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 #include "async-thread.h"
  31
  32 struct map_lookup {
  33         u64 type;
  34         int io_align;
  35         int io_width;
  36         int stripe_len;
  37         int sector_size;
  38         int num_stripes;
  39         int sub_stripes;
  40         struct btrfs_bio_stripe stripes[];
  41 };
  42
  43 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  44                             (sizeof(struct btrfs_bio_stripe) * (n)))
  45
  46 static DEFINE_MUTEX(uuid_mutex);
  47 static LIST_HEAD(fs_uuids);
  48
  49 void btrfs_lock_volumes(void)
  50 {
  51         mutex_lock(&uuid_mutex);
  52 }
  53
  54 void btrfs_unlock_volumes(void)
  55 {
  56         mutex_unlock(&uuid_mutex);
  57 }
  58
  59 static void lock_chunks(struct btrfs_root *root)
  60 {
  61         mutex_lock(&root->fs_info->alloc_mutex);
  62         mutex_lock(&root->fs_info->chunk_mutex);
  63 }
  64
  65 static void unlock_chunks(struct btrfs_root *root)
  66 {
  67         mutex_unlock(&root->fs_info->chunk_mutex);
  68         mutex_unlock(&root->fs_info->alloc_mutex);
  69 }
  70
  71 int btrfs_cleanup_fs_uuids(void)
  72 {
  73         struct btrfs_fs_devices *fs_devices;
  74         struct list_head *uuid_cur;
  75         struct list_head *devices_cur;
  76         struct btrfs_device *dev;
  77
  78         list_for_each(uuid_cur, &fs_uuids) {
  79                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  80                                         list);
  81                 while(!list_empty(&fs_devices->devices)) {
  82                         devices_cur = fs_devices->devices.next;
  83                         dev = list_entry(devices_cur, struct btrfs_device,
  84                                          dev_list);
  85                         if (dev->bdev) {
  86                                 close_bdev_excl(dev->bdev);
  87                                 fs_devices->open_devices--;
  88                         }
  89                         list_del(&dev->dev_list);
  90                         kfree(dev->name);
  91                         kfree(dev);
  92                 }
  93         }
  94         return 0;
  95 }
  96
  97 static noinline struct btrfs_device *__find_device(struct list_head *head,
  98                                                    u64 devid, u8 *uuid)
  99 {
 100         struct btrfs_device *dev;
 101         struct list_head *cur;
 102
 103         list_for_each(cur, head) {
 104                 dev = list_entry(cur, struct btrfs_device, dev_list);
 105                 if (dev->devid == devid &&
 106                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 107                         return dev;
 108                 }
 109         }
 110         return NULL;
 111 }
 112
 113 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 114 {
 115         struct list_head *cur;
 116         struct btrfs_fs_devices *fs_devices;
 117
 118         list_for_each(cur, &fs_uuids) {
 119                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
 120                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 121                         return fs_devices;
 122         }
 123         return NULL;
 124 }
 125
 126 /*
 127  * we try to collect pending bios for a device so we don't get a large
 128  * number of procs sending bios down to the same device.  This greatly
 129  * improves the schedulers ability to collect and merge the bios.
 130  *
 131  * But, it also turns into a long list of bios to process and that is sure
 132  * to eventually make the worker thread block.  The solution here is to
 133  * make some progress and then put this work struct back at the end of
 134  * the list if the block device is congested.  This way, multiple devices
 135  * can make progress from a single worker thread.
 136  */
 137 static int noinline run_scheduled_bios(struct btrfs_device *device)
 138 {
 139         struct bio *pending;
 140         struct backing_dev_info *bdi;
 141         struct btrfs_fs_info *fs_info;
 142         struct bio *tail;
 143         struct bio *cur;
 144         int again = 0;
 145         unsigned long num_run = 0;
 146         unsigned long limit;
 147
 148         bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
 149         fs_info = device->dev_root->fs_info;
 150         limit = btrfs_async_submit_limit(fs_info);
 151         limit = limit * 2 / 3;
 152
 153 loop:
 154         spin_lock(&device->io_lock);
 155
 156         /* take all the bios off the list at once and process them
 157          * later on (without the lock held).  But, remember the
 158          * tail and other pointers so the bios can be properly reinserted
 159          * into the list if we hit congestion
 160          */
 161         pending = device->pending_bios;
 162         tail = device->pending_bio_tail;
 163         WARN_ON(pending && !tail);
 164         device->pending_bios = NULL;
 165         device->pending_bio_tail = NULL;
 166
 167         /*
 168          * if pending was null this time around, no bios need processing
 169          * at all and we can stop.  Otherwise it'll loop back up again
 170          * and do an additional check so no bios are missed.
 171          *
 172          * device->running_pending is used to synchronize with the
 173          * schedule_bio code.
 174          */
 175         if (pending) {
 176                 again = 1;
 177                 device->running_pending = 1;
 178         } else {
 179                 again = 0;
 180                 device->running_pending = 0;
 181         }
 182         spin_unlock(&device->io_lock);
 183
 184         while(pending) {
 185                 cur = pending;
 186                 pending = pending->bi_next;
 187                 cur->bi_next = NULL;
 188                 atomic_dec(&fs_info->nr_async_bios);
 189
 190                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 191                     waitqueue_active(&fs_info->async_submit_wait))
 192                         wake_up(&fs_info->async_submit_wait);
 193
 194                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 195                 bio_get(cur);
 196                 submit_bio(cur->bi_rw, cur);
 197                 bio_put(cur);
 198                 num_run++;
 199
 200                 /*
 201                  * we made progress, there is more work to do and the bdi
 202                  * is now congested.  Back off and let other work structs
 203                  * run instead
 204                  */
 205                 if (pending && bdi_write_congested(bdi)) {
 206                         struct bio *old_head;
 207
 208                         spin_lock(&device->io_lock);
 209
 210                         old_head = device->pending_bios;
 211                         device->pending_bios = pending;
 212                         if (device->pending_bio_tail)
 213                                 tail->bi_next = old_head;
 214                         else
 215                                 device->pending_bio_tail = tail;
 216
 217                         spin_unlock(&device->io_lock);
 218                         btrfs_requeue_work(&device->work);
 219                         goto done;
 220                 }
 221         }
 222         if (again)
 223                 goto loop;
 224 done:
 225         return 0;
 226 }
 227
 228 void pending_bios_fn(struct btrfs_work *work)
 229 {
 230         struct btrfs_device *device;
 231
 232         device = container_of(work, struct btrfs_device, work);
 233         run_scheduled_bios(device);
 234 }
 235
 236 static noinline int device_list_add(const char *path,
 237                            struct btrfs_super_block *disk_super,
 238                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 239 {
 240         struct btrfs_device *device;
 241         struct btrfs_fs_devices *fs_devices;
 242         u64 found_transid = btrfs_super_generation(disk_super);
 243
 244         fs_devices = find_fsid(disk_super->fsid);
 245         if (!fs_devices) {
 246                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 247                 if (!fs_devices)
 248                         return -ENOMEM;
 249                 INIT_LIST_HEAD(&fs_devices->devices);
 250                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 251                 list_add(&fs_devices->list, &fs_uuids);
 252                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 253                 fs_devices->latest_devid = devid;
 254                 fs_devices->latest_trans = found_transid;
 255                 device = NULL;
 256         } else {
 257                 device = __find_device(&fs_devices->devices, devid,
 258                                        disk_super->dev_item.uuid);
 259         }
 260         if (!device) {
 261                 device = kzalloc(sizeof(*device), GFP_NOFS);
 262                 if (!device) {
 263                         /* we can safely leave the fs_devices entry around */
 264                         return -ENOMEM;
 265                 }
 266                 device->devid = devid;
 267                 device->work.func = pending_bios_fn;
 268                 memcpy(device->uuid, disk_super->dev_item.uuid,
 269                        BTRFS_UUID_SIZE);
 270                 device->barriers = 1;
 271                 spin_lock_init(&device->io_lock);
 272                 device->name = kstrdup(path, GFP_NOFS);
 273                 if (!device->name) {
 274                         kfree(device);
 275                         return -ENOMEM;
 276                 }
 277                 list_add(&device->dev_list, &fs_devices->devices);
 278                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 279                 fs_devices->num_devices++;
 280         }
 281
 282         if (found_transid > fs_devices->latest_trans) {
 283                 fs_devices->latest_devid = devid;
