == 0);
 459
 460         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 461         if (ret) {
 462                 *offset = 0;
 463         } else {
 464                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 465                                       path->slots[0]);
 466                 if (found_key.objectid != objectid)
 467                         *offset = 0;
 468                 else {
 469                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 470                                                struct btrfs_chunk);
 471                         *offset = found_key.offset +
 472                                 btrfs_chunk_length(path->nodes[0], chunk);
 473                 }
 474         }
 475         ret = 0;
 476 error:
 477         btrfs_free_path(path);
 478         return ret;
 479 }
 480
 481 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 482                            u64 *objectid)
 483 {
 484         int ret;
 485         struct btrfs_key key;
 486         struct btrfs_key found_key;
 487
 488         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 489         key.type = BTRFS_DEV_ITEM_KEY;
 490         key.offset = (u64)-1;
 491
 492         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 493         if (ret < 0)
 494                 goto error;
 495
 496         BUG_ON(ret == 0);
 497
 498         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 499                                   BTRFS_DEV_ITEM_KEY);
 500         if (ret) {
 501                 *objectid = 1;
 502         } else {
 503                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 504                                       path->slots[0]);
 505                 *objectid = found_key.offset + 1;
 506         }
 507         ret = 0;
 508 error:
 509         btrfs_release_path(path);
 510         return ret;
 511 }
 512
 513 /*
 514  * the device information is stored in the chunk root
 515  * the btrfs_device struct should be fully filled in
 516  */
 517 int btrfs_add_device(struct btrfs_trans_handle *trans,
 518                      struct btrfs_root *root,
 519                      struct btrfs_device *device)
 520 {
 521         int ret;
 522         struct btrfs_path *path;
 523         struct btrfs_dev_item *dev_item;
 524         struct extent_buffer *leaf;
 525         struct btrfs_key key;
 526         unsigned long ptr;
 527         u64 free_devid = 0;
 528
 529         root = root->fs_info->chunk_root;
 530
 531         path = btrfs_alloc_path();
 532         if (!path)
 533                 return -ENOMEM;
 534
 535         ret = find_next_devid(root, path, &free_devid);
 536         if (ret)
 537                 goto out;
 538
 539         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 540         key.type = BTRFS_DEV_ITEM_KEY;
 541         key.offset = free_devid;
 542
 543         ret = btrfs_insert_empty_item(trans, root, path, &key,
 544                                       sizeof(*dev_item));
 545         if (ret)
 546                 goto out;
 547
 548         leaf = path->nodes[0];
 549         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 550
 551         device->devid = free_devid;
 552         btrfs_set_device_id(leaf, dev_item, device->devid);
 553         btrfs_set_device_generation(leaf, dev_item, 0);
 554         btrfs_set_device_type(leaf, dev_item, device->type);
 555         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 556         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 557         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 558         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 559         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 560         btrfs_set_device_group(leaf, dev_item, 0);
 561         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 562         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 563         btrfs_set_device_start_offset(leaf, dev_item, 0);
 564
 565         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 566         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 567         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 568         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 569         btrfs_mark_buffer_dirty(leaf);
 570         ret = 0;
 571
 572 out:
 573         btrfs_free_path(path);
 574         return ret;
 575 }
 576
 577 int btrfs_update_device(struct btrfs_trans_handle *trans,
 578                         struct btrfs_device *device)
 579 {
 580         int ret;
 581         struct btrfs_path *path;
 582         struct btrfs_root *root;
 583         struct btrfs_dev_item *dev_item;
 584         struct extent_buffer *leaf;
 585         struct btrfs_key key;
 586
 587         root = device->dev_root->fs_info->chunk_root;
 588
 589         path = btrfs_alloc_path();
 590         if (!path)
 591                 return -ENOMEM;
 592
 593         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 594         key.type = BTRFS_DEV_ITEM_KEY;
 595         key.offset = device->devid;
 596
 597         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 598         if (ret < 0)
 599                 goto out;
 600
 601         if (ret > 0) {
 602                 ret = -ENOENT;
 603                 goto out;
 604         }
 605
 606         leaf = path->nodes[0];
 607         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 608
 609         btrfs_set_device_id(leaf, dev_item, device->devid);
 610         btrfs_set_device_type(leaf, dev_item, device->type);
 611         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 612         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 613         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 614         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 615         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 616         btrfs_mark_buffer_dirty(leaf);
 617
 618 out:
 619         btrfs_free_path(path);
 620         return ret;
 621 }
 622
 623 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 624                            struct btrfs_root *root,
 625                            struct btrfs_key *key,
 626                            struct btrfs_chunk *chunk, int item_size)
 627 {
 628         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
 629         struct btrfs_disk_key disk_key;
 630         u32 array_size;
 631         u8 *ptr;
 632
 633         array_size = btrfs_super_sys_array_size(super_copy);
 634         if (array_size + item_size + sizeof(disk_key)
 635                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 636                 return -EFBIG;
 637
 638         ptr = super_copy->sys_chunk_array + array_size;
 639         btrfs_cpu_key_to_disk(&disk_key, key);
 640         memcpy(ptr, &disk_key, sizeof(disk_key));
 641         ptr += sizeof(disk_key);
 642         memcpy(ptr, chunk, item_size);
 643         item_size += sizeof(disk_key);
 644         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 645         return 0;
 646 }
 647
 648 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 649                                int sub_stripes)
 650 {
 651         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
 652                 return calc_size;
 653         else if (type & BTRFS_BLOCK_GROUP_RAID10)
 654                 return calc_size * (num_stripes / sub_stripes);
 655         else if (type & BTRFS_BLOCK_GROUP_RAID5)
 656                 return calc_size * (num_stripes - 1);
 657         else if (type & BTRFS_BLOCK_GROUP_RAID6)
 658                 return calc_size * (num_stripes - 2);
 659         else
 660                 return calc_size * num_stripes;
 661 }
 662
 663
 664 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
 665 {
 666         /* TODO, add a way to store the preferred stripe size */
 667         return BTRFS_STRIPE_LEN;
 668 }
 669
 670 /*
 671  * btrfs_device_avail_bytes - count bytes available for alloc_chunk
 672  *
 673  * It is not equal to "device->total_bytes - device->bytes_used".
