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 struct btrfs_device *__find_device(struct list_head *head, u64 devid,
  98                                           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 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 int run_scheduled_bios(struct btrfs_device *device)
 138 {
 139         struct bio *pending;
 140         struct backing_dev_info *bdi;
 141         struct bio *tail;
 142         struct bio *cur;
 143         int again = 0;
 144         unsigned long num_run = 0;
 145
 146         bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
 147 loop:
 148         spin_lock(&device->io_lock);
 149
 150         /* take all the bios off the list at once and process them
 151          * later on (without the lock held).  But, remember the
 152          * tail and other pointers so the bios can be properly reinserted
 153          * into the list if we hit congestion
 154          */
 155         pending = device->pending_bios;
 156         tail = device->pending_bio_tail;
 157         WARN_ON(pending && !tail);
 158         device->pending_bios = NULL;
 159         device->pending_bio_tail = NULL;
 160
 161         /*
 162          * if pending was null this time around, no bios need processing
 163          * at all and we can stop.  Otherwise it'll loop back up again
 164          * and do an additional check so no bios are missed.
 165          *
 166          * device->running_pending is used to synchronize with the
 167          * schedule_bio code.
 168          */
 169         if (pending) {
 170                 again = 1;
 171                 device->running_pending = 1;
 172         } else {
 173                 again = 0;
 174                 device->running_pending = 0;
 175         }
 176         spin_unlock(&device->io_lock);
 177
 178         while(pending) {
 179                 cur = pending;
 180                 pending = pending->bi_next;
 181                 cur->bi_next = NULL;
 182                 atomic_dec(&device->dev_root->fs_info->nr_async_submits);
 183                 submit_bio(cur->bi_rw, cur);
 184                 num_run++;
 185
 186                 /*
 187                  * we made progress, there is more work to do and the bdi
 188                  * is now congested.  Back off and let other work structs
 189                  * run instead
 190                  */
 191                 if (pending && num_run && bdi_write_congested(bdi)) {
 192                         struct bio *old_head;
 193
 194                         spin_lock(&device->io_lock);
 195                         old_head = device->pending_bios;
 196                         device->pending_bios = pending;
 197                         if (device->pending_bio_tail)
 198                                 tail->bi_next = old_head;
 199                         else
 200                                 device->pending_bio_tail = tail;
 201
 202                         spin_unlock(&device->io_lock);
 203                         btrfs_requeue_work(&device->work);
 204                         goto done;
 205                 }
 206         }
 207         if (again)
 208                 goto loop;
 209 done:
 210         return 0;
 211 }
 212
 213 void pending_bios_fn(struct btrfs_work *work)
 214 {
 215         struct btrfs_device *device;
 216
 217         device = container_of(work, struct btrfs_device, work);
 218         run_scheduled_bios(device);
 219 }
 220
 221 static int device_list_add(const char *path,
 222                            struct btrfs_super_block *disk_super,
 223                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 224 {
 225         struct btrfs_device *device;
 226         struct btrfs_fs_devices *fs_devices;
 227         u64 found_transid = btrfs_super_generation(disk_super);
 228
 229         fs_devices = find_fsid(disk_super->fsid);
 230         if (!fs_devices) {
 231                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 232                 if (!fs_devices)
 233                         return -ENOMEM;
 234                 INIT_LIST_HEAD(&fs_devices->devices);
 235                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 236                 list_add(&fs_devices->list, &fs_uuids);
 237                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 238                 fs_devices->latest_devid = devid;
 239                 fs_devices->latest_trans = found_transid;
 240                 device = NULL;
 241         } else {
 242                 device = __find_device(&fs_devices->devices, devid,
 243                                        disk_super->dev_item.uuid);
 244         }
 245         if (!device) {
 246                 device = kzalloc(sizeof(*device), GFP_NOFS);
 247                 if (!device) {
 248                         /* we can safely leave the fs_devices entry around */
 249                         return -ENOMEM;
 250                 }
 251                 device->devid = devid;
 252                 device->work.func = pending_bios_fn;
 253                 memcpy(device->uuid, disk_super->dev_item.uuid,
 254                        BTRFS_UUID_SIZE);
 255                 device->barriers = 1;
 256                 spin_lock_init(&device->io_lock);
 257                 device->name = kstrdup(path, GFP_NOFS);
 258                 if (!device->name) {
 259                         kfree(device);
 260                         return -ENOMEM;
 261                 }
 262                 list_add(&device->dev_list, &fs_devices->devices);
 263                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 264                 fs_devices->num_devices++;
 265         }
 266
 267         if (found_transid > fs_devices->latest_trans) {
 268                 fs_devices->latest_devid = devid;
 269                 fs_devices->latest_trans = found_transid;
 270         }
 271         *fs_devices_ret = fs_devices;
 272         return 0;
 273 }
 274
 275 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 276 {
 277         struct list_head *head = &fs_devices->devices;
 278         struct list_head *cur;
 279         struct btrfs_device *device;
 280
 281         mutex_lock(&uuid_mutex);
 282 again:
 283         list_for_each(cur, head) {
 284                 device = list_entry(cur, struct btrfs_device, dev_list);
 285                 if (!device->in_fs_metadata) {
 286                         struct block_device *bdev;
 287                         list_del(&device->dev_list);
 288                         list_del(&device->dev_alloc_list);
 289                         fs_devices->num_devices--;
 290                         if (device->bdev) {
 291                                 bdev = device->bdev;
 292                                 fs_devices->open_devices--;
 293                                 mutex_unlock(&uuid_mutex);
 294                                 close_bdev_excl(bdev);
 295                                 mutex_lock(&uuid_mutex);
 296                         }
 297                         kfree(device->name);
 298                         kfree(device);
 299                         goto again;
 300                 }
 301         }
 302         mutex_unlock(&uuid_mutex);
 303         return 0;
 304 }
 305
 306 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 307 {
 308         struct list_head *head = &fs_devices->devices;
 309         struct list_head *cur;
 310         struct btrfs_device *device;
 311
 312         mutex_lock(&uuid_mutex);
 313         list_for_each(cur, head) {
 314                 device = list_entry(cur, struct btrfs_device, dev_list);
 315                 if (device->bdev) {
 316                         close_bdev_excl(device->bdev);
 317                         fs_devices->open_devices--;
 318                 }
 319                 device->bdev = NULL;
 320                 device->in_fs_metadata = 0;
 321         }
 322         fs_devices->mounted = 0;
 323         mutex_unlock(&uuid_mutex);
 324         return 0;
 325 }
 326
 327 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 328                        int flags, void *holder)
 329 {
 330         struct block_device *bdev;
 331         struct list_head *head = &fs_devices->devices;
 332         struct list_head *cur;
 333         struct btrfs_device *device;
 334         struct block_device *latest_bdev = NULL;
 335         struct buffer_head *bh;
 336         struct btrfs_super_block *disk_super;
 337         u64 latest_devid = 0;
 338         u64 latest_transid = 0;
 339         u64 transid;
 340         u64 devid;
 341         int ret = 0;
 342
 343         mutex_lock(&uuid_mutex);
 344         if (fs_devices->mounted)
 345                 goto out;
 346
 347         list_for_each(cur, head) {
 348                 device = list_entry(cur, struct btrfs_device, dev_list);
 349                 if (device->bdev)
 350                         continue;
 351
 352                 if (!device->name)
 353                         continue;
 354
 355                 bdev = open_bdev_excl(device->name, flags, holder);
 356
 357                 if (IS_ERR(bdev)) {
 358                         printk("open %s failed\n", device->name);
 359                         goto error;
 360                 }
 361                 set_blocksize(bdev, 4096);
 362
 363                 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 364                 if (!bh)
 365                         goto error_close;
 366
 367                 disk_super = (struct btrfs_super_block *)bh->b_data;
 368                 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 369                     sizeof(disk_super->magic)))
 370                         goto error_brelse;
 371
 372                 devid = le64_to_cpu(disk_super->dev_item.devid);
 373                 if (devid != device->devid)
 374                         goto error_brelse;
 375
 376                 transid = btrfs_super_generation(disk_super);
 377                 if (!latest_transid || transid > latest_transid) {
 378                         latest_devid = devid;
 379                         latest_transid = transid;
 380                         latest_bdev = bdev;
 381                 }
 382
 383                 device->bdev = bdev;
 384                 device->in_fs_metadata = 0;
 385                 fs_devices->open_devices++;
 386                 continue;
 387
 388 error_brelse:
 389                 brelse(bh);
 390 error_close:
 391                 close_bdev_excl(bdev);
 392 error:
 393                 continue;
 394         }
 395         if (fs_devices->open_devices == 0) {
 396                 ret = -EIO;
 397                 goto out;
 398         }
 399         fs_devices->mounted = 1;
 400         fs_devices->latest_bdev = latest_bdev;
 401         fs_devices->latest_devid = latest_devid;
 402         fs_devices->latest_trans = latest_transid;
 403 out:
 404         mutex_unlock(&uuid_mutex);
 405         return ret;
 406 }
 407
 408 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 409                           struct btrfs_fs_devices **fs_devices_ret)
 410 {
 411         struct btrfs_super_block *disk_super;
 412         struct block_device *bdev;
 413         struct buffer_head *bh;
 414         int ret;
 415         u64 devid;
 416         u64 transid;
 417
 418         mutex_lock(&uuid_mutex);
 419
 420         bdev = open_bdev_excl(path, flags, holder);
 421
 422         if (IS_ERR(bdev)) {
 423                 ret = PTR_ERR(bdev);
 424                 goto error;
 425         }
 426
 427         ret = set_blocksize(bdev, 4096);
 428         if (ret)
 429                 goto error_close;
 430         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 431         if (!bh) {
 432                 ret = -EIO;
 433                 goto error_close;
 434         }
 435         disk_super = (struct btrfs_super_block *)bh->b_data;
 436         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 437             sizeof(disk_super->magic))) {
 438                 ret = -EINVAL;
 439                 goto error_brelse;
 440         }
 441         devid = le64_to_cpu(disk_super->dev_item.devid);
