fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/random.h>
  23 #include <asm/div64.h>
  24 #include "ctree.h"
  25 #include "extent_map.h"
  26 #include "disk-io.h"
  27 #include "transaction.h"
  28 #include "print-tree.h"
  29 #include "volumes.h"
  30
  31 struct map_lookup {
  32         u64 type;
  33         int io_align;
  34         int io_width;
  35         int stripe_len;
  36         int sector_size;
  37         int num_stripes;
  38         int sub_stripes;
  39         struct btrfs_bio_stripe stripes[];
  40 };
  41
  42 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  43                             (sizeof(struct btrfs_bio_stripe) * (n)))
  44
  45 static DEFINE_MUTEX(uuid_mutex);
  46 static LIST_HEAD(fs_uuids);
  47
  48 int btrfs_cleanup_fs_uuids(void)
  49 {
  50         struct btrfs_fs_devices *fs_devices;
  51         struct list_head *uuid_cur;
  52         struct list_head *devices_cur;
  53         struct btrfs_device *dev;
  54
  55         list_for_each(uuid_cur, &fs_uuids) {
  56                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  57                                         list);
  58                 while(!list_empty(&fs_devices->devices)) {
  59                         devices_cur = fs_devices->devices.next;
  60                         dev = list_entry(devices_cur, struct btrfs_device,
  61                                          dev_list);
  62                         if (dev->bdev) {
  63                                 close_bdev_excl(dev->bdev);
  64                         }
  65                         list_del(&dev->dev_list);
  66                         kfree(dev);
  67                 }
  68         }
  69         return 0;
  70 }
  71
  72 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  73                                           u8 *uuid)
  74 {
  75         struct btrfs_device *dev;
  76         struct list_head *cur;
  77
  78         list_for_each(cur, head) {
  79                 dev = list_entry(cur, struct btrfs_device, dev_list);
  80                 if (dev->devid == devid &&
  81                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  82                         return dev;
  83                 }
  84         }
  85         return NULL;
  86 }
  87
  88 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
  89 {
  90         struct list_head *cur;
  91         struct btrfs_fs_devices *fs_devices;
  92
  93         list_for_each(cur, &fs_uuids) {
  94                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
  95                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  96                         return fs_devices;
  97         }
  98         return NULL;
  99 }
 100
 101 static int device_list_add(const char *path,
 102                            struct btrfs_super_block *disk_super,
 103                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 104 {
 105         struct btrfs_device *device;
 106         struct btrfs_fs_devices *fs_devices;
 107         u64 found_transid = btrfs_super_generation(disk_super);
 108
 109         fs_devices = find_fsid(disk_super->fsid);
 110         if (!fs_devices) {
 111                 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
 112                 if (!fs_devices)
 113                         return -ENOMEM;
 114                 INIT_LIST_HEAD(&fs_devices->devices);
 115                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 116                 list_add(&fs_devices->list, &fs_uuids);
 117                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 118                 fs_devices->latest_devid = devid;
 119                 fs_devices->latest_trans = found_transid;
 120                 fs_devices->lowest_devid = (u64)-1;
 121                 fs_devices->num_devices = 0;
 122                 device = NULL;
 123         } else {
 124                 device = __find_device(&fs_devices->devices, devid,
 125                                        disk_super->dev_item.uuid);
 126         }
 127         if (!device) {
 128                 device = kzalloc(sizeof(*device), GFP_NOFS);
 129                 if (!device) {
 130                         /* we can safely leave the fs_devices entry around */
 131                         return -ENOMEM;
 132                 }
 133                 device->devid = devid;
 134                 memcpy(device->uuid, disk_super->dev_item.uuid,
 135                        BTRFS_UUID_SIZE);
 136                 device->barriers = 1;
 137                 spin_lock_init(&device->io_lock);
 138                 device->name = kstrdup(path, GFP_NOFS);
 139                 if (!device->name) {
 140                         kfree(device);
 141                         return -ENOMEM;
 142                 }
 143                 list_add(&device->dev_list, &fs_devices->devices);
 144                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 145                 fs_devices->num_devices++;
 146         }
 147
 148         if (found_transid > fs_devices->latest_trans) {
 149                 fs_devices->latest_devid = devid;
 150                 fs_devices->latest_trans = found_transid;
 151         }
 152         if (fs_devices->lowest_devid > devid) {
 153                 fs_devices->lowest_devid = devid;
 154         }
 155         *fs_devices_ret = fs_devices;
 156         return 0;
 157 }
 158
 159 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 160 {
 161         struct list_head *head = &fs_devices->devices;
 162         struct list_head *cur;
 163         struct btrfs_device *device;
 164
 165         mutex_lock(&uuid_mutex);
 166         list_for_each(cur, head) {
 167                 device = list_entry(cur, struct btrfs_device, dev_list);
 168                 if (device->bdev) {
 169                         close_bdev_excl(device->bdev);
 170                 }
 171                 device->bdev = NULL;
 172         }
 173         mutex_unlock(&uuid_mutex);
 174         return 0;
 175 }
 176
 177 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 178                        int flags, void *holder)
 179 {
 180         struct block_device *bdev;
 181         struct list_head *head = &fs_devices->devices;
 182         struct list_head *cur;
 183         struct btrfs_device *device;
 184         int ret;
 185
 186         mutex_lock(&uuid_mutex);
 187         list_for_each(cur, head) {
 188                 device = list_entry(cur, struct btrfs_device, dev_list);
 189                 bdev = open_bdev_excl(device->name, flags, holder);
 190
 191                 if (IS_ERR(bdev)) {
 192                         printk("open %s failed\n", device->name);
 193                         ret = PTR_ERR(bdev);
 194                         goto fail;
 195                 }
 196                 if (device->devid == fs_devices->latest_devid)
 197                         fs_devices->latest_bdev = bdev;
 198                 if (device->devid == fs_devices->lowest_devid) {
 199                         fs_devices->lowest_bdev = bdev;
 200                 }
 201                 device->bdev = bdev;
 202         }
 203         mutex_unlock(&uuid_mutex);
 204         return 0;
 205 fail:
 206         mutex_unlock(&uuid_mutex);
 207         btrfs_close_devices(fs_devices);
 208         return ret;
 209 }
 210
 211 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 212                           struct btrfs_fs_devices **fs_devices_ret)
 213 {
 214         struct btrfs_super_block *disk_super;
 215         struct block_device *bdev;
 216         struct buffer_head *bh;
 217         int ret;
 218         u64 devid;
 219         u64 transid;
 220
 221         mutex_lock(&uuid_mutex);
 222
 223         bdev = open_bdev_excl(path, flags, holder);
 224
 225         if (IS_ERR(bdev)) {
 226                 ret = PTR_ERR(bdev);
 227                 goto error;
 228         }
 229
 230         ret = set_blocksize(bdev, 4096);
 231         if (ret)
 232                 goto error_close;
 233         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 234         if (!bh) {
 235                 ret = -EIO;
 236                 goto error_close;
 237         }
 238         disk_super = (struct btrfs_super_block *)bh->b_data;
 239         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 240             sizeof(disk_super->magic))) {
 241                 ret = -EINVAL;
 242                 goto error_brelse;
 243         }
 244         devid = le64_to_cpu(disk_super->dev_item.devid);
 245         transid = btrfs_super_generation(disk_super);
 246         if (disk_super->label[0])
 247                 printk("device label %s ", disk_super->label);
 248         else {
 249                 /* FIXME, make a readl uuid parser */
 250                 printk("device fsid %llx-%llx ",
