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/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <asm/div64.h>
  27 #include "compat.h"
  28 #include "ctree.h"
  29 #include "extent_map.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32 #include "print-tree.h"
  33 #include "volumes.h"
  34 #include "async-thread.h"
  35
  36 struct map_lookup {
  37         u64 type;
  38         int io_align;
  39         int io_width;
  40         int stripe_len;
  41         int sector_size;
  42         int num_stripes;
  43         int sub_stripes;
  44         struct btrfs_bio_stripe stripes[];
  45 };
  46
  47 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  48                                 struct btrfs_root *root,
  49                                 struct btrfs_device *device);
  50 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  51
  52 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  53                             (sizeof(struct btrfs_bio_stripe) * (n)))
  54
  55 static DEFINE_MUTEX(uuid_mutex);
  56 static LIST_HEAD(fs_uuids);
  57
  58 void btrfs_lock_volumes(void)
  59 {
  60         mutex_lock(&uuid_mutex);
  61 }
  62
  63 void btrfs_unlock_volumes(void)
  64 {
  65         mutex_unlock(&uuid_mutex);
  66 }
  67
  68 static void lock_chunks(struct btrfs_root *root)
  69 {
  70         mutex_lock(&root->fs_info->chunk_mutex);
  71 }
  72
  73 static void unlock_chunks(struct btrfs_root *root)
  74 {
  75         mutex_unlock(&root->fs_info->chunk_mutex);
  76 }
  77
  78 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  79 {
  80         struct btrfs_device *device;
  81         WARN_ON(fs_devices->opened);
  82         while (!list_empty(&fs_devices->devices)) {
  83                 device = list_entry(fs_devices->devices.next,
  84                                     struct btrfs_device, dev_list);
  85                 list_del(&device->dev_list);
  86                 kfree(device->name);
  87                 kfree(device);
  88         }
  89         kfree(fs_devices);
  90 }
  91
  92 int btrfs_cleanup_fs_uuids(void)
  93 {
  94         struct btrfs_fs_devices *fs_devices;
  95
  96         while (!list_empty(&fs_uuids)) {
  97                 fs_devices = list_entry(fs_uuids.next,
  98                                         struct btrfs_fs_devices, list);
  99                 list_del(&fs_devices->list);
 100                 free_fs_devices(fs_devices);
 101         }
 102         return 0;
 103 }
 104
 105 static noinline struct btrfs_device *__find_device(struct list_head *head,
 106                                                    u64 devid, u8 *uuid)
 107 {
 108         struct btrfs_device *dev;
 109
 110         list_for_each_entry(dev, head, dev_list) {
 111                 if (dev->devid == devid &&
 112                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 113                         return dev;
 114                 }
 115         }
 116         return NULL;
 117 }
 118
 119 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 120 {
 121         struct btrfs_fs_devices *fs_devices;
 122
 123         list_for_each_entry(fs_devices, &fs_uuids, list) {
 124                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 125                         return fs_devices;
 126         }
 127         return NULL;
 128 }
 129
 130 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 131                         struct bio *head, struct bio *tail)
 132 {
 133
 134         struct bio *old_head;
 135
 136         old_head = pending_bios->head;
 137         pending_bios->head = head;
 138         if (pending_bios->tail)
 139                 tail->bi_next = old_head;
 140         else
 141                 pending_bios->tail = tail;
 142 }
 143
 144 /*
 145  * we try to collect pending bios for a device so we don't get a large
 146  * number of procs sending bios down to the same device.  This greatly
 147  * improves the schedulers ability to collect and merge the bios.
 148  *
 149  * But, it also turns into a long list of bios to process and that is sure
 150  * to eventually make the worker thread block.  The solution here is to
 151  * make some progress and then put this work struct back at the end of
 152  * the list if the block device is congested.  This way, multiple devices
 153  * can make progress from a single worker thread.
 154  */
 155 static noinline int run_scheduled_bios(struct btrfs_device *device)
 156 {
 157         struct bio *pending;
 158         struct backing_dev_info *bdi;
 159         struct btrfs_fs_info *fs_info;
 160         struct btrfs_pending_bios *pending_bios;
 161         struct bio *tail;
 162         struct bio *cur;
 163         int again = 0;
 164         unsigned long num_run;
 165         unsigned long batch_run = 0;
 166         unsigned long limit;
 167         unsigned long last_waited = 0;
 168         int force_reg = 0;
 169
 170         bdi = blk_get_backing_dev_info(device->bdev);
 171         fs_info = device->dev_root->fs_info;
 172         limit = btrfs_async_submit_limit(fs_info);
 173         limit = limit * 2 / 3;
 174
 175 loop:
 176         spin_lock(&device->io_lock);
 177
 178 loop_lock:
 179         num_run = 0;
 180
 181         /* take all the bios off the list at once and process them
 182          * later on (without the lock held).  But, remember the
 183          * tail and other pointers so the bios can be properly reinserted
 184          * into the list if we hit congestion
 185          */
 186         if (!force_reg && device->pending_sync_bios.head) {
 187                 pending_bios = &device->pending_sync_bios;
 188                 force_reg = 1;
 189         } else {
 190                 pending_bios = &device->pending_bios;
 191                 force_reg = 0;
 192         }
 193
 194         pending = pending_bios->head;
 195         tail = pending_bios->tail;
 196         WARN_ON(pending && !tail);
 197
 198         /*
 199          * if pending was null this time around, no bios need processing
 200          * at all and we can stop.  Otherwise it'll loop back up again
 201          * and do an additional check so no bios are missed.
 202          *
 203          * device->running_pending is used to synchronize with the
 204          * schedule_bio code.
 205          */
 206         if (device->pending_sync_bios.head == NULL &&
 207             device->pending_bios.head == NULL) {
 208                 again = 0;
 209                 device->running_pending = 0;
 210         } else {
 211                 again = 1;
 212                 device->running_pending = 1;
 213         }
 214
 215         pending_bios->head = NULL;
 216         pending_bios->tail = NULL;
 217
 218         spin_unlock(&device->io_lock);
 219
 220         while (pending) {
 221
 222                 rmb();
 223                 /* we want to work on both lists, but do more bios on the
 224                  * sync list than the regular list
 225                  */
 226                 if ((num_run > 32 &&
 227                     pending_bios != &device->pending_sync_bios &&
 228                     device->pending_sync_bios.head) ||
 229                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 230                     device->pending_bios.head)) {
 231                         spin_lock(&device->io_lock);
 232                         requeue_list(pending_bios, pending, tail);
 233                         goto loop_lock;
 234                 }
 235
 236                 cur = pending;
 237                 pending = pending->bi_next;
 238                 cur->bi_next = NULL;
 239                 atomic_dec(&fs_info->nr_async_bios);
 240
 241                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 242                     waitqueue_active(&fs_info->async_submit_wait))
 243                         wake_up(&fs_info->async_submit_wait);
 244
 245                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 246
 247                 submit_bio(cur->bi_rw, cur);
 248                 num_run++;
 249                 batch_run++;
 250                 if (need_resched())
 251                         cond_resched();
 252
 253                 /*
 254                  * we made progress, there is more work to do and the bdi
 255                  * is now congested.  Back off and let other work structs
 256                  * run instead
 257                  */
 258                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 259                     fs_info->fs_devices->open_devices > 1) {
 260                         struct io_context *ioc;
 261
 262                         ioc = current->io_context;
 263
 264                         /*
 265                          * the main goal here is that we don't want to
 266                          * block if we're going to be able to submit
 267                          * more requests without blocking.
 268                          *
 269                          * This code does two great things, it pokes into
 270                          * the elevator code from a filesystem _and_
 271                          * it makes assumptions about how batching works.
 272                          */
 273                         if (ioc && ioc->nr_batch_requests > 0 &&
 274                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 275                             (last_waited == 0 ||
 276                              ioc->last_waited == last_waited)) {
 277                                 /*
 278                                  * we want to go through our batch of
 279                                  * requests and stop.  So, we copy out
 280                                  * the ioc->last_waited time and test
 281                                  * against it before looping
 282                                  */
 283                                 last_waited = ioc->last_waited;
 284                                 if (need_resched())
 285                                         cond_resched();
 286                                 continue;
 287                         }
 288                         spin_lock(&device->io_lock);
 289                         requeue_list(pending_bios, pending, tail);
 290                         device->running_pending = 1;
 291
 292                         spin_unlock(&device->io_lock);
 293                         btrfs_requeue_work(&device->work);
 294                         goto done;
 295                 }
 296         }
 297
 298         cond_resched();
 299         if (again)
 300                 goto loop;
 301
 302         spin_lock(&device->io_lock);
 303         if (device->pending_bios.head || device->pending_sync_bios.head)
 304                 goto loop_lock;
 305         spin_unlock(&device->io_lock);
 306
 307 done:
 308         return 0;
 309 }
 310
 311 static void pending_bios_fn(struct btrfs_work *work)
 312 {
 313         struct btrfs_device *device;
 314
 315         device = container_of(work, struct btrfs_device, work);
 316         run_scheduled_bios(device);
 317 }
 318
 319 static noinline int device_list_add(const char *path,
 320                            struct btrfs_super_block *disk_super,
 321                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 322 {
 323         struct btrfs_device *device;
 324         struct btrfs_fs_devices *fs_devices;
 325         u64 found_transid = btrfs_super_generation(disk_super);
 326         char *name;
 327
 328         fs_devices = find_fsid(disk_super->fsid);
 329         if (!fs_devices) {
 330                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 331                 if (!fs_devices)
 332                         return -ENOMEM;
 333                 INIT_LIST_HEAD(&fs_devices->devices);
 334                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 335                 list_add(&fs_devices->list, &fs_uuids);
 336                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 337                 fs_devices->latest_devid = devid;
 338                 fs_devices->latest_trans = found_transid;
 339                 mutex_init(&fs_devices->device_list_mutex);
 340                 device = NULL;
 341         } else {
 342                 device = __find_device(&fs_devices->devices, devid,
 343                                        disk_super->dev_item.uuid);
 344         }
 345         if (!device) {
 346                 if (fs_devices->opened)
 347                         return -EBUSY;
 348
 349                 device = kzalloc(sizeof(*device), GFP_NOFS);
 350                 if (!device) {
 351                         /* we can safely leave the fs_devices entry around */
 352                         return -ENOMEM;
 353                 }
 354                 device->devid = devid;
 355                 device->work.func = pending_bios_fn;
 356                 memcpy(device->uuid, disk_super->dev_item.uuid,
 357                        BTRFS_UUID_SIZE);
 358                 spin_lock_init(&device->io_lock);
 359                 device->name = kstrdup(path, GFP_NOFS);
 360                 if (!device->name) {
 361                         kfree(device);
 362                         return -ENOMEM;
 363                 }
 364                 INIT_LIST_HEAD(&device->dev_alloc_list);
 365
 366                 mutex_lock(&fs_devices->device_list_mutex);
 367                 list_add(&device->dev_list, &fs_devices->devices);
 368                 mutex_unlock(&fs_devices->device_list_mutex);
 369
 370                 device->fs_devices = fs_devices;
 371                 fs_devices->num_devices++;
 372         } else if (!device->name || strcmp(device->name, path)) {
 373                 name = kstrdup(path, GFP_NOFS);
 374                 if (!name)
 375                         return -ENOMEM;
 376                 kfree(device->name);
 377                 device->name = name;
 378                 if (device->missing) {
 379                         fs_devices->missing_devices--;
 380                         device->missing = 0;
 381                 }
 382         }
 383
 384         if (found_transid > fs_devices->latest_trans) {
 385                 fs_devices->latest_devid = devid;
 386                 fs_devices->latest_trans = found_transid;
 387         }
 388         *fs_devices_ret = fs_devices;
 389         return 0;
 390 }
 391
 392 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 393 {
 394         struct btrfs_fs_devices *fs_devices;
 395         struct btrfs_device *device;
 396         struct btrfs_device *orig_dev;
 397
 398         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 399         if (!fs_devices)
 400                 return ERR_PTR(-ENOMEM);
 401
 402         INIT_LIST_HEAD(&fs_devices->devices);
 403         INIT_LIST_HEAD(&fs_devices->alloc_list);
 404         INIT_LIST_HEAD(&fs_devices->list);
 405         mutex_init(&fs_devices->device_list_mutex);
 406         fs_devices->latest_devid = orig->latest_devid;
 407         fs_devices->latest_trans = orig->latest_trans;
 408         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 409
 410         mutex_lock(&orig->device_list_mutex);
 411         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 412                 device = kzalloc(sizeof(*device), GFP_NOFS);
 413                 if (!device)
 414                         goto error;
 415
 416                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 417                 if (!device->name) {
 418                         kfree(device);
 419                         goto error;
 420                 }
 421
 422                 device->devid = orig_dev->devid;
 423                 device->work.func = pending_bios_fn;
 424                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 425                 spin_lock_init(&device->io_lock);
 426                 INIT_LIST_HEAD(&device->dev_list);
 427                 INIT_LIST_HEAD(&device->dev_alloc_list);
 428
 429                 list_add(&device->dev_list, &fs_devices->devices);
 430                 device->fs_devices = fs_devices;
 431                 fs_devices->num_devices++;
 432         }
 433         mutex_unlock(&orig->device_list_mutex);
 434         return fs_devices;
 435 error:
 436         mutex_unlock(&orig->device_list_mutex);
 437         free_fs_devices(fs_devices);
 438         return ERR_PTR(-ENOMEM);
 439 }
 440
 441 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 442 {
 443         struct btrfs_device *device, *next;
 444
 445         mutex_lock(&uuid_mutex);
 446 again:
 447         mutex_lock(&fs_devices->device_list_mutex);