 674  * We do not allocate any chunk in 1M at beginning of device, and not
 675  * allowed to allocate any chunk before alloc_start if it is specified.
 676  * So search holes from max(1M, alloc_start) to device->total_bytes.
 677  */
 678 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
 679                                     struct btrfs_device *device,
 680                                     u64 *avail_bytes)
 681 {
 682         struct btrfs_path *path;
 683         struct btrfs_root *root = device->dev_root;
 684         struct btrfs_key key;
 685         struct btrfs_dev_extent *dev_extent = NULL;
 686         struct extent_buffer *l;
 687         u64 search_start = root->fs_info->alloc_start;
 688         u64 search_end = device->total_bytes;
 689         u64 extent_end = 0;
 690         u64 free_bytes = 0;
 691         int ret;
 692         int slot = 0;
 693
 694         search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
 695
 696         path = btrfs_alloc_path();
 697         if (!path)
 698                 return -ENOMEM;
 699
 700         key.objectid = device->devid;
 701         key.offset = root->fs_info->alloc_start;
 702         key.type = BTRFS_DEV_EXTENT_KEY;
 703
 704         path->reada = 2;
 705         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 706         if (ret < 0)
 707                 goto error;
 708         ret = btrfs_previous_item(root, path, 0, key.type);
 709         if (ret < 0)
 710                 goto error;
 711
 712         while (1) {
 713                 l = path->nodes[0];
 714                 slot = path->slots[0];
 715                 if (slot >= btrfs_header_nritems(l)) {
 716                         ret = btrfs_next_leaf(root, path);
 717                         if (ret == 0)
 718                                 continue;
 719                         if (ret < 0)
 720                                 goto error;
 721                         break;
 722                 }
 723                 btrfs_item_key_to_cpu(l, &key, slot);
 724
 725                 if (key.objectid < device->devid)
 726                         goto next;
 727                 if (key.objectid > device->devid)
 728                         break;
 729                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 730                         goto next;
 731                 if (key.offset > search_end)
 732                         break;
 733                 if (key.offset > search_start)
 734                         free_bytes += key.offset - search_start;
 735
 736                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 737                 extent_end = key.offset + btrfs_dev_extent_length(l,
 738                                                                   dev_extent);
 739                 if (extent_end > search_start)
 740                         search_start = extent_end;
 741                 if (search_start > search_end)
 742                         break;
 743 next:
 744                 path->slots[0]++;
 745                 cond_resched();
 746         }
 747
 748         if (search_start < search_end)
 749                 free_bytes += search_end - search_start;
 750
 751         *avail_bytes = free_bytes;
 752         ret = 0;
 753 error:
 754         btrfs_free_path(path);
 755         return ret;
 756 }
 757
 758 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
 759                         - sizeof(struct btrfs_item)             \
 760                         - sizeof(struct btrfs_chunk))           \
 761                         / sizeof(struct btrfs_stripe) + 1)
 762
 763 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
 764                                 - 2 * sizeof(struct btrfs_disk_key)     \
 765                                 - 2 * sizeof(struct btrfs_chunk))       \
 766                                 / sizeof(struct btrfs_stripe) + 1)
 767
 768 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 769                       struct btrfs_root *extent_root, u64 *start,
 770                       u64 *num_bytes, u64 type)
 771 {
 772         u64 dev_offset;
 773         struct btrfs_fs_info *info = extent_root->fs_info;
 774         struct btrfs_root *chunk_root = info->chunk_root;
 775         struct btrfs_stripe *stripes;
 776         struct btrfs_device *device = NULL;
 777         struct btrfs_chunk *chunk;
 778         struct list_head private_devs;
 779         struct list_head *dev_list = &info->fs_devices->devices;
 780         struct list_head *cur;
 781         struct map_lookup *map;
 782         int min_stripe_size = 1 * 1024 * 1024;
 783         u64 calc_size = 8 * 1024 * 1024;
 784         u64 min_free;
 785         u64 max_chunk_size = 4 * calc_size;
 786         u64 avail = 0;
 787         u64 max_avail = 0;
 788         u64 percent_max;
 789         int num_stripes = 1;
 790         int max_stripes = 0;
 791         int min_stripes = 1;
 792         int sub_stripes = 0;
 793         int looped = 0;
 794         int ret;
 795         int index;
 796         int stripe_len = BTRFS_STRIPE_LEN;
 797         struct btrfs_key key;
 798         u64 offset;
 799
 800         if (list_empty(dev_list)) {
 801                 return -ENOSPC;
 802         }
 803
 804         if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
 805                     BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
 806                     BTRFS_BLOCK_GROUP_RAID10 |
 807                     BTRFS_BLOCK_GROUP_DUP)) {
 808                 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
 809                         calc_size = 8 * 1024 * 1024;
 810                         max_chunk_size = calc_size * 2;
 811                         min_stripe_size = 1 * 1024 * 1024;
 812                         max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
 813                 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
 814                         calc_size = 1024 * 1024 * 1024;
 815                         max_chunk_size = 10 * calc_size;
 816                         min_stripe_size = 64 * 1024 * 1024;
 817                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 818                 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
 819                         calc_size = 1024 * 1024 * 1024;
 820                         max_chunk_size = 4 * calc_size;
 821                         min_stripe_size = 32 * 1024 * 1024;
 822                         max_stripes = BTRFS_MAX_DEVS(chunk_root);
 823                 }
 824         }
 825         if (type & BTRFS_BLOCK_GROUP_RAID1) {
 826                 num_stripes = min_t(u64, 2,
 827                                   btrfs_super_num_devices(info->super_copy));
 828                 if (num_stripes < 2)
 829                         return -ENOSPC;
 830                 min_stripes = 2;
 831         }
 832         if (type & BTRFS_BLOCK_GROUP_DUP) {
 833                 num_stripes = 2;
 834                 min_stripes = 2;
 835         }
 836         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
 837                 num_stripes = btrfs_super_num_devices(info->super_copy);
 838                 if (num_stripes > max_stripes)
 839                         num_stripes = max_stripes;
 840                 min_stripes = 2;
 841         }
 842         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 843                 num_stripes = btrfs_super_num_devices(info->super_copy);
 844                 if (num_stripes > max_stripes)
 845                         num_stripes = max_stripes;
 846                 if (num_stripes < 4)
 847                         return -ENOSPC;
 848                 num_stripes &= ~(u32)1;
 849                 sub_stripes = 2;
 850                 min_stripes = 4;
 851         }