 251                        *(unsigned long long *)disk_super->fsid,
 252                        *(unsigned long long *)(disk_super->fsid + 8));
 253         }
 254         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 255         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 256
 257 error_brelse:
 258         brelse(bh);
 259 error_close:
 260         close_bdev_excl(bdev);
 261 error:
 262         mutex_unlock(&uuid_mutex);
 263         return ret;
 264 }
 265
 266 /*
 267  * this uses a pretty simple search, the expectation is that it is
 268  * called very infrequently and that a given device has a small number
 269  * of extents
 270  */
 271 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 272                                 struct btrfs_device *device,
 273                                 struct btrfs_path *path,
 274                                 u64 num_bytes, u64 *start)
 275 {
 276         struct btrfs_key key;
 277         struct btrfs_root *root = device->dev_root;
 278         struct btrfs_dev_extent *dev_extent = NULL;
 279         u64 hole_size = 0;
 280         u64 last_byte = 0;
 281         u64 search_start = 0;
 282         u64 search_end = device->total_bytes;
 283         int ret;
 284         int slot = 0;
 285         int start_found;
 286         struct extent_buffer *l;
 287
 288         start_found = 0;
 289         path->reada = 2;
 290
 291         /* FIXME use last free of some kind */
 292
 293         /* we don't want to overwrite the superblock on the drive,
 294          * so we make sure to start at an offset of at least 1MB
 295          */
 296         search_start = max((u64)1024 * 1024, search_start);
 297
 298         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 299                 search_start = max(root->fs_info->alloc_start, search_start);
 300
 301         key.objectid = device->devid;
 302         key.offset = search_start;
 303         key.type = BTRFS_DEV_EXTENT_KEY;
 304         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 305         if (ret < 0)
 306                 goto error;
 307         ret = btrfs_previous_item(root, path, 0, key.type);
 308         if (ret < 0)
 309                 goto error;
 310         l = path->nodes[0];
 311         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 312         while (1) {
 313                 l = path->nodes[0];
 314                 slot = path->slots[0];
 315                 if (slot >= btrfs_header_nritems(l)) {
 316                         ret = btrfs_next_leaf(root, path);
 317                         if (ret == 0)
 318                                 continue;
 319                         if (ret < 0)
 320                                 goto error;
 321 no_more_items:
 322                         if (!start_found) {
 323                                 if (search_start >= search_end) {
 324                                         ret = -ENOSPC;
 325                                         goto error;
 326                                 }
 327                                 *start = search_start;
 328                                 start_found = 1;
 329                                 goto check_pending;
 330                         }
 331                         *start = last_byte > search_start ?
 332                                 last_byte : search_start;
 333                         if (search_end <= *start) {
 334                                 ret = -ENOSPC;
 335                                 goto error;
 336                         }
 337                         goto check_pending;
 338                 }
 339                 btrfs_item_key_to_cpu(l, &key, slot);
 340
 341                 if (key.objectid < device->devid)
 342                         goto next;
 343
 344                 if (key.objectid > device->devid)
 345                         goto no_more_items;
 346
 347                 if (key.offset >= search_start && key.offset > last_byte &&
 348                     start_found) {
 349                         if (last_byte < search_start)
 350                                 last_byte = search_start;
 351                         hole_size = key.offset - last_byte;
 352                         if (key.offset > last_byte &&
 353                             hole_size >= num_bytes) {
 354                                 *start = last_byte;
 355                                 goto check_pending;
 356                         }
 357                 }
 358                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 359                         goto next;
 360                 }
 361
 362                 start_found = 1;
 363                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 364                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 365 next:
 366                 path->slots[0]++;
 367                 cond_resched();
 368         }
 369 check_pending:
 370         /* we have to make sure we didn't find an extent that has already
 371          * been allocated by the map tree or the original allocation
 372          */
 373         btrfs_release_path(root, path);
 374         BUG_ON(*start < search_start);
 375
 376         if (*start + num_bytes > search_end) {
 377                 ret = -ENOSPC;
 378                 goto error;
 379         }
 380         /* check for pending inserts here */
 381         return 0;
 382
 383 error:
 384         btrfs_release_path(root, path);
 385         return ret;
 386 }
 387
 388 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 389                           struct btrfs_device *device,
 390                           u64 start)
 391 {
 392         int ret;
 393         struct btrfs_path *path;
 394         struct btrfs_root *root = device->dev_root;
 395         struct btrfs_key key;
 396
 397         path = btrfs_alloc_path();
 398         if (!path)
 399                 return -ENOMEM;
 400
 401         key.objectid = device->devid;
 402         key.offset = start;
 403         key.type = BTRFS_DEV_EXTENT_KEY;
 404
 405         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 406         BUG_ON(ret);
 407
 408         ret = btrfs_del_item(trans, root, path);
 409         BUG_ON(ret);
 410
 411         btrfs_free_path(path);
 412         return ret;
 413 }
 414
 415 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 416                            struct btrfs_device *device,
 417                            u64 chunk_tree, u64 chunk_objectid,
 418                            u64 chunk_offset,
 419                            u64 num_bytes, u64 *start)
 420 {
 421         int ret;
 422         struct btrfs_path *path;
 423         struct btrfs_root *root = device->dev_root;
 424         struct btrfs_dev_extent *extent;
 425         struct extent_buffer *leaf;
 426         struct btrfs_key key;
 427
 428         path = btrfs_alloc_path();
 429         if (!path)
 430                 return -ENOMEM;
 431
 432         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 433         if (ret) {
 434                 goto err;
 435         }
 436
 437         key.objectid = device->devid;
 438         key.offset = *start;
 439         key.type = BTRFS_DEV_EXTENT_KEY;
 440         ret = btrfs_insert_empty_item(trans, root, path, &key,
 441                                       sizeof(*extent));
 442         BUG_ON(ret);
 443
 444         leaf = path->nodes[0];
 445         extent = btrfs_item_ptr(leaf, path->slots[0],
 446                                 struct btrfs_dev_extent);
 447         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 448         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 449         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 450
 451         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 452                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 453                     BTRFS_UUID_SIZE);
 454
 455         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 456         btrfs_mark_buffer_dirty(leaf);
 457 err:
 458         btrfs_free_path(path);
 459         return ret;
 460 }
 461
 462 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 463 {
 464         struct btrfs_path *path;
 465         int ret;
 466         struct btrfs_key key;
 467         struct btrfs_chunk *chunk;
 468         struct btrfs_key found_key;
 469
 470         path = btrfs_alloc_path();
 471         BUG_ON(!path);
 472
 473         key.objectid = objectid;
 474         key.offset = (u64)-1;
 475         key.type = BTRFS_CHUNK_ITEM_KEY;
 476
 477         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 478         if (ret < 0)
 479                 goto error;
 480
 481         BUG_ON(ret == 0);
 482
 483         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 484         if (ret) {
 485                 *offset = 0;
 486         } else {
