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