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 num_sync_run;
 166         unsigned long batch_run = 0;
 167         unsigned long limit;
 168         unsigned long last_waited = 0;
 169         int force_reg = 0;
 170
 171         bdi = blk_get_backing_dev_info(device->bdev);
 172         fs_info = device->dev_root->fs_info;
 173         limit = btrfs_async_submit_limit(fs_info);
 174         limit = limit * 2 / 3;
 175
 176         /* we want to make sure that every time we switch from the sync
 177          * list to the normal list, we unplug
 178          */
 179         num_sync_run = 0;
 180
 181 loop:
 182         spin_lock(&device->io_lock);
 183
 184 loop_lock:
 185         num_run = 0;
 186
 187         /* take all the bios off the list at once and process them
 188          * later on (without the lock held).  But, remember the
 189          * tail and other pointers so the bios can be properly reinserted
 190          * into the list if we hit congestion
 191          */
 192         if (!force_reg && device->pending_sync_bios.head) {
 193                 pending_bios = &device->pending_sync_bios;
 194                 force_reg = 1;
 195         } else {
 196                 pending_bios = &device->pending_bios;
 197                 force_reg = 0;
 198         }
 199
 200         pending = pending_bios->head;
 201         tail = pending_bios->tail;
 202         WARN_ON(pending && !tail);
 203
 204         /*
 205          * if pending was null this time around, no bios need processing
 206          * at all and we can stop.  Otherwise it'll loop back up again
 207          * and do an additional check so no bios are missed.
 208          *
 209          * device->running_pending is used to synchronize with the
 210          * schedule_bio code.
 211          */
 212         if (device->pending_sync_bios.head == NULL &&
 213             device->pending_bios.head == NULL) {
 214                 again = 0;
 215                 device->running_pending = 0;
 216         } else {
 217                 again = 1;
 218                 device->running_pending = 1;
 219         }
 220
 221         pending_bios->head = NULL;
 222         pending_bios->tail = NULL;
 223
 224         spin_unlock(&device->io_lock);
 225
 226         /*
 227          * if we're doing the regular priority list, make sure we unplug
 228          * for any high prio bios we've sent down
 229          */
 230         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
 231                 num_sync_run = 0;
 232                 blk_run_backing_dev(bdi, NULL);
 233         }
 234
 235         while (pending) {
 236
 237                 rmb();
 238                 /* we want to work on both lists, but do more bios on the
 239                  * sync list than the regular list
 240                  */
 241                 if ((num_run > 32 &&
 242                     pending_bios != &device->pending_sync_bios &&
 243                     device->pending_sync_bios.head) ||
 244                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 245                     device->pending_bios.head)) {
 246                         spin_lock(&device->io_lock);
 247                         requeue_list(pending_bios, pending, tail);
 248                         goto loop_lock;
 249                 }
 250
 251                 cur = pending;
 252                 pending = pending->bi_next;
 253                 cur->bi_next = NULL;
 254                 atomic_dec(&fs_info->nr_async_bios);
 255
 256                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 257                     waitqueue_active(&fs_info->async_submit_wait))
 258                         wake_up(&fs_info->async_submit_wait);
 259
 260                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 261
 262                 if (cur->bi_rw & REQ_SYNC)
 263                         num_sync_run++;
 264
 265                 submit_bio(cur->bi_rw, cur);
 266                 num_run++;
 267                 batch_run++;
 268                 if (need_resched()) {
 269                         if (num_sync_run) {
 270                                 blk_run_backing_dev(bdi, NULL);
 271                                 num_sync_run = 0;
 272                         }
 273                         cond_resched();
 274                 }
 275
 276                 /*
 277                  * we made progress, there is more work to do and the bdi
 278                  * is now congested.  Back off and let other work structs
 279                  * run instead
 280                  */
 281                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 282                     fs_info->fs_devices->open_devices > 1) {
 283                         struct io_context *ioc;
 284
 285                         ioc = current->io_context;
 286
 287                         /*
 288                          * the main goal here is that we don't want to
 289                          * block if we're going to be able to submit
 290                          * more requests without blocking.
 291                          *
 292                          * This code does two great things, it pokes into
 293                          * the elevator code from a filesystem _and_
 294                          * it makes assumptions about how batching works.
 295                          */
 296                         if (ioc && ioc->nr_batch_requests > 0 &&
 297                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 298                             (last_waited == 0 ||
 299                              ioc->last_waited == last_waited)) {
 300                                 /*
 301                                  * we want to go through our batch of
 302                                  * requests and stop.  So, we copy out
 303                                  * the ioc->last_waited time and test
 304                                  * against it before looping
 305                                  */
 306                                 last_waited = ioc->last_waited;
 307                                 if (need_resched()) {
 308                                         if (num_sync_run) {
 309                                                 blk_run_backing_dev(bdi, NULL);
 310                                                 num_sync_run = 0;
 311                                         }
 312                                         cond_resched();
 313                                 }
 314                                 continue;
 315                         }
 316                         spin_lock(&device->io_lock);
 317                         requeue_list(pending_bios, pending, tail);
 318                         device->running_pending = 1;
 319
 320                         spin_unlock(&device->io_lock);
 321                         btrfs_requeue_work(&device->work);
 322                         goto done;
 323                 }
 324         }
 325
 326         if (num_sync_run) {
 327                 num_sync_run = 0;
 328                 blk_run_backing_dev(bdi, NULL);
 329         }
 330         /*
 331          * IO has already been through a long path to get here.  Checksumming,
 332          * async helper threads, perhaps compression.  We've done a pretty
 333          * good job of collecting a batch of IO and should just unplug
 334          * the device right away.
 335          *
 336          * This will help anyone who is waiting on the IO, they might have
 337          * already unplugged, but managed to do so before the bio they
 338          * cared about found its way down here.
 339          */
 340         blk_run_backing_dev(bdi, NULL);
 341
 342         cond_resched();
 343         if (again)
 344                 goto loop;
 345
 346         spin_lock(&device->io_lock);
 347         if (device->pending_bios.head || device->pending_sync_bios.head)
 348                 goto loop_lock;
 349         spin_unlock(&device->io_lock);
 350
 351 done:
 352         return 0;
 353 }
 354
 355 static void pending_bios_fn(struct btrfs_work *work)
 356 {
 357         struct btrfs_device *device;
 358
 359         device = container_of(work, struct btrfs_device, work);
 360         run_scheduled_bios(device);
 361 }
 362
 363 static noinline int device_list_add(const char *path,
 364                            struct btrfs_super_block *disk_super,
 365                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 366 {
 367         struct btrfs_device *device;
 368         struct btrfs_fs_devices *fs_devices;
 369         u64 found_transid = btrfs_super_generation(disk_super);
 370         char *name;
 371
 372         fs_devices = find_fsid(disk_super->fsid);
 373         if (!fs_devices) {
 374                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 375                 if (!fs_devices)
 376                         return -ENOMEM;
 377                 INIT_LIST_HEAD(&fs_devices->devices);
 378                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 379                 list_add(&fs_devices->list, &fs_uuids);
 380                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 381                 fs_devices->latest_devid = devid;
 382                 fs_devices->latest_trans = found_transid;
 383                 mutex_init(&fs_devices->device_list_mutex);
 384                 device = NULL;
 385         } else {
 386                 device = __find_device(&fs_devices->devices, devid,
 387                                        disk_super->dev_item.uuid);
 388         }
 389         if (!device) {
 390                 if (fs_devices->opened)
 391                         return -EBUSY;
 392
 393                 device = kzalloc(sizeof(*device), GFP_NOFS);
 394                 if (!device) {
 395                         /* we can safely leave the fs_devices entry around */
 396                         return -ENOMEM;
 397                 }
 398                 device->devid = devid;
 399                 device->work.func = pending_bios_fn;
 400                 memcpy(device->uuid, disk_super->dev_item.uuid,
 401                        BTRFS_UUID_SIZE);
 402                 device->barriers = 1;
 403                 spin_lock_init(&device->io_lock);
 404                 device->name = kstrdup(path, GFP_NOFS);
 405                 if (!device->name) {
 406                         kfree(device);
 407                         return -ENOMEM;
 408                 }
 409                 INIT_LIST_HEAD(&device->dev_alloc_list);
 410
 411                 mutex_lock(&fs_devices->device_list_mutex);
 412                 list_add(&device->dev_list, &fs_devices->devices);
 413                 mutex_unlock(&fs_devices->device_list_mutex);
 414
 415                 device->fs_devices = fs_devices;
 416                 fs_devices->num_devices++;
 417         } else if (!device->name || strcmp(device->name, path)) {
 418                 name = kstrdup(path, GFP_NOFS);
 419                 if (!name)
 420                         return -ENOMEM;
 421                 kfree(device->name);
 422                 device->name = name;
 423                 if (device->missing) {
 424                         fs_devices->missing_devices--;
 425                         device->missing = 0;
 426                 }
 427         }
 428
 429         if (found_transid > fs_devices->latest_trans) {
 430                 fs_devices->latest_devid = devid;
 431                 fs_devices->latest_trans = found_transid;
 432         }
 433         *fs_devices_ret = fs_devices;
 434         return 0;
 435 }
 436
 437 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 438 {
 439         struct btrfs_fs_devices *fs_devices;
 440         struct btrfs_device *device;
 441         struct btrfs_device *orig_dev;
 442
 443         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 444         if (!fs_devices)
 445                 return ERR_PTR(-ENOMEM);
 446
 447         INIT_LIST_HEAD(&fs_devices->devices);
 448         INIT_LIST_HEAD(&fs_devices->alloc_list);
 449         INIT_LIST_HEAD(&fs_devices->list);
 450         mutex_init(&fs_devices->device_list_mutex);
 451         fs_devices->latest_devid = orig->latest_devid;
 452         fs_devices->latest_trans = orig->latest_trans;
 453         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 454
 455         mutex_lock(&orig->device_list_mutex);
 456         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 457                 device = kzalloc(sizeof(*device), GFP_NOFS);
 458                 if (!device)
 459                         goto error;
 460
 461                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 462                 if (!device->name) {
 463                         kfree(device);
 464                         goto error;
 465                 }
 466
 467                 device->devid = orig_dev->devid;
 468                 device->work.func = pending_bios_fn;
 469                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 470                 device->barriers = 1;
