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 <linux/version.h>
  24 #include <asm/div64.h>
  25 #include "compat.h"
  26 #include "ctree.h"
  27 #include "extent_map.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "print-tree.h"
  31 #include "volumes.h"
  32 #include "async-thread.h"
  33
  34 struct map_lookup {
  35         u64 type;
  36         int io_align;
  37         int io_width;
  38         int stripe_len;
  39         int sector_size;
  40         int num_stripes;
  41         int sub_stripes;
  42         struct btrfs_bio_stripe stripes[];
  43 };
  44
  45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  46                                 struct btrfs_root *root,
  47                                 struct btrfs_device *device);
  48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  49
  50
  51 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  52                             (sizeof(struct btrfs_bio_stripe) * (n)))
  53
  54 static DEFINE_MUTEX(uuid_mutex);
  55 static LIST_HEAD(fs_uuids);
  56
  57 void btrfs_lock_volumes(void)
  58 {
  59         mutex_lock(&uuid_mutex);
  60 }
  61
  62 void btrfs_unlock_volumes(void)
  63 {
  64         mutex_unlock(&uuid_mutex);
  65 }
  66
  67 static void lock_chunks(struct btrfs_root *root)
  68 {
  69         mutex_lock(&root->fs_info->chunk_mutex);
  70 }
  71
  72 static void unlock_chunks(struct btrfs_root *root)
  73 {
  74         mutex_unlock(&root->fs_info->chunk_mutex);
  75 }
  76
  77 int btrfs_cleanup_fs_uuids(void)
  78 {
  79         struct btrfs_fs_devices *fs_devices;
  80         struct btrfs_device *dev;
  81
  82         while (!list_empty(&fs_uuids)) {
  83                 fs_devices = list_entry(fs_uuids.next,
  84                                         struct btrfs_fs_devices, list);
  85                 list_del(&fs_devices->list);
  86                 while(!list_empty(&fs_devices->devices)) {
  87                         dev = list_entry(fs_devices->devices.next,
  88                                          struct btrfs_device, dev_list);
  89                         if (dev->bdev) {
  90                                 close_bdev_exclusive(dev->bdev, dev->mode);
  91                                 fs_devices->open_devices--;
  92                         }
  93                         fs_devices->num_devices--;
  94                         if (dev->writeable)
  95                                 fs_devices->rw_devices--;
  96                         list_del(&dev->dev_list);
  97                         list_del(&dev->dev_alloc_list);
  98                         kfree(dev->name);
  99                         kfree(dev);
 100                 }
 101                 WARN_ON(fs_devices->num_devices);
 102                 WARN_ON(fs_devices->open_devices);
 103                 WARN_ON(fs_devices->rw_devices);
 104                 kfree(fs_devices);
 105         }
 106         return 0;
 107 }
 108
 109 static noinline struct btrfs_device *__find_device(struct list_head *head,
 110                                                    u64 devid, u8 *uuid)
 111 {
 112         struct btrfs_device *dev;
 113         struct list_head *cur;
 114
 115         list_for_each(cur, head) {
 116                 dev = list_entry(cur, struct btrfs_device, dev_list);
 117                 if (dev->devid == devid &&
 118                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 119                         return dev;
 120                 }
 121         }
 122         return NULL;
 123 }
 124
 125 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 126 {
 127         struct list_head *cur;
 128         struct btrfs_fs_devices *fs_devices;
 129
 130         list_for_each(cur, &fs_uuids) {
 131                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
 132                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 133                         return fs_devices;
 134         }
 135         return NULL;
 136 }
 137
 138 /*
 139  * we try to collect pending bios for a device so we don't get a large
 140  * number of procs sending bios down to the same device.  This greatly
 141  * improves the schedulers ability to collect and merge the bios.
 142  *
 143  * But, it also turns into a long list of bios to process and that is sure
 144  * to eventually make the worker thread block.  The solution here is to
 145  * make some progress and then put this work struct back at the end of
 146  * the list if the block device is congested.  This way, multiple devices
 147  * can make progress from a single worker thread.
 148  */
 149 static int noinline run_scheduled_bios(struct btrfs_device *device)
 150 {
 151         struct bio *pending;
 152         struct backing_dev_info *bdi;
 153         struct btrfs_fs_info *fs_info;
 154         struct bio *tail;
 155         struct bio *cur;
 156         int again = 0;
 157         unsigned long num_run = 0;
 158         unsigned long limit;
 159
 160         bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
 161         fs_info = device->dev_root->fs_info;
 162         limit = btrfs_async_submit_limit(fs_info);
 163         limit = limit * 2 / 3;
 164
 165 loop:
 166         spin_lock(&device->io_lock);
 167
 168         /* take all the bios off the list at once and process them
 169          * later on (without the lock held).  But, remember the
 170          * tail and other pointers so the bios can be properly reinserted
 171          * into the list if we hit congestion
 172          */
 173         pending = device->pending_bios;
 174         tail = device->pending_bio_tail;
 175         WARN_ON(pending && !tail);
 176         device->pending_bios = NULL;
 177         device->pending_bio_tail = NULL;
 178
 179         /*
 180          * if pending was null this time around, no bios need processing
 181          * at all and we can stop.  Otherwise it'll loop back up again
 182          * and do an additional check so no bios are missed.
 183          *
 184          * device->running_pending is used to synchronize with the
 185          * schedule_bio code.
 186          */
 187         if (pending) {
 188                 again = 1;
 189                 device->running_pending = 1;
 190         } else {
 191                 again = 0;
 192                 device->running_pending = 0;
 193         }
 194         spin_unlock(&device->io_lock);
 195
 196         while(pending) {
 197                 cur = pending;
 198                 pending = pending->bi_next;
 199                 cur->bi_next = NULL;
 200                 atomic_dec(&fs_info->nr_async_bios);
 201
 202                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 203                     waitqueue_active(&fs_info->async_submit_wait))
 204                         wake_up(&fs_info->async_submit_wait);
 205
 206                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 207                 bio_get(cur);
 208                 submit_bio(cur->bi_rw, cur);
 209                 bio_put(cur);
 210                 num_run++;
 211
 212                 /*
 213                  * we made progress, there is more work to do and the bdi
 214                  * is now congested.  Back off and let other work structs
 215                  * run instead
 216                  */
 217                 if (pending && bdi_write_congested(bdi) &&
 218                     fs_info->fs_devices->open_devices > 1) {
 219                         struct bio *old_head;
 220
 221                         spin_lock(&device->io_lock);
 222
 223                         old_head = device->pending_bios;
 224                         device->pending_bios = pending;
 225                         if (device->pending_bio_tail)
 226                                 tail->bi_next = old_head;
 227                         else
 228                                 device->pending_bio_tail = tail;
 229
 230                         spin_unlock(&device->io_lock);
 231                         btrfs_requeue_work(&device->work);
 232                         goto done;
 233                 }
 234         }
 235         if (again)
 236                 goto loop;
 237 done:
 238         return 0;
 239 }
 240
 241 static void pending_bios_fn(struct btrfs_work *work)
 242 {
 243         struct btrfs_device *device;
 244
 245         device = container_of(work, struct btrfs_device, work);
 246         run_scheduled_bios(device);
 247 }
 248
 249 static noinline int device_list_add(const char *path,
 250                            struct btrfs_super_block *disk_super,
 251                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 252 {
 253         struct btrfs_device *device;
 254         struct btrfs_fs_devices *fs_devices;
 255         u64 found_transid = btrfs_super_generation(disk_super);
 256
 257         fs_devices = find_fsid(disk_super->fsid);
 258         if (!fs_devices) {
 259                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 260                 if (!fs_devices)
 261                         return -ENOMEM;
 262                 INIT_LIST_HEAD(&fs_devices->devices);
 263                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 264                 list_add(&fs_devices->list, &fs_uuids);
 265                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 266                 fs_devices->latest_devid = devid;
 267                 fs_devices->latest_trans = found_transid;
 268                 device = NULL;
 269         } else {
 270                 device = __find_device(&fs_devices->devices, devid,
 271                                        disk_super->dev_item.uuid);
 272         }
 273         if (!device) {
 274                 if (fs_devices->opened)
 275                         return -EBUSY;
 276
 277                 device = kzalloc(sizeof(*device), GFP_NOFS);
 278                 if (!device) {
 279                         /* we can safely leave the fs_devices entry around */
 280                         return -ENOMEM;
 281                 }
 282                 device->devid = devid;
 283                 device->work.func = pending_bios_fn;
 284                 memcpy(device->uuid, disk_super->dev_item.uuid,
 285                        BTRFS_UUID_SIZE);
 286                 device->barriers = 1;
 287                 spin_lock_init(&device->io_lock);
 288                 device->name = kstrdup(path, GFP_NOFS);
 289                 if (!device->name) {
 290                         kfree(device);
 291                         return -ENOMEM;
 292                 }
 293                 INIT_LIST_HEAD(&device->dev_alloc_list);
 294                 list_add(&device->dev_list, &fs_devices->devices);
 295                 device->fs_devices = fs_devices;
 296                 fs_devices->num_devices++;
 297         }
 298
 299         if (found_transid > fs_devices->latest_trans) {
 300                 fs_devices->latest_devid = devid;
 301                 fs_devices->latest_trans = found_transid;
 302         }
 303         *fs_devices_ret = fs_devices;
 304         return 0;
 305 }
 306
 307 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 308 {
 309         struct list_head *tmp;
 310         struct list_head *cur;
 311         struct btrfs_device *device;
 312         int seed_devices = 0;
 313
 314         mutex_lock(&uuid_mutex);
 315 again:
 316         list_for_each_safe(cur, tmp, &fs_devices->devices) {
 317                 device = list_entry(cur, struct btrfs_device, dev_list);
 318                 if (device->in_fs_metadata)
 319                         continue;
 320
 321                 if (device->bdev) {
 322                         close_bdev_exclusive(device->bdev, device->mode);
 323                         device->bdev = NULL;
 324                         fs_devices->open_devices--;
 325                 }
 326                 if (device->writeable) {
 327                         list_del_init(&device->dev_alloc_list);
 328                         device->writeable = 0;
 329                         fs_devices->rw_devices--;
 330                 }
 331                 if (!seed_devices) {
 332                         list_del_init(&device->dev_list);
 333                         fs_devices->num_devices--;
 334                         kfree(device->name);
 335                         kfree(device);
 336                 }
 337         }
 338
 339         if (fs_devices->seed) {
 340                 fs_devices = fs_devices->seed;
 341                 seed_devices = 1;
 342                 goto again;
 343         }
 344
 345         mutex_unlock(&uuid_mutex);
 346         return 0;
 347 }
 348
 349 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 350 {
 351         struct btrfs_fs_devices *seed_devices;
 352         struct list_head *cur;
 353         struct btrfs_device *device;
 354 again:
 355         if (--fs_devices->opened > 0)
 356                 return 0;
 357
 358         list_for_each(cur, &fs_devices->devices) {
 359                 device = list_entry(cur, struct btrfs_device, dev_list);
 360                 if (device->bdev) {
 361                         close_bdev_exclusive(device->bdev, device->mode);
 362                         fs_devices->open_devices--;
 363                 }
 364                 if (device->writeable) {
 365                         list_del_init(&device->dev_alloc_list);
 366                         fs_devices->rw_devices--;
 367                 }
 368
 369                 device->bdev = NULL;
 370                 device->writeable = 0;
 371                 device->in_fs_metadata = 0;
 372         }
 373         fs_devices->opened = 0;
 374         fs_devices->seeding = 0;
 375         fs_devices->sprouted = 0;
 376
 377         seed_devices = fs_devices->seed;
 378         fs_devices->seed = NULL;
 379         if (seed_devices) {
 380                 fs_devices = seed_devices;
 381                 goto again;
 382         }
 383         return 0;
 384 }
 385
 386 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 387 {
 388         int ret;
 389
 390         mutex_lock(&uuid_mutex);
 391         ret = __btrfs_close_devices(fs_devices);
 392         mutex_unlock(&uuid_mutex);
 393         return ret;
 394 }
 395
 396 int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 397                          fmode_t flags, void *holder)
 398 {
 399         struct block_device *bdev;
 400         struct list_head *head = &fs_devices->devices;
 401         struct list_head *cur;
 402         struct btrfs_device *device;
 403         struct block_device *latest_bdev = NULL;
 404         struct buffer_head *bh;
 405         struct btrfs_super_block *disk_super;
 406         u64 latest_devid = 0;
 407         u64 latest_transid = 0;
 408         u64 devid;
 409         int seeding = 1;
 410         int ret = 0;
 411
 412         list_for_each(cur, head) {
