git.oblomov.eu Git - 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 <asm/div64.h>
  24 #include "compat.h"
  25 #include "ctree.h"
  26 #include "extent_map.h"
  27 #include "disk-io.h"
  28 #include "transaction.h"
  29 #include "print-tree.h"
  30 #include "volumes.h"
  31 #include "async-thread.h"
  32
  33 struct map_lookup {
  34         u64 type;
  35         int io_align;
  36         int io_width;
  37         int stripe_len;
  38         int sector_size;
  39         int num_stripes;
  40         int sub_stripes;
  41         struct btrfs_bio_stripe stripes[];
  42 };
  43
  44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45                                 struct btrfs_root *root,
  46                                 struct btrfs_device *device);
  47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48
  49 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  50                             (sizeof(struct btrfs_bio_stripe) * (n)))
  51
  52 static DEFINE_MUTEX(uuid_mutex);
  53 static LIST_HEAD(fs_uuids);
  54
  55 void btrfs_lock_volumes(void)
  56 {
  57         mutex_lock(&uuid_mutex);
  58 }
  59
  60 void btrfs_unlock_volumes(void)
  61 {
  62         mutex_unlock(&uuid_mutex);
  63 }
  64
  65 static void lock_chunks(struct btrfs_root *root)
  66 {
  67         mutex_lock(&root->fs_info->chunk_mutex);
  68 }
  69
  70 static void unlock_chunks(struct btrfs_root *root)
  71 {
  72         mutex_unlock(&root->fs_info->chunk_mutex);
  73 }
  74
  75 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  76 {
  77         struct btrfs_device *device;
  78         WARN_ON(fs_devices->opened);
  79         while (!list_empty(&fs_devices->devices)) {
  80                 device = list_entry(fs_devices->devices.next,
  81                                     struct btrfs_device, dev_list);
  82                 list_del(&device->dev_list);
  83                 kfree(device->name);
  84                 kfree(device);
  85         }
  86         kfree(fs_devices);
  87 }
  88
  89 int btrfs_cleanup_fs_uuids(void)
  90 {
  91         struct btrfs_fs_devices *fs_devices;
  92
  93         while (!list_empty(&fs_uuids)) {
  94                 fs_devices = list_entry(fs_uuids.next,
  95                                         struct btrfs_fs_devices, list);
  96                 list_del(&fs_devices->list);
  97                 free_fs_devices(fs_devices);
  98         }
  99         return 0;
 100 }
 101
 102 static noinline struct btrfs_device *__find_device(struct list_head *head,
 103                                                    u64 devid, u8 *uuid)
 104 {
 105         struct btrfs_device *dev;
 106
 107         list_for_each_entry(dev, head, dev_list) {
 108                 if (dev->devid == devid &&
 109                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 110                         return dev;
 111                 }
 112         }
 113         return NULL;
 114 }
 115
 116 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 117 {
 118         struct btrfs_fs_devices *fs_devices;
 119
 120         list_for_each_entry(fs_devices, &fs_uuids, list) {
 121                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 122                         return fs_devices;
 123         }
 124         return NULL;
 125 }
 126
 127 /*
 128  * we try to collect pending bios for a device so we don't get a large
 129  * number of procs sending bios down to the same device.  This greatly
 130  * improves the schedulers ability to collect and merge the bios.
 131  *
 132  * But, it also turns into a long list of bios to process and that is sure
 133  * to eventually make the worker thread block.  The solution here is to
 134  * make some progress and then put this work struct back at the end of
 135  * the list if the block device is congested.  This way, multiple devices
 136  * can make progress from a single worker thread.
 137  */
 138 static noinline int run_scheduled_bios(struct btrfs_device *device)
 139 {
 140         struct bio *pending;
 141         struct backing_dev_info *bdi;
 142         struct btrfs_fs_info *fs_info;
 143         struct bio *tail;
 144         struct bio *cur;
 145         int again = 0;
 146         unsigned long num_run = 0;
 147         unsigned long limit;
 148
 149         bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
 150         fs_info = device->dev_root->fs_info;
 151         limit = btrfs_async_submit_limit(fs_info);
 152         limit = limit * 2 / 3;
 153
 154 loop:
 155         spin_lock(&device->io_lock);
 156
 157 loop_lock:
 158         /* take all the bios off the list at once and process them
 159          * later on (without the lock held).  But, remember the
 160          * tail and other pointers so the bios can be properly reinserted
 161          * into the list if we hit congestion
 162          */
 163         pending = device->pending_bios;
 164         tail = device->pending_bio_tail;
 165         WARN_ON(pending && !tail);
 166         device->pending_bios = NULL;
 167         device->pending_bio_tail = NULL;
 168
 169         /*
 170          * if pending was null this time around, no bios need processing
 171          * at all and we can stop.  Otherwise it'll loop back up again
 172          * and do an additional check so no bios are missed.
 173          *
 174          * device->running_pending is used to synchronize with the
 175          * schedule_bio code.
 176          */
 177         if (pending) {
 178                 again = 1;
 179                 device->running_pending = 1;
 180         } else {
 181                 again = 0;
 182                 device->running_pending = 0;
 183         }
 184         spin_unlock(&device->io_lock);
 185
 186         while (pending) {
 187                 cur = pending;
 188                 pending = pending->bi_next;
 189                 cur->bi_next = NULL;
 190                 atomic_dec(&fs_info->nr_async_bios);
 191
 192                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 193                     waitqueue_active(&fs_info->async_submit_wait))
 194                         wake_up(&fs_info->async_submit_wait);
 195
 196                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 197                 bio_get(cur);
 198                 submit_bio(cur->bi_rw, cur);
 199                 bio_put(cur);
 200                 num_run++;
 201
 202                 /*
 203                  * we made progress, there is more work to do and the bdi
 204                  * is now congested.  Back off and let other work structs
 205                  * run instead
 206                  */
 207                 if (pending && bdi_write_congested(bdi) && num_run > 16 &&
 208                     fs_info->fs_devices->open_devices > 1) {
 209                         struct bio *old_head;
 210
 211                         spin_lock(&device->io_lock);
 212
 213                         old_head = device->pending_bios;
 214                         device->pending_bios = pending;
 215                         if (device->pending_bio_tail)
 216                                 tail->bi_next = old_head;
 217                         else
 218                                 device->pending_bio_tail = tail;
 219
 220                         device->running_pending = 1;
 221
 222                         spin_unlock(&device->io_lock);
 223                         btrfs_requeue_work(&device->work);
 224                         goto done;
 225                 }
 226         }
 227         if (again)
 228                 goto loop;
 229
 230         spin_lock(&device->io_lock);
 231         if (device->pending_bios)
 232                 goto loop_lock;
 233         spin_unlock(&device->io_lock);
 234 done:
 235         return 0;
 236 }
 237
 238 static void pending_bios_fn(struct btrfs_work *work)
 239 {
 240         struct btrfs_device *device;
 241
 242         device = container_of(work, struct btrfs_device, work);
 243         run_scheduled_bios(device);
 244 }
 245
 246 static noinline int device_list_add(const char *path,
 247                            struct btrfs_super_block *disk_super,
 248                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 249 {
 250         struct btrfs_device *device;
 251         struct btrfs_fs_devices *fs_devices;
 252         u64 found_transid = btrfs_super_generation(disk_super);
 253
 254         fs_devices = find_fsid(disk_super->fsid);
 255         if (!fs_devices) {
 256                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 257                 if (!fs_devices)
 258                         return -ENOMEM;
 259                 INIT_LIST_HEAD(&fs_devices->devices);
 260                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 261                 list_add(&fs_devices->list, &fs_uuids);
 262                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 263                 fs_devices->latest_devid = devid;
 264                 fs_devices->latest_trans = found_transid;
 265                 device = NULL;
 266         } else {
 267                 device = __find_device(&fs_devices->devices, devid,
 268                                        disk_super->dev_item.uuid);
 269         }
 270         if (!device) {
 271                 if (fs_devices->opened)
 272                         return -EBUSY;
 273
 274                 device = kzalloc(sizeof(*device), GFP_NOFS);
 275                 if (!device) {
 276                         /* we can safely leave the fs_devices entry around */
 277                         return -ENOMEM;
 278                 }
 279                 device->devid = devid;
 280                 device->work.func = pending_bios_fn;
 281                 memcpy(device->uuid, disk_super->dev_item.uuid,
 282                        BTRFS_UUID_SIZE);
 283                 device->barriers = 1;
 284                 spin_lock_init(&device->io_lock);
 285                 device->name = kstrdup(path, GFP_NOFS);
 286                 if (!device->name) {
 287                         kfree(device);
 288                         return -ENOMEM;
 289                 }
 290                 INIT_LIST_HEAD(&device->dev_alloc_list);
 291                 list_add(&device->dev_list, &fs_devices->devices);
 292                 device->fs_devices = fs_devices;
 293                 fs_devices->num_devices++;
 294         }
 295
 296         if (found_transid > fs_devices->latest_trans) {
 297                 fs_devices->latest_devid = devid;
 298                 fs_devices->latest_trans = found_transid;
 299         }
 300         *fs_devices_ret = fs_devices;
 301         return 0;
 302 }
 303
 304 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 305 {
 306         struct btrfs_fs_devices *fs_devices;
 307         struct btrfs_device *device;
 308         struct btrfs_device *orig_dev;
 309
 310         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 311         if (!fs_devices)
 312                 return ERR_PTR(-ENOMEM);
 313
 314         INIT_LIST_HEAD(&fs_devices->devices);
 315         INIT_LIST_HEAD(&fs_devices->alloc_list);
 316         INIT_LIST_HEAD(&fs_devices->list);
 317         fs_devices->latest_devid = orig->latest_devid;
 318         fs_devices->latest_trans = orig->latest_trans;
 319         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 320
 321         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 322                 device = kzalloc(sizeof(*device), GFP_NOFS);
 323                 if (!device)
 324                         goto error;
 325
 326                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 327                 if (!device->name)
 328                         goto error;
 329
 330                 device->devid = orig_dev->devid;
 331                 device->work.func = pending_bios_fn;
 332                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 333                 device->barriers = 1;
 334                 spin_lock_init(&device->io_lock);
 335                 INIT_LIST_HEAD(&device->dev_list);
 336                 INIT_LIST_HEAD(&device->dev_alloc_list);
 337
 338                 list_add(&device->dev_list, &fs_devices->devices);
 339                 device->fs_devices = fs_devices;
 340                 fs_devices->num_devices++;
 341         }
 342         return fs_devices;
 343 error:
 344         free_fs_devices(fs_devices);
 345         return ERR_PTR(-ENOMEM);
 346 }
 347
 348 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 349 {
 350         struct btrfs_device *device, *next;
 351
 352         mutex_lock(&uuid_mutex);
 353 again:
 354         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 355                 if (device->in_fs_metadata)
 356                         continue;
 357
 358                 if (device->bdev) {
 359                         close_bdev_exclusive(device->bdev, device->mode);
 360                         device->bdev = NULL;
 361                         fs_devices->open_devices--;
 362                 }
 363                 if (device->writeable) {
 364                         list_del_init(&device->dev_alloc_list);
 365                         device->writeable = 0;
 366                         fs_devices->rw_devices--;
 367                 }
 368                 list_del_init(&device->dev_list);
 369                 fs_devices->num_devices--;
 370                 kfree(device->name);
 371                 kfree(device);
 372         }
 373
 374         if (fs_devices->seed) {
 375                 fs_devices = fs_devices->seed;
 376                 goto again;
 377         }
 378
 379         mutex_unlock(&uuid_mutex);
 380         return 0;
 381 }
 382
 383 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 384 {
 385         struct btrfs_device *device;
 386
 387         if (--fs_devices->opened > 0)
 388                 return 0;
 389
 390         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 391                 if (device->bdev) {
 392                         close_bdev_exclusive(device->bdev, device->mode);
 393                         fs_devices->open_devices--;
 394                 }
 395                 if (device->writeable) {
 396                         list_del_init(&device->dev_alloc_list);
 397                         fs_devices->rw_devices--;
 398                 }
 399
 400                 device->bdev = NULL;
 401                 device->writeable = 0;
 402                 device->in_fs_metadata = 0;
 403         }
 404         WARN_ON(fs_devices->open_devices);
 405         WARN_ON(fs_devices->rw_devices);
 406         fs_devices->opened = 0;
 407         fs_devices->seeding = 0;
 408
 409         return 0;
 410 }
 411
 412 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 413 {
 414         struct btrfs_fs_devices *seed_devices = NULL;
 415         int ret;
 416
 417         mutex_lock(&uuid_mutex);
 418         ret = __btrfs_close_devices(fs_devices);