 442         transid = btrfs_super_generation(disk_super);
 443         if (disk_super->label[0])
 444                 printk("device label %s ", disk_super->label);
 445         else {
 446                 /* FIXME, make a readl uuid parser */
 447                 printk("device fsid %llx-%llx ",
 448                        *(unsigned long long *)disk_super->fsid,
 449                        *(unsigned long long *)(disk_super->fsid + 8));
 450         }
 451         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 452         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 453
 454 error_brelse:
 455         brelse(bh);
 456 error_close:
 457         close_bdev_excl(bdev);
 458 error:
 459         mutex_unlock(&uuid_mutex);
 460         return ret;
 461 }
 462
 463 /*
 464  * this uses a pretty simple search, the expectation is that it is
 465  * called very infrequently and that a given device has a small number
 466  * of extents
 467  */
 468 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
 469                                 struct btrfs_device *device,
 470                                 struct btrfs_path *path,
 471                                 u64 num_bytes, u64 *start)
 472 {
 473         struct btrfs_key key;
 474         struct btrfs_root *root = device->dev_root;
 475         struct btrfs_dev_extent *dev_extent = NULL;
 476         u64 hole_size = 0;
 477         u64 last_byte = 0;
 478         u64 search_start = 0;
 479         u64 search_end = device->total_bytes;
 480         int ret;
 481         int slot = 0;
 482         int start_found;
 483         struct extent_buffer *l;
 484
 485         start_found = 0;
 486         path->reada = 2;
 487
 488         /* FIXME use last free of some kind */
 489
 490         /* we don't want to overwrite the superblock on the drive,
 491          * so we make sure to start at an offset of at least 1MB
 492          */
 493         search_start = max((u64)1024 * 1024, search_start);
 494
 495         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 496                 search_start = max(root->fs_info->alloc_start, search_start);
 497
 498         key.objectid = device->devid;
 499         key.offset = search_start;
 500         key.type = BTRFS_DEV_EXTENT_KEY;
 501         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 502         if (ret < 0)
 503                 goto error;
 504         ret = btrfs_previous_item(root, path, 0, key.type);
 505         if (ret < 0)
 506                 goto error;
 507         l = path->nodes[0];
 508         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 509         while (1) {
 510                 l = path->nodes[0];
 511                 slot = path->slots[0];
 512                 if (slot >= btrfs_header_nritems(l)) {
 513                         ret = btrfs_next_leaf(root, path);
 514                         if (ret == 0)
 515                                 continue;
 516                         if (ret < 0)
 517                                 goto error;
 518 no_more_items:
 519                         if (!start_found) {
 520                                 if (search_start >= search_end) {
 521                                         ret = -ENOSPC;
 522                                         goto error;
 523                                 }
 524                                 *start = search_start;
 525                                 start_found = 1;
 526                                 goto check_pending;
 527                         }
 528                         *start = last_byte > search_start ?
 529                                 last_byte : search_start;
 530                         if (search_end <= *start) {
 531                                 ret = -ENOSPC;
 532                                 goto error;
 533                         }
 534                         goto check_pending;
 535                 }
 536                 btrfs_item_key_to_cpu(l, &key, slot);
 537
 538                 if (key.objectid < device->devid)
 539                         goto next;
 540
 541                 if (key.objectid > device->devid)
 542                         goto no_more_items;
 543
 544                 if (key.offset >= search_start && key.offset > last_byte &&
 545                     start_found) {
 546                         if (last_byte < search_start)
 547                                 last_byte = search_start;
 548                         hole_size = key.offset - last_byte;
 549                         if (key.offset > last_byte &&
 550                             hole_size >= num_bytes) {
 551                                 *start = last_byte;
 552                                 goto check_pending;
 553                         }
 554                 }
 555                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 556                         goto next;
 557                 }
 558
 559                 start_found = 1;
 560                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 561                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 562 next:
 563                 path->slots[0]++;
 564                 cond_resched();
 565         }
 566 check_pending:
 567         /* we have to make sure we didn't find an extent that has already
 568          * been allocated by the map tree or the original allocation
 569          */
 570         btrfs_release_path(root, path);
 571         BUG_ON(*start < search_start);
 572
 573         if (*start + num_bytes > search_end) {
 574                 ret = -ENOSPC;
 575                 goto error;
 576         }
 577         /* check for pending inserts here */
 578         return 0;
 579
 580 error:
 581         btrfs_release_path(root, path);
 582         return ret;
 583 }
 584
 585 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 586                           struct btrfs_device *device,
 587                           u64 start)
 588 {
 589         int ret;
 590         struct btrfs_path *path;
 591         struct btrfs_root *root = device->dev_root;
 592         struct btrfs_key key;
 593         struct btrfs_key found_key;
 594         struct extent_buffer *leaf = NULL;
 595         struct btrfs_dev_extent *extent = NULL;
 596
 597         path = btrfs_alloc_path();
 598         if (!path)
 599                 return -ENOMEM;
 600
 601         key.objectid = device->devid;
 602         key.offset = start;
 603         key.type = BTRFS_DEV_EXTENT_KEY;
 604
 605         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 606         if (ret > 0) {
 607                 ret = btrfs_previous_item(root, path, key.objectid,
 608                                           BTRFS_DEV_EXTENT_KEY);
 609                 BUG_ON(ret);
 610                 leaf = path->nodes[0];
 611                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 612                 extent = btrfs_item_ptr(leaf, path->slots[0],
 613                                         struct btrfs_dev_extent);
 614                 BUG_ON(found_key.offset > start || found_key.offset +
 615                        btrfs_dev_extent_length(leaf, extent) < start);
 616                 ret = 0;
 617         } else if (ret == 0) {
 618                 leaf = path->nodes[0];
 619                 extent = btrfs_item_ptr(leaf, path->slots[0],
 620                                         struct btrfs_dev_extent);
 621         }
 622         BUG_ON(ret);
 623
 624         if (device->bytes_used > 0)
 625                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 626         ret = btrfs_del_item(trans, root, path);
 627         BUG_ON(ret);
 628
 629         btrfs_free_path(path);
 630         return ret;
 631 }
 632
 633 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 634                            struct btrfs_device *device,
 635                            u64 chunk_tree, u64 chunk_objectid,
 636                            u64 chunk_offset,
 637                            u64 num_bytes, u64 *start)
 638 {
 639         int ret;
 640         struct btrfs_path *path;
 641         struct btrfs_root *root = device->dev_root;
 642         struct btrfs_dev_extent *extent;
 643         struct extent_buffer *leaf;
 644         struct btrfs_key key;
 645
 646         WARN_ON(!device->in_fs_metadata);
 647         path = btrfs_alloc_path();
 648         if (!path)
 649                 return -ENOMEM;
 650
 651         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 652         if (ret) {
 653                 goto err;
 654         }
 655
 656         key.objectid = device->devid;
 657         key.offset = *start;
 658         key.type = BTRFS_DEV_EXTENT_KEY;
 659         ret = btrfs_insert_empty_item(trans, root, path, &key,
 660                                       sizeof(*extent));
 661         BUG_ON(ret);
 662
 663         leaf = path->nodes[0];
 664         extent = btrfs_item_ptr(leaf, path->slots[0],
 665                                 struct btrfs_dev_extent);
 666         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 667         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 668         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 669
 670         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 671                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 672                     BTRFS_UUID_SIZE);
 673
 674         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 675         btrfs_mark_buffer_dirty(leaf);
 676 err:
 677         btrfs_free_path(path);
 678         return ret;
 679 }
 680
 681 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
 682 {
 683         struct btrfs_path *path;
 684         int ret;
 685         struct btrfs_key key;
 686         struct btrfs_chunk *chunk;
 687         struct btrfs_key found_key;
 688
 689         path = btrfs_alloc_path();
 690         BUG_ON(!path);
 691
 692         key.objectid = objectid;
 693         key.offset = (u64)-1;
 694         key.type = BTRFS_CHUNK_ITEM_KEY;
 695
 696         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 697         if (ret < 0)
 698                 goto error;
 699
 700         BUG_ON(ret == 0);
 701
 702         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 703         if (ret) {
 704                 *offset = 0;
 705         } else {
 706                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 707                                       path->slots[0]);
 708                 if (found_key.objectid != objectid)
 709                         *offset = 0;
 710                 else {
 711                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 712                                                struct btrfs_chunk);
 713                         *offset = found_key.offset +
 714                                 btrfs_chunk_length(path->nodes[0], chunk);
 715                 }
 716         }
 717         ret = 0;
 718 error:
 719         btrfs_free_path(path);
 720         return ret;
 721 }
 722
 723 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
 724                            u64 *objectid)
 725 {
 726         int ret;
 727         struct btrfs_key key;
 728         struct btrfs_key found_key;
 729
 730         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 731         key.type = BTRFS_DEV_ITEM_KEY;
 732         key.offset = (u64)-1;
 733
 734         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 735         if (ret < 0)
 736                 goto error;
 737
 738         BUG_ON(ret == 0);
 739
 740         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 741                                   BTRFS_DEV_ITEM_KEY);
 742         if (ret) {
 743                 *objectid = 1;
 744         } else {
 745                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 746                                       path->slots[0]);
 747                 *objectid = found_key.offset + 1;
 748         }
 749         ret = 0;
 750 error:
 751         btrfs_release_path(root, path);
 752         return ret;
 753 }
 754
 755 /*
 756  * the device information is stored in the chunk root