 448         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 449                 if (device->in_fs_metadata)
 450                         continue;
 451
 452                 if (device->bdev) {
 453                         blkdev_put(device->bdev, device->mode);
 454                         device->bdev = NULL;
 455                         fs_devices->open_devices--;
 456                 }
 457                 if (device->writeable) {
 458                         list_del_init(&device->dev_alloc_list);
 459                         device->writeable = 0;
 460                         fs_devices->rw_devices--;
 461                 }
 462                 list_del_init(&device->dev_list);
 463                 fs_devices->num_devices--;
 464                 kfree(device->name);
 465                 kfree(device);
 466         }
 467         mutex_unlock(&fs_devices->device_list_mutex);
 468
 469         if (fs_devices->seed) {
 470                 fs_devices = fs_devices->seed;
 471                 goto again;
 472         }
 473
 474         mutex_unlock(&uuid_mutex);
 475         return 0;
 476 }
 477
 478 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 479 {
 480         struct btrfs_device *device;
 481
 482         if (--fs_devices->opened > 0)
 483                 return 0;
 484
 485         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 486                 if (device->bdev) {
 487                         blkdev_put(device->bdev, device->mode);
 488                         fs_devices->open_devices--;
 489                 }
 490                 if (device->writeable) {
 491                         list_del_init(&device->dev_alloc_list);
 492                         fs_devices->rw_devices--;
 493                 }
 494
 495                 device->bdev = NULL;
 496                 device->writeable = 0;
 497                 device->in_fs_metadata = 0;
 498         }
 499         WARN_ON(fs_devices->open_devices);
 500         WARN_ON(fs_devices->rw_devices);
 501         fs_devices->opened = 0;
 502         fs_devices->seeding = 0;
 503
 504         return 0;
 505 }
 506
 507 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 508 {
 509         struct btrfs_fs_devices *seed_devices = NULL;
 510         int ret;
 511
 512         mutex_lock(&uuid_mutex);
 513         ret = __btrfs_close_devices(fs_devices);
 514         if (!fs_devices->opened) {
 515                 seed_devices = fs_devices->seed;
 516                 fs_devices->seed = NULL;
 517         }
 518         mutex_unlock(&uuid_mutex);
 519
 520         while (seed_devices) {
 521                 fs_devices = seed_devices;
 522                 seed_devices = fs_devices->seed;
 523                 __btrfs_close_devices(fs_devices);
 524                 free_fs_devices(fs_devices);
 525         }
 526         return ret;
 527 }
 528
 529 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 530                                 fmode_t flags, void *holder)
 531 {
 532         struct block_device *bdev;
 533         struct list_head *head = &fs_devices->devices;
 534         struct btrfs_device *device;
 535         struct block_device *latest_bdev = NULL;
 536         struct buffer_head *bh;
 537         struct btrfs_super_block *disk_super;
 538         u64 latest_devid = 0;
 539         u64 latest_transid = 0;
 540         u64 devid;
 541         int seeding = 1;
 542         int ret = 0;
 543
 544         flags |= FMODE_EXCL;
 545
 546         list_for_each_entry(device, head, dev_list) {
 547                 if (device->bdev)
 548                         continue;
 549                 if (!device->name)
 550                         continue;
 551
 552                 bdev = blkdev_get_by_path(device->name, flags, holder);
 553                 if (IS_ERR(bdev)) {
 554                         printk(KERN_INFO "open %s failed\n", device->name);
 555                         goto error;
 556                 }
 557                 set_blocksize(bdev, 4096);
 558
 559                 bh = btrfs_read_dev_super(bdev);
 560                 if (!bh) {
 561                         ret = -EINVAL;
 562                         goto error_close;
 563                 }
 564
 565                 disk_super = (struct btrfs_super_block *)bh->b_data;
 566                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 567                 if (devid != device->devid)
 568                         goto error_brelse;
 569
 570                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 571                            BTRFS_UUID_SIZE))
 572                         goto error_brelse;
 573
 574                 device->generation = btrfs_super_generation(disk_super);
 575                 if (!latest_transid || device->generation > latest_transid) {
 576                         latest_devid = devid;
 577                         latest_transid = device->generation;
 578                         latest_bdev = bdev;
 579                 }
 580
 581                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 582                         device->writeable = 0;
 583                 } else {
 584                         device->writeable = !bdev_read_only(bdev);
 585                         seeding = 0;
 586                 }
 587
 588                 device->bdev = bdev;
 589                 device->in_fs_metadata = 0;
 590                 device->mode = flags;
 591
 592                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 593                         fs_devices->rotating = 1;
 594
 595                 fs_devices->open_devices++;
 596                 if (device->writeable) {
 597                         fs_devices->rw_devices++;
 598                         list_add(&device->dev_alloc_list,
 599                                  &fs_devices->alloc_list);
 600                 }
 601                 continue;
 602
 603 error_brelse:
 604                 brelse(bh);
 605 error_close:
 606                 blkdev_put(bdev, flags);
 607 error:
 608                 continue;
 609         }
 610         if (fs_devices->open_devices == 0) {
 611                 ret = -EIO;
 612                 goto out;
 613         }
 614         fs_devices->seeding = seeding;
 615         fs_devices->opened = 1;
 616         fs_devices->latest_bdev = latest_bdev;
 617         fs_devices->latest_devid = latest_devid;
 618         fs_devices->latest_trans = latest_transid;
 619         fs_devices->total_rw_bytes = 0;
 620 out:
 621         return ret;
 622 }
 623
 624 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 625                        fmode_t flags, void *holder)
 626 {
 627         int ret;
 628
 629         mutex_lock(&uuid_mutex);
 630         if (fs_devices->opened) {
 631                 fs_devices->opened++;
 632                 ret = 0;
 633         } else {
 634                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 635         }
 636         mutex_unlock(&uuid_mutex);
 637         return ret;
 638 }
 639
 640 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 641                           struct btrfs_fs_devices **fs_devices_ret)
 642 {
 643         struct btrfs_super_block *disk_super;
 644         struct block_device *bdev;
 645         struct buffer_head *bh;
 646         int ret;
 647         u64 devid;
 648         u64 transid;
 649
 650         mutex_lock(&uuid_mutex);
 651
 652         flags |= FMODE_EXCL;
 653         bdev = blkdev_get_by_path(path, flags, holder);
 654
 655         if (IS_ERR(bdev)) {
 656                 ret = PTR_ERR(bdev);
 657                 goto error;
 658         }
 659
 660         ret = set_blocksize(bdev, 4096);
 661         if (ret)
 662                 goto error_close;
 663         bh = btrfs_read_dev_super(bdev);
 664         if (!bh) {
 665                 ret = -EINVAL;
 666                 goto error_close;
 667         }
 668         disk_super = (struct btrfs_super_block *)bh->b_data;
 669         devid = btrfs_stack_device_id(&disk_super->dev_item);
 670         transid = btrfs_super_generation(disk_super);
 671         if (disk_super->label[0])
 672                 printk(KERN_INFO "device label %s ", disk_super->label);
 673         else {
 674                 /* FIXME, make a readl uuid parser */
 675                 printk(KERN_INFO "device fsid %llx-%llx ",
 676                        *(unsigned long long *)disk_super->fsid,
 677                        *(unsigned long long *)(disk_super->fsid + 8));
 678         }
 679         printk(KERN_CONT "devid %llu transid %llu %s\n",
 680                (unsigned long long)devid, (unsigned long long)transid, path);
 681         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 682
 683         brelse(bh);
 684 error_close:
 685         blkdev_put(bdev, flags);
 686 error:
 687         mutex_unlock(&uuid_mutex);
 688         return ret;
 689 }
 690
 691 /* helper to account the used device space in the range */
 692 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 693                                    u64 end, u64 *length)
 694 {
 695         struct btrfs_key key;
 696         struct btrfs_root *root = device->dev_root;
 697         struct btrfs_dev_extent *dev_extent;
 698         struct btrfs_path *path;
 699         u64 extent_end;
 700         int ret;
 701         int slot;
 702         struct extent_buffer *l;
 703
 704         *length = 0;
 705
 706         if (start >= device->total_bytes)
 707                 return 0;
 708
 709         path = btrfs_alloc_path();
 710         if (!path)
 711                 return -ENOMEM;
 712         path->reada = 2;
 713
 714         key.objectid = device->devid;
 715         key.offset = start;
 716         key.type = BTRFS_DEV_EXTENT_KEY;
 717
 718         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 719         if (ret < 0)
 720                 goto out;
 721         if (ret > 0) {
 722                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 723                 if (ret < 0)
 724                         goto out;
 725         }
 726
 727         while (1) {
 728                 l = path->nodes[0];
 729                 slot = path->slots[0];
 730                 if (slot >= btrfs_header_nritems(l)) {
 731                         ret = btrfs_next_leaf(root, path);
 732                         if (ret == 0)
 733                                 continue;
 734                         if (ret < 0)
 735                                 goto out;
 736
 737                         break;
 738                 }
 739                 btrfs_item_key_to_cpu(l, &key, slot);
 740
 741                 if (key.objectid < device->devid)
 742                         goto next;
 743
 744                 if (key.objectid > device->devid)
 745                         break;
 746
 747                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 748                         goto next;
 749
 750                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 751                 extent_end = key.offset + btrfs_dev_extent_length(l,
 752                                                                   dev_extent);
 753                 if (key.offset <= start && extent_end > end) {
 754                         *length = end - start + 1;
 755                         break;
 756                 } else if (key.offset <= start && extent_end > start)
 757                         *length += extent_end - start;
 758                 else if (key.offset > start && extent_end <= end)
 759                         *length += extent_end - key.offset;
 760                 else if (key.offset > start && key.offset <= end) {
 761                         *length += end - key.offset + 1;
 762                         break;
 763                 } else if (key.offset > end)
 764                         break;
 765
 766 next:
 767                 path->slots[0]++;
 768         }
 769         ret = 0;
 770 out:
 771         btrfs_free_path(path);
 772         return ret;
 773 }
 774
 775 /*
 776  * find_free_dev_extent - find free space in the specified device
 777  * @trans:      transaction handler
 778  * @device:     the device which we search the free space in
 779  * @num_bytes:  the size of the free space that we need
 780  * @start:      store the start of the free space.
 781  * @len:        the size of the free space. that we find, or the size of the max
 782  *              free space if we don't find suitable free space
 783  *
 784  * this uses a pretty simple search, the expectation is that it is
 785  * called very infrequently and that a given device has a small number
 786  * of extents
 787  *
 788  * @start is used to store the start of the free space if we find. But if we
 789  * don't find suitable free space, it will be used to store the start position
 790  * of the max free space.
 791  *
 792  * @len is used to store the size of the free space that we find.
 793  * But if we don't find suitable free space, it is used to store the size of
 794  * the max free space.
 795  */
 796 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 797                          struct btrfs_device *device, u64 num_bytes,
 798                          u64 *start, u64 *len)
 799 {
 800         struct btrfs_key key;
 801         struct btrfs_root *root = device->dev_root;
 802         struct btrfs_dev_extent *dev_extent;
 803         struct btrfs_path *path;
 804         u64 hole_size;
 805         u64 max_hole_start;
 806         u64 max_hole_size;
 807         u64 extent_end;
 808         u64 search_start;
 809         u64 search_end = device->total_bytes;
 810         int ret;
 811         int slot;
 812         struct extent_buffer *l;
 813
 814         /* FIXME use last free of some kind */
 815
 816         /* we don't want to overwrite the superblock on the drive,
 817          * so we make sure to start at an offset of at least 1MB
 818          */
 819         search_start = 1024 * 1024;
 820
 821         if (root->fs_info->alloc_start + num_bytes <= search_end)
 822                 search_start = max(root->fs_info->alloc_start, search_start);
 823
 824         max_hole_start = search_start;
 825         max_hole_size = 0;
 826
 827         if (search_start >= search_end) {
 828                 ret = -ENOSPC;
 829                 goto error;
 830         }
 831
 832         path = btrfs_alloc_path();
 833         if (!path) {
 834                 ret = -ENOMEM;
 835                 goto error;
 836         }
 837         path->reada = 2;
 838
 839         key.objectid = device->devid;
 840         key.offset = search_start;
 841         key.type = BTRFS_DEV_EXTENT_KEY;
 842
 843         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 844         if (ret < 0)
 845                 goto out;
 846         if (ret > 0) {
 847                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 848                 if (ret < 0)
 849                         goto out;
 850         }
 851
 852         while (1) {
 853                 l = path->nodes[0];
 854                 slot = path->slots[0];
 855                 if (slot >= btrfs_header_nritems(l)) {
 856                         ret = btrfs_next_leaf(root, path);
 857                         if (ret == 0)
 858                                 continue;
 859                         if (ret < 0)
 860                                 goto out;
 861
 862                         break;
 863                 }
 864                 btrfs_item_key_to_cpu(l, &key, slot);
 865
 866                 if (key.objectid < device->devid)
 867                         goto next;
 868
 869                 if (key.objectid > device->devid)
 870                         break;
 871
 872                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 873                         goto next;
 874
 875                 if (key.offset > search_start) {
 876                         hole_size = key.offset - search_start;
 877
 878                         if (hole_size > max_hole_size) {
 879                                 max_hole_start = search_start;
 880                                 max_hole_size = hole_size;
 881                         }
 882
 883                         /*
 884                          * If this free space is greater than which we need,
 885                          * it must be the max free space that we have found
 886                          * until now, so max_hole_start must point to the start
 887                          * of this free space and the length of this free space
 888                          * is stored in max_hole_size. Thus, we return
 889                          * max_hole_start and max_hole_size and go back to the
 890                          * caller.