 419         if (!fs_devices->opened) {
 420                 seed_devices = fs_devices->seed;
 421                 fs_devices->seed = NULL;
 422         }
 423         mutex_unlock(&uuid_mutex);
 424
 425         while (seed_devices) {
 426                 fs_devices = seed_devices;
 427                 seed_devices = fs_devices->seed;
 428                 __btrfs_close_devices(fs_devices);
 429                 free_fs_devices(fs_devices);
 430         }
 431         return ret;
 432 }
 433
 434 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 435                                 fmode_t flags, void *holder)
 436 {
 437         struct block_device *bdev;
 438         struct list_head *head = &fs_devices->devices;
 439         struct btrfs_device *device;
 440         struct block_device *latest_bdev = NULL;
 441         struct buffer_head *bh;
 442         struct btrfs_super_block *disk_super;
 443         u64 latest_devid = 0;
 444         u64 latest_transid = 0;
 445         u64 devid;
 446         int seeding = 1;
 447         int ret = 0;
 448
 449         list_for_each_entry(device, head, dev_list) {
 450                 if (device->bdev)
 451                         continue;
 452                 if (!device->name)
 453                         continue;
 454
 455                 bdev = open_bdev_exclusive(device->name, flags, holder);
 456                 if (IS_ERR(bdev)) {
 457                         printk(KERN_INFO "open %s failed\n", device->name);
 458                         goto error;
 459                 }
 460                 set_blocksize(bdev, 4096);
 461
 462                 bh = btrfs_read_dev_super(bdev);
 463                 if (!bh)
 464                         goto error_close;
 465
 466                 disk_super = (struct btrfs_super_block *)bh->b_data;
 467                 devid = le64_to_cpu(disk_super->dev_item.devid);
 468                 if (devid != device->devid)
 469                         goto error_brelse;
 470
 471                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 472                            BTRFS_UUID_SIZE))
 473                         goto error_brelse;
 474
 475                 device->generation = btrfs_super_generation(disk_super);
 476                 if (!latest_transid || device->generation > latest_transid) {
 477                         latest_devid = devid;
 478                         latest_transid = device->generation;
 479                         latest_bdev = bdev;
 480                 }
 481
 482                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 483                         device->writeable = 0;
 484                 } else {
 485                         device->writeable = !bdev_read_only(bdev);
 486                         seeding = 0;
 487                 }
 488
 489                 device->bdev = bdev;
 490                 device->in_fs_metadata = 0;
 491                 device->mode = flags;
 492
 493                 fs_devices->open_devices++;
 494                 if (device->writeable) {
 495                         fs_devices->rw_devices++;
 496                         list_add(&device->dev_alloc_list,
 497                                  &fs_devices->alloc_list);
 498                 }
 499                 continue;
 500
 501 error_brelse:
 502                 brelse(bh);
 503 error_close:
 504                 close_bdev_exclusive(bdev, FMODE_READ);
 505 error:
 506                 continue;
 507         }
 508         if (fs_devices->open_devices == 0) {
 509                 ret = -EIO;
 510                 goto out;
 511         }
 512         fs_devices->seeding = seeding;
 513         fs_devices->opened = 1;
 514         fs_devices->latest_bdev = latest_bdev;
 515         fs_devices->latest_devid = latest_devid;
 516         fs_devices->latest_trans = latest_transid;
 517         fs_devices->total_rw_bytes = 0;
 518 out:
 519         return ret;
 520 }
 521
 522 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 523                        fmode_t flags, void *holder)
 524 {
 525         int ret;
 526
 527         mutex_lock(&uuid_mutex);
 528         if (fs_devices->opened) {
 529                 fs_devices->opened++;
 530                 ret = 0;
 531         } else {
 532                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 533         }
 534         mutex_unlock(&uuid_mutex);
 535         return ret;
 536 }
 537
 538 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 539                           struct btrfs_fs_devices **fs_devices_ret)
 540 {
 541         struct btrfs_super_block *disk_super;
 542         struct block_device *bdev;
 543         struct buffer_head *bh;
 544         int ret;
 545         u64 devid;
 546         u64 transid;
 547
 548         mutex_lock(&uuid_mutex);
 549
 550         bdev = open_bdev_exclusive(path, flags, holder);
 551
 552         if (IS_ERR(bdev)) {
 553                 ret = PTR_ERR(bdev);
 554                 goto error;
 555         }
 556
 557         ret = set_blocksize(bdev, 4096);
 558         if (ret)
 559                 goto error_close;
 560         bh = btrfs_read_dev_super(bdev);
 561         if (!bh) {
 562                 ret = -EIO;
 563                 goto error_close;
 564         }
 565         disk_super = (struct btrfs_super_block *)bh->b_data;
 566         devid = le64_to_cpu(disk_super->dev_item.devid);
 567         transid = btrfs_super_generation(disk_super);
 568         if (disk_super->label[0])
 569                 printk(KERN_INFO "device label %s ", disk_super->label);
 570         else {
 571                 /* FIXME, make a readl uuid parser */
 572                 printk(KERN_INFO "device fsid %llx-%llx ",
 573                        *(unsigned long long *)disk_super->fsid,
 574                        *(unsigned long long *)(disk_super->fsid + 8));
 575         }
 576         printk(KERN_CONT "devid %llu transid %llu %s\n",
 577                (unsigned long long)devid, (unsigned long long)transid, path);
 578         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 579
 580         brelse(bh);
 581 error_close:
 582         close_bdev_exclusive(bdev, flags);
 583 error:
 584         mutex_unlock(&uuid_mutex);
 585         return ret;
 586 }
 587
 588 /*
 589  * this uses a pretty simple search, the expectation is that it is
 590  * called very infrequently and that a given device has a small number
 591  * of extents
 592  */
 593 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
 594                                          struct btrfs_device *device,
 595                                          u64 num_bytes, u64 *start)
 596 {
 597         struct btrfs_key key;
 598         struct btrfs_root *root = device->dev_root;
 599         struct btrfs_dev_extent *dev_extent = NULL;
 600         struct btrfs_path *path;
 601         u64 hole_size = 0;
 602         u64 last_byte = 0;
 603         u64 search_start = 0;
 604         u64 search_end = device->total_bytes;
 605         int ret;
 606         int slot = 0;
 607         int start_found;
 608         struct extent_buffer *l;
 609
 610         path = btrfs_alloc_path();
 611         if (!path)
 612                 return -ENOMEM;
 613         path->reada = 2;
 614         start_found = 0;
 615
 616         /* FIXME use last free of some kind */
 617
 618         /* we don't want to overwrite the superblock on the drive,
 619          * so we make sure to start at an offset of at least 1MB
 620          */
 621         search_start = max((u64)1024 * 1024, search_start);
 622
 623         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 624                 search_start = max(root->fs_info->alloc_start, search_start);
 625
 626         key.objectid = device->devid;
 627         key.offset = search_start;
 628         key.type = BTRFS_DEV_EXTENT_KEY;
 629         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 630         if (ret < 0)
 631                 goto error;
 632         ret = btrfs_previous_item(root, path, 0, key.type);
 633         if (ret < 0)
 634                 goto error;
 635         l = path->nodes[0];
 636         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 637         while (1) {
 638                 l = path->nodes[0];
 639                 slot = path->slots[0];
 640                 if (slot >= btrfs_header_nritems(l)) {
 641                         ret = btrfs_next_leaf(root, path);
 642                         if (ret == 0)
 643                                 continue;
 644                         if (ret < 0)
 645                                 goto error;
 646 no_more_items:
 647                         if (!start_found) {
 648                                 if (search_start >= search_end) {
 649                                         ret = -ENOSPC;
 650                                         goto error;
 651                                 }
 652                                 *start = search_start;
 653                                 start_found = 1;
 654                                 goto check_pending;
 655                         }
 656                         *start = last_byte > search_start ?
 657                                 last_byte : search_start;
 658                         if (search_end <= *start) {
 659                                 ret = -ENOSPC;
 660                                 goto error;
 661                         }
 662                         goto check_pending;
 663                 }
 664                 btrfs_item_key_to_cpu(l, &key, slot);
 665
 666                 if (key.objectid < device->devid)
 667                         goto next;
 668
 669                 if (key.objectid > device->devid)
 670                         goto no_more_items;
 671
 672                 if (key.offset >= search_start && key.offset > last_byte &&
 673                     start_found) {
 674                         if (last_byte < search_start)
 675                                 last_byte = search_start;
 676                         hole_size = key.offset - last_byte;
 677                         if (key.offset > last_byte &&
 678                             hole_size >= num_bytes) {
 679                                 *start = last_byte;
 680                                 goto check_pending;
 681                         }
 682                 }
 683                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 684                         goto next;
 685
 686                 start_found = 1;
 687                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 688                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 689 next:
 690                 path->slots[0]++;
 691                 cond_resched();
 692         }
 693 check_pending:
 694         /* we have to make sure we didn't find an extent that has already
 695          * been allocated by the map tree or the original allocation
 696          */
 697         BUG_ON(*start < search_start);
 698
 699         if (*start + num_bytes > search_end) {
 700                 ret = -ENOSPC;
 701                 goto error;
 702         }
 703         /* check for pending inserts here */
 704         ret = 0;
 705
 706 error:
 707         btrfs_free_path(path);
 708         return ret;
 709 }
 710
 711 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 712                           struct btrfs_device *device,
 713                           u64 start)
 714 {
 715         int ret;
 716         struct btrfs_path *path;
 717         struct btrfs_root *root = device->dev_root;
 718         struct btrfs_key key;
 719         struct btrfs_key found_key;
 720         struct extent_buffer *leaf = NULL;
 721         struct btrfs_dev_extent *extent = NULL;
 722
 723         path = btrfs_alloc_path();
 724         if (!path)
 725                 return -ENOMEM;
 726
 727         key.objectid = device->devid;
 728         key.offset = start;
 729         key.type = BTRFS_DEV_EXTENT_KEY;
 730
 731         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 732         if (ret > 0) {
 733                 ret = btrfs_previous_item(root, path, key.objectid,
 734                                           BTRFS_DEV_EXTENT_KEY);
 735                 BUG_ON(ret);
 736                 leaf = path->nodes[0];
 737                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 738                 extent = btrfs_item_ptr(leaf, path->slots[0],
 739                                         struct btrfs_dev_extent);
 740                 BUG_ON(found_key.offset > start || found_key.offset +
 741                        btrfs_dev_extent_length(leaf, extent) < start);
 742                 ret = 0;
 743         } else if (ret == 0) {
 744                 leaf = path->nodes[0];
 745                 extent = btrfs_item_ptr(leaf, path->slots[0],
 746                                         struct btrfs_dev_extent);
 747         }
 748         BUG_ON(ret);
 749
 750         if (device->bytes_used > 0)
 751                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 752         ret = btrfs_del_item(trans, root, path);
 753         BUG_ON(ret);
 754
 755         btrfs_free_path(path);
 756         return ret;
 757 }
 758
 759 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 760                            struct btrfs_device *device,
 761                            u64 chunk_tree, u64 chunk_objectid,
 762                            u64 chunk_offset, u64 start, u64 num_bytes)
 763 {
 764         int ret;
 765         struct btrfs_path *path;
 766         struct btrfs_root *root = device->dev_root;
 767         struct btrfs_dev_extent *extent;
 768         struct extent_buffer *leaf;
 769         struct btrfs_key key;
 770
 771         WARN_ON(!device->in_fs_metadata);
 772         path = btrfs_alloc_path();
 773         if (!path)
 774                 return -ENOMEM;
 775
 776         key.objectid = device->devid;
 777         key.offset = start;
 778         key.type = BTRFS_DEV_EXTENT_KEY;
 779         ret = btrfs_insert_empty_item(trans, root, path, &key,
 780                                       sizeof(*extent));
 781         BUG_ON(ret);
 782
 783         leaf = path->nodes[0];
 784         extent = btrfs_item_ptr(leaf, path->slots[0],
 785                                 struct btrfs_dev_extent);
 786         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 787         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 788         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 789
 790         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 791                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 792                     BTRFS_UUID_SIZE);
 793
 794         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 795         btrfs_mark_buffer_dirty(leaf);
 796         btrfs_free_path(path);
 797         return ret;
 798 }
 799
