fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/random.h>
  23 #include <asm/div64.h>
  24 #include "ctree.h"
  25 #include "extent_map.h"
  26 #include "disk-io.h"
  27 #include "transaction.h"
  28 #include "print-tree.h"
  29 #include "volumes.h"
  30 #include "async-thread.h"
  31
  32 struct map_lookup {
  33         u64 type;
  34         int io_align;
  35         int io_width;
  36         int stripe_len;
  37         int sector_size;
  38         int num_stripes;
  39         int sub_stripes;
  40         struct btrfs_bio_stripe stripes[];
  41 };
  42
  43 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  44                             (sizeof(struct btrfs_bio_stripe) * (n)))
  45
  46 static DEFINE_MUTEX(uuid_mutex);
  47 static LIST_HEAD(fs_uuids);
  48
  49 void btrfs_lock_volumes(void)
  50 {
  51         mutex_lock(&uuid_mutex);
  52 }
  53
  54 void btrfs_unlock_volumes(void)
  55 {
  56         mutex_unlock(&uuid_mutex);
  57 }
  58
  59 static void lock_chunks(struct btrfs_root *root)
  60 {
  61         mutex_lock(&root->fs_info->chunk_mutex);
  62 }
  63
  64 static void unlock_chunks(struct btrfs_root *root)
  65 {
  66         mutex_unlock(&root->fs_info->chunk_mutex);
  67 }
  68
  69 int btrfs_cleanup_fs_uuids(void)
  70 {
  71         struct btrfs_fs_devices *fs_devices;
  72         struct list_head *uuid_cur;
  73         struct list_head *devices_cur;
  74         struct btrfs_device *dev;
  75
  76         list_for_each(uuid_cur, &fs_uuids) {
  77                 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
  78                                         list);
  79                 while(!list_empty(&fs_devices->devices)) {
  80                         devices_cur = fs_devices->devices.next;
  81                         dev = list_entry(devices_cur, struct btrfs_device,
  82                                          dev_list);
  83                         if (dev->bdev) {
  84                                 close_bdev_excl(dev->bdev);
  85                                 fs_devices->open_devices--;
  86                         }
  87                         list_del(&dev->dev_list);
  88                         kfree(dev->name);
  89                         kfree(dev);
  90                 }
  91         }
  92         return 0;
  93 }
  94
  95 static noinline struct btrfs_device *__find_device(struct list_head *head,
  96                                                    u64 devid, u8 *uuid)
  97 {
  98         struct btrfs_device *dev;
  99         struct list_head *cur;
 100
 101         list_for_each(cur, head) {
 102                 dev = list_entry(cur, struct btrfs_device, dev_list);
 103                 if (dev->devid == devid &&
 104                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 105                         return dev;
 106                 }
 107         }
 108         return NULL;
 109 }
 110
 111 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 112 {
 113         struct list_head *cur;
 114         struct btrfs_fs_devices *fs_devices;
 115
 116         list_for_each(cur, &fs_uuids) {
 117                 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
 118                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 119                         return fs_devices;
 120         }
 121         return NULL;
 122 }
 123
 124 /*
 125  * we try to collect pending bios for a device so we don't get a large
 126  * number of procs sending bios down to the same device.  This greatly
 127  * improves the schedulers ability to collect and merge the bios.
 128  *
 129  * But, it also turns into a long list of bios to process and that is sure
 130  * to eventually make the worker thread block.  The solution here is to
 131  * make some progress and then put this work struct back at the end of
 132  * the list if the block device is congested.  This way, multiple devices
 133  * can make progress from a single worker thread.
 134  */
 135 static int noinline run_scheduled_bios(struct btrfs_device *device)
 136 {
 137         struct bio *pending;
 138         struct backing_dev_info *bdi;
 139         struct btrfs_fs_info *fs_info;
 140         struct bio *tail;
 141         struct bio *cur;
 142         int again = 0;
 143         unsigned long num_run = 0;
 144         unsigned long limit;
 145
 146         bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
 147         fs_info = device->dev_root->fs_info;
 148         limit = btrfs_async_submit_limit(fs_info);
 149         limit = limit * 2 / 3;
 150
 151 loop:
 152         spin_lock(&device->io_lock);
 153
 154         /* take all the bios off the list at once and process them
 155          * later on (without the lock held).  But, remember the
 156          * tail and other pointers so the bios can be properly reinserted
 157          * into the list if we hit congestion
 158          */
 159         pending = device->pending_bios;
 160         tail = device->pending_bio_tail;
 161         WARN_ON(pending && !tail);
 162         device->pending_bios = NULL;
 163         device->pending_bio_tail = NULL;
 164
 165         /*
 166          * if pending was null this time around, no bios need processing
 167          * at all and we can stop.  Otherwise it'll loop back up again
 168          * and do an additional check so no bios are missed.
 169          *
 170          * device->running_pending is used to synchronize with the
 171          * schedule_bio code.
 172          */
 173         if (pending) {
 174                 again = 1;
 175                 device->running_pending = 1;
 176         } else {
 177                 again = 0;
 178                 device->running_pending = 0;
 179         }
 180         spin_unlock(&device->io_lock);
 181
 182         while(pending) {
 183                 cur = pending;
 184                 pending = pending->bi_next;
 185                 cur->bi_next = NULL;
 186                 atomic_dec(&fs_info->nr_async_bios);
 187
 188                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 189                     waitqueue_active(&fs_info->async_submit_wait))
 190                         wake_up(&fs_info->async_submit_wait);
 191
 192                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 193                 bio_get(cur);
 194                 submit_bio(cur->bi_rw, cur);
 195                 bio_put(cur);
 196                 num_run++;
 197
 198                 /*
 199                  * we made progress, there is more work to do and the bdi
 200                  * is now congested.  Back off and let other work structs
 201                  * run instead
 202                  */
 203                 if (pending && bdi_write_congested(bdi)) {
 204                         struct bio *old_head;
 205
 206                         spin_lock(&device->io_lock);
 207
 208                         old_head = device->pending_bios;
 209                         device->pending_bios = pending;
 210                         if (device->pending_bio_tail)
 211                                 tail->bi_next = old_head;
 212                         else
 213                                 device->pending_bio_tail = tail;
 214
 215                         spin_unlock(&device->io_lock);
 216                         btrfs_requeue_work(&device->work);
 217                         goto done;
 218                 }
 219         }
 220         if (again)
 221                 goto loop;
 222 done:
 223         return 0;
 224 }
 225
 226 void pending_bios_fn(struct btrfs_work *work)
 227 {
 228         struct btrfs_device *device;
 229
 230         device = container_of(work, struct btrfs_device, work);
 231         run_scheduled_bios(device);
 232 }
 233
 234 static noinline int device_list_add(const char *path,
 235                            struct btrfs_super_block *disk_super,
 236                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 237 {
 238         struct btrfs_device *device;
 239         struct btrfs_fs_devices *fs_devices;
 240         u64 found_transid = btrfs_super_generation(disk_super);
 241
 242         fs_devices = find_fsid(disk_super->fsid);
 243         if (!fs_devices) {
 244                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 245                 if (!fs_devices)
 246                         return -ENOMEM;
 247                 INIT_LIST_HEAD(&fs_devices->devices);
 248                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 249                 list_add(&fs_devices->list, &fs_uuids);
 250                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 251                 fs_devices->latest_devid = devid;
 252                 fs_devices->latest_trans = found_transid;
 253                 device = NULL;
 254         } else {
 255                 device = __find_device(&fs_devices->devices, devid,
 256                                        disk_super->dev_item.uuid);
 257         }
 258         if (!device) {
 259                 device = kzalloc(sizeof(*device), GFP_NOFS);
 260                 if (!device) {
 261                         /* we can safely leave the fs_devices entry around */
 262                         return -ENOMEM;
 263                 }
 264                 device->devid = devid;
 265                 device->work.func = pending_bios_fn;
 266                 memcpy(device->uuid, disk_super->dev_item.uuid,
 267                        BTRFS_UUID_SIZE);
 268                 device->barriers = 1;
 269                 spin_lock_init(&device->io_lock);
 270                 device->name = kstrdup(path, GFP_NOFS);
 271                 if (!device->name) {
 272                         kfree(device);
 273                         return -ENOMEM;
 274                 }
 275                 list_add(&device->dev_list, &fs_devices->devices);
 276                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
 277                 fs_devices->num_devices++;
 278         }
 279
 280         if (found_transid > fs_devices->latest_trans) {