 471                 spin_lock_init(&device->io_lock);
 472                 INIT_LIST_HEAD(&device->dev_list);
 473                 INIT_LIST_HEAD(&device->dev_alloc_list);
 474
 475                 list_add(&device->dev_list, &fs_devices->devices);
 476                 device->fs_devices = fs_devices;
 477                 fs_devices->num_devices++;
 478         }
 479         mutex_unlock(&orig->device_list_mutex);
 480         return fs_devices;
 481 error:
 482         mutex_unlock(&orig->device_list_mutex);
 483         free_fs_devices(fs_devices);
 484         return ERR_PTR(-ENOMEM);
 485 }
 486
 487 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 488 {
 489         struct btrfs_device *device, *next;
 490
 491         mutex_lock(&uuid_mutex);
 492 again:
 493         mutex_lock(&fs_devices->device_list_mutex);
 494         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 495                 if (device->in_fs_metadata)
 496                         continue;
 497
 498                 if (device->bdev) {
 499                         close_bdev_exclusive(device->bdev, device->mode);
 500                         device->bdev = NULL;
 501                         fs_devices->open_devices--;
 502                 }
 503                 if (device->writeable) {
 504                         list_del_init(&device->dev_alloc_list);
 505                         device->writeable = 0;
 506                         fs_devices->rw_devices--;
 507                 }
 508                 list_del_init(&device->dev_list);
 509                 fs_devices->num_devices--;
 510                 kfree(device->name);
 511                 kfree(device);
 512         }
 513         mutex_unlock(&fs_devices->device_list_mutex);
 514
 515         if (fs_devices->seed) {
 516                 fs_devices = fs_devices->seed;
 517                 goto again;
 518         }
 519
 520         mutex_unlock(&uuid_mutex);
 521         return 0;
 522 }
 523
 524 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 525 {
 526         struct btrfs_device *device;
 527
 528         if (--fs_devices->opened > 0)
 529                 return 0;
 530
 531         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 532                 if (device->bdev) {
 533                         close_bdev_exclusive(device->bdev, device->mode);
 534                         fs_devices->open_devices--;
 535                 }
 536                 if (device->writeable) {
 537                         list_del_init(&device->dev_alloc_list);
 538                         fs_devices->rw_devices--;
 539                 }
 540
 541                 device->bdev = NULL;
 542                 device->writeable = 0;
 543                 device->in_fs_metadata = 0;
 544         }
 545         WARN_ON(fs_devices->open_devices);
 546         WARN_ON(fs_devices->rw_devices);
 547         fs_devices->opened = 0;
 548         fs_devices->seeding = 0;
 549
 550         return 0;
 551 }
 552
 553 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 554 {
 555         struct btrfs_fs_devices *seed_devices = NULL;
 556         int ret;
 557
 558         mutex_lock(&uuid_mutex);
 559         ret = __btrfs_close_devices(fs_devices);
 560         if (!fs_devices->opened) {
 561                 seed_devices = fs_devices->seed;
 562                 fs_devices->seed = NULL;
 563         }
 564         mutex_unlock(&uuid_mutex);
 565
 566         while (seed_devices) {
 567                 fs_devices = seed_devices;
 568                 seed_devices = fs_devices->seed;
 569                 __btrfs_close_devices(fs_devices);
 570                 free_fs_devices(fs_devices);
 571         }
 572         return ret;
 573 }
 574
 575 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 576                                 fmode_t flags, void *holder)
 577 {
 578         struct block_device *bdev;
 579         struct list_head *head = &fs_devices->devices;
 580         struct btrfs_device *device;
 581         struct block_device *latest_bdev = NULL;
 582         struct buffer_head *bh;
 583         struct btrfs_super_block *disk_super;
 584         u64 latest_devid = 0;
 585         u64 latest_transid = 0;
 586         u64 devid;
 587         int seeding = 1;
 588         int ret = 0;
 589
 590         list_for_each_entry(device, head, dev_list) {
 591                 if (device->bdev)
 592                         continue;
 593                 if (!device->name)
 594                         continue;
 595
 596                 bdev = open_bdev_exclusive(device->name, flags, holder);
 597                 if (IS_ERR(bdev)) {
 598                         printk(KERN_INFO "open %s failed\n", device->name);
 599                         goto error;
 600                 }
 601                 set_blocksize(bdev, 4096);
 602
 603                 bh = btrfs_read_dev_super(bdev);
 604                 if (!bh) {
 605                         ret = -EINVAL;
 606                         goto error_close;
 607                 }
 608
 609                 disk_super = (struct btrfs_super_block *)bh->b_data;
 610                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 611                 if (devid != device->devid)
 612                         goto error_brelse;
 613
 614                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 615                            BTRFS_UUID_SIZE))
 616                         goto error_brelse;
 617
 618                 device->generation = btrfs_super_generation(disk_super);
 619                 if (!latest_transid || device->generation > latest_transid) {
 620                         latest_devid = devid;
 621                         latest_transid = device->generation;
 622                         latest_bdev = bdev;
 623                 }
 624
 625                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 626                         device->writeable = 0;
 627                 } else {
 628                         device->writeable = !bdev_read_only(bdev);
 629                         seeding = 0;
 630                 }
 631
 632                 device->bdev = bdev;
 633                 device->in_fs_metadata = 0;
 634                 device->mode = flags;
 635
 636                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 637                         fs_devices->rotating = 1;
 638
 639                 fs_devices->open_devices++;
 640                 if (device->writeable) {
 641                         fs_devices->rw_devices++;
 642                         list_add(&device->dev_alloc_list,
 643                                  &fs_devices->alloc_list);
 644                 }
 645                 continue;
 646
 647 error_brelse:
 648                 brelse(bh);
 649 error_close:
 650                 close_bdev_exclusive(bdev, FMODE_READ);
 651 error:
 652                 continue;
 653         }
 654         if (fs_devices->open_devices == 0) {
 655                 ret = -EIO;
 656                 goto out;
 657         }
 658         fs_devices->seeding = seeding;
 659         fs_devices->opened = 1;
 660         fs_devices->latest_bdev = latest_bdev;
 661         fs_devices->latest_devid = latest_devid;
 662         fs_devices->latest_trans = latest_transid;
 663         fs_devices->total_rw_bytes = 0;
 664 out:
 665         return ret;
 666 }
 667
 668 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 669                        fmode_t flags, void *holder)
 670 {
 671         int ret;
 672
 673         mutex_lock(&uuid_mutex);
 674         if (fs_devices->opened) {
 675                 fs_devices->opened++;
 676                 ret = 0;
 677         } else {
 678                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 679         }
 680         mutex_unlock(&uuid_mutex);
 681         return ret;
 682 }
 683
 684 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 685                           struct btrfs_fs_devices **fs_devices_ret)
 686 {
 687         struct btrfs_super_block *disk_super;
 688         struct block_device *bdev;
 689         struct buffer_head *bh;
 690         int ret;
 691         u64 devid;
 692         u64 transid;
 693
 694         mutex_lock(&uuid_mutex);
 695
 696         bdev = open_bdev_exclusive(path, flags, holder);
 697
 698         if (IS_ERR(bdev)) {
 699                 ret = PTR_ERR(bdev);
 700                 goto error;
 701         }
 702
 703         ret = set_blocksize(bdev, 4096);
 704         if (ret)
 705                 goto error_close;
 706         bh = btrfs_read_dev_super(bdev);
 707         if (!bh) {
 708                 ret = -EINVAL;
 709                 goto error_close;
 710         }
 711         disk_super = (struct btrfs_super_block *)bh->b_data;
 712         devid = btrfs_stack_device_id(&disk_super->dev_item);
 713         transid = btrfs_super_generation(disk_super);
 714         if (disk_super->label[0])
 715                 printk(KERN_INFO "device label %s ", disk_super->label);
 716         else {
 717                 /* FIXME, make a readl uuid parser */
 718                 printk(KERN_INFO "device fsid %llx-%llx ",
 719                        *(unsigned long long *)disk_super->fsid,
 720                        *(unsigned long long *)(disk_super->fsid + 8));
 721         }
 722         printk(KERN_CONT "devid %llu transid %llu %s\n",
 723                (unsigned long long)devid, (unsigned long long)transid, path);
 724         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 725
 726         brelse(bh);
 727 error_close:
 728         close_bdev_exclusive(bdev, flags);
 729 error:
 730         mutex_unlock(&uuid_mutex);
 731         return ret;
 732 }
 733
 734 /* helper to account the used device space in the range */
 735 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 736                                    u64 end, u64 *length)
 737 {
 738         struct btrfs_key key;
 739         struct btrfs_root *root = device->dev_root;
 740         struct btrfs_dev_extent *dev_extent;
 741         struct btrfs_path *path;
 742         u64 extent_end;
 743         int ret;
 744         int slot;
 745         struct extent_buffer *l;
 746
 747         *length = 0;
 748
 749         if (start >= device->total_bytes)
 750                 return 0;
 751
 752         path = btrfs_alloc_path();
 753         if (!path)
 754                 return -ENOMEM;
 755         path->reada = 2;
 756
 757         key.objectid = device->devid;
 758         key.offset = start;
 759         key.type = BTRFS_DEV_EXTENT_KEY;
 760
 761         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 762         if (ret < 0)
 763                 goto out;
 764         if (ret > 0) {
 765                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 766                 if (ret < 0)
 767                         goto out;
 768         }
 769
 770         while (1) {
 771                 l = path->nodes[0];
 772                 slot = path->slots[0];
 773                 if (slot >= btrfs_header_nritems(l)) {
 774                         ret = btrfs_next_leaf(root, path);
 775                         if (ret == 0)
 776                                 continue;
 777                         if (ret < 0)
 778                                 goto out;
 779
 780                         break;
 781                 }
 782                 btrfs_item_key_to_cpu(l, &key, slot);
 783
 784                 if (key.objectid < device->devid)
 785                         goto next;
 786
 787                 if (key.objectid > device->devid)
 788                         break;
 789
 790                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 791                         goto next;
 792
 793                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 794                 extent_end = key.offset + btrfs_dev_extent_length(l,
 795                                                                   dev_extent);
 796                 if (key.offset <= start && extent_end > end) {
 797                         *length = end - start + 1;
 798                         break;
 799                 } else if (key.offset <= start && extent_end > start)
 800                         *length += extent_end - start;
 801                 else if (key.offset > start && extent_end <= end)
 802                         *length += extent_end - key.offset;
 803                 else if (key.offset > start && key.offset <= end) {
 804                         *length += end - key.offset + 1;
 805                         break;
 806                 } else if (key.offset > end)
 807                         break;
 808
 809 next:
 810                 path->slots[0]++;
 811         }
 812         ret = 0;
 813 out:
 814         btrfs_free_path(path);
 815         return ret;
 816 }
 817
 818 /*
 819  * find_free_dev_extent - find free space in the specified device
 820  * @trans:      transaction handler
 821  * @device:     the device which we search the free space in
 822  * @num_bytes:  the size of the free space that we need
 823  * @start:      store the start of the free space.