 757  * the btrfs_device struct should be fully filled in
 758  */
 759 int btrfs_add_device(struct btrfs_trans_handle *trans,
 760                      struct btrfs_root *root,
 761                      struct btrfs_device *device)
 762 {
 763         int ret;
 764         struct btrfs_path *path;
 765         struct btrfs_dev_item *dev_item;
 766         struct extent_buffer *leaf;
 767         struct btrfs_key key;
 768         unsigned long ptr;
 769         u64 free_devid = 0;
 770
 771         root = root->fs_info->chunk_root;
 772
 773         path = btrfs_alloc_path();
 774         if (!path)
 775                 return -ENOMEM;
 776
 777         ret = find_next_devid(root, path, &free_devid);
 778         if (ret)
 779                 goto out;
 780
 781         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 782         key.type = BTRFS_DEV_ITEM_KEY;
 783         key.offset = free_devid;
 784
 785         ret = btrfs_insert_empty_item(trans, root, path, &key,
 786                                       sizeof(*dev_item));
 787         if (ret)
 788                 goto out;
 789
 790         leaf = path->nodes[0];
 791         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 792
 793         device->devid = free_devid;
 794         btrfs_set_device_id(leaf, dev_item, device->devid);
 795         btrfs_set_device_type(leaf, dev_item, device->type);
 796         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 797         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 798         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 799         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 800         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 801         btrfs_set_device_group(leaf, dev_item, 0);
 802         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 803         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 804
 805         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 806         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 807         btrfs_mark_buffer_dirty(leaf);
 808         ret = 0;
 809
 810 out:
 811         btrfs_free_path(path);
 812         return ret;
 813 }
 814
 815 static int btrfs_rm_dev_item(struct btrfs_root *root,
 816                              struct btrfs_device *device)
 817 {
 818         int ret;
 819         struct btrfs_path *path;
 820         struct block_device *bdev = device->bdev;
 821         struct btrfs_device *next_dev;
 822         struct btrfs_key key;
 823         u64 total_bytes;
 824         struct btrfs_fs_devices *fs_devices;
 825         struct btrfs_trans_handle *trans;
 826
 827         root = root->fs_info->chunk_root;
 828
 829         path = btrfs_alloc_path();
 830         if (!path)
 831                 return -ENOMEM;
 832
 833         trans = btrfs_start_transaction(root, 1);
 834         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 835         key.type = BTRFS_DEV_ITEM_KEY;
 836         key.offset = device->devid;
 837         lock_chunks(root);
 838
 839         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 840         if (ret < 0)
 841                 goto out;
 842
 843         if (ret > 0) {
 844                 ret = -ENOENT;
 845                 goto out;
 846         }
 847
 848         ret = btrfs_del_item(trans, root, path);
 849         if (ret)
 850                 goto out;
 851
 852         /*
 853          * at this point, the device is zero sized.  We want to
 854          * remove it from the devices list and zero out the old super
 855          */
 856         list_del_init(&device->dev_list);
 857         list_del_init(&device->dev_alloc_list);
 858         fs_devices = root->fs_info->fs_devices;
 859
 860         next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
 861                               dev_list);
 862         if (bdev == root->fs_info->sb->s_bdev)
 863                 root->fs_info->sb->s_bdev = next_dev->bdev;
 864         if (bdev == fs_devices->latest_bdev)
 865                 fs_devices->latest_bdev = next_dev->bdev;
 866
 867         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 868         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 869                                     total_bytes - 1);
 870 out:
 871         btrfs_free_path(path);
 872         unlock_chunks(root);
 873         btrfs_commit_transaction(trans, root);
 874         return ret;
 875 }
 876
 877 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 878 {
 879         struct btrfs_device *device;
 880         struct block_device *bdev;
 881         struct buffer_head *bh = NULL;
 882         struct btrfs_super_block *disk_super;
 883         u64 all_avail;
 884         u64 devid;
 885         int ret = 0;
 886
 887         mutex_lock(&uuid_mutex);
 888         mutex_lock(&root->fs_info->volume_mutex);
 889
 890         all_avail = root->fs_info->avail_data_alloc_bits |
 891                 root->fs_info->avail_system_alloc_bits |
 892                 root->fs_info->avail_metadata_alloc_bits;
 893
 894         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 895             btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
 896                 printk("btrfs: unable to go below four devices on raid10\n");
 897                 ret = -EINVAL;
 898                 goto out;
 899         }
 900
 901         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 902             btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
 903                 printk("btrfs: unable to go below two devices on raid1\n");
 904                 ret = -EINVAL;
 905                 goto out;
 906         }
 907
 908         if (strcmp(device_path, "missing") == 0) {
 909                 struct list_head *cur;
 910                 struct list_head *devices;
 911                 struct btrfs_device *tmp;
 912
 913                 device = NULL;
 914                 devices = &root->fs_info->fs_devices->devices;
 915                 list_for_each(cur, devices) {
 916                         tmp = list_entry(cur, struct btrfs_device, dev_list);
 917                         if (tmp->in_fs_metadata && !tmp->bdev) {
 918                                 device = tmp;
 919                                 break;
 920                         }
 921                 }
 922                 bdev = NULL;
 923                 bh = NULL;
 924                 disk_super = NULL;
 925                 if (!device) {
 926                         printk("btrfs: no missing devices found to remove\n");
 927                         goto out;
 928                 }
 929
 930         } else {
 931                 bdev = open_bdev_excl(device_path, 0,
 932                                       root->fs_info->bdev_holder);
 933                 if (IS_ERR(bdev)) {
 934                         ret = PTR_ERR(bdev);
 935                         goto out;
 936                 }
 937
 938                 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 939                 if (!bh) {
 940                         ret = -EIO;
 941                         goto error_close;
 942                 }
 943                 disk_super = (struct btrfs_super_block *)bh->b_data;
 944                 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 945                     sizeof(disk_super->magic))) {
 946                         ret = -ENOENT;
 947                         goto error_brelse;
 948                 }
 949                 if (memcmp(disk_super->fsid, root->fs_info->fsid,
 950                            BTRFS_FSID_SIZE)) {
 951                         ret = -ENOENT;
 952                         goto error_brelse;
 953                 }
 954                 devid = le64_to_cpu(disk_super->dev_item.devid);
 955                 device = btrfs_find_device(root, devid, NULL);
 956                 if (!device) {
 957                         ret = -ENOENT;
 958                         goto error_brelse;
 959                 }
 960
 961         }
 962         root->fs_info->fs_devices->num_devices--;
 963         root->fs_info->fs_devices->open_devices--;
 964
 965         ret = btrfs_shrink_device(device, 0);
 966         if (ret)
 967                 goto error_brelse;
 968
 969
 970         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
 971         if (ret)
 972                 goto error_brelse;
 973
 974         if (bh) {
 975                 /* make sure this device isn't detected as part of
 976                  * the FS anymore
 977                  */
 978                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
 979                 set_buffer_dirty(bh);
 980                 sync_dirty_buffer(bh);
 981
 982                 brelse(bh);
 983         }
 984
 985         if (device->bdev) {
 986                 /* one close for the device struct or super_block */
 987                 close_bdev_excl(device->bdev);
 988         }
 989         if (bdev) {
 990                 /* one close for us */
 991                 close_bdev_excl(bdev);
 992         }
 993         kfree(device->name);
 994         kfree(device);
 995         ret = 0;
 996         goto out;
 997
 998 error_brelse:
 999         brelse(bh);
1000 error_close:
1001         if (bdev)
1002                 close_bdev_excl(bdev);
1003 out:
1004         mutex_unlock(&root->fs_info->volume_mutex);
1005         mutex_unlock(&uuid_mutex);
1006         return ret;
1007 }
1008
1009 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1010 {
1011         struct btrfs_trans_handle *trans;
1012         struct btrfs_device *device;
1013         struct block_device *bdev;
1014         struct list_head *cur;
1015         struct list_head *devices;
1016         u64 total_bytes;
1017         int ret = 0;
1018
1019
1020         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1021         if (!bdev) {
1022                 return -EIO;
1023         }
1024
1025         mutex_lock(&root->fs_info->volume_mutex);
1026
1027         trans = btrfs_start_transaction(root, 1);
1028         lock_chunks(root);
1029         devices = &root->fs_info->fs_devices->devices;
1030         list_for_each(cur, devices) {
1031                 device = list_entry(cur, struct btrfs_device, dev_list);
1032                 if (device->bdev == bdev) {
1033                         ret = -EEXIST;
1034                         goto out;
1035                 }
1036         }
1037
1038         device = kzalloc(sizeof(*device), GFP_NOFS);
1039         if (!device) {
1040                 /* we can safely leave the fs_devices entry around */
1041                 ret = -ENOMEM;
1042                 goto out_close_bdev;
1043         }
1044
1045         device->barriers = 1;
1046         device->work.func = pending_bios_fn;
1047         generate_random_uuid(device->uuid);
1048         spin_lock_init(&device->io_lock);
1049         device->name = kstrdup(device_path, GFP_NOFS);
1050         if (!device->name) {
1051                 kfree(device);
1052                 goto out_close_bdev;
1053         }
1054         device->io_width = root->sectorsize;
1055         device->io_align = root->sectorsize;
1056         device->sector_size = root->sectorsize;
1057         device->total_bytes = i_size_read(bdev->bd_inode);
1058         device->dev_root = root->fs_info->dev_root;
1059         device->bdev = bdev;
1060         device->in_fs_metadata = 1;
1061
1062         ret = btrfs_add_device(trans, root, device);
1063         if (ret)
1064                 goto out_close_bdev;
1065
1066         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1067         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1068                                     total_bytes + device->total_bytes);
1069
1070         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1071         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1072                                     total_bytes + 1);
1073
1074         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1075         list_add(&device->dev_alloc_list,
1076                  &root->fs_info->fs_devices->alloc_list);