 800 static noinline int find_next_chunk(struct btrfs_root *root,
 801                                     u64 objectid, u64 *offset)
 802 {
 803         struct btrfs_path *path;
 804         int ret;
 805         struct btrfs_key key;
 806         struct btrfs_chunk *chunk;
 807         struct btrfs_key found_key;
 808
 809         path = btrfs_alloc_path();
 810         BUG_ON(!path);
 811
 812         key.objectid = objectid;
 813         key.offset = (u64)-1;
 814         key.type = BTRFS_CHUNK_ITEM_KEY;
 815
 816         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 817         if (ret < 0)
 818                 goto error;
 819
 820         BUG_ON(ret == 0);
 821
 822         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 823         if (ret) {
 824                 *offset = 0;
 825         } else {
 826                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 827                                       path->slots[0]);
 828                 if (found_key.objectid != objectid)
 829                         *offset = 0;
 830                 else {
 831                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 832                                                struct btrfs_chunk);
 833                         *offset = found_key.offset +
 834                                 btrfs_chunk_length(path->nodes[0], chunk);
 835                 }
 836         }
 837         ret = 0;
 838 error:
 839         btrfs_free_path(path);
 840         return ret;
 841 }
 842
 843 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
 844 {
 845         int ret;
 846         struct btrfs_key key;
 847         struct btrfs_key found_key;
 848         struct btrfs_path *path;
 849
 850         root = root->fs_info->chunk_root;
 851
 852         path = btrfs_alloc_path();
 853         if (!path)
 854                 return -ENOMEM;
 855
 856         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 857         key.type = BTRFS_DEV_ITEM_KEY;
 858         key.offset = (u64)-1;
 859
 860         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 861         if (ret < 0)
 862                 goto error;
 863
 864         BUG_ON(ret == 0);
 865
 866         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 867                                   BTRFS_DEV_ITEM_KEY);
 868         if (ret) {
 869                 *objectid = 1;
 870         } else {
 871                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 872                                       path->slots[0]);
 873                 *objectid = found_key.offset + 1;
 874         }
 875         ret = 0;
 876 error:
 877         btrfs_free_path(path);
 878         return ret;
 879 }
 880
 881 /*
 882  * the device information is stored in the chunk root
 883  * the btrfs_device struct should be fully filled in
 884  */
 885 int btrfs_add_device(struct btrfs_trans_handle *trans,
 886                      struct btrfs_root *root,
 887                      struct btrfs_device *device)
 888 {
 889         int ret;
 890         struct btrfs_path *path;
 891         struct btrfs_dev_item *dev_item;
 892         struct extent_buffer *leaf;
 893         struct btrfs_key key;
 894         unsigned long ptr;
 895
 896         root = root->fs_info->chunk_root;
 897
 898         path = btrfs_alloc_path();
 899         if (!path)
 900                 return -ENOMEM;
 901
 902         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 903         key.type = BTRFS_DEV_ITEM_KEY;
 904         key.offset = device->devid;
 905
 906         ret = btrfs_insert_empty_item(trans, root, path, &key,
 907                                       sizeof(*dev_item));
 908         if (ret)
 909                 goto out;
 910
 911         leaf = path->nodes[0];
 912         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 913
 914         btrfs_set_device_id(leaf, dev_item, device->devid);
 915         btrfs_set_device_generation(leaf, dev_item, 0);
 916         btrfs_set_device_type(leaf, dev_item, device->type);
 917         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 918         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 919         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 920         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 921         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 922         btrfs_set_device_group(leaf, dev_item, 0);
 923         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 924         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 925         btrfs_set_device_start_offset(leaf, dev_item, 0);
 926
 927         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 928         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 929         ptr = (unsigned long)btrfs_device_fsid(dev_item);
 930         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
 931         btrfs_mark_buffer_dirty(leaf);
 932
 933         ret = 0;
 934 out:
 935         btrfs_free_path(path);
 936         return ret;
 937 }
 938
 939 static int btrfs_rm_dev_item(struct btrfs_root *root,
 940                              struct btrfs_device *device)
 941 {
 942         int ret;
 943         struct btrfs_path *path;
 944         struct btrfs_key key;
 945         struct btrfs_trans_handle *trans;
 946
 947         root = root->fs_info->chunk_root;
 948
 949         path = btrfs_alloc_path();
 950         if (!path)
 951                 return -ENOMEM;
 952
 953         trans = btrfs_start_transaction(root, 1);
 954         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 955         key.type = BTRFS_DEV_ITEM_KEY;
 956         key.offset = device->devid;
 957         lock_chunks(root);
 958
 959         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 960         if (ret < 0)
 961                 goto out;
 962
 963         if (ret > 0) {
 964                 ret = -ENOENT;
 965                 goto out;
 966         }
 967
 968         ret = btrfs_del_item(trans, root, path);
 969         if (ret)
 970                 goto out;
 971 out:
 972         btrfs_free_path(path);
 973         unlock_chunks(root);
 974         btrfs_commit_transaction(trans, root);
 975         return ret;
 976 }
 977
 978 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 979 {
 980         struct btrfs_device *device;
 981         struct btrfs_device *next_device;
 982         struct block_device *bdev;
 983         struct buffer_head *bh = NULL;
 984         struct btrfs_super_block *disk_super;
 985         u64 all_avail;
 986         u64 devid;
 987         u64 num_devices;
 988         u8 *dev_uuid;
 989         int ret = 0;
 990
 991         mutex_lock(&uuid_mutex);
 992         mutex_lock(&root->fs_info->volume_mutex);
 993
 994         all_avail = root->fs_info->avail_data_alloc_bits |
 995                 root->fs_info->avail_system_alloc_bits |
 996                 root->fs_info->avail_metadata_alloc_bits;
 997
 998         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 999             root->fs_info->fs_devices->rw_devices <= 4) {
1000                 printk(KERN_ERR "btrfs: unable to go below four devices "
1001                        "on raid10\n");
1002                 ret = -EINVAL;
1003                 goto out;
1004         }
1005
1006         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1007             root->fs_info->fs_devices->rw_devices <= 2) {
1008                 printk(KERN_ERR "btrfs: unable to go below two "
1009                        "devices on raid1\n");
1010                 ret = -EINVAL;
1011                 goto out;
1012         }
1013
1014         if (strcmp(device_path, "missing") == 0) {
1015                 struct list_head *devices;
1016                 struct btrfs_device *tmp;
1017
1018                 device = NULL;
1019                 devices = &root->fs_info->fs_devices->devices;
1020                 list_for_each_entry(tmp, devices, dev_list) {
1021                         if (tmp->in_fs_metadata && !tmp->bdev) {
1022                                 device = tmp;
1023                                 break;
1024                         }
1025                 }
1026                 bdev = NULL;
1027                 bh = NULL;
1028                 disk_super = NULL;
1029                 if (!device) {
1030                         printk(KERN_ERR "btrfs: no missing devices found to "
1031                                "remove\n");
1032                         goto out;
1033                 }
1034         } else {
1035                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1036                                       root->fs_info->bdev_holder);
1037                 if (IS_ERR(bdev)) {
1038                         ret = PTR_ERR(bdev);
1039                         goto out;
1040                 }
1041
1042                 set_blocksize(bdev, 4096);
1043                 bh = btrfs_read_dev_super(bdev);
1044                 if (!bh) {
1045                         ret = -EIO;
1046                         goto error_close;
1047                 }
1048                 disk_super = (struct btrfs_super_block *)bh->b_data;
1049                 devid = le64_to_cpu(disk_super->dev_item.devid);
1050                 dev_uuid = disk_super->dev_item.uuid;
1051                 device = btrfs_find_device(root, devid, dev_uuid,
1052                                            disk_super->fsid);
1053                 if (!device) {
1054                         ret = -ENOENT;
1055                         goto error_brelse;
1056                 }
1057         }
1058
1059         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1060                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1061                        "device\n");
1062                 ret = -EINVAL;
1063                 goto error_brelse;
1064         }
1065
1066         if (device->writeable) {
1067                 list_del_init(&device->dev_alloc_list);
1068                 root->fs_info->fs_devices->rw_devices--;
1069         }
1070
1071         ret = btrfs_shrink_device(device, 0);
1072         if (ret)
1073                 goto error_brelse;
1074
1075         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1076         if (ret)
1077                 goto error_brelse;
1078
1079         device->in_fs_metadata = 0;
1080         list_del_init(&device->dev_list);
1081         device->fs_devices->num_devices--;
1082
1083         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1084                                  struct btrfs_device, dev_list);
1085         if (device->bdev == root->fs_info->sb->s_bdev)
1086                 root->fs_info->sb->s_bdev = next_device->bdev;
1087         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1088                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1089
1090         if (device->bdev) {
1091                 close_bdev_exclusive(device->bdev, device->mode);
1092                 device->bdev = NULL;
1093                 device->fs_devices->open_devices--;
1094         }
1095
1096         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1097         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1098
1099         if (device->fs_devices->open_devices == 0) {
1100                 struct btrfs_fs_devices *fs_devices;
1101                 fs_devices = root->fs_info->fs_devices;
1102                 while (fs_devices) {
1103                         if (fs_devices->seed == device->fs_devices)
1104                                 break;
1105                         fs_devices = fs_devices->seed;
1106                 }
1107                 fs_devices->seed = device->fs_devices->seed;
1108                 device->fs_devices->seed = NULL;
1109                 __btrfs_close_devices(device->fs_devices);
1110                 free_fs_devices(device->fs_devices);
1111         }
1112
1113         /*
1114          * at this point, the device is zero sized.  We want to
1115          * remove it from the devices list and zero out the old super
1116          */
1117         if (device->writeable) {
1118                 /* make sure this device isn't detected as part of
1119                  * the FS anymore
1120                  */
1121                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1122                 set_buffer_dirty(bh);
1123                 sync_dirty_buffer(bh);
1124         }
1125
1126         kfree(device->name);
1127         kfree(device);
1128         ret = 0;
1129
1130 error_brelse:
1131         brelse(bh);
1132 error_close:
1133         if (bdev)
1134                 close_bdev_exclusive(bdev, FMODE_READ);
1135 out:
1136         mutex_unlock(&root->fs_info->volume_mutex);
1137         mutex_unlock(&uuid_mutex);
1138         return ret;
1139 }
1140
1141 /*
1142  * does all the dirty work required for changing file system's UUID.
1143  */
1144 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1145                                 struct btrfs_root *root)
1146 {
1147         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1148         struct btrfs_fs_devices *old_devices;
1149         struct btrfs_fs_devices *seed_devices;
1150         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1151         struct btrfs_device *device;
1152         u64 super_flags;
1153
1154         BUG_ON(!mutex_is_locked(&uuid_mutex));
1155         if (!fs_devices->seeding)
1156                 return -EINVAL;
1157
1158         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1159         if (!seed_devices)
1160                 return -ENOMEM;
1161
1162         old_devices = clone_fs_devices(fs_devices);
1163         if (IS_ERR(old_devices)) {
1164                 kfree(seed_devices);
1165                 return PTR_ERR(old_devices);
1166         }
1167
1168         list_add(&old_devices->list, &fs_uuids);
1169
1170         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1171         seed_devices->opened = 1;
1172         INIT_LIST_HEAD(&seed_devices->devices);
1173         INIT_LIST_HEAD(&seed_devices->alloc_list);
1174         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1175         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1176         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1177                 device->fs_devices = seed_devices;
1178         }
1179
1180         fs_devices->seeding = 0;
1181         fs_devices->num_devices = 0;
1182         fs_devices->open_devices = 0;
1183         fs_devices->seed = seed_devices;
1184
1185         generate_random_uuid(fs_devices->fsid);
1186         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1187         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1188         super_flags = btrfs_super_flags(disk_super) &
1189                       ~BTRFS_SUPER_FLAG_SEEDING;
1190         btrfs_set_super_flags(disk_super, super_flags);
1191
1192         return 0;
1193 }
1194
1195 /*
1196  * strore the expected generation for seed devices in device items.