 284                 fs_devices->latest_trans = found_transid;
 285         }
 286         *fs_devices_ret = fs_devices;
 287         return 0;
 288 }
 289
 290 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 291 {
 292         struct list_head *head = &fs_devices->devices;
 293         struct list_head *cur;
 294         struct btrfs_device *device;
 295
 296         mutex_lock(&uuid_mutex);
 297 again:
 298         list_for_each(cur, head) {
 299                 device = list_entry(cur, struct btrfs_device, dev_list);
 300                 if (!device->in_fs_metadata) {
 301                         struct block_device *bdev;
 302                         list_del(&device->dev_list);
 303                         list_del(&device->dev_alloc_list);
 304                         fs_devices->num_devices--;
 305                         if (device->bdev) {
 306                                 bdev = device->bdev;
 307                                 fs_devices->open_devices--;
 308                                 mutex_unlock(&uuid_mutex);
 309                                 close_bdev_excl(bdev);
 310                                 mutex_lock(&uuid_mutex);
 311                         }
 312                         kfree(device->name);
 313                         kfree(device);
 314                         goto again;
 315                 }
 316         }
 317         mutex_unlock(&uuid_mutex);
 318         return 0;
 319 }
 320
 321 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 322 {
 323         struct list_head *head = &fs_devices->devices;
 324         struct list_head *cur;
 325         struct btrfs_device *device;
 326
 327         mutex_lock(&uuid_mutex);
 328         list_for_each(cur, head) {
 329                 device = list_entry(cur, struct btrfs_device, dev_list);
 330                 if (device->bdev) {
 331                         close_bdev_excl(device->bdev);
 332                         fs_devices->open_devices--;
 333                 }
 334                 device->bdev = NULL;
 335                 device->in_fs_metadata = 0;
 336         }
 337         fs_devices->mounted = 0;
 338         mutex_unlock(&uuid_mutex);
 339         return 0;
 340 }
 341
 342 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 343                        int flags, void *holder)
 344 {
 345         struct block_device *bdev;
 346         struct list_head *head = &fs_devices->devices;
 347         struct list_head *cur;
 348         struct btrfs_device *device;
 349         struct block_device *latest_bdev = NULL;
 350         struct buffer_head *bh;
 351         struct btrfs_super_block *disk_super;
 352         u64 latest_devid = 0;
 353         u64 latest_transid = 0;
 354         u64 transid;
 355         u64 devid;
 356         int ret = 0;
 357
 358         mutex_lock(&uuid_mutex);
 359         if (fs_devices->mounted)
 360                 goto out;
 361
 362         list_for_each(cur, head) {
 363                 device = list_entry(cur, struct btrfs_device, dev_list);
 364                 if (device->bdev)
 365                         continue;
 366
 367                 if (!device->name)
 368                         continue;
 369
 370                 bdev = open_bdev_excl(device->name, flags, holder);
 371
 372                 if (IS_ERR(bdev)) {
 373                         printk("open %s failed\n", device->name);
 374                         goto error;
 375                 }
 376                 set_blocksize(bdev, 4096);
 377
 378                 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 379                 if (!bh)
 380                         goto error_close;
 381
 382                 disk_super = (struct btrfs_super_block *)bh->b_data;
 383                 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 384                     sizeof(disk_super->magic)))
 385                         goto error_brelse;
 386
 387                 devid = le64_to_cpu(disk_super->dev_item.devid);
 388                 if (devid != device->devid)
 389                         goto error_brelse;
 390
 391                 transid = btrfs_super_generation(disk_super);
 392                 if (!latest_transid || transid > latest_transid) {
 393                         latest_devid = devid;
 394                         latest_transid = transid;
 395                         latest_bdev = bdev;
 396                 }
 397
 398                 device->bdev = bdev;
 399                 device->in_fs_metadata = 0;
 400                 fs_devices->open_devices++;
 401                 continue;
 402
 403 error_brelse:
 404                 brelse(bh);
 405 error_close:
 406                 close_bdev_excl(bdev);
 407 error:
 408                 continue;
 409         }
 410         if (fs_devices->open_devices == 0) {
 411                 ret = -EIO;
 412                 goto out;
 413         }
 414         fs_devices->mounted = 1;
 415         fs_devices->latest_bdev = latest_bdev;
 416         fs_devices->latest_devid = latest_devid;
 417         fs_devices->latest_trans = latest_transid;
 418 out:
 419         mutex_unlock(&uuid_mutex);
 420         return ret;
 421 }
 422
 423 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 424                           struct btrfs_fs_devices **fs_devices_ret)
 425 {
 426         struct btrfs_super_block *disk_super;
 427         struct block_device *bdev;
 428         struct buffer_head *bh;
 429         int ret;
 430         u64 devid;
 431         u64 transid;
 432
 433         mutex_lock(&uuid_mutex);
 434
 435         bdev = open_bdev_excl(path, flags, holder);
 436
 437         if (IS_ERR(bdev)) {
 438                 ret = PTR_ERR(bdev);
 439                 goto error;
 440         }
 441
 442         ret = set_blocksize(bdev, 4096);
 443         if (ret)
 444                 goto error_close;
 445         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 446         if (!bh) {
 447                 ret = -EIO;
 448                 goto error_close;
 449         }
 450         disk_super = (struct btrfs_super_block *)bh->b_data;
 451         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 452             sizeof(disk_super->magic))) {
 453                 ret = -EINVAL;
 454                 goto error_brelse;
 455         }
 456         devid = le64_to_cpu(disk_super->dev_item.devid);
 457         transid = btrfs_super_generation(disk_super);
 458         if (disk_super->label[0])
 459                 printk("device label %s ", disk_super->label);
 460         else {
 461                 /* FIXME, make a readl uuid parser */
 462                 printk("device fsid %llx-%llx ",
 463                        *(unsigned long long *)disk_super->fsid,
 464                        *(unsigned long long *)(disk_super->fsid + 8));
 465         }
 466         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 467         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 468
 469 error_brelse:
 470         brelse(bh);
 471 error_close:
 472         close_bdev_excl(bdev);
 473 error:
 474         mutex_unlock(&uuid_mutex);
 475         return ret;
 476 }
 477
 478 /*
 479  * this uses a pretty simple search, the expectation is that it is
 480  * called very infrequently and that a given device has a small number
 481  * of extents
 482  */
 483 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
 484                                          struct btrfs_device *device,
 485                                          struct btrfs_path *path,
 486                                          u64 num_bytes, u64 *start)
 487 {
 488         struct btrfs_key key;
 489         struct btrfs_root *root = device->dev_root;
 490         struct btrfs_dev_extent *dev_extent = NULL;
 491         u64 hole_size = 0;
 492         u64 last_byte = 0;
 493         u64 search_start = 0;
 494         u64 search_end = device->total_bytes;
 495         int ret;
 496         int slot = 0;
 497         int start_found;
 498         struct extent_buffer *l;
 499
 500         start_found = 0;
 501         path->reada = 2;
 502
 503         /* FIXME use last free of some kind */
 504
 505         /* we don't want to overwrite the superblock on the drive,
 506          * so we make sure to start at an offset of at least 1MB
 507          */
 508         search_start = max((u64)1024 * 1024, search_start);
 509
 510         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 511                 search_start = max(root->fs_info->alloc_start, search_start);
 512
 513         key.objectid = device->devid;
 514         key.offset = search_start;
 515         key.type = BTRFS_DEV_EXTENT_KEY;
 516         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 517         if (ret < 0)
 518                 goto error;
 519         ret = btrfs_previous_item(root, path, 0, key.type);
 520         if (ret < 0)
 521                 goto error;
 522         l = path->nodes[0];
 523         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 524         while (1) {
 525                 l = path->nodes[0];
 526                 slot = path->slots[0];
 527                 if (slot >= btrfs_header_nritems(l)) {
 528                         ret = btrfs_next_leaf(root, path);
 529                         if (ret == 0)
 530                                 continue;
 531                         if (ret < 0)
 532                                 goto error;
 533 no_more_items:
 534                         if (!start_found) {
 535                                 if (search_start >= search_end) {
 536                                         ret = -ENOSPC;
 537                                         goto error;
 538                                 }
 539                                 *start = search_start;
 540                                 start_found = 1;
 541                                 goto check_pending;
 542                         }
 543                         *start = last_byte > search_start ?