 281                 fs_devices->latest_devid = devid;
 282                 fs_devices->latest_trans = found_transid;
 283         }
 284         *fs_devices_ret = fs_devices;
 285         return 0;
 286 }
 287
 288 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 289 {
 290         struct list_head *head = &fs_devices->devices;
 291         struct list_head *cur;
 292         struct btrfs_device *device;
 293
 294         mutex_lock(&uuid_mutex);
 295 again:
 296         list_for_each(cur, head) {
 297                 device = list_entry(cur, struct btrfs_device, dev_list);
 298                 if (!device->in_fs_metadata) {
 299                         struct block_device *bdev;
 300                         list_del(&device->dev_list);
 301                         list_del(&device->dev_alloc_list);
 302                         fs_devices->num_devices--;
 303                         if (device->bdev) {
 304                                 bdev = device->bdev;
 305                                 fs_devices->open_devices--;
 306                                 mutex_unlock(&uuid_mutex);
 307                                 close_bdev_excl(bdev);
 308                                 mutex_lock(&uuid_mutex);
 309                         }
 310                         kfree(device->name);
 311                         kfree(device);
 312                         goto again;
 313                 }
 314         }
 315         mutex_unlock(&uuid_mutex);
 316         return 0;
 317 }
 318
 319 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 320 {
 321         struct list_head *head = &fs_devices->devices;
 322         struct list_head *cur;
 323         struct btrfs_device *device;
 324
 325         mutex_lock(&uuid_mutex);
 326         list_for_each(cur, head) {
 327                 device = list_entry(cur, struct btrfs_device, dev_list);
 328                 if (device->bdev) {
 329                         close_bdev_excl(device->bdev);
 330                         fs_devices->open_devices--;
 331                 }
 332                 device->bdev = NULL;
 333                 device->in_fs_metadata = 0;
 334         }
 335         fs_devices->mounted = 0;
 336         mutex_unlock(&uuid_mutex);
 337         return 0;
 338 }
 339
 340 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 341                        int flags, void *holder)
 342 {
 343         struct block_device *bdev;
 344         struct list_head *head = &fs_devices->devices;
 345         struct list_head *cur;
 346         struct btrfs_device *device;
 347         struct block_device *latest_bdev = NULL;
 348         struct buffer_head *bh;
 349         struct btrfs_super_block *disk_super;
 350         u64 latest_devid = 0;
 351         u64 latest_transid = 0;
 352         u64 transid;
 353         u64 devid;
 354         int ret = 0;
 355
 356         mutex_lock(&uuid_mutex);
 357         if (fs_devices->mounted)
 358                 goto out;
 359
 360         list_for_each(cur, head) {
 361                 device = list_entry(cur, struct btrfs_device, dev_list);
 362                 if (device->bdev)
 363                         continue;
 364
 365                 if (!device->name)
 366                         continue;
 367
 368                 bdev = open_bdev_excl(device->name, flags, holder);
 369
 370                 if (IS_ERR(bdev)) {
 371                         printk("open %s failed\n", device->name);
 372                         goto error;
 373                 }
 374                 set_blocksize(bdev, 4096);
 375
 376                 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 377                 if (!bh)
 378                         goto error_close;
 379
 380                 disk_super = (struct btrfs_super_block *)bh->b_data;
 381                 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 382                     sizeof(disk_super->magic)))
 383                         goto error_brelse;
 384
 385                 devid = le64_to_cpu(disk_super->dev_item.devid);
 386                 if (devid != device->devid)
 387                         goto error_brelse;
 388
 389                 transid = btrfs_super_generation(disk_super);
 390                 if (!latest_transid || transid > latest_transid) {
 391                         latest_devid = devid;
 392                         latest_transid = transid;
 393                         latest_bdev = bdev;
 394                 }
 395
 396                 device->bdev = bdev;
 397                 device->in_fs_metadata = 0;
 398                 fs_devices->open_devices++;
 399                 continue;
 400
 401 error_brelse:
 402                 brelse(bh);
 403 error_close:
 404                 close_bdev_excl(bdev);
 405 error:
 406                 continue;
 407         }
 408         if (fs_devices->open_devices == 0) {
 409                 ret = -EIO;
 410                 goto out;
 411         }
 412         fs_devices->mounted = 1;
 413         fs_devices->latest_bdev = latest_bdev;
 414         fs_devices->latest_devid = latest_devid;
 415         fs_devices->latest_trans = latest_transid;
 416 out:
 417         mutex_unlock(&uuid_mutex);
 418         return ret;
 419 }
 420
 421 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 422                           struct btrfs_fs_devices **fs_devices_ret)
 423 {
 424         struct btrfs_super_block *disk_super;
 425         struct block_device *bdev;
 426         struct buffer_head *bh;
 427         int ret;
 428         u64 devid;
 429         u64 transid;
 430
 431         mutex_lock(&uuid_mutex);
 432
 433         bdev = open_bdev_excl(path, flags, holder);
 434
 435         if (IS_ERR(bdev)) {
 436                 ret = PTR_ERR(bdev);
 437                 goto error;
 438         }
 439
 440         ret = set_blocksize(bdev, 4096);
 441         if (ret)
 442                 goto error_close;
 443         bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 444         if (!bh) {
 445                 ret = -EIO;
 446                 goto error_close;
 447         }
 448         disk_super = (struct btrfs_super_block *)bh->b_data;
 449         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 450             sizeof(disk_super->magic))) {
 451                 ret = -EINVAL;
 452                 goto error_brelse;
 453         }
 454         devid = le64_to_cpu(disk_super->dev_item.devid);
 455         transid = btrfs_super_generation(disk_super);
 456         if (disk_super->label[0])
 457                 printk("device label %s ", disk_super->label);
 458         else {
 459                 /* FIXME, make a readl uuid parser */
 460                 printk("device fsid %llx-%llx ",
 461                        *(unsigned long long *)disk_super->fsid,
 462                        *(unsigned long long *)(disk_super->fsid + 8));
 463         }
 464         printk("devid %Lu transid %Lu %s\n", devid, transid, path);
 465         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 466
 467 error_brelse:
 468         brelse(bh);
 469 error_close:
 470         close_bdev_excl(bdev);
 471 error:
 472         mutex_unlock(&uuid_mutex);
 473         return ret;
 474 }
 475
 476 /*
 477  * this uses a pretty simple search, the expectation is that it is
 478  * called very infrequently and that a given device has a small number
 479  * of extents
 480  */
 481 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
 482                                          struct btrfs_device *device,
 483                                          struct btrfs_path *path,
 484                                          u64 num_bytes, u64 *start)
 485 {
 486         struct btrfs_key key;
 487         struct btrfs_root *root = device->dev_root;
 488         struct btrfs_dev_extent *dev_extent = NULL;
 489         u64 hole_size = 0;
 490         u64 last_byte = 0;
 491         u64 search_start = 0;
 492         u64 search_end = device->total_bytes;
 493         int ret;
 494         int slot = 0;
 495         int start_found;
 496         struct extent_buffer *l;
 497
 498         start_found = 0;
 499         path->reada = 2;
 500
 501         /* FIXME use last free of some kind */
 502
 503         /* we don't want to overwrite the superblock on the drive,
 504          * so we make sure to start at an offset of at least 1MB
 505          */
 506         search_start = max((u64)1024 * 1024, search_start);
 507
 508         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
 509                 search_start = max(root->fs_info->alloc_start, search_start);
 510
 511         key.objectid = device->devid;
 512         key.offset = search_start;
 513         key.type = BTRFS_DEV_EXTENT_KEY;
 514         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 515         if (ret < 0)
 516                 goto error;
 517         ret = btrfs_previous_item(root, path, 0, key.type);
 518         if (ret < 0)
 519                 goto error;
 520         l = path->nodes[0];
 521         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 522         while (1) {
 523                 l = path->nodes[0];
 524                 slot = path->slots[0];
 525                 if (slot >= btrfs_header_nritems(l)) {
 526                         ret = btrfs_next_leaf(root, path);
 527                         if (ret == 0)
 528                                 continue;
 529                         if (ret < 0)
 530                                 goto error;
 531 no_more_items:
 532                         if (!start_found) {
 533                                 if (search_start >= search_end) {
 534                                         ret = -ENOSPC;
 535                                         goto error;
 536                                 }
 537                                 *start = search_start;
 538                                 start_found = 1;
 539                                 goto check_pending;
 540                         }
 541                         *start = last_byte > search_start ?