 413                 device = list_entry(cur, struct btrfs_device, dev_list);
 414                 if (device->bdev)
 415                         continue;
 416                 if (!device->name)
 417                         continue;
 418
 419                 bdev = open_bdev_exclusive(device->name, flags, holder);
 420                 if (IS_ERR(bdev)) {
 421                         printk("open %s failed\n", device->name);
 422                         goto error;
 423                 }
 424                 set_blocksize(bdev, 4096);
 425
 426                 bh = btrfs_read_dev_super(bdev);
 427                 if (!bh)
 428                         goto error_close;
 429
 430                 disk_super = (struct btrfs_super_block *)bh->b_data;
 431                 devid = le64_to_cpu(disk_super->dev_item.devid);
 432                 if (devid != device->devid)
 433                         goto error_brelse;
 434
 435                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 436                            BTRFS_UUID_SIZE))
 437                         goto error_brelse;
 438
 439                 device->generation = btrfs_super_generation(disk_super);
 440                 if (!latest_transid || device->generation > latest_transid) {
 441                         latest_devid = devid;
 442                         latest_transid = device->generation;
 443                         latest_bdev = bdev;
 444                 }
 445
 446                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 447                         device->writeable = 0;
 448                 } else {
 449                         device->writeable = !bdev_read_only(bdev);
 450                         seeding = 0;
 451                 }
 452
 453                 device->bdev = bdev;
 454                 device->in_fs_metadata = 0;
 455                 device->mode = flags;
 456
 457                 fs_devices->open_devices++;
 458                 if (device->writeable) {
 459                         fs_devices->rw_devices++;
 460                         list_add(&device->dev_alloc_list,
 461                                  &fs_devices->alloc_list);
 462                 }
 463                 continue;
 464
 465 error_brelse:
 466                 brelse(bh);
 467 error_close:
 468                 close_bdev_exclusive(bdev, FMODE_READ);
 469 error:
 470                 continue;
 471         }
 472         if (fs_devices->open_devices == 0) {
 473                 ret = -EIO;
 474                 goto out;
 475         }
 476         fs_devices->seeding = seeding;
 477         fs_devices->opened = 1;
 478         fs_devices->latest_bdev = latest_bdev;
 479         fs_devices->latest_devid = latest_devid;
 480         fs_devices->latest_trans = latest_transid;
 481         fs_devices->total_rw_bytes = 0;
 482 out:
 483         return ret;
 484 }
 485
 486 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 487                        fmode_t flags, void *holder)
 488 {
 489         int ret;
 490
 491         mutex_lock(&uuid_mutex);
 492         if (fs_devices->opened) {
 493                 if (fs_devices->sprouted) {
 494                         ret = -EBUSY;
 495                 } else {
 496                         fs_devices->opened++;
 497                         ret = 0;
 498                 }
 499         } else {
 500                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 501         }
 502         mutex_unlock(&uuid_mutex);
 503         return ret;
 504 }
 505
 506 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 507                           struct btrfs_fs_devices **fs_devices_ret)
 508 {
 509         struct btrfs_super_block *disk_super;
 510         struct block_device *bdev;
 511         struct buffer_head *bh;
 512         int ret;
 513         u64 devid;
 514         u64 transid;
 515
 516         mutex_lock(&uuid_mutex);
 517
 518         bdev = open_bdev_exclusive(path, flags, holder);
 519
 520         if (IS_ERR(bdev)) {
 521                 ret = PTR_ERR(bdev);
 522                 goto error;
 523         }
 524
 525         ret = set_blocksize(bdev, 4096);
 526         if (ret)
 527                 goto error_close;
 528         bh = btrfs_read_dev_super(bdev);
 529         if (!bh) {
 530                 ret = -EIO;
 531                 goto error_close;
 532         }
 533         disk_super = (struct btrfs_super_block *)bh->b_data;
 534         devid = le64_to_cpu(disk_super->dev_item.devid);
 535         transid = btrfs_super_generation(disk_super);
 536         if (disk_super->label[0])
 537                 printk("device label %s ", disk_super->label);
 538         else {
 539                 /* FIXME, make a readl uuid parser */
 540                 printk("device fsid %llx-%llx ",
 541                        *(unsigned long long *)disk_super->fsid,
 542                        *(unsigned long long *)(disk_super->fsid + 8));
 543         }
 544         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 545         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 546
 547         brelse(bh);
 548 error_close:
 549         close_bdev_exclusive(bdev, flags);
 550 error:
 551         mutex_unlock(&uuid_mutex);
 552         return ret;
 553 }
 554
 555 /*
 556  * this uses a pretty simple search, the expectation is that it is
 557  * called very infrequently and that a given device has a small number
 558  * of extents
 559  */
 560 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
 561                                          struct btrfs_device *device,
 562                                          u64 num_bytes, u64 *start)
 563 {
 564         struct btrfs_key key;
 565         struct btrfs_root *root = device->dev_root;
 566         struct btrfs_dev_extent *dev_extent = NULL;
 567         struct btrfs_path *path;
 568         u64 hole_size = 0;
 569         u64 last_byte = 0;
 570         u64 search_start = 0;
 571         u64 search_end = device->total_bytes;
 572         int ret;
 573         int slot = 0;
 574         int start_found;
 575         struct extent_buffer *l;
 576
 577         path = btrfs_alloc_path();
 578         if (!path)
 579                 return -ENOMEM;
 580         path->reada = 2;
 581         start_found = 0;
 582
 583         /* FIXME use last free of some kind */
 584
 585         /* we don't want to overwrite the superblock on the drive,
 586          * so we make sure to start at an offset of at least 1MB
 587          */
 588         search_start = max((u64)1024 * 1024, search_start);
 589
 590         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 591                 search_start = max(root->fs_info->alloc_start, search_start);
 592
 593         key.objectid = device->devid;
 594         key.offset = search_start;
 595         key.type = BTRFS_DEV_EXTENT_KEY;
 596         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 597         if (ret < 0)
 598                 goto error;
 599         ret = btrfs_previous_item(root, path, 0, key.type);
 600         if (ret < 0)
 601                 goto error;
 602         l = path->nodes[0];
 603         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 604         while (1) {
 605                 l = path->nodes[0];
 606                 slot = path->slots[0];
 607                 if (slot >= btrfs_header_nritems(l)) {
 608                         ret = btrfs_next_leaf(root, path);
 609                         if (ret == 0)
 610                                 continue;
 611                         if (ret < 0)
 612                                 goto error;
 613 no_more_items:
 614                         if (!start_found) {
 615                                 if (search_start >= search_end) {
 616                                         ret = -ENOSPC;
 617                                         goto error;
 618                                 }
 619                                 *start = search_start;
 620                                 start_found = 1;
 621                                 goto check_pending;
 622                         }
 623                         *start = last_byte > search_start ?
 624                                 last_byte : search_start;
 625                         if (search_end <= *start) {
 626                                 ret = -ENOSPC;
 627                                 goto error;
 628                         }
 629                         goto check_pending;
 630                 }
 631                 btrfs_item_key_to_cpu(l, &key, slot);
 632
 633                 if (key.objectid < device->devid)
 634                         goto next;
 635
 636                 if (key.objectid > device->devid)
 637                         goto no_more_items;
 638
 639                 if (key.offset >= search_start && key.offset > last_byte &&
 640                     start_found) {
 641                         if (last_byte < search_start)
 642                                 last_byte = search_start;
 643                         hole_size = key.offset - last_byte;
 644                         if (key.offset > last_byte &&
 645                             hole_size >= num_bytes) {
 646                                 *start = last_byte;
 647                                 goto check_pending;
 648                         }
 649                 }
 650                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 651                         goto next;
 652                 }
 653
 654                 start_found = 1;
 655                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 656                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 657 next:
 658                 path->slots[0]++;
 659                 cond_resched();
 660         }
 661 check_pending:
 662         /* we have to make sure we didn't find an extent that has already
 663          * been allocated by the map tree or the original allocation
 664          */
 665         BUG_ON(*start < search_start);
 666
 667         if (*start + num_bytes > search_end) {
 668                 ret = -ENOSPC;
 669                 goto error;
 670         }
 671         /* check for pending inserts here */
 672         ret = 0;
 673
 674 error:
 675         btrfs_free_path(path);
 676         return ret;
 677 }
 678
 679 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 680                           struct btrfs_device *device,
 681                           u64 start)
 682 {
 683         int ret;
 684         struct btrfs_path *path;
 685         struct btrfs_root *root = device->dev_root;
 686         struct btrfs_key key;
 687         struct btrfs_key found_key;
 688         struct extent_buffer *leaf = NULL;
 689         struct btrfs_dev_extent *extent = NULL;
 690
 691         path = btrfs_alloc_path();
 692         if (!path)
 693                 return -ENOMEM;
 694
 695         key.objectid = device->devid;
 696         key.offset = start;
 697         key.type = BTRFS_DEV_EXTENT_KEY;
 698
 699         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 700         if (ret > 0) {
 701                 ret = btrfs_previous_item(root, path, key.objectid,
 702                                           BTRFS_DEV_EXTENT_KEY);
 703                 BUG_ON(ret);
 704                 leaf = path->nodes[0];
 705                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 706                 extent = btrfs_item_ptr(leaf, path->slots[0],
 707                                         struct btrfs_dev_extent);
 708                 BUG_ON(found_key.offset > start || found_key.offset +
 709                        btrfs_dev_extent_length(leaf, extent) < start);
 710                 ret = 0;
 711         } else if (ret == 0) {
 712                 leaf = path->nodes[0];
 713                 extent = btrfs_item_ptr(leaf, path->slots[0],
 714                                         struct btrfs_dev_extent);
 715         }
 716         BUG_ON(ret);
 717
 718         if (device->bytes_used > 0)
 719                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 720         ret = btrfs_del_item(trans, root, path);
 721         BUG_ON(ret);
 722
 723         btrfs_free_path(path);
 724         return ret;
 725 }
 726
 727 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 728                            struct btrfs_device *device,
 729                            u64 chunk_tree, u64 chunk_objectid,
 730                            u64 chunk_offset, u64 start, u64 num_bytes)
 731 {
 732         int ret;
 733         struct btrfs_path *path;
 734         struct btrfs_root *root = device->dev_root;
 735         struct btrfs_dev_extent *extent;
 736         struct extent_buffer *leaf;
 737         struct btrfs_key key;
 738
 739         WARN_ON(!device->in_fs_metadata);
 740         path = btrfs_alloc_path();
 741         if (!path)
 742                 return -ENOMEM;
 743
 744         key.objectid = device->devid;
 745         key.offset = start;
 746         key.type = BTRFS_DEV_EXTENT_KEY;
 747         ret = btrfs_insert_empty_item(trans, root, path, &key,
 748                                       sizeof(*extent));
 749         BUG_ON(ret);
 750
 751         leaf = path->nodes[0];
 752         extent = btrfs_item_ptr(leaf, path->slots[0],
 753                                 struct btrfs_dev_extent);
 754         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 755         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 756         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 757
 758         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 759                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 760                     BTRFS_UUID_SIZE);
 761
 762         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 763         btrfs_mark_buffer_dirty(leaf);
 764         btrfs_free_path(path);
 765         return ret;
 766 }
 767
 768 static noinline int find_next_chunk(struct btrfs_root *root,
 769                                     u64 objectid, u64 *offset)
 770 {
 771         struct btrfs_path *path;
 772         int ret;
 773         struct btrfs_key key;
 774         struct btrfs_chunk *chunk;
 775         struct btrfs_key found_key;
 776
 777         path = btrfs_alloc_path();
 778         BUG_ON(!path);
 779
 780         key.objectid = objectid;
 781         key.offset = (u64)-1;
 782         key.type = BTRFS_CHUNK_ITEM_KEY;
 783
 784         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 785         if (ret < 0)
 786                 goto error;
 787
 788         BUG_ON(ret == 0);
 789
 790         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 791         if (ret) {
 792                 *offset = 0;
 793         } else {
 794                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 795                                       path->slots[0]);