 852         if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
 853                 num_stripes = btrfs_super_num_devices(info->super_copy);
 854                 if (num_stripes > max_stripes)
 855                         num_stripes = max_stripes;
 856                 if (num_stripes < 2)
 857                         return -ENOSPC;
 858                 min_stripes = 2;
 859                 stripe_len = find_raid56_stripe_len(num_stripes - 1,
 860                                     btrfs_super_stripesize(info->super_copy));
 861         }
 862         if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
 863                 num_stripes = btrfs_super_num_devices(info->super_copy);
 864                 if (num_stripes > max_stripes)
 865                         num_stripes = max_stripes;
 866                 if (num_stripes < 3)
 867                         return -ENOSPC;
 868                 min_stripes = 3;
 869                 stripe_len = find_raid56_stripe_len(num_stripes - 2,
 870                                     btrfs_super_stripesize(info->super_copy));
 871         }
 872
 873         /* we don't want a chunk larger than 10% of the FS */
 874         percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
 875         max_chunk_size = min(percent_max, max_chunk_size);
 876
 877 again:
 878         if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
 879             max_chunk_size) {
 880                 calc_size = max_chunk_size;
 881                 calc_size /= num_stripes;
 882                 calc_size /= stripe_len;
 883                 calc_size *= stripe_len;
 884         }
 885         /* we don't want tiny stripes */
 886         calc_size = max_t(u64, calc_size, min_stripe_size);
 887
 888         calc_size /= stripe_len;
 889         calc_size *= stripe_len;
 890         INIT_LIST_HEAD(&private_devs);
 891         cur = dev_list->next;
 892         index = 0;
 893
 894         if (type & BTRFS_BLOCK_GROUP_DUP)
 895                 min_free = calc_size * 2;
 896         else
 897                 min_free = calc_size;
 898
 899         /* build a private list of devices we will allocate from */
 900         while(index < num_stripes) {
 901                 device = list_entry(cur, struct btrfs_device, dev_list);
 902                 ret = btrfs_device_avail_bytes(trans, device, &avail);
 903                 if (ret)
 904                         return ret;
 905                 cur = cur->next;
 906                 if (avail >= min_free) {
 907                         list_move_tail(&device->dev_list, &private_devs);
 908                         index++;
 909                         if (type & BTRFS_BLOCK_GROUP_DUP)
 910                                 index++;
 911                 } else if (avail > max_avail)
 912                         max_avail = avail;
 913                 if (cur == dev_list)
 914                         break;
 915         }
 916         if (index < num_stripes) {
 917                 list_splice(&private_devs, dev_list);
 918                 if (index >= min_stripes) {
 919                         num_stripes = index;
 920                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
 921                                 num_stripes /= sub_stripes;
 922                                 num_stripes *= sub_stripes;
 923                         }
 924                         looped = 1;
 925                         goto again;
 926                 }
 927                 if (!looped && max_avail > 0) {
 928                         looped = 1;
 929                         calc_size = max_avail;
 930                         goto again;
 931                 }
 932                 return -ENOSPC;
 933         }
 934         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 935                               &offset);
 936         if (ret)
 937                 return ret;
 938         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
 939         key.type = BTRFS_CHUNK_ITEM_KEY;
 940         key.offset = offset;
 941
 942         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
 943         if (!chunk)
 944                 return -ENOMEM;
 945
 946         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
 947         if (!map) {
 948                 kfree(chunk);
 949                 return -ENOMEM;
 950         }
 951
 952         stripes = &chunk->stripe;
 953         *num_bytes = chunk_bytes_by_type(type, calc_size,
 954                                          num_stripes, sub_stripes);
 955         index = 0;
 956         while(index < num_stripes) {
 957                 struct btrfs_stripe *stripe;
 958                 BUG_ON(list_empty(&private_devs));
 959                 cur = private_devs.next;
 960                 device = list_entry(cur, struct btrfs_device, dev_list);
 961
 962                 /* loop over this device again if we're doing a dup group */
 963                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
 964                     (index == num_stripes - 1))
 965                         list_move_tail(&device->dev_list, dev_list);
 966
 967                 ret = btrfs_alloc_dev_extent(trans, device,
 968                              info->chunk_root->root_key.objectid,
 969                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
 970                              calc_size, &dev_offset);
 971                 BUG_ON(ret);
 972
 973                 device->bytes_used += calc_size;
 974                 ret = btrfs_update_device(trans, device);
 975                 BUG_ON(ret);
 976
 977                 map->stripes[index].dev = device;
 978                 map->stripes[index].physical = dev_offset;
 979                 stripe = stripes + index;
 980                 btrfs_set_stack_stripe_devid(stripe, device->devid);
 981                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
 982                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
 983                 index++;
 984         }
 985         BUG_ON(!list_empty(&private_devs));
 986
 987         /* key was set above */
 988         btrfs_set_stack_chunk_length(chunk, *num_bytes);
 989         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
 990         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 991         btrfs_set_stack_chunk_type(chunk, type);
 992         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
 993         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
 994         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 995         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
 996         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
 997         map->sector_size = extent_root->sectorsize;
 998         map->stripe_len = stripe_len;
 999         map->io_align = stripe_len;
1000         map->io_width = stripe_len;
1001         map->type = type;
1002         map->num_stripes = num_stripes;
1003         map->sub_stripes = sub_stripes;
1004
1005         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1006                                 btrfs_chunk_item_size(num_stripes));
1007         BUG_ON(ret);
1008         *start = key.offset;;
1009
1010         map->ce.start = key.offset;
1011         map->ce.size = *num_bytes;
1012
1013         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1014         BUG_ON(ret);
1015
1016         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1017                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1018                                     chunk, btrfs_chunk_item_size(num_stripes));
1019                 BUG_ON(ret);
1020         }
1021
1022         kfree(chunk);
1023         return ret;
1024 }
1025
1026 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1027                            struct btrfs_root *extent_root, u64 *start,