 544                                 last_byte : search_start;
 545                         if (search_end <= *start) {
 546                                 ret = -ENOSPC;
 547                                 goto error;
 548                         }
 549                         goto check_pending;
 550                 }
 551                 btrfs_item_key_to_cpu(l, &key, slot);
 552
 553                 if (key.objectid < device->devid)
 554                         goto next;
 555
 556                 if (key.objectid > device->devid)
 557                         goto no_more_items;
 558
 559                 if (key.offset >= search_start && key.offset > last_byte &&
 560                     start_found) {
 561                         if (last_byte < search_start)
 562                                 last_byte = search_start;
 563                         hole_size = key.offset - last_byte;
 564                         if (key.offset > last_byte &&
 565                             hole_size >= num_bytes) {
 566                                 *start = last_byte;
 567                                 goto check_pending;
 568                         }
 569                 }
 570                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 571                         goto next;
 572                 }
 573
 574                 start_found = 1;
 575                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 576                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 577 next:
 578                 path->slots[0]++;
 579                 cond_resched();
 580         }
 581 check_pending:
 582         /* we have to make sure we didn't find an extent that has already
 583          * been allocated by the map tree or the original allocation
 584          */
 585         btrfs_release_path(root, path);
 586         BUG_ON(*start < search_start);
 587
 588         if (*start + num_bytes > search_end) {
 589                 ret = -ENOSPC;
 590                 goto error;
 591         }
 592         /* check for pending inserts here */
 593         return 0;
 594
 595 error:
 596         btrfs_release_path(root, path);
 597         return ret;
 598 }
 599
 600 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 601                           struct btrfs_device *device,
 602                           u64 start)
 603 {
 604         int ret;
 605         struct btrfs_path *path;
 606         struct btrfs_root *root = device->dev_root;
 607         struct btrfs_key key;
 608         struct btrfs_key found_key;
 609         struct extent_buffer *leaf = NULL;
 610         struct btrfs_dev_extent *extent = NULL;
 611
 612         path = btrfs_alloc_path();
 613         if (!path)
 614                 return -ENOMEM;
 615
 616         key.objectid = device->devid;
 617         key.offset = start;
 618         key.type = BTRFS_DEV_EXTENT_KEY;
 619
 620         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 621         if (ret > 0) {
 622                 ret = btrfs_previous_item(root, path, key.objectid,
 623                                           BTRFS_DEV_EXTENT_KEY);
 624                 BUG_ON(ret);
 625                 leaf = path->nodes[0];
 626                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 627                 extent = btrfs_item_ptr(leaf, path->slots[0],
 628                                         struct btrfs_dev_extent);
 629                 BUG_ON(found_key.offset > start || found_key.offset +
 630                        btrfs_dev_extent_length(leaf, extent) < start);
 631                 ret = 0;
 632         } else if (ret == 0) {
 633                 leaf = path->nodes[0];
 634                 extent = btrfs_item_ptr(leaf, path->slots[0],
 635                                         struct btrfs_dev_extent);
 636         }
 637         BUG_ON(ret);
 638
 639         if (device->bytes_used > 0)
 640                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 641         ret = btrfs_del_item(trans, root, path);
 642         BUG_ON(ret);
 643
 644         btrfs_free_path(path);
 645         return ret;
 646 }
 647
 648 int noinline btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 649                            struct btrfs_device *device,
 650                            u64 chunk_tree, u64 chunk_objectid,
 651                            u64 chunk_offset,
 652                            u64 num_bytes, u64 *start)
 653 {
 654         int ret;
 655         struct btrfs_path *path;
 656         struct btrfs_root *root = device->dev_root;
 657         struct btrfs_dev_extent *extent;
 658         struct extent_buffer *leaf;
 659         struct btrfs_key key;
 660
 661         WARN_ON(!device->in_fs_metadata);
 662         path = btrfs_alloc_path();
 663         if (!path)
 664                 return -ENOMEM;
 665
 666         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 667         if (ret) {
 668                 goto err;
 669         }
 670
 671         key.objectid = device->devid;
 672         key.offset = *start;
 673         key.type = BTRFS_DEV_EXTENT_KEY;
 674         ret = btrfs_insert_empty_item(trans, root, path, &key,
 675                                       sizeof(*extent));
 676         BUG_ON(ret);
 677
 678         leaf = path->nodes[0];
 679         extent = btrfs_item_ptr(leaf, path->slots[0],
 680                                 struct btrfs_dev_extent);
 681         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 682         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 683         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 684
 685         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 686                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 687                     BTRFS_UUID_SIZE);
 688
 689         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 690         btrfs_mark_buffer_dirty(leaf);
 691 err:
 692         btrfs_free_path(path);
 693         return ret;
 694 }
 695
 696 static noinline int find_next_chunk(struct btrfs_root *root,
 697                                     u64 objectid, u64 *offset)
 698 {
 699         struct btrfs_path *path;
 700         int ret;
 701         struct btrfs_key key;
 702         struct btrfs_chunk *chunk;
 703         struct btrfs_key found_key;
 704
 705         path = btrfs_alloc_path();
 706         BUG_ON(!path);
 707
 708         key.objectid = objectid;
 709         key.offset = (u64)-1;
 710         key.type = BTRFS_CHUNK_ITEM_KEY;
 711
 712         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 713         if (ret < 0)
 714                 goto error;
 715
 716         BUG_ON(ret == 0);
 717
 718         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 719         if (ret) {
 720                 *offset = 0;
 721         } else {
 722                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 723                                       path->slots[0]);
 724                 if (found_key.objectid != objectid)
 725                         *offset = 0;
 726                 else {
 727                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 728                                                struct btrfs_chunk);
 729                         *offset = found_key.offset +
 730                                 btrfs_chunk_length(path->nodes[0], chunk);
 731                 }
 732         }
 733         ret = 0;
 734 error:
 735         btrfs_free_path(path);
 736         return ret;
 737 }
 738
 739 static noinline int find_next_devid(struct btrfs_root *root,
 740                                     struct btrfs_path *path, u64 *objectid)
 741 {
 742         int ret;
 743         struct btrfs_key key;
 744         struct btrfs_key found_key;
 745
 746         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 747         key.type = BTRFS_DEV_ITEM_KEY;
 748         key.offset = (u64)-1;
 749
 750         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 751         if (ret < 0)
 752                 goto error;
 753
 754         BUG_ON(ret == 0);
 755
 756         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 757                                   BTRFS_DEV_ITEM_KEY);
 758         if (ret) {
 759                 *objectid = 1;
 760         } else {
 761                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 762                                       path->slots[0]);
 763                 *objectid = found_key.offset + 1;
 764         }
 765         ret = 0;
 766 error:
 767         btrfs_release_path(root, path);
 768         return ret;
 769 }
 770
 771 /*
 772  * the device information is stored in the chunk root
 773  * the btrfs_device struct should be fully filled in
 774  */
 775 int btrfs_add_device(struct btrfs_trans_handle *trans,
 776                      struct btrfs_root *root,
 777                      struct btrfs_device *device)
 778 {
 779         int ret;
 780         struct btrfs_path *path;
 781         struct btrfs_dev_item *dev_item;
 782         struct extent_buffer *leaf;
 783         struct btrfs_key key;
 784         unsigned long ptr;
 785         u64 free_devid = 0;
 786
 787         root = root->fs_info->chunk_root;
 788
 789         path = btrfs_alloc_path();
 790         if (!path)
 791                 return -ENOMEM;
 792
 793         ret = find_next_devid(root, path, &free_devid);
 794         if (ret)
 795                 goto out;
 796
 797         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 798         key.type = BTRFS_DEV_ITEM_KEY;
 799         key.offset = free_devid;
 800
 801         ret = btrfs_insert_empty_item(trans, root, path, &key,
 802                                       sizeof(*dev_item));
 803         if (ret)
 804                 goto out;
 805
 806         leaf = path->nodes[0];
 807         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 808
 809         device->devid = free_devid;
 810         btrfs_set_device_id(leaf, dev_item, device->devid);
 811         btrfs_set_device_type(leaf, dev_item, device->type);
 812         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 813         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 814         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 815         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 816         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 817         btrfs_set_device_group(leaf, dev_item, 0);
 818         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 819         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 820
 821         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 822         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 823         btrfs_mark_buffer_dirty(leaf);
 824         ret = 0;
 825
 826 out:
 827         btrfs_free_path(path);
 828         return ret;
 829 }
 830
 831 static int btrfs_rm_dev_item(struct btrfs_root *root,
 832                              struct btrfs_device *device)
 833 {
 834         int ret;
 835         struct btrfs_path *path;
 836         struct block_device *bdev = device->bdev;
 837         struct btrfs_device *next_dev;
 838         struct btrfs_key key;
 839         u64 total_bytes;
 840         struct btrfs_fs_devices *fs_devices;
 841         struct btrfs_trans_handle *trans;
 842
 843         root = root->fs_info->chunk_root;
 844
 845         path = btrfs_alloc_path();
 846         if (!path)
 847                 return -ENOMEM;
 848
 849         trans = btrfs_start_transaction(root, 1);
 850         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 851         key.type = BTRFS_DEV_ITEM_KEY;
 852         key.offset = device->devid;
 853         lock_chunks(root);
 854
 855         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 856         if (ret < 0)
 857                 goto out;
 858
 859         if (ret > 0) {
 860                 ret = -ENOENT;
 861                 goto out;
 862         }
 863
 864         ret = btrfs_del_item(trans, root, path);
 865         if (ret)
 866                 goto out;
 867
 868         /*
 869          * at this point, the device is zero sized.  We want to
 870          * remove it from the devices list and zero out the old super
 871          */
 872         list_del_init(&device->dev_list);
 873         list_del_init(&device->dev_alloc_list);
 874         fs_devices = root->fs_info->fs_devices;
 875
 876         next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
 877                               dev_list);
 878         if (bdev == root->fs_info->sb->s_bdev)
 879                 root->fs_info->sb->s_bdev = next_dev->bdev;
 880         if (bdev == fs_devices->latest_bdev)
 881                 fs_devices->latest_bdev = next_dev->bdev;
 882
 883         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 884         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 885                                     total_bytes - 1);