1077         root->fs_info->fs_devices->num_devices++;
1078         root->fs_info->fs_devices->open_devices++;
1079 out:
1080         unlock_chunks(root);
1081         btrfs_end_transaction(trans, root);
1082         mutex_unlock(&root->fs_info->volume_mutex);
1083
1084         return ret;
1085
1086 out_close_bdev:
1087         close_bdev_excl(bdev);
1088         goto out;
1089 }
1090
1091 int btrfs_update_device(struct btrfs_trans_handle *trans,
1092                         struct btrfs_device *device)
1093 {
1094         int ret;
1095         struct btrfs_path *path;
1096         struct btrfs_root *root;
1097         struct btrfs_dev_item *dev_item;
1098         struct extent_buffer *leaf;
1099         struct btrfs_key key;
1100
1101         root = device->dev_root->fs_info->chunk_root;
1102
1103         path = btrfs_alloc_path();
1104         if (!path)
1105                 return -ENOMEM;
1106
1107         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1108         key.type = BTRFS_DEV_ITEM_KEY;
1109         key.offset = device->devid;
1110
1111         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1112         if (ret < 0)
1113                 goto out;
1114
1115         if (ret > 0) {
1116                 ret = -ENOENT;
1117                 goto out;
1118         }
1119
1120         leaf = path->nodes[0];
1121         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1122
1123         btrfs_set_device_id(leaf, dev_item, device->devid);
1124         btrfs_set_device_type(leaf, dev_item, device->type);
1125         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1126         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1127         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1128         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1129         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1130         btrfs_mark_buffer_dirty(leaf);
1131
1132 out:
1133         btrfs_free_path(path);
1134         return ret;
1135 }
1136
1137 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1138                       struct btrfs_device *device, u64 new_size)
1139 {
1140         struct btrfs_super_block *super_copy =
1141                 &device->dev_root->fs_info->super_copy;
1142         u64 old_total = btrfs_super_total_bytes(super_copy);
1143         u64 diff = new_size - device->total_bytes;
1144
1145         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1146         return btrfs_update_device(trans, device);
1147 }
1148
1149 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1150                       struct btrfs_device *device, u64 new_size)
1151 {
1152         int ret;
1153         lock_chunks(device->dev_root);
1154         ret = __btrfs_grow_device(trans, device, new_size);
1155         unlock_chunks(device->dev_root);
1156         return ret;
1157 }
1158
1159 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1160                             struct btrfs_root *root,
1161                             u64 chunk_tree, u64 chunk_objectid,
1162                             u64 chunk_offset)
1163 {
1164         int ret;
1165         struct btrfs_path *path;
1166         struct btrfs_key key;
1167
1168         root = root->fs_info->chunk_root;
1169         path = btrfs_alloc_path();
1170         if (!path)
1171                 return -ENOMEM;
1172
1173         key.objectid = chunk_objectid;
1174         key.offset = chunk_offset;
1175         key.type = BTRFS_CHUNK_ITEM_KEY;
1176
1177         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1178         BUG_ON(ret);
1179
1180         ret = btrfs_del_item(trans, root, path);
1181         BUG_ON(ret);
1182
1183         btrfs_free_path(path);
1184         return 0;
1185 }
1186
1187 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1188                         chunk_offset)
1189 {
1190         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1191         struct btrfs_disk_key *disk_key;
1192         struct btrfs_chunk *chunk;
1193         u8 *ptr;
1194         int ret = 0;
1195         u32 num_stripes;
1196         u32 array_size;
1197         u32 len = 0;
1198         u32 cur;
1199         struct btrfs_key key;
1200
1201         array_size = btrfs_super_sys_array_size(super_copy);
1202
1203         ptr = super_copy->sys_chunk_array;
1204         cur = 0;
1205
1206         while (cur < array_size) {
1207                 disk_key = (struct btrfs_disk_key *)ptr;
1208                 btrfs_disk_key_to_cpu(&key, disk_key);
1209
1210                 len = sizeof(*disk_key);
1211
1212                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1213                         chunk = (struct btrfs_chunk *)(ptr + len);
1214                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1215                         len += btrfs_chunk_item_size(num_stripes);
1216                 } else {
1217                         ret = -EIO;
1218                         break;
1219                 }
1220                 if (key.objectid == chunk_objectid &&
1221                     key.offset == chunk_offset) {
1222                         memmove(ptr, ptr + len, array_size - (cur + len));
1223                         array_size -= len;
1224                         btrfs_set_super_sys_array_size(super_copy, array_size);
1225                 } else {
1226                         ptr += len;
1227                         cur += len;
1228                 }
1229         }
1230         return ret;
1231 }
1232
1233
1234 int btrfs_relocate_chunk(struct btrfs_root *root,
1235                          u64 chunk_tree, u64 chunk_objectid,
1236                          u64 chunk_offset)
1237 {
1238         struct extent_map_tree *em_tree;
1239         struct btrfs_root *extent_root;
1240         struct btrfs_trans_handle *trans;
1241         struct extent_map *em;
1242         struct map_lookup *map;
1243         int ret;
1244         int i;
1245
1246         printk("btrfs relocating chunk %llu\n",
1247                (unsigned long long)chunk_offset);
1248         root = root->fs_info->chunk_root;
1249         extent_root = root->fs_info->extent_root;
1250         em_tree = &root->fs_info->mapping_tree.map_tree;
1251
1252         /* step one, relocate all the extents inside this chunk */
1253         ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1254         BUG_ON(ret);
1255
1256         trans = btrfs_start_transaction(root, 1);
1257         BUG_ON(!trans);
1258
1259         lock_chunks(root);
1260
1261         /*
1262          * step two, delete the device extents and the
1263          * chunk tree entries
1264          */
1265         spin_lock(&em_tree->lock);
1266         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1267         spin_unlock(&em_tree->lock);
1268
1269         BUG_ON(em->start > chunk_offset ||
1270                em->start + em->len < chunk_offset);
1271         map = (struct map_lookup *)em->bdev;
1272
1273         for (i = 0; i < map->num_stripes; i++) {
1274                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1275                                             map->stripes[i].physical);
1276                 BUG_ON(ret);
1277
1278                 if (map->stripes[i].dev) {
1279                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1280                         BUG_ON(ret);
1281                 }
1282         }
1283         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1284                                chunk_offset);
1285
1286         BUG_ON(ret);
1287
1288         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1289                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1290                 BUG_ON(ret);
1291         }
1292
1293         spin_lock(&em_tree->lock);
1294         remove_extent_mapping(em_tree, em);
1295         kfree(map);
1296         em->bdev = NULL;
1297
1298         /* once for the tree */
1299         free_extent_map(em);
1300         spin_unlock(&em_tree->lock);
1301
1302         /* once for us */
1303         free_extent_map(em);
1304
1305         unlock_chunks(root);
1306         btrfs_end_transaction(trans, root);
1307         return 0;
1308 }
1309
1310 static u64 div_factor(u64 num, int factor)
1311 {
1312         if (factor == 10)
1313                 return num;
1314         num *= factor;
1315         do_div(num, 10);
1316         return num;
1317 }
1318
1319
1320 int btrfs_balance(struct btrfs_root *dev_root)
1321 {
1322         int ret;
1323         struct list_head *cur;
1324         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1325         struct btrfs_device *device;
1326         u64 old_size;
1327         u64 size_to_free;
1328         struct btrfs_path *path;
1329         struct btrfs_key key;
1330         struct btrfs_chunk *chunk;
1331         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1332         struct btrfs_trans_handle *trans;
1333         struct btrfs_key found_key;
1334
1335
1336         mutex_lock(&dev_root->fs_info->volume_mutex);
1337         dev_root = dev_root->fs_info->dev_root;
1338
1339         /* step one make some room on all the devices */
1340         list_for_each(cur, devices) {
1341                 device = list_entry(cur, struct btrfs_device, dev_list);
1342                 old_size = device->total_bytes;
1343                 size_to_free = div_factor(old_size, 1);
1344                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1345                 if (device->total_bytes - device->bytes_used > size_to_free)
1346                         continue;
1347
1348                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1349                 BUG_ON(ret);
1350
1351                 trans = btrfs_start_transaction(dev_root, 1);
1352                 BUG_ON(!trans);
1353
1354                 ret = btrfs_grow_device(trans, device, old_size);
1355                 BUG_ON(ret);
1356
1357                 btrfs_end_transaction(trans, dev_root);
1358         }
1359
1360         /* step two, relocate all the chunks */
1361         path = btrfs_alloc_path();
1362         BUG_ON(!path);
1363
1364         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1365         key.offset = (u64)-1;
1366         key.type = BTRFS_CHUNK_ITEM_KEY;
1367
1368         while(1) {
1369                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1370                 if (ret < 0)
1371                         goto error;
1372
1373                 /*
1374                  * this shouldn't happen, it means the last relocate
1375                  * failed
1376                  */
1377                 if (ret == 0)
1378                         break;
1379
1380                 ret = btrfs_previous_item(chunk_root, path, 0,
1381                                           BTRFS_CHUNK_ITEM_KEY);
1382                 if (ret)
1383                         break;
1384
1385                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1386                                       path->slots[0]);
1387                 if (found_key.objectid != key.objectid)
1388                         break;
1389
1390                 chunk = btrfs_item_ptr(path->nodes[0],
1391                                        path->slots[0],
1392                                        struct btrfs_chunk);
1393                 key.offset = found_key.offset;
1394                 /* chunk zero is special */
1395                 if (key.offset == 0)
1396                         break;
1397
1398                 btrfs_release_path(chunk_root, path);
1399                 ret = btrfs_relocate_chunk(chunk_root,
1400                                            chunk_root->root_key.objectid,
1401                                            found_key.objectid,
1402                                            found_key.offset);
1403                 BUG_ON(ret);
1404         }
1405         ret = 0;
1406 error:
1407         btrfs_free_path(path);
1408         mutex_unlock(&dev_root->fs_info->volume_mutex);
1409         return ret;
1410 }
1411
1412 /*
1413  * shrinking a device means finding all of the device extents past
1414  * the new size, and then following the back refs to the chunks.