1028                            u64 num_bytes, u64 type)
1029 {
1030         u64 dev_offset;
1031         struct btrfs_fs_info *info = extent_root->fs_info;
1032         struct btrfs_root *chunk_root = info->chunk_root;
1033         struct btrfs_stripe *stripes;
1034         struct btrfs_device *device = NULL;
1035         struct btrfs_chunk *chunk;
1036         struct list_head *dev_list = &info->fs_devices->devices;
1037         struct list_head *cur;
1038         struct map_lookup *map;
1039         u64 calc_size = 8 * 1024 * 1024;
1040         int num_stripes = 1;
1041         int sub_stripes = 0;
1042         int ret;
1043         int index;
1044         int stripe_len = BTRFS_STRIPE_LEN;
1045         struct btrfs_key key;
1046
1047         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1048         key.type = BTRFS_CHUNK_ITEM_KEY;
1049         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1050                               &key.offset);
1051         if (ret)
1052                 return ret;
1053
1054         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1055         if (!chunk)
1056                 return -ENOMEM;
1057
1058         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1059         if (!map) {
1060                 kfree(chunk);
1061                 return -ENOMEM;
1062         }
1063
1064         stripes = &chunk->stripe;
1065         calc_size = num_bytes;
1066
1067         index = 0;
1068         cur = dev_list->next;
1069         device = list_entry(cur, struct btrfs_device, dev_list);
1070
1071         while (index < num_stripes) {
1072                 struct btrfs_stripe *stripe;
1073
1074                 ret = btrfs_alloc_dev_extent(trans, device,
1075                              info->chunk_root->root_key.objectid,
1076                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1077                              calc_size, &dev_offset);
1078                 BUG_ON(ret);
1079
1080                 device->bytes_used += calc_size;
1081                 ret = btrfs_update_device(trans, device);
1082                 BUG_ON(ret);
1083
1084                 map->stripes[index].dev = device;
1085                 map->stripes[index].physical = dev_offset;
1086                 stripe = stripes + index;
1087                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1088                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1089                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1090                 index++;
1091         }
1092
1093         /* key was set above */
1094         btrfs_set_stack_chunk_length(chunk, num_bytes);
1095         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1096         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1097         btrfs_set_stack_chunk_type(chunk, type);
1098         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1099         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1100         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1101         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1102         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1103         map->sector_size = extent_root->sectorsize;
1104         map->stripe_len = stripe_len;
1105         map->io_align = stripe_len;
1106         map->io_width = stripe_len;
1107         map->type = type;
1108         map->num_stripes = num_stripes;
1109         map->sub_stripes = sub_stripes;
1110
1111         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1112                                 btrfs_chunk_item_size(num_stripes));
1113         BUG_ON(ret);
1114         *start = key.offset;
1115
1116         map->ce.start = key.offset;
1117         map->ce.size = num_bytes;
1118
1119         ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1120         BUG_ON(ret);
1121
1122         kfree(chunk);
1123         return ret;
1124 }
1125
1126 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1127 {
1128         struct cache_extent *ce;
1129         struct map_lookup *map;
1130         int ret;
1131
1132         ce = search_cache_extent(&map_tree->cache_tree, logical);
1133         BUG_ON(!ce);
1134         BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1135         map = container_of(ce, struct map_lookup, ce);
1136
1137         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1138                 ret = map->num_stripes;
1139         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1140                 ret = map->sub_stripes;
1141         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1142                 ret = 2;
1143         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1144                 ret = 3;
1145         else
1146                 ret = 1;
1147         return ret;
1148 }
1149
1150 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1151                         u64 *size)
1152 {
1153         struct cache_extent *ce;
1154         struct map_lookup *map;
1155
1156         ce = search_cache_extent(&map_tree->cache_tree, *logical);
1157
1158         while (ce) {
1159                 ce = next_cache_extent(ce);
1160                 if (!ce)
1161                         return -ENOENT;
1162
1163                 map = container_of(ce, struct map_lookup, ce);
1164                 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1165                         *logical = ce->start;
1166                         *size = ce->size;
1167                         return 0;
1168                 }
1169         }
1170
1171         return -ENOENT;
1172 }
1173
1174 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1175                      u64 chunk_start, u64 physical, u64 devid,
1176                      u64 **logical, int *naddrs, int *stripe_len)
1177 {
1178         struct cache_extent *ce;
1179         struct map_lookup *map;
1180         u64 *buf;
1181         u64 bytenr;
1182         u64 length;
1183         u64 stripe_nr;
1184         u64 rmap_len;
1185         int i, j, nr = 0;
1186
1187         ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1188         BUG_ON(!ce);
1189         map = container_of(ce, struct map_lookup, ce);
1190
1191         length = ce->size;
1192         rmap_len = map->stripe_len;
1193         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1194                 length = ce->size / (map->num_stripes / map->sub_stripes);
1195         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1196                 length = ce->size / map->num_stripes;
1197         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1198                               BTRFS_BLOCK_GROUP_RAID6)) {
1199                 length = ce->size / nr_data_stripes(map);
1200                 rmap_len = map->stripe_len * nr_data_stripes(map);
1201         }
1202
1203         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1204
1205         for (i = 0; i < map->num_stripes; i++) {
1206                 if (devid && map->stripes[i].dev->devid != devid)
1207                         continue;
1208                 if (map->stripes[i].physical > physical ||
1209                     map->stripes[i].physical + length <= physical)
1210                         continue;
1211
1212                 stripe_nr = (physical - map->stripes[i].physical) /
1213                             map->stripe_len;
1214
1215                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1216                         stripe_nr = (stripe_nr * map->num_stripes + i) /
1217                                     map->sub_stripes;
1218                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1219                         stripe_nr = stripe_nr * map->num_stripes + i;
1220                 } /* else if RAID[56], multiply by nr_data_stripes().