 886 out:
 887         btrfs_free_path(path);
 888         unlock_chunks(root);
 889         btrfs_commit_transaction(trans, root);
 890         return ret;
 891 }
 892
 893 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 894 {
 895         struct btrfs_device *device;
 896         struct block_device *bdev;
 897         struct buffer_head *bh = NULL;
 898         struct btrfs_super_block *disk_super;
 899         u64 all_avail;
 900         u64 devid;
 901         int ret = 0;
 902
 903         mutex_lock(&uuid_mutex);
 904         mutex_lock(&root->fs_info->volume_mutex);
 905
 906         all_avail = root->fs_info->avail_data_alloc_bits |
 907                 root->fs_info->avail_system_alloc_bits |
 908                 root->fs_info->avail_metadata_alloc_bits;
 909
 910         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 911             btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
 912                 printk("btrfs: unable to go below four devices on raid10\n");
 913                 ret = -EINVAL;
 914                 goto out;
 915         }
 916
 917         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 918             btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
 919                 printk("btrfs: unable to go below two devices on raid1\n");
 920                 ret = -EINVAL;
 921                 goto out;
 922         }
 923
 924         if (strcmp(device_path, "missing") == 0) {
 925                 struct list_head *cur;
 926                 struct list_head *devices;
 927                 struct btrfs_device *tmp;
 928
 929                 device = NULL;
 930                 devices = &root->fs_info->fs_devices->devices;
 931                 list_for_each(cur, devices) {
 932                         tmp = list_entry(cur, struct btrfs_device, dev_list);
 933                         if (tmp->in_fs_metadata && !tmp->bdev) {
 934                                 device = tmp;
 935                                 break;
 936                         }
 937                 }
 938                 bdev = NULL;
 939                 bh = NULL;
 940                 disk_super = NULL;
 941                 if (!device) {
 942                         printk("btrfs: no missing devices found to remove\n");
 943                         goto out;
 944                 }
 945
 946         } else {
 947                 bdev = open_bdev_excl(device_path, 0,
 948                                       root->fs_info->bdev_holder);
 949                 if (IS_ERR(bdev)) {
 950                         ret = PTR_ERR(bdev);
 951                         goto out;
 952                 }
 953
 954                 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 955                 if (!bh) {
 956                         ret = -EIO;
 957                         goto error_close;
 958                 }
 959                 disk_super = (struct btrfs_super_block *)bh->b_data;
 960                 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 961                     sizeof(disk_super->magic))) {
 962                         ret = -ENOENT;
 963                         goto error_brelse;
 964                 }
 965                 if (memcmp(disk_super->fsid, root->fs_info->fsid,
 966                            BTRFS_FSID_SIZE)) {
 967                         ret = -ENOENT;
 968                         goto error_brelse;
 969                 }
 970                 devid = le64_to_cpu(disk_super->dev_item.devid);
 971                 device = btrfs_find_device(root, devid, NULL);
 972                 if (!device) {
 973                         ret = -ENOENT;
 974                         goto error_brelse;
 975                 }
 976
 977         }
 978         root->fs_info->fs_devices->num_devices--;
 979         root->fs_info->fs_devices->open_devices--;
 980
 981         ret = btrfs_shrink_device(device, 0);
 982         if (ret)
 983                 goto error_brelse;
 984
 985
 986         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
 987         if (ret)
 988                 goto error_brelse;
 989
 990         if (bh) {
 991                 /* make sure this device isn't detected as part of
 992                  * the FS anymore
 993                  */
 994                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
 995                 set_buffer_dirty(bh);
 996                 sync_dirty_buffer(bh);
 997
 998                 brelse(bh);
 999         }
1000
1001         if (device->bdev) {
1002                 /* one close for the device struct or super_block */
1003                 close_bdev_excl(device->bdev);
1004         }
1005         if (bdev) {
1006                 /* one close for us */
1007                 close_bdev_excl(bdev);
1008         }
1009         kfree(device->name);
1010         kfree(device);
1011         ret = 0;
1012         goto out;
1013
1014 error_brelse:
1015         brelse(bh);
1016 error_close:
1017         if (bdev)
1018                 close_bdev_excl(bdev);
1019 out:
1020         mutex_unlock(&root->fs_info->volume_mutex);
1021         mutex_unlock(&uuid_mutex);
1022         return ret;
1023 }
1024
1025 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1026 {
1027         struct btrfs_trans_handle *trans;
1028         struct btrfs_device *device;
1029         struct block_device *bdev;
1030         struct list_head *cur;
1031         struct list_head *devices;
1032         u64 total_bytes;
1033         int ret = 0;
1034
1035
1036         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1037         if (!bdev) {
1038                 return -EIO;
1039         }
1040
1041         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1042         mutex_lock(&root->fs_info->volume_mutex);
1043
1044         trans = btrfs_start_transaction(root, 1);
1045         lock_chunks(root);
1046         devices = &root->fs_info->fs_devices->devices;
1047         list_for_each(cur, devices) {
1048                 device = list_entry(cur, struct btrfs_device, dev_list);
1049                 if (device->bdev == bdev) {
1050                         ret = -EEXIST;
1051                         goto out;
1052                 }
1053         }
1054
1055         device = kzalloc(sizeof(*device), GFP_NOFS);
1056         if (!device) {
1057                 /* we can safely leave the fs_devices entry around */
1058                 ret = -ENOMEM;
1059                 goto out_close_bdev;
1060         }
1061
1062         device->barriers = 1;
1063         device->work.func = pending_bios_fn;
1064         generate_random_uuid(device->uuid);
1065         spin_lock_init(&device->io_lock);
1066         device->name = kstrdup(device_path, GFP_NOFS);
1067         if (!device->name) {
1068                 kfree(device);
1069                 goto out_close_bdev;
1070         }
1071         device->io_width = root->sectorsize;
1072         device->io_align = root->sectorsize;
1073         device->sector_size = root->sectorsize;
1074         device->total_bytes = i_size_read(bdev->bd_inode);
1075         device->dev_root = root->fs_info->dev_root;
1076         device->bdev = bdev;
1077         device->in_fs_metadata = 1;
1078
1079         ret = btrfs_add_device(trans, root, device);
1080         if (ret)
1081                 goto out_close_bdev;
1082
1083         set_blocksize(device->bdev, 4096);
1084
1085         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1086         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1087                                     total_bytes + device->total_bytes);
1088
1089         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1090         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1091                                     total_bytes + 1);
1092
1093         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1094         list_add(&device->dev_alloc_list,
1095                  &root->fs_info->fs_devices->alloc_list);
1096         root->fs_info->fs_devices->num_devices++;
1097         root->fs_info->fs_devices->open_devices++;
1098 out:
1099         unlock_chunks(root);
1100         btrfs_end_transaction(trans, root);
1101         mutex_unlock(&root->fs_info->volume_mutex);
1102
1103         return ret;
1104
1105 out_close_bdev:
1106         close_bdev_excl(bdev);
1107         goto out;
1108 }
1109
1110 int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1111                                  struct btrfs_device *device)
1112 {
1113         int ret;
1114         struct btrfs_path *path;
1115         struct btrfs_root *root;
1116         struct btrfs_dev_item *dev_item;
1117         struct extent_buffer *leaf;
1118         struct btrfs_key key;
1119
1120         root = device->dev_root->fs_info->chunk_root;
1121
1122         path = btrfs_alloc_path();
1123         if (!path)
1124                 return -ENOMEM;
1125
1126         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1127         key.type = BTRFS_DEV_ITEM_KEY;
1128         key.offset = device->devid;
1129
1130         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1131         if (ret < 0)
1132                 goto out;
1133
1134         if (ret > 0) {
1135                 ret = -ENOENT;
1136                 goto out;
1137         }
1138
1139         leaf = path->nodes[0];
1140         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1141
1142         btrfs_set_device_id(leaf, dev_item, device->devid);
1143         btrfs_set_device_type(leaf, dev_item, device->type);
1144         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1145         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1146         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1147         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1148         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1149         btrfs_mark_buffer_dirty(leaf);
1150
1151 out:
1152         btrfs_free_path(path);
1153         return ret;
1154 }
1155
1156 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1157                       struct btrfs_device *device, u64 new_size)
1158 {
1159         struct btrfs_super_block *super_copy =
1160                 &device->dev_root->fs_info->super_copy;
1161         u64 old_total = btrfs_super_total_bytes(super_copy);
1162         u64 diff = new_size - device->total_bytes;
1163
1164         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1165         return btrfs_update_device(trans, device);
1166 }
1167
1168 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1169                       struct btrfs_device *device, u64 new_size)
1170 {
1171         int ret;
1172         lock_chunks(device->dev_root);
1173         ret = __btrfs_grow_device(trans, device, new_size);
1174         unlock_chunks(device->dev_root);
1175         return ret;
1176 }
1177
1178 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1179                             struct btrfs_root *root,
1180                             u64 chunk_tree, u64 chunk_objectid,
1181                             u64 chunk_offset)
1182 {
1183         int ret;
1184         struct btrfs_path *path;
1185         struct btrfs_key key;
1186
1187         root = root->fs_info->chunk_root;
1188         path = btrfs_alloc_path();
1189         if (!path)
1190                 return -ENOMEM;
1191
1192         key.objectid = chunk_objectid;
1193         key.offset = chunk_offset;
1194         key.type = BTRFS_CHUNK_ITEM_KEY;
1195
1196         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1197         BUG_ON(ret);
1198
1199         ret = btrfs_del_item(trans, root, path);
1200         BUG_ON(ret);
1201
1202         btrfs_free_path(path);
1203         return 0;
1204 }
1205
1206 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1207                         chunk_offset)
1208 {
1209         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1210         struct btrfs_disk_key *disk_key;
1211         struct btrfs_chunk *chunk;
1212         u8 *ptr;
1213         int ret = 0;
1214         u32 num_stripes;
1215         u32 array_size;
1216         u32 len = 0;
1217         u32 cur;
1218         struct btrfs_key key;
1219
1220         array_size = btrfs_super_sys_array_size(super_copy);
1221
1222         ptr = super_copy->sys_chunk_array;
1223         cur = 0;
1224
1225         while (cur < array_size) {
1226                 disk_key = (struct btrfs_disk_key *)ptr;
1227                 btrfs_disk_key_to_cpu(&key, disk_key);
1228
1229                 len = sizeof(*disk_key);
1230
1231                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1232                         chunk = (struct btrfs_chunk *)(ptr + len);
1233                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1234                         len += btrfs_chunk_item_size(num_stripes);
1235                 } else {
1236                         ret = -EIO;
1237                         break;
