SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
 894         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 895         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 896         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 897         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 898         btrfs_mark_buffer_dirty(leaf);
 899
 900         ret = 0;
 901 out:
 902         btrfs_free_path(path);
 903         return ret;
 904 }
 905
 906 static int btrfs_rm_dev_item(struct btrfs_root *root,
 907                              struct btrfs_device *device)
 908 {
 909         int ret;
 910         struct btrfs_path *path;
 911         struct btrfs_key key;
 912         struct btrfs_trans_handle *trans;
 913
 914         root = root->fs_info->chunk_root;
 915
 916         path = btrfs_alloc_path();
 917         if (!path)
 918                 return -ENOMEM;
 919
 920         trans = btrfs_start_transaction(root, 1);
 921         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 922         key.type = BTRFS_DEV_ITEM_KEY;
 923         key.offset = device->devid;
 924         lock_chunks(root);
 925
 926         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 927         if (ret < 0)
 928                 goto out;
 929
 930         if (ret > 0) {
 931                 ret = -ENOENT;
 932                 goto out;
 933         }
 934
 935         ret = btrfs_del_item(trans, root, path);
 936         if (ret)
 937                 goto out;
 938 out:
 939         btrfs_free_path(path);
 940         unlock_chunks(root);
 941         btrfs_commit_transaction(trans, root);
 942         return ret;
 943 }
 944
 945 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 946 {
 947         struct btrfs_device *device;
 948         struct btrfs_device *next_device;
 949         struct block_device *bdev;
 950         struct buffer_head *bh = NULL;
 951         struct btrfs_super_block *disk_super;
 952         u64 all_avail;
 953         u64 devid;
 954         u64 num_devices;
 955         u8 *dev_uuid;
 956         int ret = 0;
 957
 958         mutex_lock(&uuid_mutex);
 959         mutex_lock(&root->fs_info->volume_mutex);
 960
 961         all_avail = root->fs_info->avail_data_alloc_bits |
 962                 root->fs_info->avail_system_alloc_bits |
 963                 root->fs_info->avail_metadata_alloc_bits;
 964
 965         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 966             root->fs_info->fs_devices->rw_devices <= 4) {
 967                 printk("btrfs: unable to go below four devices on raid10\n");
 968                 ret = -EINVAL;
 969                 goto out;
 970         }
 971
 972         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 973             root->fs_info->fs_devices->rw_devices <= 2) {
 974                 printk("btrfs: unable to go below two devices on raid1\n");
 975                 ret = -EINVAL;
 976                 goto out;
 977         }
 978
 979         if (strcmp(device_path, "missing") == 0) {
 980                 struct list_head *cur;
 981                 struct list_head *devices;
 982                 struct btrfs_device *tmp;
 983
 984                 device = NULL;
 985                 devices = &root->fs_info->fs_devices->devices;
 986                 list_for_each(cur, devices) {
 987                         tmp = list_entry(cur, struct btrfs_device, dev_list);
 988                         if (tmp->in_fs_metadata && !tmp->bdev) {
 989                                 device = tmp;
 990                                 break;
 991                         }
 992                 }
 993                 bdev = NULL;
 994                 bh = NULL;
 995                 disk_super = NULL;
 996                 if (!device) {
 997                         printk("btrfs: no missing devices found to remove\n");
 998                         goto out;
 999                 }
1000         } else {
1001                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1002                                       root->fs_info->bdev_holder);
1003                 if (IS_ERR(bdev)) {
1004                         ret = PTR_ERR(bdev);
1005                         goto out;
1006                 }
1007
1008                 set_blocksize(bdev, 4096);
1009                 bh = btrfs_read_dev_super(bdev);
1010                 if (!bh) {
1011                         ret = -EIO;
1012                         goto error_close;
1013                 }
1014                 disk_super = (struct btrfs_super_block *)bh->b_data;
1015                 devid = le64_to_cpu(disk_super->dev_item.devid);
1016                 dev_uuid = disk_super->dev_item.uuid;
1017                 device = btrfs_find_device(root, devid, dev_uuid,
1018                                            disk_super->fsid);
1019                 if (!device) {
1020                         ret = -ENOENT;
1021                         goto error_brelse;
1022                 }
1023         }
1024
1025         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1026                 printk("btrfs: unable to remove the only writeable device\n");
1027                 ret = -EINVAL;
1028                 goto error_brelse;
1029         }
1030
1031         if (device->writeable) {
1032                 list_del_init(&device->dev_alloc_list);
1033                 root->fs_info->fs_devices->rw_devices--;
1034         }
1035
1036         ret = btrfs_shrink_device(device, 0);
1037         if (ret)
1038                 goto error_brelse;
1039
1040         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1041         if (ret)
1042                 goto error_brelse;
1043
1044         device->in_fs_metadata = 0;
1045         if (device->fs_devices == root->fs_info->fs_devices) {
1046                 list_del_init(&device->dev_list);
1047                 root->fs_info->fs_devices->num_devices--;
1048                 if (device->bdev)
1049                         device->fs_devices->open_devices--;
1050         }
1051
1052         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1053                                  struct btrfs_device, dev_list);
1054         if (device->bdev == root->fs_info->sb->s_bdev)
1055                 root->fs_info->sb->s_bdev = next_device->bdev;
1056         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1057                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1058
1059         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1060         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1061
1062         if (device->fs_devices != root->fs_info->fs_devices) {
1063                 BUG_ON(device->writeable);
1064                 brelse(bh);
1065                 if (bdev)
1066                         close_bdev_exclusive(bdev, FMODE_READ);
1067
1068                 if (device->bdev) {
1069                         close_bdev_exclusive(device->bdev, device->mode);
1070                         device->bdev = NULL;
1071                         device->fs_devices->open_devices--;
1072                 }
1073                 if (device->fs_devices->open_devices == 0) {
1074                         struct btrfs_fs_devices *fs_devices;
1075                         fs_devices = root->fs_info->fs_devices;
1076                         while (fs_devices) {
1077                                 if (fs_devices->seed == device->fs_devices)
1078                                         break;
1079                                 fs_devices = fs_devices->seed;
1080                         }
1081                         fs_devices->seed = device->fs_devices->seed;
1082                         device->fs_devices->seed = NULL;
1083                         __btrfs_close_devices(device->fs_devices);
1084                 }
1085                 ret = 0;
1086                 goto out;
1087         }
1088
1089         /*
1090          * at this point, the device is zero sized.  We want to
1091          * remove it from the devices list and zero out the old super
1092          */
1093         if (device->writeable) {
1094                 /* make sure this device isn't detected as part of
1095                  * the FS anymore
1096                  */
1097                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1098                 set_buffer_dirty(bh);
1099                 sync_dirty_buffer(bh);
1100         }
1101         brelse(bh);
1102
1103         if (device->bdev) {
1104                 /* one close for the device struct or super_block */
1105                 close_bdev_exclusive(device->bdev, device->mode);
1106         }
1107         if (bdev) {
1108                 /* one close for us */
1109                 close_bdev_exclusive(bdev, FMODE_READ);
1110         }
1111         kfree(device->name);
1112         kfree(device);
1113         ret = 0;
1114         goto out;
1115
1116 error_brelse:
1117         brelse(bh);
1118 error_close:
1119         if (bdev)
1120                 close_bdev_exclusive(bdev, FMODE_READ);
1121 out:
1122         mutex_unlock(&root->fs_info->volume_mutex);
1123         mutex_unlock(&uuid_mutex);
1124         return ret;
1125 }
1126
1127 /*
1128  * does all the dirty work required for changing file system's UUID.
1129  */
1130 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1131                                 struct btrfs_root *root)
1132 {
1133         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1134         struct btrfs_fs_devices *old_devices;
1135         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1136         struct btrfs_device *device;
1137         u64 super_flags;
1138
1139         BUG_ON(!mutex_is_locked(&uuid_mutex));
1140         if (!fs_devices->seeding || fs_devices->opened != 1)
1141                 return -EINVAL;
1142
1143         old_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1144         if (!old_devices)
1145                 return -ENOMEM;
1146
1147         memcpy(old_devices, fs_devices, sizeof(*old_devices));
1148         old_devices->opened = 1;
1149         old_devices->sprouted = 1;
1150         INIT_LIST_HEAD(&old_devices->devices);
1151         INIT_LIST_HEAD(&old_devices->alloc_list);
1152         list_splice_init(&fs_devices->devices, &old_devices->devices);
1153         list_splice_init(&fs_devices->alloc_list, &old_devices->alloc_list);
1154         list_for_each_entry(device, &old_devices->devices, dev_list) {
1155                 device->fs_devices = old_devices;
1156         }
1157         list_add(&old_devices->list, &fs_uuids);
1158
1159         fs_devices->seeding = 0;
1160         fs_devices->num_devices = 0;
1161         fs_devices->open_devices = 0;
1162         fs_devices->seed = old_devices;
1163
1164         generate_random_uuid(fs_devices->fsid);
1165         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1166         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1167         super_flags = btrfs_super_flags(disk_super) &
1168                       ~BTRFS_SUPER_FLAG_SEEDING;
1169         btrfs_set_super_flags(disk_super, super_flags);
1170
1171         return 0;
1172 }
1173
1174 /*
1175  * strore the expected generation for seed devices in device items.