1221                    * Alternatively, just use rmap_len below instead of
1222                    * map->stripe_len */
1223
1224                 bytenr = ce->start + stripe_nr * rmap_len;
1225                 for (j = 0; j < nr; j++) {
1226                         if (buf[j] == bytenr)
1227                                 break;
1228                 }
1229                 if (j == nr)
1230                         buf[nr++] = bytenr;
1231         }
1232
1233         *logical = buf;
1234         *naddrs = nr;
1235         *stripe_len = rmap_len;
1236
1237         return 0;
1238 }
1239
1240 static inline int parity_smaller(u64 a, u64 b)
1241 {
1242         return a > b;
1243 }
1244
1245 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1246 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1247 {
1248         struct btrfs_bio_stripe s;
1249         int i;
1250         u64 l;
1251         int again = 1;
1252
1253         while (again) {
1254                 again = 0;
1255                 for (i = 0; i < bbio->num_stripes - 1; i++) {
1256                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
1257                                 s = bbio->stripes[i];
1258                                 l = raid_map[i];
1259                                 bbio->stripes[i] = bbio->stripes[i+1];
1260                                 raid_map[i] = raid_map[i+1];
1261                                 bbio->stripes[i+1] = s;
1262                                 raid_map[i+1] = l;
1263                                 again = 1;
1264                         }
1265                 }
1266         }
1267 }
1268
1269 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1270                     u64 logical, u64 *length,
1271                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1272                     u64 **raid_map_ret)
1273 {
1274         return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1275                                  multi_ret, mirror_num, raid_map_ret);
1276 }
1277
1278 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1279                     u64 logical, u64 *length, u64 *type,
1280                     struct btrfs_multi_bio **multi_ret, int mirror_num,
1281                     u64 **raid_map_ret)
1282 {
1283         struct cache_extent *ce;
1284         struct map_lookup *map;
1285         u64 offset;
1286         u64 stripe_offset;
1287         u64 stripe_nr;
1288         u64 *raid_map = NULL;
1289         int stripes_allocated = 8;
1290         int stripes_required = 1;
1291         int stripe_index;
1292         int i;
1293         struct btrfs_multi_bio *multi = NULL;
1294
1295         if (multi_ret && rw == READ) {
1296                 stripes_allocated = 1;
1297         }
1298 again:
1299         ce = search_cache_extent(&map_tree->cache_tree, logical);
1300         if (!ce) {
1301                 kfree(multi);
1302                 return -ENOENT;
1303         }
1304         if (ce->start > logical || ce->start + ce->size < logical) {
1305                 kfree(multi);
1306                 return -ENOENT;
1307         }
1308
1309         if (multi_ret) {
1310                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1311                                 GFP_NOFS);
1312                 if (!multi)
1313                         return -ENOMEM;
1314         }
1315         map = container_of(ce, struct map_lookup, ce);
1316         offset = logical - ce->start;
1317
1318         if (rw == WRITE) {
1319                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1320                                  BTRFS_BLOCK_GROUP_DUP)) {
1321                         stripes_required = map->num_stripes;
1322                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1323                         stripes_required = map->sub_stripes;
1324                 }
1325         }
1326         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1327             && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1328                     /* RAID[56] write or recovery. Return all stripes */
1329                     stripes_required = map->num_stripes;
1330
1331                     /* Only allocate the map if we've already got a large enough multi_ret */
1332                     if (stripes_allocated >= stripes_required) {
1333                             raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1334                             if (!raid_map) {
1335                                     kfree(multi);
1336                                     return -ENOMEM;
1337                             }
1338                     }
1339         }
1340
1341         /* if our multi bio struct is too small, back off and try again */
1342         if (multi_ret && stripes_allocated < stripes_required) {
1343                 stripes_allocated = stripes_required;
1344                 kfree(multi);
1345                 multi = NULL;
1346                 goto again;
1347         }
1348         stripe_nr = offset;
1349         /*
1350          * stripe_nr counts the total number of stripes we have to stride
1351          * to get to this block
1352          */
1353         stripe_nr = stripe_nr / map->stripe_len;
1354
1355         stripe_offset = stripe_nr * map->stripe_len;
1356         BUG_ON(offset < stripe_offset);
1357
1358         /* stripe_offset is the offset of this block in its stripe*/
1359         stripe_offset = offset - stripe_offset;
1360
1361         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1362                          BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1363                          BTRFS_BLOCK_GROUP_RAID10 |
1364                          BTRFS_BLOCK_GROUP_DUP)) {
1365                 /* we limit the length of each bio to what fits in a stripe */
1366                 *length = min_t(u64, ce->size - offset,
1367                               map->stripe_len - stripe_offset);
1368         } else {
1369                 *length = ce->size - offset;
1370         }
1371
1372         if (!multi_ret)
1373                 goto out;
1374
1375         multi->num_stripes = 1;
1376         stripe_index = 0;
1377         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1378                 if (rw == WRITE)
1379                         multi->num_stripes = map->num_stripes;
1380                 else if (mirror_num)
1381                         stripe_index = mirror_num - 1;
1382                 else
1383                         stripe_index = stripe_nr % map->num_stripes;
1384         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1385                 int factor = map->num_stripes / map->sub_stripes;
1386
1387                 stripe_index = stripe_nr % factor;
1388                 stripe_index *= map->sub_stripes;
1389
1390                 if (rw == WRITE)
1391                         multi->num_stripes = map->sub_stripes;
1392                 else if (mirror_num)
1393                         stripe_index += mirror_num - 1;
1394
1395                 stripe_nr = stripe_nr / factor;
1396         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1397                 if (rw == WRITE)
1398                         multi->num_stripes = map->num_stripes;
1399                 else if (mirror_num)
1400                         stripe_index = mirror_num - 1;
1401         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1402                                 BTRFS_BLOCK_GROUP_RAID6)) {
1403
1404                 if (raid_map) {
1405                         int rot;
1406                         u64 tmp;
1407                         u64 raid56_full_stripe_start;
1408                         u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1409
1410                         /*
1411                          * align the start of our data stripe in the logical
1412                          * address space
1413                          */
1414                         raid56_full_stripe_start = offset / full_stripe_len;
1415                         raid56_full_stripe_start *= full_stripe_len;
1416
1417                         /* get the data stripe number */
1418                         stripe_nr = raid56_full_stripe_start / map->stripe_len;
1419                         stripe_nr = stripe_nr / nr_data_stripes(map);
1420
1421                         /* Work out the disk rotation on this stripe-set */
1422                         rot = stripe_nr % map->num_stripes;
1423
1424                         /* Fill in the logical address of each stripe */
1425                         tmp = stripe_nr * nr_data_stripes(map);
1426
1427                         for (i = 0; i < nr_data_stripes(map); i++)
1428                                 raid_map[(i+rot) % map->num_stripes] =
1429                                         ce->start + (tmp + i) * map->stripe_len;
1430
1431                         raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1432                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1433                                 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1434
1435                         *length = map->stripe_len;
1436                         stripe_index = 0;
1437                         stripe_offset = 0;
1438                         multi->num_stripes = map->num_stripes;
1439                 } else {
1440                         stripe_index = stripe_nr % nr_data_stripes(map);
1441                         stripe_nr = stripe_nr / nr_data_stripes(map);
1442
1443                         /*
1444                          * Mirror #0 or #1 means the original data block.