1238                 }
1239                 if (key.objectid == chunk_objectid &&
1240                     key.offset == chunk_offset) {
1241                         memmove(ptr, ptr + len, array_size - (cur + len));
1242                         array_size -= len;
1243                         btrfs_set_super_sys_array_size(super_copy, array_size);
1244                 } else {
1245                         ptr += len;
1246                         cur += len;
1247                 }
1248         }
1249         return ret;
1250 }
1251
1252
1253 int btrfs_relocate_chunk(struct btrfs_root *root,
1254                          u64 chunk_tree, u64 chunk_objectid,
1255                          u64 chunk_offset)
1256 {
1257         struct extent_map_tree *em_tree;
1258         struct btrfs_root *extent_root;
1259         struct btrfs_trans_handle *trans;
1260         struct extent_map *em;
1261         struct map_lookup *map;
1262         int ret;
1263         int i;
1264
1265         printk("btrfs relocating chunk %llu\n",
1266                (unsigned long long)chunk_offset);
1267         root = root->fs_info->chunk_root;
1268         extent_root = root->fs_info->extent_root;
1269         em_tree = &root->fs_info->mapping_tree.map_tree;
1270
1271         /* step one, relocate all the extents inside this chunk */
1272         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1273         BUG_ON(ret);
1274
1275         trans = btrfs_start_transaction(root, 1);
1276         BUG_ON(!trans);
1277
1278         lock_chunks(root);
1279
1280         /*
1281          * step two, delete the device extents and the
1282          * chunk tree entries
1283          */
1284         spin_lock(&em_tree->lock);
1285         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1286         spin_unlock(&em_tree->lock);
1287
1288         BUG_ON(em->start > chunk_offset ||
1289                em->start + em->len < chunk_offset);
1290         map = (struct map_lookup *)em->bdev;
1291
1292         for (i = 0; i < map->num_stripes; i++) {
1293                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1294                                             map->stripes[i].physical);
1295                 BUG_ON(ret);
1296
1297                 if (map->stripes[i].dev) {
1298                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1299                         BUG_ON(ret);
1300                 }
1301         }
1302         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1303                                chunk_offset);
1304
1305         BUG_ON(ret);
1306
1307         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1308                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1309                 BUG_ON(ret);
1310         }
1311
1312         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1313         BUG_ON(ret);
1314
1315         spin_lock(&em_tree->lock);
1316         remove_extent_mapping(em_tree, em);
1317         spin_unlock(&em_tree->lock);
1318
1319         kfree(map);
1320         em->bdev = NULL;
1321
1322         /* once for the tree */
1323         free_extent_map(em);
1324         /* once for us */
1325         free_extent_map(em);
1326
1327         unlock_chunks(root);
1328         btrfs_end_transaction(trans, root);
1329         return 0;
1330 }
1331
1332 static u64 div_factor(u64 num, int factor)
1333 {
1334         if (factor == 10)
1335                 return num;
1336         num *= factor;
1337         do_div(num, 10);
1338         return num;
1339 }
1340
1341
1342 int btrfs_balance(struct btrfs_root *dev_root)
1343 {
1344         int ret;
1345         struct list_head *cur;
1346         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1347         struct btrfs_device *device;
1348         u64 old_size;
1349         u64 size_to_free;
1350         struct btrfs_path *path;
1351         struct btrfs_key key;
1352         struct btrfs_chunk *chunk;
1353         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1354         struct btrfs_trans_handle *trans;
1355         struct btrfs_key found_key;
1356
1357
1358         mutex_lock(&dev_root->fs_info->volume_mutex);
1359         dev_root = dev_root->fs_info->dev_root;
1360
1361         /* step one make some room on all the devices */
1362         list_for_each(cur, devices) {
1363                 device = list_entry(cur, struct btrfs_device, dev_list);
1364                 old_size = device->total_bytes;
1365                 size_to_free = div_factor(old_size, 1);
1366                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1367                 if (device->total_bytes - device->bytes_used > size_to_free)
1368                         continue;
1369
1370                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1371                 BUG_ON(ret);
1372
1373                 trans = btrfs_start_transaction(dev_root, 1);
1374                 BUG_ON(!trans);
1375
1376                 ret = btrfs_grow_device(trans, device, old_size);
1377                 BUG_ON(ret);
1378
1379                 btrfs_end_transaction(trans, dev_root);
1380         }
1381
1382         /* step two, relocate all the chunks */
1383         path = btrfs_alloc_path();
1384         BUG_ON(!path);
1385
1386         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1387         key.offset = (u64)-1;
1388         key.type = BTRFS_CHUNK_ITEM_KEY;
1389
1390         while(1) {
1391                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1392                 if (ret < 0)
1393                         goto error;
1394
1395                 /*
1396                  * this shouldn't happen, it means the last relocate
1397                  * failed
1398                  */
1399                 if (ret == 0)
1400                         break;
1401
1402                 ret = btrfs_previous_item(chunk_root, path, 0,
1403                                           BTRFS_CHUNK_ITEM_KEY);
1404                 if (ret)
1405                         break;
1406
1407                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1408                                       path->slots[0]);
1409                 if (found_key.objectid != key.objectid)
1410                         break;
1411
1412                 chunk = btrfs_item_ptr(path->nodes[0],
1413                                        path->slots[0],
1414                                        struct btrfs_chunk);
1415                 key.offset = found_key.offset;
1416                 /* chunk zero is special */
1417                 if (key.offset == 0)
1418                         break;
1419
1420                 btrfs_release_path(chunk_root, path);
1421                 ret = btrfs_relocate_chunk(chunk_root,
1422                                            chunk_root->root_key.objectid,
1423                                            found_key.objectid,
1424                                            found_key.offset);
1425                 BUG_ON(ret);
1426         }
1427         ret = 0;
1428 error:
1429         btrfs_free_path(path);
1430         mutex_unlock(&dev_root->fs_info->volume_mutex);
1431         return ret;
1432 }
1433
1434 /*
1435  * shrinking a device means finding all of the device extents past
1436  * the new size, and then following the back refs to the chunks.
1437  * The chunk relocation code actually frees the device extent
1438  */
1439 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1440 {
1441         struct btrfs_trans_handle *trans;
1442         struct btrfs_root *root = device->dev_root;
1443         struct btrfs_dev_extent *dev_extent = NULL;
1444         struct btrfs_path *path;
1445         u64 length;
1446         u64 chunk_tree;
1447         u64 chunk_objectid;
1448         u64 chunk_offset;
1449         int ret;
1450         int slot;
1451         struct extent_buffer *l;
1452         struct btrfs_key key;
1453         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1454         u64 old_total = btrfs_super_total_bytes(super_copy);
1455         u64 diff = device->total_bytes - new_size;
1456
1457
1458         path = btrfs_alloc_path();
1459         if (!path)
1460                 return -ENOMEM;
1461
1462         trans = btrfs_start_transaction(root, 1);
1463         if (!trans) {
1464                 ret = -ENOMEM;
1465                 goto done;
1466         }
1467
1468         path->reada = 2;
1469
1470         lock_chunks(root);
1471
1472         device->total_bytes = new_size;
1473         ret = btrfs_update_device(trans, device);
1474         if (ret) {
1475                 unlock_chunks(root);
1476                 btrfs_end_transaction(trans, root);
1477                 goto done;
1478         }
1479         WARN_ON(diff > old_total);
1480         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1481         unlock_chunks(root);
1482         btrfs_end_transaction(trans, root);
1483
1484         key.objectid = device->devid;
1485         key.offset = (u64)-1;
1486         key.type = BTRFS_DEV_EXTENT_KEY;
1487
1488         while (1) {
1489                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1490                 if (ret < 0)
1491                         goto done;
1492
1493                 ret = btrfs_previous_item(root, path, 0, key.type);
1494                 if (ret < 0)
1495                         goto done;
1496                 if (ret) {
1497                         ret = 0;
1498                         goto done;
1499                 }
1500
1501                 l = path->nodes[0];
1502                 slot = path->slots[0];
1503                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1504
1505                 if (key.objectid != device->devid)
1506                         goto done;
1507
1508                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1509                 length = btrfs_dev_extent_length(l, dev_extent);
1510
1511                 if (key.offset + length <= new_size)
1512                         goto done;
1513
1514                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1515                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1516                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1517                 btrfs_release_path(root, path);
1518
1519                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1520                                            chunk_offset);
1521                 if (ret)
1522                         goto done;
1523         }
1524
1525 done:
1526         btrfs_free_path(path);
1527         return ret;
1528 }
1529
1530 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1531                            struct btrfs_root *root,
1532                            struct btrfs_key *key,
1533                            struct btrfs_chunk *chunk, int item_size)
1534 {
1535         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1536         struct btrfs_disk_key disk_key;
1537         u32 array_size;
1538         u8 *ptr;
1539
1540         array_size = btrfs_super_sys_array_size(super_copy);
1541         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1542                 return -EFBIG;
1543
1544         ptr = super_copy->sys_chunk_array + array_size;
1545         btrfs_cpu_key_to_disk(&disk_key, key);
1546         memcpy(ptr, &disk_key, sizeof(disk_key));
1547         ptr += sizeof(disk_key);
1548         memcpy(ptr, chunk, item_size);
1549         item_size += sizeof(disk_key);
1550         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1551         return 0;
1552 }
1553
1554 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1555                                         int num_stripes, int sub_stripes)
1556 {
1557         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1558                 return calc_size;
1559         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1560                 return calc_size * (num_stripes / sub_stripes);
1561         else
1562                 return calc_size * num_stripes;
1563 }
1564
1565
1566 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1567                       struct btrfs_root *extent_root, u64 *start,
1568                       u64 *num_bytes, u64 type)
1569 {
1570         u64 dev_offset;
1571         struct btrfs_fs_info *info = extent_root->fs_info;
1572         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1573         struct btrfs_path *path;
1574         struct btrfs_stripe *stripes;
1575         struct btrfs_device *device = NULL;
1576         struct btrfs_chunk *chunk;
1577         struct list_head private_devs;
1578         struct list_head *dev_list;
1579         struct list_head *cur;
1580         struct extent_map_tree *em_tree;
1581         struct map_lookup *map;
1582         struct extent_map *em;
1583         int min_stripe_size = 1 * 1024 * 1024;
1584         u64 physical;
1585         u64 calc_size = 1024 * 1024 * 1024;
1586         u64 max_chunk_size = calc_size;
