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->alloc_mutex);
  68         mutex_unlock(&root->fs_info->chunk_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         mutex_lock(&root->fs_info->volume_mutex);
1042
1043         trans = btrfs_start_transaction(root, 1);
1044         lock_chunks(root);
1045         devices = &root->fs_info->fs_devices->devices;
1046         list_for_each(cur, devices) {
1047                 device = list_entry(cur, struct btrfs_device, dev_list);
1048                 if (device->bdev == bdev) {
1049                         ret = -EEXIST;
1050                         goto out;
1051                 }
1052         }
1053
1054         device = kzalloc(sizeof(*device), GFP_NOFS);
1055         if (!device) {
1056                 /* we can safely leave the fs_devices entry around */
1057                 ret = -ENOMEM;
1058                 goto out_close_bdev;
1059         }
1060
1061         device->barriers = 1;
1062         device->work.func = pending_bios_fn;
1063         generate_random_uuid(device->uuid);
1064         spin_lock_init(&device->io_lock);
1065         device->name = kstrdup(device_path, GFP_NOFS);
1066         if (!device->name) {
1067                 kfree(device);
1068                 goto out_close_bdev;
1069         }
1070         device->io_width = root->sectorsize;
1071         device->io_align = root->sectorsize;
1072         device->sector_size = root->sectorsize;
1073         device->total_bytes = i_size_read(bdev->bd_inode);
1074         device->dev_root = root->fs_info->dev_root;
1075         device->bdev = bdev;
1076         device->in_fs_metadata = 1;
1077
1078         ret = btrfs_add_device(trans, root, device);
1079         if (ret)
1080                 goto out_close_bdev;
1081
1082         set_blocksize(device->bdev, 4096);
1083
1084         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1085         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1086                                     total_bytes + device->total_bytes);
1087
1088         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1089         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1090                                     total_bytes + 1);
1091
1092         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1093         list_add(&device->dev_alloc_list,
1094                  &root->fs_info->fs_devices->alloc_list);
1095         root->fs_info->fs_devices->num_devices++;
1096         root->fs_info->fs_devices->open_devices++;
1097 out:
1098         unlock_chunks(root);
1099         btrfs_end_transaction(trans, root);
1100         mutex_unlock(&root->fs_info->volume_mutex);
1101
1102         return ret;
1103
1104 out_close_bdev:
1105         close_bdev_excl(bdev);
1106         goto out;
1107 }
1108
1109 int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1110                                  struct btrfs_device *device)
1111 {
1112         int ret;
1113         struct btrfs_path *path;
1114         struct btrfs_root *root;
1115         struct btrfs_dev_item *dev_item;
1116         struct extent_buffer *leaf;
1117         struct btrfs_key key;
1118
1119         root = device->dev_root->fs_info->chunk_root;
1120
1121         path = btrfs_alloc_path();
1122         if (!path)
1123                 return -ENOMEM;
1124
1125         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1126         key.type = BTRFS_DEV_ITEM_KEY;
1127         key.offset = device->devid;
1128
1129         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1130         if (ret < 0)
1131                 goto out;
1132
1133         if (ret > 0) {
1134                 ret = -ENOENT;
1135                 goto out;
1136         }
1137
1138         leaf = path->nodes[0];
1139         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1140
1141         btrfs_set_device_id(leaf, dev_item, device->devid);
1142         btrfs_set_device_type(leaf, dev_item, device->type);
1143         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1144         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1145         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1146         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1147         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1148         btrfs_mark_buffer_dirty(leaf);
1149
1150 out:
1151         btrfs_free_path(path);
1152         return ret;
1153 }
1154
1155 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1156                       struct btrfs_device *device, u64 new_size)
1157 {
1158         struct btrfs_super_block *super_copy =
1159                 &device->dev_root->fs_info->super_copy;
1160         u64 old_total = btrfs_super_total_bytes(super_copy);
1161         u64 diff = new_size - device->total_bytes;
1162
1163         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1164         return btrfs_update_device(trans, device);
1165 }
1166
1167 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1168                       struct btrfs_device *device, u64 new_size)
1169 {
1170         int ret;
1171         lock_chunks(device->dev_root);
1172         ret = __btrfs_grow_device(trans, device, new_size);
1173         unlock_chunks(device->dev_root);
1174         return ret;
1175 }
1176
1177 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1178                             struct btrfs_root *root,
1179                             u64 chunk_tree, u64 chunk_objectid,
1180                             u64 chunk_offset)
1181 {
1182         int ret;
1183         struct btrfs_path *path;
1184         struct btrfs_key key;
1185
1186         root = root->fs_info->chunk_root;
1187         path = btrfs_alloc_path();
1188         if (!path)
1189                 return -ENOMEM;
1190
1191         key.objectid = chunk_objectid;
1192         key.offset = chunk_offset;
1193         key.type = BTRFS_CHUNK_ITEM_KEY;
1194
1195         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1196         BUG_ON(ret);
1197
1198         ret = btrfs_del_item(trans, root, path);
1199         BUG_ON(ret);
1200
1201         btrfs_free_path(path);
1202         return 0;
1203 }
1204
1205 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1206                         chunk_offset)
1207 {
1208         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1209         struct btrfs_disk_key *disk_key;
1210         struct btrfs_chunk *chunk;
1211         u8 *ptr;
1212         int ret = 0;
1213         u32 num_stripes;
1214         u32 array_size;
1215         u32 len = 0;
1216         u32 cur;
1217         struct btrfs_key key;
1218
1219         array_size = btrfs_super_sys_array_size(super_copy);
1220
1221         ptr = super_copy->sys_chunk_array;
1222         cur = 0;
1223
1224         while (cur < array_size) {
1225                 disk_key = (struct btrfs_disk_key *)ptr;
1226                 btrfs_disk_key_to_cpu(&key, disk_key);
1227
1228                 len = sizeof(*disk_key);
1229
1230                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1231                         chunk = (struct btrfs_chunk *)(ptr + len);
1232                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1233                         len += btrfs_chunk_item_size(num_stripes);
1234                 } else {