 824  * @len:        the size of the free space. that we find, or the size of the max
 825  *              free space if we don't find suitable free space
 826  *
 827  * this uses a pretty simple search, the expectation is that it is
 828  * called very infrequently and that a given device has a small number
 829  * of extents
 830  *
 831  * @start is used to store the start of the free space if we find. But if we
 832  * don't find suitable free space, it will be used to store the start position
 833  * of the max free space.
 834  *
 835  * @len is used to store the size of the free space that we find.
 836  * But if we don't find suitable free space, it is used to store the size of
 837  * the max free space.
 838  */
 839 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 840                          struct btrfs_device *device, u64 num_bytes,
 841                          u64 *start, u64 *len)
 842 {
 843         struct btrfs_key key;
 844         struct btrfs_root *root = device->dev_root;
 845         struct btrfs_dev_extent *dev_extent;
 846         struct btrfs_path *path;
 847         u64 hole_size;
 848         u64 max_hole_start;
 849         u64 max_hole_size;
 850         u64 extent_end;
 851         u64 search_start;
 852         u64 search_end = device->total_bytes;
 853         int ret;
 854         int slot;
 855         struct extent_buffer *l;
 856
 857         /* FIXME use last free of some kind */
 858
 859         /* we don't want to overwrite the superblock on the drive,
 860          * so we make sure to start at an offset of at least 1MB
 861          */
 862         search_start = 1024 * 1024;
 863
 864         if (root->fs_info->alloc_start + num_bytes <= search_end)
 865                 search_start = max(root->fs_info->alloc_start, search_start);
 866
 867         max_hole_start = search_start;
 868         max_hole_size = 0;
 869
 870         if (search_start >= search_end) {
 871                 ret = -ENOSPC;
 872                 goto error;
 873         }
 874
 875         path = btrfs_alloc_path();
 876         if (!path) {
 877                 ret = -ENOMEM;
 878                 goto error;
 879         }
 880         path->reada = 2;
 881
 882         key.objectid = device->devid;
 883         key.offset = search_start;
 884         key.type = BTRFS_DEV_EXTENT_KEY;
 885
 886         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 887         if (ret < 0)
 888                 goto out;
 889         if (ret > 0) {
 890                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 891                 if (ret < 0)
 892                         goto out;
 893         }
 894
 895         while (1) {
 896                 l = path->nodes[0];
 897                 slot = path->slots[0];
 898                 if (slot >= btrfs_header_nritems(l)) {
 899                         ret = btrfs_next_leaf(root, path);
 900                         if (ret == 0)
 901                                 continue;
 902                         if (ret < 0)
 903                                 goto out;
 904
 905                         break;
 906                 }
 907                 btrfs_item_key_to_cpu(l, &key, slot);
 908
 909                 if (key.objectid < device->devid)
 910                         goto next;
 911
 912                 if (key.objectid > device->devid)
 913                         break;
 914
 915                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 916                         goto next;
 917
 918                 if (key.offset > search_start) {
 919                         hole_size = key.offset - search_start;
 920
 921                         if (hole_size > max_hole_size) {
 922                                 max_hole_start = search_start;
 923                                 max_hole_size = hole_size;
 924                         }
 925
 926                         /*
 927                          * If this free space is greater than which we need,
 928                          * it must be the max free space that we have found
 929                          * until now, so max_hole_start must point to the start
 930                          * of this free space and the length of this free space
 931                          * is stored in max_hole_size. Thus, we return
 932                          * max_hole_start and max_hole_size and go back to the
 933                          * caller.
 934                          */
 935                         if (hole_size >= num_bytes) {
 936                                 ret = 0;
 937                                 goto out;
 938                         }
 939                 }
 940
 941                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 942                 extent_end = key.offset + btrfs_dev_extent_length(l,
 943                                                                   dev_extent);
 944                 if (extent_end > search_start)
 945                         search_start = extent_end;
 946 next:
 947                 path->slots[0]++;
 948                 cond_resched();
 949         }
 950
 951         hole_size = search_end- search_start;
 952         if (hole_size > max_hole_size) {
 953                 max_hole_start = search_start;
 954                 max_hole_size = hole_size;
 955         }
 956
 957         /* See above. */
 958         if (hole_size < num_bytes)
 959                 ret = -ENOSPC;
 960         else
 961                 ret = 0;
 962
 963 out:
 964         btrfs_free_path(path);
 965 error:
 966         *start = max_hole_start;
 967         if (len)
 968                 *len = max_hole_size;
 969         return ret;
 970 }
 971
 972 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 973                           struct btrfs_device *device,
 974                           u64 start)
 975 {
 976         int ret;
 977         struct btrfs_path *path;
 978         struct btrfs_root *root = device->dev_root;
 979         struct btrfs_key key;
 980         struct btrfs_key found_key;
 981         struct extent_buffer *leaf = NULL;
 982         struct btrfs_dev_extent *extent = NULL;
 983
 984         path = btrfs_alloc_path();
 985         if (!path)
 986                 return -ENOMEM;
 987
 988         key.objectid = device->devid;
 989         key.offset = start;
 990         key.type = BTRFS_DEV_EXTENT_KEY;
 991
 992         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 993         if (ret > 0) {
 994                 ret = btrfs_previous_item(root, path, key.objectid,
 995                                           BTRFS_DEV_EXTENT_KEY);
 996                 BUG_ON(ret);
 997                 leaf = path->nodes[0];
 998                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 999                 extent = btrfs_item_ptr(leaf, path->slots[0],
1000                                         struct btrfs_dev_extent);
1001                 BUG_ON(found_key.offset > start || found_key.offset +
1002                        btrfs_dev_extent_length(leaf, extent) < start);
1003                 ret = 0;
1004         } else if (ret == 0) {
1005                 leaf = path->nodes[0];
1006                 extent = btrfs_item_ptr(leaf, path->slots[0],
1007                                         struct btrfs_dev_extent);
1008         }
1009         BUG_ON(ret);
1010
1011         if (device->bytes_used > 0)
1012                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
1013         ret = btrfs_del_item(trans, root, path);
1014         BUG_ON(ret);
1015
1016         btrfs_free_path(path);
1017         return ret;
1018 }
1019
1020 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1021                            struct btrfs_device *device,
1022                            u64 chunk_tree, u64 chunk_objectid,
1023                            u64 chunk_offset, u64 start, u64 num_bytes)
1024 {
1025         int ret;
1026         struct btrfs_path *path;
1027         struct btrfs_root *root = device->dev_root;
1028         struct btrfs_dev_extent *extent;
1029         struct extent_buffer *leaf;
1030         struct btrfs_key key;
1031
1032         WARN_ON(!device->in_fs_metadata);
1033         path = btrfs_alloc_path();
1034         if (!path)
1035                 return -ENOMEM;
1036
1037         key.objectid = device->devid;
1038         key.offset = start;
1039         key.type = BTRFS_DEV_EXTENT_KEY;
1040         ret = btrfs_insert_empty_item(trans, root, path, &key,
1041                                       sizeof(*extent));
1042         BUG_ON(ret);
1043
1044         leaf = path->nodes[0];
1045         extent = btrfs_item_ptr(leaf, path->slots[0],
1046                                 struct btrfs_dev_extent);
1047         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1048         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1049         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1050
1051         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1052                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1053                     BTRFS_UUID_SIZE);
1054
1055         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1056         btrfs_mark_buffer_dirty(leaf);
1057         btrfs_free_path(path);
1058         return ret;
1059 }
1060
1061 static noinline int find_next_chunk(struct btrfs_root *root,
1062                                     u64 objectid, u64 *offset)
1063 {
1064         struct btrfs_path *path;
1065         int ret;
1066         struct btrfs_key key;
1067         struct btrfs_chunk *chunk;
1068         struct btrfs_key found_key;
1069
1070         path = btrfs_alloc_path();
1071         BUG_ON(!path);
1072
1073         key.objectid = objectid;
1074         key.offset = (u64)-1;
1075         key.type = BTRFS_CHUNK_ITEM_KEY;
1076
1077         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1078         if (ret < 0)
1079                 goto error;
1080
1081         BUG_ON(ret == 0);
1082
1083         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1084         if (ret) {
1085                 *offset = 0;
1086         } else {
1087                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1088                                       path->slots[0]);
1089                 if (found_key.objectid != objectid)
1090                         *offset = 0;
1091                 else {
1092                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1093                                                struct btrfs_chunk);
1094                         *offset = found_key.offset +
1095                                 btrfs_chunk_length(path->nodes[0], chunk);
1096                 }
1097         }
1098         ret = 0;
1099 error:
1100         btrfs_free_path(path);
1101         return ret;
1102 }
1103
1104 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1105 {
1106         int ret;
1107         struct btrfs_key key;
1108         struct btrfs_key found_key;
1109         struct btrfs_path *path;
1110
1111         root = root->fs_info->chunk_root;
1112
1113         path = btrfs_alloc_path();
1114         if (!path)
1115                 return -ENOMEM;
1116
1117         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1118         key.type = BTRFS_DEV_ITEM_KEY;
1119         key.offset = (u64)-1;
1120
1121         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1122         if (ret < 0)