1587         u64 min_free;
1588         u64 avail;
1589         u64 max_avail = 0;
1590         u64 percent_max;
1591         int num_stripes = 1;
1592         int min_stripes = 1;
1593         int sub_stripes = 0;
1594         int looped = 0;
1595         int ret;
1596         int index;
1597         int stripe_len = 64 * 1024;
1598         struct btrfs_key key;
1599
1600         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1601             (type & BTRFS_BLOCK_GROUP_DUP)) {
1602                 WARN_ON(1);
1603                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1604         }
1605         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1606         if (list_empty(dev_list))
1607                 return -ENOSPC;
1608
1609         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1610                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1611                 min_stripes = 2;
1612         }
1613         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1614                 num_stripes = 2;
1615                 min_stripes = 2;
1616         }
1617         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1618                 num_stripes = min_t(u64, 2,
1619                             extent_root->fs_info->fs_devices->open_devices);
1620                 if (num_stripes < 2)
1621                         return -ENOSPC;
1622                 min_stripes = 2;
1623         }
1624         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1625                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1626                 if (num_stripes < 4)
1627                         return -ENOSPC;
1628                 num_stripes &= ~(u32)1;
1629                 sub_stripes = 2;
1630                 min_stripes = 4;
1631         }
1632
1633         if (type & BTRFS_BLOCK_GROUP_DATA) {
1634                 max_chunk_size = 10 * calc_size;
1635                 min_stripe_size = 64 * 1024 * 1024;
1636         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1637                 max_chunk_size = 4 * calc_size;
1638                 min_stripe_size = 32 * 1024 * 1024;
1639         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1640                 calc_size = 8 * 1024 * 1024;
1641                 max_chunk_size = calc_size * 2;
1642                 min_stripe_size = 1 * 1024 * 1024;
1643         }
1644
1645         path = btrfs_alloc_path();
1646         if (!path)
1647                 return -ENOMEM;
1648
1649         /* we don't want a chunk larger than 10% of the FS */
1650         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1651         max_chunk_size = min(percent_max, max_chunk_size);
1652
1653 again:
1654         if (calc_size * num_stripes > max_chunk_size) {
1655                 calc_size = max_chunk_size;
1656                 do_div(calc_size, num_stripes);
1657                 do_div(calc_size, stripe_len);
1658                 calc_size *= stripe_len;
1659         }
1660         /* we don't want tiny stripes */
1661         calc_size = max_t(u64, min_stripe_size, calc_size);
1662
1663         do_div(calc_size, stripe_len);
1664         calc_size *= stripe_len;
1665
1666         INIT_LIST_HEAD(&private_devs);
1667         cur = dev_list->next;
1668         index = 0;
1669
1670         if (type & BTRFS_BLOCK_GROUP_DUP)
1671                 min_free = calc_size * 2;
1672         else
1673                 min_free = calc_size;
1674
1675         /*
1676          * we add 1MB because we never use the first 1MB of the device, unless
1677          * we've looped, then we are likely allocating the maximum amount of
1678          * space left already
1679          */
1680         if (!looped)
1681                 min_free += 1024 * 1024;
1682
1683         /* build a private list of devices we will allocate from */
1684         while(index < num_stripes) {
1685                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1686
1687                 if (device->total_bytes > device->bytes_used)
1688                         avail = device->total_bytes - device->bytes_used;
1689                 else
1690                         avail = 0;
1691                 cur = cur->next;
1692
1693                 if (device->in_fs_metadata && avail >= min_free) {
1694                         u64 ignored_start = 0;
1695                         ret = find_free_dev_extent(trans, device, path,
1696                                                    min_free,
1697                                                    &ignored_start);
1698                         if (ret == 0) {
1699                                 list_move_tail(&device->dev_alloc_list,
1700                                                &private_devs);
1701                                 index++;
1702                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1703                                         index++;
1704                         }
1705                 } else if (device->in_fs_metadata && avail > max_avail)
1706                         max_avail = avail;
1707                 if (cur == dev_list)
1708                         break;
1709         }
1710         if (index < num_stripes) {
1711                 list_splice(&private_devs, dev_list);
1712                 if (index >= min_stripes) {
1713                         num_stripes = index;
1714                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1715                                 num_stripes /= sub_stripes;
1716                                 num_stripes *= sub_stripes;
1717                         }
1718                         looped = 1;
1719                         goto again;
1720                 }
1721                 if (!looped && max_avail > 0) {
1722                         looped = 1;
1723                         calc_size = max_avail;
1724                         goto again;
1725                 }
1726                 btrfs_free_path(path);
1727                 return -ENOSPC;
1728         }
1729         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1730         key.type = BTRFS_CHUNK_ITEM_KEY;
1731         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1732                               &key.offset);
1733         if (ret) {
1734                 btrfs_free_path(path);
1735                 return ret;
1736         }
1737
1738         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1739         if (!chunk) {
1740                 btrfs_free_path(path);
1741                 return -ENOMEM;
1742         }
1743
1744         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1745         if (!map) {
1746                 kfree(chunk);
1747                 btrfs_free_path(path);
1748                 return -ENOMEM;
1749         }
1750         btrfs_free_path(path);
1751         path = NULL;
1752
1753         stripes = &chunk->stripe;
1754         *num_bytes = chunk_bytes_by_type(type, calc_size,
1755                                          num_stripes, sub_stripes);
1756
1757         index = 0;
1758         while(index < num_stripes) {
1759                 struct btrfs_stripe *stripe;
1760                 BUG_ON(list_empty(&private_devs));
1761                 cur = private_devs.next;
1762                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1763
1764                 /* loop over this device again if we're doing a dup group */
1765                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1766                     (index == num_stripes - 1))
1767                         list_move_tail(&device->dev_alloc_list, dev_list);
1768
1769                 ret = btrfs_alloc_dev_extent(trans, device,
1770                              info->chunk_root->root_key.objectid,
1771                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1772                              calc_size, &dev_offset);
1773                 BUG_ON(ret);
1774                 device->bytes_used += calc_size;
1775                 ret = btrfs_update_device(trans, device);
1776                 BUG_ON(ret);
1777
1778                 map->stripes[index].dev = device;
1779                 map->stripes[index].physical = dev_offset;
1780                 stripe = stripes + index;
1781                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1782                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1783                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1784                 physical = dev_offset;
1785                 index++;
1786         }
1787         BUG_ON(!list_empty(&private_devs));
1788
1789         /* key was set above */
1790         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1791         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1792         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1793         btrfs_set_stack_chunk_type(chunk, type);
1794         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1795         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1796         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1797         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1798         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1799         map->sector_size = extent_root->sectorsize;
1800         map->stripe_len = stripe_len;
1801         map->io_align = stripe_len;
1802         map->io_width = stripe_len;
1803         map->type = type;
1804         map->num_stripes = num_stripes;
1805         map->sub_stripes = sub_stripes;
1806
1807         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1808                                 btrfs_chunk_item_size(num_stripes));
1809         BUG_ON(ret);
1810         *start = key.offset;;
1811
1812         em = alloc_extent_map(GFP_NOFS);
1813         if (!em)
1814                 return -ENOMEM;
1815         em->bdev = (struct block_device *)map;
1816         em->start = key.offset;
1817         em->len = *num_bytes;
1818         em->block_start = 0;
1819
1820         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1821                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1822                                     chunk, btrfs_chunk_item_size(num_stripes));
1823                 BUG_ON(ret);
1824         }
1825         kfree(chunk);
1826
1827         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1828         spin_lock(&em_tree->lock);
1829         ret = add_extent_mapping(em_tree, em);
1830         spin_unlock(&em_tree->lock);
1831         BUG_ON(ret);
1832         free_extent_map(em);
1833         return ret;
1834 }
1835
1836 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1837 {
1838         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1839 }
1840
1841 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1842 {
1843         struct extent_map *em;
1844
1845         while(1) {
1846                 spin_lock(&tree->map_tree.lock);
1847                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1848                 if (em)
1849                         remove_extent_mapping(&tree->map_tree, em);
1850                 spin_unlock(&tree->map_tree.lock);
1851                 if (!em)
1852                         break;
1853                 kfree(em->bdev);
1854                 /* once for us */
1855                 free_extent_map(em);
1856                 /* once for the tree */
1857                 free_extent_map(em);
1858         }
1859 }
1860
1861 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1862 {
1863         struct extent_map *em;
1864         struct map_lookup *map;
1865         struct extent_map_tree *em_tree = &map_tree->map_tree;
1866         int ret;
1867
1868         spin_lock(&em_tree->lock);
1869         em = lookup_extent_mapping(em_tree, logical, len);
1870         spin_unlock(&em_tree->lock);
1871         BUG_ON(!em);
1872
1873         BUG_ON(em->start > logical || em->start + em->len < logical);
1874         map = (struct map_lookup *)em->bdev;
1875         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1876                 ret = map->num_stripes;
1877         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1878                 ret = map->sub_stripes;
1879         else
1880                 ret = 1;
1881         free_extent_map(em);
1882         return ret;
1883 }
1884
1885 static int find_live_mirror(struct map_lookup *map, int first, int num,
1886                             int optimal)
1887 {
1888         int i;
1889         if (map->stripes[optimal].dev->bdev)
1890                 return optimal;
1891         for (i = first; i < first + num; i++) {
1892                 if (map->stripes[i].dev->bdev)
1893                         return i;
1894         }
1895         /* we couldn't find one that doesn't fail.  Just return something
1896          * and the io error handling code will clean up eventually
1897          */
1898         return optimal;
1899 }
1900