1176  */
1177 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1178                                struct btrfs_root *root)
1179 {
1180         struct btrfs_path *path;
1181         struct extent_buffer *leaf;
1182         struct btrfs_dev_item *dev_item;
1183         struct btrfs_device *device;
1184         struct btrfs_key key;
1185         u8 fs_uuid[BTRFS_UUID_SIZE];
1186         u8 dev_uuid[BTRFS_UUID_SIZE];
1187         u64 devid;
1188         int ret;
1189
1190         path = btrfs_alloc_path();
1191         if (!path)
1192                 return -ENOMEM;
1193
1194         root = root->fs_info->chunk_root;
1195         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1196         key.offset = 0;
1197         key.type = BTRFS_DEV_ITEM_KEY;
1198
1199         while (1) {
1200                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1201                 if (ret < 0)
1202                         goto error;
1203
1204                 leaf = path->nodes[0];
1205 next_slot:
1206                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1207                         ret = btrfs_next_leaf(root, path);
1208                         if (ret > 0)
1209                                 break;
1210                         if (ret < 0)
1211                                 goto error;
1212                         leaf = path->nodes[0];
1213                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1214                         btrfs_release_path(root, path);
1215                         continue;
1216                 }
1217
1218                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1219                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1220                     key.type != BTRFS_DEV_ITEM_KEY)
1221                         break;
1222
1223                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1224                                           struct btrfs_dev_item);
1225                 devid = btrfs_device_id(leaf, dev_item);
1226                 read_extent_buffer(leaf, dev_uuid,
1227                                    (unsigned long)btrfs_device_uuid(dev_item),
1228                                    BTRFS_UUID_SIZE);
1229                 read_extent_buffer(leaf, fs_uuid,
1230                                    (unsigned long)btrfs_device_fsid(dev_item),
1231                                    BTRFS_UUID_SIZE);
1232                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1233                 BUG_ON(!device);
1234
1235                 if (device->fs_devices->seeding) {
1236                         btrfs_set_device_generation(leaf, dev_item,
1237                                                     device->generation);
1238                         btrfs_mark_buffer_dirty(leaf);
1239                 }
1240
1241                 path->slots[0]++;
1242                 goto next_slot;
1243         }
1244         ret = 0;
1245 error:
1246         btrfs_free_path(path);
1247         return ret;
1248 }
1249
1250 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1251 {
1252         struct btrfs_trans_handle *trans;
1253         struct btrfs_device *device;
1254         struct block_device *bdev;
1255         struct list_head *cur;
1256         struct list_head *devices;
1257         struct super_block *sb = root->fs_info->sb;
1258         u64 total_bytes;
1259         int seeding_dev = 0;
1260         int ret = 0;
1261
1262         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1263                 return -EINVAL;
1264
1265         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1266         if (!bdev) {
1267                 return -EIO;
1268         }
1269
1270         if (root->fs_info->fs_devices->seeding) {
1271                 seeding_dev = 1;
1272                 down_write(&sb->s_umount);
1273                 mutex_lock(&uuid_mutex);
1274         }
1275
1276         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1277         mutex_lock(&root->fs_info->volume_mutex);
1278
1279         devices = &root->fs_info->fs_devices->devices;
1280         list_for_each(cur, devices) {
1281                 device = list_entry(cur, struct btrfs_device, dev_list);
1282                 if (device->bdev == bdev) {
1283                         ret = -EEXIST;
1284                         goto error;
1285                 }
1286         }
1287
1288         device = kzalloc(sizeof(*device), GFP_NOFS);
1289         if (!device) {
1290                 /* we can safely leave the fs_devices entry around */
1291                 ret = -ENOMEM;
1292                 goto error;
1293         }
1294
1295         device->name = kstrdup(device_path, GFP_NOFS);
1296         if (!device->name) {
1297                 kfree(device);
1298                 ret = -ENOMEM;
1299                 goto error;
1300         }
1301
1302         ret = find_next_devid(root, &device->devid);
1303         if (ret) {
1304                 kfree(device);
1305                 goto error;
1306         }
1307
1308         trans = btrfs_start_transaction(root, 1);
1309         lock_chunks(root);
1310
1311         device->barriers = 1;
1312         device->writeable = 1;
1313         device->work.func = pending_bios_fn;
1314         generate_random_uuid(device->uuid);
1315         spin_lock_init(&device->io_lock);
1316         device->generation = trans->transid;
1317         device->io_width = root->sectorsize;
1318         device->io_align = root->sectorsize;
1319         device->sector_size = root->sectorsize;
1320         device->total_bytes = i_size_read(bdev->bd_inode);
1321         device->dev_root = root->fs_info->dev_root;
1322         device->bdev = bdev;
1323         device->in_fs_metadata = 1;
1324         device->mode = 0;
1325         set_blocksize(device->bdev, 4096);
1326
1327         if (seeding_dev) {
1328                 sb->s_flags &= ~MS_RDONLY;
1329                 ret = btrfs_prepare_sprout(trans, root);
1330                 BUG_ON(ret);
1331         }
1332
1333         device->fs_devices = root->fs_info->fs_devices;
1334         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1335         list_add(&device->dev_alloc_list,
1336                  &root->fs_info->fs_devices->alloc_list);
1337         root->fs_info->fs_devices->num_devices++;
1338         root->fs_info->fs_devices->open_devices++;
1339         root->fs_info->fs_devices->rw_devices++;
1340         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1341
1342         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1343         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1344                                     total_bytes + device->total_bytes);
1345
1346         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1347         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1348                                     total_bytes + 1);
1349
1350         if (seeding_dev) {
1351                 ret = init_first_rw_device(trans, root, device);
1352                 BUG_ON(ret);
1353                 ret = btrfs_finish_sprout(trans, root);
1354                 BUG_ON(ret);
1355         } else {
1356                 ret = btrfs_add_device(trans, root, device);
1357         }
1358
1359         unlock_chunks(root);
1360         btrfs_commit_transaction(trans, root);
1361
1362         if (seeding_dev) {
1363                 mutex_unlock(&uuid_mutex);
1364                 up_write(&sb->s_umount);
1365
1366                 ret = btrfs_relocate_sys_chunks(root);
1367                 BUG_ON(ret);
1368         }
1369 out:
1370         mutex_unlock(&root->fs_info->volume_mutex);
1371         return ret;
1372 error:
1373         close_bdev_exclusive(bdev, 0);
1374         if (seeding_dev) {
1375                 mutex_unlock(&uuid_mutex);
1376                 up_write(&sb->s_umount);
1377         }
1378         goto out;
1379 }
1380
1381 static int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1382                                  struct btrfs_device *device)
1383 {
1384         int ret;
1385         struct btrfs_path *path;
1386         struct btrfs_root *root;
1387         struct btrfs_dev_item *dev_item;
1388         struct extent_buffer *leaf;
1389         struct btrfs_key key;
1390
1391         root = device->dev_root->fs_info->chunk_root;
1392
1393         path = btrfs_alloc_path();
1394         if (!path)
1395                 return -ENOMEM;
1396
1397         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1398         key.type = BTRFS_DEV_ITEM_KEY;
1399         key.offset = device->devid;
1400
1401         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1402         if (ret < 0)
1403                 goto out;
1404
1405         if (ret > 0) {
1406                 ret = -ENOENT;
1407                 goto out;
1408         }
1409
1410         leaf = path->nodes[0];
1411         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1412
1413         btrfs_set_device_id(leaf, dev_item, device->devid);
1414         btrfs_set_device_type(leaf, dev_item, device->type);
1415         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1416         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1417         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1418         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1419         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1420         btrfs_mark_buffer_dirty(leaf);
1421
1422 out:
1423         btrfs_free_path(path);
1424         return ret;
1425 }
1426
1427 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1428                       struct btrfs_device *device, u64 new_size)
1429 {
1430         struct btrfs_super_block *super_copy =
1431                 &device->dev_root->fs_info->super_copy;
1432         u64 old_total = btrfs_super_total_bytes(super_copy);
1433         u64 diff = new_size - device->total_bytes;
1434
1435         if (!device->writeable)
1436                 return -EACCES;
1437         if (new_size <= device->total_bytes)
1438                 return -EINVAL;
1439
1440         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1441         device->fs_devices->total_rw_bytes += diff;
1442
1443         device->total_bytes = new_size;
1444         return btrfs_update_device(trans, device);
1445 }
1446
1447 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1448                       struct btrfs_device *device, u64 new_size)
1449 {
1450         int ret;
1451         lock_chunks(device->dev_root);
1452         ret = __btrfs_grow_device(trans, device, new_size);
1453         unlock_chunks(device->dev_root);
1454         return ret;
1455 }
1456
1457 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1458                             struct btrfs_root *root,
1459                             u64 chunk_tree, u64 chunk_objectid,
1460                             u64 chunk_offset)
1461 {
1462         int ret;
1463         struct btrfs_path *path;
1464         struct btrfs_key key;
1465
1466         root = root->fs_info->chunk_root;
1467         path = btrfs_alloc_path();
1468         if (!path)
1469                 return -ENOMEM;
1470
1471         key.objectid = chunk_objectid;
1472         key.offset = chunk_offset;