 487                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 488                                       path->slots[0]);
 489                 if (found_key.objectid != objectid)
 490                         *offset = 0;
 491                 else {
 492                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 493                                                struct btrfs_chunk);
 494                         *offset = found_key.offset +
 495                                 btrfs_chunk_length(path->nodes[0], chunk);
 496                 }
 497         }
 498         ret = 0;
 499 error:
 500         btrfs_free_path(path);
 501         return ret;
 502 }
 503
 504 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 505                            u64 *objectid)
 506 {
 507         int ret;
 508         struct btrfs_key key;
 509         struct btrfs_key found_key;
 510
 511         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 512         key.type = BTRFS_DEV_ITEM_KEY;
 513         key.offset = (u64)-1;
 514
 515         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 516         if (ret < 0)
 517                 goto error;
 518
 519         BUG_ON(ret == 0);
 520
 521         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 522                                   BTRFS_DEV_ITEM_KEY);
 523         if (ret) {
 524                 *objectid = 1;
 525         } else {
 526                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 527                                       path->slots[0]);
 528                 *objectid = found_key.offset + 1;
 529         }
 530         ret = 0;
 531 error:
 532         btrfs_release_path(root, path);
 533         return ret;
 534 }
 535
 536 /*
 537  * the device information is stored in the chunk root
 538  * the btrfs_device struct should be fully filled in
 539  */
 540 int btrfs_add_device(struct btrfs_trans_handle *trans,
 541                      struct btrfs_root *root,
 542                      struct btrfs_device *device)
 543 {
 544         int ret;
 545         struct btrfs_path *path;
 546         struct btrfs_dev_item *dev_item;
 547         struct extent_buffer *leaf;
 548         struct btrfs_key key;
 549         unsigned long ptr;
 550         u64 free_devid;
 551
 552         root = root->fs_info->chunk_root;
 553
 554         path = btrfs_alloc_path();
 555         if (!path)
 556                 return -ENOMEM;
 557
 558         ret = find_next_devid(root, path, &free_devid);
 559         if (ret)
 560                 goto out;
 561
 562         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 563         key.type = BTRFS_DEV_ITEM_KEY;
 564         key.offset = free_devid;
 565
 566         ret = btrfs_insert_empty_item(trans, root, path, &key,
 567                                       sizeof(*dev_item));
 568         if (ret)
 569                 goto out;
 570
 571         leaf = path->nodes[0];
 572         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 573
 574         device->devid = free_devid;
 575         btrfs_set_device_id(leaf, dev_item, device->devid);
 576         btrfs_set_device_type(leaf, dev_item, device->type);
 577         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 578         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 579         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 580         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 581         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 582         btrfs_set_device_group(leaf, dev_item, 0);
 583         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 584         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 585
 586         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 587         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 588         btrfs_mark_buffer_dirty(leaf);
 589         ret = 0;
 590
 591 out:
 592         btrfs_free_path(path);
 593         return ret;
 594 }
 595
 596 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 597 {
 598         struct btrfs_trans_handle *trans;
 599         struct btrfs_device *device;
 600         struct block_device *bdev;
 601         struct list_head *cur;
 602         struct list_head *devices;
 603         u64 total_bytes;
 604         int ret = 0;
 605
 606
 607         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
 608         if (!bdev) {
 609                 return -EIO;
 610         }
 611         mutex_lock(&root->fs_info->fs_mutex);
 612         trans = btrfs_start_transaction(root, 1);
 613         devices = &root->fs_info->fs_devices->devices;
 614         list_for_each(cur, devices) {
 615                 device = list_entry(cur, struct btrfs_device, dev_list);
 616                 if (device->bdev == bdev) {
 617                         ret = -EEXIST;
 618                         goto out;
 619                 }
 620         }
 621
 622         device = kzalloc(sizeof(*device), GFP_NOFS);
 623         if (!device) {
 624                 /* we can safely leave the fs_devices entry around */
 625                 ret = -ENOMEM;
 626                 goto out_close_bdev;
 627         }
 628
 629         device->barriers = 1;
 630         generate_random_uuid(device->uuid);
 631         spin_lock_init(&device->io_lock);
 632         device->name = kstrdup(device_path, GFP_NOFS);
 633         if (!device->name) {
 634                 kfree(device);
 635                 goto out_close_bdev;
 636         }
 637         device->io_width = root->sectorsize;
 638         device->io_align = root->sectorsize;
 639         device->sector_size = root->sectorsize;
 640         device->total_bytes = i_size_read(bdev->bd_inode);
 641         device->dev_root = root->fs_info->dev_root;
 642         device->bdev = bdev;
 643
 644         ret = btrfs_add_device(trans, root, device);
 645         if (ret)
 646                 goto out_close_bdev;
 647
 648         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 649         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
 650                                     total_bytes + device->total_bytes);
 651
 652         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 653         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 654                                     total_bytes + 1);
 655
 656         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
 657         list_add(&device->dev_alloc_list,
 658                  &root->fs_info->fs_devices->alloc_list);
 659         root->fs_info->fs_devices->num_devices++;
 660 out:
 661         btrfs_end_transaction(trans, root);
 662         mutex_unlock(&root->fs_info->fs_mutex);
 663         return ret;
 664
 665 out_close_bdev:
 666         close_bdev_excl(bdev);
 667         goto out;
 668 }
 669
 670 int btrfs_update_device(struct btrfs_trans_handle *trans,
 671                         struct btrfs_device *device)
 672 {
 673         int ret;
 674         struct btrfs_path *path;
 675         struct btrfs_root *root;
 676         struct btrfs_dev_item *dev_item;
 677         struct extent_buffer *leaf;
 678         struct btrfs_key key;
 679
 680         root = device->dev_root->fs_info->chunk_root;
 681
 682         path = btrfs_alloc_path();
 683         if (!path)
 684                 return -ENOMEM;
 685
 686         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 687         key.type = BTRFS_DEV_ITEM_KEY;
 688         key.offset = device->devid;
 689
 690         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 691         if (ret < 0)
 692                 goto out;
 693
 694         if (ret > 0) {
 695                 ret = -ENOENT;
 696                 goto out;
 697         }
 698
 699         leaf = path->nodes[0];
 700         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 701
 702         btrfs_set_device_id(leaf, dev_item, device->devid);
 703         btrfs_set_device_type(leaf, dev_item, device->type);
 704         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 705         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 706         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 707         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 708         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 709         btrfs_mark_buffer_dirty(leaf);
 710
 711 out:
 712         btrfs_free_path(path);
 713         return ret;
 714 }
 715
 716 int btrfs_grow_device(struct btrfs_trans_handle *trans,
 717                       struct btrfs_device *device, u64 new_size)
 718 {
 719         struct btrfs_super_block *super_copy =
 720                 &device->dev_root->fs_info->super_copy;
 721         u64 old_total = btrfs_super_total_bytes(super_copy);
 722         u64 diff = new_size - device->total_bytes;
 723