1901 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1902                              u64 logical, u64 *length,
1903                              struct btrfs_multi_bio **multi_ret,
1904                              int mirror_num, struct page *unplug_page)
1905 {
1906         struct extent_map *em;
1907         struct map_lookup *map;
1908         struct extent_map_tree *em_tree = &map_tree->map_tree;
1909         u64 offset;
1910         u64 stripe_offset;
1911         u64 stripe_nr;
1912         int stripes_allocated = 8;
1913         int stripes_required = 1;
1914         int stripe_index;
1915         int i;
1916         int num_stripes;
1917         int max_errors = 0;
1918         struct btrfs_multi_bio *multi = NULL;
1919
1920         if (multi_ret && !(rw & (1 << BIO_RW))) {
1921                 stripes_allocated = 1;
1922         }
1923 again:
1924         if (multi_ret) {
1925                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1926                                 GFP_NOFS);
1927                 if (!multi)
1928                         return -ENOMEM;
1929
1930                 atomic_set(&multi->error, 0);
1931         }
1932
1933         spin_lock(&em_tree->lock);
1934         em = lookup_extent_mapping(em_tree, logical, *length);
1935         spin_unlock(&em_tree->lock);
1936
1937         if (!em && unplug_page)
1938                 return 0;
1939
1940         if (!em) {
1941                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1942                 BUG();
1943         }
1944
1945         BUG_ON(em->start > logical || em->start + em->len < logical);
1946         map = (struct map_lookup *)em->bdev;
1947         offset = logical - em->start;
1948
1949         if (mirror_num > map->num_stripes)
1950                 mirror_num = 0;
1951
1952         /* if our multi bio struct is too small, back off and try again */
1953         if (rw & (1 << BIO_RW)) {
1954                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1955                                  BTRFS_BLOCK_GROUP_DUP)) {
1956                         stripes_required = map->num_stripes;
1957                         max_errors = 1;
1958                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1959                         stripes_required = map->sub_stripes;
1960                         max_errors = 1;
1961                 }
1962         }
1963         if (multi_ret && rw == WRITE &&
1964             stripes_allocated < stripes_required) {
1965                 stripes_allocated = map->num_stripes;
1966                 free_extent_map(em);
1967                 kfree(multi);
1968                 goto again;
1969         }
1970         stripe_nr = offset;
1971         /*
1972          * stripe_nr counts the total number of stripes we have to stride
1973          * to get to this block
1974          */
1975         do_div(stripe_nr, map->stripe_len);
1976
1977         stripe_offset = stripe_nr * map->stripe_len;
1978         BUG_ON(offset < stripe_offset);
1979
1980         /* stripe_offset is the offset of this block in its stripe*/
1981         stripe_offset = offset - stripe_offset;
1982
1983         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1984                          BTRFS_BLOCK_GROUP_RAID10 |
1985                          BTRFS_BLOCK_GROUP_DUP)) {
1986                 /* we limit the length of each bio to what fits in a stripe */
1987                 *length = min_t(u64, em->len - offset,
1988                               map->stripe_len - stripe_offset);
1989         } else {
1990                 *length = em->len - offset;
1991         }
1992
1993         if (!multi_ret && !unplug_page)
1994                 goto out;
1995
1996         num_stripes = 1;
1997         stripe_index = 0;
1998         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1999                 if (unplug_page || (rw & (1 << BIO_RW)))
2000                         num_stripes = map->num_stripes;
2001                 else if (mirror_num)
2002                         stripe_index = mirror_num - 1;
2003                 else {
2004                         stripe_index = find_live_mirror(map, 0,
2005                                             map->num_stripes,
2006                                             current->pid % map->num_stripes);
2007                 }
2008
2009         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2010                 if (rw & (1 << BIO_RW))
2011                         num_stripes = map->num_stripes;
2012                 else if (mirror_num)
2013                         stripe_index = mirror_num - 1;
2014
2015         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2016                 int factor = map->num_stripes / map->sub_stripes;
2017
2018                 stripe_index = do_div(stripe_nr, factor);
2019                 stripe_index *= map->sub_stripes;
2020
2021                 if (unplug_page || (rw & (1 << BIO_RW)))
2022                         num_stripes = map->sub_stripes;
2023                 else if (mirror_num)
2024                         stripe_index += mirror_num - 1;
2025                 else {
2026                         stripe_index = find_live_mirror(map, stripe_index,
2027                                               map->sub_stripes, stripe_index +
2028                                               current->pid % map->sub_stripes);
2029                 }
2030         } else {
2031                 /*
2032                  * after this do_div call, stripe_nr is the number of stripes
2033                  * on this device we have to walk to find the data, and
2034                  * stripe_index is the number of our device in the stripe array
2035                  */
2036                 stripe_index = do_div(stripe_nr, map->num_stripes);
2037         }
2038         BUG_ON(stripe_index >= map->num_stripes);
2039
2040         for (i = 0; i < num_stripes; i++) {
2041                 if (unplug_page) {
2042                         struct btrfs_device *device;
2043                         struct backing_dev_info *bdi;
2044
2045                         device = map->stripes[stripe_index].dev;
2046                         if (device->bdev) {
2047                                 bdi = blk_get_backing_dev_info(device->bdev);
2048                                 if (bdi->unplug_io_fn) {
2049                                         bdi->unplug_io_fn(bdi, unplug_page);
2050                                 }
2051                         }
2052                 } else {
2053                         multi->stripes[i].physical =
2054                                 map->stripes[stripe_index].physical +
2055                                 stripe_offset + stripe_nr * map->stripe_len;
2056                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2057                 }
2058                 stripe_index++;
2059         }
2060         if (multi_ret) {
2061                 *multi_ret = multi;
2062                 multi->num_stripes = num_stripes;
2063                 multi->max_errors = max_errors;
2064         }
2065 out:
2066         free_extent_map(em);
2067         return 0;
2068 }
2069
2070 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2071                       u64 logical, u64 *length,
2072                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2073 {
2074         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2075                                  mirror_num, NULL);
2076 }
2077
2078 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2079                       u64 logical, struct page *page)
2080 {
2081         u64 length = PAGE_CACHE_SIZE;
2082         return __btrfs_map_block(map_tree, READ, logical, &length,
2083                                  NULL, 0, page);
2084 }
2085
2086
2087 static void end_bio_multi_stripe(struct bio *bio, int err)
2088 {
2089         struct btrfs_multi_bio *multi = bio->bi_private;
2090         int is_orig_bio = 0;
2091
2092         if (err)
2093                 atomic_inc(&multi->error);
2094
2095         if (bio == multi->orig_bio)
2096                 is_orig_bio = 1;
2097
2098         if (atomic_dec_and_test(&multi->stripes_pending)) {
2099                 if (!is_orig_bio) {
2100                         bio_put(bio);
2101                         bio = multi->orig_bio;
2102                 }
2103                 bio->bi_private = multi->private;
2104                 bio->bi_end_io = multi->end_io;
2105                 /* only send an error to the higher layers if it is
2106                  * beyond the tolerance of the multi-bio
2107                  */
2108                 if (atomic_read(&multi->error) > multi->max_errors) {
2109                         err = -EIO;
2110                 } else if (err) {
2111                         /*
2112                          * this bio is actually up to date, we didn't
2113                          * go over the max number of errors
2114                          */
2115                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2116                         err = 0;
2117                 }
2118                 kfree(multi);
2119
2120                 bio_endio(bio, err);
2121         } else if (!is_orig_bio) {
2122                 bio_put(bio);
2123         }
2124 }
2125
2126 struct async_sched {
2127         struct bio *bio;
2128         int rw;
2129         struct btrfs_fs_info *info;
2130         struct btrfs_work work;
2131 };
2132
2133 /*
2134  * see run_scheduled_bios for a description of why bios are collected for
2135  * async submit.
2136  *
2137  * This will add one bio to the pending list for a device and make sure
2138  * the work struct is scheduled.
2139  */
2140 static int noinline schedule_bio(struct btrfs_root *root,
2141                                  struct btrfs_device *device,
2142                                  int rw, struct bio *bio)
2143 {
2144         int should_queue = 1;
2145
2146         /* don't bother with additional async steps for reads, right now */
2147         if (!(rw & (1 << BIO_RW))) {
2148                 bio_get(bio);
2149                 submit_bio(rw, bio);
2150                 bio_put(bio);
2151                 return 0;
2152         }
2153
2154         /*
2155          * nr_async_bios allows us to reliably return congestion to the
2156          * higher layers.  Otherwise, the async bio makes it appear we have
2157          * made progress against dirty pages when we've really just put it
2158          * on a queue for later
2159          */
2160         atomic_inc(&root->fs_info->nr_async_bios);
2161         WARN_ON(bio->bi_next);
2162         bio->bi_next = NULL;
2163         bio->bi_rw |= rw;
2164
2165         spin_lock(&device->io_lock);
2166
2167         if (device->pending_bio_tail)
2168                 device->pending_bio_tail->bi_next = bio;
2169
2170         device->pending_bio_tail = bio;
2171         if (!device->pending_bios)
2172                 device->pending_bios = bio;
2173         if (device->running_pending)
2174                 should_queue = 0;
2175
2176         spin_unlock(&device->io_lock);
2177
2178         if (should_queue)
2179                 btrfs_queue_worker(&root->fs_info->submit_workers,
2180                                    &device->work);
2181         return 0;
2182 }
2183
2184 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2185                   int mirror_num, int async_submit)
2186 {
2187         struct btrfs_mapping_tree *map_tree;
2188         struct btrfs_device *dev;
2189         struct bio *first_bio = bio;
2190         u64 logical = bio->bi_sector << 9;
2191         u64 length = 0;
2192         u64 map_length;
2193         struct btrfs_multi_bio *multi = NULL;
2194         int ret;
2195         int dev_nr = 0;
2196         int total_devs = 1;
2197
2198         length = bio->bi_size;
2199         map_tree = &root->fs_info->mapping_tree;
2200         map_length = length;
2201
2202         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2203                               mirror_num);
2204         BUG_ON(ret);
2205
2206         total_devs = multi->num_stripes;
2207         if (map_length < length) {
2208                 printk("mapping failed logical %Lu bio len %Lu "
2209                        "len %Lu\n", logical, length, map_length);
2210                 BUG();
2211         }
2212         multi->end_io = first_bio->bi_end_io;
2213         multi->private = first_bio->bi_private;
2214         multi->orig_bio = first_bio;
2215         atomic_set(&multi->stripes_pending, multi->num_stripes);
2216
2217         while(dev_nr < total_devs) {
2218                 if (total_devs > 1) {
2219                         if (dev_nr < total_devs - 1) {
2220                                 bio = bio_clone(first_bio, GFP_NOFS);
2221                                 BUG_ON(!bio);
2222                         } else {
2223                                 bio = first_bio;
2224                         }