1415  * The chunk relocation code actually frees the device extent
1416  */
1417 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1418 {
1419         struct btrfs_trans_handle *trans;
1420         struct btrfs_root *root = device->dev_root;
1421         struct btrfs_dev_extent *dev_extent = NULL;
1422         struct btrfs_path *path;
1423         u64 length;
1424         u64 chunk_tree;
1425         u64 chunk_objectid;
1426         u64 chunk_offset;
1427         int ret;
1428         int slot;
1429         struct extent_buffer *l;
1430         struct btrfs_key key;
1431         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1432         u64 old_total = btrfs_super_total_bytes(super_copy);
1433         u64 diff = device->total_bytes - new_size;
1434
1435
1436         path = btrfs_alloc_path();
1437         if (!path)
1438                 return -ENOMEM;
1439
1440         trans = btrfs_start_transaction(root, 1);
1441         if (!trans) {
1442                 ret = -ENOMEM;
1443                 goto done;
1444         }
1445
1446         path->reada = 2;
1447
1448         lock_chunks(root);
1449
1450         device->total_bytes = new_size;
1451         ret = btrfs_update_device(trans, device);
1452         if (ret) {
1453                 unlock_chunks(root);
1454                 btrfs_end_transaction(trans, root);
1455                 goto done;
1456         }
1457         WARN_ON(diff > old_total);
1458         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1459         unlock_chunks(root);
1460         btrfs_end_transaction(trans, root);
1461
1462         key.objectid = device->devid;
1463         key.offset = (u64)-1;
1464         key.type = BTRFS_DEV_EXTENT_KEY;
1465
1466         while (1) {
1467                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1468                 if (ret < 0)
1469                         goto done;
1470
1471                 ret = btrfs_previous_item(root, path, 0, key.type);
1472                 if (ret < 0)
1473                         goto done;
1474                 if (ret) {
1475                         ret = 0;
1476                         goto done;
1477                 }
1478
1479                 l = path->nodes[0];
1480                 slot = path->slots[0];
1481                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1482
1483                 if (key.objectid != device->devid)
1484                         goto done;
1485
1486                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1487                 length = btrfs_dev_extent_length(l, dev_extent);
1488
1489                 if (key.offset + length <= new_size)
1490                         goto done;
1491
1492                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1493                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1494                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1495                 btrfs_release_path(root, path);
1496
1497                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1498                                            chunk_offset);
1499                 if (ret)
1500                         goto done;
1501         }
1502
1503 done:
1504         btrfs_free_path(path);
1505         return ret;
1506 }
1507
1508 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1509                            struct btrfs_root *root,
1510                            struct btrfs_key *key,
1511                            struct btrfs_chunk *chunk, int item_size)
1512 {
1513         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1514         struct btrfs_disk_key disk_key;
1515         u32 array_size;
1516         u8 *ptr;
1517
1518         array_size = btrfs_super_sys_array_size(super_copy);
1519         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1520                 return -EFBIG;
1521
1522         ptr = super_copy->sys_chunk_array + array_size;
1523         btrfs_cpu_key_to_disk(&disk_key, key);
1524         memcpy(ptr, &disk_key, sizeof(disk_key));
1525         ptr += sizeof(disk_key);
1526         memcpy(ptr, chunk, item_size);
1527         item_size += sizeof(disk_key);
1528         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1529         return 0;
1530 }
1531
1532 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1533                                int sub_stripes)
1534 {
1535         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1536                 return calc_size;
1537         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1538                 return calc_size * (num_stripes / sub_stripes);
1539         else
1540                 return calc_size * num_stripes;
1541 }
1542
1543
1544 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1545                       struct btrfs_root *extent_root, u64 *start,
1546                       u64 *num_bytes, u64 type)
1547 {
1548         u64 dev_offset;
1549         struct btrfs_fs_info *info = extent_root->fs_info;
1550         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1551         struct btrfs_path *path;
1552         struct btrfs_stripe *stripes;
1553         struct btrfs_device *device = NULL;
1554         struct btrfs_chunk *chunk;
1555         struct list_head private_devs;
1556         struct list_head *dev_list;
1557         struct list_head *cur;
1558         struct extent_map_tree *em_tree;
1559         struct map_lookup *map;
1560         struct extent_map *em;
1561         int min_stripe_size = 1 * 1024 * 1024;
1562         u64 physical;
1563         u64 calc_size = 1024 * 1024 * 1024;
1564         u64 max_chunk_size = calc_size;
1565         u64 min_free;
1566         u64 avail;
1567         u64 max_avail = 0;
1568         u64 percent_max;
1569         int num_stripes = 1;
1570         int min_stripes = 1;
1571         int sub_stripes = 0;
1572         int looped = 0;
1573         int ret;
1574         int index;
1575         int stripe_len = 64 * 1024;
1576         struct btrfs_key key;
1577
1578         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1579             (type & BTRFS_BLOCK_GROUP_DUP)) {
1580                 WARN_ON(1);
1581                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1582         }
1583         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1584         if (list_empty(dev_list))
1585                 return -ENOSPC;
1586
1587         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1588                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1589                 min_stripes = 2;
1590         }
1591         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1592                 num_stripes = 2;
1593                 min_stripes = 2;
1594         }
1595         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1596                 num_stripes = min_t(u64, 2,
1597                             extent_root->fs_info->fs_devices->open_devices);
1598                 if (num_stripes < 2)
1599                         return -ENOSPC;
1600                 min_stripes = 2;
1601         }
1602         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1603                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1604                 if (num_stripes < 4)
1605                         return -ENOSPC;
1606                 num_stripes &= ~(u32)1;
1607                 sub_stripes = 2;
1608                 min_stripes = 4;
1609         }
1610
1611         if (type & BTRFS_BLOCK_GROUP_DATA) {
1612                 max_chunk_size = 10 * calc_size;
1613                 min_stripe_size = 64 * 1024 * 1024;
1614         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1615                 max_chunk_size = 4 * calc_size;
1616                 min_stripe_size = 32 * 1024 * 1024;
1617         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1618                 calc_size = 8 * 1024 * 1024;
1619                 max_chunk_size = calc_size * 2;
1620                 min_stripe_size = 1 * 1024 * 1024;
1621         }
1622
1623         path = btrfs_alloc_path();
1624         if (!path)
1625                 return -ENOMEM;
1626
1627         /* we don't want a chunk larger than 10% of the FS */
1628         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1629         max_chunk_size = min(percent_max, max_chunk_size);
1630
1631 again:
1632         if (calc_size * num_stripes > max_chunk_size) {
1633                 calc_size = max_chunk_size;
1634                 do_div(calc_size, num_stripes);
1635                 do_div(calc_size, stripe_len);
1636                 calc_size *= stripe_len;
1637         }
1638         /* we don't want tiny stripes */
1639         calc_size = max_t(u64, min_stripe_size, calc_size);
1640
1641         do_div(calc_size, stripe_len);
1642         calc_size *= stripe_len;
1643
1644         INIT_LIST_HEAD(&private_devs);
1645         cur = dev_list->next;
1646         index = 0;
1647
1648         if (type & BTRFS_BLOCK_GROUP_DUP)
1649                 min_free = calc_size * 2;
1650         else
1651                 min_free = calc_size;
1652
1653         /* we add 1MB because we never use the first 1MB of the device */
1654         min_free += 1024 * 1024;
1655
1656         /* build a private list of devices we will allocate from */
1657         while(index < num_stripes) {
1658                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1659
1660                 if (device->total_bytes > device->bytes_used)
1661                         avail = device->total_bytes - device->bytes_used;
1662                 else
1663                         avail = 0;
1664                 cur = cur->next;
1665
1666                 if (device->in_fs_metadata && avail >= min_free) {
1667                         u64 ignored_start = 0;
1668                         ret = find_free_dev_extent(trans, device, path,
1669                                                    min_free,
1670                                                    &ignored_start);
1671                         if (ret == 0) {
1672                                 list_move_tail(&device->dev_alloc_list,
1673                                                &private_devs);
1674                                 index++;
1675                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1676                                         index++;
1677                         }
1678                 } else if (device->in_fs_metadata && avail > max_avail)
1679                         max_avail = avail;
1680                 if (cur == dev_list)
1681                         break;
1682         }
1683         if (index < num_stripes) {
1684                 list_splice(&private_devs, dev_list);
1685                 if (index >= min_stripes) {
1686                         num_stripes = index;
1687                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1688                                 num_stripes /= sub_stripes;
1689                                 num_stripes *= sub_stripes;
1690                         }
1691                         looped = 1;
1692                         goto again;
1693                 }