 724         btrfs_set_super_total_bytes(super_copy, old_total + diff);
 725         return btrfs_update_device(trans, device);
 726 }
 727
 728 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
 729                             struct btrfs_root *root,
 730                             u64 chunk_tree, u64 chunk_objectid,
 731                             u64 chunk_offset)
 732 {
 733         int ret;
 734         struct btrfs_path *path;
 735         struct btrfs_key key;
 736
 737         root = root->fs_info->chunk_root;
 738         path = btrfs_alloc_path();
 739         if (!path)
 740                 return -ENOMEM;
 741
 742         key.objectid = chunk_objectid;
 743         key.offset = chunk_offset;
 744         key.type = BTRFS_CHUNK_ITEM_KEY;
 745
 746         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 747         BUG_ON(ret);
 748
 749         ret = btrfs_del_item(trans, root, path);
 750         BUG_ON(ret);
 751
 752         btrfs_free_path(path);
 753         return 0;
 754 }
 755
 756 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
 757                         chunk_offset)
 758 {
 759         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 760         struct btrfs_disk_key *disk_key;
 761         struct btrfs_chunk *chunk;
 762         u8 *ptr;
 763         int ret = 0;
 764         u32 num_stripes;
 765         u32 array_size;
 766         u32 len = 0;
 767         u32 cur;
 768         struct btrfs_key key;
 769
 770         array_size = btrfs_super_sys_array_size(super_copy);
 771
 772         ptr = super_copy->sys_chunk_array;
 773         cur = 0;
 774
 775         while (cur < array_size) {
 776                 disk_key = (struct btrfs_disk_key *)ptr;
 777                 btrfs_disk_key_to_cpu(&key, disk_key);
 778
 779                 len = sizeof(*disk_key);
 780
 781                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
 782                         chunk = (struct btrfs_chunk *)(ptr + len);
 783                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
 784                         len += btrfs_chunk_item_size(num_stripes);
 785                 } else {
 786                         ret = -EIO;
 787                         break;
 788                 }
 789                 if (key.objectid == chunk_objectid &&
 790                     key.offset == chunk_offset) {
 791                         memmove(ptr, ptr + len, array_size - (cur + len));
 792                         array_size -= len;
 793                         btrfs_set_super_sys_array_size(super_copy, array_size);
 794                 } else {
 795                         ptr += len;
 796                         cur += len;
 797                 }
 798         }
 799         return ret;
 800 }
 801
 802
 803 int btrfs_relocate_chunk(struct btrfs_root *root,
 804                          u64 chunk_tree, u64 chunk_objectid,
 805                          u64 chunk_offset)
 806 {
 807         struct extent_map_tree *em_tree;
 808         struct btrfs_root *extent_root;
 809         struct btrfs_trans_handle *trans;
 810         struct extent_map *em;
 811         struct map_lookup *map;
 812         int ret;
 813         int i;
 814
 815         root = root->fs_info->chunk_root;
 816         extent_root = root->fs_info->extent_root;
 817         em_tree = &root->fs_info->mapping_tree.map_tree;
 818
 819         /* step one, relocate all the extents inside this chunk */
 820         ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
 821         BUG_ON(ret);
 822
 823         trans = btrfs_start_transaction(root, 1);
 824         BUG_ON(!trans);
 825
 826         /*
 827          * step two, delete the device extents and the
 828          * chunk tree entries
 829          */
 830         spin_lock(&em_tree->lock);
 831         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
 832         spin_unlock(&em_tree->lock);
 833
 834         BUG_ON(em->start > chunk_offset || em->start + em->len < chunk_offset);
 835         map = (struct map_lookup *)em->bdev;
 836
 837         for (i = 0; i < map->num_stripes; i++) {
 838                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
 839                                             map->stripes[i].physical);
 840                 BUG_ON(ret);
 841         }
 842         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
 843                                chunk_offset);
 844
 845         BUG_ON(ret);
 846
 847         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
 848                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
 849                 BUG_ON(ret);
 850                 goto out;
 851         }
 852
 853
 854
 855         spin_lock(&em_tree->lock);
 856         remove_extent_mapping(em_tree, em);
 857         kfree(map);
 858         em->bdev = NULL;
 859
 860         /* once for the tree */
 861         free_extent_map(em);
 862         spin_unlock(&em_tree->lock);
 863
 864 out:
 865         /* once for us */
 866         free_extent_map(em);
 867
 868         btrfs_end_transaction(trans, root);
 869         return 0;
 870 }
 871
 872 /*
 873  * shrinking a device means finding all of the device extents past
 874  * the new size, and then following the back refs to the chunks.
 875  * The chunk relocation code actually frees the device extent
 876  */
 877 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
 878 {
 879         struct btrfs_trans_handle *trans;
 880         struct btrfs_root *root = device->dev_root;
 881         struct btrfs_dev_extent *dev_extent = NULL;
 882         struct btrfs_path *path;
 883         u64 length;
 884         u64 chunk_tree;
 885         u64 chunk_objectid;
 886         u64 chunk_offset;
 887         int ret;
 888         int slot;
 889         struct extent_buffer *l;
 890         struct btrfs_key key;
 891         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 892         u64 old_total = btrfs_super_total_bytes(super_copy);
 893         u64 diff = device->total_bytes - new_size;
 894
 895
 896         path = btrfs_alloc_path();
 897         if (!path)
 898                 return -ENOMEM;
 899
 900         trans = btrfs_start_transaction(root, 1);
 901         if (!trans) {
 902                 ret = -ENOMEM;
 903                 goto done;
 904         }
 905
 906         path->reada = 2;
 907
 908         device->total_bytes = new_size;
 909         ret = btrfs_update_device(trans, device);
 910         if (ret) {
 911                 btrfs_end_transaction(trans, root);
 912                 goto done;
 913         }
 914         WARN_ON(diff > old_total);
 915         btrfs_set_super_total_bytes(super_copy, old_total - diff);
 916         btrfs_end_transaction(trans, root);
 917
 918         key.objectid = device->devid;
 919         key.offset = (u64)-1;
 920         key.type = BTRFS_DEV_EXTENT_KEY;
 921
 922         while (1) {
 923                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 924                 if (ret < 0)
 925                         goto done;
 926
 927                 ret = btrfs_previous_item(root, path, 0, key.type);
 928                 if (ret < 0)
 929                         goto done;
 930                 if (ret) {
 931                         ret = 0;
 932                         goto done;
 933                 }
 934
 935                 l = path->nodes[0];
 936                 slot = path->slots[0];
 937                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 938
 939                 if (key.objectid != device->devid)
 940                         goto done;
 941
 942                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 943                 length = btrfs_dev_extent_length(l, dev_extent);
 944
 945                 if (key.offset + length <= new_size)
 946                         goto done;
 947
 948                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
 949                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
 950                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
 951                 btrfs_release_path(root, path);
 952
 953                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
 954                                            chunk_offset);
 955                 if (ret)
 956                         goto done;
 957         }
 958
 959 done:
 960         btrfs_free_path(path);
 961         return ret;
 962 }
 963
 964 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 965                            struct btrfs_root *root,
 966                            struct btrfs_key *key,
 967                            struct btrfs_chunk *chunk, int item_size)
 968 {
 969         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
 970         struct btrfs_disk_key disk_key;
 971         u32 array_size;
 972         u8 *ptr;
 973