1197  */
1198 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1199                                struct btrfs_root *root)
1200 {
1201         struct btrfs_path *path;
1202         struct extent_buffer *leaf;
1203         struct btrfs_dev_item *dev_item;
1204         struct btrfs_device *device;
1205         struct btrfs_key key;
1206         u8 fs_uuid[BTRFS_UUID_SIZE];
1207         u8 dev_uuid[BTRFS_UUID_SIZE];
1208         u64 devid;
1209         int ret;
1210
1211         path = btrfs_alloc_path();
1212         if (!path)
1213                 return -ENOMEM;
1214
1215         root = root->fs_info->chunk_root;
1216         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1217         key.offset = 0;
1218         key.type = BTRFS_DEV_ITEM_KEY;
1219
1220         while (1) {
1221                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1222                 if (ret < 0)
1223                         goto error;
1224
1225                 leaf = path->nodes[0];
1226 next_slot:
1227                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1228                         ret = btrfs_next_leaf(root, path);
1229                         if (ret > 0)
1230                                 break;
1231                         if (ret < 0)
1232                                 goto error;
1233                         leaf = path->nodes[0];
1234                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1235                         btrfs_release_path(root, path);
1236                         continue;
1237                 }
1238
1239                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1240                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1241                     key.type != BTRFS_DEV_ITEM_KEY)
1242                         break;
1243
1244                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1245                                           struct btrfs_dev_item);
1246                 devid = btrfs_device_id(leaf, dev_item);
1247                 read_extent_buffer(leaf, dev_uuid,
1248                                    (unsigned long)btrfs_device_uuid(dev_item),
1249                                    BTRFS_UUID_SIZE);
1250                 read_extent_buffer(leaf, fs_uuid,
1251                                    (unsigned long)btrfs_device_fsid(dev_item),
1252                                    BTRFS_UUID_SIZE);
1253                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1254                 BUG_ON(!device);
1255
1256                 if (device->fs_devices->seeding) {
1257                         btrfs_set_device_generation(leaf, dev_item,
1258                                                     device->generation);
1259                         btrfs_mark_buffer_dirty(leaf);
1260                 }
1261
1262                 path->slots[0]++;
1263                 goto next_slot;
1264         }
1265         ret = 0;
1266 error:
1267         btrfs_free_path(path);
1268         return ret;
1269 }
1270
1271 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1272 {
1273         struct btrfs_trans_handle *trans;
1274         struct btrfs_device *device;
1275         struct block_device *bdev;
1276         struct list_head *devices;
1277         struct super_block *sb = root->fs_info->sb;
1278         u64 total_bytes;
1279         int seeding_dev = 0;
1280         int ret = 0;
1281
1282         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1283                 return -EINVAL;
1284
1285         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1286         if (!bdev)
1287                 return -EIO;
1288
1289         if (root->fs_info->fs_devices->seeding) {
1290                 seeding_dev = 1;
1291                 down_write(&sb->s_umount);
1292                 mutex_lock(&uuid_mutex);
1293         }
1294
1295         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1296         mutex_lock(&root->fs_info->volume_mutex);
1297
1298         devices = &root->fs_info->fs_devices->devices;
1299         list_for_each_entry(device, devices, dev_list) {
1300                 if (device->bdev == bdev) {
1301                         ret = -EEXIST;
1302                         goto error;
1303                 }
1304         }
1305
1306         device = kzalloc(sizeof(*device), GFP_NOFS);
1307         if (!device) {
1308                 /* we can safely leave the fs_devices entry around */
1309                 ret = -ENOMEM;
1310                 goto error;
1311         }
1312
1313         device->name = kstrdup(device_path, GFP_NOFS);
1314         if (!device->name) {
1315                 kfree(device);
1316                 ret = -ENOMEM;
1317                 goto error;
1318         }
1319
1320         ret = find_next_devid(root, &device->devid);
1321         if (ret) {
1322                 kfree(device);
1323                 goto error;
1324         }
1325
1326         trans = btrfs_start_transaction(root, 1);
1327         lock_chunks(root);
1328
1329         device->barriers = 1;
1330         device->writeable = 1;
1331         device->work.func = pending_bios_fn;
1332         generate_random_uuid(device->uuid);
1333         spin_lock_init(&device->io_lock);
1334         device->generation = trans->transid;
1335         device->io_width = root->sectorsize;
1336         device->io_align = root->sectorsize;
1337         device->sector_size = root->sectorsize;
1338         device->total_bytes = i_size_read(bdev->bd_inode);
1339         device->dev_root = root->fs_info->dev_root;
1340         device->bdev = bdev;
1341         device->in_fs_metadata = 1;
1342         device->mode = 0;
1343         set_blocksize(device->bdev, 4096);
1344
1345         if (seeding_dev) {
1346                 sb->s_flags &= ~MS_RDONLY;
1347                 ret = btrfs_prepare_sprout(trans, root);
1348                 BUG_ON(ret);
1349         }
1350
1351         device->fs_devices = root->fs_info->fs_devices;
1352         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1353         list_add(&device->dev_alloc_list,
1354                  &root->fs_info->fs_devices->alloc_list);
1355         root->fs_info->fs_devices->num_devices++;
1356         root->fs_info->fs_devices->open_devices++;
1357         root->fs_info->fs_devices->rw_devices++;
1358         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1359
1360         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1361         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1362                                     total_bytes + device->total_bytes);
1363
1364         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1365         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1366                                     total_bytes + 1);
1367
1368         if (seeding_dev) {
1369                 ret = init_first_rw_device(trans, root, device);
1370                 BUG_ON(ret);
1371                 ret = btrfs_finish_sprout(trans, root);
1372                 BUG_ON(ret);
1373         } else {
1374                 ret = btrfs_add_device(trans, root, device);
1375         }
1376
1377         unlock_chunks(root);
1378         btrfs_commit_transaction(trans, root);
1379
1380         if (seeding_dev) {
1381                 mutex_unlock(&uuid_mutex);
1382                 up_write(&sb->s_umount);
1383
1384                 ret = btrfs_relocate_sys_chunks(root);
1385                 BUG_ON(ret);
1386         }
1387 out:
1388         mutex_unlock(&root->fs_info->volume_mutex);
1389         return ret;
1390 error:
1391         close_bdev_exclusive(bdev, 0);
1392         if (seeding_dev) {
1393                 mutex_unlock(&uuid_mutex);
1394                 up_write(&sb->s_umount);
1395         }
1396         goto out;
1397 }
1398
1399 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1400                                         struct btrfs_device *device)
1401 {
1402         int ret;
1403         struct btrfs_path *path;
1404         struct btrfs_root *root;
1405         struct btrfs_dev_item *dev_item;
1406         struct extent_buffer *leaf;
1407         struct btrfs_key key;
1408
1409         root = device->dev_root->fs_info->chunk_root;
1410
1411         path = btrfs_alloc_path();
1412         if (!path)
1413                 return -ENOMEM;
1414
1415         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1416         key.type = BTRFS_DEV_ITEM_KEY;
1417         key.offset = device->devid;
1418
1419         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1420         if (ret < 0)
1421                 goto out;
1422
1423         if (ret > 0) {
1424                 ret = -ENOENT;
1425                 goto out;
1426         }
1427
1428         leaf = path->nodes[0];
1429         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1430
1431         btrfs_set_device_id(leaf, dev_item, device->devid);
1432         btrfs_set_device_type(leaf, dev_item, device->type);
1433         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1434         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1435         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1436         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1437         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1438         btrfs_mark_buffer_dirty(leaf);
1439
1440 out:
1441         btrfs_free_path(path);
1442         return ret;
1443 }
1444
1445 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1446                       struct btrfs_device *device, u64 new_size)
1447 {
1448         struct btrfs_super_block *super_copy =
1449                 &device->dev_root->fs_info->super_copy;
1450         u64 old_total = btrfs_super_total_bytes(super_copy);
1451         u64 diff = new_size - device->total_bytes;
1452
1453         if (!device->writeable)
1454                 return -EACCES;
1455         if (new_size <= device->total_bytes)
1456                 return -EINVAL;
1457
1458         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1459         device->fs_devices->total_rw_bytes += diff;
1460
1461         device->total_bytes = new_size;
1462         btrfs_clear_space_info_full(device->dev_root->fs_info);
1463
1464         return btrfs_update_device(trans, device);
1465 }
1466
1467 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1468                       struct btrfs_device *device, u64 new_size)
1469 {
1470         int ret;
1471         lock_chunks(device->dev_root);
1472         ret = __btrfs_grow_device(trans, device, new_size);
1473         unlock_chunks(device->dev_root);
1474         return ret;
1475 }
1476
1477 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1478                             struct btrfs_root *root,
1479                             u64 chunk_tree, u64 chunk_objectid,
1480                             u64 chunk_offset)
1481 {
1482         int ret;
1483         struct btrfs_path *path;
1484         struct btrfs_key key;
1485
1486         root = root->fs_info->chunk_root;
1487         path = btrfs_alloc_path();
1488         if (!path)
1489                 return -ENOMEM;
1490
1491         key.objectid = chunk_objectid;
1492         key.offset = chunk_offset;
1493         key.type = BTRFS_CHUNK_ITEM_KEY;
1494
1495         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1496         BUG_ON(ret);
1497
1498         ret = btrfs_del_item(trans, root, path);
1499         BUG_ON(ret);
1500
1501         btrfs_free_path(path);
1502         return 0;
1503 }
1504
1505 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1506                         chunk_offset)
1507 {
1508         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1509         struct btrfs_disk_key *disk_key;
1510         struct btrfs_chunk *chunk;
1511         u8 *ptr;
1512         int ret = 0;
1513         u32 num_stripes;
1514         u32 array_size;
1515         u32 len = 0;
1516         u32 cur;
1517         struct btrfs_key key;
1518
1519         array_size = btrfs_super_sys_array_size(super_copy);
1520
1521         ptr = super_copy->sys_chunk_array;
1522         cur = 0;
1523
1524         while (cur < array_size) {
1525                 disk_key = (struct btrfs_disk_key *)ptr;
1526                 btrfs_disk_key_to_cpu(&key, disk_key);
1527
1528                 len = sizeof(*disk_key);
1529
1530                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1531                         chunk = (struct btrfs_chunk *)(ptr + len);
1532                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1533                         len += btrfs_chunk_item_size(num_stripes);
1534                 } else {
1535                         ret = -EIO;
1536                         break;
1537                 }
1538                 if (key.objectid == chunk_objectid &&
1539                     key.offset == chunk_offset) {
1540                         memmove(ptr, ptr + len, array_size - (cur + len));
1541                         array_size -= len;
1542                         btrfs_set_super_sys_array_size(super_copy, array_size);
1543                 } else {
1544                         ptr += len;
1545                         cur += len;
1546                 }
1547         }
1548         return ret;
1549 }
1550
1551 static int btrfs_relocate_chunk(struct btrfs_root *root,
1552                          u64 chunk_tree, u64 chunk_objectid,
1553                          u64 chunk_offset)
1554 {
1555         struct extent_map_tree *em_tree;
1556         struct btrfs_root *extent_root;
1557         struct btrfs_trans_handle *trans;
1558         struct extent_map *em;
1559         struct map_lookup *map;
1560         int ret;
1561         int i;
1562
1563         printk(KERN_INFO "btrfs relocating chunk %llu\n",
1564                (unsigned long long)chunk_offset);
1565         root = root->fs_info->chunk_root;
1566         extent_root = root->fs_info->extent_root;
1567         em_tree = &root->fs_info->mapping_tree.map_tree;
1568
1569         /* step one, relocate all the extents inside this chunk */
1570         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1571         BUG_ON(ret);
1572
1573         trans = btrfs_start_transaction(root, 1);
1574         BUG_ON(!trans);
1575
1576         lock_chunks(root);
1577
1578         /*
1579          * step two, delete the device extents and the
1580          * chunk tree entries
1581          */
1582         spin_lock(&em_tree->lock);
1583         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1584         spin_unlock(&em_tree->lock);
1585
1586         BUG_ON(em->start > chunk_offset ||
1587                em->start + em->len < chunk_offset);
1588         map = (struct map_lookup *)em->bdev;
1589
1590         for (i = 0; i < map->num_stripes; i++) {
1591                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1592                                             map->stripes[i].physical);