1123                 goto error;
1124
1125         BUG_ON(ret == 0);
1126
1127         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1128                                   BTRFS_DEV_ITEM_KEY);
1129         if (ret) {
1130                 *objectid = 1;
1131         } else {
1132                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1133                                       path->slots[0]);
1134                 *objectid = found_key.offset + 1;
1135         }
1136         ret = 0;
1137 error:
1138         btrfs_free_path(path);
1139         return ret;
1140 }
1141
1142 /*
1143  * the device information is stored in the chunk root
1144  * the btrfs_device struct should be fully filled in
1145  */
1146 int btrfs_add_device(struct btrfs_trans_handle *trans,
1147                      struct btrfs_root *root,
1148                      struct btrfs_device *device)
1149 {
1150         int ret;
1151         struct btrfs_path *path;
1152         struct btrfs_dev_item *dev_item;
1153         struct extent_buffer *leaf;
1154         struct btrfs_key key;
1155         unsigned long ptr;
1156
1157         root = root->fs_info->chunk_root;
1158
1159         path = btrfs_alloc_path();
1160         if (!path)
1161                 return -ENOMEM;
1162
1163         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1164         key.type = BTRFS_DEV_ITEM_KEY;
1165         key.offset = device->devid;
1166
1167         ret = btrfs_insert_empty_item(trans, root, path, &key,
1168                                       sizeof(*dev_item));
1169         if (ret)
1170                 goto out;
1171
1172         leaf = path->nodes[0];
1173         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1174
1175         btrfs_set_device_id(leaf, dev_item, device->devid);
1176         btrfs_set_device_generation(leaf, dev_item, 0);
1177         btrfs_set_device_type(leaf, dev_item, device->type);
1178         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1179         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1180         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1181         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1182         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1183         btrfs_set_device_group(leaf, dev_item, 0);
1184         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1185         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1186         btrfs_set_device_start_offset(leaf, dev_item, 0);
1187
1188         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1189         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1190         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1191         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1192         btrfs_mark_buffer_dirty(leaf);
1193
1194         ret = 0;
1195 out:
1196         btrfs_free_path(path);
1197         return ret;
1198 }
1199
1200 static int btrfs_rm_dev_item(struct btrfs_root *root,
1201                              struct btrfs_device *device)
1202 {
1203         int ret;
1204         struct btrfs_path *path;
1205         struct btrfs_key key;
1206         struct btrfs_trans_handle *trans;
1207
1208         root = root->fs_info->chunk_root;
1209
1210         path = btrfs_alloc_path();
1211         if (!path)
1212                 return -ENOMEM;
1213
1214         trans = btrfs_start_transaction(root, 0);
1215         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1216         key.type = BTRFS_DEV_ITEM_KEY;
1217         key.offset = device->devid;
1218         lock_chunks(root);
1219
1220         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1221         if (ret < 0)
1222                 goto out;
1223
1224         if (ret > 0) {
1225                 ret = -ENOENT;
1226                 goto out;
1227         }
1228
1229         ret = btrfs_del_item(trans, root, path);
1230         if (ret)
1231                 goto out;
1232 out:
1233         btrfs_free_path(path);
1234         unlock_chunks(root);
1235         btrfs_commit_transaction(trans, root);
1236         return ret;
1237 }
1238
1239 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1240 {
1241         struct btrfs_device *device;
1242         struct btrfs_device *next_device;
1243         struct block_device *bdev;
1244         struct buffer_head *bh = NULL;
1245         struct btrfs_super_block *disk_super;
1246         u64 all_avail;
1247         u64 devid;
1248         u64 num_devices;
1249         u8 *dev_uuid;
1250         int ret = 0;
1251
1252         mutex_lock(&uuid_mutex);
1253         mutex_lock(&root->fs_info->volume_mutex);
1254
1255         all_avail = root->fs_info->avail_data_alloc_bits |
1256                 root->fs_info->avail_system_alloc_bits |
1257                 root->fs_info->avail_metadata_alloc_bits;
1258
1259         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1260             root->fs_info->fs_devices->num_devices <= 4) {
1261                 printk(KERN_ERR "btrfs: unable to go below four devices "
1262                        "on raid10\n");
1263                 ret = -EINVAL;
1264                 goto out;
1265         }
1266
1267         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1268             root->fs_info->fs_devices->num_devices <= 2) {
1269                 printk(KERN_ERR "btrfs: unable to go below two "
1270                        "devices on raid1\n");
1271                 ret = -EINVAL;
1272                 goto out;
1273         }
1274
1275         if (strcmp(device_path, "missing") == 0) {
1276                 struct list_head *devices;
1277                 struct btrfs_device *tmp;
1278
1279                 device = NULL;
1280                 devices = &root->fs_info->fs_devices->devices;
1281                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1282                 list_for_each_entry(tmp, devices, dev_list) {
1283                         if (tmp->in_fs_metadata && !tmp->bdev) {
1284                                 device = tmp;
1285                                 break;
1286                         }
1287                 }
1288                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1289                 bdev = NULL;
1290                 bh = NULL;
1291                 disk_super = NULL;
1292                 if (!device) {
1293                         printk(KERN_ERR "btrfs: no missing devices found to "
1294                                "remove\n");
1295                         goto out;
1296                 }
1297         } else {
1298                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1299                                       root->fs_info->bdev_holder);
1300                 if (IS_ERR(bdev)) {
1301                         ret = PTR_ERR(bdev);
1302                         goto out;
1303                 }
1304
1305                 set_blocksize(bdev, 4096);
1306                 bh = btrfs_read_dev_super(bdev);
1307                 if (!bh) {
1308                         ret = -EINVAL;
1309                         goto error_close;
1310                 }
1311                 disk_super = (struct btrfs_super_block *)bh->b_data;
1312                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1313                 dev_uuid = disk_super->dev_item.uuid;
1314                 device = btrfs_find_device(root, devid, dev_uuid,
1315                                            disk_super->fsid);
1316                 if (!device) {
1317                         ret = -ENOENT;
1318                         goto error_brelse;
1319                 }
1320         }
1321
1322         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1323                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1324                        "device\n");
1325                 ret = -EINVAL;
1326                 goto error_brelse;
1327         }
1328
1329         if (device->writeable) {
1330                 list_del_init(&device->dev_alloc_list);
1331                 root->fs_info->fs_devices->rw_devices--;
1332         }
1333
1334         ret = btrfs_shrink_device(device, 0);
1335         if (ret)
1336                 goto error_brelse;
1337
1338         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1339         if (ret)
1340                 goto error_brelse;
1341
1342         device->in_fs_metadata = 0;
1343
1344         /*
1345          * the device list mutex makes sure that we don't change
1346          * the device list while someone else is writing out all
1347          * the device supers.
1348          */
1349         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1350         list_del_init(&device->dev_list);
1351         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1352
1353         device->fs_devices->num_devices--;
1354
1355         if (device->missing)
1356                 root->fs_info->fs_devices->missing_devices--;
1357
1358         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1359                                  struct btrfs_device, dev_list);
1360         if (device->bdev == root->fs_info->sb->s_bdev)
1361                 root->fs_info->sb->s_bdev = next_device->bdev;
1362         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1363                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1364
1365         if (device->bdev) {
1366                 close_bdev_exclusive(device->bdev, device->mode);
1367                 device->bdev = NULL;
1368                 device->fs_devices->open_devices--;
1369         }
1370
1371         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1372         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1373
1374         if (device->fs_devices->open_devices == 0) {
1375                 struct btrfs_fs_devices *fs_devices;
1376                 fs_devices = root->fs_info->fs_devices;
1377                 while (fs_devices) {
1378                         if (fs_devices->seed == device->fs_devices)
1379                                 break;
1380                         fs_devices = fs_devices->seed;
1381                 }
1382                 fs_devices->seed = device->fs_devices->seed;
1383                 device->fs_devices->seed = NULL;
1384                 __btrfs_close_devices(device->fs_devices);
1385                 free_fs_devices(device->fs_devices);
1386         }
1387
1388         /*
1389          * at this point, the device is zero sized.  We want to
1390          * remove it from the devices list and zero out the old super
1391          */
1392         if (device->writeable) {
1393                 /* make sure this device isn't detected as part of
1394                  * the FS anymore
1395                  */
1396                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1397                 set_buffer_dirty(bh);
1398                 sync_dirty_buffer(bh);
1399         }
1400
1401         kfree(device->name);
1402         kfree(device);
1403         ret = 0;
1404
1405 error_brelse:
1406         brelse(bh);
1407 error_close:
1408         if (bdev)
1409                 close_bdev_exclusive(bdev, FMODE_READ);
1410 out:
1411         mutex_unlock(&root->fs_info->volume_mutex);
1412         mutex_unlock(&uuid_mutex);
1413         return ret;
1414 }
1415
1416 /*
1417  * does all the dirty work required for changing file system's UUID.