 974         array_size = btrfs_super_sys_array_size(super_copy);
 975         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
 976                 return -EFBIG;
 977
 978         ptr = super_copy->sys_chunk_array + array_size;
 979         btrfs_cpu_key_to_disk(&disk_key, key);
 980         memcpy(ptr, &disk_key, sizeof(disk_key));
 981         ptr += sizeof(disk_key);
 982         memcpy(ptr, chunk, item_size);
 983         item_size += sizeof(disk_key);
 984         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
 985         return 0;
 986 }
 987
 988 static u64 div_factor(u64 num, int factor)
 989 {
 990         if (factor == 10)
 991                 return num;
 992         num *= factor;
 993         do_div(num, 10);
 994         return num;
 995 }
 996
 997 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
 998                                int sub_stripes)
 999 {
1000         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1001                 return calc_size;
1002         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1003                 return calc_size * (num_stripes / sub_stripes);
1004         else
1005                 return calc_size * num_stripes;
1006 }
1007
1008
1009 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1010                       struct btrfs_root *extent_root, u64 *start,
1011                       u64 *num_bytes, u64 type)
1012 {
1013         u64 dev_offset;
1014         struct btrfs_fs_info *info = extent_root->fs_info;
1015         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1016         struct btrfs_path *path;
1017         struct btrfs_stripe *stripes;
1018         struct btrfs_device *device = NULL;
1019         struct btrfs_chunk *chunk;
1020         struct list_head private_devs;
1021         struct list_head *dev_list;
1022         struct list_head *cur;
1023         struct extent_map_tree *em_tree;
1024         struct map_lookup *map;
1025         struct extent_map *em;
1026         int min_stripe_size = 1 * 1024 * 1024;
1027         u64 physical;
1028         u64 calc_size = 1024 * 1024 * 1024;
1029         u64 max_chunk_size = calc_size;
1030         u64 min_free;
1031         u64 avail;
1032         u64 max_avail = 0;
1033         u64 percent_max;
1034         int num_stripes = 1;
1035         int min_stripes = 1;
1036         int sub_stripes = 0;
1037         int looped = 0;
1038         int ret;
1039         int index;
1040         int stripe_len = 64 * 1024;
1041         struct btrfs_key key;
1042
1043         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1044         if (list_empty(dev_list))
1045                 return -ENOSPC;
1046
1047         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1048                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1049                 min_stripes = 2;
1050         }
1051         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1052                 num_stripes = 2;
1053                 min_stripes = 2;
1054         }
1055         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1056                 num_stripes = min_t(u64, 2,
1057                                   btrfs_super_num_devices(&info->super_copy));
1058                 if (num_stripes < 2)
1059                         return -ENOSPC;
1060                 min_stripes = 2;
1061         }
1062         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1063                 num_stripes = btrfs_super_num_devices(&info->super_copy);
1064                 if (num_stripes < 4)
1065                         return -ENOSPC;
1066                 num_stripes &= ~(u32)1;
1067                 sub_stripes = 2;
1068                 min_stripes = 4;
1069         }
1070
1071         if (type & BTRFS_BLOCK_GROUP_DATA) {
1072                 max_chunk_size = 10 * calc_size;
1073                 min_stripe_size = 64 * 1024 * 1024;
1074         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1075                 max_chunk_size = 4 * calc_size;
1076                 min_stripe_size = 32 * 1024 * 1024;
1077         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1078                 calc_size = 8 * 1024 * 1024;
1079                 max_chunk_size = calc_size * 2;
1080                 min_stripe_size = 1 * 1024 * 1024;
1081         }
1082
1083         path = btrfs_alloc_path();
1084         if (!path)
1085                 return -ENOMEM;
1086
1087         /* we don't want a chunk larger than 10% of the FS */
1088         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1089         max_chunk_size = min(percent_max, max_chunk_size);
1090
1091 again:
1092         if (calc_size * num_stripes > max_chunk_size) {
1093                 calc_size = max_chunk_size;
1094                 do_div(calc_size, num_stripes);
1095                 do_div(calc_size, stripe_len);
1096                 calc_size *= stripe_len;
1097         }
1098         /* we don't want tiny stripes */
1099         calc_size = max_t(u64, min_stripe_size, calc_size);
1100
1101         do_div(calc_size, stripe_len);
1102         calc_size *= stripe_len;
1103
1104         INIT_LIST_HEAD(&private_devs);
1105         cur = dev_list->next;
1106         index = 0;
1107
1108         if (type & BTRFS_BLOCK_GROUP_DUP)
1109                 min_free = calc_size * 2;
1110         else
1111                 min_free = calc_size;
1112
1113         /* we add 1MB because we never use the first 1MB of the device */
1114         min_free += 1024 * 1024;
1115
1116         /* build a private list of devices we will allocate from */
1117         while(index < num_stripes) {
1118                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1119
1120                 avail = device->total_bytes - device->bytes_used;
1121                 cur = cur->next;
1122
1123                 if (avail >= min_free) {
1124                         u64 ignored_start = 0;
1125                         ret = find_free_dev_extent(trans, device, path,
1126                                                    min_free,
1127                                                    &ignored_start);
1128                         if (ret == 0) {
1129                                 list_move_tail(&device->dev_alloc_list,
1130                                                &private_devs);
1131                                 index++;
1132                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1133                                         index++;
1134                         }
1135                 } else if (avail > max_avail)
1136                         max_avail = avail;
1137                 if (cur == dev_list)
1138                         break;
1139         }
1140         if (index < num_stripes) {
1141                 list_splice(&private_devs, dev_list);
1142                 if (index >= min_stripes) {
1143                         num_stripes = index;
1144                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1145                                 num_stripes /= sub_stripes;
1146                                 num_stripes *= sub_stripes;
1147                         }
1148                         looped = 1;
1149                         goto again;
1150                 }
1151                 if (!looped && max_avail > 0) {
1152                         looped = 1;
1153                         calc_size = max_avail;
1154                         goto again;
1155                 }
1156                 btrfs_free_path(path);
1157                 return -ENOSPC;
1158         }
1159         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1160         key.type = BTRFS_CHUNK_ITEM_KEY;
1161         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1162                               &key.offset);
1163         if (ret) {
1164                 btrfs_free_path(path);
1165                 return ret;
1166         }
1167
1168         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1169         if (!chunk) {
1170                 btrfs_free_path(path);
1171                 return -ENOMEM;
1172         }
1173
1174         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1175         if (!map) {
1176                 kfree(chunk);
1177                 btrfs_free_path(path);
1178                 return -ENOMEM;
1179         }
1180         btrfs_free_path(path);
1181         path = NULL;
1182
1183         stripes = &chunk->stripe;
1184         *num_bytes = chunk_bytes_by_type(type, calc_size,
1185                                          num_stripes, sub_stripes);
1186
1187         index = 0;
1188 printk("new chunk type %Lu start %Lu size %Lu\n", type, key.offset, *num_bytes);
1189         while(index < num_stripes) {
1190                 struct btrfs_stripe *stripe;
1191                 BUG_ON(list_empty(&private_devs));
1192                 cur = private_devs.next;