 796                 if (found_key.objectid != objectid)
 797                         *offset = 0;
 798                 else {
 799                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 800                                                struct btrfs_chunk);
 801                         *offset = found_key.offset +
 802                                 btrfs_chunk_length(path->nodes[0], chunk);
 803                 }
 804         }
 805         ret = 0;
 806 error:
 807         btrfs_free_path(path);
 808         return ret;
 809 }
 810
 811 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
 812 {
 813         int ret;
 814         struct btrfs_key key;
 815         struct btrfs_key found_key;
 816         struct btrfs_path *path;
 817
 818         root = root->fs_info->chunk_root;
 819
 820         path = btrfs_alloc_path();
 821         if (!path)
 822                 return -ENOMEM;
 823
 824         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 825         key.type = BTRFS_DEV_ITEM_KEY;
 826         key.offset = (u64)-1;
 827
 828         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 829         if (ret < 0)
 830                 goto error;
 831
 832         BUG_ON(ret == 0);
 833
 834         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 835                                   BTRFS_DEV_ITEM_KEY);
 836         if (ret) {
 837                 *objectid = 1;
 838         } else {
 839                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 840                                       path->slots[0]);
 841                 *objectid = found_key.offset + 1;
 842         }
 843         ret = 0;
 844 error:
 845         btrfs_free_path(path);
 846         return ret;
 847 }
 848
 849 /*
 850  * the device information is stored in the chunk root
 851  * the btrfs_device struct should be fully filled in
 852  */
 853 int btrfs_add_device(struct btrfs_trans_handle *trans,
 854                      struct btrfs_root *root,
 855                      struct btrfs_device *device)
 856 {
 857         int ret;
 858         struct btrfs_path *path;
 859         struct btrfs_dev_item *dev_item;
 860         struct extent_buffer *leaf;
 861         struct btrfs_key key;
 862         unsigned long ptr;
 863
 864         root = root->fs_info->chunk_root;
 865
 866         path = btrfs_alloc_path();
 867         if (!path)
 868                 return -ENOMEM;
 869
 870         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 871         key.type = BTRFS_DEV_ITEM_KEY;
 872         key.offset = device->devid;
 873
 874         ret = btrfs_insert_empty_item(trans, root, path, &key,
 875                                       sizeof(*dev_item));
 876         if (ret)
 877                 goto out;
 878
 879         leaf = path->nodes[0];
 880         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 881
 882         btrfs_set_device_id(leaf, dev_item, device->devid);
 883         btrfs_set_device_generation(leaf, dev_item, 0);
 884         btrfs_set_device_type(leaf, dev_item, device->type);
 885         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 886         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 887         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 888         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 889         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 890         btrfs_set_device_group(leaf, dev_item, 0);
 891         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 892         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 893         btrfs_set_device_start_offset(leaf, dev_item, 0);
 894
 895         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 896         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 897         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 898         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 899         btrfs_mark_buffer_dirty(leaf);
 900
 901         ret = 0;
 902 out:
 903         btrfs_free_path(path);
 904         return ret;
 905 }
 906
 907 static int btrfs_rm_dev_item(struct btrfs_root *root,
 908                              struct btrfs_device *device)
 909 {
 910         int ret;
 911         struct btrfs_path *path;
 912         struct btrfs_key key;
 913         struct btrfs_trans_handle *trans;
 914
 915         root = root->fs_info->chunk_root;
 916
 917         path = btrfs_alloc_path();
 918         if (!path)
 919                 return -ENOMEM;
 920
 921         trans = btrfs_start_transaction(root, 1);
 922         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 923         key.type = BTRFS_DEV_ITEM_KEY;
 924         key.offset = device->devid;
 925         lock_chunks(root);
 926
 927         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 928         if (ret < 0)
 929                 goto out;
 930
 931         if (ret > 0) {
 932                 ret = -ENOENT;
 933                 goto out;
 934         }
 935
 936         ret = btrfs_del_item(trans, root, path);
 937         if (ret)
 938                 goto out;
 939 out:
 940         btrfs_free_path(path);
 941         unlock_chunks(root);
 942         btrfs_commit_transaction(trans, root);
 943         return ret;
 944 }
 945
 946 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 947 {
 948         struct btrfs_device *device;
 949         struct btrfs_device *next_device;
 950         struct block_device *bdev;
 951         struct buffer_head *bh = NULL;
 952         struct btrfs_super_block *disk_super;
 953         u64 all_avail;
 954         u64 devid;
 955         u64 num_devices;
 956         u8 *dev_uuid;
 957         int ret = 0;
 958
 959         mutex_lock(&uuid_mutex);
 960         mutex_lock(&root->fs_info->volume_mutex);
 961
 962         all_avail = root->fs_info->avail_data_alloc_bits |
 963                 root->fs_info->avail_system_alloc_bits |
 964                 root->fs_info->avail_metadata_alloc_bits;
 965
 966         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 967             root->fs_info->fs_devices->rw_devices <= 4) {
 968                 printk("btrfs: unable to go below four devices on raid10\n");
 969                 ret = -EINVAL;
 970                 goto out;
 971         }
 972
 973         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 974             root->fs_info->fs_devices->rw_devices <= 2) {
 975                 printk("btrfs: unable to go below two devices on raid1\n");
 976                 ret = -EINVAL;
 977                 goto out;
 978         }
 979
 980         if (strcmp(device_path, "missing") == 0) {
 981                 struct list_head *cur;
 982                 struct list_head *devices;
 983                 struct btrfs_device *tmp;
 984
 985                 device = NULL;
 986                 devices = &root->fs_info->fs_devices->devices;
 987                 list_for_each(cur, devices) {
 988                         tmp = list_entry(cur, struct btrfs_device, dev_list);
 989                         if (tmp->in_fs_metadata && !tmp->bdev) {
 990                                 device = tmp;
 991                                 break;
 992                         }
 993                 }
 994                 bdev = NULL;
 995                 bh = NULL;
 996                 disk_super = NULL;
 997                 if (!device) {
 998                         printk("btrfs: no missing devices found to remove\n");
 999                         goto out;
1000                 }
1001         } else {
1002                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1003                                       root->fs_info->bdev_holder);
1004                 if (IS_ERR(bdev)) {
1005                         ret = PTR_ERR(bdev);
1006                         goto out;
1007                 }
1008
1009                 set_blocksize(bdev, 4096);
1010                 bh = btrfs_read_dev_super(bdev);
1011                 if (!bh) {
1012                         ret = -EIO;
1013                         goto error_close;
1014                 }
1015                 disk_super = (struct btrfs_super_block *)bh->b_data;
1016                 devid = le64_to_cpu(disk_super->dev_item.devid);
1017                 dev_uuid = disk_super->dev_item.uuid;
1018                 device = btrfs_find_device(root, devid, dev_uuid,
1019                                            disk_super->fsid);
1020                 if (!device) {
1021                         ret = -ENOENT;
1022                         goto error_brelse;
1023                 }
1024         }
1025
1026         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1027                 printk("btrfs: unable to remove the only writeable device\n");
1028                 ret = -EINVAL;
1029                 goto error_brelse;
1030         }
1031
1032         if (device->writeable) {
1033                 list_del_init(&device->dev_alloc_list);
1034                 root->fs_info->fs_devices->rw_devices--;
1035         }
1036
1037         ret = btrfs_shrink_device(device, 0);
1038         if (ret)
1039                 goto error_brelse;
1040
1041         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1042         if (ret)
1043                 goto error_brelse;
1044
1045         device->in_fs_metadata = 0;
1046         if (device->fs_devices == root->fs_info->fs_devices) {
1047                 list_del_init(&device->dev_list);
1048                 root->fs_info->fs_devices->num_devices--;
1049                 if (device->bdev)
1050                         device->fs_devices->open_devices--;
1051         }
1052
1053         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1054                                  struct btrfs_device, dev_list);
1055         if (device->bdev == root->fs_info->sb->s_bdev)
1056                 root->fs_info->sb->s_bdev = next_device->bdev;
1057         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1058                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1059
1060         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1061         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1062
1063         if (device->fs_devices != root->fs_info->fs_devices) {
1064                 BUG_ON(device->writeable);
1065                 brelse(bh);
1066                 if (bdev)
1067                         close_bdev_exclusive(bdev, FMODE_READ);
1068
1069                 if (device->bdev) {
1070                         close_bdev_exclusive(device->bdev, device->mode);
1071                         device->bdev = NULL;
1072                         device->fs_devices->open_devices--;
1073                 }
1074                 if (device->fs_devices->open_devices == 0) {
1075                         struct btrfs_fs_devices *fs_devices;
1076                         fs_devices = root->fs_info->fs_devices;
1077                         while (fs_devices) {
1078                                 if (fs_devices->seed == device->fs_devices)
1079                                         break;
1080                                 fs_devices = fs_devices->seed;
1081                         }
1082                         fs_devices->seed = device->fs_devices->seed;
1083                         device->fs_devices->seed = NULL;
1084                         __btrfs_close_devices(device->fs_devices);
1085                 }
1086                 ret = 0;
1087                 goto out;
1088         }
1089
1090         /*
1091          * at this point, the device is zero sized.  We want to
1092          * remove it from the devices list and zero out the old super
1093          */
1094         if (device->writeable) {
1095                 /* make sure this device isn't detected as part of
1096                  * the FS anymore
1097                  */
1098                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1099                 set_buffer_dirty(bh);
1100                 sync_dirty_buffer(bh);
1101         }
1102         brelse(bh);
1103
1104         if (device->bdev) {
1105                 /* one close for the device struct or super_block */
1106                 close_bdev_exclusive(device->bdev, device->mode);
1107         }
1108         if (bdev) {
1109                 /* one close for us */
1110                 close_bdev_exclusive(bdev, FMODE_READ);
1111         }
1112         kfree(device->name);
1113         kfree(device);
1114         ret = 0;
1115         goto out;
1116
1117 error_brelse:
1118         brelse(bh);
1119 error_close:
1120         if (bdev)
1121                 close_bdev_exclusive(bdev, FMODE_READ);
1122 out:
1123         mutex_unlock(&root->fs_info->volume_mutex);
1124         mutex_unlock(&uuid_mutex);
1125         return ret;
1126 }
1127
1128 /*
1129  * does all the dirty work required for changing file system's UUID.
1130  */
1131 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1132                                 struct btrfs_root *root)
1133 {
1134         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1135         struct btrfs_fs_devices *old_devices;
1136         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1137         struct btrfs_device *device;
1138         u64 super_flags;
1139
1140         BUG_ON(!mutex_is_locked(&uuid_mutex));
1141         if (!fs_devices->seeding || fs_devices->opened != 1)
1142                 return -EINVAL;
1143
1144         old_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1145         if (!old_devices)
1146                 return -ENOMEM;
1147
1148         memcpy(old_devices, fs_devices, sizeof(*old_devices));
1149         old_devices->opened = 1;
1150         old_devices->sprouted = 1;
1151         INIT_LIST_HEAD(&old_devices->devices);
1152         INIT_LIST_HEAD(&old_devices->alloc_list);
1153         list_splice_init(&fs_devices->devices, &old_devices->devices);
1154         list_splice_init(&fs_devices->alloc_list, &old_devices->alloc_list);
1155         list_for_each_entry(device, &old_devices->devices, dev_list) {
1156                 device->fs_devices = old_devices;
1157         }
1158         list_add(&old_devices->list, &fs_uuids);
1159
1160         fs_devices->seeding = 0;
1161         fs_devices->num_devices = 0;
1162         fs_devices->open_devices = 0;
1163         fs_devices->seed = old_devices;
1164
1165         generate_random_uuid(fs_devices->fsid);
1166         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1167         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1168         super_flags = btrfs_super_flags(disk_super) &
1169                       ~BTRFS_SUPER_FLAG_SEEDING;
1170         btrfs_set_super_flags(disk_super, super_flags);
1171
1172         return 0;
1173 }
1174
1175 /*
1176  * strore the expected generation for seed devices in device items.