1235                         ret = -EIO;
1236                         break;
1237                 }
1238                 if (key.objectid == chunk_objectid &&
1239                     key.offset == chunk_offset) {
1240                         memmove(ptr, ptr + len, array_size - (cur + len));
1241                         array_size -= len;
1242                         btrfs_set_super_sys_array_size(super_copy, array_size);
1243                 } else {
1244                         ptr += len;
1245                         cur += len;
1246                 }
1247         }
1248         return ret;
1249 }
1250
1251
1252 int btrfs_relocate_chunk(struct btrfs_root *root,
1253                          u64 chunk_tree, u64 chunk_objectid,
1254                          u64 chunk_offset)
1255 {
1256         struct extent_map_tree *em_tree;
1257         struct btrfs_root *extent_root;
1258         struct btrfs_trans_handle *trans;
1259         struct extent_map *em;
1260         struct map_lookup *map;
1261         int ret;
1262         int i;
1263
1264         printk("btrfs relocating chunk %llu\n",
1265                (unsigned long long)chunk_offset);
1266         root = root->fs_info->chunk_root;
1267         extent_root = root->fs_info->extent_root;
1268         em_tree = &root->fs_info->mapping_tree.map_tree;
1269
1270         /* step one, relocate all the extents inside this chunk */
1271         ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1272         BUG_ON(ret);
1273
1274         trans = btrfs_start_transaction(root, 1);
1275         BUG_ON(!trans);
1276
1277         lock_chunks(root);
1278
1279         /*
1280          * step two, delete the device extents and the
1281          * chunk tree entries
1282          */
1283         spin_lock(&em_tree->lock);
1284         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1285         spin_unlock(&em_tree->lock);
1286
1287         BUG_ON(em->start > chunk_offset ||
1288                em->start + em->len < chunk_offset);
1289         map = (struct map_lookup *)em->bdev;
1290
1291         for (i = 0; i < map->num_stripes; i++) {
1292                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1293                                             map->stripes[i].physical);
1294                 BUG_ON(ret);
1295
1296                 if (map->stripes[i].dev) {
1297                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1298                         BUG_ON(ret);
1299                 }
1300         }
1301         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1302                                chunk_offset);
1303
1304         BUG_ON(ret);
1305
1306         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1307                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1308                 BUG_ON(ret);
1309         }
1310
1311         spin_lock(&em_tree->lock);
1312         remove_extent_mapping(em_tree, em);
1313         kfree(map);
1314         em->bdev = NULL;
1315
1316         /* once for the tree */
1317         free_extent_map(em);
1318         spin_unlock(&em_tree->lock);
1319
1320         /* once for us */
1321         free_extent_map(em);
1322
1323         unlock_chunks(root);
1324         btrfs_end_transaction(trans, root);
1325         return 0;
1326 }
1327
1328 static u64 div_factor(u64 num, int factor)
1329 {
1330         if (factor == 10)
1331                 return num;
1332         num *= factor;
1333         do_div(num, 10);
1334         return num;
1335 }
1336
1337
1338 int btrfs_balance(struct btrfs_root *dev_root)
1339 {
1340         int ret;
1341         struct list_head *cur;
1342         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1343         struct btrfs_device *device;
1344         u64 old_size;
1345         u64 size_to_free;
1346         struct btrfs_path *path;
1347         struct btrfs_key key;
1348         struct btrfs_chunk *chunk;
1349         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1350         struct btrfs_trans_handle *trans;
1351         struct btrfs_key found_key;
1352
1353
1354         mutex_lock(&dev_root->fs_info->volume_mutex);
1355         dev_root = dev_root->fs_info->dev_root;
1356
1357         /* step one make some room on all the devices */
1358         list_for_each(cur, devices) {
1359                 device = list_entry(cur, struct btrfs_device, dev_list);
1360                 old_size = device->total_bytes;
1361                 size_to_free = div_factor(old_size, 1);
1362                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1363                 if (device->total_bytes - device->bytes_used > size_to_free)
1364                         continue;
1365
1366                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1367                 BUG_ON(ret);
1368
1369                 trans = btrfs_start_transaction(dev_root, 1);
1370                 BUG_ON(!trans);
1371
1372                 ret = btrfs_grow_device(trans, device, old_size);
1373                 BUG_ON(ret);
1374
1375                 btrfs_end_transaction(trans, dev_root);
1376         }
1377
1378         /* step two, relocate all the chunks */
1379         path = btrfs_alloc_path();
1380         BUG_ON(!path);
1381
1382         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1383         key.offset = (u64)-1;
1384         key.type = BTRFS_CHUNK_ITEM_KEY;
1385
1386         while(1) {
1387                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1388                 if (ret < 0)
1389                         goto error;
1390
1391                 /*
1392                  * this shouldn't happen, it means the last relocate
1393                  * failed
1394                  */
1395                 if (ret == 0)
1396                         break;
1397
1398                 ret = btrfs_previous_item(chunk_root, path, 0,
1399                                           BTRFS_CHUNK_ITEM_KEY);
1400                 if (ret)
1401                         break;
1402
1403                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1404                                       path->slots[0]);
1405                 if (found_key.objectid != key.objectid)
1406                         break;
1407
1408                 chunk = btrfs_item_ptr(path->nodes[0],
1409                                        path->slots[0],
1410                                        struct btrfs_chunk);
1411                 key.offset = found_key.offset;
1412                 /* chunk zero is special */
1413                 if (key.offset == 0)
1414                         break;
1415
1416                 btrfs_release_path(chunk_root, path);
1417                 ret = btrfs_relocate_chunk(chunk_root,
1418                                            chunk_root->root_key.objectid,
1419                                            found_key.objectid,
1420                                            found_key.offset);
1421                 BUG_ON(ret);
1422         }
1423         ret = 0;
1424 error:
1425         btrfs_free_path(path);
1426         mutex_unlock(&dev_root->fs_info->volume_mutex);
1427         return ret;
1428 }
1429
1430 /*
1431  * shrinking a device means finding all of the device extents past
1432  * the new size, and then following the back refs to the chunks.