1193                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1194
1195                 /* loop over this device again if we're doing a dup group */
1196                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1197                     (index == num_stripes - 1))
1198                         list_move_tail(&device->dev_alloc_list, dev_list);
1199
1200                 ret = btrfs_alloc_dev_extent(trans, device,
1201                              info->chunk_root->root_key.objectid,
1202                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1203                              calc_size, &dev_offset);
1204                 BUG_ON(ret);
1205 printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.offset, calc_size, device->devid, type);
1206                 device->bytes_used += calc_size;
1207                 ret = btrfs_update_device(trans, device);
1208                 BUG_ON(ret);
1209
1210                 map->stripes[index].dev = device;
1211                 map->stripes[index].physical = dev_offset;
1212                 stripe = stripes + index;
1213                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1214                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1215                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1216                 physical = dev_offset;
1217                 index++;
1218         }
1219         BUG_ON(!list_empty(&private_devs));
1220
1221         /* key was set above */
1222         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1223         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1224         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1225         btrfs_set_stack_chunk_type(chunk, type);
1226         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1227         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1228         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1229         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1230         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1231         map->sector_size = extent_root->sectorsize;
1232         map->stripe_len = stripe_len;
1233         map->io_align = stripe_len;
1234         map->io_width = stripe_len;
1235         map->type = type;
1236         map->num_stripes = num_stripes;
1237         map->sub_stripes = sub_stripes;
1238
1239         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1240                                 btrfs_chunk_item_size(num_stripes));
1241         BUG_ON(ret);
1242         *start = key.offset;;
1243
1244         em = alloc_extent_map(GFP_NOFS);
1245         if (!em)
1246                 return -ENOMEM;
1247         em->bdev = (struct block_device *)map;
1248         em->start = key.offset;
1249         em->len = *num_bytes;
1250         em->block_start = 0;
1251
1252         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1253                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1254                                     chunk, btrfs_chunk_item_size(num_stripes));
1255                 BUG_ON(ret);
1256         }
1257         kfree(chunk);
1258
1259         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1260         spin_lock(&em_tree->lock);
1261         ret = add_extent_mapping(em_tree, em);
1262         spin_unlock(&em_tree->lock);
1263         BUG_ON(ret);
1264         free_extent_map(em);
1265         return ret;
1266 }
1267
1268 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1269 {
1270         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1271 }
1272
1273 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1274 {
1275         struct extent_map *em;
1276
1277         while(1) {
1278                 spin_lock(&tree->map_tree.lock);
1279                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1280                 if (em)
1281                         remove_extent_mapping(&tree->map_tree, em);
1282                 spin_unlock(&tree->map_tree.lock);
1283                 if (!em)
1284                         break;
1285                 kfree(em->bdev);
1286                 /* once for us */
1287                 free_extent_map(em);
1288                 /* once for the tree */
1289                 free_extent_map(em);
1290         }
1291 }
1292
1293 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1294 {
1295         struct extent_map *em;
1296         struct map_lookup *map;
1297         struct extent_map_tree *em_tree = &map_tree->map_tree;
1298         int ret;
1299
1300         spin_lock(&em_tree->lock);
1301         em = lookup_extent_mapping(em_tree, logical, len);
1302         spin_unlock(&em_tree->lock);
1303         BUG_ON(!em);
1304
1305         BUG_ON(em->start > logical || em->start + em->len < logical);
1306         map = (struct map_lookup *)em->bdev;
1307         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1308                 ret = map->num_stripes;
1309         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1310                 ret = map->sub_stripes;
1311         else
1312                 ret = 1;
1313         free_extent_map(em);
1314         return ret;
1315 }
1316
1317 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1318                              u64 logical, u64 *length,
1319                              struct btrfs_multi_bio **multi_ret,
1320                              int mirror_num, struct page *unplug_page)
1321 {
1322         struct extent_map *em;
1323         struct map_lookup *map;
1324         struct extent_map_tree *em_tree = &map_tree->map_tree;
1325         u64 offset;
1326         u64 stripe_offset;
1327         u64 stripe_nr;
1328         int stripes_allocated = 8;
1329         int stripes_required = 1;
1330         int stripe_index;
1331         int i;
1332         int num_stripes;
1333         struct btrfs_multi_bio *multi = NULL;
1334
1335         if (multi_ret && !(rw & (1 << BIO_RW))) {
1336                 stripes_allocated = 1;
1337         }
1338 again:
1339         if (multi_ret) {
1340                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1341                                 GFP_NOFS);
1342                 if (!multi)
1343                         return -ENOMEM;
1344         }
1345
1346         spin_lock(&em_tree->lock);
1347         em = lookup_extent_mapping(em_tree, logical, *length);
1348         spin_unlock(&em_tree->lock);
1349
1350         if (!em && unplug_page)
1351                 return 0;
1352
1353         if (!em) {
1354                 printk("unable to find logical %Lu\n", logical);
1355                 BUG();
1356         }
1357
1358         BUG_ON(em->start > logical || em->start + em->len < logical);
1359         map = (struct map_lookup *)em->bdev;
1360         offset = logical - em->start;
1361
1362         if (mirror_num > map->num_stripes)
1363                 mirror_num = 0;
1364
1365         /* if our multi bio struct is too small, back off and try again */
1366         if (rw & (1 << BIO_RW)) {
1367                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1368                                  BTRFS_BLOCK_GROUP_DUP)) {
1369                         stripes_required = map->num_stripes;
1370                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1371                         stripes_required = map->sub_stripes;
1372                 }
1373         }
1374         if (multi_ret && rw == WRITE &&
1375             stripes_allocated < stripes_required) {
1376                 stripes_allocated = map->num_stripes;
1377                 free_extent_map(em);
1378                 kfree(multi);
1379                 goto again;
1380         }
1381         stripe_nr = offset;
1382         /*
1383          * stripe_nr counts the total number of stripes we have to stride
1384          * to get to this block
1385          */
1386         do_div(stripe_nr, map->stripe_len);
1387
1388         stripe_offset = stripe_nr * map->stripe_len;
1389         BUG_ON(offset < stripe_offset);
1390
1391         /* stripe_offset is the offset of this block in its stripe*/
1392         stripe_offset = offset - stripe_offset;
1393
1394         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1395                          BTRFS_BLOCK_GROUP_RAID10 |
1396                          BTRFS_BLOCK_GROUP_DUP)) {
1397                 /* we limit the length of each bio to what fits in a stripe */
1398                 *length = min_t(u64, em->len - offset,
1399                               map->stripe_len - stripe_offset);
1400         } else {
1401                 *length = em->len - offset;
1402         }
1403
1404         if (!multi_ret && !unplug_page)
1405                 goto out;
1406
1407         num_stripes = 1;
1408         stripe_index = 0;
1409         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1410                 if (unplug_page || (rw & (1 << BIO_RW)))