1473         key.type = BTRFS_CHUNK_ITEM_KEY;
1474
1475         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1476         BUG_ON(ret);
1477
1478         ret = btrfs_del_item(trans, root, path);
1479         BUG_ON(ret);
1480
1481         btrfs_free_path(path);
1482         return 0;
1483 }
1484
1485 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1486                         chunk_offset)
1487 {
1488         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1489         struct btrfs_disk_key *disk_key;
1490         struct btrfs_chunk *chunk;
1491         u8 *ptr;
1492         int ret = 0;
1493         u32 num_stripes;
1494         u32 array_size;
1495         u32 len = 0;
1496         u32 cur;
1497         struct btrfs_key key;
1498
1499         array_size = btrfs_super_sys_array_size(super_copy);
1500
1501         ptr = super_copy->sys_chunk_array;
1502         cur = 0;
1503
1504         while (cur < array_size) {
1505                 disk_key = (struct btrfs_disk_key *)ptr;
1506                 btrfs_disk_key_to_cpu(&key, disk_key);
1507
1508                 len = sizeof(*disk_key);
1509
1510                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1511                         chunk = (struct btrfs_chunk *)(ptr + len);
1512                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1513                         len += btrfs_chunk_item_size(num_stripes);
1514                 } else {
1515                         ret = -EIO;
1516                         break;
1517                 }
1518                 if (key.objectid == chunk_objectid &&
1519                     key.offset == chunk_offset) {
1520                         memmove(ptr, ptr + len, array_size - (cur + len));
1521                         array_size -= len;
1522                         btrfs_set_super_sys_array_size(super_copy, array_size);
1523                 } else {
1524                         ptr += len;
1525                         cur += len;
1526                 }
1527         }
1528         return ret;
1529 }
1530
1531 static int btrfs_relocate_chunk(struct btrfs_root *root,
1532                          u64 chunk_tree, u64 chunk_objectid,
1533                          u64 chunk_offset)
1534 {
1535         struct extent_map_tree *em_tree;
1536         struct btrfs_root *extent_root;
1537         struct btrfs_trans_handle *trans;
1538         struct extent_map *em;
1539         struct map_lookup *map;
1540         int ret;
1541         int i;
1542
1543         printk("btrfs relocating chunk %llu\n",
1544                (unsigned long long)chunk_offset);
1545         root = root->fs_info->chunk_root;
1546         extent_root = root->fs_info->extent_root;
1547         em_tree = &root->fs_info->mapping_tree.map_tree;
1548
1549         /* step one, relocate all the extents inside this chunk */
1550         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1551         BUG_ON(ret);
1552
1553         trans = btrfs_start_transaction(root, 1);
1554         BUG_ON(!trans);
1555
1556         lock_chunks(root);
1557
1558         /*
1559          * step two, delete the device extents and the
1560          * chunk tree entries
1561          */
1562         spin_lock(&em_tree->lock);
1563         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1564         spin_unlock(&em_tree->lock);
1565
1566         BUG_ON(em->start > chunk_offset ||
1567                em->start + em->len < chunk_offset);
1568         map = (struct map_lookup *)em->bdev;
1569
1570         for (i = 0; i < map->num_stripes; i++) {
1571                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1572                                             map->stripes[i].physical);
1573                 BUG_ON(ret);
1574
1575                 if (map->stripes[i].dev) {
1576                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1577                         BUG_ON(ret);
1578                 }
1579         }
1580         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1581                                chunk_offset);
1582
1583         BUG_ON(ret);
1584
1585         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1586                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1587                 BUG_ON(ret);
1588         }
1589
1590         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1591         BUG_ON(ret);
1592
1593         spin_lock(&em_tree->lock);
1594         remove_extent_mapping(em_tree, em);
1595         spin_unlock(&em_tree->lock);
1596
1597         kfree(map);
1598         em->bdev = NULL;
1599
1600         /* once for the tree */
1601         free_extent_map(em);
1602         /* once for us */
1603         free_extent_map(em);
1604
1605         unlock_chunks(root);
1606         btrfs_end_transaction(trans, root);
1607         return 0;
1608 }
1609
1610 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1611 {
1612         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1613         struct btrfs_path *path;
1614         struct extent_buffer *leaf;
1615         struct btrfs_chunk *chunk;
1616         struct btrfs_key key;
1617         struct btrfs_key found_key;
1618         u64 chunk_tree = chunk_root->root_key.objectid;
1619         u64 chunk_type;
1620         int ret;
1621
1622         path = btrfs_alloc_path();
1623         if (!path)
1624                 return -ENOMEM;
1625
1626         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1627         key.offset = (u64)-1;
1628         key.type = BTRFS_CHUNK_ITEM_KEY;
1629
1630         while (1) {
1631                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1632                 if (ret < 0)
1633                         goto error;
1634                 BUG_ON(ret == 0);
1635
1636                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1637                                           key.type);
1638                 if (ret < 0)
1639                         goto error;
1640                 if (ret > 0)
1641                         break;
1642
1643                 leaf = path->nodes[0];
1644                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1645
1646                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1647                                        struct btrfs_chunk);
1648                 chunk_type = btrfs_chunk_type(leaf, chunk);
1649                 btrfs_release_path(chunk_root, path);
1650
1651                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1652                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1653                                                    found_key.objectid,
1654                                                    found_key.offset);
1655                         BUG_ON(ret);
1656                 }
1657
1658                 if (found_key.offset == 0)
1659                         break;
1660                 key.offset = found_key.offset - 1;
1661         }
1662         ret = 0;
1663 error:
1664         btrfs_free_path(path);
1665         return ret;
1666 }
1667
1668 static u64 div_factor(u64 num, int factor)
1669 {
1670         if (factor == 10)
1671                 return num;
1672         num *= factor;
1673         do_div(num, 10);
1674         return num;
1675 }
1676
1677 int btrfs_balance(struct btrfs_root *dev_root)
1678 {
1679         int ret;
1680         struct list_head *cur;
1681         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1682         struct btrfs_device *device;
1683         u64 old_size;
1684         u64 size_to_free;
1685         struct btrfs_path *path;
1686         struct btrfs_key key;
1687         struct btrfs_chunk *chunk;
1688         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1689         struct btrfs_trans_handle *trans;
1690         struct btrfs_key found_key;
1691
1692         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1693                 return -EROFS;
1694
1695         mutex_lock(&dev_root->fs_info->volume_mutex);
1696         dev_root = dev_root->fs_info->dev_root;
1697
1698         /* step one make some room on all the devices */
1699         list_for_each(cur, devices) {
1700                 device = list_entry(cur, struct btrfs_device, dev_list);
1701                 old_size = device->total_bytes;
1702                 size_to_free = div_factor(old_size, 1);
1703                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1704                 if (!device->writeable ||
1705                     device->total_bytes - device->bytes_used > size_to_free)
1706                         continue;
1707
1708                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1709                 BUG_ON(ret);
1710
1711                 trans = btrfs_start_transaction(dev_root, 1);
1712                 BUG_ON(!trans);
1713
1714                 ret = btrfs_grow_device(trans, device, old_size);
1715                 BUG_ON(ret);
1716
1717                 btrfs_end_transaction(trans, dev_root);
1718         }
1719
1720         /* step two, relocate all the chunks */
1721         path = btrfs_alloc_path();
1722         BUG_ON(!path);
1723
1724         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1725         key.offset = (u64)-1;
1726         key.type = BTRFS_CHUNK_ITEM_KEY;
1727
1728         while(1) {
1729                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1730                 if (ret < 0)
1731                         goto error;
1732
1733                 /*
1734                  * this shouldn't happen, it means the last relocate
1735                  * failed
1736                  */
1737                 if (ret == 0)
1738                         break;
1739
1740                 ret = btrfs_previous_item(chunk_root, path, 0,
1741                                           BTRFS_CHUNK_ITEM_KEY);
1742                 if (ret)
1743                         break;
1744
1745                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1746                                       path->slots[0]);
1747                 if (found_key.objectid != key.objectid)
1748                         break;
1749
1750                 chunk = btrfs_item_ptr(path->nodes[0],
1751                                        path->slots[0],
1752                                        struct btrfs_chunk);
1753                 key.offset = found_key.offset;
1754                 /* chunk zero is special */
1755                 if (key.offset == 0)
1756                         break;
1757
1758                 btrfs_release_path(chunk_root, path);
1759                 ret = btrfs_relocate_chunk(chunk_root,
1760                                            chunk_root->root_key.objectid,
1761                                            found_key.objectid,
1762                                            found_key.offset);
1763                 BUG_ON(ret);
1764         }
1765         ret = 0;
1766 error:
1767         btrfs_free_path(path);
1768         mutex_unlock(&dev_root->fs_info->volume_mutex);
1769         return ret;
1770 }
1771
1772 /*
1773  * shrinking a device means finding all of the device extents past
1774  * the new size, and then following the back refs to the chunks.