1433  * The chunk relocation code actually frees the device extent
1434  */
1435 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1436 {
1437         struct btrfs_trans_handle *trans;
1438         struct btrfs_root *root = device->dev_root;
1439         struct btrfs_dev_extent *dev_extent = NULL;
1440         struct btrfs_path *path;
1441         u64 length;
1442         u64 chunk_tree;
1443         u64 chunk_objectid;
1444         u64 chunk_offset;
1445         int ret;
1446         int slot;
1447         struct extent_buffer *l;
1448         struct btrfs_key key;
1449         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1450         u64 old_total = btrfs_super_total_bytes(super_copy);
1451         u64 diff = device->total_bytes - new_size;
1452
1453
1454         path = btrfs_alloc_path();
1455         if (!path)
1456                 return -ENOMEM;
1457
1458         trans = btrfs_start_transaction(root, 1);
1459         if (!trans) {
1460                 ret = -ENOMEM;
1461                 goto done;
1462         }
1463
1464         path->reada = 2;
1465
1466         lock_chunks(root);
1467
1468         device->total_bytes = new_size;
1469         ret = btrfs_update_device(trans, device);
1470         if (ret) {
1471                 unlock_chunks(root);
1472                 btrfs_end_transaction(trans, root);
1473                 goto done;
1474         }
1475         WARN_ON(diff > old_total);
1476         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1477         unlock_chunks(root);
1478         btrfs_end_transaction(trans, root);
1479
1480         key.objectid = device->devid;
1481         key.offset = (u64)-1;
1482         key.type = BTRFS_DEV_EXTENT_KEY;
1483
1484         while (1) {
1485                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1486                 if (ret < 0)
1487                         goto done;
1488
1489                 ret = btrfs_previous_item(root, path, 0, key.type);
1490                 if (ret < 0)
1491                         goto done;
1492                 if (ret) {
1493                         ret = 0;
1494                         goto done;
1495                 }
1496
1497                 l = path->nodes[0];
1498                 slot = path->slots[0];
1499                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1500
1501                 if (key.objectid != device->devid)
1502                         goto done;
1503
1504                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1505                 length = btrfs_dev_extent_length(l, dev_extent);
1506
1507                 if (key.offset + length <= new_size)
1508                         goto done;
1509
1510                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1511                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1512                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1513                 btrfs_release_path(root, path);
1514
1515                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1516                                            chunk_offset);
1517                 if (ret)
1518                         goto done;
1519         }
1520
1521 done:
1522         btrfs_free_path(path);
1523         return ret;
1524 }
1525
1526 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1527                            struct btrfs_root *root,
1528                            struct btrfs_key *key,
1529                            struct btrfs_chunk *chunk, int item_size)
1530 {
1531         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1532         struct btrfs_disk_key disk_key;
1533         u32 array_size;
1534         u8 *ptr;
1535
1536         array_size = btrfs_super_sys_array_size(super_copy);
1537         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1538                 return -EFBIG;
1539
1540         ptr = super_copy->sys_chunk_array + array_size;
1541         btrfs_cpu_key_to_disk(&disk_key, key);
1542         memcpy(ptr, &disk_key, sizeof(disk_key));
1543         ptr += sizeof(disk_key);
1544         memcpy(ptr, chunk, item_size);
1545         item_size += sizeof(disk_key);
1546         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1547         return 0;
1548 }
1549
1550 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1551                                         int num_stripes, int sub_stripes)
1552 {
1553         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1554                 return calc_size;
1555         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1556                 return calc_size * (num_stripes / sub_stripes);
1557         else
1558                 return calc_size * num_stripes;
1559 }
1560
1561
1562 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1563                       struct btrfs_root *extent_root, u64 *start,
1564                       u64 *num_bytes, u64 type)
1565 {
1566         u64 dev_offset;
1567         struct btrfs_fs_info *info = extent_root->fs_info;
1568         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1569         struct btrfs_path *path;
1570         struct btrfs_stripe *stripes;
1571         struct btrfs_device *device = NULL;
1572         struct btrfs_chunk *chunk;
1573         struct list_head private_devs;
1574         struct list_head *dev_list;
1575         struct list_head *cur;
1576         struct extent_map_tree *em_tree;
1577         struct map_lookup *map;
1578         struct extent_map *em;
1579         int min_stripe_size = 1 * 1024 * 1024;
1580         u64 physical;
1581         u64 calc_size = 1024 * 1024 * 1024;
1582         u64 max_chunk_size = calc_size;
1583         u64 min_free;
1584         u64 avail;
1585         u64 max_avail = 0;
1586         u64 percent_max;
1587         int num_stripes = 1;
1588         int min_stripes = 1;
1589         int sub_stripes = 0;
1590         int looped = 0;
1591         int ret;
1592         int index;
1593         int stripe_len = 64 * 1024;
1594         struct btrfs_key key;
1595
1596         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1597             (type & BTRFS_BLOCK_GROUP_DUP)) {
1598                 WARN_ON(1);
1599                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1600         }
1601         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1602         if (list_empty(dev_list))
1603                 return -ENOSPC;
1604
1605         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1606                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1607                 min_stripes = 2;
1608         }
1609         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1610                 num_stripes = 2;
1611                 min_stripes = 2;
1612         }
1613         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1614                 num_stripes = min_t(u64, 2,