1411                         num_stripes = map->num_stripes;
1412                 else if (mirror_num) {
1413                         stripe_index = mirror_num - 1;
1414                 } else {
1415                         u64 orig_stripe_nr = stripe_nr;
1416                         stripe_index = do_div(orig_stripe_nr, num_stripes);
1417                 }
1418         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1419                 if (rw & (1 << BIO_RW))
1420                         num_stripes = map->num_stripes;
1421                 else if (mirror_num)
1422                         stripe_index = mirror_num - 1;
1423         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1424                 int factor = map->num_stripes / map->sub_stripes;
1425
1426                 stripe_index = do_div(stripe_nr, factor);
1427                 stripe_index *= map->sub_stripes;
1428
1429                 if (unplug_page || (rw & (1 << BIO_RW)))
1430                         num_stripes = map->sub_stripes;
1431                 else if (mirror_num)
1432                         stripe_index += mirror_num - 1;
1433                 else {
1434                         u64 orig_stripe_nr = stripe_nr;
1435                         stripe_index += do_div(orig_stripe_nr,
1436                                                map->sub_stripes);
1437                 }
1438         } else {
1439                 /*
1440                  * after this do_div call, stripe_nr is the number of stripes
1441                  * on this device we have to walk to find the data, and
1442                  * stripe_index is the number of our device in the stripe array
1443                  */
1444                 stripe_index = do_div(stripe_nr, map->num_stripes);
1445         }
1446         BUG_ON(stripe_index >= map->num_stripes);
1447
1448         for (i = 0; i < num_stripes; i++) {
1449                 if (unplug_page) {
1450                         struct btrfs_device *device;
1451                         struct backing_dev_info *bdi;
1452
1453                         device = map->stripes[stripe_index].dev;
1454                         bdi = blk_get_backing_dev_info(device->bdev);
1455                         if (bdi->unplug_io_fn) {
1456                                 bdi->unplug_io_fn(bdi, unplug_page);
1457                         }
1458                 } else {
1459                         multi->stripes[i].physical =
1460                                 map->stripes[stripe_index].physical +
1461                                 stripe_offset + stripe_nr * map->stripe_len;
1462                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
1463                 }
1464                 stripe_index++;
1465         }
1466         if (multi_ret) {
1467                 *multi_ret = multi;
1468                 multi->num_stripes = num_stripes;
1469         }
1470 out:
1471         free_extent_map(em);
1472         return 0;
1473 }
1474
1475 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1476                       u64 logical, u64 *length,
1477                       struct btrfs_multi_bio **multi_ret, int mirror_num)
1478 {
1479         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1480                                  mirror_num, NULL);
1481 }
1482
1483 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1484                       u64 logical, struct page *page)
1485 {
1486         u64 length = PAGE_CACHE_SIZE;
1487         return __btrfs_map_block(map_tree, READ, logical, &length,
1488                                  NULL, 0, page);
1489 }
1490
1491
1492 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1493 static void end_bio_multi_stripe(struct bio *bio, int err)
1494 #else
1495 static int end_bio_multi_stripe(struct bio *bio,
1496                                    unsigned int bytes_done, int err)
1497 #endif
1498 {
1499         struct btrfs_multi_bio *multi = bio->bi_private;
1500
1501 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1502         if (bio->bi_size)
1503                 return 1;
1504 #endif
1505         if (err)
1506                 multi->error = err;
1507
1508         if (atomic_dec_and_test(&multi->stripes_pending)) {
1509                 bio->bi_private = multi->private;
1510                 bio->bi_end_io = multi->end_io;
1511
1512                 if (!err && multi->error)
1513                         err = multi->error;
1514                 kfree(multi);
1515
1516 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1517                 bio_endio(bio, bio->bi_size, err);
1518 #else
1519                 bio_endio(bio, err);
1520 #endif
1521         } else {
1522                 bio_put(bio);
1523         }
1524 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1525         return 0;
1526 #endif
1527 }
1528
1529 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1530                   int mirror_num)
1531 {
1532         struct btrfs_mapping_tree *map_tree;
1533         struct btrfs_device *dev;
1534         struct bio *first_bio = bio;
1535         u64 logical = bio->bi_sector << 9;
1536         u64 length = 0;
1537         u64 map_length;
1538         struct btrfs_multi_bio *multi = NULL;
1539         int ret;
1540         int dev_nr = 0;
1541         int total_devs = 1;
1542
1543         length = bio->bi_size;
1544         map_tree = &root->fs_info->mapping_tree;
1545         map_length = length;
1546
1547         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1548                               mirror_num);
1549         BUG_ON(ret);
1550
1551         total_devs = multi->num_stripes;
1552         if (map_length < length) {
1553                 printk("mapping failed logical %Lu bio len %Lu "
1554                        "len %Lu\n", logical, length, map_length);
1555                 BUG();
1556         }
1557         multi->end_io = first_bio->bi_end_io;
1558         multi->private = first_bio->bi_private;
1559         atomic_set(&multi->stripes_pending, multi->num_stripes);
1560
1561         while(dev_nr < total_devs) {
1562                 if (total_devs > 1) {
1563                         if (dev_nr < total_devs - 1) {
1564                                 bio = bio_clone(first_bio, GFP_NOFS);
1565                                 BUG_ON(!bio);
1566                         } else {
1567                                 bio = first_bio;
1568                         }
1569                         bio->bi_private = multi;
1570                         bio->bi_end_io = end_bio_multi_stripe;
1571                 }
1572                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1573                 dev = multi->stripes[dev_nr].dev;
1574
1575                 bio->bi_bdev = dev->bdev;
1576                 spin_lock(&dev->io_lock);
1577                 dev->total_ios++;
1578                 spin_unlock(&dev->io_lock);
1579                 submit_bio(rw, bio);
1580                 dev_nr++;
1581         }
1582         if (total_devs == 1)
1583                 kfree(multi);
1584         return 0;
1585 }
1586
1587 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1588                                        u8 *uuid)
1589 {
1590         struct list_head *head = &root->fs_info->fs_devices->devices;
1591
1592         return __find_device(head, devid, uuid);
1593 }
1594
1595 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1596                           struct extent_buffer *leaf,
1597                           struct btrfs_chunk *chunk)
1598 {
1599         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1600         struct map_lookup *map;
1601         struct extent_map *em;
1602         u64 logical;
1603         u64 length;
1604         u64 devid;
1605         u8 uuid[BTRFS_UUID_SIZE];
1606         int num_stripes;
1607         int ret;
1608         int i;
1609
1610         logical = key->offset;
1611         length = btrfs_chunk_length(leaf, chunk);
1612         spin_lock(&map_tree->map_tree.lock);
1613         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1614         spin_unlock(&map_tree->map_tree.lock);
1615
1616         /* already mapped? */
1617         if (em && em->start <= logical && em->start + em->len > logical) {
1618                 free_extent_map(em);
1619                 return 0;
1620         } else if (em) {
1621                 free_extent_map(em);
1622         }
1623
1624         map = kzalloc(sizeof(*map), GFP_NOFS);
1625         if (!map)
1626                 return -ENOMEM;
1627
1628         em = alloc_extent_map(GFP_NOFS);
1629         if (!em)
1630                 return -ENOMEM;
1631         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1632         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1633         if (!map) {
1634                 free_extent_map(em);
1635                 return -ENOMEM;