 542                                 last_byte : search_start;
 543                         if (search_end <= *start) {
 544                                 ret = -ENOSPC;
 545                                 goto error;
 546                         }
 547                         goto check_pending;
 548                 }
 549                 btrfs_item_key_to_cpu(l, &key, slot);
 550
 551                 if (key.objectid < device->devid)
 552                         goto next;
 553
 554                 if (key.objectid > device->devid)
 555                         goto no_more_items;
 556
 557                 if (key.offset >= search_start && key.offset > last_byte &&
 558                     start_found) {
 559                         if (last_byte < search_start)
 560                                 last_byte = search_start;
 561                         hole_size = key.offset - last_byte;
 562                         if (key.offset > last_byte &&
 563                             hole_size >= num_bytes) {
 564                                 *start = last_byte;
 565                                 goto check_pending;
 566                         }
 567                 }
 568                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
 569                         goto next;
 570                 }
 571
 572                 start_found = 1;
 573                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 574                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
 575 next:
 576                 path->slots[0]++;
 577                 cond_resched();
 578         }
 579 check_pending:
 580         /* we have to make sure we didn't find an extent that has already
 581          * been allocated by the map tree or the original allocation
 582          */
 583         btrfs_release_path(root, path);
 584         BUG_ON(*start < search_start);
 585
 586         if (*start + num_bytes > search_end) {
 587                 ret = -ENOSPC;
 588                 goto error;
 589         }
 590         /* check for pending inserts here */
 591         return 0;
 592
 593 error:
 594         btrfs_release_path(root, path);
 595         return ret;
 596 }
 597
 598 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 599                           struct btrfs_device *device,
 600                           u64 start)
 601 {
 602         int ret;
 603         struct btrfs_path *path;
 604         struct btrfs_root *root = device->dev_root;
 605         struct btrfs_key key;
 606         struct btrfs_key found_key;
 607         struct extent_buffer *leaf = NULL;
 608         struct btrfs_dev_extent *extent = NULL;
 609
 610         path = btrfs_alloc_path();
 611         if (!path)
 612                 return -ENOMEM;
 613
 614         key.objectid = device->devid;
 615         key.offset = start;
 616         key.type = BTRFS_DEV_EXTENT_KEY;
 617
 618         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 619         if (ret > 0) {
 620                 ret = btrfs_previous_item(root, path, key.objectid,
 621                                           BTRFS_DEV_EXTENT_KEY);
 622                 BUG_ON(ret);
 623                 leaf = path->nodes[0];
 624                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 625                 extent = btrfs_item_ptr(leaf, path->slots[0],
 626                                         struct btrfs_dev_extent);
 627                 BUG_ON(found_key.offset > start || found_key.offset +
 628                        btrfs_dev_extent_length(leaf, extent) < start);
 629                 ret = 0;
 630         } else if (ret == 0) {
 631                 leaf = path->nodes[0];
 632                 extent = btrfs_item_ptr(leaf, path->slots[0],
 633                                         struct btrfs_dev_extent);
 634         }
 635         BUG_ON(ret);
 636
 637         if (device->bytes_used > 0)
 638                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 639         ret = btrfs_del_item(trans, root, path);
 640         BUG_ON(ret);
 641
 642         btrfs_free_path(path);
 643         return ret;
 644 }
 645
 646 int noinline btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 647                            struct btrfs_device *device,
 648                            u64 chunk_tree, u64 chunk_objectid,
 649                            u64 chunk_offset,
 650                            u64 num_bytes, u64 *start)
 651 {
 652         int ret;
 653         struct btrfs_path *path;
 654         struct btrfs_root *root = device->dev_root;
 655         struct btrfs_dev_extent *extent;
 656         struct extent_buffer *leaf;
 657         struct btrfs_key key;
 658
 659         WARN_ON(!device->in_fs_metadata);
 660         path = btrfs_alloc_path();
 661         if (!path)
 662                 return -ENOMEM;
 663
 664         ret = find_free_dev_extent(trans, device, path, num_bytes, start);
 665         if (ret) {
 666                 goto err;
 667         }
 668
 669         key.objectid = device->devid;
 670         key.offset = *start;
 671         key.type = BTRFS_DEV_EXTENT_KEY;
 672         ret = btrfs_insert_empty_item(trans, root, path, &key,
 673                                       sizeof(*extent));
 674         BUG_ON(ret);
 675
 676         leaf = path->nodes[0];
 677         extent = btrfs_item_ptr(leaf, path->slots[0],
 678                                 struct btrfs_dev_extent);
 679         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
 680         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
 681         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
 682
 683         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 684                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
 685                     BTRFS_UUID_SIZE);
 686
 687         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
 688         btrfs_mark_buffer_dirty(leaf);
 689 err:
 690         btrfs_free_path(path);
 691         return ret;
 692 }
 693
 694 static noinline int find_next_chunk(struct btrfs_root *root,
 695                                     u64 objectid, u64 *offset)
 696 {
 697         struct btrfs_path *path;
 698         int ret;
 699         struct btrfs_key key;
 700         struct btrfs_chunk *chunk;
 701         struct btrfs_key found_key;
 702
 703         path = btrfs_alloc_path();
 704         BUG_ON(!path);
 705
 706         key.objectid = objectid;
 707         key.offset = (u64)-1;
 708         key.type = BTRFS_CHUNK_ITEM_KEY;
 709
 710         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 711         if (ret < 0)
 712                 goto error;
 713
 714         BUG_ON(ret == 0);
 715
 716         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
 717         if (ret) {
 718                 *offset = 0;
 719         } else {
 720                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 721                                       path->slots[0]);
 722                 if (found_key.objectid != objectid)
 723                         *offset = 0;
 724                 else {
 725                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
 726                                                struct btrfs_chunk);
 727                         *offset = found_key.offset +
 728                                 btrfs_chunk_length(path->nodes[0], chunk);
 729                 }
 730         }
 731         ret = 0;
 732 error:
 733         btrfs_free_path(path);
 734         return ret;
 735 }
 736
 737 static noinline int find_next_devid(struct btrfs_root *root,
 738                                     struct btrfs_path *path, u64 *objectid)
 739 {
 740         int ret;
 741         struct btrfs_key key;
 742         struct btrfs_key found_key;
 743
 744         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 745         key.type = BTRFS_DEV_ITEM_KEY;
 746         key.offset = (u64)-1;
 747
 748         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 749         if (ret < 0)
 750                 goto error;
 751
 752         BUG_ON(ret == 0);
 753
 754         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
 755                                   BTRFS_DEV_ITEM_KEY);
 756         if (ret) {
 757                 *objectid = 1;
 758         } else {
 759                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 760                                       path->slots[0]);
 761                 *objectid = found_key.offset + 1;
 762         }
 763         ret = 0;
 764 error:
 765         btrfs_release_path(root, path);
 766         return ret;
 767 }
 768
 769 /*
 770  * the device information is stored in the chunk root
 771  * the btrfs_device struct should be fully filled in
 772  */
 773 int btrfs_add_device(struct btrfs_trans_handle *trans,
 774                      struct btrfs_root *root,
 775                      struct btrfs_device *device)
 776 {
 777         int ret;
 778         struct btrfs_path *path;
 779         struct btrfs_dev_item *dev_item;
 780         struct extent_buffer *leaf;
 781         struct btrfs_key key;
 782         unsigned long ptr;
 783         u64 free_devid = 0;
 784
 785         root = root->fs_info->chunk_root;
 786
 787         path = btrfs_alloc_path();
 788         if (!path)
 789                 return -ENOMEM;
 790
 791         ret = find_next_devid(root, path, &free_devid);
 792         if (ret)
 793                 goto out;
 794
 795         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 796         key.type = BTRFS_DEV_ITEM_KEY;
 797         key.offset = free_devid;
 798
 799         ret = btrfs_insert_empty_item(trans, root, path, &key,
 800                                       sizeof(*dev_item));
 801         if (ret)
 802                 goto out;
 803
 804         leaf = path->nodes[0];
 805         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
 806
 807         device->devid = free_devid;
 808         btrfs_set_device_id(leaf, dev_item, device->devid);
 809         btrfs_set_device_type(leaf, dev_item, device->type);
 810         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
 811         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
 812         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
 813         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
 814         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
 815         btrfs_set_device_group(leaf, dev_item, 0);
 816         btrfs_set_device_seek_speed(leaf, dev_item, 0);
 817         btrfs_set_device_bandwidth(leaf, dev_item, 0);
 818
 819         ptr = (unsigned long)btrfs_device_uuid(dev_item);
 820         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
 821         btrfs_mark_buffer_dirty(leaf);
 822         ret = 0;
 823
 824 out:
 825         btrfs_free_path(path);
 826         return ret;
 827 }
 828
 829 static int btrfs_rm_dev_item(struct btrfs_root *root,
 830                              struct btrfs_device *device)
 831 {
 832         int ret;
 833         struct btrfs_path *path;
 834         struct block_device *bdev = device->bdev;
 835         struct btrfs_device *next_dev;
 836         struct btrfs_key key;
 837         u64 total_bytes;
 838         struct btrfs_fs_devices *fs_devices;
 839         struct btrfs_trans_handle *trans;
 840
 841         root = root->fs_info->chunk_root;
 842
 843         path = btrfs_alloc_path();
 844         if (!path)
 845                 return -ENOMEM;
 846
 847         trans = btrfs_start_transaction(root, 1);
 848         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 849         key.type = BTRFS_DEV_ITEM_KEY;
 850         key.offset = device->devid;
 851         lock_chunks(root);
 852
 853         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 854         if (ret < 0)
 855                 goto out;
 856
 857         if (ret > 0) {
 858                 ret = -ENOENT;
 859                 goto out;
 860         }
 861
 862         ret = btrfs_del_item(trans, root, path);
 863         if (ret)
 864                 goto out;
 865
 866         /*
 867          * at this point, the device is zero sized.  We want to
 868          * remove it from the devices list and zero out the old super
 869          */
 870         list_del_init(&device->dev_list);
 871         list_del_init(&device->dev_alloc_list);
 872         fs_devices = root->fs_info->fs_devices;
 873
 874         next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
 875                               dev_list);
 876         if (bdev == root->fs_info->sb->s_bdev)
 877                 root->fs_info->sb->s_bdev = next_dev->bdev;
 878         if (bdev == fs_devices->latest_bdev)
 879                 fs_devices->latest_bdev = next_dev->bdev;
 880
 881         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 882         btrfs_set_super_num_devices(&root->fs_info->super_copy,
 883                                     total_bytes - 1);