1177  */
1178 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1179                                struct btrfs_root *root)
1180 {
1181         struct btrfs_path *path;
1182         struct extent_buffer *leaf;
1183         struct btrfs_dev_item *dev_item;
1184         struct btrfs_device *device;
1185         struct btrfs_key key;
1186         u8 fs_uuid[BTRFS_UUID_SIZE];
1187         u8 dev_uuid[BTRFS_UUID_SIZE];
1188         u64 devid;
1189         int ret;
1190
1191         path = btrfs_alloc_path();
1192         if (!path)
1193                 return -ENOMEM;
1194
1195         root = root->fs_info->chunk_root;
1196         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1197         key.offset = 0;
1198         key.type = BTRFS_DEV_ITEM_KEY;
1199
1200         while (1) {
1201                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1202                 if (ret < 0)
1203                         goto error;
1204
1205                 leaf = path->nodes[0];
1206 next_slot:
1207                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1208                         ret = btrfs_next_leaf(root, path);
1209                         if (ret > 0)
1210                                 break;
1211                         if (ret < 0)
1212                                 goto error;
1213                         leaf = path->nodes[0];
1214                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1215                         btrfs_release_path(root, path);
1216                         continue;
1217                 }
1218
1219                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1220                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1221                     key.type != BTRFS_DEV_ITEM_KEY)
1222                         break;
1223
1224                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1225                                           struct btrfs_dev_item);
1226                 devid = btrfs_device_id(leaf, dev_item);
1227                 read_extent_buffer(leaf, dev_uuid,
1228                                    (unsigned long)btrfs_device_uuid(dev_item),
1229                                    BTRFS_UUID_SIZE);
1230                 read_extent_buffer(leaf, fs_uuid,
1231                                    (unsigned long)btrfs_device_fsid(dev_item),
1232                                    BTRFS_UUID_SIZE);
1233                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1234                 BUG_ON(!device);
1235
1236                 if (device->fs_devices->seeding) {
1237                         btrfs_set_device_generation(leaf, dev_item,
1238                                                     device->generation);
1239                         btrfs_mark_buffer_dirty(leaf);
1240                 }
1241
1242                 path->slots[0]++;
1243                 goto next_slot;
1244         }
1245         ret = 0;
1246 error:
1247         btrfs_free_path(path);
1248         return ret;
1249 }
1250
1251 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1252 {
1253         struct btrfs_trans_handle *trans;
1254         struct btrfs_device *device;
1255         struct block_device *bdev;
1256         struct list_head *cur;
1257         struct list_head *devices;
1258         struct super_block *sb = root->fs_info->sb;
1259         u64 total_bytes;
1260         int seeding_dev = 0;
1261         int ret = 0;
1262
1263         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1264                 return -EINVAL;
1265
1266         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1267         if (!bdev) {
1268                 return -EIO;
1269         }
1270
1271         if (root->fs_info->fs_devices->seeding) {
1272                 seeding_dev = 1;
1273                 down_write(&sb->s_umount);
1274                 mutex_lock(&uuid_mutex);
1275         }
1276
1277         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1278         mutex_lock(&root->fs_info->volume_mutex);
1279
1280         devices = &root->fs_info->fs_devices->devices;
1281         list_for_each(cur, devices) {
1282                 device = list_entry(cur, struct btrfs_device, dev_list);
1283                 if (device->bdev == bdev) {
1284                         ret = -EEXIST;
1285                         goto error;
1286                 }
1287         }
1288
1289         device = kzalloc(sizeof(*device), GFP_NOFS);
1290         if (!device) {
1291                 /* we can safely leave the fs_devices entry around */
1292                 ret = -ENOMEM;
1293                 goto error;
1294         }
1295
1296         device->name = kstrdup(device_path, GFP_NOFS);
1297         if (!device->name) {
1298                 kfree(device);
1299                 ret = -ENOMEM;
1300                 goto error;
1301         }
1302
1303         ret = find_next_devid(root, &device->devid);
1304         if (ret) {
1305                 kfree(device);
1306                 goto error;
1307         }
1308
1309         trans = btrfs_start_transaction(root, 1);
1310         lock_chunks(root);
1311
1312         device->barriers = 1;
1313         device->writeable = 1;
1314         device->work.func = pending_bios_fn;
1315         generate_random_uuid(device->uuid);
1316         spin_lock_init(&device->io_lock);
1317         device->generation = trans->transid;
1318         device->io_width = root->sectorsize;
1319         device->io_align = root->sectorsize;
1320         device->sector_size = root->sectorsize;
1321         device->total_bytes = i_size_read(bdev->bd_inode);
1322         device->dev_root = root->fs_info->dev_root;
1323         device->bdev = bdev;
1324         device->in_fs_metadata = 1;
1325         device->mode = 0;
1326         set_blocksize(device->bdev, 4096);
1327
1328         if (seeding_dev) {
1329                 sb->s_flags &= ~MS_RDONLY;
1330                 ret = btrfs_prepare_sprout(trans, root);
1331                 BUG_ON(ret);
1332         }
1333
1334         device->fs_devices = root->fs_info->fs_devices;
1335         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1336         list_add(&device->dev_alloc_list,
1337                  &root->fs_info->fs_devices->alloc_list);
1338         root->fs_info->fs_devices->num_devices++;
1339         root->fs_info->fs_devices->open_devices++;
1340         root->fs_info->fs_devices->rw_devices++;
1341         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1342
1343         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1344         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1345                                     total_bytes + device->total_bytes);
1346
1347         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1348         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1349                                     total_bytes + 1);
1350
1351         if (seeding_dev) {
1352                 ret = init_first_rw_device(trans, root, device);
1353                 BUG_ON(ret);
1354                 ret = btrfs_finish_sprout(trans, root);
1355                 BUG_ON(ret);
1356         } else {
1357                 ret = btrfs_add_device(trans, root, device);
1358         }
1359
1360         unlock_chunks(root);
1361         btrfs_commit_transaction(trans, root);
1362
1363         if (seeding_dev) {
1364                 mutex_unlock(&uuid_mutex);
1365                 up_write(&sb->s_umount);
1366
1367                 ret = btrfs_relocate_sys_chunks(root);
1368                 BUG_ON(ret);
1369         }
1370 out:
1371         mutex_unlock(&root->fs_info->volume_mutex);
1372         return ret;
1373 error:
1374         close_bdev_exclusive(bdev, 0);
1375         if (seeding_dev) {
1376                 mutex_unlock(&uuid_mutex);
1377                 up_write(&sb->s_umount);
1378         }
1379         goto out;
1380 }
1381
1382 static int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1383                                  struct btrfs_device *device)
1384 {
1385         int ret;
1386         struct btrfs_path *path;
1387         struct btrfs_root *root;
1388         struct btrfs_dev_item *dev_item;
1389         struct extent_buffer *leaf;
1390         struct btrfs_key key;
1391
1392         root = device->dev_root->fs_info->chunk_root;
1393
1394         path = btrfs_alloc_path();
1395         if (!path)
1396                 return -ENOMEM;
1397
1398         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1399         key.type = BTRFS_DEV_ITEM_KEY;
1400         key.offset = device->devid;
1401
1402         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1403         if (ret < 0)
1404                 goto out;
1405
1406         if (ret > 0) {
1407                 ret = -ENOENT;
1408                 goto out;
1409         }
1410
1411         leaf = path->nodes[0];
1412         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1413
1414         btrfs_set_device_id(leaf, dev_item, device->devid);
1415         btrfs_set_device_type(leaf, dev_item, device->type);
1416         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1417         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1418         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1419         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1420         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1421         btrfs_mark_buffer_dirty(leaf);
1422
1423 out:
1424         btrfs_free_path(path);
1425         return ret;
1426 }
1427
1428 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1429                       struct btrfs_device *device, u64 new_size)
1430 {
1431         struct btrfs_super_block *super_copy =
1432                 &device->dev_root->fs_info->super_copy;
1433         u64 old_total = btrfs_super_total_bytes(super_copy);
1434         u64 diff = new_size - device->total_bytes;
1435
1436         if (!device->writeable)
1437                 return -EACCES;
1438         if (new_size <= device->total_bytes)
1439                 return -EINVAL;
1440
1441         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1442         device->fs_devices->total_rw_bytes += diff;
1443
1444         device->total_bytes = new_size;
1445         return btrfs_update_device(trans, device);
1446 }
1447
1448 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1449                       struct btrfs_device *device, u64 new_size)
1450 {
1451         int ret;
1452         lock_chunks(device->dev_root);
1453         ret = __btrfs_grow_device(trans, device, new_size);
1454         unlock_chunks(device->dev_root);
1455         return ret;
1456 }
1457
1458 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1459                             struct btrfs_root *root,
1460                             u64 chunk_tree, u64 chunk_objectid,
1461                             u64 chunk_offset)
1462 {
1463         int ret;
1464         struct btrfs_path *path;
1465         struct btrfs_key key;
1466
1467         root = root->fs_info->chunk_root;
1468         path = btrfs_alloc_path();
1469         if (!path)
1470                 return -ENOMEM;
1471
1472         key.objectid = chunk_objectid;
1473         key.offset = chunk_offset;
1474         key.type = BTRFS_CHUNK_ITEM_KEY;
1475
1476         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1477         BUG_ON(ret);
1478
1479         ret = btrfs_del_item(trans, root, path);
1480         BUG_ON(ret);
1481
1482         btrfs_free_path(path);
1483         return 0;
1484 }
1485
1486 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1487                         chunk_offset)
1488 {
1489         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1490         struct btrfs_disk_key *disk_key;
1491         struct btrfs_chunk *chunk;
1492         u8 *ptr;
1493         int ret = 0;
1494         u32 num_stripes;
1495         u32 array_size;
1496         u32 len = 0;
1497         u32 cur;
1498         struct btrfs_key key;
1499
1500         array_size = btrfs_super_sys_array_size(super_copy);
1501
1502         ptr = super_copy->sys_chunk_array;
1503         cur = 0;
1504
1505         while (cur < array_size) {
1506                 disk_key = (struct btrfs_disk_key *)ptr;
1507                 btrfs_disk_key_to_cpu(&key, disk_key);
1508
1509                 len = sizeof(*disk_key);
1510
1511                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1512                         chunk = (struct btrfs_chunk *)(ptr + len);
1513                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1514                         len += btrfs_chunk_item_size(num_stripes);
1515                 } else {
1516                         ret = -EIO;
1517                         break;
1518                 }
1519                 if (key.objectid == chunk_objectid &&
1520                     key.offset == chunk_offset) {
1521                         memmove(ptr, ptr + len, array_size - (cur + len));
1522                         array_size -= len;
1523                         btrfs_set_super_sys_array_size(super_copy, array_size);
1524                 } else {
1525                         ptr += len;
1526                         cur += len;
1527                 }
1528         }
1529         return ret;
1530 }
1531
1532 static int btrfs_relocate_chunk(struct btrfs_root *root,
1533                          u64 chunk_tree, u64 chunk_objectid,
1534                          u64 chunk_offset)
1535 {
1536         struct extent_map_tree *em_tree;
1537         struct btrfs_root *extent_root;
1538         struct btrfs_trans_handle *trans;
1539         struct extent_map *em;
1540         struct map_lookup *map;
1541         int ret;
1542         int i;
1543
1544         printk("btrfs relocating chunk %llu\n",
1545                (unsigned long long)chunk_offset);
1546         root = root->fs_info->chunk_root;
1547         extent_root = root->fs_info->extent_root;
1548         em_tree = &root->fs_info->mapping_tree.map_tree;
1549
1550         /* step one, relocate all the extents inside this chunk */
1551         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1552         BUG_ON(ret);
1553
1554         trans = btrfs_start_transaction(root, 1);
1555         BUG_ON(!trans);
1556
1557         lock_chunks(root);
1558
1559         /*
1560          * step two, delete the device extents and the
1561          * chunk tree entries
1562          */
1563         spin_lock(&em_tree->lock);
1564         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1565         spin_unlock(&em_tree->lock);
1566
1567         BUG_ON(em->start > chunk_offset ||
1568                em->start + em->len < chunk_offset);
1569         map = (struct map_lookup *)em->bdev;
1570
1571         for (i = 0; i < map->num_stripes; i++) {
1572                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1573                                             map->stripes[i].physical);