1615                             extent_root->fs_info->fs_devices->open_devices);
1616                 if (num_stripes < 2)
1617                         return -ENOSPC;
1618                 min_stripes = 2;
1619         }
1620         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1621                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1622                 if (num_stripes < 4)
1623                         return -ENOSPC;
1624                 num_stripes &= ~(u32)1;
1625                 sub_stripes = 2;
1626                 min_stripes = 4;
1627         }
1628
1629         if (type & BTRFS_BLOCK_GROUP_DATA) {
1630                 max_chunk_size = 10 * calc_size;
1631                 min_stripe_size = 64 * 1024 * 1024;
1632         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1633                 max_chunk_size = 4 * calc_size;
1634                 min_stripe_size = 32 * 1024 * 1024;
1635         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1636                 calc_size = 8 * 1024 * 1024;
1637                 max_chunk_size = calc_size * 2;
1638                 min_stripe_size = 1 * 1024 * 1024;
1639         }
1640
1641         path = btrfs_alloc_path();
1642         if (!path)
1643                 return -ENOMEM;
1644
1645         /* we don't want a chunk larger than 10% of the FS */
1646         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1647         max_chunk_size = min(percent_max, max_chunk_size);
1648
1649 again:
1650         if (calc_size * num_stripes > max_chunk_size) {
1651                 calc_size = max_chunk_size;
1652                 do_div(calc_size, num_stripes);
1653                 do_div(calc_size, stripe_len);
1654                 calc_size *= stripe_len;
1655         }
1656         /* we don't want tiny stripes */
1657         calc_size = max_t(u64, min_stripe_size, calc_size);
1658
1659         do_div(calc_size, stripe_len);
1660         calc_size *= stripe_len;
1661
1662         INIT_LIST_HEAD(&private_devs);
1663         cur = dev_list->next;
1664         index = 0;
1665
1666         if (type & BTRFS_BLOCK_GROUP_DUP)
1667                 min_free = calc_size * 2;
1668         else
1669                 min_free = calc_size;
1670
1671         /* we add 1MB because we never use the first 1MB of the device */
1672         min_free += 1024 * 1024;
1673
1674         /* build a private list of devices we will allocate from */
1675         while(index < num_stripes) {
1676                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1677
1678                 if (device->total_bytes > device->bytes_used)
1679                         avail = device->total_bytes - device->bytes_used;
1680                 else
1681                         avail = 0;
1682                 cur = cur->next;
1683
1684                 if (device->in_fs_metadata && avail >= min_free) {
1685                         u64 ignored_start = 0;
1686                         ret = find_free_dev_extent(trans, device, path,
1687                                                    min_free,
1688                                                    &ignored_start);
1689                         if (ret == 0) {
1690                                 list_move_tail(&device->dev_alloc_list,
1691                                                &private_devs);
1692                                 index++;
1693                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1694                                         index++;
1695                         }
1696                 } else if (device->in_fs_metadata && avail > max_avail)
1697                         max_avail = avail;
1698                 if (cur == dev_list)
1699                         break;
1700         }
1701         if (index < num_stripes) {
1702                 list_splice(&private_devs, dev_list);
1703                 if (index >= min_stripes) {
1704                         num_stripes = index;
1705                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1706                                 num_stripes /= sub_stripes;
1707                                 num_stripes *= sub_stripes;
1708                         }
1709                         looped = 1;
1710                         goto again;
1711                 }
1712                 if (!looped && max_avail > 0) {
1713                         looped = 1;
1714                         calc_size = max_avail;
1715                         goto again;
1716                 }
1717                 btrfs_free_path(path);
1718                 return -ENOSPC;
1719         }
1720         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1721         key.type = BTRFS_CHUNK_ITEM_KEY;
1722         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1723                               &key.offset);
1724         if (ret) {
1725                 btrfs_free_path(path);
1726                 return ret;
1727         }
1728
1729         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1730         if (!chunk) {
1731                 btrfs_free_path(path);
1732                 return -ENOMEM;
1733         }
1734
1735         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1736         if (!map) {
1737                 kfree(chunk);
1738                 btrfs_free_path(path);
1739                 return -ENOMEM;
1740         }
1741         btrfs_free_path(path);
1742         path = NULL;
1743
1744         stripes = &chunk->stripe;
1745         *num_bytes = chunk_bytes_by_type(type, calc_size,
1746                                          num_stripes, sub_stripes);
1747
1748         index = 0;
1749         while(index < num_stripes) {
1750                 struct btrfs_stripe *stripe;
1751                 BUG_ON(list_empty(&private_devs));
1752                 cur = private_devs.next;
1753                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1754
1755                 /* loop over this device again if we're doing a dup group */
1756                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1757                     (index == num_stripes - 1))
1758                         list_move_tail(&device->dev_alloc_list, dev_list);
1759
1760                 ret = btrfs_alloc_dev_extent(trans, device,
1761                              info->chunk_root->root_key.objectid,
1762                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1763                              calc_size, &dev_offset);
1764                 BUG_ON(ret);
1765                 device->bytes_used += calc_size;
1766                 ret = btrfs_update_device(trans, device);
1767                 BUG_ON(ret);
1768
1769                 map->stripes[index].dev = device;
1770                 map->stripes[index].physical = dev_offset;
1771                 stripe = stripes + index;