 884 out:
 885         btrfs_free_path(path);
 886         unlock_chunks(root);
 887         btrfs_commit_transaction(trans, root);
 888         return ret;
 889 }
 890
 891 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 892 {
 893         struct btrfs_device *device;
 894         struct block_device *bdev;
 895         struct buffer_head *bh = NULL;
 896         struct btrfs_super_block *disk_super;
 897         u64 all_avail;
 898         u64 devid;
 899         int ret = 0;
 900
 901         mutex_lock(&uuid_mutex);
 902         mutex_lock(&root->fs_info->volume_mutex);
 903
 904         all_avail = root->fs_info->avail_data_alloc_bits |
 905                 root->fs_info->avail_system_alloc_bits |
 906                 root->fs_info->avail_metadata_alloc_bits;
 907
 908         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
 909             btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
 910                 printk("btrfs: unable to go below four devices on raid10\n");
 911                 ret = -EINVAL;
 912                 goto out;
 913         }
 914
 915         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
 916             btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
 917                 printk("btrfs: unable to go below two devices on raid1\n");
 918                 ret = -EINVAL;
 919                 goto out;
 920         }
 921
 922         if (strcmp(device_path, "missing") == 0) {
 923                 struct list_head *cur;
 924                 struct list_head *devices;
 925                 struct btrfs_device *tmp;
 926
 927                 device = NULL;
 928                 devices = &root->fs_info->fs_devices->devices;
 929                 list_for_each(cur, devices) {
 930                         tmp = list_entry(cur, struct btrfs_device, dev_list);
 931                         if (tmp->in_fs_metadata && !tmp->bdev) {
 932                                 device = tmp;
 933                                 break;
 934                         }
 935                 }
 936                 bdev = NULL;
 937                 bh = NULL;
 938                 disk_super = NULL;
 939                 if (!device) {
 940                         printk("btrfs: no missing devices found to remove\n");
 941                         goto out;
 942                 }
 943
 944         } else {
 945                 bdev = open_bdev_excl(device_path, 0,
 946                                       root->fs_info->bdev_holder);
 947                 if (IS_ERR(bdev)) {
 948                         ret = PTR_ERR(bdev);
 949                         goto out;
 950                 }
 951
 952                 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
 953                 if (!bh) {
 954                         ret = -EIO;
 955                         goto error_close;
 956                 }
 957                 disk_super = (struct btrfs_super_block *)bh->b_data;
 958                 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 959                     sizeof(disk_super->magic))) {
 960                         ret = -ENOENT;
 961                         goto error_brelse;
 962                 }
 963                 if (memcmp(disk_super->fsid, root->fs_info->fsid,
 964                            BTRFS_FSID_SIZE)) {
 965                         ret = -ENOENT;
 966                         goto error_brelse;
 967                 }
 968                 devid = le64_to_cpu(disk_super->dev_item.devid);
 969                 device = btrfs_find_device(root, devid, NULL);
 970                 if (!device) {
 971                         ret = -ENOENT;
 972                         goto error_brelse;
 973                 }
 974
 975         }
 976         root->fs_info->fs_devices->num_devices--;
 977         root->fs_info->fs_devices->open_devices--;
 978
 979         ret = btrfs_shrink_device(device, 0);
 980         if (ret)
 981                 goto error_brelse;
 982
 983
 984         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
 985         if (ret)
 986                 goto error_brelse;
 987
 988         if (bh) {
 989                 /* make sure this device isn't detected as part of
 990                  * the FS anymore
 991                  */
 992                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
 993                 set_buffer_dirty(bh);
 994                 sync_dirty_buffer(bh);
 995
 996                 brelse(bh);
 997         }
 998
 999         if (device->bdev) {
1000                 /* one close for the device struct or super_block */
1001                 close_bdev_excl(device->bdev);
1002         }
1003         if (bdev) {
1004                 /* one close for us */
1005                 close_bdev_excl(bdev);
1006         }
1007         kfree(device->name);
1008         kfree(device);
1009         ret = 0;
1010         goto out;
1011
1012 error_brelse:
1013         brelse(bh);
1014 error_close:
1015         if (bdev)
1016                 close_bdev_excl(bdev);
1017 out:
1018         mutex_unlock(&root->fs_info->volume_mutex);
1019         mutex_unlock(&uuid_mutex);
1020         return ret;
1021 }
1022
1023 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1024 {
1025         struct btrfs_trans_handle *trans;
1026         struct btrfs_device *device;
1027         struct block_device *bdev;
1028         struct list_head *cur;
1029         struct list_head *devices;
1030         u64 total_bytes;
1031         int ret = 0;
1032
1033
1034         bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1035         if (!bdev) {
1036                 return -EIO;
1037         }
1038
1039         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1040         mutex_lock(&root->fs_info->volume_mutex);
1041
1042         trans = btrfs_start_transaction(root, 1);
1043         lock_chunks(root);
1044         devices = &root->fs_info->fs_devices->devices;
1045         list_for_each(cur, devices) {
1046                 device = list_entry(cur, struct btrfs_device, dev_list);
1047                 if (device->bdev == bdev) {
1048                         ret = -EEXIST;
1049                         goto out;
1050                 }
1051         }
1052
1053         device = kzalloc(sizeof(*device), GFP_NOFS);
1054         if (!device) {
1055                 /* we can safely leave the fs_devices entry around */
1056                 ret = -ENOMEM;
1057                 goto out_close_bdev;
1058         }
1059
1060         device->barriers = 1;
1061         device->work.func = pending_bios_fn;
1062         generate_random_uuid(device->uuid);
1063         spin_lock_init(&device->io_lock);
1064         device->name = kstrdup(device_path, GFP_NOFS);
1065         if (!device->name) {
1066                 kfree(device);
1067                 goto out_close_bdev;
1068         }
1069         device->io_width = root->sectorsize;
1070         device->io_align = root->sectorsize;
1071         device->sector_size = root->sectorsize;
1072         device->total_bytes = i_size_read(bdev->bd_inode);
1073         device->dev_root = root->fs_info->dev_root;
1074         device->bdev = bdev;
1075         device->in_fs_metadata = 1;
1076
1077         ret = btrfs_add_device(trans, root, device);
1078         if (ret)
1079                 goto out_close_bdev;
1080
1081         set_blocksize(device->bdev, 4096);
1082
1083         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1084         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1085                                     total_bytes + device->total_bytes);
1086
1087         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1088         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1089                                     total_bytes + 1);
1090
1091         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1092         list_add(&device->dev_alloc_list,
1093                  &root->fs_info->fs_devices->alloc_list);
1094         root->fs_info->fs_devices->num_devices++;
1095         root->fs_info->fs_devices->open_devices++;
1096 out:
1097         unlock_chunks(root);
1098         btrfs_end_transaction(trans, root);
1099         mutex_unlock(&root->fs_info->volume_mutex);
1100
1101         return ret;
1102
1103 out_close_bdev:
1104         close_bdev_excl(bdev);
1105         goto out;
1106 }
1107
1108 int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1109                                  struct btrfs_device *device)
1110 {
1111         int ret;
1112         struct btrfs_path *path;
1113         struct btrfs_root *root;
1114         struct btrfs_dev_item *dev_item;
1115         struct extent_buffer *leaf;
1116         struct btrfs_key key;
1117
1118         root = device->dev_root->fs_info->chunk_root;
1119
1120         path = btrfs_alloc_path();
1121         if (!path)
1122                 return -ENOMEM;
1123
1124         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1125         key.type = BTRFS_DEV_ITEM_KEY;
1126         key.offset = device->devid;
1127
1128         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1129         if (ret < 0)
1130                 goto out;
1131
1132         if (ret > 0) {
1133                 ret = -ENOENT;
1134                 goto out;
1135         }
1136
1137         leaf = path->nodes[0];
1138         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1139
1140         btrfs_set_device_id(leaf, dev_item, device->devid);
1141         btrfs_set_device_type(leaf, dev_item, device->type);
1142         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1143         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1144         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1145         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1146         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1147         btrfs_mark_buffer_dirty(leaf);
1148
1149 out:
1150         btrfs_free_path(path);
1151         return ret;
1152 }
1153
1154 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1155                       struct btrfs_device *device, u64 new_size)
1156 {
1157         struct btrfs_super_block *super_copy =
1158                 &device->dev_root->fs_info->super_copy;
1159         u64 old_total = btrfs_super_total_bytes(super_copy);
1160         u64 diff = new_size - device->total_bytes;
1161
1162         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1163         return btrfs_update_device(trans, device);
1164 }
1165
1166 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1167                       struct btrfs_device *device, u64 new_size)
1168 {
1169         int ret;
1170         lock_chunks(device->dev_root);
1171         ret = __btrfs_grow_device(trans, device, new_size);
1172         unlock_chunks(device->dev_root);
1173         return ret;
1174 }
1175
1176 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1177                             struct btrfs_root *root,
1178                             u64 chunk_tree, u64 chunk_objectid,
1179                             u64 chunk_offset)
1180 {
1181         int ret;
1182         struct btrfs_path *path;
1183         struct btrfs_key key;
1184
1185         root = root->fs_info->chunk_root;
1186         path = btrfs_alloc_path();
1187         if (!path)
1188                 return -ENOMEM;
1189
1190         key.objectid = chunk_objectid;
1191         key.offset = chunk_offset;
1192         key.type = BTRFS_CHUNK_ITEM_KEY;
1193
1194         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1195         BUG_ON(ret);
1196
1197         ret = btrfs_del_item(trans, root, path);
1198         BUG_ON(ret);
1199
1200         btrfs_free_path(path);
1201         return 0;
1202 }
1203
1204 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1205                         chunk_offset)
1206 {
1207         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1208         struct btrfs_disk_key *disk_key;
1209         struct btrfs_chunk *chunk;
1210         u8 *ptr;
1211         int ret = 0;
1212         u32 num_stripes;
1213         u32 array_size;
1214         u32 len = 0;
1215         u32 cur;
1216         struct btrfs_key key;
1217
1218         array_size = btrfs_super_sys_array_size(super_copy);
1219
1220         ptr = super_copy->sys_chunk_array;
1221         cur = 0;
1222
1223         while (cur < array_size) {
1224                 disk_key = (struct btrfs_disk_key *)ptr;
1225                 btrfs_disk_key_to_cpu(&key, disk_key);
1226
1227                 len = sizeof(*disk_key);
1228
1229                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1230                         chunk = (struct btrfs_chunk *)(ptr + len);
1231                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1232                         len += btrfs_chunk_item_size(num_stripes);
1233                 } else {
1234                         ret = -EIO;
1235                         break;
1236                 }