1775  * The chunk relocation code actually frees the device extent
1776  */
1777 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1778 {
1779         struct btrfs_trans_handle *trans;
1780         struct btrfs_root *root = device->dev_root;
1781         struct btrfs_dev_extent *dev_extent = NULL;
1782         struct btrfs_path *path;
1783         u64 length;
1784         u64 chunk_tree;
1785         u64 chunk_objectid;
1786         u64 chunk_offset;
1787         int ret;
1788         int slot;
1789         struct extent_buffer *l;
1790         struct btrfs_key key;
1791         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1792         u64 old_total = btrfs_super_total_bytes(super_copy);
1793         u64 diff = device->total_bytes - new_size;
1794
1795         if (new_size >= device->total_bytes)
1796                 return -EINVAL;
1797
1798         path = btrfs_alloc_path();
1799         if (!path)
1800                 return -ENOMEM;
1801
1802         trans = btrfs_start_transaction(root, 1);
1803         if (!trans) {
1804                 ret = -ENOMEM;
1805                 goto done;
1806         }
1807
1808         path->reada = 2;
1809
1810         lock_chunks(root);
1811
1812         device->total_bytes = new_size;
1813         if (device->writeable)
1814                 device->fs_devices->total_rw_bytes -= diff;
1815         ret = btrfs_update_device(trans, device);
1816         if (ret) {
1817                 unlock_chunks(root);
1818                 btrfs_end_transaction(trans, root);
1819                 goto done;
1820         }
1821         WARN_ON(diff > old_total);
1822         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1823         unlock_chunks(root);
1824         btrfs_end_transaction(trans, root);
1825
1826         key.objectid = device->devid;
1827         key.offset = (u64)-1;
1828         key.type = BTRFS_DEV_EXTENT_KEY;
1829
1830         while (1) {
1831                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1832                 if (ret < 0)
1833                         goto done;
1834
1835                 ret = btrfs_previous_item(root, path, 0, key.type);
1836                 if (ret < 0)
1837                         goto done;
1838                 if (ret) {
1839                         ret = 0;
1840                         goto done;
1841                 }
1842
1843                 l = path->nodes[0];
1844                 slot = path->slots[0];
1845                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1846
1847                 if (key.objectid != device->devid)
1848                         goto done;
1849
1850                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1851                 length = btrfs_dev_extent_length(l, dev_extent);
1852
1853                 if (key.offset + length <= new_size)
1854                         goto done;
1855
1856                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1857                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1858                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1859                 btrfs_release_path(root, path);
1860
1861                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1862                                            chunk_offset);
1863                 if (ret)
1864                         goto done;
1865         }
1866
1867 done:
1868         btrfs_free_path(path);
1869         return ret;
1870 }
1871
1872 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1873                            struct btrfs_root *root,
1874                            struct btrfs_key *key,
1875                            struct btrfs_chunk *chunk, int item_size)
1876 {
1877         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1878         struct btrfs_disk_key disk_key;
1879         u32 array_size;
1880         u8 *ptr;
1881
1882         array_size = btrfs_super_sys_array_size(super_copy);
1883         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1884                 return -EFBIG;
1885
1886         ptr = super_copy->sys_chunk_array + array_size;
1887         btrfs_cpu_key_to_disk(&disk_key, key);
1888         memcpy(ptr, &disk_key, sizeof(disk_key));
1889         ptr += sizeof(disk_key);
1890         memcpy(ptr, chunk, item_size);
1891         item_size += sizeof(disk_key);
1892         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1893         return 0;
1894 }
1895
1896 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1897                                         int num_stripes, int sub_stripes)
1898 {
1899         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1900                 return calc_size;
1901         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1902                 return calc_size * (num_stripes / sub_stripes);
1903         else
1904                 return calc_size * num_stripes;
1905 }
1906
1907 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1908                                struct btrfs_root *extent_root,
1909                                struct map_lookup **map_ret,
1910                                u64 *num_bytes, u64 *stripe_size,
1911                                u64 start, u64 type)
1912 {
1913         struct btrfs_fs_info *info = extent_root->fs_info;
1914         struct btrfs_device *device = NULL;
1915         struct btrfs_fs_devices *fs_devices = info->fs_devices;
1916         struct list_head *cur;
1917         struct map_lookup *map = NULL;
1918         struct extent_map_tree *em_tree;
1919         struct extent_map *em;
1920         struct list_head private_devs;
1921         int min_stripe_size = 1 * 1024 * 1024;
1922         u64 calc_size = 1024 * 1024 * 1024;
1923         u64 max_chunk_size = calc_size;
1924         u64 min_free;
1925         u64 avail;
1926         u64 max_avail = 0;
1927         u64 dev_offset;
1928         int num_stripes = 1;
1929         int min_stripes = 1;
1930         int sub_stripes = 0;
1931         int looped = 0;
1932         int ret;
1933         int index;
1934         int stripe_len = 64 * 1024;
1935
1936         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1937             (type & BTRFS_BLOCK_GROUP_DUP)) {
1938                 WARN_ON(1);
1939                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1940         }
1941         if (list_empty(&fs_devices->alloc_list))
1942                 return -ENOSPC;
1943
1944         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1945                 num_stripes = fs_devices->rw_devices;
1946                 min_stripes = 2;
1947         }
1948         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1949                 num_stripes = 2;
1950                 min_stripes = 2;
1951         }
1952         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1953                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1954                 if (num_stripes < 2)
1955                         return -ENOSPC;
1956                 min_stripes = 2;
1957         }
1958         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1959                 num_stripes = fs_devices->rw_devices;
1960                 if (num_stripes < 4)
1961                         return -ENOSPC;
1962                 num_stripes &= ~(u32)1;
1963                 sub_stripes = 2;
1964                 min_stripes = 4;
1965         }
1966
1967         if (type & BTRFS_BLOCK_GROUP_DATA) {
1968                 max_chunk_size = 10 * calc_size;
1969                 min_stripe_size = 64 * 1024 * 1024;
1970         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1971                 max_chunk_size = 4 * calc_size;
1972                 min_stripe_size = 32 * 1024 * 1024;
1973         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1974                 calc_size = 8 * 1024 * 1024;
1975                 max_chunk_size = calc_size * 2;
1976                 min_stripe_size = 1 * 1024 * 1024;
1977         }
1978
1979         /* we don't want a chunk larger than 10% of writeable space */
1980         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
1981                              max_chunk_size);
1982
1983 again:
1984         if (!map || map->num_stripes != num_stripes) {
1985                 kfree(map);
1986                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1987                 if (!map)
1988                         return -ENOMEM;
1989                 map->num_stripes = num_stripes;
1990         }
1991
1992         if (calc_size * num_stripes > max_chunk_size) {
1993                 calc_size = max_chunk_size;
1994                 do_div(calc_size, num_stripes);
1995                 do_div(calc_size, stripe_len);
1996                 calc_size *= stripe_len;
1997         }
1998         /* we don't want tiny stripes */
1999         calc_size = max_t(u64, min_stripe_size, calc_size);
2000
2001         do_div(calc_size, stripe_len);
2002         calc_size *= stripe_len;
2003
2004         cur = fs_devices->alloc_list.next;
2005         index = 0;
2006
2007         if (type & BTRFS_BLOCK_GROUP_DUP)
2008                 min_free = calc_size * 2;
2009         else
2010                 min_free = calc_size;
2011
2012         /*
2013          * we add 1MB because we never use the first 1MB of the device, unless
2014          * we've looped, then we are likely allocating the maximum amount of
2015          * space left already
2016          */
2017         if (!looped)
2018                 min_free += 1024 * 1024;
2019
2020         INIT_LIST_HEAD(&private_devs);
2021         while(index < num_stripes) {
2022                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2023                 BUG_ON(!device->writeable);
2024                 if (device->total_bytes > device->bytes_used)
2025                         avail = device->total_bytes - device->bytes_used;
2026                 else
2027                         avail = 0;
2028                 cur = cur->next;
2029
2030                 if (device->in_fs_metadata && avail >= min_free) {
2031                         ret = find_free_dev_extent(trans, device,
2032                                                    min_free, &dev_offset);