1772                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1773                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1774                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1775                 physical = dev_offset;
1776                 index++;
1777         }
1778         BUG_ON(!list_empty(&private_devs));
1779
1780         /* key was set above */
1781         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1782         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1783         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1784         btrfs_set_stack_chunk_type(chunk, type);
1785         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1786         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1787         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1788         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1789         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1790         map->sector_size = extent_root->sectorsize;
1791         map->stripe_len = stripe_len;
1792         map->io_align = stripe_len;
1793         map->io_width = stripe_len;
1794         map->type = type;
1795         map->num_stripes = num_stripes;
1796         map->sub_stripes = sub_stripes;
1797
1798         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1799                                 btrfs_chunk_item_size(num_stripes));
1800         BUG_ON(ret);
1801         *start = key.offset;;
1802
1803         em = alloc_extent_map(GFP_NOFS);
1804         if (!em)
1805                 return -ENOMEM;
1806         em->bdev = (struct block_device *)map;
1807         em->start = key.offset;
1808         em->len = *num_bytes;
1809         em->block_start = 0;
1810
1811         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1812                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1813                                     chunk, btrfs_chunk_item_size(num_stripes));
1814                 BUG_ON(ret);
1815         }
1816         kfree(chunk);
1817
1818         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1819         spin_lock(&em_tree->lock);
1820         ret = add_extent_mapping(em_tree, em);
1821         spin_unlock(&em_tree->lock);
1822         BUG_ON(ret);
1823         free_extent_map(em);
1824         return ret;
1825 }
1826
1827 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1828 {
1829         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1830 }
1831
1832 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1833 {
1834         struct extent_map *em;
1835
1836         while(1) {
1837                 spin_lock(&tree->map_tree.lock);
1838                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1839                 if (em)
1840                         remove_extent_mapping(&tree->map_tree, em);
1841                 spin_unlock(&tree->map_tree.lock);
1842                 if (!em)
1843                         break;
1844                 kfree(em->bdev);
1845                 /* once for us */
1846                 free_extent_map(em);
1847                 /* once for the tree */
1848                 free_extent_map(em);
1849         }
1850 }
1851
1852 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1853 {
1854         struct extent_map *em;
1855         struct map_lookup *map;
1856         struct extent_map_tree *em_tree = &map_tree->map_tree;
1857         int ret;
1858
1859         spin_lock(&em_tree->lock);
1860         em = lookup_extent_mapping(em_tree, logical, len);
1861         spin_unlock(&em_tree->lock);
1862         BUG_ON(!em);
1863
1864         BUG_ON(em->start > logical || em->start + em->len < logical);
1865         map = (struct map_lookup *)em->bdev;
1866         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1867                 ret = map->num_stripes;
1868         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1869                 ret = map->sub_stripes;
1870         else
1871                 ret = 1;
1872         free_extent_map(em);
1873         return ret;
1874 }
1875
1876 static int find_live_mirror(struct map_lookup *map, int first, int num,
1877                             int optimal)
1878 {
1879         int i;
1880         if (map->stripes[optimal].dev->bdev)
1881                 return optimal;
1882         for (i = first; i < first + num; i++) {
1883                 if (map->stripes[i].dev->bdev)
1884                         return i;
1885         }
1886         /* we couldn't find one that doesn't fail.  Just return something
1887          * and the io error handling code will clean up eventually
1888          */
1889         return optimal;
1890 }
1891
1892 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1893                              u64 logical, u64 *length,
1894                              struct btrfs_multi_bio **multi_ret,
1895                              int mirror_num, struct page *unplug_page)
1896 {
1897         struct extent_map *em;
1898         struct map_lookup *map;
1899         struct extent_map_tree *em_tree = &map_tree->map_tree;
1900         u64 offset;
1901         u64 stripe_offset;
1902         u64 stripe_nr;
1903         int stripes_allocated = 8;
1904         int stripes_required = 1;
1905         int stripe_index;
1906         int i;
1907         int num_stripes;
1908         int max_errors = 0;
1909         struct btrfs_multi_bio *multi = NULL;
1910
1911         if (multi_ret && !(rw & (1 << BIO_RW))) {
1912                 stripes_allocated = 1;
1913         }
1914 again:
1915         if (multi_ret) {
1916                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1917                                 GFP_NOFS);
1918                 if (!multi)
1919                         return -ENOMEM;
1920
1921                 atomic_set(&multi->error, 0);
1922         }
1923
1924         spin_lock(&em_tree->lock);
1925         em = lookup_extent_mapping(em_tree, logical, *length);
1926         spin_unlock(&em_tree->lock);
1927
1928         if (!em && unplug_page)
1929                 return 0;
1930
1931         if (!em) {
1932                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1933                 BUG();
1934         }
1935
1936         BUG_ON(em->start > logical || em->start + em->len < logical);
1937         map = (struct map_lookup *)em->bdev;
1938         offset = logical - em->start;
1939
1940         if (mirror_num > map->num_stripes)
1941                 mirror_num = 0;
1942
1943         /* if our multi bio struct is too small, back off and try again */
1944         if (rw & (1 << BIO_RW)) {
1945                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1946                                  BTRFS_BLOCK_GROUP_DUP)) {
1947                         stripes_required = map->num_stripes;
1948                         max_errors = 1;
1949                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1950                         stripes_required = map->sub_stripes;