1418  */
1419 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1420                                 struct btrfs_root *root)
1421 {
1422         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1423         struct btrfs_fs_devices *old_devices;
1424         struct btrfs_fs_devices *seed_devices;
1425         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1426         struct btrfs_device *device;
1427         u64 super_flags;
1428
1429         BUG_ON(!mutex_is_locked(&uuid_mutex));
1430         if (!fs_devices->seeding)
1431                 return -EINVAL;
1432
1433         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1434         if (!seed_devices)
1435                 return -ENOMEM;
1436
1437         old_devices = clone_fs_devices(fs_devices);
1438         if (IS_ERR(old_devices)) {
1439                 kfree(seed_devices);
1440                 return PTR_ERR(old_devices);
1441         }
1442
1443         list_add(&old_devices->list, &fs_uuids);
1444
1445         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1446         seed_devices->opened = 1;
1447         INIT_LIST_HEAD(&seed_devices->devices);
1448         INIT_LIST_HEAD(&seed_devices->alloc_list);
1449         mutex_init(&seed_devices->device_list_mutex);
1450         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1451         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1452         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1453                 device->fs_devices = seed_devices;
1454         }
1455
1456         fs_devices->seeding = 0;
1457         fs_devices->num_devices = 0;
1458         fs_devices->open_devices = 0;
1459         fs_devices->seed = seed_devices;
1460
1461         generate_random_uuid(fs_devices->fsid);
1462         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1463         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1464         super_flags = btrfs_super_flags(disk_super) &
1465                       ~BTRFS_SUPER_FLAG_SEEDING;
1466         btrfs_set_super_flags(disk_super, super_flags);
1467
1468         return 0;
1469 }
1470
1471 /*
1472  * strore the expected generation for seed devices in device items.
1473  */
1474 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1475                                struct btrfs_root *root)
1476 {
1477         struct btrfs_path *path;
1478         struct extent_buffer *leaf;
1479         struct btrfs_dev_item *dev_item;
1480         struct btrfs_device *device;
1481         struct btrfs_key key;
1482         u8 fs_uuid[BTRFS_UUID_SIZE];
1483         u8 dev_uuid[BTRFS_UUID_SIZE];
1484         u64 devid;
1485         int ret;
1486
1487         path = btrfs_alloc_path();
1488         if (!path)
1489                 return -ENOMEM;
1490
1491         root = root->fs_info->chunk_root;
1492         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1493         key.offset = 0;
1494         key.type = BTRFS_DEV_ITEM_KEY;
1495
1496         while (1) {
1497                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1498                 if (ret < 0)
1499                         goto error;
1500
1501                 leaf = path->nodes[0];
1502 next_slot:
1503                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1504                         ret = btrfs_next_leaf(root, path);
1505                         if (ret > 0)
1506                                 break;
1507                         if (ret < 0)
1508                                 goto error;
1509                         leaf = path->nodes[0];
1510                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1511                         btrfs_release_path(root, path);
1512                         continue;
1513                 }
1514
1515                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1516                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1517                     key.type != BTRFS_DEV_ITEM_KEY)
1518                         break;
1519
1520                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1521                                           struct btrfs_dev_item);
1522                 devid = btrfs_device_id(leaf, dev_item);
1523                 read_extent_buffer(leaf, dev_uuid,
1524                                    (unsigned long)btrfs_device_uuid(dev_item),
1525                                    BTRFS_UUID_SIZE);
1526                 read_extent_buffer(leaf, fs_uuid,
1527                                    (unsigned long)btrfs_device_fsid(dev_item),
1528                                    BTRFS_UUID_SIZE);
1529                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1530                 BUG_ON(!device);
1531
1532                 if (device->fs_devices->seeding) {
1533                         btrfs_set_device_generation(leaf, dev_item,
1534                                                     device->generation);
1535                         btrfs_mark_buffer_dirty(leaf);
1536                 }
1537
1538                 path->slots[0]++;
1539                 goto next_slot;
1540         }
1541         ret = 0;
1542 error:
1543         btrfs_free_path(path);
1544         return ret;
1545 }
1546
1547 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1548 {
1549         struct btrfs_trans_handle *trans;
1550         struct btrfs_device *device;
1551         struct block_device *bdev;
1552         struct list_head *devices;
1553         struct super_block *sb = root->fs_info->sb;
1554         u64 total_bytes;
1555         int seeding_dev = 0;
1556         int ret = 0;
1557
1558         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1559                 return -EINVAL;
1560
1561         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1562         if (IS_ERR(bdev))
1563                 return PTR_ERR(bdev);
1564
1565         if (root->fs_info->fs_devices->seeding) {
1566                 seeding_dev = 1;
1567                 down_write(&sb->s_umount);
1568                 mutex_lock(&uuid_mutex);
1569         }
1570
1571         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1572         mutex_lock(&root->fs_info->volume_mutex);
1573
1574         devices = &root->fs_info->fs_devices->devices;
1575         /*
1576          * we have the volume lock, so we don't need the extra
1577          * device list mutex while reading the list here.
1578          */
1579         list_for_each_entry(device, devices, dev_list) {
1580                 if (device->bdev == bdev) {
1581                         ret = -EEXIST;
1582                         goto error;
1583                 }
1584         }
1585
1586         device = kzalloc(sizeof(*device), GFP_NOFS);
1587         if (!device) {
1588                 /* we can safely leave the fs_devices entry around */
1589                 ret = -ENOMEM;
1590                 goto error;
1591         }
1592
1593         device->name = kstrdup(device_path, GFP_NOFS);
1594         if (!device->name) {
1595                 kfree(device);
1596                 ret = -ENOMEM;
1597                 goto error;
1598         }
1599
1600         ret = find_next_devid(root, &device->devid);
1601         if (ret) {
1602                 kfree(device);
1603                 goto error;
1604         }
1605
1606         trans = btrfs_start_transaction(root, 0);
1607         lock_chunks(root);
1608
1609         device->barriers = 1;
1610         device->writeable = 1;
1611         device->work.func = pending_bios_fn;
1612         generate_random_uuid(device->uuid);
1613         spin_lock_init(&device->io_lock);
1614         device->generation = trans->transid;
1615         device->io_width = root->sectorsize;
1616         device->io_align = root->sectorsize;
1617         device->sector_size = root->sectorsize;
1618         device->total_bytes = i_size_read(bdev->bd_inode);
1619         device->disk_total_bytes = device->total_bytes;
1620         device->dev_root = root->fs_info->dev_root;
1621         device->bdev = bdev;
1622         device->in_fs_metadata = 1;
1623         device->mode = 0;
1624         set_blocksize(device->bdev, 4096);
1625
1626         if (seeding_dev) {
1627                 sb->s_flags &= ~MS_RDONLY;
1628                 ret = btrfs_prepare_sprout(trans, root);
1629                 BUG_ON(ret);
1630         }
1631
1632         device->fs_devices = root->fs_info->fs_devices;
1633
1634         /*
1635          * we don't want write_supers to jump in here with our device
1636          * half setup
1637          */
1638         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1639         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1640         list_add(&device->dev_alloc_list,
1641                  &root->fs_info->fs_devices->alloc_list);
1642         root->fs_info->fs_devices->num_devices++;
1643         root->fs_info->fs_devices->open_devices++;
1644         root->fs_info->fs_devices->rw_devices++;
1645         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1646
1647         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1648                 root->fs_info->fs_devices->rotating = 1;
1649
1650         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1651         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1652                                     total_bytes + device->total_bytes);
1653
1654         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1655         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1656                                     total_bytes + 1);
1657         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1658
1659         if (seeding_dev) {
1660                 ret = init_first_rw_device(trans, root, device);
1661                 BUG_ON(ret);
1662                 ret = btrfs_finish_sprout(trans, root);
1663                 BUG_ON(ret);
1664         } else {
1665                 ret = btrfs_add_device(trans, root, device);
1666         }
1667
1668         /*
1669          * we've got more storage, clear any full flags on the space
1670          * infos
1671          */
1672         btrfs_clear_space_info_full(root->fs_info);
1673
1674         unlock_chunks(root);
1675         btrfs_commit_transaction(trans, root);
1676
1677         if (seeding_dev) {
1678                 mutex_unlock(&uuid_mutex);
1679                 up_write(&sb->s_umount);
1680
1681                 ret = btrfs_relocate_sys_chunks(root);
1682                 BUG_ON(ret);
1683         }
1684 out:
1685         mutex_unlock(&root->fs_info->volume_mutex);
1686         return ret;
1687 error:
1688         close_bdev_exclusive(bdev, 0);
1689         if (seeding_dev) {
1690                 mutex_unlock(&uuid_mutex);
1691                 up_write(&sb->s_umount);
1692         }
1693         goto out;
1694 }
1695
1696 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1697                                         struct btrfs_device *device)
1698 {
1699         int ret;
1700         struct btrfs_path *path;
1701         struct btrfs_root *root;
1702         struct btrfs_dev_item *dev_item;
1703         struct extent_buffer *leaf;
1704         struct btrfs_key key;
1705
1706         root = device->dev_root->fs_info->chunk_root;
1707
1708         path = btrfs_alloc_path();
1709         if (!path)
1710                 return -ENOMEM;
1711
1712         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1713         key.type = BTRFS_DEV_ITEM_KEY;
1714         key.offset = device->devid;
1715
1716         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1717         if (ret < 0)
1718                 goto out;
1719
1720         if (ret > 0) {
1721                 ret = -ENOENT;
1722                 goto out;
1723         }
1724
1725         leaf = path->nodes[0];
1726         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1727
1728         btrfs_set_device_id(leaf, dev_item, device->devid);
1729         btrfs_set_device_type(leaf, dev_item, device->type);
1730         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1731         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1732         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1733         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1734         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1735         btrfs_mark_buffer_dirty(leaf);
1736
1737 out:
1738         btrfs_free_path(path);
1739         return ret;
1740 }
1741
1742 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1743                       struct btrfs_device *device, u64 new_size)
1744 {
1745         struct btrfs_super_block *super_copy =
1746                 &device->dev_root->fs_info->super_copy;
1747         u64 old_total = btrfs_super_total_bytes(super_copy);
1748         u64 diff = new_size - device->total_bytes;
1749
1750         if (!device->writeable)
1751                 return -EACCES;
1752         if (new_size <= device->total_bytes)