1694                 if (!looped && max_avail > 0) {
1695                         looped = 1;
1696                         calc_size = max_avail;
1697                         goto again;
1698                 }
1699                 btrfs_free_path(path);
1700                 return -ENOSPC;
1701         }
1702         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1703         key.type = BTRFS_CHUNK_ITEM_KEY;
1704         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1705                               &key.offset);
1706         if (ret) {
1707                 btrfs_free_path(path);
1708                 return ret;
1709         }
1710
1711         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1712         if (!chunk) {
1713                 btrfs_free_path(path);
1714                 return -ENOMEM;
1715         }
1716
1717         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1718         if (!map) {
1719                 kfree(chunk);
1720                 btrfs_free_path(path);
1721                 return -ENOMEM;
1722         }
1723         btrfs_free_path(path);
1724         path = NULL;
1725
1726         stripes = &chunk->stripe;
1727         *num_bytes = chunk_bytes_by_type(type, calc_size,
1728                                          num_stripes, sub_stripes);
1729
1730         index = 0;
1731         while(index < num_stripes) {
1732                 struct btrfs_stripe *stripe;
1733                 BUG_ON(list_empty(&private_devs));
1734                 cur = private_devs.next;
1735                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1736
1737                 /* loop over this device again if we're doing a dup group */
1738                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1739                     (index == num_stripes - 1))
1740                         list_move_tail(&device->dev_alloc_list, dev_list);
1741
1742                 ret = btrfs_alloc_dev_extent(trans, device,
1743                              info->chunk_root->root_key.objectid,
1744                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1745                              calc_size, &dev_offset);
1746                 BUG_ON(ret);
1747                 device->bytes_used += calc_size;
1748                 ret = btrfs_update_device(trans, device);
1749                 BUG_ON(ret);
1750
1751                 map->stripes[index].dev = device;
1752                 map->stripes[index].physical = dev_offset;
1753                 stripe = stripes + index;
1754                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1755                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1756                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1757                 physical = dev_offset;
1758                 index++;
1759         }
1760         BUG_ON(!list_empty(&private_devs));
1761
1762         /* key was set above */
1763         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1764         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1765         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1766         btrfs_set_stack_chunk_type(chunk, type);
1767         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1768         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1769         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1770         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1771         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1772         map->sector_size = extent_root->sectorsize;
1773         map->stripe_len = stripe_len;
1774         map->io_align = stripe_len;
1775         map->io_width = stripe_len;
1776         map->type = type;
1777         map->num_stripes = num_stripes;
1778         map->sub_stripes = sub_stripes;
1779
1780         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1781                                 btrfs_chunk_item_size(num_stripes));
1782         BUG_ON(ret);
1783         *start = key.offset;;
1784
1785         em = alloc_extent_map(GFP_NOFS);
1786         if (!em)
1787                 return -ENOMEM;
1788         em->bdev = (struct block_device *)map;
1789         em->start = key.offset;
1790         em->len = *num_bytes;
1791         em->block_start = 0;
1792
1793         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1794                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1795                                     chunk, btrfs_chunk_item_size(num_stripes));
1796                 BUG_ON(ret);
1797         }
1798         kfree(chunk);
1799
1800         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1801         spin_lock(&em_tree->lock);
1802         ret = add_extent_mapping(em_tree, em);
1803         spin_unlock(&em_tree->lock);
1804         BUG_ON(ret);
1805         free_extent_map(em);
1806         return ret;
1807 }
1808
1809 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1810 {
1811         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1812 }
1813
1814 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1815 {
1816         struct extent_map *em;
1817
1818         while(1) {
1819                 spin_lock(&tree->map_tree.lock);
1820                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1821                 if (em)
1822                         remove_extent_mapping(&tree->map_tree, em);
1823                 spin_unlock(&tree->map_tree.lock);
1824                 if (!em)
1825                         break;
1826                 kfree(em->bdev);
1827                 /* once for us */
1828                 free_extent_map(em);
1829                 /* once for the tree */
1830                 free_extent_map(em);
1831         }
1832 }
1833
1834 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1835 {
1836         struct extent_map *em;
1837         struct map_lookup *map;
1838         struct extent_map_tree *em_tree = &map_tree->map_tree;
1839         int ret;
1840
1841         spin_lock(&em_tree->lock);
1842         em = lookup_extent_mapping(em_tree, logical, len);
1843         spin_unlock(&em_tree->lock);
1844         BUG_ON(!em);
1845
1846         BUG_ON(em->start > logical || em->start + em->len < logical);
1847         map = (struct map_lookup *)em->bdev;
1848         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1849                 ret = map->num_stripes;
1850         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1851                 ret = map->sub_stripes;
1852         else
1853                 ret = 1;
1854         free_extent_map(em);
1855         return ret;
1856 }
1857
1858 static int find_live_mirror(struct map_lookup *map, int first, int num,
1859                             int optimal)
1860 {
1861         int i;
1862         if (map->stripes[optimal].dev->bdev)
1863                 return optimal;
1864         for (i = first; i < first + num; i++) {
1865                 if (map->stripes[i].dev->bdev)
1866                         return i;
1867         }
1868         /* we couldn't find one that doesn't fail.  Just return something
1869          * and the io error handling code will clean up eventually
1870          */
1871         return optimal;
1872 }
1873
1874 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1875                              u64 logical, u64 *length,
1876                              struct btrfs_multi_bio **multi_ret,
1877                              int mirror_num, struct page *unplug_page)
1878 {
1879         struct extent_map *em;
1880         struct map_lookup *map;
1881         struct extent_map_tree *em_tree = &map_tree->map_tree;
1882         u64 offset;
1883         u64 stripe_offset;
1884         u64 stripe_nr;
1885         int stripes_allocated = 8;
1886         int stripes_required = 1;
1887         int stripe_index;
1888         int i;
1889         int num_stripes;
1890         int max_errors = 0;
1891         struct btrfs_multi_bio *multi = NULL;
1892
1893         if (multi_ret && !(rw & (1 << BIO_RW))) {
1894                 stripes_allocated = 1;
1895         }
1896 again:
1897         if (multi_ret) {
1898                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1899                                 GFP_NOFS);
1900                 if (!multi)
1901                         return -ENOMEM;
1902
1903                 atomic_set(&multi->error, 0);
1904         }
1905
1906         spin_lock(&em_tree->lock);
1907         em = lookup_extent_mapping(em_tree, logical, *length);
1908         spin_unlock(&em_tree->lock);
1909
1910         if (!em && unplug_page)
1911                 return 0;
1912
1913         if (!em) {
1914                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1915                 BUG();
1916         }
1917
1918         BUG_ON(em->start > logical || em->start + em->len < logical);
1919         map = (struct map_lookup *)em->bdev;
1920         offset = logical - em->start;
1921
1922         if (mirror_num > map->num_stripes)
1923                 mirror_num = 0;
1924
1925         /* if our multi bio struct is too small, back off and try again */
1926         if (rw & (1 << BIO_RW)) {
1927                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1928                                  BTRFS_BLOCK_GROUP_DUP)) {
1929                         stripes_required = map->num_stripes;
1930                         max_errors = 1;
1931                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1932                         stripes_required = map->sub_stripes;
1933                         max_errors = 1;
1934                 }
1935         }
1936         if (multi_ret && rw == WRITE &&
1937             stripes_allocated < stripes_required) {
1938                 stripes_allocated = map->num_stripes;
1939                 free_extent_map(em);
1940                 kfree(multi);
1941                 goto again;
1942         }
1943         stripe_nr = offset;
1944         /*
1945          * stripe_nr counts the total number of stripes we have to stride
1946          * to get to this block
1947          */
1948         do_div(stripe_nr, map->stripe_len);
1949
1950         stripe_offset = stripe_nr * map->stripe_len;
1951         BUG_ON(offset < stripe_offset);
1952
1953         /* stripe_offset is the offset of this block in its stripe*/
1954         stripe_offset = offset - stripe_offset;
1955
1956         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1957                          BTRFS_BLOCK_GROUP_RAID10 |
1958                          BTRFS_BLOCK_GROUP_DUP)) {
1959                 /* we limit the length of each bio to what fits in a stripe */
1960                 *length = min_t(u64, em->len - offset,
1961                               map->stripe_len - stripe_offset);
1962         } else {
1963                 *length = em->len - offset;
1964         }
1965
1966         if (!multi_ret && !unplug_page)
1967                 goto out;
1968
1969         num_stripes = 1;
1970         stripe_index = 0;
1971         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1972                 if (unplug_page || (rw & (1 << BIO_RW)))