2033                         if (ret == 0) {
2034                                 list_move_tail(&device->dev_alloc_list,
2035                                                &private_devs);
2036                                 map->stripes[index].dev = device;
2037                                 map->stripes[index].physical = dev_offset;
2038                                 index++;
2039                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2040                                         map->stripes[index].dev = device;
2041                                         map->stripes[index].physical =
2042                                                 dev_offset + calc_size;
2043                                         index++;
2044                                 }
2045                         }
2046                 } else if (device->in_fs_metadata && avail > max_avail)
2047                         max_avail = avail;
2048                 if (cur == &fs_devices->alloc_list)
2049                         break;
2050         }
2051         list_splice(&private_devs, &fs_devices->alloc_list);
2052         if (index < num_stripes) {
2053                 if (index >= min_stripes) {
2054                         num_stripes = index;
2055                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2056                                 num_stripes /= sub_stripes;
2057                                 num_stripes *= sub_stripes;
2058                         }
2059                         looped = 1;
2060                         goto again;
2061                 }
2062                 if (!looped && max_avail > 0) {
2063                         looped = 1;
2064                         calc_size = max_avail;
2065                         goto again;
2066                 }
2067                 kfree(map);
2068                 return -ENOSPC;
2069         }
2070         map->sector_size = extent_root->sectorsize;
2071         map->stripe_len = stripe_len;
2072         map->io_align = stripe_len;
2073         map->io_width = stripe_len;
2074         map->type = type;
2075         map->num_stripes = num_stripes;
2076         map->sub_stripes = sub_stripes;
2077
2078         *map_ret = map;
2079         *stripe_size = calc_size;
2080         *num_bytes = chunk_bytes_by_type(type, calc_size,
2081                                          num_stripes, sub_stripes);
2082
2083         em = alloc_extent_map(GFP_NOFS);
2084         if (!em) {
2085                 kfree(map);
2086                 return -ENOMEM;
2087         }
2088         em->bdev = (struct block_device *)map;
2089         em->start = start;
2090         em->len = *num_bytes;
2091         em->block_start = 0;
2092         em->block_len = em->len;
2093
2094         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2095         spin_lock(&em_tree->lock);
2096         ret = add_extent_mapping(em_tree, em);
2097         spin_unlock(&em_tree->lock);
2098         BUG_ON(ret);
2099         free_extent_map(em);
2100
2101         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2102                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2103                                      start, *num_bytes);
2104         BUG_ON(ret);
2105
2106         index = 0;
2107         while (index < map->num_stripes) {
2108                 device = map->stripes[index].dev;
2109                 dev_offset = map->stripes[index].physical;
2110
2111                 ret = btrfs_alloc_dev_extent(trans, device,
2112                                 info->chunk_root->root_key.objectid,
2113                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2114                                 start, dev_offset, calc_size);
2115                 BUG_ON(ret);
2116                 index++;
2117         }
2118
2119         return 0;
2120 }
2121
2122 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2123                                 struct btrfs_root *extent_root,
2124                                 struct map_lookup *map, u64 chunk_offset,
2125                                 u64 chunk_size, u64 stripe_size)
2126 {
2127         u64 dev_offset;
2128         struct btrfs_key key;
2129         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2130         struct btrfs_device *device;
2131         struct btrfs_chunk *chunk;
2132         struct btrfs_stripe *stripe;
2133         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2134         int index = 0;
2135         int ret;
2136
2137         chunk = kzalloc(item_size, GFP_NOFS);
2138         if (!chunk)
2139                 return -ENOMEM;
2140
2141         index = 0;
2142         while (index < map->num_stripes) {
2143                 device = map->stripes[index].dev;
2144                 device->bytes_used += stripe_size;
2145                 ret = btrfs_update_device(trans, device);
2146                 BUG_ON(ret);
2147                 index++;
2148         }
2149
2150         index = 0;
2151         stripe = &chunk->stripe;
2152         while (index < map->num_stripes) {
2153                 device = map->stripes[index].dev;
2154                 dev_offset = map->stripes[index].physical;
2155
2156                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2157                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2158                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2159                 stripe++;
2160                 index++;
2161         }
2162
2163         btrfs_set_stack_chunk_length(chunk, chunk_size);
2164         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2165         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2166         btrfs_set_stack_chunk_type(chunk, map->type);
2167         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2168         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2169         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2170         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2171         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2172
2173         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2174         key.type = BTRFS_CHUNK_ITEM_KEY;
2175         key.offset = chunk_offset;
2176
2177         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2178         BUG_ON(ret);
2179
2180         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2181                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2182                                              item_size);
2183                 BUG_ON(ret);
2184         }
2185         kfree(chunk);
2186         return 0;
2187 }
2188
2189 /*
2190  * Chunk allocation falls into two parts. The first part does works
2191  * that make the new allocated chunk useable, but not do any operation
2192  * that modifies the chunk tree. The second part does the works that
2193  * require modifying the chunk tree. This division is important for the
2194  * bootstrap process of adding storage to a seed btrfs.
2195  */
2196 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2197                       struct btrfs_root *extent_root, u64 type)
2198 {
2199         u64 chunk_offset;
2200         u64 chunk_size;
2201         u64 stripe_size;
2202         struct map_lookup *map;
2203         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2204         int ret;
2205
2206         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2207                               &chunk_offset);
2208         if (ret)
2209                 return ret;
2210
2211         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2212                                   &stripe_size, chunk_offset, type);
2213         if (ret)
2214                 return ret;
2215
2216         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2217                                    chunk_size, stripe_size);
2218         BUG_ON(ret);
2219         return 0;
2220 }
2221
2222 static int noinline init_first_rw_device(struct btrfs_trans_handle *trans,
2223                                          struct btrfs_root *root,
2224                                          struct btrfs_device *device)
2225 {
2226         u64 chunk_offset;
2227         u64 sys_chunk_offset;
2228         u64 chunk_size;
2229         u64 sys_chunk_size;
2230         u64 stripe_size;
2231         u64 sys_stripe_size;
2232         u64 alloc_profile;
2233         struct map_lookup *map;
2234         struct map_lookup *sys_map;
2235         struct btrfs_fs_info *fs_info = root->fs_info;
2236         struct btrfs_root *extent_root = fs_info->extent_root;
2237         int ret;
2238
2239         ret = find_next_chunk(fs_info->chunk_root,
2240                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2241         BUG_ON(ret);
2242
2243         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2244                         (fs_info->metadata_alloc_profile &
2245                          fs_info->avail_metadata_alloc_bits);
2246         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2247
2248         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2249                                   &stripe_size, chunk_offset, alloc_profile);
2250         BUG_ON(ret);
2251
2252         sys_chunk_offset = chunk_offset + chunk_size;
2253
2254         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2255                         (fs_info->system_alloc_profile &
2256                          fs_info->avail_system_alloc_bits);
2257         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2258
2259         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2260                                   &sys_chunk_size, &sys_stripe_size,
2261                                   sys_chunk_offset, alloc_profile);
2262         BUG_ON(ret);
2263
2264         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2265         BUG_ON(ret);
2266
2267         /*
2268          * Modifying chunk tree needs allocating new blocks from both
2269          * system block group and metadata block group. So we only can
2270          * do operations require modifying the chunk tree after both
2271          * block groups were created.