1593                 BUG_ON(ret);
1594
1595                 if (map->stripes[i].dev) {
1596                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1597                         BUG_ON(ret);
1598                 }
1599         }
1600         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1601                                chunk_offset);
1602
1603         BUG_ON(ret);
1604
1605         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1606                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1607                 BUG_ON(ret);
1608         }
1609
1610         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1611         BUG_ON(ret);
1612
1613         spin_lock(&em_tree->lock);
1614         remove_extent_mapping(em_tree, em);
1615         spin_unlock(&em_tree->lock);
1616
1617         kfree(map);
1618         em->bdev = NULL;
1619
1620         /* once for the tree */
1621         free_extent_map(em);
1622         /* once for us */
1623         free_extent_map(em);
1624
1625         unlock_chunks(root);
1626         btrfs_end_transaction(trans, root);
1627         return 0;
1628 }
1629
1630 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1631 {
1632         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1633         struct btrfs_path *path;
1634         struct extent_buffer *leaf;
1635         struct btrfs_chunk *chunk;
1636         struct btrfs_key key;
1637         struct btrfs_key found_key;
1638         u64 chunk_tree = chunk_root->root_key.objectid;
1639         u64 chunk_type;
1640         int ret;
1641
1642         path = btrfs_alloc_path();
1643         if (!path)
1644                 return -ENOMEM;
1645
1646         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1647         key.offset = (u64)-1;
1648         key.type = BTRFS_CHUNK_ITEM_KEY;
1649
1650         while (1) {
1651                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1652                 if (ret < 0)
1653                         goto error;
1654                 BUG_ON(ret == 0);
1655
1656                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1657                                           key.type);
1658                 if (ret < 0)
1659                         goto error;
1660                 if (ret > 0)
1661                         break;
1662
1663                 leaf = path->nodes[0];
1664                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1665
1666                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1667                                        struct btrfs_chunk);
1668                 chunk_type = btrfs_chunk_type(leaf, chunk);
1669                 btrfs_release_path(chunk_root, path);
1670
1671                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1672                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1673                                                    found_key.objectid,
1674                                                    found_key.offset);
1675                         BUG_ON(ret);
1676                 }
1677
1678                 if (found_key.offset == 0)
1679                         break;
1680                 key.offset = found_key.offset - 1;
1681         }
1682         ret = 0;
1683 error:
1684         btrfs_free_path(path);
1685         return ret;
1686 }
1687
1688 static u64 div_factor(u64 num, int factor)
1689 {
1690         if (factor == 10)
1691                 return num;
1692         num *= factor;
1693         do_div(num, 10);
1694         return num;
1695 }
1696
1697 int btrfs_balance(struct btrfs_root *dev_root)
1698 {
1699         int ret;
1700         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1701         struct btrfs_device *device;
1702         u64 old_size;
1703         u64 size_to_free;
1704         struct btrfs_path *path;
1705         struct btrfs_key key;
1706         struct btrfs_chunk *chunk;
1707         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1708         struct btrfs_trans_handle *trans;
1709         struct btrfs_key found_key;
1710
1711         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1712                 return -EROFS;
1713
1714         mutex_lock(&dev_root->fs_info->volume_mutex);
1715         dev_root = dev_root->fs_info->dev_root;
1716
1717         /* step one make some room on all the devices */
1718         list_for_each_entry(device, devices, dev_list) {
1719                 old_size = device->total_bytes;
1720                 size_to_free = div_factor(old_size, 1);
1721                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1722                 if (!device->writeable ||
1723                     device->total_bytes - device->bytes_used > size_to_free)
1724                         continue;
1725
1726                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1727                 BUG_ON(ret);
1728
1729                 trans = btrfs_start_transaction(dev_root, 1);
1730                 BUG_ON(!trans);
1731
1732                 ret = btrfs_grow_device(trans, device, old_size);
1733                 BUG_ON(ret);
1734
1735                 btrfs_end_transaction(trans, dev_root);
1736         }
1737
1738         /* step two, relocate all the chunks */
1739         path = btrfs_alloc_path();
1740         BUG_ON(!path);
1741
1742         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1743         key.offset = (u64)-1;
1744         key.type = BTRFS_CHUNK_ITEM_KEY;
1745
1746         while (1) {
1747                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1748                 if (ret < 0)
1749                         goto error;
1750
1751                 /*
1752                  * this shouldn't happen, it means the last relocate
1753                  * failed
1754                  */
1755                 if (ret == 0)
1756                         break;
1757
1758                 ret = btrfs_previous_item(chunk_root, path, 0,
1759                                           BTRFS_CHUNK_ITEM_KEY);
1760                 if (ret)
1761                         break;
1762
1763                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1764                                       path->slots[0]);
1765                 if (found_key.objectid != key.objectid)
1766                         break;
1767
1768                 chunk = btrfs_item_ptr(path->nodes[0],
1769                                        path->slots[0],
1770                                        struct btrfs_chunk);
1771                 key.offset = found_key.offset;
1772                 /* chunk zero is special */
1773                 if (key.offset == 0)
1774                         break;
1775
1776                 btrfs_release_path(chunk_root, path);
1777                 ret = btrfs_relocate_chunk(chunk_root,
1778                                            chunk_root->root_key.objectid,
1779                                            found_key.objectid,
1780                                            found_key.offset);
1781                 BUG_ON(ret);
1782         }
1783         ret = 0;
1784 error:
1785         btrfs_free_path(path);
1786         mutex_unlock(&dev_root->fs_info->volume_mutex);
1787         return ret;
1788 }
1789
1790 /*
1791  * shrinking a device means finding all of the device extents past
1792  * the new size, and then following the back refs to the chunks.
1793  * The chunk relocation code actually frees the device extent
1794  */
1795 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1796 {
1797         struct btrfs_trans_handle *trans;
1798         struct btrfs_root *root = device->dev_root;
1799         struct btrfs_dev_extent *dev_extent = NULL;
1800         struct btrfs_path *path;
1801         u64 length;
1802         u64 chunk_tree;
1803         u64 chunk_objectid;
1804         u64 chunk_offset;
1805         int ret;
1806         int slot;
1807         struct extent_buffer *l;
1808         struct btrfs_key key;
1809         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1810         u64 old_total = btrfs_super_total_bytes(super_copy);
1811         u64 diff = device->total_bytes - new_size;
1812
1813         if (new_size >= device->total_bytes)
1814                 return -EINVAL;
1815
1816         path = btrfs_alloc_path();
1817         if (!path)
1818                 return -ENOMEM;
1819
1820         trans = btrfs_start_transaction(root, 1);
1821         if (!trans) {
1822                 ret = -ENOMEM;
1823                 goto done;
1824         }
1825
1826         path->reada = 2;
1827
1828         lock_chunks(root);
1829
1830         device->total_bytes = new_size;
1831         if (device->writeable)
1832                 device->fs_devices->total_rw_bytes -= diff;
1833         ret = btrfs_update_device(trans, device);
1834         if (ret) {
1835                 unlock_chunks(root);
1836                 btrfs_end_transaction(trans, root);
1837                 goto done;
1838         }
1839         WARN_ON(diff > old_total);
1840         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1841         unlock_chunks(root);
1842         btrfs_end_transaction(trans, root);
1843
1844         key.objectid = device->devid;
1845         key.offset = (u64)-1;
1846         key.type = BTRFS_DEV_EXTENT_KEY;
1847
1848         while (1) {
1849                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1850                 if (ret < 0)
1851                         goto done;
1852
1853                 ret = btrfs_previous_item(root, path, 0, key.type);
1854                 if (ret < 0)
1855                         goto done;
1856                 if (ret) {
1857                         ret = 0;
1858                         goto done;
1859                 }
1860
1861                 l = path->nodes[0];
1862                 slot = path->slots[0];
1863                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1864
1865                 if (key.objectid != device->devid)
1866                         goto done;
1867
1868                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1869                 length = btrfs_dev_extent_length(l, dev_extent);
1870
1871                 if (key.offset + length <= new_size)
1872                         goto done;
1873
1874                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1875                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1876                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1877                 btrfs_release_path(root, path);
1878
1879                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1880                                            chunk_offset);
1881                 if (ret)
1882                         goto done;
1883         }
1884
1885 done:
1886         btrfs_free_path(path);
1887         return ret;
1888 }
1889
1890 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1891                            struct btrfs_root *root,
1892                            struct btrfs_key *key,
1893                            struct btrfs_chunk *chunk, int item_size)
1894 {
1895         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1896         struct btrfs_disk_key disk_key;
1897         u32 array_size;
1898         u8 *ptr;
1899
1900         array_size = btrfs_super_sys_array_size(super_copy);
1901         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1902                 return -EFBIG;
1903
1904         ptr = super_copy->sys_chunk_array + array_size;
1905         btrfs_cpu_key_to_disk(&disk_key, key);
1906         memcpy(ptr, &disk_key, sizeof(disk_key));
1907         ptr += sizeof(disk_key);
1908         memcpy(ptr, chunk, item_size);
1909         item_size += sizeof(disk_key);
1910         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1911         return 0;
1912 }
1913
1914 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
1915                                         int num_stripes, int sub_stripes)
1916 {
1917         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1918                 return calc_size;
1919         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1920                 return calc_size * (num_stripes / sub_stripes);
1921         else
1922                 return calc_size * num_stripes;
1923 }
1924
1925 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1926                                struct btrfs_root *extent_root,
1927                                struct map_lookup **map_ret,
1928                                u64 *num_bytes, u64 *stripe_size,
1929                                u64 start, u64 type)
1930 {
1931         struct btrfs_fs_info *info = extent_root->fs_info;
1932         struct btrfs_device *device = NULL;
1933         struct btrfs_fs_devices *fs_devices = info->fs_devices;
1934         struct list_head *cur;
1935         struct map_lookup *map = NULL;
1936         struct extent_map_tree *em_tree;
1937         struct extent_map *em;
1938         struct list_head private_devs;
1939         int min_stripe_size = 1 * 1024 * 1024;
1940         u64 calc_size = 1024 * 1024 * 1024;
1941         u64 max_chunk_size = calc_size;
1942         u64 min_free;
1943         u64 avail;
1944         u64 max_avail = 0;
1945         u64 dev_offset;
1946         int num_stripes = 1;
1947         int min_stripes = 1;
1948         int sub_stripes = 0;
1949         int looped = 0;
1950         int ret;
1951         int index;
1952         int stripe_len = 64 * 1024;
1953
1954         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1955             (type & BTRFS_BLOCK_GROUP_DUP)) {
1956                 WARN_ON(1);
1957                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1958         }
1959         if (list_empty(&fs_devices->alloc_list))
1960                 return -ENOSPC;
1961
1962         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1963                 num_stripes = fs_devices->rw_devices;
1964                 min_stripes = 2;
1965         }
1966         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1967                 num_stripes = 2;
1968                 min_stripes = 2;
1969         }
1970         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1971                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1972                 if (num_stripes < 2)
1973                         return -ENOSPC;
1974                 min_stripes = 2;
1975         }
1976         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1977                 num_stripes = fs_devices->rw_devices;
1978                 if (num_stripes < 4)
1979                         return -ENOSPC;
1980                 num_stripes &= ~(u32)1;
1981                 sub_stripes = 2;
1982                 min_stripes = 4;
1983         }
1984
1985         if (type & BTRFS_BLOCK_GROUP_DATA) {