1753                 return -EINVAL;
1754
1755         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1756         device->fs_devices->total_rw_bytes += diff;
1757
1758         device->total_bytes = new_size;
1759         device->disk_total_bytes = new_size;
1760         btrfs_clear_space_info_full(device->dev_root->fs_info);
1761
1762         return btrfs_update_device(trans, device);
1763 }
1764
1765 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1766                       struct btrfs_device *device, u64 new_size)
1767 {
1768         int ret;
1769         lock_chunks(device->dev_root);
1770         ret = __btrfs_grow_device(trans, device, new_size);
1771         unlock_chunks(device->dev_root);
1772         return ret;
1773 }
1774
1775 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1776                             struct btrfs_root *root,
1777                             u64 chunk_tree, u64 chunk_objectid,
1778                             u64 chunk_offset)
1779 {
1780         int ret;
1781         struct btrfs_path *path;
1782         struct btrfs_key key;
1783
1784         root = root->fs_info->chunk_root;
1785         path = btrfs_alloc_path();
1786         if (!path)
1787                 return -ENOMEM;
1788
1789         key.objectid = chunk_objectid;
1790         key.offset = chunk_offset;
1791         key.type = BTRFS_CHUNK_ITEM_KEY;
1792
1793         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1794         BUG_ON(ret);
1795
1796         ret = btrfs_del_item(trans, root, path);
1797         BUG_ON(ret);
1798
1799         btrfs_free_path(path);
1800         return 0;
1801 }
1802
1803 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1804                         chunk_offset)
1805 {
1806         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1807         struct btrfs_disk_key *disk_key;
1808         struct btrfs_chunk *chunk;
1809         u8 *ptr;
1810         int ret = 0;
1811         u32 num_stripes;
1812         u32 array_size;
1813         u32 len = 0;
1814         u32 cur;
1815         struct btrfs_key key;
1816
1817         array_size = btrfs_super_sys_array_size(super_copy);
1818
1819         ptr = super_copy->sys_chunk_array;
1820         cur = 0;
1821
1822         while (cur < array_size) {
1823                 disk_key = (struct btrfs_disk_key *)ptr;
1824                 btrfs_disk_key_to_cpu(&key, disk_key);
1825
1826                 len = sizeof(*disk_key);
1827
1828                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1829                         chunk = (struct btrfs_chunk *)(ptr + len);
1830                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1831                         len += btrfs_chunk_item_size(num_stripes);
1832                 } else {
1833                         ret = -EIO;
1834                         break;
1835                 }
1836                 if (key.objectid == chunk_objectid &&
1837                     key.offset == chunk_offset) {
1838                         memmove(ptr, ptr + len, array_size - (cur + len));
1839                         array_size -= len;
1840                         btrfs_set_super_sys_array_size(super_copy, array_size);
1841                 } else {
1842                         ptr += len;
1843                         cur += len;
1844                 }
1845         }
1846         return ret;
1847 }
1848
1849 static int btrfs_relocate_chunk(struct btrfs_root *root,
1850                          u64 chunk_tree, u64 chunk_objectid,
1851                          u64 chunk_offset)
1852 {
1853         struct extent_map_tree *em_tree;
1854         struct btrfs_root *extent_root;
1855         struct btrfs_trans_handle *trans;
1856         struct extent_map *em;
1857         struct map_lookup *map;
1858         int ret;
1859         int i;
1860
1861         root = root->fs_info->chunk_root;
1862         extent_root = root->fs_info->extent_root;
1863         em_tree = &root->fs_info->mapping_tree.map_tree;
1864
1865         ret = btrfs_can_relocate(extent_root, chunk_offset);
1866         if (ret)
1867                 return -ENOSPC;
1868
1869         /* step one, relocate all the extents inside this chunk */
1870         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1871         if (ret)
1872                 return ret;
1873
1874         trans = btrfs_start_transaction(root, 0);
1875         BUG_ON(!trans);
1876
1877         lock_chunks(root);
1878
1879         /*
1880          * step two, delete the device extents and the
1881          * chunk tree entries
1882          */
1883         read_lock(&em_tree->lock);
1884         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1885         read_unlock(&em_tree->lock);
1886
1887         BUG_ON(em->start > chunk_offset ||
1888                em->start + em->len < chunk_offset);
1889         map = (struct map_lookup *)em->bdev;
1890
1891         for (i = 0; i < map->num_stripes; i++) {
1892                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1893                                             map->stripes[i].physical);
1894                 BUG_ON(ret);
1895
1896                 if (map->stripes[i].dev) {
1897                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1898                         BUG_ON(ret);
1899                 }
1900         }
1901         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1902                                chunk_offset);
1903
1904         BUG_ON(ret);
1905
1906         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1907                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1908                 BUG_ON(ret);
1909         }
1910
1911         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1912         BUG_ON(ret);
1913
1914         write_lock(&em_tree->lock);
1915         remove_extent_mapping(em_tree, em);
1916         write_unlock(&em_tree->lock);
1917
1918         kfree(map);
1919         em->bdev = NULL;
1920
1921         /* once for the tree */
1922         free_extent_map(em);
1923         /* once for us */
1924         free_extent_map(em);
1925
1926         unlock_chunks(root);
1927         btrfs_end_transaction(trans, root);
1928         return 0;
1929 }
1930
1931 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1932 {
1933         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1934         struct btrfs_path *path;
1935         struct extent_buffer *leaf;
1936         struct btrfs_chunk *chunk;
1937         struct btrfs_key key;
1938         struct btrfs_key found_key;
1939         u64 chunk_tree = chunk_root->root_key.objectid;
1940         u64 chunk_type;
1941         bool retried = false;
1942         int failed = 0;
1943         int ret;
1944
1945         path = btrfs_alloc_path();
1946         if (!path)
1947                 return -ENOMEM;
1948
1949 again:
1950         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1951         key.offset = (u64)-1;
1952         key.type = BTRFS_CHUNK_ITEM_KEY;
1953
1954         while (1) {
1955                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1956                 if (ret < 0)
1957                         goto error;
1958                 BUG_ON(ret == 0);
1959
1960                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1961                                           key.type);
1962                 if (ret < 0)
1963                         goto error;
1964                 if (ret > 0)
1965                         break;
1966
1967                 leaf = path->nodes[0];
1968                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1969
1970                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1971                                        struct btrfs_chunk);
1972                 chunk_type = btrfs_chunk_type(leaf, chunk);
1973                 btrfs_release_path(chunk_root, path);
1974
1975                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1976                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1977                                                    found_key.objectid,
1978                                                    found_key.offset);
1979                         if (ret == -ENOSPC)
1980                                 failed++;
1981                         else if (ret)
1982                                 BUG();
1983                 }
1984
1985                 if (found_key.offset == 0)
1986                         break;
1987                 key.offset = found_key.offset - 1;
1988         }
1989         ret = 0;
1990         if (failed && !retried) {
1991                 failed = 0;
1992                 retried = true;
1993                 goto again;
1994         } else if (failed && retried) {
1995                 WARN_ON(1);
1996                 ret = -ENOSPC;
1997         }
1998 error:
1999         btrfs_free_path(path);
2000         return ret;
2001 }
2002
2003 static u64 div_factor(u64 num, int factor)
2004 {
2005         if (factor == 10)
2006                 return num;
2007         num *= factor;
2008         do_div(num, 10);
2009         return num;
2010 }
2011
2012 int btrfs_balance(struct btrfs_root *dev_root)
2013 {
2014         int ret;
2015         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2016         struct btrfs_device *device;
2017         u64 old_size;
2018         u64 size_to_free;
2019         struct btrfs_path *path;
2020         struct btrfs_key key;
2021         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2022         struct btrfs_trans_handle *trans;
2023         struct btrfs_key found_key;
2024
2025         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2026                 return -EROFS;
2027
2028         if (!capable(CAP_SYS_ADMIN))
2029                 return -EPERM;
2030
2031         mutex_lock(&dev_root->fs_info->volume_mutex);
2032         dev_root = dev_root->fs_info->dev_root;
2033
2034         /* step one make some room on all the devices */
2035         list_for_each_entry(device, devices, dev_list) {
2036                 old_size = device->total_bytes;
2037                 size_to_free = div_factor(old_size, 1);
2038                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2039                 if (!device->writeable ||
2040                     device->total_bytes - device->bytes_used > size_to_free)
2041                         continue;
2042
2043                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2044                 if (ret == -ENOSPC)
2045                         break;
2046                 BUG_ON(ret);
2047
2048                 trans = btrfs_start_transaction(dev_root, 0);
2049                 BUG_ON(!trans);
2050
2051                 ret = btrfs_grow_device(trans, device, old_size);
2052                 BUG_ON(ret);
2053
2054                 btrfs_end_transaction(trans, dev_root);
2055         }
2056
2057         /* step two, relocate all the chunks */
2058         path = btrfs_alloc_path();
2059         BUG_ON(!path);
2060
2061         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2062         key.offset = (u64)-1;
2063         key.type = BTRFS_CHUNK_ITEM_KEY;
2064
2065         while (1) {
2066                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2067                 if (ret < 0)
2068                         goto error;
2069
2070                 /*
2071                  * this shouldn't happen, it means the last relocate
2072                  * failed
2073                  */
2074                 if (ret == 0)
2075                         break;
2076
2077                 ret = btrfs_previous_item(chunk_root, path, 0,
2078                                           BTRFS_CHUNK_ITEM_KEY);
2079                 if (ret)
2080                         break;
2081
2082                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2083                                       path->slots[0]);
2084                 if (found_key.objectid != key.objectid)
2085                         break;
2086
2087                 /* chunk zero is special */
2088                 if (found_key.offset == 0)
2089                         break;
2090
2091                 btrfs_release_path(chunk_root, path);
2092                 ret = btrfs_relocate_chunk(chunk_root,
2093                                            chunk_root->root_key.objectid,
2094                                            found_key.objectid,
2095                                            found_key.offset);
2096                 BUG_ON(ret && ret != -ENOSPC);
2097                 key.offset = found_key.offset - 1;
2098         }
2099         ret = 0;
2100 error:
2101         btrfs_free_path(path);
2102         mutex_unlock(&dev_root->fs_info->volume_mutex);
2103         return ret;
2104 }
2105
2106 /*
2107  * shrinking a device means finding all of the device extents past
2108  * the new size, and then following the back refs to the chunks.