 891                          */
 892                         if (hole_size >= num_bytes) {
 893                                 ret = 0;
 894                                 goto out;
 895                         }
 896                 }
 897
 898                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 899                 extent_end = key.offset + btrfs_dev_extent_length(l,
 900                                                                   dev_extent);
 901                 if (extent_end > search_start)
 902                         search_start = extent_end;
 903 next:
 904                 path->slots[0]++;
 905                 cond_resched();
 906         }
 907
 908         hole_size = search_end- search_start;
 909         if (hole_size > max_hole_size) {
 910                 max_hole_start = search_start;
 911                 max_hole_size = hole_size;
 912         }
 913
 914         /* See above. */
 915         if (hole_size < num_bytes)
 916                 ret = -ENOSPC;
 917         else
 918                 ret = 0;
 919
 920 out:
 921         btrfs_free_path(path);
 922 error:
 923         *start = max_hole_start;
 924         if (len)
 925                 *len = max_hole_size;
 926         return ret;
 927 }
 928
 929 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 930                           struct btrfs_device *device,
 931                           u64 start)
 932 {
 933         int ret;
 934         struct btrfs_path *path;
 935         struct btrfs_root *root = device->dev_root;
 936         struct btrfs_key key;
 937         struct btrfs_key found_key;
 938         struct extent_buffer *leaf = NULL;
 939         struct btrfs_dev_extent *extent = NULL;
 940
 941         path = btrfs_alloc_path();
 942         if (!path)
 943                 return -ENOMEM;
 944
 945         key.objectid = device->devid;
 946         key.offset = start;
 947         key.type = BTRFS_DEV_EXTENT_KEY;
 948
 949         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 950         if (ret > 0) {
 951                 ret = btrfs_previous_item(root, path, key.objectid,
 952                                           BTRFS_DEV_EXTENT_KEY);
 953                 BUG_ON(ret);
 954                 leaf = path->nodes[0];
 955                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 956                 extent = btrfs_item_ptr(leaf, path->slots[0],
 957                                         struct btrfs_dev_extent);
 958                 BUG_ON(found_key.offset > start || found_key.offset +
 959                        btrfs_dev_extent_length(leaf, extent) < start);
 960                 ret = 0;
 961         } else if (ret == 0) {
 962                 leaf = path->nodes[0];
 963                 extent = btrfs_item_ptr(leaf, path->slots[0],
 964                                         struct btrfs_dev_extent);
 965         }
 966         BUG_ON(ret);
 967
 968         if (device->bytes_used > 0)
 969                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
 970         ret = btrfs_del_item(trans, root, path);
 971         BUG_ON(ret);
 972
 973         btrfs_free_path(path);
 974         return ret;
 975 }
 976
 977 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
 978                            struct btrfs_device *device,
 979                            u64 chunk_tree, u64 chunk_objectid,
 980                            u64 chunk_offset, u64 start, u64 num_bytes)
 981 {
 982         int ret;
 983         struct btrfs_path *path;
 984         struct btrfs_root *root = device->dev_root;
 985         struct btrfs_dev_extent *extent;
 986         struct extent_buffer *leaf;
 987         struct btrfs_key key;
 988
 989         WARN_ON(!device->in_fs_metadata);
 990         path = btrfs_alloc_path();
 991         if (!path)
 992                 return -ENOMEM;
 993
 994         key.objectid = device->devid;
 995         key.offset = start;
 996         key.type = BTRFS_DEV_EXTENT_KEY;
 997         ret = btrfs_insert_empty_item(trans, root, path, &key,
 998                                       sizeof(*extent));
 999         BUG_ON(ret);
1000
1001         leaf = path->nodes[0];
1002         extent = btrfs_item_ptr(leaf, path->slots[0],
1003                                 struct btrfs_dev_extent);
1004         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1005         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1006         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1007
1008         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1009                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1010                     BTRFS_UUID_SIZE);
1011
1012         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1013         btrfs_mark_buffer_dirty(leaf);
1014         btrfs_free_path(path);
1015         return ret;
1016 }
1017
1018 static noinline int find_next_chunk(struct btrfs_root *root,
1019                                     u64 objectid, u64 *offset)
1020 {
1021         struct btrfs_path *path;
1022         int ret;
1023         struct btrfs_key key;
1024         struct btrfs_chunk *chunk;
1025         struct btrfs_key found_key;
1026
1027         path = btrfs_alloc_path();
1028         BUG_ON(!path);
1029
1030         key.objectid = objectid;
1031         key.offset = (u64)-1;
1032         key.type = BTRFS_CHUNK_ITEM_KEY;
1033
1034         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1035         if (ret < 0)
1036                 goto error;
1037
1038         BUG_ON(ret == 0);
1039
1040         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1041         if (ret) {
1042                 *offset = 0;
1043         } else {
1044                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1045                                       path->slots[0]);
1046                 if (found_key.objectid != objectid)
1047                         *offset = 0;
1048                 else {
1049                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1050                                                struct btrfs_chunk);
1051                         *offset = found_key.offset +
1052                                 btrfs_chunk_length(path->nodes[0], chunk);
1053                 }
1054         }
1055         ret = 0;
1056 error:
1057         btrfs_free_path(path);
1058         return ret;
1059 }
1060
1061 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1062 {
1063         int ret;
1064         struct btrfs_key key;
1065         struct btrfs_key found_key;
1066         struct btrfs_path *path;
1067
1068         root = root->fs_info->chunk_root;
1069
1070         path = btrfs_alloc_path();
1071         if (!path)
1072                 return -ENOMEM;
1073
1074         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1075         key.type = BTRFS_DEV_ITEM_KEY;
1076         key.offset = (u64)-1;
1077
1078         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1079         if (ret < 0)
1080                 goto error;
1081
1082         BUG_ON(ret == 0);
1083
1084         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1085                                   BTRFS_DEV_ITEM_KEY);
1086         if (ret) {
1087                 *objectid = 1;
1088         } else {
1089                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1090                                       path->slots[0]);
1091                 *objectid = found_key.offset + 1;
1092         }
1093         ret = 0;
1094 error:
1095         btrfs_free_path(path);
1096         return ret;
1097 }
1098
1099 /*
1100  * the device information is stored in the chunk root
1101  * the btrfs_device struct should be fully filled in
1102  */
1103 int btrfs_add_device(struct btrfs_trans_handle *trans,
1104                      struct btrfs_root *root,
1105                      struct btrfs_device *device)
1106 {
1107         int ret;
1108         struct btrfs_path *path;
1109         struct btrfs_dev_item *dev_item;
1110         struct extent_buffer *leaf;
1111         struct btrfs_key key;
1112         unsigned long ptr;
1113
1114         root = root->fs_info->chunk_root;
1115
1116         path = btrfs_alloc_path();
1117         if (!path)
1118                 return -ENOMEM;
1119
1120         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1121         key.type = BTRFS_DEV_ITEM_KEY;
1122         key.offset = device->devid;
1123
1124         ret = btrfs_insert_empty_item(trans, root, path, &key,
1125                                       sizeof(*dev_item));
1126         if (ret)
1127                 goto out;
1128
1129         leaf = path->nodes[0];
1130         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1131
1132         btrfs_set_device_id(leaf, dev_item, device->devid);
1133         btrfs_set_device_generation(leaf, dev_item, 0);
1134         btrfs_set_device_type(leaf, dev_item, device->type);
1135         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1136         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1137         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1138         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1139         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1140         btrfs_set_device_group(leaf, dev_item, 0);
1141         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1142         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1143         btrfs_set_device_start_offset(leaf, dev_item, 0);
1144
1145         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1146         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1147         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1148         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1149         btrfs_mark_buffer_dirty(leaf);
1150
1151         ret = 0;
1152 out:
1153         btrfs_free_path(path);
1154         return ret;
1155 }
1156
1157 static int btrfs_rm_dev_item(struct btrfs_root *root,
1158                              struct btrfs_device *device)
1159 {
1160         int ret;
1161         struct btrfs_path *path;
1162         struct btrfs_key key;
1163         struct btrfs_trans_handle *trans;
1164
1165         root = root->fs_info->chunk_root;
1166
1167         path = btrfs_alloc_path();
1168         if (!path)
1169                 return -ENOMEM;
1170
1171         trans = btrfs_start_transaction(root, 0);
1172         if (IS_ERR(trans)) {
1173                 btrfs_free_path(path);
1174                 return PTR_ERR(trans);
1175         }
1176         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1177         key.type = BTRFS_DEV_ITEM_KEY;
1178         key.offset = device->devid;
1179         lock_chunks(root);
1180
1181         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1182         if (ret < 0)
1183                 goto out;
1184
1185         if (ret > 0) {
1186                 ret = -ENOENT;
1187                 goto out;
1188         }
1189
1190         ret = btrfs_del_item(trans, root, path);
1191         if (ret)
1192                 goto out;
1193 out:
1194         btrfs_free_path(path);
1195         unlock_chunks(root);
1196         btrfs_commit_transaction(trans, root);
1197         return ret;
1198 }
1199
1200 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1201 {
1202         struct btrfs_device *device;
1203         struct btrfs_device *next_device;
1204         struct block_device *bdev;
1205         struct buffer_head *bh = NULL;
1206         struct btrfs_super_block *disk_super;
1207         u64 all_avail;
1208         u64 devid;
1209         u64 num_devices;
1210         u8 *dev_uuid;
1211         int ret = 0;
1212
1213         mutex_lock(&uuid_mutex);
1214         mutex_lock(&root->fs_info->volume_mutex);
1215
1216         all_avail = root->fs_info->avail_data_alloc_bits |
1217                 root->fs_info->avail_system_alloc_bits |
1218                 root->fs_info->avail_metadata_alloc_bits;
1219
1220         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1221             root->fs_info->fs_devices->num_devices <= 4) {
1222                 printk(KERN_ERR "btrfs: unable to go below four devices "
1223                        "on raid10\n");
1224                 ret = -EINVAL;
1225                 goto out;
1226         }
1227
1228         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1229             root->fs_info->fs_devices->num_devices <= 2) {
1230                 printk(KERN_ERR "btrfs: unable to go below two "
1231                        "devices on raid1\n");
1232                 ret = -EINVAL;
1233                 goto out;
1234         }
1235
1236         if (strcmp(device_path, "missing") == 0) {
1237                 struct list_head *devices;
1238                 struct btrfs_device *tmp;
1239
1240                 device = NULL;
1241                 devices = &root->fs_info->fs_devices->devices;
1242                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1243                 list_for_each_entry(tmp, devices, dev_list) {
1244                         if (tmp->in_fs_metadata && !tmp->bdev) {
1245                                 device = tmp;
1246                                 break;
1247                         }
1248                 }
1249                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1250                 bdev = NULL;
1251                 bh = NULL;
1252                 disk_super = NULL;
1253                 if (!device) {
1254                         printk(KERN_ERR "btrfs: no missing devices found to "
1255                                "remove\n");
1256                         goto out;
1257                 }
1258         } else {
1259                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1260                                           root->fs_info->bdev_holder);
1261                 if (IS_ERR(bdev)) {
1262                         ret = PTR_ERR(bdev);
1263                         goto out;
1264                 }
1265
1266                 set_blocksize(bdev, 4096);
1267                 bh = btrfs_read_dev_super(bdev);
1268                 if (!bh) {
1269                         ret = -EINVAL;
1270                         goto error_close;
1271                 }
1272                 disk_super = (struct btrfs_super_block *)bh->b_data;
1273                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1274                 dev_uuid = disk_super->dev_item.uuid;
1275                 device = btrfs_find_device(root, devid, dev_uuid,
1276                                            disk_super->fsid);
1277                 if (!device) {
1278                         ret = -ENOENT;
1279                         goto error_brelse;
1280                 }
1281         }
1282
1283         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1284                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1285                        "device\n");
1286                 ret = -EINVAL;
1287                 goto error_brelse;
1288         }
1289
1290         if (device->writeable) {
1291                 list_del_init(&device->dev_alloc_list);
1292                 root->fs_info->fs_devices->rw_devices--;
1293         }
1294
1295         ret = btrfs_shrink_device(device, 0);
1296         if (ret)
1297                 goto error_brelse;
1298
1299         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1300         if (ret)
1301                 goto error_brelse;
1302
1303         device->in_fs_metadata = 0;
1304
1305         /*
1306          * the device list mutex makes sure that we don't change
1307          * the device list while someone else is writing out all
1308          * the device supers.
1309          */
1310         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1311         list_del_init(&device->dev_list);
1312         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1313
1314         device->fs_devices->num_devices--;
1315
1316         if (device->missing)
1317                 root->fs_info->fs_devices->missing_devices--;
1318
1319         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1320                                  struct btrfs_device, dev_list);
1321         if (device->bdev == root->fs_info->sb->s_bdev)
1322                 root->fs_info->sb->s_bdev = next_device->bdev;
1323         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1324                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1325
1326         if (device->bdev) {
1327                 blkdev_put(device->bdev, device->mode);
1328                 device->bdev = NULL;
1329                 device->fs_devices->open_devices--;
1330         }
1331
1332         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1333         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1334
1335         if (device->fs_devices->open_devices == 0) {
1336                 struct btrfs_fs_devices *fs_devices;
1337                 fs_devices = root->fs_info->fs_devices;
1338                 while (fs_devices) {
1339                         if (fs_devices->seed == device->fs_devices)
1340                                 break;
1341                         fs_devices = fs_devices->seed;
1342                 }
1343                 fs_devices->seed = device->fs_devices->seed;
1344                 device->fs_devices->seed = NULL;
1345                 __btrfs_close_devices(device->fs_devices);
1346                 free_fs_devices(device->fs_devices);
1347         }
1348
1349         /*
1350          * at this point, the device is zero sized.  We want to
1351          * remove it from the devices list and zero out the old super
1352          */
1353         if (device->writeable) {
1354                 /* make sure this device isn't detected as part of
1355                  * the FS anymore
1356                  */
1357                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1358                 set_buffer_dirty(bh);
1359                 sync_dirty_buffer(bh);
1360         }
1361
1362         kfree(device->name);
1363         kfree(device);
1364         ret = 0;
1365
1366 error_brelse:
1367         brelse(bh);
1368 error_close:
1369         if (bdev)
1370                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1371 out:
1372         mutex_unlock(&root->fs_info->volume_mutex);
1373         mutex_unlock(&uuid_mutex);
1374         return ret;
1375 }
1376
1377 /*
1378  * does all the dirty work required for changing file system's UUID.