1986                 max_chunk_size = 10 * calc_size;
1987                 min_stripe_size = 64 * 1024 * 1024;
1988         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1989                 max_chunk_size = 4 * calc_size;
1990                 min_stripe_size = 32 * 1024 * 1024;
1991         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1992                 calc_size = 8 * 1024 * 1024;
1993                 max_chunk_size = calc_size * 2;
1994                 min_stripe_size = 1 * 1024 * 1024;
1995         }
1996
1997         /* we don't want a chunk larger than 10% of writeable space */
1998         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
1999                              max_chunk_size);
2000
2001 again:
2002         if (!map || map->num_stripes != num_stripes) {
2003                 kfree(map);
2004                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2005                 if (!map)
2006                         return -ENOMEM;
2007                 map->num_stripes = num_stripes;
2008         }
2009
2010         if (calc_size * num_stripes > max_chunk_size) {
2011                 calc_size = max_chunk_size;
2012                 do_div(calc_size, num_stripes);
2013                 do_div(calc_size, stripe_len);
2014                 calc_size *= stripe_len;
2015         }
2016         /* we don't want tiny stripes */
2017         calc_size = max_t(u64, min_stripe_size, calc_size);
2018
2019         do_div(calc_size, stripe_len);
2020         calc_size *= stripe_len;
2021
2022         cur = fs_devices->alloc_list.next;
2023         index = 0;
2024
2025         if (type & BTRFS_BLOCK_GROUP_DUP)
2026                 min_free = calc_size * 2;
2027         else
2028                 min_free = calc_size;
2029
2030         /*
2031          * we add 1MB because we never use the first 1MB of the device, unless
2032          * we've looped, then we are likely allocating the maximum amount of
2033          * space left already
2034          */
2035         if (!looped)
2036                 min_free += 1024 * 1024;
2037
2038         INIT_LIST_HEAD(&private_devs);
2039         while (index < num_stripes) {
2040                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2041                 BUG_ON(!device->writeable);
2042                 if (device->total_bytes > device->bytes_used)
2043                         avail = device->total_bytes - device->bytes_used;
2044                 else
2045                         avail = 0;
2046                 cur = cur->next;
2047
2048                 if (device->in_fs_metadata && avail >= min_free) {
2049                         ret = find_free_dev_extent(trans, device,
2050                                                    min_free, &dev_offset);
2051                         if (ret == 0) {
2052                                 list_move_tail(&device->dev_alloc_list,
2053                                                &private_devs);
2054                                 map->stripes[index].dev = device;
2055                                 map->stripes[index].physical = dev_offset;
2056                                 index++;
2057                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2058                                         map->stripes[index].dev = device;
2059                                         map->stripes[index].physical =
2060                                                 dev_offset + calc_size;
2061                                         index++;
2062                                 }
2063                         }
2064                 } else if (device->in_fs_metadata && avail > max_avail)
2065                         max_avail = avail;
2066                 if (cur == &fs_devices->alloc_list)
2067                         break;
2068         }
2069         list_splice(&private_devs, &fs_devices->alloc_list);
2070         if (index < num_stripes) {
2071                 if (index >= min_stripes) {
2072                         num_stripes = index;
2073                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2074                                 num_stripes /= sub_stripes;
2075                                 num_stripes *= sub_stripes;
2076                         }
2077                         looped = 1;
2078                         goto again;
2079                 }
2080                 if (!looped && max_avail > 0) {
2081                         looped = 1;
2082                         calc_size = max_avail;
2083                         goto again;
2084                 }
2085                 kfree(map);
2086                 return -ENOSPC;
2087         }
2088         map->sector_size = extent_root->sectorsize;
2089         map->stripe_len = stripe_len;
2090         map->io_align = stripe_len;
2091         map->io_width = stripe_len;
2092         map->type = type;
2093         map->num_stripes = num_stripes;
2094         map->sub_stripes = sub_stripes;
2095
2096         *map_ret = map;
2097         *stripe_size = calc_size;
2098         *num_bytes = chunk_bytes_by_type(type, calc_size,
2099                                          num_stripes, sub_stripes);
2100
2101         em = alloc_extent_map(GFP_NOFS);
2102         if (!em) {
2103                 kfree(map);
2104                 return -ENOMEM;
2105         }
2106         em->bdev = (struct block_device *)map;
2107         em->start = start;
2108         em->len = *num_bytes;
2109         em->block_start = 0;
2110         em->block_len = em->len;
2111
2112         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2113         spin_lock(&em_tree->lock);
2114         ret = add_extent_mapping(em_tree, em);
2115         spin_unlock(&em_tree->lock);
2116         BUG_ON(ret);
2117         free_extent_map(em);
2118
2119         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2120                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2121                                      start, *num_bytes);
2122         BUG_ON(ret);
2123
2124         index = 0;
2125         while (index < map->num_stripes) {
2126                 device = map->stripes[index].dev;
2127                 dev_offset = map->stripes[index].physical;
2128
2129                 ret = btrfs_alloc_dev_extent(trans, device,
2130                                 info->chunk_root->root_key.objectid,
2131                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2132                                 start, dev_offset, calc_size);
2133                 BUG_ON(ret);
2134                 index++;
2135         }
2136
2137         return 0;
2138 }
2139
2140 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2141                                 struct btrfs_root *extent_root,
2142                                 struct map_lookup *map, u64 chunk_offset,
2143                                 u64 chunk_size, u64 stripe_size)
2144 {
2145         u64 dev_offset;
2146         struct btrfs_key key;
2147         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2148         struct btrfs_device *device;
2149         struct btrfs_chunk *chunk;
2150         struct btrfs_stripe *stripe;
2151         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2152         int index = 0;
2153         int ret;
2154
2155         chunk = kzalloc(item_size, GFP_NOFS);
2156         if (!chunk)
2157                 return -ENOMEM;
2158
2159         index = 0;
2160         while (index < map->num_stripes) {
2161                 device = map->stripes[index].dev;
2162                 device->bytes_used += stripe_size;
2163                 ret = btrfs_update_device(trans, device);
2164                 BUG_ON(ret);
2165                 index++;
2166         }
2167
2168         index = 0;
2169         stripe = &chunk->stripe;
2170         while (index < map->num_stripes) {
2171                 device = map->stripes[index].dev;
2172                 dev_offset = map->stripes[index].physical;
2173
2174                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2175                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2176                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2177                 stripe++;
2178                 index++;
2179         }
2180
2181         btrfs_set_stack_chunk_length(chunk, chunk_size);
2182         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2183         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2184         btrfs_set_stack_chunk_type(chunk, map->type);
2185         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2186         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2187         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2188         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2189         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2190
2191         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2192         key.type = BTRFS_CHUNK_ITEM_KEY;
2193         key.offset = chunk_offset;
2194
2195         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2196         BUG_ON(ret);
2197
2198         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2199                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2200                                              item_size);
2201                 BUG_ON(ret);
2202         }
2203         kfree(chunk);
2204         return 0;
2205 }
2206
2207 /*
2208  * Chunk allocation falls into two parts. The first part does works
2209  * that make the new allocated chunk useable, but not do any operation
2210  * that modifies the chunk tree. The second part does the works that
2211  * require modifying the chunk tree. This division is important for the
2212  * bootstrap process of adding storage to a seed btrfs.
2213  */
2214 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2215                       struct btrfs_root *extent_root, u64 type)
2216 {
2217         u64 chunk_offset;
2218         u64 chunk_size;
2219         u64 stripe_size;
2220         struct map_lookup *map;
2221         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2222         int ret;
2223
2224         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2225                               &chunk_offset);
2226         if (ret)
2227                 return ret;
2228
2229         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2230                                   &stripe_size, chunk_offset, type);
2231         if (ret)
2232                 return ret;
2233
2234         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2235                                    chunk_size, stripe_size);
2236         BUG_ON(ret);
2237         return 0;
2238 }
2239
2240 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2241                                          struct btrfs_root *root,
2242                                          struct btrfs_device *device)
2243 {
2244         u64 chunk_offset;
2245         u64 sys_chunk_offset;
2246         u64 chunk_size;
2247         u64 sys_chunk_size;
2248         u64 stripe_size;
2249         u64 sys_stripe_size;
2250         u64 alloc_profile;
2251         struct map_lookup *map;
2252         struct map_lookup *sys_map;
2253         struct btrfs_fs_info *fs_info = root->fs_info;
2254         struct btrfs_root *extent_root = fs_info->extent_root;
2255         int ret;
2256
2257         ret = find_next_chunk(fs_info->chunk_root,
2258                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2259         BUG_ON(ret);
2260
2261         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2262                         (fs_info->metadata_alloc_profile &
2263                          fs_info->avail_metadata_alloc_bits);
2264         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2265
2266         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2267                                   &stripe_size, chunk_offset, alloc_profile);
2268         BUG_ON(ret);
2269
2270         sys_chunk_offset = chunk_offset + chunk_size;
2271
2272         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2273                         (fs_info->system_alloc_profile &
2274                          fs_info->avail_system_alloc_bits);
2275         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2276
2277         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2278                                   &sys_chunk_size, &sys_stripe_size,
2279                                   sys_chunk_offset, alloc_profile);
2280         BUG_ON(ret);
2281
2282         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2283         BUG_ON(ret);
2284
2285         /*
2286          * Modifying chunk tree needs allocating new blocks from both
2287          * system block group and metadata block group. So we only can
2288          * do operations require modifying the chunk tree after both
2289          * block groups were created.
2290          */
2291         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2292                                    chunk_size, stripe_size);
2293         BUG_ON(ret);
2294
2295         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2296                                    sys_chunk_offset, sys_chunk_size,
2297                                    sys_stripe_size);
2298         BUG_ON(ret);
2299         return 0;
2300 }
2301
2302 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2303 {
2304         struct extent_map *em;
2305         struct map_lookup *map;
2306         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2307         int readonly = 0;
2308         int i;
2309
2310         spin_lock(&map_tree->map_tree.lock);
2311         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2312         spin_unlock(&map_tree->map_tree.lock);
2313         if (!em)
2314                 return 1;
2315
2316         map = (struct map_lookup *)em->bdev;
2317         for (i = 0; i < map->num_stripes; i++) {
2318                 if (!map->stripes[i].dev->writeable) {
2319                         readonly = 1;
2320                         break;
2321                 }
2322         }
2323         free_extent_map(em);
2324         return readonly;
2325 }
2326
2327 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2328 {
2329         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2330 }
2331
2332 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2333 {
2334         struct extent_map *em;
2335
2336         while (1) {
2337                 spin_lock(&tree->map_tree.lock);
2338                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2339                 if (em)
2340                         remove_extent_mapping(&tree->map_tree, em);
2341                 spin_unlock(&tree->map_tree.lock);
2342                 if (!em)
2343                         break;
2344                 kfree(em->bdev);