1237                 if (key.objectid == chunk_objectid &&
1238                     key.offset == chunk_offset) {
1239                         memmove(ptr, ptr + len, array_size - (cur + len));
1240                         array_size -= len;
1241                         btrfs_set_super_sys_array_size(super_copy, array_size);
1242                 } else {
1243                         ptr += len;
1244                         cur += len;
1245                 }
1246         }
1247         return ret;
1248 }
1249
1250
1251 int btrfs_relocate_chunk(struct btrfs_root *root,
1252                          u64 chunk_tree, u64 chunk_objectid,
1253                          u64 chunk_offset)
1254 {
1255         struct extent_map_tree *em_tree;
1256         struct btrfs_root *extent_root;
1257         struct btrfs_trans_handle *trans;
1258         struct extent_map *em;
1259         struct map_lookup *map;
1260         int ret;
1261         int i;
1262
1263         printk("btrfs relocating chunk %llu\n",
1264                (unsigned long long)chunk_offset);
1265         root = root->fs_info->chunk_root;
1266         extent_root = root->fs_info->extent_root;
1267         em_tree = &root->fs_info->mapping_tree.map_tree;
1268
1269         /* step one, relocate all the extents inside this chunk */
1270         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1271         BUG_ON(ret);
1272
1273         trans = btrfs_start_transaction(root, 1);
1274         BUG_ON(!trans);
1275
1276         lock_chunks(root);
1277
1278         /*
1279          * step two, delete the device extents and the
1280          * chunk tree entries
1281          */
1282         spin_lock(&em_tree->lock);
1283         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1284         spin_unlock(&em_tree->lock);
1285
1286         BUG_ON(em->start > chunk_offset ||
1287                em->start + em->len < chunk_offset);
1288         map = (struct map_lookup *)em->bdev;
1289
1290         for (i = 0; i < map->num_stripes; i++) {
1291                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1292                                             map->stripes[i].physical);
1293                 BUG_ON(ret);
1294
1295                 if (map->stripes[i].dev) {
1296                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1297                         BUG_ON(ret);
1298                 }
1299         }
1300         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1301                                chunk_offset);
1302
1303         BUG_ON(ret);
1304
1305         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1306                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1307                 BUG_ON(ret);
1308         }
1309
1310         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1311         BUG_ON(ret);
1312
1313         spin_lock(&em_tree->lock);
1314         remove_extent_mapping(em_tree, em);
1315         spin_unlock(&em_tree->lock);
1316
1317         kfree(map);
1318         em->bdev = NULL;
1319
1320         /* once for the tree */
1321         free_extent_map(em);
1322         /* once for us */
1323         free_extent_map(em);
1324
1325         unlock_chunks(root);
1326         btrfs_end_transaction(trans, root);
1327         return 0;
1328 }
1329
1330 static u64 div_factor(u64 num, int factor)
1331 {
1332         if (factor == 10)
1333                 return num;
1334         num *= factor;
1335         do_div(num, 10);
1336         return num;
1337 }
1338
1339
1340 int btrfs_balance(struct btrfs_root *dev_root)
1341 {
1342         int ret;
1343         struct list_head *cur;
1344         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1345         struct btrfs_device *device;
1346         u64 old_size;
1347         u64 size_to_free;
1348         struct btrfs_path *path;
1349         struct btrfs_key key;
1350         struct btrfs_chunk *chunk;
1351         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1352         struct btrfs_trans_handle *trans;
1353         struct btrfs_key found_key;
1354
1355
1356         mutex_lock(&dev_root->fs_info->volume_mutex);
1357         dev_root = dev_root->fs_info->dev_root;
1358
1359         /* step one make some room on all the devices */
1360         list_for_each(cur, devices) {
1361                 device = list_entry(cur, struct btrfs_device, dev_list);
1362                 old_size = device->total_bytes;
1363                 size_to_free = div_factor(old_size, 1);
1364                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1365                 if (device->total_bytes - device->bytes_used > size_to_free)
1366                         continue;
1367
1368                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1369                 BUG_ON(ret);
1370
1371                 trans = btrfs_start_transaction(dev_root, 1);
1372                 BUG_ON(!trans);
1373
1374                 ret = btrfs_grow_device(trans, device, old_size);
1375                 BUG_ON(ret);
1376
1377                 btrfs_end_transaction(trans, dev_root);
1378         }
1379
1380         /* step two, relocate all the chunks */
1381         path = btrfs_alloc_path();
1382         BUG_ON(!path);
1383
1384         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1385         key.offset = (u64)-1;
1386         key.type = BTRFS_CHUNK_ITEM_KEY;
1387
1388         while(1) {
1389                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1390                 if (ret < 0)
1391                         goto error;
1392
1393                 /*
1394                  * this shouldn't happen, it means the last relocate
1395                  * failed
1396                  */
1397                 if (ret == 0)
1398                         break;
1399
1400                 ret = btrfs_previous_item(chunk_root, path, 0,
1401                                           BTRFS_CHUNK_ITEM_KEY);
1402                 if (ret)
1403                         break;
1404
1405                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1406                                       path->slots[0]);
1407                 if (found_key.objectid != key.objectid)
1408                         break;
1409
1410                 chunk = btrfs_item_ptr(path->nodes[0],
1411                                        path->slots[0],
1412                                        struct btrfs_chunk);
1413                 key.offset = found_key.offset;
1414                 /* chunk zero is special */
1415                 if (key.offset == 0)
1416                         break;
1417
1418                 btrfs_release_path(chunk_root, path);
1419                 ret = btrfs_relocate_chunk(chunk_root,
1420                                            chunk_root->root_key.objectid,
1421                                            found_key.objectid,
1422                                            found_key.offset);
1423                 BUG_ON(ret);
1424         }
1425         ret = 0;
1426 error:
1427         btrfs_free_path(path);
1428         mutex_unlock(&dev_root->fs_info->volume_mutex);
1429         return ret;
1430 }
1431
1432 /*
1433  * shrinking a device means finding all of the device extents past
1434  * the new size, and then following the back refs to the chunks.
1435  * The chunk relocation code actually frees the device extent
1436  */
1437 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1438 {
1439         struct btrfs_trans_handle *trans;
1440         struct btrfs_root *root = device->dev_root;
1441         struct btrfs_dev_extent *dev_extent = NULL;
1442         struct btrfs_path *path;
1443         u64 length;
1444         u64 chunk_tree;
1445         u64 chunk_objectid;
1446         u64 chunk_offset;
1447         int ret;
1448         int slot;
1449         struct extent_buffer *l;
1450         struct btrfs_key key;
1451         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1452         u64 old_total = btrfs_super_total_bytes(super_copy);
1453         u64 diff = device->total_bytes - new_size;
1454
1455
1456         path = btrfs_alloc_path();
1457         if (!path)
1458                 return -ENOMEM;
1459
1460         trans = btrfs_start_transaction(root, 1);
1461         if (!trans) {
1462                 ret = -ENOMEM;
1463                 goto done;
1464         }
1465
1466         path->reada = 2;
1467
1468         lock_chunks(root);
1469
1470         device->total_bytes = new_size;
1471         ret = btrfs_update_device(trans, device);
1472         if (ret) {
1473                 unlock_chunks(root);
1474                 btrfs_end_transaction(trans, root);
1475                 goto done;
1476         }
1477         WARN_ON(diff > old_total);
1478         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1479         unlock_chunks(root);
1480         btrfs_end_transaction(trans, root);
1481
1482         key.objectid = device->devid;
1483         key.offset = (u64)-1;
1484         key.type = BTRFS_DEV_EXTENT_KEY;
1485
1486         while (1) {
1487                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1488                 if (ret < 0)
1489                         goto done;
1490
1491                 ret = btrfs_previous_item(root, path, 0, key.type);
1492                 if (ret < 0)
1493                         goto done;
1494                 if (ret) {
1495                         ret = 0;
1496                         goto done;
1497                 }
1498
1499                 l = path->nodes[0];
1500                 slot = path->slots[0];
1501                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1502
1503                 if (key.objectid != device->devid)
1504                         goto done;
1505
1506                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1507                 length = btrfs_dev_extent_length(l, dev_extent);
1508
1509                 if (key.offset + length <= new_size)
1510                         goto done;
1511
1512                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1513                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1514                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1515                 btrfs_release_path(root, path);
1516
1517                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1518                                            chunk_offset);
1519                 if (ret)
1520                         goto done;
1521         }
1522
1523 done:
1524         btrfs_free_path(path);
1525         return ret;
1526 }
1527
1528 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1529                            struct btrfs_root *root,
1530                            struct btrfs_key *key,
1531                            struct btrfs_chunk *chunk, int item_size)
1532 {
1533         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1534         struct btrfs_disk_key disk_key;
1535         u32 array_size;
1536         u8 *ptr;
1537
1538         array_size = btrfs_super_sys_array_size(super_copy);
1539         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1540                 return -EFBIG;
1541
1542         ptr = super_copy->sys_chunk_array + array_size;
1543         btrfs_cpu_key_to_disk(&disk_key, key);
1544         memcpy(ptr, &disk_key, sizeof(disk_key));
1545         ptr += sizeof(disk_key);
1546         memcpy(ptr, chunk, item_size);
1547         item_size += sizeof(disk_key);
1548         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1549         return 0;
1550 }
1551
1552 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1553                                         int num_stripes, int sub_stripes)
1554 {
1555         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1556                 return calc_size;
1557         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1558                 return calc_size * (num_stripes / sub_stripes);
1559         else
1560                 return calc_size * num_stripes;
1561 }
1562
1563
1564 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1565                       struct btrfs_root *extent_root, u64 *start,
1566                       u64 *num_bytes, u64 type)
1567 {
1568         u64 dev_offset;
1569         struct btrfs_fs_info *info = extent_root->fs_info;
1570         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1571         struct btrfs_path *path;
1572         struct btrfs_stripe *stripes;
1573         struct btrfs_device *device = NULL;
1574         struct btrfs_chunk *chunk;
1575         struct list_head private_devs;
1576         struct list_head *dev_list;
1577         struct list_head *cur;
1578         struct extent_map_tree *em_tree;
1579         struct map_lookup *map;
1580         struct extent_map *em;
1581         int min_stripe_size = 1 * 1024 * 1024;
1582         u64 physical;
1583         u64 calc_size = 1024 * 1024 * 1024;
1584         u64 max_chunk_size = calc_size;
1585         u64 min_free;