1973                         num_stripes = map->num_stripes;
1974                 else if (mirror_num)
1975                         stripe_index = mirror_num - 1;
1976                 else {
1977                         stripe_index = find_live_mirror(map, 0,
1978                                             map->num_stripes,
1979                                             current->pid % map->num_stripes);
1980                 }
1981
1982         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1983                 if (rw & (1 << BIO_RW))
1984                         num_stripes = map->num_stripes;
1985                 else if (mirror_num)
1986                         stripe_index = mirror_num - 1;
1987
1988         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1989                 int factor = map->num_stripes / map->sub_stripes;
1990
1991                 stripe_index = do_div(stripe_nr, factor);
1992                 stripe_index *= map->sub_stripes;
1993
1994                 if (unplug_page || (rw & (1 << BIO_RW)))
1995                         num_stripes = map->sub_stripes;
1996                 else if (mirror_num)
1997                         stripe_index += mirror_num - 1;
1998                 else {
1999                         stripe_index = find_live_mirror(map, stripe_index,
2000                                               map->sub_stripes, stripe_index +
2001                                               current->pid % map->sub_stripes);
2002                 }
2003         } else {
2004                 /*
2005                  * after this do_div call, stripe_nr is the number of stripes
2006                  * on this device we have to walk to find the data, and
2007                  * stripe_index is the number of our device in the stripe array
2008                  */
2009                 stripe_index = do_div(stripe_nr, map->num_stripes);
2010         }
2011         BUG_ON(stripe_index >= map->num_stripes);
2012
2013         for (i = 0; i < num_stripes; i++) {
2014                 if (unplug_page) {
2015                         struct btrfs_device *device;
2016                         struct backing_dev_info *bdi;
2017
2018                         device = map->stripes[stripe_index].dev;
2019                         if (device->bdev) {
2020                                 bdi = blk_get_backing_dev_info(device->bdev);
2021                                 if (bdi->unplug_io_fn) {
2022                                         bdi->unplug_io_fn(bdi, unplug_page);
2023                                 }
2024                         }
2025                 } else {
2026                         multi->stripes[i].physical =
2027                                 map->stripes[stripe_index].physical +
2028                                 stripe_offset + stripe_nr * map->stripe_len;
2029                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2030                 }
2031                 stripe_index++;
2032         }
2033         if (multi_ret) {
2034                 *multi_ret = multi;
2035                 multi->num_stripes = num_stripes;
2036                 multi->max_errors = max_errors;
2037         }
2038 out:
2039         free_extent_map(em);
2040         return 0;
2041 }
2042
2043 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2044                       u64 logical, u64 *length,
2045                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2046 {
2047         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2048                                  mirror_num, NULL);
2049 }
2050
2051 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2052                       u64 logical, struct page *page)
2053 {
2054         u64 length = PAGE_CACHE_SIZE;
2055         return __btrfs_map_block(map_tree, READ, logical, &length,
2056                                  NULL, 0, page);
2057 }
2058
2059
2060 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2061 static void end_bio_multi_stripe(struct bio *bio, int err)
2062 #else
2063 static int end_bio_multi_stripe(struct bio *bio,
2064                                    unsigned int bytes_done, int err)
2065 #endif
2066 {
2067         struct btrfs_multi_bio *multi = bio->bi_private;
2068
2069 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2070         if (bio->bi_size)
2071                 return 1;
2072 #endif
2073         if (err)
2074                 atomic_inc(&multi->error);
2075
2076         if (atomic_dec_and_test(&multi->stripes_pending)) {
2077                 bio->bi_private = multi->private;
2078                 bio->bi_end_io = multi->end_io;
2079                 /* only send an error to the higher layers if it is
2080                  * beyond the tolerance of the multi-bio
2081                  */
2082                 if (atomic_read(&multi->error) > multi->max_errors) {
2083                         err = -EIO;
2084                 } else if (err) {
2085                         /*
2086                          * this bio is actually up to date, we didn't
2087                          * go over the max number of errors
2088                          */
2089                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2090                         err = 0;
2091                 }
2092                 kfree(multi);
2093
2094 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2095                 bio_endio(bio, bio->bi_size, err);
2096 #else
2097                 bio_endio(bio, err);
2098 #endif
2099         } else {
2100                 bio_put(bio);
2101         }
2102 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2103         return 0;
2104 #endif
2105 }
2106
2107 struct async_sched {
2108         struct bio *bio;
2109         int rw;
2110         struct btrfs_fs_info *info;
2111         struct btrfs_work work;
2112 };
2113
2114 /*
2115  * see run_scheduled_bios for a description of why bios are collected for
2116  * async submit.
2117  *
2118  * This will add one bio to the pending list for a device and make sure
2119  * the work struct is scheduled.
2120  */
2121 int schedule_bio(struct btrfs_root *root, struct btrfs_device *device,
2122                  int rw, struct bio *bio)
2123 {
2124         int should_queue = 1;
2125
2126         /* don't bother with additional async steps for reads, right now */
2127         if (!(rw & (1 << BIO_RW))) {
2128                 submit_bio(rw, bio);
2129                 return 0;
2130         }
2131
2132         /*
2133          * nr_async_sumbits allows us to reliably return congestion to the
2134          * higher layers.  Otherwise, the async bio makes it appear we have
2135          * made progress against dirty pages when we've really just put it
2136          * on a queue for later
2137          */
2138         atomic_inc(&root->fs_info->nr_async_submits);
2139         bio->bi_next = NULL;
2140         bio->bi_rw |= rw;
2141
2142         spin_lock(&device->io_lock);
2143
2144         if (device->pending_bio_tail)
2145                 device->pending_bio_tail->bi_next = bio;
2146
2147         device->pending_bio_tail = bio;
2148         if (!device->pending_bios)
2149                 device->pending_bios = bio;
2150         if (device->running_pending)
2151                 should_queue = 0;
2152
2153         spin_unlock(&device->io_lock);
2154
2155         if (should_queue)
2156                 btrfs_queue_worker(&root->fs_info->submit_workers,
2157                                    &device->work);
2158         return 0;
2159 }
2160
2161 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2162                   int mirror_num, int async_submit)
2163 {
2164         struct btrfs_mapping_tree *map_tree;
2165         struct btrfs_device *dev;
2166         struct bio *first_bio = bio;
2167         u64 logical = bio->bi_sector << 9;
2168         u64 length = 0;
2169         u64 map_length;
2170         struct btrfs_multi_bio *multi = NULL;
2171         int ret;
2172         int dev_nr = 0;
2173         int total_devs = 1;
2174
2175         length = bio->bi_size;
2176         map_tree = &root->fs_info->mapping_tree;
2177         map_length = length;
2178
2179         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2180                               mirror_num);
2181         BUG_ON(ret);
2182
2183         total_devs = multi->num_stripes;
2184         if (map_length < length) {
2185                 printk("mapping failed logical %Lu bio len %Lu "
2186                        "len %Lu\n", logical, length, map_length);
2187                 BUG();
2188         }
2189         multi->end_io = first_bio->bi_end_io;
2190         multi->private = first_bio->bi_private;
2191         atomic_set(&multi->stripes_pending, multi->num_stripes);
2192
2193         while(dev_nr < total_devs) {
2194                 if (total_devs > 1) {
2195                         if (dev_nr < total_devs - 1) {
2196                                 bio = bio_clone(first_bio, GFP_NOFS);
2197                                 BUG_ON(!bio);
2198                         } else {
2199                                 bio = first_bio;
2200                         }
2201                         bio->bi_private = multi;
2202                         bio->bi_end_io = end_bio_multi_stripe;
2203                 }
2204                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2205                 dev = multi->stripes[dev_nr].dev;
2206                 if (dev && dev->bdev) {
2207                         bio->bi_bdev = dev->bdev;
2208                         if (async_submit)
2209                                 schedule_bio(root, dev, rw, bio);
2210                         else
2211                                 submit_bio(rw, bio);
2212                 } else {
2213                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2214                         bio->bi_sector = logical >> 9;
2215 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2216                         bio_endio(bio, bio->bi_size, -EIO);
2217 #else
2218                         bio_endio(bio, -EIO);
2219 #endif
2220                 }
2221                 dev_nr++;
2222         }
2223         if (total_devs == 1)
2224                 kfree(multi);
2225         return 0;
2226 }
2227
2228 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2229                                        u8 *uuid)
2230 {
2231         struct list_head *head = &root->fs_info->fs_devices->devices;
2232
2233         return __find_device(head, devid, uuid);
2234 }
2235
2236 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2237                                             u64 devid, u8 *dev_uuid)
2238 {
2239         struct btrfs_device *device;
2240         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2241
2242         device = kzalloc(sizeof(*device), GFP_NOFS);
2243         list_add(&device->dev_list,
2244                  &fs_devices->devices);
2245         list_add(&device->dev_alloc_list,
2246                  &fs_devices->alloc_list);
2247         device->barriers = 1;
2248         device->dev_root = root->fs_info->dev_root;
2249         device->devid = devid;