2345                 /* once for us */
2346                 free_extent_map(em);
2347                 /* once for the tree */
2348                 free_extent_map(em);
2349         }
2350 }
2351
2352 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2353 {
2354         struct extent_map *em;
2355         struct map_lookup *map;
2356         struct extent_map_tree *em_tree = &map_tree->map_tree;
2357         int ret;
2358
2359         spin_lock(&em_tree->lock);
2360         em = lookup_extent_mapping(em_tree, logical, len);
2361         spin_unlock(&em_tree->lock);
2362         BUG_ON(!em);
2363
2364         BUG_ON(em->start > logical || em->start + em->len < logical);
2365         map = (struct map_lookup *)em->bdev;
2366         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2367                 ret = map->num_stripes;
2368         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2369                 ret = map->sub_stripes;
2370         else
2371                 ret = 1;
2372         free_extent_map(em);
2373         return ret;
2374 }
2375
2376 static int find_live_mirror(struct map_lookup *map, int first, int num,
2377                             int optimal)
2378 {
2379         int i;
2380         if (map->stripes[optimal].dev->bdev)
2381                 return optimal;
2382         for (i = first; i < first + num; i++) {
2383                 if (map->stripes[i].dev->bdev)
2384                         return i;
2385         }
2386         /* we couldn't find one that doesn't fail.  Just return something
2387          * and the io error handling code will clean up eventually
2388          */
2389         return optimal;
2390 }
2391
2392 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2393                              u64 logical, u64 *length,
2394                              struct btrfs_multi_bio **multi_ret,
2395                              int mirror_num, struct page *unplug_page)
2396 {
2397         struct extent_map *em;
2398         struct map_lookup *map;
2399         struct extent_map_tree *em_tree = &map_tree->map_tree;
2400         u64 offset;
2401         u64 stripe_offset;
2402         u64 stripe_nr;
2403         int stripes_allocated = 8;
2404         int stripes_required = 1;
2405         int stripe_index;
2406         int i;
2407         int num_stripes;
2408         int max_errors = 0;
2409         struct btrfs_multi_bio *multi = NULL;
2410
2411         if (multi_ret && !(rw & (1 << BIO_RW)))
2412                 stripes_allocated = 1;
2413 again:
2414         if (multi_ret) {
2415                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2416                                 GFP_NOFS);
2417                 if (!multi)
2418                         return -ENOMEM;
2419
2420                 atomic_set(&multi->error, 0);
2421         }
2422
2423         spin_lock(&em_tree->lock);
2424         em = lookup_extent_mapping(em_tree, logical, *length);
2425         spin_unlock(&em_tree->lock);
2426
2427         if (!em && unplug_page)
2428                 return 0;
2429
2430         if (!em) {
2431                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2432                        (unsigned long long)logical,
2433                        (unsigned long long)*length);
2434                 BUG();
2435         }
2436
2437         BUG_ON(em->start > logical || em->start + em->len < logical);
2438         map = (struct map_lookup *)em->bdev;
2439         offset = logical - em->start;
2440
2441         if (mirror_num > map->num_stripes)
2442                 mirror_num = 0;
2443
2444         /* if our multi bio struct is too small, back off and try again */
2445         if (rw & (1 << BIO_RW)) {
2446                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2447                                  BTRFS_BLOCK_GROUP_DUP)) {
2448                         stripes_required = map->num_stripes;
2449                         max_errors = 1;
2450                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2451                         stripes_required = map->sub_stripes;
2452                         max_errors = 1;
2453                 }
2454         }
2455         if (multi_ret && rw == WRITE &&
2456             stripes_allocated < stripes_required) {
2457                 stripes_allocated = map->num_stripes;
2458                 free_extent_map(em);
2459                 kfree(multi);
2460                 goto again;
2461         }
2462         stripe_nr = offset;
2463         /*
2464          * stripe_nr counts the total number of stripes we have to stride
2465          * to get to this block
2466          */
2467         do_div(stripe_nr, map->stripe_len);
2468
2469         stripe_offset = stripe_nr * map->stripe_len;
2470         BUG_ON(offset < stripe_offset);
2471
2472         /* stripe_offset is the offset of this block in its stripe*/
2473         stripe_offset = offset - stripe_offset;
2474
2475         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2476                          BTRFS_BLOCK_GROUP_RAID10 |
2477                          BTRFS_BLOCK_GROUP_DUP)) {
2478                 /* we limit the length of each bio to what fits in a stripe */
2479                 *length = min_t(u64, em->len - offset,
2480                               map->stripe_len - stripe_offset);
2481         } else {
2482                 *length = em->len - offset;
2483         }
2484
2485         if (!multi_ret && !unplug_page)
2486                 goto out;
2487
2488         num_stripes = 1;
2489         stripe_index = 0;
2490         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2491                 if (unplug_page || (rw & (1 << BIO_RW)))
2492                         num_stripes = map->num_stripes;
2493                 else if (mirror_num)
2494                         stripe_index = mirror_num - 1;
2495                 else {
2496                         stripe_index = find_live_mirror(map, 0,
2497                                             map->num_stripes,
2498                                             current->pid % map->num_stripes);
2499                 }
2500
2501         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2502                 if (rw & (1 << BIO_RW))
2503                         num_stripes = map->num_stripes;
2504                 else if (mirror_num)
2505                         stripe_index = mirror_num - 1;
2506
2507         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2508                 int factor = map->num_stripes / map->sub_stripes;
2509
2510                 stripe_index = do_div(stripe_nr, factor);
2511                 stripe_index *= map->sub_stripes;
2512
2513                 if (unplug_page || (rw & (1 << BIO_RW)))
2514                         num_stripes = map->sub_stripes;
2515                 else if (mirror_num)
2516                         stripe_index += mirror_num - 1;
2517                 else {
2518                         stripe_index = find_live_mirror(map, stripe_index,
2519                                               map->sub_stripes, stripe_index +
2520                                               current->pid % map->sub_stripes);
2521                 }
2522         } else {
2523                 /*
2524                  * after this do_div call, stripe_nr is the number of stripes
2525                  * on this device we have to walk to find the data, and
2526                  * stripe_index is the number of our device in the stripe array
2527                  */
2528                 stripe_index = do_div(stripe_nr, map->num_stripes);
2529         }
2530         BUG_ON(stripe_index >= map->num_stripes);
2531
2532         for (i = 0; i < num_stripes; i++) {
2533                 if (unplug_page) {
2534                         struct btrfs_device *device;
2535                         struct backing_dev_info *bdi;
2536
2537                         device = map->stripes[stripe_index].dev;
2538                         if (device->bdev) {
2539                                 bdi = blk_get_backing_dev_info(device->bdev);
2540                                 if (bdi->unplug_io_fn)
2541                                         bdi->unplug_io_fn(bdi, unplug_page);
2542                         }
2543                 } else {
2544                         multi->stripes[i].physical =
2545                                 map->stripes[stripe_index].physical +
2546                                 stripe_offset + stripe_nr * map->stripe_len;
2547                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2548                 }
2549                 stripe_index++;
2550         }
2551         if (multi_ret) {
2552                 *multi_ret = multi;
2553                 multi->num_stripes = num_stripes;
2554                 multi->max_errors = max_errors;
2555         }
2556 out:
2557         free_extent_map(em);
2558         return 0;
2559 }
2560
2561 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2562                       u64 logical, u64 *length,
2563                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2564 {
2565         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2566                                  mirror_num, NULL);
2567 }
2568
2569 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2570                      u64 chunk_start, u64 physical, u64 devid,
2571                      u64 **logical, int *naddrs, int *stripe_len)
2572 {
2573         struct extent_map_tree *em_tree = &map_tree->map_tree;
2574         struct extent_map *em;
2575         struct map_lookup *map;
2576         u64 *buf;
2577         u64 bytenr;
2578         u64 length;
2579         u64 stripe_nr;
2580         int i, j, nr = 0;
2581
2582         spin_lock(&em_tree->lock);
2583         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2584         spin_unlock(&em_tree->lock);
2585
2586         BUG_ON(!em || em->start != chunk_start);
2587         map = (struct map_lookup *)em->bdev;
2588
2589         length = em->len;
2590         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2591                 do_div(length, map->num_stripes / map->sub_stripes);
2592         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2593                 do_div(length, map->num_stripes);
2594
2595         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2596         BUG_ON(!buf);
2597
2598         for (i = 0; i < map->num_stripes; i++) {
2599                 if (devid && map->stripes[i].dev->devid != devid)
2600                         continue;
2601                 if (map->stripes[i].physical > physical ||
2602                     map->stripes[i].physical + length <= physical)
2603                         continue;
2604
2605                 stripe_nr = physical - map->stripes[i].physical;
2606                 do_div(stripe_nr, map->stripe_len);
2607
2608                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2609                         stripe_nr = stripe_nr * map->num_stripes + i;
2610                         do_div(stripe_nr, map->sub_stripes);
2611                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2612                         stripe_nr = stripe_nr * map->num_stripes + i;
2613                 }
2614                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2615                 WARN_ON(nr >= map->num_stripes);
2616                 for (j = 0; j < nr; j++) {
2617                         if (buf[j] == bytenr)
2618                                 break;
2619                 }
2620                 if (j == nr) {
2621                         WARN_ON(nr >= map->num_stripes);
2622                         buf[nr++] = bytenr;
2623                 }
2624         }
2625
2626         for (i = 0; i > nr; i++) {
2627                 struct btrfs_multi_bio *multi;
2628                 struct btrfs_bio_stripe *stripe;
2629                 int ret;
2630
2631                 length = 1;
2632                 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2633                                       &length, &multi, 0);
2634                 BUG_ON(ret);
2635
2636                 stripe = multi->stripes;
2637                 for (j = 0; j < multi->num_stripes; j++) {
2638                         if (stripe->physical >= physical &&
2639                             physical < stripe->physical + length)
2640                                 break;
2641                 }
2642                 BUG_ON(j >= multi->num_stripes);
2643                 kfree(multi);
2644         }
2645
2646         *logical = buf;
2647         *naddrs = nr;
2648         *stripe_len = map->stripe_len;
2649
2650         free_extent_map(em);
2651         return 0;
2652 }
2653
2654 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2655                       u64 logical, struct page *page)
2656 {
2657         u64 length = PAGE_CACHE_SIZE;
2658         return __btrfs_map_block(map_tree, READ, logical, &length,
2659                                  NULL, 0, page);
2660 }
2661
2662 static void end_bio_multi_stripe(struct bio *bio, int err)
2663 {
2664         struct btrfs_multi_bio *multi = bio->bi_private;
2665         int is_orig_bio = 0;
2666
2667         if (err)
2668                 atomic_inc(&multi->error);
2669
2670         if (bio == multi->orig_bio)
2671                 is_orig_bio = 1;
2672
2673         if (atomic_dec_and_test(&multi->stripes_pending)) {
2674                 if (!is_orig_bio) {
2675                         bio_put(bio);
2676                         bio = multi->orig_bio;
2677                 }
2678                 bio->bi_private = multi->private;
2679                 bio->bi_end_io = multi->end_io;
2680                 /* only send an error to the higher layers if it is
2681                  * beyond the tolerance of the multi-bio
2682                  */
2683                 if (atomic_read(&multi->error) > multi->max_errors) {
2684                         err = -EIO;
2685                 } else if (err) {
2686                         /*
2687                          * this bio is actually up to date, we didn't
2688                          * go over the max number of errors
2689                          */
2690                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2691                         err = 0;
2692                 }
2693                 kfree(multi);
2694
2695                 bio_endio(bio, err);
2696         } else if (!is_orig_bio) {
2697                 bio_put(bio);
2698         }
2699 }
2700
2701 struct async_sched {
2702         struct bio *bio;
2703         int rw;
2704         struct btrfs_fs_info *info;
2705         struct btrfs_work work;
2706 };
2707
2708 /*
2709  * see run_scheduled_bios for a description of why bios are collected for
2710  * async submit.
2711  *
2712  * This will add one bio to the pending list for a device and make sure
2713  * the work struct is scheduled.