1586         u64 avail;
1587         u64 max_avail = 0;
1588         u64 percent_max;
1589         int num_stripes = 1;
1590         int min_stripes = 1;
1591         int sub_stripes = 0;
1592         int looped = 0;
1593         int ret;
1594         int index;
1595         int stripe_len = 64 * 1024;
1596         struct btrfs_key key;
1597
1598         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1599             (type & BTRFS_BLOCK_GROUP_DUP)) {
1600                 WARN_ON(1);
1601                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1602         }
1603         dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1604         if (list_empty(dev_list))
1605                 return -ENOSPC;
1606
1607         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1608                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1609                 min_stripes = 2;
1610         }
1611         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1612                 num_stripes = 2;
1613                 min_stripes = 2;
1614         }
1615         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1616                 num_stripes = min_t(u64, 2,
1617                             extent_root->fs_info->fs_devices->open_devices);
1618                 if (num_stripes < 2)
1619                         return -ENOSPC;
1620                 min_stripes = 2;
1621         }
1622         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1623                 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1624                 if (num_stripes < 4)
1625                         return -ENOSPC;
1626                 num_stripes &= ~(u32)1;
1627                 sub_stripes = 2;
1628                 min_stripes = 4;
1629         }
1630
1631         if (type & BTRFS_BLOCK_GROUP_DATA) {
1632                 max_chunk_size = 10 * calc_size;
1633                 min_stripe_size = 64 * 1024 * 1024;
1634         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1635                 max_chunk_size = 4 * calc_size;
1636                 min_stripe_size = 32 * 1024 * 1024;
1637         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1638                 calc_size = 8 * 1024 * 1024;
1639                 max_chunk_size = calc_size * 2;
1640                 min_stripe_size = 1 * 1024 * 1024;
1641         }
1642
1643         path = btrfs_alloc_path();
1644         if (!path)
1645                 return -ENOMEM;
1646
1647         /* we don't want a chunk larger than 10% of the FS */
1648         percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1649         max_chunk_size = min(percent_max, max_chunk_size);
1650
1651 again:
1652         if (calc_size * num_stripes > max_chunk_size) {
1653                 calc_size = max_chunk_size;
1654                 do_div(calc_size, num_stripes);
1655                 do_div(calc_size, stripe_len);
1656                 calc_size *= stripe_len;
1657         }
1658         /* we don't want tiny stripes */
1659         calc_size = max_t(u64, min_stripe_size, calc_size);
1660
1661         do_div(calc_size, stripe_len);
1662         calc_size *= stripe_len;
1663
1664         INIT_LIST_HEAD(&private_devs);
1665         cur = dev_list->next;
1666         index = 0;
1667
1668         if (type & BTRFS_BLOCK_GROUP_DUP)
1669                 min_free = calc_size * 2;
1670         else
1671                 min_free = calc_size;
1672
1673         /*
1674          * we add 1MB because we never use the first 1MB of the device, unless
1675          * we've looped, then we are likely allocating the maximum amount of
1676          * space left already
1677          */
1678         if (!looped)
1679                 min_free += 1024 * 1024;
1680
1681         /* build a private list of devices we will allocate from */
1682         while(index < num_stripes) {
1683                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1684
1685                 if (device->total_bytes > device->bytes_used)
1686                         avail = device->total_bytes - device->bytes_used;
1687                 else
1688                         avail = 0;
1689                 cur = cur->next;
1690
1691                 if (device->in_fs_metadata && avail >= min_free) {
1692                         u64 ignored_start = 0;
1693                         ret = find_free_dev_extent(trans, device, path,
1694                                                    min_free,
1695                                                    &ignored_start);
1696                         if (ret == 0) {
1697                                 list_move_tail(&device->dev_alloc_list,
1698                                                &private_devs);
1699                                 index++;
1700                                 if (type & BTRFS_BLOCK_GROUP_DUP)
1701                                         index++;
1702                         }
1703                 } else if (device->in_fs_metadata && avail > max_avail)
1704                         max_avail = avail;
1705                 if (cur == dev_list)
1706                         break;
1707         }
1708         if (index < num_stripes) {
1709                 list_splice(&private_devs, dev_list);
1710                 if (index >= min_stripes) {
1711                         num_stripes = index;
1712                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1713                                 num_stripes /= sub_stripes;
1714                                 num_stripes *= sub_stripes;
1715                         }
1716                         looped = 1;
1717                         goto again;
1718                 }
1719                 if (!looped && max_avail > 0) {
1720                         looped = 1;
1721                         calc_size = max_avail;
1722                         goto again;
1723                 }
1724                 btrfs_free_path(path);
1725                 return -ENOSPC;
1726         }
1727         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1728         key.type = BTRFS_CHUNK_ITEM_KEY;
1729         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1730                               &key.offset);
1731         if (ret) {
1732                 btrfs_free_path(path);
1733                 return ret;
1734         }
1735
1736         chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1737         if (!chunk) {
1738                 btrfs_free_path(path);
1739                 return -ENOMEM;
1740         }
1741
1742         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1743         if (!map) {
1744                 kfree(chunk);
1745                 btrfs_free_path(path);
1746                 return -ENOMEM;
1747         }
1748         btrfs_free_path(path);
1749         path = NULL;
1750
1751         stripes = &chunk->stripe;
1752         *num_bytes = chunk_bytes_by_type(type, calc_size,
1753                                          num_stripes, sub_stripes);
1754
1755         index = 0;
1756         while(index < num_stripes) {
1757                 struct btrfs_stripe *stripe;
1758                 BUG_ON(list_empty(&private_devs));
1759                 cur = private_devs.next;
1760                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1761
1762                 /* loop over this device again if we're doing a dup group */
1763                 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1764                     (index == num_stripes - 1))
1765                         list_move_tail(&device->dev_alloc_list, dev_list);
1766
1767                 ret = btrfs_alloc_dev_extent(trans, device,
1768                              info->chunk_root->root_key.objectid,
1769                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1770                              calc_size, &dev_offset);
1771                 BUG_ON(ret);
1772                 device->bytes_used += calc_size;
1773                 ret = btrfs_update_device(trans, device);
1774                 BUG_ON(ret);
1775
1776                 map->stripes[index].dev = device;
1777                 map->stripes[index].physical = dev_offset;
1778                 stripe = stripes + index;
1779                 btrfs_set_stack_stripe_devid(stripe, device->devid);
1780                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1781                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1782                 physical = dev_offset;
1783                 index++;
1784         }
1785         BUG_ON(!list_empty(&private_devs));
1786
1787         /* key was set above */
1788         btrfs_set_stack_chunk_length(chunk, *num_bytes);
1789         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1790         btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1791         btrfs_set_stack_chunk_type(chunk, type);
1792         btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1793         btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1794         btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1795         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1796         btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1797         map->sector_size = extent_root->sectorsize;
1798         map->stripe_len = stripe_len;
1799         map->io_align = stripe_len;
1800         map->io_width = stripe_len;
1801         map->type = type;
1802         map->num_stripes = num_stripes;
1803         map->sub_stripes = sub_stripes;
1804
1805         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1806                                 btrfs_chunk_item_size(num_stripes));
1807         BUG_ON(ret);
1808         *start = key.offset;;
1809
1810         em = alloc_extent_map(GFP_NOFS);
1811         if (!em)
1812                 return -ENOMEM;
1813         em->bdev = (struct block_device *)map;
1814         em->start = key.offset;
1815         em->len = *num_bytes;
1816         em->block_start = 0;
1817         em->block_len = em->len;
1818
1819         if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1820                 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1821                                     chunk, btrfs_chunk_item_size(num_stripes));
1822                 BUG_ON(ret);
1823         }
1824         kfree(chunk);
1825
1826         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1827         spin_lock(&em_tree->lock);
1828         ret = add_extent_mapping(em_tree, em);
1829         spin_unlock(&em_tree->lock);
1830         BUG_ON(ret);
1831         free_extent_map(em);
1832         return ret;
1833 }
1834
1835 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1836 {
1837         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1838 }
1839
1840 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1841 {
1842         struct extent_map *em;
1843
1844         while(1) {
1845                 spin_lock(&tree->map_tree.lock);
1846                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1847                 if (em)
1848                         remove_extent_mapping(&tree->map_tree, em);
1849                 spin_unlock(&tree->map_tree.lock);
1850                 if (!em)
1851                         break;
1852                 kfree(em->bdev);
1853                 /* once for us */
1854                 free_extent_map(em);
1855                 /* once for the tree */
1856                 free_extent_map(em);
1857         }
1858 }
1859
1860 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1861 {
1862         struct extent_map *em;
1863         struct map_lookup *map;
1864         struct extent_map_tree *em_tree = &map_tree->map_tree;
1865         int ret;
1866
1867         spin_lock(&em_tree->lock);
1868         em = lookup_extent_mapping(em_tree, logical, len);
1869         spin_unlock(&em_tree->lock);
1870         BUG_ON(!em);
1871
1872         BUG_ON(em->start > logical || em->start + em->len < logical);
1873         map = (struct map_lookup *)em->bdev;
1874         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1875                 ret = map->num_stripes;
1876         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1877                 ret = map->sub_stripes;
1878         else
1879                 ret = 1;
1880         free_extent_map(em);
1881         return ret;
1882 }
1883
1884 static int find_live_mirror(struct map_lookup *map, int first, int num,
1885                             int optimal)
1886 {
1887         int i;
1888         if (map->stripes[optimal].dev->bdev)
1889                 return optimal;
1890         for (i = first; i < first + num; i++) {
1891                 if (map->stripes[i].dev->bdev)
1892                         return i;
1893         }
1894         /* we couldn't find one that doesn't fail.  Just return something
1895          * and the io error handling code will clean up eventually
1896          */
1897         return optimal;
1898 }
1899