2109  * The chunk relocation code actually frees the device extent
2110  */
2111 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2112 {
2113         struct btrfs_trans_handle *trans;
2114         struct btrfs_root *root = device->dev_root;
2115         struct btrfs_dev_extent *dev_extent = NULL;
2116         struct btrfs_path *path;
2117         u64 length;
2118         u64 chunk_tree;
2119         u64 chunk_objectid;
2120         u64 chunk_offset;
2121         int ret;
2122         int slot;
2123         int failed = 0;
2124         bool retried = false;
2125         struct extent_buffer *l;
2126         struct btrfs_key key;
2127         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2128         u64 old_total = btrfs_super_total_bytes(super_copy);
2129         u64 old_size = device->total_bytes;
2130         u64 diff = device->total_bytes - new_size;
2131
2132         if (new_size >= device->total_bytes)
2133                 return -EINVAL;
2134
2135         path = btrfs_alloc_path();
2136         if (!path)
2137                 return -ENOMEM;
2138
2139         path->reada = 2;
2140
2141         lock_chunks(root);
2142
2143         device->total_bytes = new_size;
2144         if (device->writeable)
2145                 device->fs_devices->total_rw_bytes -= diff;
2146         unlock_chunks(root);
2147
2148 again:
2149         key.objectid = device->devid;
2150         key.offset = (u64)-1;
2151         key.type = BTRFS_DEV_EXTENT_KEY;
2152
2153         while (1) {
2154                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2155                 if (ret < 0)
2156                         goto done;
2157
2158                 ret = btrfs_previous_item(root, path, 0, key.type);
2159                 if (ret < 0)
2160                         goto done;
2161                 if (ret) {
2162                         ret = 0;
2163                         btrfs_release_path(root, path);
2164                         break;
2165                 }
2166
2167                 l = path->nodes[0];
2168                 slot = path->slots[0];
2169                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2170
2171                 if (key.objectid != device->devid) {
2172                         btrfs_release_path(root, path);
2173                         break;
2174                 }
2175
2176                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2177                 length = btrfs_dev_extent_length(l, dev_extent);
2178
2179                 if (key.offset + length <= new_size) {
2180                         btrfs_release_path(root, path);
2181                         break;
2182                 }
2183
2184                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2185                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2186                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2187                 btrfs_release_path(root, path);
2188
2189                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2190                                            chunk_offset);
2191                 if (ret && ret != -ENOSPC)
2192                         goto done;
2193                 if (ret == -ENOSPC)
2194                         failed++;
2195                 key.offset -= 1;
2196         }
2197
2198         if (failed && !retried) {
2199                 failed = 0;
2200                 retried = true;
2201                 goto again;
2202         } else if (failed && retried) {
2203                 ret = -ENOSPC;
2204                 lock_chunks(root);
2205
2206                 device->total_bytes = old_size;
2207                 if (device->writeable)
2208                         device->fs_devices->total_rw_bytes += diff;
2209                 unlock_chunks(root);
2210                 goto done;
2211         }
2212
2213         /* Shrinking succeeded, else we would be at "done". */
2214         trans = btrfs_start_transaction(root, 0);
2215         lock_chunks(root);
2216
2217         device->disk_total_bytes = new_size;
2218         /* Now btrfs_update_device() will change the on-disk size. */
2219         ret = btrfs_update_device(trans, device);
2220         if (ret) {
2221                 unlock_chunks(root);
2222                 btrfs_end_transaction(trans, root);
2223                 goto done;
2224         }
2225         WARN_ON(diff > old_total);
2226         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2227         unlock_chunks(root);
2228         btrfs_end_transaction(trans, root);
2229 done:
2230         btrfs_free_path(path);
2231         return ret;
2232 }
2233
2234 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2235                            struct btrfs_root *root,
2236                            struct btrfs_key *key,
2237                            struct btrfs_chunk *chunk, int item_size)
2238 {
2239         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2240         struct btrfs_disk_key disk_key;
2241         u32 array_size;
2242         u8 *ptr;
2243
2244         array_size = btrfs_super_sys_array_size(super_copy);
2245         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2246                 return -EFBIG;
2247
2248         ptr = super_copy->sys_chunk_array + array_size;
2249         btrfs_cpu_key_to_disk(&disk_key, key);
2250         memcpy(ptr, &disk_key, sizeof(disk_key));
2251         ptr += sizeof(disk_key);
2252         memcpy(ptr, chunk, item_size);
2253         item_size += sizeof(disk_key);
2254         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2255         return 0;
2256 }
2257
2258 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2259                                         int num_stripes, int sub_stripes)
2260 {
2261         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2262                 return calc_size;
2263         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2264                 return calc_size * (num_stripes / sub_stripes);
2265         else
2266                 return calc_size * num_stripes;
2267 }
2268
2269 /* Used to sort the devices by max_avail(descending sort) */
2270 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2271 {
2272         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2273             ((struct btrfs_device_info *)dev_info2)->max_avail)
2274                 return -1;
2275         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2276                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2277                 return 1;
2278         else
2279                 return 0;
2280 }
2281
2282 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2283                                  int *num_stripes, int *min_stripes,
2284                                  int *sub_stripes)
2285 {
2286         *num_stripes = 1;
2287         *min_stripes = 1;
2288         *sub_stripes = 0;
2289
2290         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2291                 *num_stripes = fs_devices->rw_devices;
2292                 *min_stripes = 2;
2293         }
2294         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2295                 *num_stripes = 2;
2296                 *min_stripes = 2;
2297         }
2298         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2299                 if (fs_devices->rw_devices < 2)
2300                         return -ENOSPC;
2301                 *num_stripes = 2;
2302                 *min_stripes = 2;
2303         }
2304         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2305                 *num_stripes = fs_devices->rw_devices;
2306                 if (*num_stripes < 4)
2307                         return -ENOSPC;
2308                 *num_stripes &= ~(u32)1;
2309                 *sub_stripes = 2;
2310                 *min_stripes = 4;
2311         }
2312
2313         return 0;
2314 }
2315
2316 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2317                                     u64 proposed_size, u64 type,
2318                                     int num_stripes, int small_stripe)
2319 {
2320         int min_stripe_size = 1 * 1024 * 1024;
2321         u64 calc_size = proposed_size;
2322         u64 max_chunk_size = calc_size;
2323         int ncopies = 1;
2324
2325         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2326                     BTRFS_BLOCK_GROUP_DUP |
2327                     BTRFS_BLOCK_GROUP_RAID10))
2328                 ncopies = 2;
2329
2330         if (type & BTRFS_BLOCK_GROUP_DATA) {
2331                 max_chunk_size = 10 * calc_size;
2332                 min_stripe_size = 64 * 1024 * 1024;
2333         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2334                 max_chunk_size = 256 * 1024 * 1024;
2335                 min_stripe_size = 32 * 1024 * 1024;
2336         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2337                 calc_size = 8 * 1024 * 1024;
2338                 max_chunk_size = calc_size * 2;
2339                 min_stripe_size = 1 * 1024 * 1024;
2340         }
2341
2342         /* we don't want a chunk larger than 10% of writeable space */
2343         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2344                              max_chunk_size);
2345
2346         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2347                 calc_size = max_chunk_size * ncopies;
2348                 do_div(calc_size, num_stripes);
2349                 do_div(calc_size, BTRFS_STRIPE_LEN);
2350                 calc_size *= BTRFS_STRIPE_LEN;
2351         }
2352
2353         /* we don't want tiny stripes */
2354         if (!small_stripe)
2355                 calc_size = max_t(u64, min_stripe_size, calc_size);
2356
2357         /*
2358          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2359          * we end up with something bigger than a stripe
2360          */
2361         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2362
2363         do_div(calc_size, BTRFS_STRIPE_LEN);
2364         calc_size *= BTRFS_STRIPE_LEN;
2365
2366         return calc_size;
2367 }
2368
2369 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2370                                                       int num_stripes)
2371 {
2372         struct map_lookup *new;
2373         size_t len = map_lookup_size(num_stripes);
2374
2375         BUG_ON(map->num_stripes < num_stripes);
2376
2377         if (map->num_stripes == num_stripes)
2378                 return map;
2379
2380         new = kmalloc(len, GFP_NOFS);
2381         if (!new) {
2382                 /* just change map->num_stripes */
2383                 map->num_stripes = num_stripes;
2384                 return map;
2385         }
2386
2387         memcpy(new, map, len);
2388         new->num_stripes = num_stripes;
2389         kfree(map);
2390         return new;
2391 }
2392
2393 /*
2394  * helper to allocate device space from btrfs_device_info, in which we stored
2395  * max free space information of every device. It is used when we can not
2396  * allocate chunks by default size.
2397  *
2398  * By this helper, we can allocate a new chunk as larger as possible.