2714  */
2715 static noinline int schedule_bio(struct btrfs_root *root,
2716                                  struct btrfs_device *device,
2717                                  int rw, struct bio *bio)
2718 {
2719         int should_queue = 1;
2720
2721         /* don't bother with additional async steps for reads, right now */
2722         if (!(rw & (1 << BIO_RW))) {
2723                 bio_get(bio);
2724                 submit_bio(rw, bio);
2725                 bio_put(bio);
2726                 return 0;
2727         }
2728
2729         /*
2730          * nr_async_bios allows us to reliably return congestion to the
2731          * higher layers.  Otherwise, the async bio makes it appear we have
2732          * made progress against dirty pages when we've really just put it
2733          * on a queue for later
2734          */
2735         atomic_inc(&root->fs_info->nr_async_bios);
2736         WARN_ON(bio->bi_next);
2737         bio->bi_next = NULL;
2738         bio->bi_rw |= rw;
2739
2740         spin_lock(&device->io_lock);
2741
2742         if (device->pending_bio_tail)
2743                 device->pending_bio_tail->bi_next = bio;
2744
2745         device->pending_bio_tail = bio;
2746         if (!device->pending_bios)
2747                 device->pending_bios = bio;
2748         if (device->running_pending)
2749                 should_queue = 0;
2750
2751         spin_unlock(&device->io_lock);
2752
2753         if (should_queue)
2754                 btrfs_queue_worker(&root->fs_info->submit_workers,
2755                                    &device->work);
2756         return 0;
2757 }
2758
2759 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2760                   int mirror_num, int async_submit)
2761 {
2762         struct btrfs_mapping_tree *map_tree;
2763         struct btrfs_device *dev;
2764         struct bio *first_bio = bio;
2765         u64 logical = (u64)bio->bi_sector << 9;
2766         u64 length = 0;
2767         u64 map_length;
2768         struct btrfs_multi_bio *multi = NULL;
2769         int ret;
2770         int dev_nr = 0;
2771         int total_devs = 1;
2772
2773         length = bio->bi_size;
2774         map_tree = &root->fs_info->mapping_tree;
2775         map_length = length;
2776
2777         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2778                               mirror_num);
2779         BUG_ON(ret);
2780
2781         total_devs = multi->num_stripes;
2782         if (map_length < length) {
2783                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2784                        "len %llu\n", (unsigned long long)logical,
2785                        (unsigned long long)length,
2786                        (unsigned long long)map_length);
2787                 BUG();
2788         }
2789         multi->end_io = first_bio->bi_end_io;
2790         multi->private = first_bio->bi_private;
2791         multi->orig_bio = first_bio;
2792         atomic_set(&multi->stripes_pending, multi->num_stripes);
2793
2794         while (dev_nr < total_devs) {
2795                 if (total_devs > 1) {
2796                         if (dev_nr < total_devs - 1) {
2797                                 bio = bio_clone(first_bio, GFP_NOFS);
2798                                 BUG_ON(!bio);
2799                         } else {
2800                                 bio = first_bio;
2801                         }
2802                         bio->bi_private = multi;
2803                         bio->bi_end_io = end_bio_multi_stripe;
2804                 }
2805                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2806                 dev = multi->stripes[dev_nr].dev;
2807                 BUG_ON(rw == WRITE && !dev->writeable);
2808                 if (dev && dev->bdev) {
2809                         bio->bi_bdev = dev->bdev;
2810                         if (async_submit)
2811                                 schedule_bio(root, dev, rw, bio);
2812                         else
2813                                 submit_bio(rw, bio);
2814                 } else {
2815                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2816                         bio->bi_sector = logical >> 9;
2817                         bio_endio(bio, -EIO);
2818                 }
2819                 dev_nr++;
2820         }
2821         if (total_devs == 1)
2822                 kfree(multi);
2823         return 0;
2824 }
2825
2826 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2827                                        u8 *uuid, u8 *fsid)
2828 {
2829         struct btrfs_device *device;
2830         struct btrfs_fs_devices *cur_devices;
2831
2832         cur_devices = root->fs_info->fs_devices;
2833         while (cur_devices) {
2834                 if (!fsid ||
2835                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2836                         device = __find_device(&cur_devices->devices,
2837                                                devid, uuid);
2838                         if (device)
2839                                 return device;
2840                 }
2841                 cur_devices = cur_devices->seed;
2842         }
2843         return NULL;
2844 }
2845
2846 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2847                                             u64 devid, u8 *dev_uuid)
2848 {
2849         struct btrfs_device *device;
2850         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2851
2852         device = kzalloc(sizeof(*device), GFP_NOFS);
2853         if (!device)
2854                 return NULL;
2855         list_add(&device->dev_list,
2856                  &fs_devices->devices);
2857         device->barriers = 1;
2858         device->dev_root = root->fs_info->dev_root;
2859         device->devid = devid;
2860         device->work.func = pending_bios_fn;
2861         device->fs_devices = fs_devices;
2862         fs_devices->num_devices++;
2863         spin_lock_init(&device->io_lock);
2864         INIT_LIST_HEAD(&device->dev_alloc_list);
2865         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2866         return device;
2867 }
2868
2869 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2870                           struct extent_buffer *leaf,
2871                           struct btrfs_chunk *chunk)
2872 {
2873         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2874         struct map_lookup *map;
2875         struct extent_map *em;
2876         u64 logical;
2877         u64 length;
2878         u64 devid;
2879         u8 uuid[BTRFS_UUID_SIZE];
2880         int num_stripes;
2881         int ret;
2882         int i;
2883
2884         logical = key->offset;
2885         length = btrfs_chunk_length(leaf, chunk);
2886
2887         spin_lock(&map_tree->map_tree.lock);
2888         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2889         spin_unlock(&map_tree->map_tree.lock);
2890
2891         /* already mapped? */
2892         if (em && em->start <= logical && em->start + em->len > logical) {
2893                 free_extent_map(em);
2894                 return 0;
2895         } else if (em) {
2896                 free_extent_map(em);
2897         }
2898
2899         em = alloc_extent_map(GFP_NOFS);
2900         if (!em)
2901                 return -ENOMEM;
2902         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2903         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2904         if (!map) {
2905                 free_extent_map(em);
2906                 return -ENOMEM;
2907         }
2908
2909         em->bdev = (struct block_device *)map;
2910         em->start = logical;
2911         em->len = length;
2912         em->block_start = 0;
2913         em->block_len = em->len;
2914
2915         map->num_stripes = num_stripes;
2916         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2917         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2918         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2919         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2920         map->type = btrfs_chunk_type(leaf, chunk);
2921         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2922         for (i = 0; i < num_stripes; i++) {
2923                 map->stripes[i].physical =
2924                         btrfs_stripe_offset_nr(leaf, chunk, i);
2925                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2926                 read_extent_buffer(leaf, uuid, (unsigned long)
2927                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2928                                    BTRFS_UUID_SIZE);
2929                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2930                                                         NULL);
2931                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2932                         kfree(map);
2933                         free_extent_map(em);
2934                         return -EIO;
2935                 }
2936                 if (!map->stripes[i].dev) {
2937                         map->stripes[i].dev =
2938                                 add_missing_dev(root, devid, uuid);
2939                         if (!map->stripes[i].dev) {
2940                                 kfree(map);
2941                                 free_extent_map(em);
2942                                 return -EIO;
2943                         }
2944                 }
2945                 map->stripes[i].dev->in_fs_metadata = 1;
2946         }
2947
2948         spin_lock(&map_tree->map_tree.lock);
2949         ret = add_extent_mapping(&map_tree->map_tree, em);
2950         spin_unlock(&map_tree->map_tree.lock);
2951         BUG_ON(ret);
2952         free_extent_map(em);
2953
2954         return 0;
2955 }
2956
2957 static int fill_device_from_item(struct extent_buffer *leaf,
2958                                  struct btrfs_dev_item *dev_item,
2959                                  struct btrfs_device *device)
2960 {
2961         unsigned long ptr;
2962
2963         device->devid = btrfs_device_id(leaf, dev_item);
2964         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2965         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2966         device->type = btrfs_device_type(leaf, dev_item);
2967         device->io_align = btrfs_device_io_align(leaf, dev_item);
2968         device->io_width = btrfs_device_io_width(leaf, dev_item);
2969         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2970
2971         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2972         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2973
2974         return 0;
2975 }
2976
2977 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2978 {
2979         struct btrfs_fs_devices *fs_devices;
2980         int ret;
2981
2982         mutex_lock(&uuid_mutex);
2983
2984         fs_devices = root->fs_info->fs_devices->seed;
2985         while (fs_devices) {
2986                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2987                         ret = 0;
2988                         goto out;
2989                 }
2990                 fs_devices = fs_devices->seed;
2991         }
2992
2993         fs_devices = find_fsid(fsid);
2994         if (!fs_devices) {
2995                 ret = -ENOENT;
2996                 goto out;
2997         }
2998
2999         fs_devices = clone_fs_devices(fs_devices);
3000         if (IS_ERR(fs_devices)) {
3001                 ret = PTR_ERR(fs_devices);
3002                 goto out;
3003         }
3004
3005         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3006                                    root->fs_info->bdev_holder);
3007         if (ret)
3008                 goto out;
3009
3010         if (!fs_devices->seeding) {
3011                 __btrfs_close_devices(fs_devices);
3012                 free_fs_devices(fs_devices);
3013                 ret = -EINVAL;
3014                 goto out;
3015         }
3016
3017         fs_devices->seed = root->fs_info->fs_devices->seed;
3018         root->fs_info->fs_devices->seed = fs_devices;
3019 out:
3020         mutex_unlock(&uuid_mutex);
3021         return ret;
3022 }
3023
3024 static int read_one_dev(struct btrfs_root *root,
3025                         struct extent_buffer *leaf,
3026                         struct btrfs_dev_item *dev_item)
3027 {
3028         struct btrfs_device *device;
3029         u64 devid;
3030         int ret;
3031         u8 fs_uuid[BTRFS_UUID_SIZE];
3032         u8 dev_uuid[BTRFS_UUID_SIZE];
3033
3034         devid = btrfs_device_id(leaf, dev_item);
3035         read_extent_buffer(leaf, dev_uuid,
3036                            (unsigned long)btrfs_device_uuid(dev_item),
3037                            BTRFS_UUID_SIZE);
3038         read_extent_buffer(leaf, fs_uuid,
3039                            (unsigned long)btrfs_device_fsid(dev_item),
3040                            BTRFS_UUID_SIZE);
3041
3042         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3043                 ret = open_seed_devices(root, fs_uuid);
3044                 if (ret && !btrfs_test_opt(root, DEGRADED))
3045                         return ret;
3046         }
3047
3048         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3049         if (!device || !device->bdev) {
3050                 if (!btrfs_test_opt(root, DEGRADED))
3051                         return -EIO;
3052
3053                 if (!device) {
3054                         printk(KERN_WARNING "warning devid %llu missing\n",
3055                                (unsigned long long)devid);
3056                         device = add_missing_dev(root, devid, dev_uuid);
3057                         if (!device)
3058                                 return -ENOMEM;
3059                 }
3060         }
3061
3062         if (device->fs_devices != root->fs_info->fs_devices) {
3063                 BUG_ON(device->writeable);
3064                 if (device->generation !=
3065                     btrfs_device_generation(leaf, dev_item))
3066                         return -EINVAL;
3067         }
3068
3069         fill_device_from_item(leaf, dev_item, device);
3070         device->dev_root = root->fs_info->dev_root;
3071         device->in_fs_metadata = 1;
3072         if (device->writeable)
3073                 device->fs_devices->total_rw_bytes += device->total_bytes;
3074         ret = 0;
3075         return ret;
3076 }
3077
3078 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3079 {
3080         struct btrfs_dev_item *dev_item;
3081
3082         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3083                                                      dev_item);
3084         return read_one_dev(root, buf, dev_item);
3085 }
3086
3087 int btrfs_read_sys_array(struct btrfs_root *root)
3088 {
3089         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3090         struct extent_buffer *sb;
3091         struct btrfs_disk_key *disk_key;
3092         struct btrfs_chunk *chunk;
3093         u8 *ptr;
3094         unsigned long sb_ptr;
3095         int ret = 0;
3096         u32 num_stripes;
3097         u32 array_size;
3098         u32 len = 0;
3099         u32 cur;
3100         struct btrfs_key key;
3101
3102         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3103                                           BTRFS_SUPER_INFO_SIZE);
3104         if (!sb)
3105                 return -ENOMEM;
3106         btrfs_set_buffer_uptodate(sb);
3107         btrfs_set_buffer_lockdep_class(sb, 0);
3108
3109         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3110         array_size = btrfs_super_sys_array_size(super_copy);
3111
3112         ptr = super_copy->sys_chunk_array;
3113         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3114         cur = 0;
3115
3116         while (cur < array_size) {
3117                 disk_key = (struct btrfs_disk_key *)ptr;
3118                 btrfs_disk_key_to_cpu(&key, disk_key);
3119
3120                 len = sizeof(*disk_key); ptr += len;
3121                 sb_ptr += len;
3122                 cur += len;
3123
3124                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3125                         chunk = (struct btrfs_chunk *)sb_ptr;
3126                         ret = read_one_chunk(root, &key, sb, chunk);
3127                         if (ret)
3128                                 break;
3129                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3130                         len = btrfs_chunk_item_size(num_stripes);
3131                 } else {
3132                         ret = -EIO;
3133                         break;
3134                 }
3135                 ptr += len;
3136                 sb_ptr += len;
3137                 cur += len;
3138         }
3139         free_extent_buffer(sb);
3140         return ret;
3141 }
3142
3143 int btrfs_read_chunk_tree(struct btrfs_root *root)
3144 {
3145         struct btrfs_path *path;
3146         struct extent_buffer *leaf;
3147         struct btrfs_key key;
3148         struct btrfs_key found_key;
3149         int ret;
3150         int slot;
3151
3152         root = root->fs_info->chunk_root;
3153
3154         path = btrfs_alloc_path();
3155         if (!path)
3156                 return -ENOMEM;
3157
3158         /* first we search for all of the device items, and then we
3159          * read in all of the chunk items.  This way we can create chunk
3160          * mappings that reference all of the devices that are afound
3161          */
3162         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3163         key.offset = 0;
3164         key.type = 0;
3165 again:
3166         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3167         while (1) {
3168                 leaf = path->nodes[0];
3169                 slot = path->slots[0];
3170                 if (slot >= btrfs_header_nritems(leaf)) {
3171                         ret = btrfs_next_leaf(root, path);
3172                         if (ret == 0)
3173                                 continue;
3174                         if (ret < 0)
3175                                 goto error;
3176                         break;
3177                 }
3178                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3179                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3180                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3181                                 break;
3182                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3183                                 struct btrfs_dev_item *dev_item;
3184                                 dev_item = btrfs_item_ptr(leaf, slot,
3185                                                   struct btrfs_dev_item);
3186                                 ret = read_one_dev(root, leaf, dev_item);
3187                                 if (ret)
3188                                         goto error;
3189                         }
3190                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3191                         struct btrfs_chunk *chunk;
3192                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3193                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3194                         if (ret)
3195                                 goto error;
3196                 }
3197                 path->slots[0]++;
3198         }
3199         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3200                 key.objectid = 0;
3201                 btrfs_release_path(root, path);
3202                 goto again;
3203         }
3204         ret = 0;
3205 error:
3206         btrfs_free_path(path);
3207         return ret;
3208 }