2225                         bio->bi_private = multi;
2226                         bio->bi_end_io = end_bio_multi_stripe;
2227                 }
2228                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2229                 dev = multi->stripes[dev_nr].dev;
2230                 if (dev && dev->bdev) {
2231                         bio->bi_bdev = dev->bdev;
2232                         if (async_submit)
2233                                 schedule_bio(root, dev, rw, bio);
2234                         else
2235                                 submit_bio(rw, bio);
2236                 } else {
2237                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2238                         bio->bi_sector = logical >> 9;
2239                         bio_endio(bio, -EIO);
2240                 }
2241                 dev_nr++;
2242         }
2243         if (total_devs == 1)
2244                 kfree(multi);
2245         return 0;
2246 }
2247
2248 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2249                                        u8 *uuid)
2250 {
2251         struct list_head *head = &root->fs_info->fs_devices->devices;
2252
2253         return __find_device(head, devid, uuid);
2254 }
2255
2256 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2257                                             u64 devid, u8 *dev_uuid)
2258 {
2259         struct btrfs_device *device;
2260         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2261
2262         device = kzalloc(sizeof(*device), GFP_NOFS);
2263         list_add(&device->dev_list,
2264                  &fs_devices->devices);
2265         list_add(&device->dev_alloc_list,
2266                  &fs_devices->alloc_list);
2267         device->barriers = 1;
2268         device->dev_root = root->fs_info->dev_root;
2269         device->devid = devid;
2270         device->work.func = pending_bios_fn;
2271         fs_devices->num_devices++;
2272         spin_lock_init(&device->io_lock);
2273         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2274         return device;
2275 }
2276
2277
2278 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2279                           struct extent_buffer *leaf,
2280                           struct btrfs_chunk *chunk)
2281 {
2282         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2283         struct map_lookup *map;
2284         struct extent_map *em;
2285         u64 logical;
2286         u64 length;
2287         u64 devid;
2288         u8 uuid[BTRFS_UUID_SIZE];
2289         int num_stripes;
2290         int ret;
2291         int i;
2292
2293         logical = key->offset;
2294         length = btrfs_chunk_length(leaf, chunk);
2295
2296         spin_lock(&map_tree->map_tree.lock);
2297         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2298         spin_unlock(&map_tree->map_tree.lock);
2299
2300         /* already mapped? */
2301         if (em && em->start <= logical && em->start + em->len > logical) {
2302                 free_extent_map(em);
2303                 return 0;
2304         } else if (em) {
2305                 free_extent_map(em);
2306         }
2307
2308         map = kzalloc(sizeof(*map), GFP_NOFS);
2309         if (!map)
2310                 return -ENOMEM;
2311
2312         em = alloc_extent_map(GFP_NOFS);
2313         if (!em)
2314                 return -ENOMEM;
2315         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2316         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2317         if (!map) {
2318                 free_extent_map(em);
2319                 return -ENOMEM;
2320         }
2321
2322         em->bdev = (struct block_device *)map;
2323         em->start = logical;
2324         em->len = length;
2325         em->block_start = 0;
2326
2327         map->num_stripes = num_stripes;
2328         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2329         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2330         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2331         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2332         map->type = btrfs_chunk_type(leaf, chunk);
2333         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2334         for (i = 0; i < num_stripes; i++) {
2335                 map->stripes[i].physical =
2336                         btrfs_stripe_offset_nr(leaf, chunk, i);
2337                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2338                 read_extent_buffer(leaf, uuid, (unsigned long)
2339                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2340                                    BTRFS_UUID_SIZE);
2341                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2342
2343                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2344                         kfree(map);
2345                         free_extent_map(em);
2346                         return -EIO;
2347                 }
2348                 if (!map->stripes[i].dev) {
2349                         map->stripes[i].dev =
2350                                 add_missing_dev(root, devid, uuid);
2351                         if (!map->stripes[i].dev) {
2352                                 kfree(map);
2353                                 free_extent_map(em);
2354                                 return -EIO;
2355                         }
2356                 }
2357                 map->stripes[i].dev->in_fs_metadata = 1;
2358         }
2359
2360         spin_lock(&map_tree->map_tree.lock);
2361         ret = add_extent_mapping(&map_tree->map_tree, em);
2362         spin_unlock(&map_tree->map_tree.lock);
2363         BUG_ON(ret);
2364         free_extent_map(em);
2365
2366         return 0;
2367 }
2368
2369 static int fill_device_from_item(struct extent_buffer *leaf,
2370                                  struct btrfs_dev_item *dev_item,
2371                                  struct btrfs_device *device)
2372 {
2373         unsigned long ptr;
2374
2375         device->devid = btrfs_device_id(leaf, dev_item);
2376         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2377         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2378         device->type = btrfs_device_type(leaf, dev_item);
2379         device->io_align = btrfs_device_io_align(leaf, dev_item);
2380         device->io_width = btrfs_device_io_width(leaf, dev_item);
2381         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2382
2383         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2384         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2385
2386         return 0;
2387 }
2388
2389 static int read_one_dev(struct btrfs_root *root,
2390                         struct extent_buffer *leaf,
2391                         struct btrfs_dev_item *dev_item)
2392 {
2393         struct btrfs_device *device;
2394         u64 devid;
2395         int ret;
2396         u8 dev_uuid[BTRFS_UUID_SIZE];
2397
2398         devid = btrfs_device_id(leaf, dev_item);
2399         read_extent_buffer(leaf, dev_uuid,
2400                            (unsigned long)btrfs_device_uuid(dev_item),
2401                            BTRFS_UUID_SIZE);
2402         device = btrfs_find_device(root, devid, dev_uuid);
2403         if (!device) {
2404                 printk("warning devid %Lu missing\n", devid);
2405                 device = add_missing_dev(root, devid, dev_uuid);
2406                 if (!device)
2407                         return -ENOMEM;
2408         }
2409
2410         fill_device_from_item(leaf, dev_item, device);
2411         device->dev_root = root->fs_info->dev_root;
2412         device->in_fs_metadata = 1;
2413         ret = 0;
2414 #if 0
2415         ret = btrfs_open_device(device);
2416         if (ret) {
2417                 kfree(device);
2418         }
2419 #endif
2420         return ret;
2421 }
2422
2423 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2424 {
2425         struct btrfs_dev_item *dev_item;
2426
2427         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2428                                                      dev_item);
2429         return read_one_dev(root, buf, dev_item);
2430 }
2431
2432 int btrfs_read_sys_array(struct btrfs_root *root)
2433 {
2434         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2435         struct extent_buffer *sb;
2436         struct btrfs_disk_key *disk_key;
2437         struct btrfs_chunk *chunk;
2438         u8 *ptr;
2439         unsigned long sb_ptr;
2440         int ret = 0;
2441         u32 num_stripes;
2442         u32 array_size;
2443         u32 len = 0;
2444         u32 cur;
2445         struct btrfs_key key;
2446
2447         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2448                                           BTRFS_SUPER_INFO_SIZE);
2449         if (!sb)
2450                 return -ENOMEM;
2451         btrfs_set_buffer_uptodate(sb);
2452         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2453         array_size = btrfs_super_sys_array_size(super_copy);
2454
2455         ptr = super_copy->sys_chunk_array;
2456         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2457         cur = 0;
2458
2459         while (cur < array_size) {
2460                 disk_key = (struct btrfs_disk_key *)ptr;
2461                 btrfs_disk_key_to_cpu(&key, disk_key);
2462
2463                 len = sizeof(*disk_key); ptr += len;
2464                 sb_ptr += len;
2465                 cur += len;
2466
2467                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2468                         chunk = (struct btrfs_chunk *)sb_ptr;
2469                         ret = read_one_chunk(root, &key, sb, chunk);
2470                         if (ret)
2471                                 break;
2472                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2473                         len = btrfs_chunk_item_size(num_stripes);
2474                 } else {
2475                         ret = -EIO;
2476                         break;
2477                 }
2478                 ptr += len;
2479                 sb_ptr += len;
2480                 cur += len;
2481         }
2482         free_extent_buffer(sb);
2483         return ret;
2484 }
2485
2486 int btrfs_read_chunk_tree(struct btrfs_root *root)
2487 {
2488         struct btrfs_path *path;
2489         struct extent_buffer *leaf;
2490         struct btrfs_key key;
2491         struct btrfs_key found_key;
2492         int ret;
2493         int slot;
2494
2495         root = root->fs_info->chunk_root;
2496
2497         path = btrfs_alloc_path();
2498         if (!path)
2499                 return -ENOMEM;
2500
2501         /* first we search for all of the device items, and then we
2502          * read in all of the chunk items.  This way we can create chunk
2503          * mappings that reference all of the devices that are afound
2504          */
2505         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2506         key.offset = 0;
2507         key.type = 0;
2508 again:
2509         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2510         while(1) {
2511                 leaf = path->nodes[0];
2512                 slot = path->slots[0];
2513                 if (slot >= btrfs_header_nritems(leaf)) {
2514                         ret = btrfs_next_leaf(root, path);
2515                         if (ret == 0)
2516                                 continue;
2517                         if (ret < 0)
2518                                 goto error;
2519                         break;
2520                 }
2521                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2522                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2523                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2524                                 break;
2525                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2526                                 struct btrfs_dev_item *dev_item;
2527                                 dev_item = btrfs_item_ptr(leaf, slot,
2528                                                   struct btrfs_dev_item);
2529                                 ret = read_one_dev(root, leaf, dev_item);
2530                                 BUG_ON(ret);
2531                         }
2532                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2533                         struct btrfs_chunk *chunk;
2534                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2535                         ret = read_one_chunk(root, &found_key, leaf, chunk);
2536                 }
2537                 path->slots[0]++;
2538         }
2539         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2540                 key.objectid = 0;
2541                 btrfs_release_path(root, path);
2542                 goto again;
2543         }
2544
2545         btrfs_free_path(path);
2546         ret = 0;
2547 error:
2548         return ret;
2549 }