1379  */
1380 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1381                                 struct btrfs_root *root)
1382 {
1383         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1384         struct btrfs_fs_devices *old_devices;
1385         struct btrfs_fs_devices *seed_devices;
1386         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1387         struct btrfs_device *device;
1388         u64 super_flags;
1389
1390         BUG_ON(!mutex_is_locked(&uuid_mutex));
1391         if (!fs_devices->seeding)
1392                 return -EINVAL;
1393
1394         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1395         if (!seed_devices)
1396                 return -ENOMEM;
1397
1398         old_devices = clone_fs_devices(fs_devices);
1399         if (IS_ERR(old_devices)) {
1400                 kfree(seed_devices);
1401                 return PTR_ERR(old_devices);
1402         }
1403
1404         list_add(&old_devices->list, &fs_uuids);
1405
1406         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1407         seed_devices->opened = 1;
1408         INIT_LIST_HEAD(&seed_devices->devices);
1409         INIT_LIST_HEAD(&seed_devices->alloc_list);
1410         mutex_init(&seed_devices->device_list_mutex);
1411         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1412         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1413         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1414                 device->fs_devices = seed_devices;
1415         }
1416
1417         fs_devices->seeding = 0;
1418         fs_devices->num_devices = 0;
1419         fs_devices->open_devices = 0;
1420         fs_devices->seed = seed_devices;
1421
1422         generate_random_uuid(fs_devices->fsid);
1423         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1424         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1425         super_flags = btrfs_super_flags(disk_super) &
1426                       ~BTRFS_SUPER_FLAG_SEEDING;
1427         btrfs_set_super_flags(disk_super, super_flags);
1428
1429         return 0;
1430 }
1431
1432 /*
1433  * strore the expected generation for seed devices in device items.
1434  */
1435 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1436                                struct btrfs_root *root)
1437 {
1438         struct btrfs_path *path;
1439         struct extent_buffer *leaf;
1440         struct btrfs_dev_item *dev_item;
1441         struct btrfs_device *device;
1442         struct btrfs_key key;
1443         u8 fs_uuid[BTRFS_UUID_SIZE];
1444         u8 dev_uuid[BTRFS_UUID_SIZE];
1445         u64 devid;
1446         int ret;
1447
1448         path = btrfs_alloc_path();
1449         if (!path)
1450                 return -ENOMEM;
1451
1452         root = root->fs_info->chunk_root;
1453         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1454         key.offset = 0;
1455         key.type = BTRFS_DEV_ITEM_KEY;
1456
1457         while (1) {
1458                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1459                 if (ret < 0)
1460                         goto error;
1461
1462                 leaf = path->nodes[0];
1463 next_slot:
1464                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1465                         ret = btrfs_next_leaf(root, path);
1466                         if (ret > 0)
1467                                 break;
1468                         if (ret < 0)
1469                                 goto error;
1470                         leaf = path->nodes[0];
1471                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1472                         btrfs_release_path(root, path);
1473                         continue;
1474                 }
1475
1476                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1477                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1478                     key.type != BTRFS_DEV_ITEM_KEY)
1479                         break;
1480
1481                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1482                                           struct btrfs_dev_item);
1483                 devid = btrfs_device_id(leaf, dev_item);
1484                 read_extent_buffer(leaf, dev_uuid,
1485                                    (unsigned long)btrfs_device_uuid(dev_item),
1486                                    BTRFS_UUID_SIZE);
1487                 read_extent_buffer(leaf, fs_uuid,
1488                                    (unsigned long)btrfs_device_fsid(dev_item),
1489                                    BTRFS_UUID_SIZE);
1490                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1491                 BUG_ON(!device);
1492
1493                 if (device->fs_devices->seeding) {
1494                         btrfs_set_device_generation(leaf, dev_item,
1495                                                     device->generation);
1496                         btrfs_mark_buffer_dirty(leaf);
1497                 }
1498
1499                 path->slots[0]++;
1500                 goto next_slot;
1501         }
1502         ret = 0;
1503 error:
1504         btrfs_free_path(path);
1505         return ret;
1506 }
1507
1508 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1509 {
1510         struct btrfs_trans_handle *trans;
1511         struct btrfs_device *device;
1512         struct block_device *bdev;
1513         struct list_head *devices;
1514         struct super_block *sb = root->fs_info->sb;
1515         u64 total_bytes;
1516         int seeding_dev = 0;
1517         int ret = 0;
1518
1519         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1520                 return -EINVAL;
1521
1522         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1523                                   root->fs_info->bdev_holder);
1524         if (IS_ERR(bdev))
1525                 return PTR_ERR(bdev);
1526
1527         if (root->fs_info->fs_devices->seeding) {
1528                 seeding_dev = 1;
1529                 down_write(&sb->s_umount);
1530                 mutex_lock(&uuid_mutex);
1531         }
1532
1533         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1534         mutex_lock(&root->fs_info->volume_mutex);
1535
1536         devices = &root->fs_info->fs_devices->devices;
1537         /*
1538          * we have the volume lock, so we don't need the extra
1539          * device list mutex while reading the list here.
1540          */
1541         list_for_each_entry(device, devices, dev_list) {
1542                 if (device->bdev == bdev) {
1543                         ret = -EEXIST;
1544                         goto error;
1545                 }
1546         }
1547
1548         device = kzalloc(sizeof(*device), GFP_NOFS);
1549         if (!device) {
1550                 /* we can safely leave the fs_devices entry around */
1551                 ret = -ENOMEM;
1552                 goto error;
1553         }
1554
1555         device->name = kstrdup(device_path, GFP_NOFS);
1556         if (!device->name) {
1557                 kfree(device);
1558                 ret = -ENOMEM;
1559                 goto error;
1560         }
1561
1562         ret = find_next_devid(root, &device->devid);
1563         if (ret) {
1564                 kfree(device->name);
1565                 kfree(device);
1566                 goto error;
1567         }
1568
1569         trans = btrfs_start_transaction(root, 0);
1570         if (IS_ERR(trans)) {
1571                 kfree(device->name);
1572                 kfree(device);
1573                 ret = PTR_ERR(trans);
1574                 goto error;
1575         }
1576
1577         lock_chunks(root);
1578
1579         device->writeable = 1;
1580         device->work.func = pending_bios_fn;
1581         generate_random_uuid(device->uuid);
1582         spin_lock_init(&device->io_lock);
1583         device->generation = trans->transid;
1584         device->io_width = root->sectorsize;
1585         device->io_align = root->sectorsize;
1586         device->sector_size = root->sectorsize;
1587         device->total_bytes = i_size_read(bdev->bd_inode);
1588         device->disk_total_bytes = device->total_bytes;
1589         device->dev_root = root->fs_info->dev_root;
1590         device->bdev = bdev;
1591         device->in_fs_metadata = 1;
1592         device->mode = 0;
1593         set_blocksize(device->bdev, 4096);
1594
1595         if (seeding_dev) {
1596                 sb->s_flags &= ~MS_RDONLY;
1597                 ret = btrfs_prepare_sprout(trans, root);
1598                 BUG_ON(ret);
1599         }
1600
1601         device->fs_devices = root->fs_info->fs_devices;
1602
1603         /*
1604          * we don't want write_supers to jump in here with our device
1605          * half setup
1606          */
1607         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1608         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1609         list_add(&device->dev_alloc_list,
1610                  &root->fs_info->fs_devices->alloc_list);
1611         root->fs_info->fs_devices->num_devices++;
1612         root->fs_info->fs_devices->open_devices++;
1613         root->fs_info->fs_devices->rw_devices++;
1614         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1615
1616         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1617                 root->fs_info->fs_devices->rotating = 1;
1618
1619         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1620         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1621                                     total_bytes + device->total_bytes);
1622
1623         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1624         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1625                                     total_bytes + 1);
1626         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1627
1628         if (seeding_dev) {
1629                 ret = init_first_rw_device(trans, root, device);
1630                 BUG_ON(ret);
1631                 ret = btrfs_finish_sprout(trans, root);
1632                 BUG_ON(ret);
1633         } else {
1634                 ret = btrfs_add_device(trans, root, device);
1635         }
1636
1637         /*
1638          * we've got more storage, clear any full flags on the space
1639          * infos
1640          */
1641         btrfs_clear_space_info_full(root->fs_info);
1642
1643         unlock_chunks(root);
1644         btrfs_commit_transaction(trans, root);
1645
1646         if (seeding_dev) {
1647                 mutex_unlock(&uuid_mutex);
1648                 up_write(&sb->s_umount);
1649
1650                 ret = btrfs_relocate_sys_chunks(root);
1651                 BUG_ON(ret);
1652         }
1653 out:
1654         mutex_unlock(&root->fs_info->volume_mutex);
1655         return ret;
1656 error:
1657         blkdev_put(bdev, FMODE_EXCL);
1658         if (seeding_dev) {
1659                 mutex_unlock(&uuid_mutex);
1660                 up_write(&sb->s_umount);
1661         }
1662         goto out;
1663 }
1664
1665 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1666                                         struct btrfs_device *device)
1667 {
1668         int ret;
1669         struct btrfs_path *path;
1670         struct btrfs_root *root;
1671         struct btrfs_dev_item *dev_item;
1672         struct extent_buffer *leaf;
1673         struct btrfs_key key;
1674
1675         root = device->dev_root->fs_info->chunk_root;
1676
1677         path = btrfs_alloc_path();
1678         if (!path)
1679                 return -ENOMEM;
1680
1681         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1682         key.type = BTRFS_DEV_ITEM_KEY;
1683         key.offset = device->devid;
1684
1685         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1686         if (ret < 0)
1687                 goto out;
1688
1689         if (ret > 0) {
1690                 ret = -ENOENT;
1691                 goto out;
1692         }
1693
1694         leaf = path->nodes[0];
1695         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1696
1697         btrfs_set_device_id(leaf, dev_item, device->devid);
1698         btrfs_set_device_type(leaf, dev_item, device->type);
1699         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1700         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1701         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1702         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1703         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1704         btrfs_mark_buffer_dirty(leaf);
1705
1706 out:
1707         btrfs_free_path(path);
1708         return ret;
1709 }
1710
1711 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1712                       struct btrfs_device *device, u64 new_size)
1713 {
1714         struct btrfs_super_block *super_copy =
1715                 &device->dev_root->fs_info->super_copy;
1716         u64 old_total = btrfs_super_total_bytes(super_copy);
1717         u64 diff = new_size - device->total_bytes;
1718
1719         if (!device->writeable)
1720                 return -EACCES;
1721         if (new_size <= device->total_bytes)
1722                 return -EINVAL;
1723
1724         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1725         device->fs_devices->total_rw_bytes += diff;
1726
1727         device->total_bytes = new_size;
1728         device->disk_total_bytes = new_size;
1729         btrfs_clear_space_info_full(device->dev_root->fs_info);
1730
1731         return btrfs_update_device(trans, device);
1732 }
1733
1734 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1735                       struct btrfs_device *device, u64 new_size)
1736 {
1737         int ret;
1738         lock_chunks(device->dev_root);
1739         ret = __btrfs_grow_device(trans, device, new_size);
1740         unlock_chunks(device->dev_root);
1741         return ret;
1742 }
1743
1744 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1745                             struct btrfs_root *root,
1746                             u64 chunk_tree, u64 chunk_objectid,
1747                             u64 chunk_offset)
1748 {
1749         int ret;
1750         struct btrfs_path *path;
1751         struct btrfs_key key;
1752
1753         root = root->fs_info->chunk_root;
1754         path = btrfs_alloc_path();
1755         if (!path)
1756                 return -ENOMEM;
1757
1758         key.objectid = chunk_objectid;
1759         key.offset = chunk_offset;
1760         key.type = BTRFS_CHUNK_ITEM_KEY;
1761
1762         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1763         BUG_ON(ret);
1764
1765         ret = btrfs_del_item(trans, root, path);
1766         BUG_ON(ret);
1767
1768         btrfs_free_path(path);
1769         return 0;
1770 }
1771
1772 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1773                         chunk_offset)
1774 {
1775         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1776         struct btrfs_disk_key *disk_key;
1777         struct btrfs_chunk *chunk;
1778         u8 *ptr;
1779         int ret = 0;
1780         u32 num_stripes;
1781         u32 array_size;
1782         u32 len = 0;
1783         u32 cur;
1784         struct btrfs_key key;
1785
1786         array_size = btrfs_super_sys_array_size(super_copy);
1787
1788         ptr = super_copy->sys_chunk_array;
1789         cur = 0;
1790
1791         while (cur < array_size) {
1792                 disk_key = (struct btrfs_disk_key *)ptr;
1793                 btrfs_disk_key_to_cpu(&key, disk_key);
1794
1795                 len = sizeof(*disk_key);
1796
1797                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1798                         chunk = (struct btrfs_chunk *)(ptr + len);
1799                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1800                         len += btrfs_chunk_item_size(num_stripes);
1801                 } else {
1802                         ret = -EIO;
1803                         break;
1804                 }
1805                 if (key.objectid == chunk_objectid &&
1806                     key.offset == chunk_offset) {
1807                         memmove(ptr, ptr + len, array_size - (cur + len));
1808                         array_size -= len;
1809                         btrfs_set_super_sys_array_size(super_copy, array_size);
1810                 } else {
1811                         ptr += len;
1812                         cur += len;
1813                 }
1814         }
1815         return ret;
1816 }
1817
1818 static int btrfs_relocate_chunk(struct btrfs_root *root,
1819                          u64 chunk_tree, u64 chunk_objectid,
1820                          u64 chunk_offset)
1821 {
1822         struct extent_map_tree *em_tree;
1823         struct btrfs_root *extent_root;
1824         struct btrfs_trans_handle *trans;
1825         struct extent_map *em;
1826         struct map_lookup *map;
1827         int ret;
1828         int i;
1829
1830         root = root->fs_info->chunk_root;
1831         extent_root = root->fs_info->extent_root;
1832         em_tree = &root->fs_info->mapping_tree.map_tree;
1833
1834         ret = btrfs_can_relocate(extent_root, chunk_offset);
1835         if (ret)
1836                 return -ENOSPC;
1837
1838         /* step one, relocate all the extents inside this chunk */
1839         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1840         if (ret)
1841                 return ret;
1842
1843         trans = btrfs_start_transaction(root, 0);
1844         BUG_ON(IS_ERR(trans));
1845
1846         lock_chunks(root);
1847
1848         /*
1849          * step two, delete the device extents and the
1850          * chunk tree entries
1851          */
1852         read_lock(&em_tree->lock);
1853         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1854         read_unlock(&em_tree->lock);
1855
1856         BUG_ON(em->start > chunk_offset ||
1857                em->start + em->len < chunk_offset);
1858         map = (struct map_lookup *)em->bdev;
1859
1860         for (i = 0; i < map->num_stripes; i++) {
1861                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1862                                             map->stripes[i].physical);
1863                 BUG_ON(ret);
1864
1865                 if (map->stripes[i].dev) {
1866                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1867                         BUG_ON(ret);
1868                 }
1869         }
1870         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1871                                chunk_offset);
1872