1900 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1901                              u64 logical, u64 *length,
1902                              struct btrfs_multi_bio **multi_ret,
1903                              int mirror_num, struct page *unplug_page)
1904 {
1905         struct extent_map *em;
1906         struct map_lookup *map;
1907         struct extent_map_tree *em_tree = &map_tree->map_tree;
1908         u64 offset;
1909         u64 stripe_offset;
1910         u64 stripe_nr;
1911         int stripes_allocated = 8;
1912         int stripes_required = 1;
1913         int stripe_index;
1914         int i;
1915         int num_stripes;
1916         int max_errors = 0;
1917         struct btrfs_multi_bio *multi = NULL;
1918
1919         if (multi_ret && !(rw & (1 << BIO_RW))) {
1920                 stripes_allocated = 1;
1921         }
1922 again:
1923         if (multi_ret) {
1924                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1925                                 GFP_NOFS);
1926                 if (!multi)
1927                         return -ENOMEM;
1928
1929                 atomic_set(&multi->error, 0);
1930         }
1931
1932         spin_lock(&em_tree->lock);
1933         em = lookup_extent_mapping(em_tree, logical, *length);
1934         spin_unlock(&em_tree->lock);
1935
1936         if (!em && unplug_page)
1937                 return 0;
1938
1939         if (!em) {
1940                 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1941                 BUG();
1942         }
1943
1944         BUG_ON(em->start > logical || em->start + em->len < logical);
1945         map = (struct map_lookup *)em->bdev;
1946         offset = logical - em->start;
1947
1948         if (mirror_num > map->num_stripes)
1949                 mirror_num = 0;
1950
1951         /* if our multi bio struct is too small, back off and try again */
1952         if (rw & (1 << BIO_RW)) {
1953                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1954                                  BTRFS_BLOCK_GROUP_DUP)) {
1955                         stripes_required = map->num_stripes;
1956                         max_errors = 1;
1957                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1958                         stripes_required = map->sub_stripes;
1959                         max_errors = 1;
1960                 }
1961         }
1962         if (multi_ret && rw == WRITE &&
1963             stripes_allocated < stripes_required) {
1964                 stripes_allocated = map->num_stripes;
1965                 free_extent_map(em);
1966                 kfree(multi);
1967                 goto again;
1968         }
1969         stripe_nr = offset;
1970         /*
1971          * stripe_nr counts the total number of stripes we have to stride
1972          * to get to this block
1973          */
1974         do_div(stripe_nr, map->stripe_len);
1975
1976         stripe_offset = stripe_nr * map->stripe_len;
1977         BUG_ON(offset < stripe_offset);
1978
1979         /* stripe_offset is the offset of this block in its stripe*/
1980         stripe_offset = offset - stripe_offset;
1981
1982         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1983                          BTRFS_BLOCK_GROUP_RAID10 |
1984                          BTRFS_BLOCK_GROUP_DUP)) {
1985                 /* we limit the length of each bio to what fits in a stripe */
1986                 *length = min_t(u64, em->len - offset,
1987                               map->stripe_len - stripe_offset);
1988         } else {
1989                 *length = em->len - offset;
1990         }
1991
1992         if (!multi_ret && !unplug_page)
1993                 goto out;
1994
1995         num_stripes = 1;
1996         stripe_index = 0;
1997         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1998                 if (unplug_page || (rw & (1 << BIO_RW)))
1999                         num_stripes = map->num_stripes;
2000                 else if (mirror_num)
2001                         stripe_index = mirror_num - 1;
2002                 else {
2003                         stripe_index = find_live_mirror(map, 0,
2004                                             map->num_stripes,
2005                                             current->pid % map->num_stripes);
2006                 }
2007
2008         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2009                 if (rw & (1 << BIO_RW))
2010                         num_stripes = map->num_stripes;
2011                 else if (mirror_num)
2012                         stripe_index = mirror_num - 1;
2013
2014         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2015                 int factor = map->num_stripes / map->sub_stripes;
2016
2017                 stripe_index = do_div(stripe_nr, factor);
2018                 stripe_index *= map->sub_stripes;
2019
2020                 if (unplug_page || (rw & (1 << BIO_RW)))
2021                         num_stripes = map->sub_stripes;
2022                 else if (mirror_num)
2023                         stripe_index += mirror_num - 1;
2024                 else {
2025                         stripe_index = find_live_mirror(map, stripe_index,
2026                                               map->sub_stripes, stripe_index +
2027                                               current->pid % map->sub_stripes);
2028                 }
2029         } else {
2030                 /*
2031                  * after this do_div call, stripe_nr is the number of stripes
2032                  * on this device we have to walk to find the data, and
2033                  * stripe_index is the number of our device in the stripe array
2034                  */
2035                 stripe_index = do_div(stripe_nr, map->num_stripes);
2036         }
2037         BUG_ON(stripe_index >= map->num_stripes);
2038
2039         for (i = 0; i < num_stripes; i++) {
2040                 if (unplug_page) {
2041                         struct btrfs_device *device;
2042                         struct backing_dev_info *bdi;
2043
2044                         device = map->stripes[stripe_index].dev;
2045                         if (device->bdev) {
2046                                 bdi = blk_get_backing_dev_info(device->bdev);
2047                                 if (bdi->unplug_io_fn) {
2048                                         bdi->unplug_io_fn(bdi, unplug_page);
2049                                 }
2050                         }
2051                 } else {
2052                         multi->stripes[i].physical =
2053                                 map->stripes[stripe_index].physical +
2054                                 stripe_offset + stripe_nr * map->stripe_len;
2055                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2056                 }
2057                 stripe_index++;
2058         }
2059         if (multi_ret) {
2060                 *multi_ret = multi;
2061                 multi->num_stripes = num_stripes;
2062                 multi->max_errors = max_errors;
2063         }
2064 out:
2065         free_extent_map(em);
2066         return 0;
2067 }
2068
2069 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2070                       u64 logical, u64 *length,
2071                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2072 {
2073         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2074                                  mirror_num, NULL);
2075 }
2076
2077 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2078                       u64 logical, struct page *page)
2079 {
2080         u64 length = PAGE_CACHE_SIZE;
2081         return __btrfs_map_block(map_tree, READ, logical, &length,
2082                                  NULL, 0, page);
2083 }
2084
2085
2086 static void end_bio_multi_stripe(struct bio *bio, int err)
2087 {
2088         struct btrfs_multi_bio *multi = bio->bi_private;
2089         int is_orig_bio = 0;
2090
2091         if (err)
2092                 atomic_inc(&multi->error);
2093
2094         if (bio == multi->orig_bio)
2095                 is_orig_bio = 1;
2096
2097         if (atomic_dec_and_test(&multi->stripes_pending)) {
2098                 if (!is_orig_bio) {
2099                         bio_put(bio);
2100                         bio = multi->orig_bio;
2101                 }
2102                 bio->bi_private = multi->private;
2103                 bio->bi_end_io = multi->end_io;
2104                 /* only send an error to the higher layers if it is
2105                  * beyond the tolerance of the multi-bio
2106                  */
2107                 if (atomic_read(&multi->error) > multi->max_errors) {
2108                         err = -EIO;
2109                 } else if (err) {
2110                         /*
2111                          * this bio is actually up to date, we didn't
2112                          * go over the max number of errors
2113                          */
2114                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2115                         err = 0;
2116                 }
2117                 kfree(multi);
2118
2119                 bio_endio(bio, err);
2120         } else if (!is_orig_bio) {
2121                 bio_put(bio);
2122         }
2123 }
2124
2125 struct async_sched {
2126         struct bio *bio;
2127         int rw;
2128         struct btrfs_fs_info *info;
2129         struct btrfs_work work;
2130 };
2131
2132 /*
2133  * see run_scheduled_bios for a description of why bios are collected for
2134  * async submit.
2135  *
2136  * This will add one bio to the pending list for a device and make sure
2137  * the work struct is scheduled.
2138  */
2139 static int noinline schedule_bio(struct btrfs_root *root,
2140                                  struct btrfs_device *device,
2141                                  int rw, struct bio *bio)
2142 {
2143         int should_queue = 1;
2144
2145         /* don't bother with additional async steps for reads, right now */
2146         if (!(rw & (1 << BIO_RW))) {
2147                 bio_get(bio);
2148                 submit_bio(rw, bio);
2149                 bio_put(bio);
2150                 return 0;
2151         }
2152
2153         /*
2154          * nr_async_bios allows us to reliably return congestion to the
2155          * higher layers.  Otherwise, the async bio makes it appear we have
2156          * made progress against dirty pages when we've really just put it
2157          * on a queue for later
2158          */
2159         atomic_inc(&root->fs_info->nr_async_bios);
2160         WARN_ON(bio->bi_next);
2161         bio->bi_next = NULL;
2162         bio->bi_rw |= rw;
2163
2164         spin_lock(&device->io_lock);
2165
2166         if (device->pending_bio_tail)
2167                 device->pending_bio_tail->bi_next = bio;
2168
2169         device->pending_bio_tail = bio;
2170         if (!device->pending_bios)
2171                 device->pending_bios = bio;
2172         if (device->running_pending)
2173                 should_queue = 0;
2174
2175         spin_unlock(&device->io_lock);
2176
2177         if (should_queue)
2178                 btrfs_queue_worker(&root->fs_info->submit_workers,
2179                                    &device->work);
2180         return 0;
2181 }
2182
2183 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2184                   int mirror_num, int async_submit)
2185 {
2186         struct btrfs_mapping_tree *map_tree;
2187         struct btrfs_device *dev;
2188         struct bio *first_bio = bio;
2189         u64 logical = (u64)bio->bi_sector << 9;
2190         u64 length = 0;
2191         u64 map_length;
2192         struct btrfs_multi_bio *multi = NULL;
2193         int ret;
2194         int dev_nr = 0;
2195         int total_devs = 1;
2196
2197         length = bio->bi_size;
2198         map_tree = &root->fs_info->mapping_tree;
2199         map_length = length;
2200
2201         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2202                               mirror_num);
2203         BUG_ON(ret);
2204
2205         total_devs = multi->num_stripes;
2206         if (map_length < length) {
2207                 printk("mapping failed logical %Lu bio len %Lu "
2208                        "len %Lu\n", logical, length, map_length);
2209                 BUG();
2210         }
2211         multi->end_io = first_bio->bi_end_io;
2212         multi->private = first_bio->bi_private;
2213         multi->orig_bio = first_bio;
2214         atomic_set(&multi->stripes_pending, multi->num_stripes);
2215
2216         while(dev_nr < total_devs) {
2217                 if (total_devs > 1) {
2218                         if (dev_nr < total_devs - 1) {
2219                                 bio = bio_clone(first_bio, GFP_NOFS);
2220                                 BUG_ON(!bio);
2221                         } else {
2222                                 bio = first_bio;
2223                         }
2224                         bio->bi_private = multi;