1445                          * Mirror #2 is RAID5 parity block.
1446                          * Mirror #3 is RAID6 Q block.
1447                          */
1448                         if (mirror_num > 1)
1449                                 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1450
1451                         /* We distribute the parity blocks across stripes */
1452                         stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1453                 }
1454         } else {
1455                 /*
1456                  * after this do_div call, stripe_nr is the number of stripes
1457                  * on this device we have to walk to find the data, and
1458                  * stripe_index is the number of our device in the stripe array
1459                  */
1460                 stripe_index = stripe_nr % map->num_stripes;
1461                 stripe_nr = stripe_nr / map->num_stripes;
1462         }
1463         BUG_ON(stripe_index >= map->num_stripes);
1464
1465         for (i = 0; i < multi->num_stripes; i++) {
1466                 multi->stripes[i].physical =
1467                         map->stripes[stripe_index].physical + stripe_offset +
1468                         stripe_nr * map->stripe_len;
1469                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1470                 stripe_index++;
1471         }
1472         *multi_ret = multi;
1473
1474         if (type)
1475                 *type = map->type;
1476
1477         if (raid_map) {
1478                 sort_parity_stripes(multi, raid_map);
1479                 *raid_map_ret = raid_map;
1480         }
1481 out:
1482         return 0;
1483 }
1484
1485 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1486                                        u8 *uuid, u8 *fsid)
1487 {
1488         struct btrfs_device *device;
1489         struct btrfs_fs_devices *cur_devices;
1490
1491         cur_devices = root->fs_info->fs_devices;
1492         while (cur_devices) {
1493                 if (!fsid ||
1494                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1495                         device = __find_device(&cur_devices->devices,
1496                                                devid, uuid);
1497                         if (device)
1498                                 return device;
1499                 }
1500                 cur_devices = cur_devices->seed;
1501         }
1502         return NULL;
1503 }
1504
1505 struct btrfs_device *
1506 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1507                            u64 devid, int instance)
1508 {
1509         struct list_head *head = &fs_devices->devices;
1510         struct btrfs_device *dev;
1511         int num_found = 0;
1512
1513         list_for_each_entry(dev, head, dev_list) {
1514                 if (dev->devid == devid && num_found++ == instance)
1515                         return dev;
1516         }
1517         return NULL;
1518 }
1519
1520 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1521 {
1522         struct cache_extent *ce;
1523         struct map_lookup *map;
1524         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1525         int readonly = 0;
1526         int i;
1527
1528         /*
1529          * During chunk recovering, we may fail to find block group's
1530          * corresponding chunk, we will rebuild it later
1531          */
1532         ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1533         if (!root->fs_info->is_chunk_recover)
1534                 BUG_ON(!ce);
1535         else
1536                 return 0;
1537
1538         map = container_of(ce, struct map_lookup, ce);
1539         for (i = 0; i < map->num_stripes; i++) {
1540                 if (!map->stripes[i].dev->writeable) {
1541                         readonly = 1;
1542                         break;
1543                 }
1544         }
1545
1546         return readonly;
1547 }
1548
1549 static struct btrfs_device *fill_missing_device(u64 devid)
1550 {
1551         struct btrfs_device *device;
1552
1553         device = kzalloc(sizeof(*device), GFP_NOFS);
1554         device->devid = devid;
1555         device->fd = -1;
1556         return device;
1557 }
1558
1559 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1560                           struct extent_buffer *leaf,
1561                           struct btrfs_chunk *chunk)
1562 {
1563         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1564         struct map_lookup *map;
1565         struct cache_extent *ce;
1566         u64 logical;
1567         u64 length;
1568         u64 devid;
1569         u8 uuid[BTRFS_UUID_SIZE];
1570         int num_stripes;
1571         int ret;
1572         int i;
1573
1574         logical = key->offset;
1575         length = btrfs_chunk_length(leaf, chunk);
1576
1577         ce = search_cache_extent(&map_tree->cache_tree, logical);
1578
1579         /* already mapped? */
1580         if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1581                 return 0;
1582         }
1583
1584         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1585         map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1586         if (!map)
1587                 return -ENOMEM;
1588
1589         map->ce.start = logical;
1590         map->ce.size = length;
1591         map->num_stripes = num_stripes;
1592         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1593         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1594         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1595         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1596         map->type = btrfs_chunk_type(leaf, chunk);
1597         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1598
1599         for (i = 0; i < num_stripes; i++) {
1600                 map->stripes[i].physical =
1601                         btrfs_stripe_offset_nr(leaf, chunk, i);
1602                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1603                 read_extent_buffer(leaf, uuid, (unsigned long)
1604                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1605                                    BTRFS_UUID_SIZE);
1606                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1607                                                         NULL);
1608                 if (!map->stripes[i].dev) {
1609                         map->stripes[i].dev = fill_missing_device(devid);
1610                         printf("warning, device %llu is missing\n",
1611                                (unsigned long long)devid);
1612                 }
1613
1614         }
1615         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1616         BUG_ON(ret);
1617
1618         return 0;
1619 }
1620
1621 static int fill_device_from_item(struct extent_buffer *leaf,
1622                                  struct btrfs_dev_item *dev_item,
1623                                  struct btrfs_device *device)
1624 {
1625         unsigned long ptr;
1626