1873         BUG_ON(ret);
1874
1875         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1876                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1877                 BUG_ON(ret);
1878         }
1879
1880         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1881         BUG_ON(ret);
1882
1883         write_lock(&em_tree->lock);
1884         remove_extent_mapping(em_tree, em);
1885         write_unlock(&em_tree->lock);
1886
1887         kfree(map);
1888         em->bdev = NULL;
1889
1890         /* once for the tree */
1891         free_extent_map(em);
1892         /* once for us */
1893         free_extent_map(em);
1894
1895         unlock_chunks(root);
1896         btrfs_end_transaction(trans, root);
1897         return 0;
1898 }
1899
1900 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1901 {
1902         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1903         struct btrfs_path *path;
1904         struct extent_buffer *leaf;
1905         struct btrfs_chunk *chunk;
1906         struct btrfs_key key;
1907         struct btrfs_key found_key;
1908         u64 chunk_tree = chunk_root->root_key.objectid;
1909         u64 chunk_type;
1910         bool retried = false;
1911         int failed = 0;
1912         int ret;
1913
1914         path = btrfs_alloc_path();
1915         if (!path)
1916                 return -ENOMEM;
1917
1918 again:
1919         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1920         key.offset = (u64)-1;
1921         key.type = BTRFS_CHUNK_ITEM_KEY;
1922
1923         while (1) {
1924                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1925                 if (ret < 0)
1926                         goto error;
1927                 BUG_ON(ret == 0);
1928
1929                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1930                                           key.type);
1931                 if (ret < 0)
1932                         goto error;
1933                 if (ret > 0)
1934                         break;
1935
1936                 leaf = path->nodes[0];
1937                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1938
1939                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1940                                        struct btrfs_chunk);
1941                 chunk_type = btrfs_chunk_type(leaf, chunk);
1942                 btrfs_release_path(chunk_root, path);
1943
1944                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1945                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1946                                                    found_key.objectid,
1947                                                    found_key.offset);
1948                         if (ret == -ENOSPC)
1949                                 failed++;
1950                         else if (ret)
1951                                 BUG();
1952                 }
1953
1954                 if (found_key.offset == 0)
1955                         break;
1956                 key.offset = found_key.offset - 1;
1957         }
1958         ret = 0;
1959         if (failed && !retried) {
1960                 failed = 0;
1961                 retried = true;
1962                 goto again;
1963         } else if (failed && retried) {
1964                 WARN_ON(1);
1965                 ret = -ENOSPC;
1966         }
1967 error:
1968         btrfs_free_path(path);
1969         return ret;
1970 }
1971
1972 static u64 div_factor(u64 num, int factor)
1973 {
1974         if (factor == 10)
1975                 return num;
1976         num *= factor;
1977         do_div(num, 10);
1978         return num;
1979 }
1980
1981 int btrfs_balance(struct btrfs_root *dev_root)
1982 {
1983         int ret;
1984         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1985         struct btrfs_device *device;
1986         u64 old_size;
1987         u64 size_to_free;
1988         struct btrfs_path *path;
1989         struct btrfs_key key;
1990         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1991         struct btrfs_trans_handle *trans;
1992         struct btrfs_key found_key;
1993
1994         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1995                 return -EROFS;
1996
1997         if (!capable(CAP_SYS_ADMIN))
1998                 return -EPERM;
1999
2000         mutex_lock(&dev_root->fs_info->volume_mutex);
2001         dev_root = dev_root->fs_info->dev_root;
2002
2003         /* step one make some room on all the devices */
2004         list_for_each_entry(device, devices, dev_list) {
2005                 old_size = device->total_bytes;
2006                 size_to_free = div_factor(old_size, 1);
2007                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2008                 if (!device->writeable ||
2009                     device->total_bytes - device->bytes_used > size_to_free)
2010                         continue;
2011
2012                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2013                 if (ret == -ENOSPC)
2014                         break;
2015                 BUG_ON(ret);
2016
2017                 trans = btrfs_start_transaction(dev_root, 0);
2018                 BUG_ON(IS_ERR(trans));
2019
2020                 ret = btrfs_grow_device(trans, device, old_size);
2021                 BUG_ON(ret);
2022
2023                 btrfs_end_transaction(trans, dev_root);
2024         }
2025
2026         /* step two, relocate all the chunks */
2027         path = btrfs_alloc_path();
2028         BUG_ON(!path);
2029
2030         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2031         key.offset = (u64)-1;
2032         key.type = BTRFS_CHUNK_ITEM_KEY;
2033
2034         while (1) {
2035                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2036                 if (ret < 0)
2037                         goto error;
2038
2039                 /*
2040                  * this shouldn't happen, it means the last relocate
2041                  * failed
2042                  */
2043                 if (ret == 0)
2044                         break;
2045
2046                 ret = btrfs_previous_item(chunk_root, path, 0,
2047                                           BTRFS_CHUNK_ITEM_KEY);
2048                 if (ret)
2049                         break;
2050
2051                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2052                                       path->slots[0]);
2053                 if (found_key.objectid != key.objectid)
2054                         break;
2055
2056                 /* chunk zero is special */
2057                 if (found_key.offset == 0)
2058                         break;
2059
2060                 btrfs_release_path(chunk_root, path);
2061                 ret = btrfs_relocate_chunk(chunk_root,
2062                                            chunk_root->root_key.objectid,
2063                                            found_key.objectid,
2064                                            found_key.offset);
2065                 BUG_ON(ret && ret != -ENOSPC);
2066                 key.offset = found_key.offset - 1;
2067         }
2068         ret = 0;
2069 error:
2070         btrfs_free_path(path);
2071         mutex_unlock(&dev_root->fs_info->volume_mutex);
2072         return ret;
2073 }
2074
2075 /*
2076  * shrinking a device means finding all of the device extents past
2077  * the new size, and then following the back refs to the chunks.
2078  * The chunk relocation code actually frees the device extent
2079  */
2080 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2081 {
2082         struct btrfs_trans_handle *trans;
2083         struct btrfs_root *root = device->dev_root;
2084         struct btrfs_dev_extent *dev_extent = NULL;
2085         struct btrfs_path *path;
2086         u64 length;
2087         u64 chunk_tree;
2088         u64 chunk_objectid;
2089         u64 chunk_offset;
2090         int ret;
2091         int slot;
2092         int failed = 0;
2093         bool retried = false;
2094         struct extent_buffer *l;
2095         struct btrfs_key key;
2096         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2097         u64 old_total = btrfs_super_total_bytes(super_copy);
2098         u64 old_size = device->total_bytes;
2099         u64 diff = device->total_bytes - new_size;
2100
2101         if (new_size >= device->total_bytes)
2102                 return -EINVAL;
2103
2104         path = btrfs_alloc_path();
2105         if (!path)
2106                 return -ENOMEM;
2107
2108         path->reada = 2;
2109
2110         lock_chunks(root);
2111
2112         device->total_bytes = new_size;
2113         if (device->writeable)
2114                 device->fs_devices->total_rw_bytes -= diff;
2115         unlock_chunks(root);
2116
2117 again:
2118         key.objectid = device->devid;
2119         key.offset = (u64)-1;
2120         key.type = BTRFS_DEV_EXTENT_KEY;
2121
2122         while (1) {
2123                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2124                 if (ret < 0)
2125                         goto done;
2126
2127                 ret = btrfs_previous_item(root, path, 0, key.type);
2128                 if (ret < 0)
2129                         goto done;
2130                 if (ret) {
2131                         ret = 0;
2132                         btrfs_release_path(root, path);
2133                         break;
2134                 }
2135
2136                 l = path->nodes[0];
2137                 slot = path->slots[0];
2138                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2139
2140                 if (key.objectid != device->devid) {
2141                         btrfs_release_path(root, path);
2142                         break;
2143                 }
2144
2145                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2146                 length = btrfs_dev_extent_length(l, dev_extent);
2147
2148                 if (key.offset + length <= new_size) {
2149                         btrfs_release_path(root, path);
2150                         break;
2151                 }
2152
2153                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2154                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2155                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2156                 btrfs_release_path(root, path);
2157
2158                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2159                                            chunk_offset);
2160                 if (ret && ret != -ENOSPC)
2161                         goto done;
2162                 if (ret == -ENOSPC)
2163                         failed++;
2164                 key.offset -= 1;
2165         }
2166
2167         if (failed && !retried) {
2168                 failed = 0;
2169                 retried = true;
2170                 goto again;
2171         } else if (failed && retried) {
2172                 ret = -ENOSPC;
2173                 lock_chunks(root);
2174
2175                 device->total_bytes = old_size;
2176                 if (device->writeable)
2177                         device->fs_devices->total_rw_bytes += diff;
2178                 unlock_chunks(root);
2179                 goto done;
2180         }
2181
2182         /* Shrinking succeeded, else we would be at "done". */
2183         trans = btrfs_start_transaction(root, 0);
2184         if (IS_ERR(trans)) {
2185                 ret = PTR_ERR(trans);
2186                 goto done;
2187         }
2188
2189         lock_chunks(root);
2190
2191         device->disk_total_bytes = new_size;
2192         /* Now btrfs_update_device() will change the on-disk size. */
2193         ret = btrfs_update_device(trans, device);
2194         if (ret) {
2195                 unlock_chunks(root);
2196                 btrfs_end_transaction(trans, root);
2197                 goto done;
2198         }
2199         WARN_ON(diff > old_total);
2200         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2201         unlock_chunks(root);
2202         btrfs_end_transaction(trans, root);
2203 done:
2204         btrfs_free_path(path);
2205         return ret;
2206 }
2207
2208 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2209                            struct btrfs_root *root,
2210                            struct btrfs_key *key,
2211                            struct btrfs_chunk *chunk, int item_size)
2212 {
2213         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2214         struct btrfs_disk_key disk_key;
2215         u32 array_size;
2216         u8 *ptr;
2217
2218         array_size = btrfs_super_sys_array_size(super_copy);
2219         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2220                 return -EFBIG;
2221
2222         ptr = super_copy->sys_chunk_array + array_size;
2223         btrfs_cpu_key_to_disk(&disk_key, key);
2224         memcpy(ptr, &disk_key, sizeof(disk_key));
2225         ptr += sizeof(disk_key);
2226         memcpy(ptr, chunk, item_size);
2227         item_size += sizeof(disk_key);
2228         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2229         return 0;
2230 }
2231
2232 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2233                                         int num_stripes, int sub_stripes)
2234 {
2235         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2236                 return calc_size;
2237         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2238                 return calc_size * (num_stripes / sub_stripes);
2239         else
2240                 return calc_size * num_stripes;
2241 }
2242
2243 /* Used to sort the devices by max_avail(descending sort) */
2244 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2245 {
2246         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2247             ((struct btrfs_device_info *)dev_info2)->max_avail)
2248                 return -1;
2249         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2250                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2251                 return 1;
2252         else
2253                 return 0;
2254 }
2255
2256 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2257                                  int *num_stripes, int *min_stripes,
2258                                  int *sub_stripes)
2259 {
2260         *num_stripes = 1;
2261         *min_stripes = 1;
2262         *sub_stripes = 0;
2263
2264         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2265                 *num_stripes = fs_devices->rw_devices;
2266                 *min_stripes = 2;
2267         }
2268         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2269                 *num_stripes = 2;
2270                 *min_stripes = 2;
2271         }
2272         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2273                 if (fs_devices->rw_devices < 2)
2274                         return -ENOSPC;
2275                 *num_stripes = 2;
2276                 *min_stripes = 2;
2277         }
2278         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2279                 *num_stripes = fs_devices->rw_devices;
2280                 if (*num_stripes < 4)
2281                         return -ENOSPC;
2282                 *num_stripes &= ~(u32)1;
2283                 *sub_stripes = 2;
2284                 *min_stripes = 4;
2285         }
2286
2287         return 0;
2288 }
2289
2290 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2291                                     u64 proposed_size, u64 type,
2292                                     int num_stripes, int small_stripe)
2293 {
2294         int min_stripe_size = 1 * 1024 * 1024;
2295         u64 calc_size = proposed_size;
2296         u64 max_chunk_size = calc_size;
2297         int ncopies = 1;
2298
2299         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2300                     BTRFS_BLOCK_GROUP_DUP |
2301                     BTRFS_BLOCK_GROUP_RAID10))
2302                 ncopies = 2;
2303
2304         if (type & BTRFS_BLOCK_GROUP_DATA) {
2305                 max_chunk_size = 10 * calc_size;
2306                 min_stripe_size = 64 * 1024 * 1024;
2307         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2308                 max_chunk_size = 256 * 1024 * 1024;
2309                 min_stripe_size = 32 * 1024 * 1024;
2310         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2311                 calc_size = 8 * 1024 * 1024;
2312                 max_chunk_size = calc_size * 2;
2313                 min_stripe_size = 1 * 1024 * 1024;
2314         }
2315
2316         /* we don't want a chunk larger than 10% of writeable space */
2317         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2318                              max_chunk_size);
2319
2320         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2321                 calc_size = max_chunk_size * ncopies;
2322                 do_div(calc_size, num_stripes);
2323                 do_div(calc_size, BTRFS_STRIPE_LEN);
2324                 calc_size *= BTRFS_STRIPE_LEN;
2325         }
2326
2327         /* we don't want tiny stripes */
2328         if (!small_stripe)
2329                 calc_size = max_t(u64, min_stripe_size, calc_size);
2330
2331         /*
2332          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2333          * we end up with something bigger than a stripe
2334          */
2335         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2336
2337         do_div(calc_size, BTRFS_STRIPE_LEN);
2338         calc_size *= BTRFS_STRIPE_LEN;
2339
2340         return calc_size;
2341 }
2342
2343 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2344                                                       int num_stripes)
2345 {
2346         struct map_lookup *new;
2347         size_t len = map_lookup_size(num_stripes);
2348
2349         BUG_ON(map->num_stripes < num_stripes);
2350
2351         if (map->num_stripes == num_stripes)
2352                 return map;
2353
2354         new = kmalloc(len, GFP_NOFS);
2355         if (!new) {
2356                 /* just change map->num_stripes */
2357                 map->num_stripes = num_stripes;
2358                 return map;
2359         }
2360
2361         memcpy(new, map, len);
2362         new->num_stripes = num_stripes;
2363         kfree(map);
2364         return new;
2365 }
2366
2367 /*
2368  * helper to allocate device space from btrfs_device_info, in which we stored
2369  * max free space information of every device. It is used when we can not
2370  * allocate chunks by default size.