1574                 BUG_ON(ret);
1575
1576                 if (map->stripes[i].dev) {
1577                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1578                         BUG_ON(ret);
1579                 }
1580         }
1581         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1582                                chunk_offset);
1583
1584         BUG_ON(ret);
1585
1586         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1587                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1588                 BUG_ON(ret);
1589         }
1590
1591         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1592         BUG_ON(ret);
1593
1594         spin_lock(&em_tree->lock);
1595         remove_extent_mapping(em_tree, em);
1596         spin_unlock(&em_tree->lock);
1597
1598         kfree(map);
1599         em->bdev = NULL;
1600
1601         /* once for the tree */
1602         free_extent_map(em);
1603         /* once for us */
1604         free_extent_map(em);
1605
1606         unlock_chunks(root);
1607         btrfs_end_transaction(trans, root);
1608         return 0;
1609 }
1610
1611 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1612 {
1613         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1614         struct btrfs_path *path;
1615         struct extent_buffer *leaf;
1616         struct btrfs_chunk *chunk;
1617         struct btrfs_key key;
1618         struct btrfs_key found_key;
1619         u64 chunk_tree = chunk_root->root_key.objectid;
1620         u64 chunk_type;
1621         int ret;
1622
1623         path = btrfs_alloc_path();
1624         if (!path)
1625                 return -ENOMEM;
1626
1627         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1628         key.offset = (u64)-1;
1629         key.type = BTRFS_CHUNK_ITEM_KEY;
1630
1631         while (1) {
1632                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1633                 if (ret < 0)
1634                         goto error;
1635                 BUG_ON(ret == 0);
1636
1637                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1638                                           key.type);
1639                 if (ret < 0)
1640                         goto error;
1641                 if (ret > 0)
1642                         break;
1643
1644                 leaf = path->nodes[0];
1645                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1646
1647                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1648                                        struct btrfs_chunk);
1649                 chunk_type = btrfs_chunk_type(leaf, chunk);
1650                 btrfs_release_path(chunk_root, path);
1651
1652                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1653                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1654                                                    found_key.objectid,
1655                                                    found_key.offset);
1656                         BUG_ON(ret);
1657                 }
1658
1659                 if (found_key.offset == 0)
1660                         break;
1661                 key.offset = found_key.offset - 1;
1662         }
1663         ret = 0;
1664 error:
1665         btrfs_free_path(path);
1666         return ret;
1667 }
1668
1669 static u64 div_factor(u64 num, int factor)
1670 {
1671         if (factor == 10)
1672                 return num;
1673         num *= factor;
1674         do_div(num, 10);
1675         return num;
1676 }
1677
1678 int btrfs_balance(struct btrfs_root *dev_root)
1679 {
1680         int ret;
1681         struct list_head *cur;
1682         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1683         struct btrfs_device *device;
1684         u64 old_size;
1685         u64 size_to_free;
1686         struct btrfs_path *path;
1687         struct btrfs_key key;
1688         struct btrfs_chunk *chunk;
1689         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1690         struct btrfs_trans_handle *trans;
1691         struct btrfs_key found_key;
1692
1693         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1694                 return -EROFS;
1695
1696         mutex_lock(&dev_root->fs_info->volume_mutex);
1697         dev_root = dev_root->fs_info->dev_root;
1698
1699         /* step one make some room on all the devices */
1700         list_for_each(cur, devices) {
1701                 device = list_entry(cur, struct btrfs_device, dev_list);
1702                 old_size = device->total_bytes;
1703                 size_to_free = div_factor(old_size, 1);
1704                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1705                 if (!device->writeable ||
1706                     device->total_bytes - device->bytes_used > size_to_free)
1707                         continue;
1708
1709                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1710                 BUG_ON(ret);
1711
1712                 trans = btrfs_start_transaction(dev_root, 1);
1713                 BUG_ON(!trans);
1714
1715                 ret = btrfs_grow_device(trans, device, old_size);
1716                 BUG_ON(ret);
1717
1718                 btrfs_end_transaction(trans, dev_root);
1719         }
1720
1721         /* step two, relocate all the chunks */
1722         path = btrfs_alloc_path();
1723         BUG_ON(!path);
1724
1725         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1726         key.offset = (u64)-1;
1727         key.type = BTRFS_CHUNK_ITEM_KEY;
1728
1729         while(1) {
1730                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1731                 if (ret < 0)
1732                         goto error;
1733
1734                 /*
1735                  * this shouldn't happen, it means the last relocate
1736                  * failed
1737                  */
1738                 if (ret == 0)
1739                         break;
1740
1741                 ret = btrfs_previous_item(chunk_root, path, 0,
1742                                           BTRFS_CHUNK_ITEM_KEY);
1743                 if (ret)
1744                         break;
1745
1746                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1747                                       path->slots[0]);
1748                 if (found_key.objectid != key.objectid)
1749                         break;
1750
1751                 chunk = btrfs_item_ptr(path->nodes[0],
1752                                        path->slots[0],
1753                                        struct btrfs_chunk);
1754                 key.offset = found_key.offset;
1755                 /* chunk zero is special */
1756                 if (key.offset == 0)
1757                         break;
1758
1759                 btrfs_release_path(chunk_root, path);
1760                 ret = btrfs_relocate_chunk(chunk_root,
1761                                            chunk_root->root_key.objectid,
1762                                            found_key.objectid,
1763                                            found_key.offset);
1764                 BUG_ON(ret);
1765         }
1766         ret = 0;
1767 error:
1768         btrfs_free_path(path);
1769         mutex_unlock(&dev_root->fs_info->volume_mutex);
1770         return ret;
1771 }
1772
1773 /*
1774  * shrinking a device means finding all of the device extents past
1775  * the new size, and then following the back refs to the chunks.
1776  * The chunk relocation code actually frees the device extent
1777  */
1778 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1779 {
1780         struct btrfs_trans_handle *trans;
1781         struct btrfs_root *root = device->dev_root;
1782         struct btrfs_dev_extent *dev_extent = NULL;
1783         struct btrfs_path *path;
1784         u64 length;
1785         u64 chunk_tree;
1786         u64 chunk_objectid;
1787         u64 chunk_offset;
1788         int ret;
1789         int slot;
1790         struct extent_buffer *l;
1791         struct btrfs_key key;
1792         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1793         u64 old_total = btrfs_super_total_bytes(super_copy);
1794         u64 diff = device->total_bytes - new_size;
1795
1796         if (new_size >= device->total_bytes)
1797                 return -EINVAL;
1798
1799         path = btrfs_alloc_path();
1800         if (!path)
1801                 return -ENOMEM;
1802
1803         trans = btrfs_start_transaction(root, 1);
1804         if (!trans) {
1805                 ret = -ENOMEM;
1806                 goto done;
1807         }
1808
1809         path->reada = 2;
1810
1811         lock_chunks(root);
1812
1813         device->total_bytes = new_size;
1814         if (device->writeable)
1815                 device->fs_devices->total_rw_bytes -= diff;
1816         ret = btrfs_update_device(trans, device);
1817         if (ret) {
1818                 unlock_chunks(root);
1819                 btrfs_end_transaction(trans, root);
1820                 goto done;
1821         }
1822         WARN_ON(diff > old_total);
1823         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1824         unlock_chunks(root);
1825         btrfs_end_transaction(trans, root);
1826
1827         key.objectid = device->devid;
1828         key.offset = (u64)-1;
1829         key.type = BTRFS_DEV_EXTENT_KEY;
1830
1831         while (1) {
1832                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1833                 if (ret < 0)
1834                         goto done;
1835
1836                 ret = btrfs_previous_item(root, path, 0, key.type);
1837                 if (ret < 0)
1838                         goto done;
1839                 if (ret) {
1840                         ret = 0;
1841                         goto done;
1842                 }
1843
1844                 l = path->nodes[0];
1845                 slot = path->slots[0];
1846                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1847
1848                 if (key.objectid != device->devid)
1849                         goto done;
1850
1851                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1852                 length = btrfs_dev_extent_length(l, dev_extent);
1853
1854                 if (key.offset + length <= new_size)
1855                         goto done;
1856
1857                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1858                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1859                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1860                 btrfs_release_path(root, path);
1861
1862                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1863                                            chunk_offset);
1864                 if (ret)
1865                         goto done;
1866         }
1867
1868 done:
1869         btrfs_free_path(path);
1870         return ret;
1871 }
1872
1873 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1874                            struct btrfs_root *root,
1875                            struct btrfs_key *key,
1876                            struct btrfs_chunk *chunk, int item_size)
1877 {
1878         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1879         struct btrfs_disk_key disk_key;
1880         u32 array_size;
1881         u8 *ptr;
1882
1883         array_size = btrfs_super_sys_array_size(super_copy);
1884         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1885                 return -EFBIG;
1886
1887         ptr = super_copy->sys_chunk_array + array_size;
1888         btrfs_cpu_key_to_disk(&disk_key, key);
1889         memcpy(ptr, &disk_key, sizeof(disk_key));
1890         ptr += sizeof(disk_key);
1891         memcpy(ptr, chunk, item_size);
1892         item_size += sizeof(disk_key);
1893         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1894         return 0;
1895 }
1896
1897 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1898                                         int num_stripes, int sub_stripes)
1899 {
1900         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1901                 return calc_size;
1902         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1903                 return calc_size * (num_stripes / sub_stripes);
1904         else
1905                 return calc_size * num_stripes;
1906 }
1907
1908 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1909                                struct btrfs_root *extent_root,
1910                                struct map_lookup **map_ret,
1911                                u64 *num_bytes, u64 *stripe_size,
1912                                u64 start, u64 type)
1913 {
1914         struct btrfs_fs_info *info = extent_root->fs_info;
1915         struct btrfs_device *device = NULL;
1916         struct btrfs_fs_devices *fs_devices = info->fs_devices;
1917         struct list_head *cur;
1918         struct map_lookup *map = NULL;
1919         struct extent_map_tree *em_tree;
1920         struct extent_map *em;
1921         struct list_head private_devs;
1922         int min_stripe_size = 1 * 1024 * 1024;
1923         u64 calc_size = 1024 * 1024 * 1024;
1924         u64 max_chunk_size = calc_size;
1925         u64 min_free;
1926         u64 avail;
1927         u64 max_avail = 0;
1928         u64 dev_offset;
1929         int num_stripes = 1;
1930         int min_stripes = 1;
1931         int sub_stripes = 0;
1932         int looped = 0;
1933         int ret;
1934         int index;
1935         int stripe_len = 64 * 1024;
1936
1937         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1938             (type & BTRFS_BLOCK_GROUP_DUP)) {
1939                 WARN_ON(1);
1940                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1941         }
1942         if (list_empty(&fs_devices->alloc_list))
1943                 return -ENOSPC;
1944
1945         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1946                 num_stripes = fs_devices->rw_devices;
1947                 min_stripes = 2;
1948         }
1949         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1950                 num_stripes = 2;
1951                 min_stripes = 2;
1952         }
1953         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1954                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1955                 if (num_stripes < 2)
1956                         return -ENOSPC;
1957                 min_stripes = 2;
1958         }
1959         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1960                 num_stripes = fs_devices->rw_devices;
1961                 if (num_stripes < 4)
1962                         return -ENOSPC;
1963                 num_stripes &= ~(u32)1;
1964                 sub_stripes = 2;
1965                 min_stripes = 4;