1951                         max_errors = 1;
1952                 }
1953         }
1954         if (multi_ret && rw == WRITE &&
1955             stripes_allocated < stripes_required) {
1956                 stripes_allocated = map->num_stripes;
1957                 free_extent_map(em);
1958                 kfree(multi);
1959                 goto again;
1960         }
1961         stripe_nr = offset;
1962         /*
1963          * stripe_nr counts the total number of stripes we have to stride
1964          * to get to this block
1965          */
1966         do_div(stripe_nr, map->stripe_len);
1967
1968         stripe_offset = stripe_nr * map->stripe_len;
1969         BUG_ON(offset < stripe_offset);
1970
1971         /* stripe_offset is the offset of this block in its stripe*/
1972         stripe_offset = offset - stripe_offset;
1973
1974         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1975                          BTRFS_BLOCK_GROUP_RAID10 |
1976                          BTRFS_BLOCK_GROUP_DUP)) {
1977                 /* we limit the length of each bio to what fits in a stripe */
1978                 *length = min_t(u64, em->len - offset,
1979                               map->stripe_len - stripe_offset);
1980         } else {
1981                 *length = em->len - offset;
1982         }
1983
1984         if (!multi_ret && !unplug_page)
1985                 goto out;
1986
1987         num_stripes = 1;
1988         stripe_index = 0;
1989         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1990                 if (unplug_page || (rw & (1 << BIO_RW)))
1991                         num_stripes = map->num_stripes;
1992                 else if (mirror_num)
1993                         stripe_index = mirror_num - 1;
1994                 else {
1995                         stripe_index = find_live_mirror(map, 0,
1996                                             map->num_stripes,
1997                                             current->pid % map->num_stripes);
1998                 }
1999
2000         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2001                 if (rw & (1 << BIO_RW))
2002                         num_stripes = map->num_stripes;
2003                 else if (mirror_num)
2004                         stripe_index = mirror_num - 1;
2005
2006         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2007                 int factor = map->num_stripes / map->sub_stripes;
2008
2009                 stripe_index = do_div(stripe_nr, factor);
2010                 stripe_index *= map->sub_stripes;
2011
2012                 if (unplug_page || (rw & (1 << BIO_RW)))
2013                         num_stripes = map->sub_stripes;
2014                 else if (mirror_num)
2015                         stripe_index += mirror_num - 1;
2016                 else {
2017                         stripe_index = find_live_mirror(map, stripe_index,
2018                                               map->sub_stripes, stripe_index +
2019                                               current->pid % map->sub_stripes);
2020                 }
2021         } else {
2022                 /*
2023                  * after this do_div call, stripe_nr is the number of stripes
2024                  * on this device we have to walk to find the data, and
2025                  * stripe_index is the number of our device in the stripe array
2026                  */
2027                 stripe_index = do_div(stripe_nr, map->num_stripes);
2028         }
2029         BUG_ON(stripe_index >= map->num_stripes);
2030
2031         for (i = 0; i < num_stripes; i++) {
2032                 if (unplug_page) {
2033                         struct btrfs_device *device;
2034                         struct backing_dev_info *bdi;
2035
2036                         device = map->stripes[stripe_index].dev;
2037                         if (device->bdev) {
2038                                 bdi = blk_get_backing_dev_info(device->bdev);
2039                                 if (bdi->unplug_io_fn) {
2040                                         bdi->unplug_io_fn(bdi, unplug_page);
2041                                 }
2042                         }
2043                 } else {
2044                         multi->stripes[i].physical =
2045                                 map->stripes[stripe_index].physical +
2046                                 stripe_offset + stripe_nr * map->stripe_len;
2047                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2048                 }
2049                 stripe_index++;
2050         }
2051         if (multi_ret) {
2052                 *multi_ret = multi;
2053                 multi->num_stripes = num_stripes;
2054                 multi->max_errors = max_errors;
2055         }
2056 out:
2057         free_extent_map(em);
2058         return 0;
2059 }
2060
2061 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2062                       u64 logical, u64 *length,
2063                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2064 {
2065         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2066                                  mirror_num, NULL);
2067 }
2068
2069 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2070                       u64 logical, struct page *page)
2071 {
2072         u64 length = PAGE_CACHE_SIZE;
2073         return __btrfs_map_block(map_tree, READ, logical, &length,
2074                                  NULL, 0, page);
2075 }
2076
2077
2078 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2079 static void end_bio_multi_stripe(struct bio *bio, int err)
2080 #else
2081 static int end_bio_multi_stripe(struct bio *bio,
2082                                    unsigned int bytes_done, int err)
2083 #endif
2084 {
2085         struct btrfs_multi_bio *multi = bio->bi_private;
2086         int is_orig_bio = 0;
2087
2088 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2089         if (bio->bi_size)
2090                 return 1;
2091 #endif
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 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2121                 bio_endio(bio, bio->bi_size, err);
2122 #else
2123                 bio_endio(bio, err);
2124 #endif
2125         } else if (!is_orig_bio) {
2126                 bio_put(bio);
2127         }
2128 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2129         return 0;
2130 #endif
2131 }
2132
2133 struct async_sched {
2134         struct bio *bio;
2135         int rw;
2136         struct btrfs_fs_info *info;
2137         struct btrfs_work work;
2138 };
2139
2140 /*
2141  * see run_scheduled_bios for a description of why bios are collected for
2142  * async submit.
2143  *
2144  * This will add one bio to the pending list for a device and make sure
2145  * the work struct is scheduled.