2371  *
2372  * By this helper, we can allocate a new chunk as larger as possible.
2373  */
2374 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2375                                     struct btrfs_fs_devices *fs_devices,
2376                                     struct btrfs_device_info *devices,
2377                                     int nr_device, u64 type,
2378                                     struct map_lookup **map_lookup,
2379                                     int min_stripes, u64 *stripe_size)
2380 {
2381         int i, index, sort_again = 0;
2382         int min_devices = min_stripes;
2383         u64 max_avail, min_free;
2384         struct map_lookup *map = *map_lookup;
2385         int ret;
2386
2387         if (nr_device < min_stripes)
2388                 return -ENOSPC;
2389
2390         btrfs_descending_sort_devices(devices, nr_device);
2391
2392         max_avail = devices[0].max_avail;
2393         if (!max_avail)
2394                 return -ENOSPC;
2395
2396         for (i = 0; i < nr_device; i++) {
2397                 /*
2398                  * if dev_offset = 0, it means the free space of this device
2399                  * is less than what we need, and we didn't search max avail
2400                  * extent on this device, so do it now.
2401                  */
2402                 if (!devices[i].dev_offset) {
2403                         ret = find_free_dev_extent(trans, devices[i].dev,
2404                                                    max_avail,
2405                                                    &devices[i].dev_offset,
2406                                                    &devices[i].max_avail);
2407                         if (ret != 0 && ret != -ENOSPC)
2408                                 return ret;
2409                         sort_again = 1;
2410                 }
2411         }
2412
2413         /* we update the max avail free extent of each devices, sort again */
2414         if (sort_again)
2415                 btrfs_descending_sort_devices(devices, nr_device);
2416
2417         if (type & BTRFS_BLOCK_GROUP_DUP)
2418                 min_devices = 1;
2419
2420         if (!devices[min_devices - 1].max_avail)
2421                 return -ENOSPC;
2422
2423         max_avail = devices[min_devices - 1].max_avail;
2424         if (type & BTRFS_BLOCK_GROUP_DUP)
2425                 do_div(max_avail, 2);
2426
2427         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2428                                              min_stripes, 1);
2429         if (type & BTRFS_BLOCK_GROUP_DUP)
2430                 min_free = max_avail * 2;
2431         else
2432                 min_free = max_avail;
2433
2434         if (min_free > devices[min_devices - 1].max_avail)
2435                 return -ENOSPC;
2436
2437         map = __shrink_map_lookup_stripes(map, min_stripes);
2438         *stripe_size = max_avail;
2439
2440         index = 0;
2441         for (i = 0; i < min_stripes; i++) {
2442                 map->stripes[i].dev = devices[index].dev;
2443                 map->stripes[i].physical = devices[index].dev_offset;
2444                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2445                         i++;
2446                         map->stripes[i].dev = devices[index].dev;
2447                         map->stripes[i].physical = devices[index].dev_offset +
2448                                                    max_avail;
2449                 }
2450                 index++;
2451         }
2452         *map_lookup = map;
2453
2454         return 0;
2455 }
2456
2457 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2458                                struct btrfs_root *extent_root,
2459                                struct map_lookup **map_ret,
2460                                u64 *num_bytes, u64 *stripe_size,
2461                                u64 start, u64 type)
2462 {
2463         struct btrfs_fs_info *info = extent_root->fs_info;
2464         struct btrfs_device *device = NULL;
2465         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2466         struct list_head *cur;
2467         struct map_lookup *map;
2468         struct extent_map_tree *em_tree;
2469         struct extent_map *em;
2470         struct btrfs_device_info *devices_info;
2471         struct list_head private_devs;
2472         u64 calc_size = 1024 * 1024 * 1024;
2473         u64 min_free;
2474         u64 avail;
2475         u64 dev_offset;
2476         int num_stripes;
2477         int min_stripes;
2478         int sub_stripes;
2479         int min_devices;        /* the min number of devices we need */
2480         int i;
2481         int ret;
2482         int index;
2483
2484         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2485             (type & BTRFS_BLOCK_GROUP_DUP)) {
2486                 WARN_ON(1);
2487                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2488         }
2489         if (list_empty(&fs_devices->alloc_list))
2490                 return -ENOSPC;
2491
2492         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2493                                     &min_stripes, &sub_stripes);
2494         if (ret)
2495                 return ret;
2496
2497         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2498                                GFP_NOFS);
2499         if (!devices_info)
2500                 return -ENOMEM;
2501
2502         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2503         if (!map) {
2504                 ret = -ENOMEM;
2505                 goto error;
2506         }
2507         map->num_stripes = num_stripes;
2508
2509         cur = fs_devices->alloc_list.next;
2510         index = 0;
2511         i = 0;
2512
2513         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2514                                              num_stripes, 0);
2515
2516         if (type & BTRFS_BLOCK_GROUP_DUP) {
2517                 min_free = calc_size * 2;
2518                 min_devices = 1;
2519         } else {
2520                 min_free = calc_size;
2521                 min_devices = min_stripes;
2522         }
2523
2524         INIT_LIST_HEAD(&private_devs);
2525         while (index < num_stripes) {
2526                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2527                 BUG_ON(!device->writeable);
2528                 if (device->total_bytes > device->bytes_used)
2529                         avail = device->total_bytes - device->bytes_used;
2530                 else
2531                         avail = 0;
2532                 cur = cur->next;
2533
2534                 if (device->in_fs_metadata && avail >= min_free) {
2535                         ret = find_free_dev_extent(trans, device, min_free,
2536                                                    &devices_info[i].dev_offset,
2537                                                    &devices_info[i].max_avail);
2538                         if (ret == 0) {
2539                                 list_move_tail(&device->dev_alloc_list,
2540                                                &private_devs);
2541                                 map->stripes[index].dev = device;
2542                                 map->stripes[index].physical =
2543                                                 devices_info[i].dev_offset;
2544                                 index++;
2545                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2546                                         map->stripes[index].dev = device;
2547                                         map->stripes[index].physical =
2548                                                 devices_info[i].dev_offset +
2549                                                 calc_size;
2550                                         index++;
2551                                 }
2552                         } else if (ret != -ENOSPC)
2553                                 goto error;
2554
2555                         devices_info[i].dev = device;
2556                         i++;
2557                 } else if (device->in_fs_metadata &&
2558                            avail >= BTRFS_STRIPE_LEN) {
2559                         devices_info[i].dev = device;
2560                         devices_info[i].max_avail = avail;
2561                         i++;
2562                 }
2563
2564                 if (cur == &fs_devices->alloc_list)
2565                         break;
2566         }
2567
2568         list_splice(&private_devs, &fs_devices->alloc_list);
2569         if (index < num_stripes) {
2570                 if (index >= min_stripes) {
2571                         num_stripes = index;
2572                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2573                                 num_stripes /= sub_stripes;
2574                                 num_stripes *= sub_stripes;
2575                         }
2576
2577                         map = __shrink_map_lookup_stripes(map, num_stripes);
2578                 } else if (i >= min_devices) {
2579                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2580                                                        devices_info, i, type,
2581                                                        &map, min_stripes,
2582                                                        &calc_size);
2583                         if (ret)
2584                                 goto error;
2585                 } else {
2586                         ret = -ENOSPC;
2587                         goto error;
2588                 }
2589         }
2590         map->sector_size = extent_root->sectorsize;
2591         map->stripe_len = BTRFS_STRIPE_LEN;
2592         map->io_align = BTRFS_STRIPE_LEN;
2593         map->io_width = BTRFS_STRIPE_LEN;
2594         map->type = type;
2595         map->sub_stripes = sub_stripes;
2596
2597         *map_ret = map;
2598         *stripe_size = calc_size;
2599         *num_bytes = chunk_bytes_by_type(type, calc_size,
2600                                          map->num_stripes, sub_stripes);
2601
2602         em = alloc_extent_map(GFP_NOFS);
2603         if (!em) {
2604                 ret = -ENOMEM;
2605                 goto error;
2606         }
2607         em->bdev = (struct block_device *)map;
2608         em->start = start;
2609         em->len = *num_bytes;
2610         em->block_start = 0;
2611         em->block_len = em->len;
2612
2613         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2614         write_lock(&em_tree->lock);
2615         ret = add_extent_mapping(em_tree, em);
2616         write_unlock(&em_tree->lock);
2617         BUG_ON(ret);
2618         free_extent_map(em);
2619
2620         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2621                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2622                                      start, *num_bytes);
2623         BUG_ON(ret);
2624
2625         index = 0;
2626         while (index < map->num_stripes) {
2627                 device = map->stripes[index].dev;
2628                 dev_offset = map->stripes[index].physical;
2629
2630                 ret = btrfs_alloc_dev_extent(trans, device,
2631                                 info->chunk_root->root_key.objectid,
2632                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2633                                 start, dev_offset, calc_size);
2634                 BUG_ON(ret);
2635                 index++;
2636         }
2637
2638         kfree(devices_info);
2639         return 0;
2640
2641 error:
2642         kfree(map);
2643         kfree(devices_info);
2644         return ret;
2645 }
2646
2647 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2648                                 struct btrfs_root *extent_root,
2649                                 struct map_lookup *map, u64 chunk_offset,
2650                                 u64 chunk_size, u64 stripe_size)
2651 {
2652         u64 dev_offset;
2653         struct btrfs_key key;
2654         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2655         struct btrfs_device *device;
2656         struct btrfs_chunk *chunk;
2657         struct btrfs_stripe *stripe;
2658         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2659         int index = 0;
2660         int ret;
2661
2662         chunk = kzalloc(item_size, GFP_NOFS);
2663         if (!chunk)
2664                 return -ENOMEM;
2665
2666         index = 0;
2667         while (index < map->num_stripes) {
2668                 device = map->stripes[index].dev;
2669                 device->bytes_used += stripe_size;
2670                 ret = btrfs_update_device(trans, device);
2671                 BUG_ON(ret);
2672                 index++;
2673         }
2674
2675         index = 0;
2676         stripe = &chunk->stripe;
2677         while (index < map->num_stripes) {
2678                 device = map->stripes[index].dev;
2679                 dev_offset = map->stripes[index].physical;
2680
2681                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2682                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2683                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2684                 stripe++;
2685                 index++;
2686         }
2687
2688         btrfs_set_stack_chunk_length(chunk, chunk_size);
2689         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2690         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2691         btrfs_set_stack_chunk_type(chunk, map->type);
2692         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2693         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2694         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2695         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2696         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2697
2698         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2699         key.type = BTRFS_CHUNK_ITEM_KEY;
2700         key.offset = chunk_offset;
2701
2702         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2703         BUG_ON(ret);
2704
2705         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2706                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2707                                              item_size);
2708                 BUG_ON(ret);
2709         }
2710         kfree(chunk);
2711         return 0;
2712 }
2713
2714 /*
2715  * Chunk allocation falls into two parts. The first part does works
2716  * that make the new allocated chunk useable, but not do any operation
2717  * that modifies the chunk tree. The second part does the works that
2718  * require modifying the chunk tree. This division is important for the
2719  * bootstrap process of adding storage to a seed btrfs.
2720  */
2721 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2722                       struct btrfs_root *extent_root, u64 type)
2723 {
2724         u64 chunk_offset;
2725         u64 chunk_size;
2726         u64 stripe_size;
2727         struct map_lookup *map;
2728         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2729         int ret;
2730
2731         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2732                               &chunk_offset);
2733         if (ret)
2734                 return ret;
2735
2736         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2737                                   &stripe_size, chunk_offset, type);
2738         if (ret)
2739                 return ret;
2740
2741         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2742                                    chunk_size, stripe_size);
2743         BUG_ON(ret);
2744         return 0;
2745 }
2746
2747 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2748                                          struct btrfs_root *root,
2749                                          struct btrfs_device *device)
2750 {
2751         u64 chunk_offset;
2752         u64 sys_chunk_offset;
2753         u64 chunk_size;
2754         u64 sys_chunk_size;
2755         u64 stripe_size;
2756         u64 sys_stripe_size;
2757         u64 alloc_profile;
2758         struct map_lookup *map;
2759         struct map_lookup *sys_map;
2760         struct btrfs_fs_info *fs_info = root->fs_info;
2761         struct btrfs_root *extent_root = fs_info->extent_root;
2762         int ret;
2763
2764         ret = find_next_chunk(fs_info->chunk_root,
2765                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2766         BUG_ON(ret);
2767
2768         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2769                         (fs_info->metadata_alloc_profile &
2770                          fs_info->avail_metadata_alloc_bits);
2771         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2772
2773         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2774                                   &stripe_size, chunk_offset, alloc_profile);
2775         BUG_ON(ret);
2776
2777         sys_chunk_offset = chunk_offset + chunk_size;
2778
2779         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2780                         (fs_info->system_alloc_profile &
2781                          fs_info->avail_system_alloc_bits);
2782         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2783
2784         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2785                                   &sys_chunk_size, &sys_stripe_size,
2786                                   sys_chunk_offset, alloc_profile);
2787         BUG_ON(ret);
2788
2789         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2790         BUG_ON(ret);
2791
2792         /*
2793          * Modifying chunk tree needs allocating new blocks from both
2794          * system block group and metadata block group. So we only can
2795          * do operations require modifying the chunk tree after both
2796          * block groups were created.