1627         device->devid = btrfs_device_id(leaf, dev_item);
1628         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1629         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1630         device->type = btrfs_device_type(leaf, dev_item);
1631         device->io_align = btrfs_device_io_align(leaf, dev_item);
1632         device->io_width = btrfs_device_io_width(leaf, dev_item);
1633         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1634
1635         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1636         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1637
1638         return 0;
1639 }
1640
1641 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1642 {
1643         struct btrfs_fs_devices *fs_devices;
1644         int ret;
1645
1646         fs_devices = root->fs_info->fs_devices->seed;
1647         while (fs_devices) {
1648                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1649                         ret = 0;
1650                         goto out;
1651                 }
1652                 fs_devices = fs_devices->seed;
1653         }
1654
1655         fs_devices = find_fsid(fsid);
1656         if (!fs_devices) {
1657                 ret = -ENOENT;
1658                 goto out;
1659         }
1660
1661         ret = btrfs_open_devices(fs_devices, O_RDONLY);
1662         if (ret)
1663                 goto out;
1664
1665         fs_devices->seed = root->fs_info->fs_devices->seed;
1666         root->fs_info->fs_devices->seed = fs_devices;
1667 out:
1668         return ret;
1669 }
1670
1671 static int read_one_dev(struct btrfs_root *root,
1672                         struct extent_buffer *leaf,
1673                         struct btrfs_dev_item *dev_item)
1674 {
1675         struct btrfs_device *device;
1676         u64 devid;
1677         int ret = 0;
1678         u8 fs_uuid[BTRFS_UUID_SIZE];
1679         u8 dev_uuid[BTRFS_UUID_SIZE];
1680
1681         devid = btrfs_device_id(leaf, dev_item);
1682         read_extent_buffer(leaf, dev_uuid,
1683                            (unsigned long)btrfs_device_uuid(dev_item),
1684                            BTRFS_UUID_SIZE);
1685         read_extent_buffer(leaf, fs_uuid,
1686                            (unsigned long)btrfs_device_fsid(dev_item),
1687                            BTRFS_UUID_SIZE);
1688
1689         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1690                 ret = open_seed_devices(root, fs_uuid);
1691                 if (ret)
1692                         return ret;
1693         }
1694
1695         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1696         if (!device) {
1697                 printk("warning devid %llu not found already\n",
1698                         (unsigned long long)devid);
1699                 device = kzalloc(sizeof(*device), GFP_NOFS);
1700                 if (!device)
1701                         return -ENOMEM;
1702                 device->fd = -1;
1703                 list_add(&device->dev_list,
1704                          &root->fs_info->fs_devices->devices);
1705         }
1706
1707         fill_device_from_item(leaf, dev_item, device);
1708         device->dev_root = root->fs_info->dev_root;
1709         return ret;
1710 }
1711
1712 int btrfs_read_sys_array(struct btrfs_root *root)
1713 {
1714         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1715         struct extent_buffer *sb;
1716         struct btrfs_disk_key *disk_key;
1717         struct btrfs_chunk *chunk;
1718         struct btrfs_key key;
1719         u32 num_stripes;
1720         u32 len = 0;
1721         u8 *ptr;
1722         u8 *array_end;
1723         int ret = 0;
1724
1725         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1726                                           BTRFS_SUPER_INFO_SIZE);
1727         if (!sb)
1728                 return -ENOMEM;
1729         btrfs_set_buffer_uptodate(sb);
1730         write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1731         array_end = ((u8 *)super_copy->sys_chunk_array) +
1732                     btrfs_super_sys_array_size(super_copy);
1733
1734         /*
1735          * we do this loop twice, once for the device items and
1736          * once for all of the chunks.  This way there are device
1737          * structs filled in for every chunk
1738          */
1739         ptr = super_copy->sys_chunk_array;
1740
1741         while (ptr < array_end) {
1742                 disk_key = (struct btrfs_disk_key *)ptr;
1743                 btrfs_disk_key_to_cpu(&key, disk_key);
1744
1745                 len = sizeof(*disk_key);
1746                 ptr += len;
1747
1748                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1749                         chunk = (struct btrfs_chunk *)(ptr - (u8 *)super_copy);
1750                         ret = read_one_chunk(root, &key, sb, chunk);
1751                         if (ret)
1752                                 break;
1753                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1754                         len = btrfs_chunk_item_size(num_stripes);
1755                 } else {
1756                         BUG();
1757                 }
1758                 ptr += len;
1759         }
1760         free_extent_buffer(sb);
1761         return ret;
1762 }
1763
1764 int btrfs_read_chunk_tree(struct btrfs_root *root)
1765 {
1766         struct btrfs_path *path;
1767         struct extent_buffer *leaf;
1768         struct btrfs_key key;
1769         struct btrfs_key found_key;
1770         int ret;
1771         int slot;
1772
1773         root = root->fs_info->chunk_root;
1774
1775         path = btrfs_alloc_path();
1776         if (!path)
1777                 return -ENOMEM;
1778
1779         /*
1780          * Read all device items, and then all the chunk items. All
1781          * device items are found before any chunk item (their object id
1782          * is smaller than the lowest possible object id for a chunk
1783          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1784          */
1785         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1786         key.offset = 0;
1787         key.type = 0;
1788         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1789         if (ret < 0)
1790                 goto error;
1791         while(1) {
1792                 leaf = path->nodes[0];
1793                 slot = path->slots[0];
1794                 if (slot >= btrfs_header_nritems(leaf)) {
1795                         ret = btrfs_next_leaf(root, path);
1796                         if (ret == 0)
1797                                 continue;
1798                         if (ret < 0)
1799                                 goto error;
1800                         break;
1801                 }
1802                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1803                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1804                         struct btrfs_dev_item *dev_item;
1805                         dev_item = btrfs_item_ptr(leaf, slot,