1966         }
1967
1968         if (type & BTRFS_BLOCK_GROUP_DATA) {
1969                 max_chunk_size = 10 * calc_size;
1970                 min_stripe_size = 64 * 1024 * 1024;
1971         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1972                 max_chunk_size = 4 * calc_size;
1973                 min_stripe_size = 32 * 1024 * 1024;
1974         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1975                 calc_size = 8 * 1024 * 1024;
1976                 max_chunk_size = calc_size * 2;
1977                 min_stripe_size = 1 * 1024 * 1024;
1978         }
1979
1980         /* we don't want a chunk larger than 10% of writeable space */
1981         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
1982                              max_chunk_size);
1983
1984 again:
1985         if (!map || map->num_stripes != num_stripes) {
1986                 kfree(map);
1987                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1988                 if (!map)
1989                         return -ENOMEM;
1990                 map->num_stripes = num_stripes;
1991         }
1992
1993         if (calc_size * num_stripes > max_chunk_size) {
1994                 calc_size = max_chunk_size;
1995                 do_div(calc_size, num_stripes);
1996                 do_div(calc_size, stripe_len);
1997                 calc_size *= stripe_len;
1998         }
1999         /* we don't want tiny stripes */
2000         calc_size = max_t(u64, min_stripe_size, calc_size);
2001
2002         do_div(calc_size, stripe_len);
2003         calc_size *= stripe_len;
2004
2005         cur = fs_devices->alloc_list.next;
2006         index = 0;
2007
2008         if (type & BTRFS_BLOCK_GROUP_DUP)
2009                 min_free = calc_size * 2;
2010         else
2011                 min_free = calc_size;
2012
2013         /*
2014          * we add 1MB because we never use the first 1MB of the device, unless
2015          * we've looped, then we are likely allocating the maximum amount of
2016          * space left already
2017          */
2018         if (!looped)
2019                 min_free += 1024 * 1024;
2020
2021         INIT_LIST_HEAD(&private_devs);
2022         while(index < num_stripes) {
2023                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2024                 BUG_ON(!device->writeable);
2025                 if (device->total_bytes > device->bytes_used)
2026                         avail = device->total_bytes - device->bytes_used;
2027                 else
2028                         avail = 0;
2029                 cur = cur->next;
2030
2031                 if (device->in_fs_metadata && avail >= min_free) {
2032                         ret = find_free_dev_extent(trans, device,
2033                                                    min_free, &dev_offset);
2034                         if (ret == 0) {
2035                                 list_move_tail(&device->dev_alloc_list,
2036                                                &private_devs);
2037                                 map->stripes[index].dev = device;
2038                                 map->stripes[index].physical = dev_offset;
2039                                 index++;
2040                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2041                                         map->stripes[index].dev = device;
2042                                         map->stripes[index].physical =
2043                                                 dev_offset + calc_size;
2044                                         index++;
2045                                 }
2046                         }
2047                 } else if (device->in_fs_metadata && avail > max_avail)
2048                         max_avail = avail;
2049                 if (cur == &fs_devices->alloc_list)
2050                         break;
2051         }
2052         list_splice(&private_devs, &fs_devices->alloc_list);
2053         if (index < num_stripes) {
2054                 if (index >= min_stripes) {
2055                         num_stripes = index;
2056                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2057                                 num_stripes /= sub_stripes;
2058                                 num_stripes *= sub_stripes;
2059                         }
2060                         looped = 1;
2061                         goto again;
2062                 }
2063                 if (!looped && max_avail > 0) {
2064                         looped = 1;
2065                         calc_size = max_avail;
2066                         goto again;
2067                 }
2068                 kfree(map);
2069                 return -ENOSPC;
2070         }
2071         map->sector_size = extent_root->sectorsize;
2072         map->stripe_len = stripe_len;
2073         map->io_align = stripe_len;
2074         map->io_width = stripe_len;
2075         map->type = type;
2076         map->num_stripes = num_stripes;
2077         map->sub_stripes = sub_stripes;
2078
2079         *map_ret = map;
2080         *stripe_size = calc_size;
2081         *num_bytes = chunk_bytes_by_type(type, calc_size,
2082                                          num_stripes, sub_stripes);
2083
2084         em = alloc_extent_map(GFP_NOFS);
2085         if (!em) {
2086                 kfree(map);
2087                 return -ENOMEM;
2088         }
2089         em->bdev = (struct block_device *)map;
2090         em->start = start;
2091         em->len = *num_bytes;
2092         em->block_start = 0;
2093         em->block_len = em->len;
2094
2095         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2096         spin_lock(&em_tree->lock);
2097         ret = add_extent_mapping(em_tree, em);
2098         spin_unlock(&em_tree->lock);
2099         BUG_ON(ret);
2100         free_extent_map(em);
2101
2102         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2103                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2104                                      start, *num_bytes);
2105         BUG_ON(ret);
2106
2107         index = 0;
2108         while (index < map->num_stripes) {
2109                 device = map->stripes[index].dev;
2110                 dev_offset = map->stripes[index].physical;
2111
2112                 ret = btrfs_alloc_dev_extent(trans, device,
2113                                 info->chunk_root->root_key.objectid,
2114                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2115                                 start, dev_offset, calc_size);
2116                 BUG_ON(ret);
2117                 index++;
2118         }
2119
2120         return 0;
2121 }
2122
2123 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2124                                 struct btrfs_root *extent_root,
2125                                 struct map_lookup *map, u64 chunk_offset,
2126                                 u64 chunk_size, u64 stripe_size)
2127 {
2128         u64 dev_offset;
2129         struct btrfs_key key;
2130         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2131         struct btrfs_device *device;
2132         struct btrfs_chunk *chunk;
2133         struct btrfs_stripe *stripe;
2134         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2135         int index = 0;
2136         int ret;
2137
2138         chunk = kzalloc(item_size, GFP_NOFS);
2139         if (!chunk)
2140                 return -ENOMEM;
2141
2142         index = 0;
2143         while (index < map->num_stripes) {
2144                 device = map->stripes[index].dev;
2145                 device->bytes_used += stripe_size;
2146                 ret = btrfs_update_device(trans, device);
2147                 BUG_ON(ret);
2148                 index++;
2149         }
2150
2151         index = 0;
2152         stripe = &chunk->stripe;
2153         while (index < map->num_stripes) {
2154                 device = map->stripes[index].dev;
2155                 dev_offset = map->stripes[index].physical;
2156
2157                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2158                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2159                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2160                 stripe++;
2161                 index++;
2162         }
2163
2164         btrfs_set_stack_chunk_length(chunk, chunk_size);
2165         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2166         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2167         btrfs_set_stack_chunk_type(chunk, map->type);
2168         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2169         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2170         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2171         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2172         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2173
2174         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2175         key.type = BTRFS_CHUNK_ITEM_KEY;
2176         key.offset = chunk_offset;
2177
2178         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2179         BUG_ON(ret);
2180
2181         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2182                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2183                                              item_size);
2184                 BUG_ON(ret);
2185         }
2186         kfree(chunk);
2187         return 0;
2188 }
2189
2190 /*
2191  * Chunk allocation falls into two parts. The first part does works
2192  * that make the new allocated chunk useable, but not do any operation
2193  * that modifies the chunk tree. The second part does the works that
2194  * require modifying the chunk tree. This division is important for the
2195  * bootstrap process of adding storage to a seed btrfs.
2196  */
2197 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2198                       struct btrfs_root *extent_root, u64 type)
2199 {
2200         u64 chunk_offset;
2201         u64 chunk_size;
2202         u64 stripe_size;
2203         struct map_lookup *map;
2204         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2205         int ret;
2206
2207         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2208                               &chunk_offset);
2209         if (ret)
2210                 return ret;
2211
2212         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2213                                   &stripe_size, chunk_offset, type);
2214         if (ret)
2215                 return ret;
2216
2217         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2218                                    chunk_size, stripe_size);
2219         BUG_ON(ret);
2220         return 0;
2221 }
2222
2223 static int noinline init_first_rw_device(struct btrfs_trans_handle *trans,
2224                                          struct btrfs_root *root,
2225                                          struct btrfs_device *device)
2226 {
2227         u64 chunk_offset;
2228         u64 sys_chunk_offset;
2229         u64 chunk_size;
2230         u64 sys_chunk_size;
2231         u64 stripe_size;
2232         u64 sys_stripe_size;
2233         u64 alloc_profile;
2234         struct map_lookup *map;
2235         struct map_lookup *sys_map;
2236         struct btrfs_fs_info *fs_info = root->fs_info;
2237         struct btrfs_root *extent_root = fs_info->extent_root;
2238         int ret;
2239
2240         ret = find_next_chunk(fs_info->chunk_root,
2241                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2242         BUG_ON(ret);
2243
2244         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2245                         (fs_info->metadata_alloc_profile &
2246                          fs_info->avail_metadata_alloc_bits);
2247         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2248
2249         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2250                                   &stripe_size, chunk_offset, alloc_profile);
2251         BUG_ON(ret);
2252
2253         sys_chunk_offset = chunk_offset + chunk_size;
2254
2255         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2256                         (fs_info->system_alloc_profile &
2257                          fs_info->avail_system_alloc_bits);
2258         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2259
2260         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2261                                   &sys_chunk_size, &sys_stripe_size,
2262                                   sys_chunk_offset, alloc_profile);
2263         BUG_ON(ret);
2264
2265         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2266         BUG_ON(ret);
2267
2268         /*
2269          * Modifying chunk tree needs allocating new blocks from both
2270          * system block group and metadata block group. So we only can
2271          * do operations require modifying the chunk tree after both
2272          * block groups were created.
2273          */
2274         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2275                                    chunk_size, stripe_size);
2276         BUG_ON(ret);
2277
2278         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2279                                    sys_chunk_offset, sys_chunk_size,
2280                                    sys_stripe_size);
2281         BUG_ON(ret);
2282         return 0;
2283 }
2284
2285 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2286 {
2287         struct extent_map *em;
2288         struct map_lookup *map;
2289         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2290         int readonly = 0;
2291         int i;
2292
2293         spin_lock(&map_tree->map_tree.lock);
2294         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2295         spin_unlock(&map_tree->map_tree.lock);
2296         if (!em)
2297                 return 1;
2298
2299         map = (struct map_lookup *)em->bdev;
2300         for (i = 0; i < map->num_stripes; i++) {
2301                 if (!map->stripes[i].dev->writeable) {
2302                         readonly = 1;
2303                         break;
2304                 }
2305         }
2306         free_extent_map(em);
2307         return readonly;
2308 }
2309
2310 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2311 {
2312         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2313 }
2314
2315 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2316 {
2317         struct extent_map *em;
2318
2319         while(1) {
2320                 spin_lock(&tree->map_tree.lock);
2321                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2322                 if (em)
2323                         remove_extent_mapping(&tree->map_tree, em);
2324                 spin_unlock(&tree->map_tree.lock);
2325                 if (!em)