2225                         bio->bi_end_io = end_bio_multi_stripe;
2226                 }
2227                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2228                 dev = multi->stripes[dev_nr].dev;
2229                 if (dev && dev->bdev) {
2230                         bio->bi_bdev = dev->bdev;
2231                         if (async_submit)
2232                                 schedule_bio(root, dev, rw, bio);
2233                         else
2234                                 submit_bio(rw, bio);
2235                 } else {
2236                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2237                         bio->bi_sector = logical >> 9;
2238                         bio_endio(bio, -EIO);
2239                 }
2240                 dev_nr++;
2241         }
2242         if (total_devs == 1)
2243                 kfree(multi);
2244         return 0;
2245 }
2246
2247 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2248                                        u8 *uuid)
2249 {
2250         struct list_head *head = &root->fs_info->fs_devices->devices;
2251
2252         return __find_device(head, devid, uuid);
2253 }
2254
2255 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2256                                             u64 devid, u8 *dev_uuid)
2257 {
2258         struct btrfs_device *device;
2259         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2260
2261         device = kzalloc(sizeof(*device), GFP_NOFS);
2262         list_add(&device->dev_list,
2263                  &fs_devices->devices);
2264         list_add(&device->dev_alloc_list,
2265                  &fs_devices->alloc_list);
2266         device->barriers = 1;
2267         device->dev_root = root->fs_info->dev_root;
2268         device->devid = devid;
2269         device->work.func = pending_bios_fn;
2270         fs_devices->num_devices++;
2271         spin_lock_init(&device->io_lock);
2272         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2273         return device;
2274 }
2275
2276
2277 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2278                           struct extent_buffer *leaf,
2279                           struct btrfs_chunk *chunk)
2280 {
2281         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2282         struct map_lookup *map;
2283         struct extent_map *em;
2284         u64 logical;
2285         u64 length;
2286         u64 devid;
2287         u8 uuid[BTRFS_UUID_SIZE];
2288         int num_stripes;
2289         int ret;
2290         int i;
2291
2292         logical = key->offset;
2293         length = btrfs_chunk_length(leaf, chunk);
2294
2295         spin_lock(&map_tree->map_tree.lock);
2296         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2297         spin_unlock(&map_tree->map_tree.lock);
2298
2299         /* already mapped? */
2300         if (em && em->start <= logical && em->start + em->len > logical) {
2301                 free_extent_map(em);
2302                 return 0;
2303         } else if (em) {
2304                 free_extent_map(em);
2305         }
2306
2307         map = kzalloc(sizeof(*map), GFP_NOFS);
2308         if (!map)
2309                 return -ENOMEM;
2310
2311         em = alloc_extent_map(GFP_NOFS);
2312         if (!em)
2313                 return -ENOMEM;
2314         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2315         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2316         if (!map) {
2317                 free_extent_map(em);
2318                 return -ENOMEM;
2319         }
2320
2321         em->bdev = (struct block_device *)map;
2322         em->start = logical;
2323         em->len = length;
2324         em->block_start = 0;
2325         em->block_len = em->len;
2326
2327         map->num_stripes = num_stripes;
2328         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2329         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2330         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2331         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2332         map->type = btrfs_chunk_type(leaf, chunk);
2333         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2334         for (i = 0; i < num_stripes; i++) {
2335                 map->stripes[i].physical =
2336                         btrfs_stripe_offset_nr(leaf, chunk, i);
2337                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2338                 read_extent_buffer(leaf, uuid, (unsigned long)
2339                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2340                                    BTRFS_UUID_SIZE);
2341                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2342
2343                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2344                         kfree(map);
2345                         free_extent_map(em);
2346                         return -EIO;
2347                 }
2348                 if (!map->stripes[i].dev) {
2349                         map->stripes[i].dev =
2350                                 add_missing_dev(root, devid, uuid);
2351                         if (!map->stripes[i].dev) {
2352                                 kfree(map);
2353                                 free_extent_map(em);
2354                                 return -EIO;
2355                         }
2356                 }
2357                 map->stripes[i].dev->in_fs_metadata = 1;
2358         }
2359
2360         spin_lock(&map_tree->map_tree.lock);
2361         ret = add_extent_mapping(&map_tree->map_tree, em);
2362         spin_unlock(&map_tree->map_tree.lock);
2363         BUG_ON(ret);
2364         free_extent_map(em);
2365
2366         return 0;
2367 }
2368
2369 static int fill_device_from_item(struct extent_buffer *leaf,
2370                                  struct btrfs_dev_item *dev_item,
2371                                  struct btrfs_device *device)
2372 {
2373         unsigned long ptr;
2374
2375         device->devid = btrfs_device_id(leaf, dev_item);
2376         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2377         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2378         device->type = btrfs_device_type(leaf, dev_item);
2379         device->io_align = btrfs_device_io_align(leaf, dev_item);
2380         device->io_width = btrfs_device_io_width(leaf, dev_item);
2381         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2382
2383         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2384         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2385
2386         return 0;
2387 }
2388
2389 static int read_one_dev(struct btrfs_root *root,
2390                         struct extent_buffer *leaf,
2391                         struct btrfs_dev_item *dev_item)
2392 {
2393         struct btrfs_device *device;
2394         u64 devid;
2395         int ret;
2396         u8 dev_uuid[BTRFS_UUID_SIZE];
2397
2398         devid = btrfs_device_id(leaf, dev_item);
2399         read_extent_buffer(leaf, dev_uuid,
2400                            (unsigned long)btrfs_device_uuid(dev_item),
2401                            BTRFS_UUID_SIZE);
2402         device = btrfs_find_device(root, devid, dev_uuid);
2403         if (!device) {
2404                 printk("warning devid %Lu missing\n", devid);
2405                 device = add_missing_dev(root, devid, dev_uuid);
2406                 if (!device)
2407                         return -ENOMEM;
2408         }
2409
2410         fill_device_from_item(leaf, dev_item, device);
2411         device->dev_root = root->fs_info->dev_root;
2412         device->in_fs_metadata = 1;
2413         ret = 0;
2414 #if 0
2415         ret = btrfs_open_device(device);
2416         if (ret) {
2417                 kfree(device);
2418         }
2419 #endif
2420         return ret;
2421 }
2422
2423 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2424 {
2425         struct btrfs_dev_item *dev_item;
2426
2427         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2428                                                      dev_item);
2429         return read_one_dev(root, buf, dev_item);
2430 }
2431
2432 int btrfs_read_sys_array(struct btrfs_root *root)
2433 {
2434         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2435         struct extent_buffer *sb;
2436         struct btrfs_disk_key *disk_key;
2437         struct btrfs_chunk *chunk;
2438         u8 *ptr;
2439         unsigned long sb_ptr;
2440         int ret = 0;
2441         u32 num_stripes;
2442         u32 array_size;
2443         u32 len = 0;
2444         u32 cur;
2445         struct btrfs_key key;
2446
2447         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2448                                           BTRFS_SUPER_INFO_SIZE);
2449         if (!sb)
2450                 return -ENOMEM;
2451         btrfs_set_buffer_uptodate(sb);
2452         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2453         array_size = btrfs_super_sys_array_size(super_copy);
2454
2455         ptr = super_copy->sys_chunk_array;
2456         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2457         cur = 0;
2458
2459         while (cur < array_size) {
2460                 disk_key = (struct btrfs_disk_key *)ptr;
2461                 btrfs_disk_key_to_cpu(&key, disk_key);
2462
2463                 len = sizeof(*disk_key); ptr += len;
2464                 sb_ptr += len;
2465                 cur += len;
2466
2467                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2468                         chunk = (struct btrfs_chunk *)sb_ptr;
2469                         ret = read_one_chunk(root, &key, sb, chunk);
2470                         if (ret)
2471                                 break;
2472                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2473                         len = btrfs_chunk_item_size(num_stripes);
2474                 } else {
2475                         ret = -EIO;
2476                         break;
2477                 }
2478                 ptr += len;
2479                 sb_ptr += len;
2480                 cur += len;
2481         }
2482         free_extent_buffer(sb);
2483         return ret;
2484 }
2485
2486 int btrfs_read_chunk_tree(struct btrfs_root *root)
2487 {
2488         struct btrfs_path *path;
2489         struct extent_buffer *leaf;
2490         struct btrfs_key key;
2491         struct btrfs_key found_key;
2492         int ret;
2493         int slot;
2494
2495         root = root->fs_info->chunk_root;
2496
2497         path = btrfs_alloc_path();
2498         if (!path)
2499                 return -ENOMEM;
2500
2501         /* first we search for all of the device items, and then we
2502          * read in all of the chunk items.  This way we can create chunk
2503          * mappings that reference all of the devices that are afound
2504          */
2505         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2506         key.offset = 0;
2507         key.type = 0;
2508 again:
2509         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2510         while(1) {
2511                 leaf = path->nodes[0];
2512                 slot = path->slots[0];
2513                 if (slot >= btrfs_header_nritems(leaf)) {
2514                         ret = btrfs_next_leaf(root, path);
2515                         if (ret == 0)
2516                                 continue;
2517                         if (ret < 0)
2518                                 goto error;
2519                         break;
2520                 }
2521                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2522                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2523                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2524                                 break;
2525                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2526                                 struct btrfs_dev_item *dev_item;
2527                                 dev_item = btrfs_item_ptr(leaf, slot,
2528                                                   struct btrfs_dev_item);
2529                                 ret = read_one_dev(root, leaf, dev_item);
2530                                 BUG_ON(ret);
2531                         }
2532                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2533                         struct btrfs_chunk *chunk;
2534                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2535                         ret = read_one_chunk(root, &found_key, leaf, chunk);
2536                 }
2537                 path->slots[0]++;
2538         }
2539         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2540                 key.objectid = 0;
2541                 btrfs_release_path(root, path);
2542                 goto again;
2543         }
2544
2545         btrfs_free_path(path);
2546         ret = 0;
2547 error:
2548         return ret;
2549 }