1806                                                   struct btrfs_dev_item);
1807                         ret = read_one_dev(root, leaf, dev_item);
1808                         BUG_ON(ret);
1809                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1810                         struct btrfs_chunk *chunk;
1811                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1812                         ret = read_one_chunk(root, &found_key, leaf, chunk);
1813                         BUG_ON(ret);
1814                 }
1815                 path->slots[0]++;
1816         }
1817
1818         ret = 0;
1819 error:
1820         btrfs_free_path(path);
1821         return ret;
1822 }
1823
1824 struct list_head *btrfs_scanned_uuids(void)
1825 {
1826         return &fs_uuids;
1827 }
1828
1829 static int rmw_eb(struct btrfs_fs_info *info,
1830                   struct extent_buffer *eb, struct extent_buffer *orig_eb)
1831 {
1832         int ret;
1833         unsigned long orig_off = 0;
1834         unsigned long dest_off = 0;
1835         unsigned long copy_len = eb->len;
1836
1837         ret = read_whole_eb(info, eb, 0);
1838         if (ret)
1839                 return ret;
1840
1841         if (eb->start + eb->len <= orig_eb->start ||
1842             eb->start >= orig_eb->start + orig_eb->len)
1843                 return 0;
1844         /*
1845          * | ----- orig_eb ------- |
1846          *         | ----- stripe -------  |
1847          *         | ----- orig_eb ------- |
1848          *              | ----- orig_eb ------- |
1849          */
1850         if (eb->start > orig_eb->start)
1851                 orig_off = eb->start - orig_eb->start;
1852         if (orig_eb->start > eb->start)
1853                 dest_off = orig_eb->start - eb->start;
1854
1855         if (copy_len > orig_eb->len - orig_off)
1856                 copy_len = orig_eb->len - orig_off;
1857         if (copy_len > eb->len - dest_off)
1858                 copy_len = eb->len - dest_off;
1859
1860         memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1861         return 0;
1862 }
1863
1864 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1865                                 struct extent_buffer *orig_eb,
1866                                struct extent_buffer **ebs,
1867                                u64 stripe_len, u64 *raid_map,
1868                                int num_stripes)
1869 {
1870         struct extent_buffer *eb;
1871         u64 start = orig_eb->start;
1872         u64 this_eb_start;
1873         int i;
1874         int ret;
1875
1876         for (i = 0; i < num_stripes; i++) {
1877                 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1878                         break;
1879
1880                 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1881                 if (!eb)
1882                         BUG();
1883                 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1884
1885                 eb->start = raid_map[i];
1886                 eb->len = stripe_len;
1887                 eb->refs = 1;
1888                 eb->flags = 0;
1889                 eb->fd = -1;
1890                 eb->dev_bytenr = (u64)-1;
1891
1892                 this_eb_start = raid_map[i];
1893
1894                 if (start > this_eb_start ||
1895                     start + orig_eb->len < this_eb_start + stripe_len) {
1896                         ret = rmw_eb(info, eb, orig_eb);
1897                         BUG_ON(ret);
1898                 } else {
1899                         memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1900                 }
1901                 ebs[i] = eb;
1902         }
1903 }
1904
1905 int write_raid56_with_parity(struct btrfs_fs_info *info,
1906                              struct extent_buffer *eb,
1907                              struct btrfs_multi_bio *multi,
1908                              u64 stripe_len, u64 *raid_map)
1909 {
1910         struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
1911         int i;
1912         int j;
1913         int ret;
1914         int alloc_size = eb->len;
1915
1916         ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
1917         BUG_ON(!ebs);
1918
1919         if (stripe_len > alloc_size)
1920                 alloc_size = stripe_len;
1921
1922         split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1923                             multi->num_stripes);
1924
1925         for (i = 0; i < multi->num_stripes; i++) {
1926                 struct extent_buffer *new_eb;
1927                 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1928                         ebs[i]->dev_bytenr = multi->stripes[i].physical;
1929                         ebs[i]->fd = multi->stripes[i].dev->fd;
1930                         multi->stripes[i].dev->total_ios++;
1931                         BUG_ON(ebs[i]->start != raid_map[i]);
1932                         continue;
1933                 }
1934                 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1935                 BUG_ON(!new_eb);
1936                 new_eb->dev_bytenr = multi->stripes[i].physical;
1937                 new_eb->fd = multi->stripes[i].dev->fd;
1938                 multi->stripes[i].dev->total_ios++;
1939                 new_eb->len = stripe_len;
1940
1941                 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1942                         p_eb = new_eb;
1943                 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1944                         q_eb = new_eb;
1945         }
1946         if (q_eb) {
1947                 void **pointers;
1948
1949                 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
1950                                    GFP_NOFS);
1951                 BUG_ON(!pointers);
1952
1953                 ebs[multi->num_stripes - 2] = p_eb;
1954                 ebs[multi->num_stripes - 1] = q_eb;
1955
1956                 for (i = 0; i < multi->num_stripes; i++)
1957                         pointers[i] = ebs[i]->data;
1958
1959                 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1960                 kfree(pointers);
1961         } else {
1962                 ebs[multi->num_stripes - 1] = p_eb;
1963                 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1964                 for (j = 1; j < multi->num_stripes - 1; j++) {
1965                         for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1966                                 *(unsigned long *)(p_eb->data + i) ^=
1967                                         *(unsigned long *)(ebs[j]->data + i);
1968                         }
1969                 }
1970         }
1971
1972         for (i = 0; i < multi->num_stripes; i++) {
1973                 ret = write_extent_to_disk(ebs[i]);
1974                 BUG_ON(ret);
1975                 if (ebs[i] != eb)
1976                         kfree(ebs[i]);
1977         }
1978
1979         kfree(ebs);
1980
1981         return 0;
1982 }