2272          */
2273         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2274                                    chunk_size, stripe_size);
2275         BUG_ON(ret);
2276
2277         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2278                                    sys_chunk_offset, sys_chunk_size,
2279                                    sys_stripe_size);
2280         BUG_ON(ret);
2281         return 0;
2282 }
2283
2284 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2285 {
2286         struct extent_map *em;
2287         struct map_lookup *map;
2288         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2289         int readonly = 0;
2290         int i;
2291
2292         spin_lock(&map_tree->map_tree.lock);
2293         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2294         spin_unlock(&map_tree->map_tree.lock);
2295         if (!em)
2296                 return 1;
2297
2298         map = (struct map_lookup *)em->bdev;
2299         for (i = 0; i < map->num_stripes; i++) {
2300                 if (!map->stripes[i].dev->writeable) {
2301                         readonly = 1;
2302                         break;
2303                 }
2304         }
2305         free_extent_map(em);
2306         return readonly;
2307 }
2308
2309 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2310 {
2311         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2312 }
2313
2314 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2315 {
2316         struct extent_map *em;
2317
2318         while(1) {
2319                 spin_lock(&tree->map_tree.lock);
2320                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2321                 if (em)
2322                         remove_extent_mapping(&tree->map_tree, em);
2323                 spin_unlock(&tree->map_tree.lock);
2324                 if (!em)
2325                         break;
2326                 kfree(em->bdev);
2327                 /* once for us */
2328                 free_extent_map(em);
2329                 /* once for the tree */
2330                 free_extent_map(em);
2331         }
2332 }
2333
2334 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2335 {
2336         struct extent_map *em;
2337         struct map_lookup *map;
2338         struct extent_map_tree *em_tree = &map_tree->map_tree;
2339         int ret;
2340
2341         spin_lock(&em_tree->lock);
2342         em = lookup_extent_mapping(em_tree, logical, len);
2343         spin_unlock(&em_tree->lock);
2344         BUG_ON(!em);
2345
2346         BUG_ON(em->start > logical || em->start + em->len < logical);
2347         map = (struct map_lookup *)em->bdev;
2348         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2349                 ret = map->num_stripes;
2350         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2351                 ret = map->sub_stripes;
2352         else
2353                 ret = 1;
2354         free_extent_map(em);
2355         return ret;
2356 }
2357
2358 static int find_live_mirror(struct map_lookup *map, int first, int num,
2359                             int optimal)
2360 {
2361         int i;
2362         if (map->stripes[optimal].dev->bdev)
2363                 return optimal;
2364         for (i = first; i < first + num; i++) {
2365                 if (map->stripes[i].dev->bdev)
2366                         return i;
2367         }
2368         /* we couldn't find one that doesn't fail.  Just return something
2369          * and the io error handling code will clean up eventually
2370          */
2371         return optimal;
2372 }
2373
2374 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2375                              u64 logical, u64 *length,
2376                              struct btrfs_multi_bio **multi_ret,
2377                              int mirror_num, struct page *unplug_page)
2378 {
2379         struct extent_map *em;
2380         struct map_lookup *map;
2381         struct extent_map_tree *em_tree = &map_tree->map_tree;
2382         u64 offset;
2383         u64 stripe_offset;
2384         u64 stripe_nr;
2385         int stripes_allocated = 8;
2386         int stripes_required = 1;
2387         int stripe_index;
2388         int i;
2389         int num_stripes;
2390         int max_errors = 0;
2391         struct btrfs_multi_bio *multi = NULL;
2392
2393         if (multi_ret && !(rw & (1 << BIO_RW))) {
2394                 stripes_allocated = 1;
2395         }
2396 again:
2397         if (multi_ret) {
2398                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2399                                 GFP_NOFS);
2400                 if (!multi)
2401                         return -ENOMEM;
2402
2403                 atomic_set(&multi->error, 0);
2404         }
2405
2406         spin_lock(&em_tree->lock);
2407         em = lookup_extent_mapping(em_tree, logical, *length);
2408         spin_unlock(&em_tree->lock);
2409
2410         if (!em && unplug_page)
2411                 return 0;
2412
2413         if (!em) {
2414                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
2415                 BUG();
2416         }
2417
2418         BUG_ON(em->start > logical || em->start + em->len < logical);
2419         map = (struct map_lookup *)em->bdev;
2420         offset = logical - em->start;
2421
2422         if (mirror_num > map->num_stripes)
2423                 mirror_num = 0;
2424
2425         /* if our multi bio struct is too small, back off and try again */
2426         if (rw & (1 << BIO_RW)) {
2427                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2428                                  BTRFS_BLOCK_GROUP_DUP)) {
2429                         stripes_required = map->num_stripes;
2430                         max_errors = 1;
2431                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2432                         stripes_required = map->sub_stripes;
2433                         max_errors = 1;
2434                 }
2435         }
2436         if (multi_ret && rw == WRITE &&
2437             stripes_allocated < stripes_required) {
2438                 stripes_allocated = map->num_stripes;
2439                 free_extent_map(em);
2440                 kfree(multi);
2441                 goto again;
2442         }
2443         stripe_nr = offset;
2444         /*
2445          * stripe_nr counts the total number of stripes we have to stride
2446          * to get to this block
2447          */
2448         do_div(stripe_nr, map->stripe_len);
2449
2450         stripe_offset = stripe_nr * map->stripe_len;
2451         BUG_ON(offset < stripe_offset);
2452
2453         /* stripe_offset is the offset of this block in its stripe*/
2454         stripe_offset = offset - stripe_offset;
2455
2456         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2457                          BTRFS_BLOCK_GROUP_RAID10 |
2458                          BTRFS_BLOCK_GROUP_DUP)) {
2459                 /* we limit the length of each bio to what fits in a stripe */
2460                 *length = min_t(u64, em->len - offset,
2461                               map->stripe_len - stripe_offset);
2462         } else {
2463                 *length = em->len - offset;
2464         }
2465
2466         if (!multi_ret && !unplug_page)
2467                 goto out;
2468
2469         num_stripes = 1;
2470         stripe_index = 0;
2471         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2472                 if (unplug_page || (rw & (1 << BIO_RW)))
2473                         num_stripes = map->num_stripes;
2474                 else if (mirror_num)
2475                         stripe_index = mirror_num - 1;
2476                 else {
2477                         stripe_index = find_live_mirror(map, 0,
2478                                             map->num_stripes,
2479                                             current->pid % map->num_stripes);
2480                 }
2481
2482         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2483                 if (rw & (1 << BIO_RW))
2484                         num_stripes = map->num_stripes;
2485                 else if (mirror_num)
2486                         stripe_index = mirror_num - 1;
2487
2488         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2489                 int factor = map->num_stripes / map->sub_stripes;
2490
2491                 stripe_index = do_div(stripe_nr, factor);
2492                 stripe_index *= map->sub_stripes;
2493
2494                 if (unplug_page || (rw & (1 << BIO_RW)))
2495                         num_stripes = map->sub_stripes;
2496                 else if (mirror_num)
2497                         stripe_index += mirror_num - 1;
2498                 else {
2499                         stripe_index = find_live_mirror(map, stripe_index,
2500                                               map->sub_stripes, stripe_index +
2501                                               current->pid % map->sub_stripes);
2502                 }
2503         } else {
2504                 /*
2505                  * after this do_div call, stripe_nr is the number of stripes
2506                  * on this device we have to walk to find the data, and
2507                  * stripe_index is the number of our device in the stripe array
2508                  */
2509                 stripe_index = do_div(stripe_nr, map->num_stripes);
2510         }
2511         BUG_ON(stripe_index >= map->num_stripes);
2512
2513         for (i = 0; i < num_stripes; i++) {
2514                 if (unplug_page) {
2515                         struct btrfs_device *device;
2516                         struct backing_dev_info *bdi;
2517
2518                         device = map->stripes[stripe_index].dev;
2519                         if (device->bdev) {
2520                                 bdi = blk_get_backing_dev_info(device->bdev);
2521                                 if (bdi->unplug_io_fn) {
2522                                         bdi->unplug_io_fn(bdi, unplug_page);
2523                                 }
2524                         }
2525                 } else {
2526                         multi->stripes[i].physical =
2527                                 map->stripes[stripe_index].physical +
2528                                 stripe_offset + stripe_nr * map->stripe_len;
2529                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2530                 }
2531                 stripe_index++;
2532         }
2533         if (multi_ret) {
2534                 *multi_ret = multi;
2535                 multi->num_stripes = num_stripes;
2536                 multi->max_errors = max_errors;
2537         }
2538 out:
2539         free_extent_map(em);
2540         return 0;
2541 }
2542
2543 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2544                       u64 logical, u64 *length,
2545                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2546 {
2547         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2548                                  mirror_num, NULL);
2549 }
2550
2551 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2552                      u64 chunk_start, u64 physical, u64 devid,
2553                      u64 **logical, int *naddrs, int *stripe_len)
2554 {
2555         struct extent_map_tree *em_tree = &map_tree->map_tree;
2556         struct extent_map *em;
2557         struct map_lookup *map;
2558         u64 *buf;
2559         u64 bytenr;
2560         u64 length;
2561         u64 stripe_nr;
2562         int i, j, nr = 0;
2563
2564         spin_lock(&em_tree->lock);
2565         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2566         spin_unlock(&em_tree->lock);
2567
2568         BUG_ON(!em || em->start != chunk_start);
2569         map = (struct map_lookup *)em->bdev;
2570
2571         length = em->len;
2572         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2573                 do_div(length, map->num_stripes / map->sub_stripes);
2574         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2575                 do_div(length, map->num_stripes);
2576
2577         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2578         BUG_ON(!buf);
2579
2580         for (i = 0; i < map->num_stripes; i++) {
2581                 if (devid && map->stripes[i].dev->devid != devid)
2582                         continue;
2583                 if (map->stripes[i].physical > physical ||
2584                     map->stripes[i].physical + length <= physical)
2585                         continue;
2586
2587                 stripe_nr = physical - map->stripes[i].physical;
2588                 do_div(stripe_nr, map->stripe_len);
2589
2590                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2591                         stripe_nr = stripe_nr * map->num_stripes + i;
2592                         do_div(stripe_nr, map->sub_stripes);
2593                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2594                         stripe_nr = stripe_nr * map->num_stripes + i;
2595                 }
2596                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2597                 for (j = 0; j < nr; j++) {
2598                         if (buf[j] == bytenr)
2599                                 break;
2600                 }
2601                 if (j == nr)
2602                         buf[nr++] = bytenr;
2603         }
2604
2605         for (i = 0; i > nr; i++) {
2606                 struct btrfs_multi_bio *multi;
2607                 struct btrfs_bio_stripe *stripe;
2608                 int ret;
2609
2610                 length = 1;
2611                 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2612                                       &length, &multi, 0);
2613                 BUG_ON(ret);
2614
2615                 stripe = multi->stripes;
2616                 for (j = 0; j < multi->num_stripes; j++) {
2617                         if (stripe->physical >= physical &&
2618                             physical < stripe->physical + length)
2619                                 break;
2620                 }
2621                 BUG_ON(j >= multi->num_stripes);
2622                 kfree(multi);
2623         }
2624
2625         *logical = buf;
2626         *naddrs = nr;
2627         *stripe_len = map->stripe_len;
2628
2629         free_extent_map(em);
2630         return 0;
2631 }
2632
2633 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2634                       u64 logical, struct page *page)
2635 {
2636         u64 length = PAGE_CACHE_SIZE;
2637         return __btrfs_map_block(map_tree, READ, logical, &length,
2638                                  NULL, 0, page);
2639 }
2640
2641
2642 static void end_bio_multi_stripe(struct bio *bio, int err)
2643 {
2644         struct btrfs_multi_bio *multi = bio->bi_private;
2645         int is_orig_bio = 0;
2646
2647         if (err)
2648                 atomic_inc(&multi->error);
2649
2650         if (bio == multi->orig_bio)
2651                 is_orig_bio = 1;
2652
2653         if (atomic_dec_and_test(&multi->stripes_pending)) {
2654                 if (!is_orig_bio) {
2655                         bio_put(bio);
2656                         bio = multi->orig_bio;
2657                 }
2658                 bio->bi_private = multi->private;
2659                 bio->bi_end_io = multi->end_io;
2660                 /* only send an error to the higher layers if it is
2661                  * beyond the tolerance of the multi-bio
2662                  */
2663                 if (atomic_read(&multi->error) > multi->max_errors) {
2664                         err = -EIO;
2665                 } else if (err) {
2666                         /*
2667                          * this bio is actually up to date, we didn't
2668                          * go over the max number of errors
2669                          */
2670                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2671                         err = 0;
2672                 }
2673                 kfree(multi);
2674
2675                 bio_endio(bio, err);
2676         } else if (!is_orig_bio) {
2677                 bio_put(bio);
2678         }
2679 }
2680
2681 struct async_sched {
2682         struct bio *bio;
2683         int rw;
2684         struct btrfs_fs_info *info;
2685         struct btrfs_work work;
2686 };
2687
2688 /*
2689  * see run_scheduled_bios for a description of why bios are collected for
2690  * async submit.