1636         }
1637
1638         em->bdev = (struct block_device *)map;
1639         em->start = logical;
1640         em->len = length;
1641         em->block_start = 0;
1642
1643         map->num_stripes = num_stripes;
1644         map->io_width = btrfs_chunk_io_width(leaf, chunk);
1645         map->io_align = btrfs_chunk_io_align(leaf, chunk);
1646         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1647         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1648         map->type = btrfs_chunk_type(leaf, chunk);
1649         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1650         for (i = 0; i < num_stripes; i++) {
1651                 map->stripes[i].physical =
1652                         btrfs_stripe_offset_nr(leaf, chunk, i);
1653                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1654                 read_extent_buffer(leaf, uuid, (unsigned long)
1655                                    btrfs_stripe_dev_uuid_nr(chunk, i),
1656                                    BTRFS_UUID_SIZE);
1657                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1658                 if (!map->stripes[i].dev) {
1659                         kfree(map);
1660                         free_extent_map(em);
1661                         return -EIO;
1662                 }
1663         }
1664
1665         spin_lock(&map_tree->map_tree.lock);
1666         ret = add_extent_mapping(&map_tree->map_tree, em);
1667         spin_unlock(&map_tree->map_tree.lock);
1668         BUG_ON(ret);
1669         free_extent_map(em);
1670
1671         return 0;
1672 }
1673
1674 static int fill_device_from_item(struct extent_buffer *leaf,
1675                                  struct btrfs_dev_item *dev_item,
1676                                  struct btrfs_device *device)
1677 {
1678         unsigned long ptr;
1679
1680         device->devid = btrfs_device_id(leaf, dev_item);
1681         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1682         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1683         device->type = btrfs_device_type(leaf, dev_item);
1684         device->io_align = btrfs_device_io_align(leaf, dev_item);
1685         device->io_width = btrfs_device_io_width(leaf, dev_item);
1686         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1687
1688         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1689         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1690
1691         return 0;
1692 }
1693
1694 static int read_one_dev(struct btrfs_root *root,
1695                         struct extent_buffer *leaf,
1696                         struct btrfs_dev_item *dev_item)
1697 {
1698         struct btrfs_device *device;
1699         u64 devid;
1700         int ret;
1701         u8 dev_uuid[BTRFS_UUID_SIZE];
1702
1703         devid = btrfs_device_id(leaf, dev_item);
1704         read_extent_buffer(leaf, dev_uuid,
1705                            (unsigned long)btrfs_device_uuid(dev_item),
1706                            BTRFS_UUID_SIZE);
1707         device = btrfs_find_device(root, devid, dev_uuid);
1708         if (!device) {
1709                 printk("warning devid %Lu not found already\n", devid);
1710                 device = kzalloc(sizeof(*device), GFP_NOFS);
1711                 if (!device)
1712                         return -ENOMEM;
1713                 list_add(&device->dev_list,
1714                          &root->fs_info->fs_devices->devices);
1715                 list_add(&device->dev_alloc_list,
1716                          &root->fs_info->fs_devices->alloc_list);
1717                 device->barriers = 1;
1718                 spin_lock_init(&device->io_lock);
1719         }
1720
1721         fill_device_from_item(leaf, dev_item, device);
1722         device->dev_root = root->fs_info->dev_root;
1723         ret = 0;
1724 #if 0
1725         ret = btrfs_open_device(device);
1726         if (ret) {
1727                 kfree(device);
1728         }
1729 #endif
1730         return ret;
1731 }
1732
1733 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1734 {
1735         struct btrfs_dev_item *dev_item;
1736
1737         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1738                                                      dev_item);
1739         return read_one_dev(root, buf, dev_item);
1740 }
1741
1742 int btrfs_read_sys_array(struct btrfs_root *root)
1743 {
1744         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1745         struct extent_buffer *sb = root->fs_info->sb_buffer;
1746         struct btrfs_disk_key *disk_key;
1747         struct btrfs_chunk *chunk;
1748         u8 *ptr;
1749         unsigned long sb_ptr;
1750         int ret = 0;
1751         u32 num_stripes;
1752         u32 array_size;
1753         u32 len = 0;
1754         u32 cur;
1755         struct btrfs_key key;
1756
1757         array_size = btrfs_super_sys_array_size(super_copy);
1758
1759         ptr = super_copy->sys_chunk_array;
1760         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1761         cur = 0;
1762
1763         while (cur < array_size) {
1764                 disk_key = (struct btrfs_disk_key *)ptr;
1765                 btrfs_disk_key_to_cpu(&key, disk_key);
1766
1767                 len = sizeof(*disk_key);
1768                 ptr += len;
1769                 sb_ptr += len;
1770                 cur += len;
1771
1772                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1773                         chunk = (struct btrfs_chunk *)sb_ptr;
1774                         ret = read_one_chunk(root, &key, sb, chunk);
1775                         if (ret)
1776                                 break;
1777                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1778                         len = btrfs_chunk_item_size(num_stripes);
1779                 } else {
1780                         ret = -EIO;
1781                         break;
1782                 }
1783                 ptr += len;
1784                 sb_ptr += len;
1785                 cur += len;
1786         }
1787         return ret;
1788 }
1789
1790 int btrfs_read_chunk_tree(struct btrfs_root *root)
1791 {
1792         struct btrfs_path *path;
1793         struct extent_buffer *leaf;
1794         struct btrfs_key key;
1795         struct btrfs_key found_key;
1796         int ret;
1797         int slot;
1798
1799         root = root->fs_info->chunk_root;
1800
1801         path = btrfs_alloc_path();
1802         if (!path)
1803                 return -ENOMEM;
1804
1805         /* first we search for all of the device items, and then we
1806          * read in all of the chunk items.  This way we can create chunk
1807          * mappings that reference all of the devices that are afound
1808          */
1809         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1810         key.offset = 0;
1811         key.type = 0;
1812 again:
1813         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1814         while(1) {
1815                 leaf = path->nodes[0];
1816                 slot = path->slots[0];
1817                 if (slot >= btrfs_header_nritems(leaf)) {
1818                         ret = btrfs_next_leaf(root, path);
1819                         if (ret == 0)
1820                                 continue;
1821                         if (ret < 0)
1822                                 goto error;
1823                         break;
1824                 }
1825                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1826                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1827                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1828                                 break;
1829                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1830                                 struct btrfs_dev_item *dev_item;
1831                                 dev_item = btrfs_item_ptr(leaf, slot,
1832                                                   struct btrfs_dev_item);
1833                                 ret = read_one_dev(root, leaf, dev_item);
1834                                 BUG_ON(ret);
1835                         }
1836                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1837                         struct btrfs_chunk *chunk;
1838                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1839                         ret = read_one_chunk(root, &found_key, leaf, chunk);
1840                 }
1841                 path->slots[0]++;
1842         }
1843         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1844                 key.objectid = 0;
1845                 btrfs_release_path(root, path);
1846                 goto again;
1847         }
1848
1849         btrfs_free_path(path);
1850         ret = 0;
1851 error:
1852         return ret;
1853 }
1854