2250         device->work.func = pending_bios_fn;
2251         fs_devices->num_devices++;
2252         spin_lock_init(&device->io_lock);
2253         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2254         return device;
2255 }
2256
2257
2258 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2259                           struct extent_buffer *leaf,
2260                           struct btrfs_chunk *chunk)
2261 {
2262         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2263         struct map_lookup *map;
2264         struct extent_map *em;
2265         u64 logical;
2266         u64 length;
2267         u64 devid;
2268         u8 uuid[BTRFS_UUID_SIZE];
2269         int num_stripes;
2270         int ret;
2271         int i;
2272
2273         logical = key->offset;
2274         length = btrfs_chunk_length(leaf, chunk);
2275
2276         spin_lock(&map_tree->map_tree.lock);
2277         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2278         spin_unlock(&map_tree->map_tree.lock);
2279
2280         /* already mapped? */
2281         if (em && em->start <= logical && em->start + em->len > logical) {
2282                 free_extent_map(em);
2283                 return 0;
2284         } else if (em) {
2285                 free_extent_map(em);
2286         }
2287
2288         map = kzalloc(sizeof(*map), GFP_NOFS);
2289         if (!map)
2290                 return -ENOMEM;
2291
2292         em = alloc_extent_map(GFP_NOFS);
2293         if (!em)
2294                 return -ENOMEM;
2295         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2296         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2297         if (!map) {
2298                 free_extent_map(em);
2299                 return -ENOMEM;
2300         }
2301
2302         em->bdev = (struct block_device *)map;
2303         em->start = logical;
2304         em->len = length;
2305         em->block_start = 0;
2306
2307         map->num_stripes = num_stripes;
2308         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2309         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2310         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2311         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2312         map->type = btrfs_chunk_type(leaf, chunk);
2313         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2314         for (i = 0; i < num_stripes; i++) {
2315                 map->stripes[i].physical =
2316                         btrfs_stripe_offset_nr(leaf, chunk, i);
2317                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2318                 read_extent_buffer(leaf, uuid, (unsigned long)
2319                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2320                                    BTRFS_UUID_SIZE);
2321                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2322
2323                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2324                         kfree(map);
2325                         free_extent_map(em);
2326                         return -EIO;
2327                 }
2328                 if (!map->stripes[i].dev) {
2329                         map->stripes[i].dev =
2330                                 add_missing_dev(root, devid, uuid);
2331                         if (!map->stripes[i].dev) {
2332                                 kfree(map);
2333                                 free_extent_map(em);
2334                                 return -EIO;
2335                         }
2336                 }
2337                 map->stripes[i].dev->in_fs_metadata = 1;
2338         }
2339
2340         spin_lock(&map_tree->map_tree.lock);
2341         ret = add_extent_mapping(&map_tree->map_tree, em);
2342         spin_unlock(&map_tree->map_tree.lock);
2343         BUG_ON(ret);
2344         free_extent_map(em);
2345
2346         return 0;
2347 }
2348
2349 static int fill_device_from_item(struct extent_buffer *leaf,
2350                                  struct btrfs_dev_item *dev_item,
2351                                  struct btrfs_device *device)
2352 {
2353         unsigned long ptr;
2354
2355         device->devid = btrfs_device_id(leaf, dev_item);
2356         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2357         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2358         device->type = btrfs_device_type(leaf, dev_item);
2359         device->io_align = btrfs_device_io_align(leaf, dev_item);
2360         device->io_width = btrfs_device_io_width(leaf, dev_item);
2361         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2362
2363         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2364         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2365
2366         return 0;
2367 }
2368
2369 static int read_one_dev(struct btrfs_root *root,
2370                         struct extent_buffer *leaf,
2371                         struct btrfs_dev_item *dev_item)
2372 {
2373         struct btrfs_device *device;
2374         u64 devid;
2375         int ret;
2376         u8 dev_uuid[BTRFS_UUID_SIZE];
2377
2378         devid = btrfs_device_id(leaf, dev_item);
2379         read_extent_buffer(leaf, dev_uuid,
2380                            (unsigned long)btrfs_device_uuid(dev_item),
2381                            BTRFS_UUID_SIZE);
2382         device = btrfs_find_device(root, devid, dev_uuid);
2383         if (!device) {
2384                 printk("warning devid %Lu missing\n", devid);
2385                 device = add_missing_dev(root, devid, dev_uuid);
2386                 if (!device)
2387                         return -ENOMEM;
2388         }
2389
2390         fill_device_from_item(leaf, dev_item, device);
2391         device->dev_root = root->fs_info->dev_root;
2392         device->in_fs_metadata = 1;
2393         ret = 0;
2394 #if 0
2395         ret = btrfs_open_device(device);
2396         if (ret) {
2397                 kfree(device);
2398         }
2399 #endif
2400         return ret;
2401 }
2402
2403 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2404 {
2405         struct btrfs_dev_item *dev_item;
2406
2407         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2408                                                      dev_item);
2409         return read_one_dev(root, buf, dev_item);
2410 }
2411
2412 int btrfs_read_sys_array(struct btrfs_root *root)
2413 {
2414         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2415         struct extent_buffer *sb;
2416         struct btrfs_disk_key *disk_key;
2417         struct btrfs_chunk *chunk;
2418         u8 *ptr;
2419         unsigned long sb_ptr;
2420         int ret = 0;
2421         u32 num_stripes;
2422         u32 array_size;
2423         u32 len = 0;
2424         u32 cur;
2425         struct btrfs_key key;
2426
2427         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2428                                           BTRFS_SUPER_INFO_SIZE);
2429         if (!sb)
2430                 return -ENOMEM;
2431         btrfs_set_buffer_uptodate(sb);
2432         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2433         array_size = btrfs_super_sys_array_size(super_copy);
2434
2435         ptr = super_copy->sys_chunk_array;
2436         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2437         cur = 0;
2438
2439         while (cur < array_size) {
2440                 disk_key = (struct btrfs_disk_key *)ptr;
2441                 btrfs_disk_key_to_cpu(&key, disk_key);
2442
2443                 len = sizeof(*disk_key); ptr += len;
2444                 sb_ptr += len;
2445                 cur += len;
2446
2447                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2448                         chunk = (struct btrfs_chunk *)sb_ptr;
2449                         ret = read_one_chunk(root, &key, sb, chunk);
2450                         if (ret)
2451                                 break;
2452                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2453                         len = btrfs_chunk_item_size(num_stripes);
2454                 } else {
2455                         ret = -EIO;
2456                         break;
2457                 }
2458                 ptr += len;
2459                 sb_ptr += len;
2460                 cur += len;
2461         }
2462         free_extent_buffer(sb);
2463         return ret;
2464 }
2465
2466 int btrfs_read_chunk_tree(struct btrfs_root *root)
2467 {
2468         struct btrfs_path *path;
2469         struct extent_buffer *leaf;
2470         struct btrfs_key key;
2471         struct btrfs_key found_key;
2472         int ret;
2473         int slot;
2474
2475         root = root->fs_info->chunk_root;
2476
2477         path = btrfs_alloc_path();
2478         if (!path)
2479                 return -ENOMEM;
2480
2481         /* first we search for all of the device items, and then we
2482          * read in all of the chunk items.  This way we can create chunk
2483          * mappings that reference all of the devices that are afound
2484          */
2485         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2486         key.offset = 0;
2487         key.type = 0;
2488 again:
2489         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2490         while(1) {
2491                 leaf = path->nodes[0];
2492                 slot = path->slots[0];
2493                 if (slot >= btrfs_header_nritems(leaf)) {
2494                         ret = btrfs_next_leaf(root, path);
2495                         if (ret == 0)
2496                                 continue;
2497                         if (ret < 0)
2498                                 goto error;
2499                         break;
2500                 }
2501                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2502                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2503                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2504                                 break;
2505                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2506                                 struct btrfs_dev_item *dev_item;
2507                                 dev_item = btrfs_item_ptr(leaf, slot,
2508                                                   struct btrfs_dev_item);
2509                                 ret = read_one_dev(root, leaf, dev_item);
2510                                 BUG_ON(ret);
2511                         }
2512                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2513                         struct btrfs_chunk *chunk;
2514                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2515                         ret = read_one_chunk(root, &found_key, leaf, chunk);
2516                 }
2517                 path->slots[0]++;
2518         }
2519         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2520                 key.objectid = 0;
2521                 btrfs_release_path(root, path);
2522                 goto again;
2523         }
2524
2525         btrfs_free_path(path);
2526         ret = 0;
2527 error:
2528         return ret;
2529 }