2691  *
2692  * This will add one bio to the pending list for a device and make sure
2693  * the work struct is scheduled.
2694  */
2695 static int noinline schedule_bio(struct btrfs_root *root,
2696                                  struct btrfs_device *device,
2697                                  int rw, struct bio *bio)
2698 {
2699         int should_queue = 1;
2700
2701         /* don't bother with additional async steps for reads, right now */
2702         if (!(rw & (1 << BIO_RW))) {
2703                 bio_get(bio);
2704                 submit_bio(rw, bio);
2705                 bio_put(bio);
2706                 return 0;
2707         }
2708
2709         /*
2710          * nr_async_bios allows us to reliably return congestion to the
2711          * higher layers.  Otherwise, the async bio makes it appear we have
2712          * made progress against dirty pages when we've really just put it
2713          * on a queue for later
2714          */
2715         atomic_inc(&root->fs_info->nr_async_bios);
2716         WARN_ON(bio->bi_next);
2717         bio->bi_next = NULL;
2718         bio->bi_rw |= rw;
2719
2720         spin_lock(&device->io_lock);
2721
2722         if (device->pending_bio_tail)
2723                 device->pending_bio_tail->bi_next = bio;
2724
2725         device->pending_bio_tail = bio;
2726         if (!device->pending_bios)
2727                 device->pending_bios = bio;
2728         if (device->running_pending)
2729                 should_queue = 0;
2730
2731         spin_unlock(&device->io_lock);
2732
2733         if (should_queue)
2734                 btrfs_queue_worker(&root->fs_info->submit_workers,
2735                                    &device->work);
2736         return 0;
2737 }
2738
2739 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2740                   int mirror_num, int async_submit)
2741 {
2742         struct btrfs_mapping_tree *map_tree;
2743         struct btrfs_device *dev;
2744         struct bio *first_bio = bio;
2745         u64 logical = (u64)bio->bi_sector << 9;
2746         u64 length = 0;
2747         u64 map_length;
2748         struct btrfs_multi_bio *multi = NULL;
2749         int ret;
2750         int dev_nr = 0;
2751         int total_devs = 1;
2752
2753         length = bio->bi_size;
2754         map_tree = &root->fs_info->mapping_tree;
2755         map_length = length;
2756
2757         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2758                               mirror_num);
2759         BUG_ON(ret);
2760
2761         total_devs = multi->num_stripes;
2762         if (map_length < length) {
2763                 printk("mapping failed logical %Lu bio len %Lu "
2764                        "len %Lu\n", logical, length, map_length);
2765                 BUG();
2766         }
2767         multi->end_io = first_bio->bi_end_io;
2768         multi->private = first_bio->bi_private;
2769         multi->orig_bio = first_bio;
2770         atomic_set(&multi->stripes_pending, multi->num_stripes);
2771
2772         while(dev_nr < total_devs) {
2773                 if (total_devs > 1) {
2774                         if (dev_nr < total_devs - 1) {
2775                                 bio = bio_clone(first_bio, GFP_NOFS);
2776                                 BUG_ON(!bio);
2777                         } else {
2778                                 bio = first_bio;
2779                         }
2780                         bio->bi_private = multi;
2781                         bio->bi_end_io = end_bio_multi_stripe;
2782                 }
2783                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2784                 dev = multi->stripes[dev_nr].dev;
2785                 BUG_ON(rw == WRITE && !dev->writeable);
2786                 if (dev && dev->bdev) {
2787                         bio->bi_bdev = dev->bdev;
2788                         if (async_submit)
2789                                 schedule_bio(root, dev, rw, bio);
2790                         else
2791                                 submit_bio(rw, bio);
2792                 } else {
2793                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2794                         bio->bi_sector = logical >> 9;
2795                         bio_endio(bio, -EIO);
2796                 }
2797                 dev_nr++;
2798         }
2799         if (total_devs == 1)
2800                 kfree(multi);
2801         return 0;
2802 }
2803
2804 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2805                                        u8 *uuid, u8 *fsid)
2806 {
2807         struct btrfs_device *device;
2808         struct btrfs_fs_devices *cur_devices;
2809
2810         cur_devices = root->fs_info->fs_devices;
2811         while (cur_devices) {
2812                 if (!fsid ||
2813                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2814                         device = __find_device(&cur_devices->devices,
2815                                                devid, uuid);
2816                         if (device)
2817                                 return device;
2818                 }
2819                 cur_devices = cur_devices->seed;
2820         }
2821         return NULL;
2822 }
2823
2824 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2825                                             u64 devid, u8 *dev_uuid)
2826 {
2827         struct btrfs_device *device;
2828         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2829
2830         device = kzalloc(sizeof(*device), GFP_NOFS);
2831         if (!device)
2832                 return NULL;
2833         list_add(&device->dev_list,
2834                  &fs_devices->devices);
2835         device->barriers = 1;
2836         device->dev_root = root->fs_info->dev_root;
2837         device->devid = devid;
2838         device->work.func = pending_bios_fn;
2839         fs_devices->num_devices++;
2840         spin_lock_init(&device->io_lock);
2841         INIT_LIST_HEAD(&device->dev_alloc_list);
2842         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2843         return device;
2844 }
2845
2846 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2847                           struct extent_buffer *leaf,
2848                           struct btrfs_chunk *chunk)
2849 {
2850         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2851         struct map_lookup *map;
2852         struct extent_map *em;
2853         u64 logical;
2854         u64 length;
2855         u64 devid;
2856         u8 uuid[BTRFS_UUID_SIZE];
2857         int num_stripes;
2858         int ret;
2859         int i;
2860
2861         logical = key->offset;
2862         length = btrfs_chunk_length(leaf, chunk);
2863
2864         spin_lock(&map_tree->map_tree.lock);
2865         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2866         spin_unlock(&map_tree->map_tree.lock);
2867
2868         /* already mapped? */
2869         if (em && em->start <= logical && em->start + em->len > logical) {
2870                 free_extent_map(em);
2871                 return 0;
2872         } else if (em) {
2873                 free_extent_map(em);
2874         }
2875
2876         map = kzalloc(sizeof(*map), GFP_NOFS);
2877         if (!map)
2878                 return -ENOMEM;
2879
2880         em = alloc_extent_map(GFP_NOFS);
2881         if (!em)
2882                 return -ENOMEM;
2883         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2884         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2885         if (!map) {
2886                 free_extent_map(em);
2887                 return -ENOMEM;
2888         }
2889
2890         em->bdev = (struct block_device *)map;
2891         em->start = logical;
2892         em->len = length;
2893         em->block_start = 0;
2894         em->block_len = em->len;
2895
2896         map->num_stripes = num_stripes;
2897         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2898         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2899         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2900         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2901         map->type = btrfs_chunk_type(leaf, chunk);
2902         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2903         for (i = 0; i < num_stripes; i++) {
2904                 map->stripes[i].physical =
2905                         btrfs_stripe_offset_nr(leaf, chunk, i);
2906                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2907                 read_extent_buffer(leaf, uuid, (unsigned long)
2908                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2909                                    BTRFS_UUID_SIZE);
2910                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2911                                                         NULL);
2912                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2913                         kfree(map);
2914                         free_extent_map(em);
2915                         return -EIO;
2916                 }
2917                 if (!map->stripes[i].dev) {
2918                         map->stripes[i].dev =
2919                                 add_missing_dev(root, devid, uuid);
2920                         if (!map->stripes[i].dev) {
2921                                 kfree(map);
2922                                 free_extent_map(em);
2923                                 return -EIO;
2924                         }
2925                 }
2926                 map->stripes[i].dev->in_fs_metadata = 1;
2927         }
2928
2929         spin_lock(&map_tree->map_tree.lock);
2930         ret = add_extent_mapping(&map_tree->map_tree, em);
2931         spin_unlock(&map_tree->map_tree.lock);
2932         BUG_ON(ret);
2933         free_extent_map(em);
2934
2935         return 0;
2936 }
2937
2938 static int fill_device_from_item(struct extent_buffer *leaf,