2326                         break;
2327                 kfree(em->bdev);
2328                 /* once for us */
2329                 free_extent_map(em);
2330                 /* once for the tree */
2331                 free_extent_map(em);
2332         }
2333 }
2334
2335 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2336 {
2337         struct extent_map *em;
2338         struct map_lookup *map;
2339         struct extent_map_tree *em_tree = &map_tree->map_tree;
2340         int ret;
2341
2342         spin_lock(&em_tree->lock);
2343         em = lookup_extent_mapping(em_tree, logical, len);
2344         spin_unlock(&em_tree->lock);
2345         BUG_ON(!em);
2346
2347         BUG_ON(em->start > logical || em->start + em->len < logical);
2348         map = (struct map_lookup *)em->bdev;
2349         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2350                 ret = map->num_stripes;
2351         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2352                 ret = map->sub_stripes;
2353         else
2354                 ret = 1;
2355         free_extent_map(em);
2356         return ret;
2357 }
2358
2359 static int find_live_mirror(struct map_lookup *map, int first, int num,
2360                             int optimal)
2361 {
2362         int i;
2363         if (map->stripes[optimal].dev->bdev)
2364                 return optimal;
2365         for (i = first; i < first + num; i++) {
2366                 if (map->stripes[i].dev->bdev)
2367                         return i;
2368         }
2369         /* we couldn't find one that doesn't fail.  Just return something
2370          * and the io error handling code will clean up eventually
2371          */
2372         return optimal;
2373 }
2374
2375 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2376                              u64 logical, u64 *length,
2377                              struct btrfs_multi_bio **multi_ret,
2378                              int mirror_num, struct page *unplug_page)
2379 {
2380         struct extent_map *em;
2381         struct map_lookup *map;
2382         struct extent_map_tree *em_tree = &map_tree->map_tree;
2383         u64 offset;
2384         u64 stripe_offset;
2385         u64 stripe_nr;
2386         int stripes_allocated = 8;
2387         int stripes_required = 1;
2388         int stripe_index;
2389         int i;
2390         int num_stripes;
2391         int max_errors = 0;
2392         struct btrfs_multi_bio *multi = NULL;
2393
2394         if (multi_ret && !(rw & (1 << BIO_RW))) {
2395                 stripes_allocated = 1;
2396         }
2397 again:
2398         if (multi_ret) {
2399                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2400                                 GFP_NOFS);
2401                 if (!multi)
2402                         return -ENOMEM;
2403
2404                 atomic_set(&multi->error, 0);
2405         }
2406
2407         spin_lock(&em_tree->lock);
2408         em = lookup_extent_mapping(em_tree, logical, *length);
2409         spin_unlock(&em_tree->lock);
2410
2411         if (!em && unplug_page)
2412                 return 0;
2413
2414         if (!em) {
2415                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
2416                 BUG();
2417         }
2418
2419         BUG_ON(em->start > logical || em->start + em->len < logical);
2420         map = (struct map_lookup *)em->bdev;
2421         offset = logical - em->start;
2422
2423         if (mirror_num > map->num_stripes)
2424                 mirror_num = 0;
2425
2426         /* if our multi bio struct is too small, back off and try again */
2427         if (rw & (1 << BIO_RW)) {
2428                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2429                                  BTRFS_BLOCK_GROUP_DUP)) {
2430                         stripes_required = map->num_stripes;
2431                         max_errors = 1;
2432                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2433                         stripes_required = map->sub_stripes;
2434                         max_errors = 1;
2435                 }
2436         }
2437         if (multi_ret && rw == WRITE &&
2438             stripes_allocated < stripes_required) {
2439                 stripes_allocated = map->num_stripes;
2440                 free_extent_map(em);
2441                 kfree(multi);
2442                 goto again;
2443         }
2444         stripe_nr = offset;
2445         /*
2446          * stripe_nr counts the total number of stripes we have to stride
2447          * to get to this block
2448          */
2449         do_div(stripe_nr, map->stripe_len);
2450
2451         stripe_offset = stripe_nr * map->stripe_len;
2452         BUG_ON(offset < stripe_offset);
2453
2454         /* stripe_offset is the offset of this block in its stripe*/
2455         stripe_offset = offset - stripe_offset;
2456
2457         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2458                          BTRFS_BLOCK_GROUP_RAID10 |
2459                          BTRFS_BLOCK_GROUP_DUP)) {
2460                 /* we limit the length of each bio to what fits in a stripe */
2461                 *length = min_t(u64, em->len - offset,
2462                               map->stripe_len - stripe_offset);
2463         } else {
2464                 *length = em->len - offset;
2465         }
2466
2467         if (!multi_ret && !unplug_page)
2468                 goto out;
2469
2470         num_stripes = 1;
2471         stripe_index = 0;
2472         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2473                 if (unplug_page || (rw & (1 << BIO_RW)))
2474                         num_stripes = map->num_stripes;
2475                 else if (mirror_num)
2476                         stripe_index = mirror_num - 1;
2477                 else {
2478                         stripe_index = find_live_mirror(map, 0,
2479                                             map->num_stripes,
2480                                             current->pid % map->num_stripes);
2481                 }
2482
2483         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2484                 if (rw & (1 << BIO_RW))
2485                         num_stripes = map->num_stripes;
2486                 else if (mirror_num)
2487                         stripe_index = mirror_num - 1;
2488
2489         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2490                 int factor = map->num_stripes / map->sub_stripes;
2491
2492                 stripe_index = do_div(stripe_nr, factor);
2493                 stripe_index *= map->sub_stripes;
2494
2495                 if (unplug_page || (rw & (1 << BIO_RW)))
2496                         num_stripes = map->sub_stripes;
2497                 else if (mirror_num)
2498                         stripe_index += mirror_num - 1;
2499                 else {
2500                         stripe_index = find_live_mirror(map, stripe_index,
2501                                               map->sub_stripes, stripe_index +
2502                                               current->pid % map->sub_stripes);
2503                 }
2504         } else {
2505                 /*
2506                  * after this do_div call, stripe_nr is the number of stripes
2507                  * on this device we have to walk to find the data, and
2508                  * stripe_index is the number of our device in the stripe array
2509                  */
2510                 stripe_index = do_div(stripe_nr, map->num_stripes);
2511         }
2512         BUG_ON(stripe_index >= map->num_stripes);
2513
2514         for (i = 0; i < num_stripes; i++) {
2515                 if (unplug_page) {
2516                         struct btrfs_device *device;
2517                         struct backing_dev_info *bdi;
2518
2519                         device = map->stripes[stripe_index].dev;
2520                         if (device->bdev) {
2521                                 bdi = blk_get_backing_dev_info(device->bdev);
2522                                 if (bdi->unplug_io_fn) {
2523                                         bdi->unplug_io_fn(bdi, unplug_page);
2524                                 }
2525                         }
2526                 } else {
2527                         multi->stripes[i].physical =
2528                                 map->stripes[stripe_index].physical +
2529                                 stripe_offset + stripe_nr * map->stripe_len;
2530                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2531                 }
2532                 stripe_index++;
2533         }
2534         if (multi_ret) {
2535                 *multi_ret = multi;
2536                 multi->num_stripes = num_stripes;
2537                 multi->max_errors = max_errors;
2538         }
2539 out:
2540         free_extent_map(em);
2541         return 0;
2542 }
2543
2544 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2545                       u64 logical, u64 *length,
2546                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2547 {
2548         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2549                                  mirror_num, NULL);
2550 }
2551
2552 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2553                      u64 chunk_start, u64 physical, u64 devid,
2554                      u64 **logical, int *naddrs, int *stripe_len)
2555 {
2556         struct extent_map_tree *em_tree = &map_tree->map_tree;
2557         struct extent_map *em;
2558         struct map_lookup *map;
2559         u64 *buf;
2560         u64 bytenr;
2561         u64 length;
2562         u64 stripe_nr;
2563         int i, j, nr = 0;
2564
2565         spin_lock(&em_tree->lock);
2566         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2567         spin_unlock(&em_tree->lock);
2568
2569         BUG_ON(!em || em->start != chunk_start);
2570         map = (struct map_lookup *)em->bdev;
2571
2572         length = em->len;
2573         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2574                 do_div(length, map->num_stripes / map->sub_stripes);
2575         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2576                 do_div(length, map->num_stripes);
2577
2578         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2579         BUG_ON(!buf);
2580
2581         for (i = 0; i < map->num_stripes; i++) {
2582                 if (devid && map->stripes[i].dev->devid != devid)
2583                         continue;
2584                 if (map->stripes[i].physical > physical ||
2585                     map->stripes[i].physical + length <= physical)
2586                         continue;
2587
2588                 stripe_nr = physical - map->stripes[i].physical;
2589                 do_div(stripe_nr, map->stripe_len);
2590
2591                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2592                         stripe_nr = stripe_nr * map->num_stripes + i;
2593                         do_div(stripe_nr, map->sub_stripes);
2594                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2595                         stripe_nr = stripe_nr * map->num_stripes + i;
2596                 }
2597                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2598                 WARN_ON(nr >= map->num_stripes);
2599                 for (j = 0; j < nr; j++) {
2600                         if (buf[j] == bytenr)
2601                                 break;
2602                 }
2603                 if (j == nr) {
2604                         WARN_ON(nr >= map->num_stripes);
2605                         buf[nr++] = bytenr;
2606                 }
2607         }
2608
2609         for (i = 0; i > nr; i++) {
2610                 struct btrfs_multi_bio *multi;
2611                 struct btrfs_bio_stripe *stripe;
2612                 int ret;
2613
2614                 length = 1;
2615                 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2616                                       &length, &multi, 0);
2617                 BUG_ON(ret);
2618
2619                 stripe = multi->stripes;
2620                 for (j = 0; j < multi->num_stripes; j++) {
2621                         if (stripe->physical >= physical &&
2622                             physical < stripe->physical + length)
2623                                 break;
2624                 }
2625                 BUG_ON(j >= multi->num_stripes);
2626                 kfree(multi);
2627         }
2628
2629         *logical = buf;
2630         *naddrs = nr;
2631         *stripe_len = map->stripe_len;
2632
2633         free_extent_map(em);
2634         return 0;
2635 }
2636
2637 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2638                       u64 logical, struct page *page)
2639 {
2640         u64 length = PAGE_CACHE_SIZE;
2641         return __btrfs_map_block(map_tree, READ, logical, &length,
2642                                  NULL, 0, page);
2643 }
2644
2645
2646 static void end_bio_multi_stripe(struct bio *bio, int err)
2647 {
2648         struct btrfs_multi_bio *multi = bio->bi_private;
2649         int is_orig_bio = 0;
2650
2651         if (err)
2652                 atomic_inc(&multi->error);
2653
2654         if (bio == multi->orig_bio)
2655                 is_orig_bio = 1;
2656
2657         if (atomic_dec_and_test(&multi->stripes_pending)) {
2658                 if (!is_orig_bio) {
2659                         bio_put(bio);
2660                         bio = multi->orig_bio;
2661                 }
2662                 bio->bi_private = multi->private;
2663                 bio->bi_end_io = multi->end_io;
2664                 /* only send an error to the higher layers if it is
2665                  * beyond the tolerance of the multi-bio
2666                  */
2667                 if (atomic_read(&multi->error) > multi->max_errors) {
2668                         err = -EIO;
2669                 } else if (err) {
2670                         /*
2671                          * this bio is actually up to date, we didn't
2672                          * go over the max number of errors
2673                          */
2674                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2675                         err = 0;
2676                 }
2677                 kfree(multi);
2678
2679                 bio_endio(bio, err);
2680         } else if (!is_orig_bio) {
2681                 bio_put(bio);
2682         }
2683 }
2684
2685 struct async_sched {
2686         struct bio *bio;
2687         int rw;
2688         struct btrfs_fs_info *info;
2689         struct btrfs_work work;
2690 };
2691
2692 /*
2693  * see run_scheduled_bios for a description of why bios are collected for
2694  * async submit.
2695  *
2696  * This will add one bio to the pending list for a device and make sure
2697  * the work struct is scheduled.