2797          */
2798         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2799                                    chunk_size, stripe_size);
2800         BUG_ON(ret);
2801
2802         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2803                                    sys_chunk_offset, sys_chunk_size,
2804                                    sys_stripe_size);
2805         BUG_ON(ret);
2806         return 0;
2807 }
2808
2809 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2810 {
2811         struct extent_map *em;
2812         struct map_lookup *map;
2813         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2814         int readonly = 0;
2815         int i;
2816
2817         read_lock(&map_tree->map_tree.lock);
2818         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2819         read_unlock(&map_tree->map_tree.lock);
2820         if (!em)
2821                 return 1;
2822
2823         if (btrfs_test_opt(root, DEGRADED)) {
2824                 free_extent_map(em);
2825                 return 0;
2826         }
2827
2828         map = (struct map_lookup *)em->bdev;
2829         for (i = 0; i < map->num_stripes; i++) {
2830                 if (!map->stripes[i].dev->writeable) {
2831                         readonly = 1;
2832                         break;
2833                 }
2834         }
2835         free_extent_map(em);
2836         return readonly;
2837 }
2838
2839 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2840 {
2841         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2842 }
2843
2844 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2845 {
2846         struct extent_map *em;
2847
2848         while (1) {
2849                 write_lock(&tree->map_tree.lock);
2850                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2851                 if (em)
2852                         remove_extent_mapping(&tree->map_tree, em);
2853                 write_unlock(&tree->map_tree.lock);
2854                 if (!em)
2855                         break;
2856                 kfree(em->bdev);
2857                 /* once for us */
2858                 free_extent_map(em);
2859                 /* once for the tree */
2860                 free_extent_map(em);
2861         }
2862 }
2863
2864 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2865 {
2866         struct extent_map *em;
2867         struct map_lookup *map;
2868         struct extent_map_tree *em_tree = &map_tree->map_tree;
2869         int ret;
2870
2871         read_lock(&em_tree->lock);
2872         em = lookup_extent_mapping(em_tree, logical, len);
2873         read_unlock(&em_tree->lock);
2874         BUG_ON(!em);
2875
2876         BUG_ON(em->start > logical || em->start + em->len < logical);
2877         map = (struct map_lookup *)em->bdev;
2878         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2879                 ret = map->num_stripes;
2880         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2881                 ret = map->sub_stripes;
2882         else
2883                 ret = 1;
2884         free_extent_map(em);
2885         return ret;
2886 }
2887
2888 static int find_live_mirror(struct map_lookup *map, int first, int num,
2889                             int optimal)
2890 {
2891         int i;
2892         if (map->stripes[optimal].dev->bdev)
2893                 return optimal;
2894         for (i = first; i < first + num; i++) {
2895                 if (map->stripes[i].dev->bdev)
2896                         return i;
2897         }
2898         /* we couldn't find one that doesn't fail.  Just return something
2899          * and the io error handling code will clean up eventually
2900          */
2901         return optimal;
2902 }
2903
2904 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2905                              u64 logical, u64 *length,
2906                              struct btrfs_multi_bio **multi_ret,
2907                              int mirror_num)
2908 {
2909         struct extent_map *em;
2910         struct map_lookup *map;
2911         struct extent_map_tree *em_tree = &map_tree->map_tree;
2912         u64 offset;
2913         u64 stripe_offset;
2914         u64 stripe_nr;
2915         int stripes_allocated = 8;
2916         int stripes_required = 1;
2917         int stripe_index;
2918         int i;
2919         int num_stripes;
2920         int max_errors = 0;
2921         struct btrfs_multi_bio *multi = NULL;
2922
2923         if (multi_ret && !(rw & REQ_WRITE))
2924                 stripes_allocated = 1;
2925 again:
2926         if (multi_ret) {
2927                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2928                                 GFP_NOFS);
2929                 if (!multi)
2930                         return -ENOMEM;
2931
2932                 atomic_set(&multi->error, 0);
2933         }
2934
2935         read_lock(&em_tree->lock);
2936         em = lookup_extent_mapping(em_tree, logical, *length);
2937         read_unlock(&em_tree->lock);
2938
2939         if (!em) {
2940                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2941                        (unsigned long long)logical,
2942                        (unsigned long long)*length);
2943                 BUG();
2944         }
2945
2946         BUG_ON(em->start > logical || em->start + em->len < logical);
2947         map = (struct map_lookup *)em->bdev;
2948         offset = logical - em->start;
2949
2950         if (mirror_num > map->num_stripes)
2951                 mirror_num = 0;
2952
2953         /* if our multi bio struct is too small, back off and try again */
2954         if (rw & REQ_WRITE) {
2955                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2956                                  BTRFS_BLOCK_GROUP_DUP)) {
2957                         stripes_required = map->num_stripes;
2958                         max_errors = 1;
2959                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2960                         stripes_required = map->sub_stripes;
2961                         max_errors = 1;
2962                 }
2963         }
2964         if (multi_ret && (rw & REQ_WRITE) &&
2965             stripes_allocated < stripes_required) {
2966                 stripes_allocated = map->num_stripes;
2967                 free_extent_map(em);
2968                 kfree(multi);
2969                 goto again;
2970         }
2971         stripe_nr = offset;
2972         /*
2973          * stripe_nr counts the total number of stripes we have to stride
2974          * to get to this block
2975          */
2976         do_div(stripe_nr, map->stripe_len);
2977
2978         stripe_offset = stripe_nr * map->stripe_len;
2979         BUG_ON(offset < stripe_offset);
2980
2981         /* stripe_offset is the offset of this block in its stripe*/
2982         stripe_offset = offset - stripe_offset;
2983
2984         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2985                          BTRFS_BLOCK_GROUP_RAID10 |
2986                          BTRFS_BLOCK_GROUP_DUP)) {
2987                 /* we limit the length of each bio to what fits in a stripe */
2988                 *length = min_t(u64, em->len - offset,
2989                               map->stripe_len - stripe_offset);
2990         } else {
2991                 *length = em->len - offset;
2992         }
2993
2994         if (!multi_ret)
2995                 goto out;
2996
2997         num_stripes = 1;
2998         stripe_index = 0;
2999         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3000                 if (rw & REQ_WRITE)
3001                         num_stripes = map->num_stripes;
3002                 else if (mirror_num)
3003                         stripe_index = mirror_num - 1;
3004                 else {
3005                         stripe_index = find_live_mirror(map, 0,
3006                                             map->num_stripes,
3007                                             current->pid % map->num_stripes);
3008                 }
3009
3010         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3011                 if (rw & REQ_WRITE)
3012                         num_stripes = map->num_stripes;
3013                 else if (mirror_num)
3014                         stripe_index = mirror_num - 1;
3015
3016         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3017                 int factor = map->num_stripes / map->sub_stripes;
3018
3019                 stripe_index = do_div(stripe_nr, factor);
3020                 stripe_index *= map->sub_stripes;
3021
3022                 if (rw & REQ_WRITE)
3023                         num_stripes = map->sub_stripes;
3024                 else if (mirror_num)
3025                         stripe_index += mirror_num - 1;
3026                 else {
3027                         stripe_index = find_live_mirror(map, stripe_index,
3028                                               map->sub_stripes, stripe_index +
3029                                               current->pid % map->sub_stripes);
3030                 }
3031         } else {
3032                 /*
3033                  * after this do_div call, stripe_nr is the number of stripes
3034                  * on this device we have to walk to find the data, and
3035                  * stripe_index is the number of our device in the stripe array
3036                  */
3037                 stripe_index = do_div(stripe_nr, map->num_stripes);
3038         }
3039         BUG_ON(stripe_index >= map->num_stripes);
3040
3041         for (i = 0; i < num_stripes; i++) {
3042                 multi->stripes[i].physical =
3043                         map->stripes[stripe_index].physical +
3044                         stripe_offset + stripe_nr * map->stripe_len;
3045                 multi->stripes[i].dev = map->stripes[stripe_index].dev;
3046                 stripe_index++;
3047         }
3048         if (multi_ret) {
3049                 *multi_ret = multi;
3050                 multi->num_stripes = num_stripes;
3051                 multi->max_errors = max_errors;
3052         }
3053 out:
3054         free_extent_map(em);
3055         return 0;
3056 }
3057
3058 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3059                       u64 logical, u64 *length,
3060                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3061 {
3062         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3063                                  mirror_num);
3064 }
3065
3066 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3067                      u64 chunk_start, u64 physical, u64 devid,
3068                      u64 **logical, int *naddrs, int *stripe_len)
3069 {
3070         struct extent_map_tree *em_tree = &map_tree->map_tree;
3071         struct extent_map *em;
3072         struct map_lookup *map;
3073         u64 *buf;
3074         u64 bytenr;
3075         u64 length;
3076         u64 stripe_nr;
3077         int i, j, nr = 0;
3078
3079         read_lock(&em_tree->lock);
3080         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3081         read_unlock(&em_tree->lock);
3082
3083         BUG_ON(!em || em->start != chunk_start);
3084         map = (struct map_lookup *)em->bdev;
3085
3086         length = em->len;
3087         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3088                 do_div(length, map->num_stripes / map->sub_stripes);
3089         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3090                 do_div(length, map->num_stripes);
3091
3092         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3093         BUG_ON(!buf);
3094
3095         for (i = 0; i < map->num_stripes; i++) {
3096                 if (devid && map->stripes[i].dev->devid != devid)
3097                         continue;
3098                 if (map->stripes[i].physical > physical ||
3099                     map->stripes[i].physical + length <= physical)
3100                         continue;
3101
3102                 stripe_nr = physical - map->stripes[i].physical;
3103                 do_div(stripe_nr, map->stripe_len);
3104
3105                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3106                         stripe_nr = stripe_nr * map->num_stripes + i;
3107                         do_div(stripe_nr, map->sub_stripes);
3108                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3109                         stripe_nr = stripe_nr * map->num_stripes + i;
3110                 }
3111                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3112                 WARN_ON(nr >= map->num_stripes);
3113                 for (j = 0; j < nr; j++) {
3114                         if (buf[j] == bytenr)
3115                                 break;
3116                 }
3117                 if (j == nr) {
3118                         WARN_ON(nr >= map->num_stripes);
3119                         buf[nr++] = bytenr;
3120                 }
3121         }
3122
3123         *logical = buf;
3124         *naddrs = nr;
3125         *stripe_len = map->stripe_len;
3126
3127         free_extent_map(em);
3128         return 0;
3129 }
3130
3131 static void end_bio_multi_stripe(struct bio *bio, int err)
3132 {
3133         struct btrfs_multi_bio *multi = bio->bi_private;
3134         int is_orig_bio = 0;
3135
3136         if (err)
3137                 atomic_inc(&multi->error);
3138
3139         if (bio == multi->orig_bio)
3140                 is_orig_bio = 1;
3141
3142         if (atomic_dec_and_test(&multi->stripes_pending)) {
3143                 if (!is_orig_bio) {
3144                         bio_put(bio);
3145                         bio = multi->orig_bio;
3146                 }
3147                 bio->bi_private = multi->private;
3148                 bio->bi_end_io = multi->end_io;
3149                 /* only send an error to the higher layers if it is
3150                  * beyond the tolerance of the multi-bio
3151                  */
3152                 if (atomic_read(&multi->error) > multi->max_errors) {
3153                         err = -EIO;
3154                 } else if (err) {
3155                         /*
3156                          * this bio is actually up to date, we didn't
3157                          * go over the max number of errors
3158                          */
3159                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3160                         err = 0;
3161                 }
3162                 kfree(multi);
3163
3164                 bio_endio(bio, err);
3165         } else if (!is_orig_bio) {
3166                 bio_put(bio);
3167         }
3168 }
3169
3170 struct async_sched {
3171         struct bio *bio;
3172         int rw;
3173         struct btrfs_fs_info *info;
3174         struct btrfs_work work;
3175 };
3176
3177 /*
3178  * see run_scheduled_bios for a description of why bios are collected for
3179  * async submit.
3180  *
3181  * This will add one bio to the pending list for a device and make sure
3182  * the work struct is scheduled.