2146  */
2147 static int noinline schedule_bio(struct btrfs_root *root,
2148                                  struct btrfs_device *device,
2149                                  int rw, struct bio *bio)
2150 {
2151         int should_queue = 1;
2152
2153         /* don't bother with additional async steps for reads, right now */
2154         if (!(rw & (1 << BIO_RW))) {
2155                 bio_get(bio);
2156                 submit_bio(rw, bio);
2157                 bio_put(bio);
2158                 return 0;
2159         }
2160
2161         /*
2162          * nr_async_bios allows us to reliably return congestion to the
2163          * higher layers.  Otherwise, the async bio makes it appear we have
2164          * made progress against dirty pages when we've really just put it
2165          * on a queue for later
2166          */
2167         atomic_inc(&root->fs_info->nr_async_bios);
2168         WARN_ON(bio->bi_next);
2169         bio->bi_next = NULL;
2170         bio->bi_rw |= rw;
2171
2172         spin_lock(&device->io_lock);
2173
2174         if (device->pending_bio_tail)
2175                 device->pending_bio_tail->bi_next = bio;
2176
2177         device->pending_bio_tail = bio;
2178         if (!device->pending_bios)
2179                 device->pending_bios = bio;
2180         if (device->running_pending)
2181                 should_queue = 0;
2182
2183         spin_unlock(&device->io_lock);
2184
2185         if (should_queue)
2186                 btrfs_queue_worker(&root->fs_info->submit_workers,
2187                                    &device->work);
2188         return 0;
2189 }
2190
2191 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2192                   int mirror_num, int async_submit)
2193 {
2194         struct btrfs_mapping_tree *map_tree;
2195         struct btrfs_device *dev;
2196         struct bio *first_bio = bio;
2197         u64 logical = bio->bi_sector << 9;
2198         u64 length = 0;
2199         u64 map_length;
2200         struct btrfs_multi_bio *multi = NULL;
2201         int ret;
2202         int dev_nr = 0;
2203         int total_devs = 1;
2204
2205         length = bio->bi_size;
2206         map_tree = &root->fs_info->mapping_tree;
2207         map_length = length;
2208
2209         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2210                               mirror_num);
2211         BUG_ON(ret);
2212
2213         total_devs = multi->num_stripes;
2214         if (map_length < length) {
2215                 printk("mapping failed logical %Lu bio len %Lu "
2216                        "len %Lu\n", logical, length, map_length);
2217                 BUG();
2218         }
2219         multi->end_io = first_bio->bi_end_io;
2220         multi->private = first_bio->bi_private;
2221         multi->orig_bio = first_bio;
2222         atomic_set(&multi->stripes_pending, multi->num_stripes);
2223
2224         while(dev_nr < total_devs) {
2225                 if (total_devs > 1) {
2226                         if (dev_nr < total_devs - 1) {
2227                                 bio = bio_clone(first_bio, GFP_NOFS);
2228                                 BUG_ON(!bio);
2229                         } else {
2230                                 bio = first_bio;
2231                         }
2232                         bio->bi_private = multi;
2233                         bio->bi_end_io = end_bio_multi_stripe;
2234                 }
2235                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2236                 dev = multi->stripes[dev_nr].dev;
2237                 if (dev && dev->bdev) {
2238                         bio->bi_bdev = dev->bdev;
2239                         if (async_submit)
2240                                 schedule_bio(root, dev, rw, bio);
2241                         else
2242                                 submit_bio(rw, bio);
2243                 } else {
2244                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2245                         bio->bi_sector = logical >> 9;
2246 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2247                         bio_endio(bio, bio->bi_size, -EIO);
2248 #else
2249                         bio_endio(bio, -EIO);
2250 #endif
2251                 }
2252                 dev_nr++;
2253         }
2254         if (total_devs == 1)
2255                 kfree(multi);
2256         return 0;
2257 }
2258
2259 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2260                                        u8 *uuid)
2261 {
2262         struct list_head *head = &root->fs_info->fs_devices->devices;
2263
2264         return __find_device(head, devid, uuid);
2265 }
2266
2267 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2268                                             u64 devid, u8 *dev_uuid)
2269 {
2270         struct btrfs_device *device;
2271         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2272
2273         device = kzalloc(sizeof(*device), GFP_NOFS);
2274         list_add(&device->dev_list,
2275                  &fs_devices->devices);
2276         list_add(&device->dev_alloc_list,
2277                  &fs_devices->alloc_list);
2278         device->barriers = 1;
2279         device->dev_root = root->fs_info->dev_root;
2280         device->devid = devid;
2281         device->work.func = pending_bios_fn;
2282         fs_devices->num_devices++;
2283         spin_lock_init(&device->io_lock);
2284         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2285         return device;
2286 }
2287
2288
2289 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2290                           struct extent_buffer *leaf,
2291                           struct btrfs_chunk *chunk)
2292 {
2293         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2294         struct map_lookup *map;
2295         struct extent_map *em;
2296         u64 logical;
2297         u64 length;
2298         u64 devid;
2299         u8 uuid[BTRFS_UUID_SIZE];
2300         int num_stripes;
2301         int ret;
2302         int i;
2303
2304         logical = key->offset;
2305         length = btrfs_chunk_length(leaf, chunk);
2306
2307         spin_lock(&map_tree->map_tree.lock);
2308         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2309         spin_unlock(&map_tree->map_tree.lock);
2310
2311         /* already mapped? */
2312         if (em && em->start <= logical && em->start + em->len > logical) {
2313                 free_extent_map(em);
2314                 return 0;
2315         } else if (em) {
2316                 free_extent_map(em);
2317         }
2318
2319         map = kzalloc(sizeof(*map), GFP_NOFS);
2320         if (!map)
2321                 return -ENOMEM;
2322
2323         em = alloc_extent_map(GFP_NOFS);
2324         if (!em)
2325                 return -ENOMEM;
2326         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2327         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2328         if (!map) {
2329                 free_extent_map(em);
2330                 return -ENOMEM;
2331         }
2332
2333         em->bdev = (struct block_device *)map;
2334         em->start = logical;
2335         em->len = length;
2336         em->block_start = 0;
2337
2338         map->num_stripes = num_stripes;
2339         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2340         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2341         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2342         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2343         map->type = btrfs_chunk_type(leaf, chunk);
2344         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2345         for (i = 0; i < num_stripes; i++) {
2346                 map->stripes[i].physical =
2347                         btrfs_stripe_offset_nr(leaf, chunk, i);
2348                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2349                 read_extent_buffer(leaf, uuid, (unsigned long)
2350                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2351                                    BTRFS_UUID_SIZE);
2352                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2353