2698  */
2699 static int noinline schedule_bio(struct btrfs_root *root,
2700                                  struct btrfs_device *device,
2701                                  int rw, struct bio *bio)
2702 {
2703         int should_queue = 1;
2704
2705         /* don't bother with additional async steps for reads, right now */
2706         if (!(rw & (1 << BIO_RW))) {
2707                 bio_get(bio);
2708                 submit_bio(rw, bio);
2709                 bio_put(bio);
2710                 return 0;
2711         }
2712
2713         /*
2714          * nr_async_bios allows us to reliably return congestion to the
2715          * higher layers.  Otherwise, the async bio makes it appear we have
2716          * made progress against dirty pages when we've really just put it
2717          * on a queue for later
2718          */
2719         atomic_inc(&root->fs_info->nr_async_bios);
2720         WARN_ON(bio->bi_next);
2721         bio->bi_next = NULL;
2722         bio->bi_rw |= rw;
2723
2724         spin_lock(&device->io_lock);
2725
2726         if (device->pending_bio_tail)
2727                 device->pending_bio_tail->bi_next = bio;
2728
2729         device->pending_bio_tail = bio;
2730         if (!device->pending_bios)
2731                 device->pending_bios = bio;
2732         if (device->running_pending)
2733                 should_queue = 0;
2734
2735         spin_unlock(&device->io_lock);
2736
2737         if (should_queue)
2738                 btrfs_queue_worker(&root->fs_info->submit_workers,
2739                                    &device->work);
2740         return 0;
2741 }
2742
2743 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2744                   int mirror_num, int async_submit)
2745 {
2746         struct btrfs_mapping_tree *map_tree;
2747         struct btrfs_device *dev;
2748         struct bio *first_bio = bio;
2749         u64 logical = (u64)bio->bi_sector << 9;
2750         u64 length = 0;
2751         u64 map_length;
2752         struct btrfs_multi_bio *multi = NULL;
2753         int ret;
2754         int dev_nr = 0;
2755         int total_devs = 1;
2756
2757         length = bio->bi_size;
2758         map_tree = &root->fs_info->mapping_tree;
2759         map_length = length;
2760
2761         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2762                               mirror_num);
2763         BUG_ON(ret);
2764
2765         total_devs = multi->num_stripes;
2766         if (map_length < length) {
2767                 printk("mapping failed logical %Lu bio len %Lu "
2768                        "len %Lu\n", logical, length, map_length);
2769                 BUG();
2770         }
2771         multi->end_io = first_bio->bi_end_io;
2772         multi->private = first_bio->bi_private;
2773         multi->orig_bio = first_bio;
2774         atomic_set(&multi->stripes_pending, multi->num_stripes);
2775
2776         while(dev_nr < total_devs) {
2777                 if (total_devs > 1) {
2778                         if (dev_nr < total_devs - 1) {
2779                                 bio = bio_clone(first_bio, GFP_NOFS);
2780                                 BUG_ON(!bio);
2781                         } else {
2782                                 bio = first_bio;
2783                         }
2784                         bio->bi_private = multi;
2785                         bio->bi_end_io = end_bio_multi_stripe;
2786                 }
2787                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2788                 dev = multi->stripes[dev_nr].dev;
2789                 BUG_ON(rw == WRITE && !dev->writeable);
2790                 if (dev && dev->bdev) {
2791                         bio->bi_bdev = dev->bdev;
2792                         if (async_submit)
2793                                 schedule_bio(root, dev, rw, bio);
2794                         else
2795                                 submit_bio(rw, bio);
2796                 } else {
2797                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2798                         bio->bi_sector = logical >> 9;
2799                         bio_endio(bio, -EIO);
2800                 }
2801                 dev_nr++;
2802         }
2803         if (total_devs == 1)
2804                 kfree(multi);
2805         return 0;
2806 }
2807
2808 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2809                                        u8 *uuid, u8 *fsid)
2810 {
2811         struct btrfs_device *device;
2812         struct btrfs_fs_devices *cur_devices;
2813
2814         cur_devices = root->fs_info->fs_devices;
2815         while (cur_devices) {
2816                 if (!fsid ||
2817                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2818                         device = __find_device(&cur_devices->devices,
2819                                                devid, uuid);
2820                         if (device)
2821                                 return device;
2822                 }
2823                 cur_devices = cur_devices->seed;
2824         }
2825         return NULL;
2826 }
2827
2828 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2829                                             u64 devid, u8 *dev_uuid)
2830 {
2831         struct btrfs_device *device;
2832         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2833
2834         device = kzalloc(sizeof(*device), GFP_NOFS);
2835         if (!device)
2836                 return NULL;
2837         list_add(&device->dev_list,
2838                  &fs_devices->devices);
2839         device->barriers = 1;
2840         device->dev_root = root->fs_info->dev_root;
2841         device->devid = devid;
2842         device->work.func = pending_bios_fn;
2843         fs_devices->num_devices++;
2844         spin_lock_init(&device->io_lock);
2845         INIT_LIST_HEAD(&device->dev_alloc_list);
2846         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2847         return device;
2848 }
2849
2850 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2851                           struct extent_buffer *leaf,
2852                           struct btrfs_chunk *chunk)
2853 {
2854         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2855         struct map_lookup *map;
2856         struct extent_map *em;
2857         u64 logical;
2858         u64 length;
2859         u64 devid;
2860         u8 uuid[BTRFS_UUID_SIZE];
2861         int num_stripes;
2862         int ret;
2863         int i;
2864
2865         logical = key->offset;
2866         length = btrfs_chunk_length(leaf, chunk);
2867
2868         spin_lock(&map_tree->map_tree.lock);
2869         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2870         spin_unlock(&map_tree->map_tree.lock);
2871
2872         /* already mapped? */
2873         if (em && em->start <= logical && em->start + em->len > logical) {
2874                 free_extent_map(em);
2875                 return 0;
2876         } else if (em) {
2877                 free_extent_map(em);
2878         }
2879
2880         map = kzalloc(sizeof(*map), GFP_NOFS);
2881         if (!map)
2882                 return -ENOMEM;
2883
2884         em = alloc_extent_map(GFP_NOFS);
2885         if (!em)
2886                 return -ENOMEM;
2887         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2888         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2889         if (!map) {
2890                 free_extent_map(em);
2891                 return -ENOMEM;
2892         }
2893
2894         em->bdev = (struct block_device *)map;
2895         em->start = logical;
2896         em->len = length;
2897         em->block_start = 0;
2898         em->block_len = em->len;
2899
2900         map->num_stripes = num_stripes;
2901         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2902         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2903         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2904         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2905         map->type = btrfs_chunk_type(leaf, chunk);
2906         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2907         for (i = 0; i < num_stripes; i++) {
2908                 map->stripes[i].physical =
2909                         btrfs_stripe_offset_nr(leaf, chunk, i);
2910                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2911                 read_extent_buffer(leaf, uuid, (unsigned long)
2912                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2913                                    BTRFS_UUID_SIZE);
2914                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2915                                                         NULL);
2916                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2917                         kfree(map);
2918                         free_extent_map(em);
2919                         return -EIO;
2920                 }
2921                 if (!map->stripes[i].dev) {
2922                         map->stripes[i].dev =
2923                                 add_missing_dev(root, devid, uuid);
2924                         if (!map->stripes[i].dev) {
2925                                 kfree(map);
2926                                 free_extent_map(em);
2927                                 return -EIO;
2928                         }
2929                 }
2930                 map->stripes[i].dev->in_fs_metadata = 1;
2931         }
2932
2933         spin_lock(&map_tree->map_tree.lock);
2934         ret = add_extent_mapping(&map_tree->map_tree, em);
2935         spin_unlock(&map_tree->map_tree.lock);
2936         BUG_ON(ret);
2937         free_extent_map(em);
2938
2939         return 0;
2940 }
2941
2942 static int fill_device_from_item(struct extent_buffer *leaf,
2943                                  struct btrfs_dev_item *dev_item,
2944                                  struct btrfs_device *device)
2945 {
2946         unsigned long ptr;
2947
2948         device->devid = btrfs_device_id(leaf, dev_item);
2949         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2950         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2951         device->type = btrfs_device_type(leaf, dev_item);
2952         device->io_align = btrfs_device_io_align(leaf, dev_item);
2953         device->io_width = btrfs_device_io_width(leaf, dev_item);
2954         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2955
2956         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2957         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2958
2959         return 0;
2960 }
2961
2962 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2963 {
2964         struct btrfs_fs_devices *fs_devices;
2965         int ret;
2966
2967         mutex_lock(&uuid_mutex);
2968
2969         fs_devices = root->fs_info->fs_devices->seed;
2970         while (fs_devices) {
2971                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2972                         ret = 0;
2973                         goto out;
2974                 }
2975                 fs_devices = fs_devices->seed;
2976         }
2977
2978         fs_devices = find_fsid(fsid);
2979         if (!fs_devices) {
2980                 ret = -ENOENT;
2981                 goto out;
2982         }
2983         if (fs_devices->opened) {
2984                 ret = -EBUSY;
2985                 goto out;
2986         }
2987
2988         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
2989                                    root->fs_info->bdev_holder);
2990         if (ret)
2991                 goto out;
2992
2993         if (!fs_devices->seeding) {
2994                 __btrfs_close_devices(fs_devices);
2995                 ret = -EINVAL;
2996                 goto out;
2997         }
2998
2999         fs_devices->seed = root->fs_info->fs_devices->seed;
3000         root->fs_info->fs_devices->seed = fs_devices;
3001         fs_devices->sprouted = 1;
3002 out:
3003         mutex_unlock(&uuid_mutex);
3004         return ret;
3005 }
3006
3007 static int read_one_dev(struct btrfs_root *root,
3008                         struct extent_buffer *leaf,
3009                         struct btrfs_dev_item *dev_item)
3010 {
3011         struct btrfs_device *device;
3012         u64 devid;
3013         int ret;
3014         int seed_devices = 0;
3015         u8 fs_uuid[BTRFS_UUID_SIZE];
3016         u8 dev_uuid[BTRFS_UUID_SIZE];
3017
3018         devid = btrfs_device_id(leaf, dev_item);
3019         read_extent_buffer(leaf, dev_uuid,
3020                            (unsigned long)btrfs_device_uuid(dev_item),
3021                            BTRFS_UUID_SIZE);
3022         read_extent_buffer(leaf, fs_uuid,
3023                            (unsigned long)btrfs_device_fsid(dev_item),
3024                            BTRFS_UUID_SIZE);
3025
3026         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3027                 ret = open_seed_devices(root, fs_uuid);
3028                 if (ret)
3029                         return ret;
3030                 seed_devices = 1;
3031         }
3032
3033         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3034         if (!device || !device->bdev) {
3035                 if (!btrfs_test_opt(root, DEGRADED) || seed_devices)
3036                         return -EIO;
3037
3038                 if (!device) {
3039                         printk("warning devid %Lu missing\n", devid);
3040                         device = add_missing_dev(root, devid, dev_uuid);
3041                         if (!device)
3042                                 return -ENOMEM;
3043                 }
3044         }
3045
3046         if (device->fs_devices != root->fs_info->fs_devices) {
3047                 BUG_ON(device->writeable);
3048                 if (device->generation !=
3049                     btrfs_device_generation(leaf, dev_item))
3050                         return -EINVAL;
3051         }
3052
3053         fill_device_from_item(leaf, dev_item, device);
3054         device->dev_root = root->fs_info->dev_root;
3055         device->in_fs_metadata = 1;
3056         if (device->writeable)
3057                 device->fs_devices->total_rw_bytes += device->total_bytes;
3058         ret = 0;
3059 #if 0
3060         ret = btrfs_open_device(device);
3061         if (ret) {
3062                 kfree(device);
3063         }
3064 #endif
3065         return ret;
3066 }
3067
3068 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3069 {
3070         struct btrfs_dev_item *dev_item;
3071
3072         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3073                                                      dev_item);
3074         return read_one_dev(root, buf, dev_item);
3075 }
3076
3077 int btrfs_read_sys_array(struct btrfs_root *root, u64 sb_bytenr)
3078 {
3079         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3080         struct extent_buffer *sb;
3081         struct btrfs_disk_key *disk_key;
3082         struct btrfs_chunk *chunk;
3083         u8 *ptr;
3084         unsigned long sb_ptr;
3085         int ret = 0;
3086         u32 num_stripes;
3087         u32 array_size;
3088         u32 len = 0;
3089         u32 cur;
3090         struct btrfs_key key;
3091
3092         sb = btrfs_find_create_tree_block(root, sb_bytenr,
3093                                           BTRFS_SUPER_INFO_SIZE);
3094         if (!sb)
3095                 return -ENOMEM;
3096         btrfs_set_buffer_uptodate(sb);
3097         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3098         array_size = btrfs_super_sys_array_size(super_copy);
3099
3100         ptr = super_copy->sys_chunk_array;
3101         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3102         cur = 0;
3103
3104         while (cur < array_size) {
3105                 disk_key = (struct btrfs_disk_key *)ptr;
3106                 btrfs_disk_key_to_cpu(&key, disk_key);
3107
3108                 len = sizeof(*disk_key); ptr += len;
3109                 sb_ptr += len;
3110                 cur += len;
3111
3112                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3113                         chunk = (struct btrfs_chunk *)sb_ptr;
3114                         ret = read_one_chunk(root, &key, sb, chunk);
3115                         if (ret)
3116                                 break;
3117                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3118                         len = btrfs_chunk_item_size(num_stripes);
3119                 } else {
3120                         ret = -EIO;
3121                         break;
3122                 }
3123                 ptr += len;
3124                 sb_ptr += len;
3125                 cur += len;
3126         }
3127         free_extent_buffer(sb);
3128         return ret;
3129 }
3130
3131 int btrfs_read_chunk_tree(struct btrfs_root *root)
3132 {
3133         struct btrfs_path *path;
3134         struct extent_buffer *leaf;
3135         struct btrfs_key key;
3136         struct btrfs_key found_key;
3137         int ret;
3138         int slot;
3139
3140         root = root->fs_info->chunk_root;
3141
3142         path = btrfs_alloc_path();
3143         if (!path)
3144                 return -ENOMEM;
3145
3146         /* first we search for all of the device items, and then we
3147          * read in all of the chunk items.  This way we can create chunk
3148          * mappings that reference all of the devices that are afound
3149          */
3150         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3151         key.offset = 0;
3152         key.type = 0;
3153 again:
3154         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3155         while(1) {
3156                 leaf = path->nodes[0];
3157                 slot = path->slots[0];
3158                 if (slot >= btrfs_header_nritems(leaf)) {
3159                         ret = btrfs_next_leaf(root, path);
3160                         if (ret == 0)
3161                                 continue;
3162                         if (ret < 0)
3163                                 goto error;
3164                         break;
3165                 }
3166                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3167                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3168                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3169                                 break;
3170                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3171                                 struct btrfs_dev_item *dev_item;
3172                                 dev_item = btrfs_item_ptr(leaf, slot,
3173                                                   struct btrfs_dev_item);
3174                                 ret = read_one_dev(root, leaf, dev_item);
3175                                 if (ret)
3176                                         goto error;
3177                         }
3178                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3179                         struct btrfs_chunk *chunk;
3180                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3181                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3182                         if (ret)
3183                                 goto error;
3184                 }
3185                 path->slots[0]++;
3186         }
3187         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3188                 key.objectid = 0;
3189                 btrfs_release_path(root, path);
3190                 goto again;
3191         }
3192         ret = 0;
3193 error:
3194         btrfs_free_path(path);
3195         return ret;
3196 }