3183  */
3184 static noinline int schedule_bio(struct btrfs_root *root,
3185                                  struct btrfs_device *device,
3186                                  int rw, struct bio *bio)
3187 {
3188         int should_queue = 1;
3189         struct btrfs_pending_bios *pending_bios;
3190
3191         /* don't bother with additional async steps for reads, right now */
3192         if (!(rw & REQ_WRITE)) {
3193                 bio_get(bio);
3194                 submit_bio(rw, bio);
3195                 bio_put(bio);
3196                 return 0;
3197         }
3198
3199         /*
3200          * nr_async_bios allows us to reliably return congestion to the
3201          * higher layers.  Otherwise, the async bio makes it appear we have
3202          * made progress against dirty pages when we've really just put it
3203          * on a queue for later
3204          */
3205         atomic_inc(&root->fs_info->nr_async_bios);
3206         WARN_ON(bio->bi_next);
3207         bio->bi_next = NULL;
3208         bio->bi_rw |= rw;
3209
3210         spin_lock(&device->io_lock);
3211         if (bio->bi_rw & REQ_SYNC)
3212                 pending_bios = &device->pending_sync_bios;
3213         else
3214                 pending_bios = &device->pending_bios;
3215
3216         if (pending_bios->tail)
3217                 pending_bios->tail->bi_next = bio;
3218
3219         pending_bios->tail = bio;
3220         if (!pending_bios->head)
3221                 pending_bios->head = bio;
3222         if (device->running_pending)
3223                 should_queue = 0;
3224
3225         spin_unlock(&device->io_lock);
3226
3227         if (should_queue)
3228                 btrfs_queue_worker(&root->fs_info->submit_workers,
3229                                    &device->work);
3230         return 0;
3231 }
3232
3233 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3234                   int mirror_num, int async_submit)
3235 {
3236         struct btrfs_mapping_tree *map_tree;
3237         struct btrfs_device *dev;
3238         struct bio *first_bio = bio;
3239         u64 logical = (u64)bio->bi_sector << 9;
3240         u64 length = 0;
3241         u64 map_length;
3242         struct btrfs_multi_bio *multi = NULL;
3243         int ret;
3244         int dev_nr = 0;
3245         int total_devs = 1;
3246
3247         length = bio->bi_size;
3248         map_tree = &root->fs_info->mapping_tree;
3249         map_length = length;
3250
3251         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3252                               mirror_num);
3253         BUG_ON(ret);
3254
3255         total_devs = multi->num_stripes;
3256         if (map_length < length) {
3257                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3258                        "len %llu\n", (unsigned long long)logical,
3259                        (unsigned long long)length,
3260                        (unsigned long long)map_length);
3261                 BUG();
3262         }
3263         multi->end_io = first_bio->bi_end_io;
3264         multi->private = first_bio->bi_private;
3265         multi->orig_bio = first_bio;
3266         atomic_set(&multi->stripes_pending, multi->num_stripes);
3267
3268         while (dev_nr < total_devs) {
3269                 if (total_devs > 1) {
3270                         if (dev_nr < total_devs - 1) {
3271                                 bio = bio_clone(first_bio, GFP_NOFS);
3272                                 BUG_ON(!bio);
3273                         } else {
3274                                 bio = first_bio;
3275                         }
3276                         bio->bi_private = multi;
3277                         bio->bi_end_io = end_bio_multi_stripe;
3278                 }
3279                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3280                 dev = multi->stripes[dev_nr].dev;
3281                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3282                         bio->bi_bdev = dev->bdev;
3283                         if (async_submit)
3284                                 schedule_bio(root, dev, rw, bio);
3285                         else
3286                                 submit_bio(rw, bio);
3287                 } else {
3288                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3289                         bio->bi_sector = logical >> 9;
3290                         bio_endio(bio, -EIO);
3291                 }
3292                 dev_nr++;
3293         }
3294         if (total_devs == 1)
3295                 kfree(multi);
3296         return 0;
3297 }
3298
3299 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3300                                        u8 *uuid, u8 *fsid)
3301 {
3302         struct btrfs_device *device;
3303         struct btrfs_fs_devices *cur_devices;
3304
3305         cur_devices = root->fs_info->fs_devices;
3306         while (cur_devices) {
3307                 if (!fsid ||
3308                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3309                         device = __find_device(&cur_devices->devices,
3310                                                devid, uuid);
3311                         if (device)
3312                                 return device;
3313                 }
3314                 cur_devices = cur_devices->seed;
3315         }
3316         return NULL;
3317 }
3318
3319 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3320                                             u64 devid, u8 *dev_uuid)
3321 {
3322         struct btrfs_device *device;
3323         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3324
3325         device = kzalloc(sizeof(*device), GFP_NOFS);
3326         if (!device)
3327                 return NULL;
3328         list_add(&device->dev_list,
3329                  &fs_devices->devices);
3330         device->dev_root = root->fs_info->dev_root;
3331         device->devid = devid;
3332         device->work.func = pending_bios_fn;
3333         device->fs_devices = fs_devices;
3334         device->missing = 1;
3335         fs_devices->num_devices++;
3336         fs_devices->missing_devices++;
3337         spin_lock_init(&device->io_lock);
3338         INIT_LIST_HEAD(&device->dev_alloc_list);
3339         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3340         return device;
3341 }
3342
3343 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3344                           struct extent_buffer *leaf,
3345                           struct btrfs_chunk *chunk)
3346 {
3347         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3348         struct map_lookup *map;
3349         struct extent_map *em;
3350         u64 logical;
3351         u64 length;
3352         u64 devid;
3353         u8 uuid[BTRFS_UUID_SIZE];
3354         int num_stripes;
3355         int ret;
3356         int i;
3357
3358         logical = key->offset;
3359         length = btrfs_chunk_length(leaf, chunk);
3360
3361         read_lock(&map_tree->map_tree.lock);
3362         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3363         read_unlock(&map_tree->map_tree.lock);
3364
3365         /* already mapped? */
3366         if (em && em->start <= logical && em->start + em->len > logical) {
3367                 free_extent_map(em);
3368                 return 0;
3369         } else if (em) {
3370                 free_extent_map(em);
3371         }
3372
3373         em = alloc_extent_map(GFP_NOFS);
3374         if (!em)
3375                 return -ENOMEM;
3376         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3377         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3378         if (!map) {
3379                 free_extent_map(em);
3380                 return -ENOMEM;
3381         }
3382
3383         em->bdev = (struct block_device *)map;
3384         em->start = logical;
3385         em->len = length;
3386         em->block_start = 0;
3387         em->block_len = em->len;
3388
3389         map->num_stripes = num_stripes;
3390         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3391         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3392         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3393         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3394         map->type = btrfs_chunk_type(leaf, chunk);
3395         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3396         for (i = 0; i < num_stripes; i++) {
3397                 map->stripes[i].physical =
3398                         btrfs_stripe_offset_nr(leaf, chunk, i);
3399                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3400                 read_extent_buffer(leaf, uuid, (unsigned long)
3401                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3402                                    BTRFS_UUID_SIZE);
3403                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3404                                                         NULL);
3405                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3406                         kfree(map);
3407                         free_extent_map(em);
3408                         return -EIO;
3409                 }
3410                 if (!map->stripes[i].dev) {
3411                         map->stripes[i].dev =
3412                                 add_missing_dev(root, devid, uuid);
3413                         if (!map->stripes[i].dev) {
3414                                 kfree(map);
3415                                 free_extent_map(em);
3416                                 return -EIO;
3417                         }
3418                 }
3419                 map->stripes[i].dev->in_fs_metadata = 1;
3420         }
3421
3422         write_lock(&map_tree->map_tree.lock);
3423         ret = add_extent_mapping(&map_tree->map_tree, em);
3424         write_unlock(&map_tree->map_tree.lock);
3425         BUG_ON(ret);
3426         free_extent_map(em);
3427
3428         return 0;
3429 }
3430
3431 static int fill_device_from_item(struct extent_buffer *leaf,
3432                                  struct btrfs_dev_item *dev_item,
3433                                  struct btrfs_device *device)
3434 {
3435         unsigned long ptr;
3436
3437         device->devid = btrfs_device_id(leaf, dev_item);
3438         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3439         device->total_bytes = device->disk_total_bytes;
3440         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3441         device->type = btrfs_device_type(leaf, dev_item);
3442         device->io_align = btrfs_device_io_align(leaf, dev_item);
3443         device->io_width = btrfs_device_io_width(leaf, dev_item);
3444         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3445
3446         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3447         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3448
3449         return 0;
3450 }
3451
3452 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3453 {
3454         struct btrfs_fs_devices *fs_devices;
3455         int ret;
3456
3457         mutex_lock(&uuid_mutex);
3458
3459         fs_devices = root->fs_info->fs_devices->seed;
3460         while (fs_devices) {
3461                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3462                         ret = 0;
3463                         goto out;
3464                 }
3465                 fs_devices = fs_devices->seed;
3466         }
3467
3468         fs_devices = find_fsid(fsid);
3469         if (!fs_devices) {
3470                 ret = -ENOENT;
3471                 goto out;
3472         }
3473
3474         fs_devices = clone_fs_devices(fs_devices);
3475         if (IS_ERR(fs_devices)) {
3476                 ret = PTR_ERR(fs_devices);
3477                 goto out;
3478         }
3479
3480         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3481                                    root->fs_info->bdev_holder);
3482         if (ret)
3483                 goto out;
3484
3485         if (!fs_devices->seeding) {
3486                 __btrfs_close_devices(fs_devices);
3487                 free_fs_devices(fs_devices);
3488                 ret = -EINVAL;
3489                 goto out;
3490         }
3491
3492         fs_devices->seed = root->fs_info->fs_devices->seed;
3493         root->fs_info->fs_devices->seed = fs_devices;
3494 out:
3495         mutex_unlock(&uuid_mutex);
3496         return ret;
3497 }
3498
3499 static int read_one_dev(struct btrfs_root *root,
3500                         struct extent_buffer *leaf,
3501                         struct btrfs_dev_item *dev_item)
3502 {
3503         struct btrfs_device *device;
3504         u64 devid;
3505         int ret;
3506         u8 fs_uuid[BTRFS_UUID_SIZE];
3507         u8 dev_uuid[BTRFS_UUID_SIZE];
3508
3509         devid = btrfs_device_id(leaf, dev_item);
3510         read_extent_buffer(leaf, dev_uuid,
3511                            (unsigned long)btrfs_device_uuid(dev_item),
3512                            BTRFS_UUID_SIZE);
3513         read_extent_buffer(leaf, fs_uuid,
3514                            (unsigned long)btrfs_device_fsid(dev_item),
3515                            BTRFS_UUID_SIZE);
3516
3517         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3518                 ret = open_seed_devices(root, fs_uuid);
3519                 if (ret && !btrfs_test_opt(root, DEGRADED))
3520                         return ret;
3521         }
3522
3523         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3524         if (!device || !device->bdev) {
3525                 if (!btrfs_test_opt(root, DEGRADED))
3526                         return -EIO;
3527
3528                 if (!device) {
3529                         printk(KERN_WARNING "warning devid %llu missing\n",
3530                                (unsigned long long)devid);
3531                         device = add_missing_dev(root, devid, dev_uuid);
3532                         if (!device)
3533                                 return -ENOMEM;
3534                 } else if (!device->missing) {
3535                         /*
3536                          * this happens when a device that was properly setup
3537                          * in the device info lists suddenly goes bad.
3538                          * device->bdev is NULL, and so we have to set
3539                          * device->missing to one here
3540                          */
3541                         root->fs_info->fs_devices->missing_devices++;
3542                         device->missing = 1;
3543                 }
3544         }
3545
3546         if (device->fs_devices != root->fs_info->fs_devices) {
3547                 BUG_ON(device->writeable);
3548                 if (device->generation !=
3549                     btrfs_device_generation(leaf, dev_item))
3550                         return -EINVAL;
3551         }
3552
3553         fill_device_from_item(leaf, dev_item, device);
3554         device->dev_root = root->fs_info->dev_root;
3555         device->in_fs_metadata = 1;
3556         if (device->writeable)
3557                 device->fs_devices->total_rw_bytes += device->total_bytes;
3558         ret = 0;
3559         return ret;
3560 }
3561
3562 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3563 {
3564         struct btrfs_dev_item *dev_item;
3565
3566         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3567                                                      dev_item);
3568         return read_one_dev(root, buf, dev_item);
3569 }
3570
3571 int btrfs_read_sys_array(struct btrfs_root *root)
3572 {
3573         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3574         struct extent_buffer *sb;
3575         struct btrfs_disk_key *disk_key;
3576         struct btrfs_chunk *chunk;
3577         u8 *ptr;
3578         unsigned long sb_ptr;
3579         int ret = 0;
3580         u32 num_stripes;
3581         u32 array_size;
3582         u32 len = 0;
3583         u32 cur;
3584         struct btrfs_key key;
3585
3586         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3587                                           BTRFS_SUPER_INFO_SIZE);
3588         if (!sb)
3589                 return -ENOMEM;
3590         btrfs_set_buffer_uptodate(sb);
3591         btrfs_set_buffer_lockdep_class(sb, 0);
3592
3593         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3594         array_size = btrfs_super_sys_array_size(super_copy);
3595
3596         ptr = super_copy->sys_chunk_array;
3597         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3598         cur = 0;
3599
3600         while (cur < array_size) {
3601                 disk_key = (struct btrfs_disk_key *)ptr;
3602                 btrfs_disk_key_to_cpu(&key, disk_key);
3603
3604                 len = sizeof(*disk_key); ptr += len;
3605                 sb_ptr += len;
3606                 cur += len;
3607
3608                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3609                         chunk = (struct btrfs_chunk *)sb_ptr;
3610                         ret = read_one_chunk(root, &key, sb, chunk);
3611                         if (ret)
3612                                 break;
3613                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3614                         len = btrfs_chunk_item_size(num_stripes);
3615                 } else {
3616                         ret = -EIO;
3617                         break;
3618                 }
3619                 ptr += len;
3620                 sb_ptr += len;
3621                 cur += len;
3622         }
3623         free_extent_buffer(sb);
3624         return ret;
3625 }
3626
3627 int btrfs_read_chunk_tree(struct btrfs_root *root)
3628 {
3629         struct btrfs_path *path;
3630         struct extent_buffer *leaf;
3631         struct btrfs_key key;
3632         struct btrfs_key found_key;
3633         int ret;
3634         int slot;
3635
3636         root = root->fs_info->chunk_root;
3637
3638         path = btrfs_alloc_path();
3639         if (!path)
3640                 return -ENOMEM;
3641
3642         /* first we search for all of the device items, and then we
3643          * read in all of the chunk items.  This way we can create chunk
3644          * mappings that reference all of the devices that are afound
3645          */
3646         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3647         key.offset = 0;
3648         key.type = 0;
3649 again:
3650         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3651         if (ret < 0)
3652                 goto error;
3653         while (1) {
3654                 leaf = path->nodes[0];
3655                 slot = path->slots[0];
3656                 if (slot >= btrfs_header_nritems(leaf)) {
3657                         ret = btrfs_next_leaf(root, path);
3658                         if (ret == 0)
3659                                 continue;
3660                         if (ret < 0)
3661                                 goto error;
3662                         break;
3663                 }
3664                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3665                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3666                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3667                                 break;
3668                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3669                                 struct btrfs_dev_item *dev_item;
3670                                 dev_item = btrfs_item_ptr(leaf, slot,
3671                                                   struct btrfs_dev_item);
3672                                 ret = read_one_dev(root, leaf, dev_item);
3673                                 if (ret)
3674                                         goto error;
3675                         }
3676                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3677                         struct btrfs_chunk *chunk;
3678                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3679                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3680                         if (ret)
3681                                 goto error;
3682                 }
3683                 path->slots[0]++;
3684         }
3685         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3686                 key.objectid = 0;
3687                 btrfs_release_path(root, path);
3688                 goto again;
3689         }
3690         ret = 0;
3691 error:
3692         btrfs_free_path(path);
3693         return ret;
3694 }