2354                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2355                         kfree(map);
2356                         free_extent_map(em);
2357                         return -EIO;
2358                 }
2359                 if (!map->stripes[i].dev) {
2360                         map->stripes[i].dev =
2361                                 add_missing_dev(root, devid, uuid);
2362                         if (!map->stripes[i].dev) {
2363                                 kfree(map);
2364                                 free_extent_map(em);
2365                                 return -EIO;
2366                         }
2367                 }
2368                 map->stripes[i].dev->in_fs_metadata = 1;
2369         }
2370
2371         spin_lock(&map_tree->map_tree.lock);
2372         ret = add_extent_mapping(&map_tree->map_tree, em);
2373         spin_unlock(&map_tree->map_tree.lock);
2374         BUG_ON(ret);
2375         free_extent_map(em);
2376
2377         return 0;
2378 }
2379
2380 static int fill_device_from_item(struct extent_buffer *leaf,
2381                                  struct btrfs_dev_item *dev_item,
2382                                  struct btrfs_device *device)
2383 {
2384         unsigned long ptr;
2385
2386         device->devid = btrfs_device_id(leaf, dev_item);
2387         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2388         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2389         device->type = btrfs_device_type(leaf, dev_item);
2390         device->io_align = btrfs_device_io_align(leaf, dev_item);
2391         device->io_width = btrfs_device_io_width(leaf, dev_item);
2392         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2393
2394         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2395         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2396
2397         return 0;
2398 }
2399
2400 static int read_one_dev(struct btrfs_root *root,
2401                         struct extent_buffer *leaf,
2402                         struct btrfs_dev_item *dev_item)
2403 {
2404         struct btrfs_device *device;
2405         u64 devid;
2406         int ret;
2407         u8 dev_uuid[BTRFS_UUID_SIZE];
2408
2409         devid = btrfs_device_id(leaf, dev_item);
2410         read_extent_buffer(leaf, dev_uuid,
2411                            (unsigned long)btrfs_device_uuid(dev_item),
2412                            BTRFS_UUID_SIZE);
2413         device = btrfs_find_device(root, devid, dev_uuid);
2414         if (!device) {
2415                 printk("warning devid %Lu missing\n", devid);
2416                 device = add_missing_dev(root, devid, dev_uuid);
2417                 if (!device)
2418                         return -ENOMEM;
2419         }
2420
2421         fill_device_from_item(leaf, dev_item, device);
2422         device->dev_root = root->fs_info->dev_root;
2423         device->in_fs_metadata = 1;
2424         ret = 0;
2425 #if 0
2426         ret = btrfs_open_device(device);
2427         if (ret) {
2428                 kfree(device);
2429         }
2430 #endif
2431         return ret;
2432 }
2433
2434 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2435 {
2436         struct btrfs_dev_item *dev_item;
2437
2438         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2439                                                      dev_item);
2440         return read_one_dev(root, buf, dev_item);
2441 }
2442
2443 int btrfs_read_sys_array(struct btrfs_root *root)
2444 {
2445         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2446         struct extent_buffer *sb;
2447         struct btrfs_disk_key *disk_key;
2448         struct btrfs_chunk *chunk;
2449         u8 *ptr;
2450         unsigned long sb_ptr;
2451         int ret = 0;
2452         u32 num_stripes;
2453         u32 array_size;
2454         u32 len = 0;
2455         u32 cur;
2456         struct btrfs_key key;
2457
2458         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2459                                           BTRFS_SUPER_INFO_SIZE);
2460         if (!sb)
2461                 return -ENOMEM;
2462         btrfs_set_buffer_uptodate(sb);
2463         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2464         array_size = btrfs_super_sys_array_size(super_copy);
2465
2466         ptr = super_copy->sys_chunk_array;
2467         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2468         cur = 0;
2469
2470         while (cur < array_size) {
2471                 disk_key = (struct btrfs_disk_key *)ptr;
2472                 btrfs_disk_key_to_cpu(&key, disk_key);
2473
2474                 len = sizeof(*disk_key); ptr += len;
2475                 sb_ptr += len;
2476                 cur += len;
2477
2478                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2479                         chunk = (struct btrfs_chunk *)sb_ptr;
2480                         ret = read_one_chunk(root, &key, sb, chunk);
2481                         if (ret)
2482                                 break;
2483                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2484                         len = btrfs_chunk_item_size(num_stripes);
2485                 } else {
2486                         ret = -EIO;
2487                         break;
2488                 }
2489                 ptr += len;
2490                 sb_ptr += len;
2491                 cur += len;
2492         }
2493         free_extent_buffer(sb);
2494         return ret;
2495 }
2496
2497 int btrfs_read_chunk_tree(struct btrfs_root *root)
2498 {
2499         struct btrfs_path *path;
2500         struct extent_buffer *leaf;
2501         struct btrfs_key key;
2502         struct btrfs_key found_key;
2503         int ret;
2504         int slot;
2505
2506         root = root->fs_info->chunk_root;
2507
2508         path = btrfs_alloc_path();
2509         if (!path)
2510                 return -ENOMEM;
2511
2512         /* first we search for all of the device items, and then we
2513          * read in all of the chunk items.  This way we can create chunk
2514          * mappings that reference all of the devices that are afound
2515          */
2516         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2517         key.offset = 0;
2518         key.type = 0;
2519 again:
2520         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2521         while(1) {
2522                 leaf = path->nodes[0];
2523                 slot = path->slots[0];
2524                 if (slot >= btrfs_header_nritems(leaf)) {
2525                         ret = btrfs_next_leaf(root, path);
2526                         if (ret == 0)
2527                                 continue;
2528                         if (ret < 0)
2529                                 goto error;
2530                         break;
2531                 }
2532                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2533                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2534                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2535                                 break;
2536                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2537                                 struct btrfs_dev_item *dev_item;
2538                                 dev_item = btrfs_item_ptr(leaf, slot,
2539                                                   struct btrfs_dev_item);
2540                                 ret = read_one_dev(root, leaf, dev_item);
2541                                 BUG_ON(ret);
2542                         }
2543                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2544                         struct btrfs_chunk *chunk;
2545                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2546                         ret = read_one_chunk(root, &found_key, leaf, chunk);
2547                 }
2548                 path->slots[0]++;
2549         }
2550         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2551                 key.objectid = 0;
2552                 btrfs_release_path(root, path);
2553                 goto again;
2554         }
2555
2556         btrfs_free_path(path);
2557         ret = 0;
2558 error:
2559         return ret;
2560 }