2939                                  struct btrfs_dev_item *dev_item,
2940                                  struct btrfs_device *device)
2941 {
2942         unsigned long ptr;
2943
2944         device->devid = btrfs_device_id(leaf, dev_item);
2945         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2946         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2947         device->type = btrfs_device_type(leaf, dev_item);
2948         device->io_align = btrfs_device_io_align(leaf, dev_item);
2949         device->io_width = btrfs_device_io_width(leaf, dev_item);
2950         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2951
2952         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2953         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2954
2955         return 0;
2956 }
2957
2958 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2959 {
2960         struct btrfs_fs_devices *fs_devices;
2961         int ret;
2962
2963         mutex_lock(&uuid_mutex);
2964
2965         fs_devices = root->fs_info->fs_devices->seed;
2966         while (fs_devices) {
2967                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2968                         ret = 0;
2969                         goto out;
2970                 }
2971                 fs_devices = fs_devices->seed;
2972         }
2973
2974         fs_devices = find_fsid(fsid);
2975         if (!fs_devices) {
2976                 ret = -ENOENT;
2977                 goto out;
2978         }
2979         if (fs_devices->opened) {
2980                 ret = -EBUSY;
2981                 goto out;
2982         }
2983
2984         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
2985                                    root->fs_info->bdev_holder);
2986         if (ret)
2987                 goto out;
2988
2989         if (!fs_devices->seeding) {
2990                 __btrfs_close_devices(fs_devices);
2991                 ret = -EINVAL;
2992                 goto out;
2993         }
2994
2995         fs_devices->seed = root->fs_info->fs_devices->seed;
2996         root->fs_info->fs_devices->seed = fs_devices;
2997         fs_devices->sprouted = 1;
2998 out:
2999         mutex_unlock(&uuid_mutex);
3000         return ret;
3001 }
3002
3003 static int read_one_dev(struct btrfs_root *root,
3004                         struct extent_buffer *leaf,
3005                         struct btrfs_dev_item *dev_item)
3006 {
3007         struct btrfs_device *device;
3008         u64 devid;
3009         int ret;
3010         int seed_devices = 0;
3011         u8 fs_uuid[BTRFS_UUID_SIZE];
3012         u8 dev_uuid[BTRFS_UUID_SIZE];
3013
3014         devid = btrfs_device_id(leaf, dev_item);
3015         read_extent_buffer(leaf, dev_uuid,
3016                            (unsigned long)btrfs_device_uuid(dev_item),
3017                            BTRFS_UUID_SIZE);
3018         read_extent_buffer(leaf, fs_uuid,
3019                            (unsigned long)btrfs_device_fsid(dev_item),
3020                            BTRFS_UUID_SIZE);
3021
3022         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3023                 ret = open_seed_devices(root, fs_uuid);
3024                 if (ret)
3025                         return ret;
3026                 seed_devices = 1;
3027         }
3028
3029         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3030         if (!device || !device->bdev) {
3031                 if (!btrfs_test_opt(root, DEGRADED) || seed_devices)
3032                         return -EIO;
3033
3034                 if (!device) {
3035                         printk("warning devid %Lu missing\n", devid);
3036                         device = add_missing_dev(root, devid, dev_uuid);
3037                         if (!device)
3038                                 return -ENOMEM;
3039                 }
3040         }
3041
3042         if (device->fs_devices != root->fs_info->fs_devices) {
3043                 BUG_ON(device->writeable);
3044                 if (device->generation !=
3045                     btrfs_device_generation(leaf, dev_item))
3046                         return -EINVAL;
3047         }
3048
3049         fill_device_from_item(leaf, dev_item, device);
3050         device->dev_root = root->fs_info->dev_root;
3051         device->in_fs_metadata = 1;
3052         if (device->writeable)
3053                 device->fs_devices->total_rw_bytes += device->total_bytes;
3054         ret = 0;
3055 #if 0
3056         ret = btrfs_open_device(device);
3057         if (ret) {
3058                 kfree(device);
3059         }
3060 #endif
3061         return ret;
3062 }
3063
3064 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3065 {
3066         struct btrfs_dev_item *dev_item;
3067
3068         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3069                                                      dev_item);
3070         return read_one_dev(root, buf, dev_item);
3071 }
3072
3073 int btrfs_read_sys_array(struct btrfs_root *root, u64 sb_bytenr)
3074 {
3075         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3076         struct extent_buffer *sb;
3077         struct btrfs_disk_key *disk_key;
3078         struct btrfs_chunk *chunk;
3079         u8 *ptr;
3080         unsigned long sb_ptr;
3081         int ret = 0;
3082         u32 num_stripes;
3083         u32 array_size;
3084         u32 len = 0;
3085         u32 cur;
3086         struct btrfs_key key;
3087
3088         sb = btrfs_find_create_tree_block(root, sb_bytenr,
3089                                           BTRFS_SUPER_INFO_SIZE);
3090         if (!sb)
3091                 return -ENOMEM;
3092         btrfs_set_buffer_uptodate(sb);
3093         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3094         array_size = btrfs_super_sys_array_size(super_copy);
3095
3096         ptr = super_copy->sys_chunk_array;
3097         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3098         cur = 0;
3099
3100         while (cur < array_size) {
3101                 disk_key = (struct btrfs_disk_key *)ptr;
3102                 btrfs_disk_key_to_cpu(&key, disk_key);
3103
3104                 len = sizeof(*disk_key); ptr += len;
3105                 sb_ptr += len;
3106                 cur += len;
3107
3108                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3109                         chunk = (struct btrfs_chunk *)sb_ptr;
3110                         ret = read_one_chunk(root, &key, sb, chunk);
3111                         if (ret)
3112                                 break;
3113                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3114                         len = btrfs_chunk_item_size(num_stripes);
3115                 } else {
3116                         ret = -EIO;
3117                         break;
3118                 }
3119                 ptr += len;
3120                 sb_ptr += len;
3121                 cur += len;
3122         }
3123         free_extent_buffer(sb);
3124         return ret;
3125 }
3126
3127 int btrfs_read_chunk_tree(struct btrfs_root *root)
3128 {
3129         struct btrfs_path *path;
3130         struct extent_buffer *leaf;
3131         struct btrfs_key key;
3132         struct btrfs_key found_key;
3133         int ret;
3134         int slot;
3135
3136         root = root->fs_info->chunk_root;
3137
3138         path = btrfs_alloc_path();
3139         if (!path)
3140                 return -ENOMEM;
3141
3142         /* first we search for all of the device items, and then we
3143          * read in all of the chunk items.  This way we can create chunk
3144          * mappings that reference all of the devices that are afound
3145          */
3146         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3147         key.offset = 0;
3148         key.type = 0;
3149 again:
3150         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3151         while(1) {
3152                 leaf = path->nodes[0];
3153                 slot = path->slots[0];
3154                 if (slot >= btrfs_header_nritems(leaf)) {
3155                         ret = btrfs_next_leaf(root, path);
3156                         if (ret == 0)
3157                                 continue;
3158                         if (ret < 0)
3159                                 goto error;
3160                         break;
3161                 }
3162                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3163                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3164                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3165                                 break;
3166                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3167                                 struct btrfs_dev_item *dev_item;
3168                                 dev_item = btrfs_item_ptr(leaf, slot,
3169                                                   struct btrfs_dev_item);
3170                                 ret = read_one_dev(root, leaf, dev_item);
3171                                 if (ret)
3172                                         goto error;
3173                         }
3174                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3175                         struct btrfs_chunk *chunk;
3176                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3177                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3178                         if (ret)
3179                                 goto error;
3180                 }
3181                 path->slots[0]++;
3182         }
3183         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3184                 key.objectid = 0;
3185                 btrfs_release_path(root, path);
3186                 goto again;
3187         }
3188         ret = 0;
3189 error:
3190         btrfs_free_path(path);
3191         return ret;
3192 }