2399  */
2400 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2401                                     struct btrfs_fs_devices *fs_devices,
2402                                     struct btrfs_device_info *devices,
2403                                     int nr_device, u64 type,
2404                                     struct map_lookup **map_lookup,
2405                                     int min_stripes, u64 *stripe_size)
2406 {
2407         int i, index, sort_again = 0;
2408         int min_devices = min_stripes;
2409         u64 max_avail, min_free;
2410         struct map_lookup *map = *map_lookup;
2411         int ret;
2412
2413         if (nr_device < min_stripes)
2414                 return -ENOSPC;
2415
2416         btrfs_descending_sort_devices(devices, nr_device);
2417
2418         max_avail = devices[0].max_avail;
2419         if (!max_avail)
2420                 return -ENOSPC;
2421
2422         for (i = 0; i < nr_device; i++) {
2423                 /*
2424                  * if dev_offset = 0, it means the free space of this device
2425                  * is less than what we need, and we didn't search max avail
2426                  * extent on this device, so do it now.
2427                  */
2428                 if (!devices[i].dev_offset) {
2429                         ret = find_free_dev_extent(trans, devices[i].dev,
2430                                                    max_avail,
2431                                                    &devices[i].dev_offset,
2432                                                    &devices[i].max_avail);
2433                         if (ret != 0 && ret != -ENOSPC)
2434                                 return ret;
2435                         sort_again = 1;
2436                 }
2437         }
2438
2439         /* we update the max avail free extent of each devices, sort again */
2440         if (sort_again)
2441                 btrfs_descending_sort_devices(devices, nr_device);
2442
2443         if (type & BTRFS_BLOCK_GROUP_DUP)
2444                 min_devices = 1;
2445
2446         if (!devices[min_devices - 1].max_avail)
2447                 return -ENOSPC;
2448
2449         max_avail = devices[min_devices - 1].max_avail;
2450         if (type & BTRFS_BLOCK_GROUP_DUP)
2451                 do_div(max_avail, 2);
2452
2453         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2454                                              min_stripes, 1);
2455         if (type & BTRFS_BLOCK_GROUP_DUP)
2456                 min_free = max_avail * 2;
2457         else
2458                 min_free = max_avail;
2459
2460         if (min_free > devices[min_devices - 1].max_avail)
2461                 return -ENOSPC;
2462
2463         map = __shrink_map_lookup_stripes(map, min_stripes);
2464         *stripe_size = max_avail;
2465
2466         index = 0;
2467         for (i = 0; i < min_stripes; i++) {
2468                 map->stripes[i].dev = devices[index].dev;
2469                 map->stripes[i].physical = devices[index].dev_offset;
2470                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2471                         i++;
2472                         map->stripes[i].dev = devices[index].dev;
2473                         map->stripes[i].physical = devices[index].dev_offset +
2474                                                    max_avail;
2475                 }
2476                 index++;
2477         }
2478         *map_lookup = map;
2479
2480         return 0;
2481 }
2482
2483 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2484                                struct btrfs_root *extent_root,
2485                                struct map_lookup **map_ret,
2486                                u64 *num_bytes, u64 *stripe_size,
2487                                u64 start, u64 type)
2488 {
2489         struct btrfs_fs_info *info = extent_root->fs_info;
2490         struct btrfs_device *device = NULL;
2491         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2492         struct list_head *cur;
2493         struct map_lookup *map;
2494         struct extent_map_tree *em_tree;
2495         struct extent_map *em;
2496         struct btrfs_device_info *devices_info;
2497         struct list_head private_devs;
2498         u64 calc_size = 1024 * 1024 * 1024;
2499         u64 min_free;
2500         u64 avail;
2501         u64 dev_offset;
2502         int num_stripes;
2503         int min_stripes;
2504         int sub_stripes;
2505         int min_devices;        /* the min number of devices we need */
2506         int i;
2507         int ret;
2508         int index;
2509
2510         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2511             (type & BTRFS_BLOCK_GROUP_DUP)) {
2512                 WARN_ON(1);
2513                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2514         }
2515         if (list_empty(&fs_devices->alloc_list))
2516                 return -ENOSPC;
2517
2518         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2519                                     &min_stripes, &sub_stripes);
2520         if (ret)
2521                 return ret;
2522
2523         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2524                                GFP_NOFS);
2525         if (!devices_info)
2526                 return -ENOMEM;
2527
2528         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2529         if (!map) {
2530                 ret = -ENOMEM;
2531                 goto error;
2532         }
2533         map->num_stripes = num_stripes;
2534
2535         cur = fs_devices->alloc_list.next;
2536         index = 0;
2537         i = 0;
2538
2539         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2540                                              num_stripes, 0);
2541
2542         if (type & BTRFS_BLOCK_GROUP_DUP) {
2543                 min_free = calc_size * 2;
2544                 min_devices = 1;
2545         } else {
2546                 min_free = calc_size;
2547                 min_devices = min_stripes;
2548         }
2549
2550         INIT_LIST_HEAD(&private_devs);
2551         while (index < num_stripes) {
2552                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2553                 BUG_ON(!device->writeable);
2554                 if (device->total_bytes > device->bytes_used)
2555                         avail = device->total_bytes - device->bytes_used;
2556                 else
2557                         avail = 0;
2558                 cur = cur->next;
2559
2560                 if (device->in_fs_metadata && avail >= min_free) {
2561                         ret = find_free_dev_extent(trans, device, min_free,
2562                                                    &devices_info[i].dev_offset,
2563                                                    &devices_info[i].max_avail);
2564                         if (ret == 0) {
2565                                 list_move_tail(&device->dev_alloc_list,
2566                                                &private_devs);
2567                                 map->stripes[index].dev = device;
2568                                 map->stripes[index].physical =
2569                                                 devices_info[i].dev_offset;
2570                                 index++;
2571                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2572                                         map->stripes[index].dev = device;
2573                                         map->stripes[index].physical =
2574                                                 devices_info[i].dev_offset +
2575                                                 calc_size;
2576                                         index++;
2577                                 }
2578                         } else if (ret != -ENOSPC)
2579                                 goto error;
2580
2581                         devices_info[i].dev = device;
2582                         i++;
2583                 } else if (device->in_fs_metadata &&
2584                            avail >= BTRFS_STRIPE_LEN) {
2585                         devices_info[i].dev = device;
2586                         devices_info[i].max_avail = avail;
2587                         i++;
2588                 }
2589
2590                 if (cur == &fs_devices->alloc_list)
2591                         break;
2592         }
2593
2594         list_splice(&private_devs, &fs_devices->alloc_list);
2595         if (index < num_stripes) {
2596                 if (index >= min_stripes) {
2597                         num_stripes = index;
2598                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2599                                 num_stripes /= sub_stripes;
2600                                 num_stripes *= sub_stripes;
2601                         }
2602
2603                         map = __shrink_map_lookup_stripes(map, num_stripes);
2604                 } else if (i >= min_devices) {
2605                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2606                                                        devices_info, i, type,
2607                                                        &map, min_stripes,
2608                                                        &calc_size);
2609                         if (ret)
2610                                 goto error;
2611                 } else {
2612                         ret = -ENOSPC;
2613                         goto error;
2614                 }
2615         }
2616         map->sector_size = extent_root->sectorsize;
2617         map->stripe_len = BTRFS_STRIPE_LEN;
2618         map->io_align = BTRFS_STRIPE_LEN;
2619         map->io_width = BTRFS_STRIPE_LEN;
2620         map->type = type;
2621         map->sub_stripes = sub_stripes;
2622
2623         *map_ret = map;
2624         *stripe_size = calc_size;
2625         *num_bytes = chunk_bytes_by_type(type, calc_size,
2626                                          map->num_stripes, sub_stripes);
2627
2628         em = alloc_extent_map(GFP_NOFS);
2629         if (!em) {
2630                 ret = -ENOMEM;
2631                 goto error;
2632         }
2633         em->bdev = (struct block_device *)map;
2634         em->start = start;
2635         em->len = *num_bytes;
2636         em->block_start = 0;
2637         em->block_len = em->len;
2638
2639         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2640         write_lock(&em_tree->lock);
2641         ret = add_extent_mapping(em_tree, em);
2642         write_unlock(&em_tree->lock);
2643         BUG_ON(ret);
2644         free_extent_map(em);
2645
2646         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2647                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2648                                      start, *num_bytes);
2649         BUG_ON(ret);
2650
2651         index = 0;
2652         while (index < map->num_stripes) {
2653                 device = map->stripes[index].dev;
2654                 dev_offset = map->stripes[index].physical;
2655
2656                 ret = btrfs_alloc_dev_extent(trans, device,
2657                                 info->chunk_root->root_key.objectid,
2658                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2659                                 start, dev_offset, calc_size);
2660                 BUG_ON(ret);
2661                 index++;
2662         }
2663
2664         kfree(devices_info);
2665         return 0;
2666
2667 error:
2668         kfree(map);
2669         kfree(devices_info);
2670         return ret;
2671 }
2672
2673 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2674                                 struct btrfs_root *extent_root,
2675                                 struct map_lookup *map, u64 chunk_offset,
2676                                 u64 chunk_size, u64 stripe_size)
2677 {
2678         u64 dev_offset;
2679         struct btrfs_key key;
2680         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2681         struct btrfs_device *device;
2682         struct btrfs_chunk *chunk;
2683         struct btrfs_stripe *stripe;
2684         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2685         int index = 0;
2686         int ret;
2687
2688         chunk = kzalloc(item_size, GFP_NOFS);
2689         if (!chunk)
2690                 return -ENOMEM;
2691
2692         index = 0;
2693         while (index < map->num_stripes) {
2694                 device = map->stripes[index].dev;
2695                 device->bytes_used += stripe_size;
2696                 ret = btrfs_update_device(trans, device);
2697                 BUG_ON(ret);
2698                 index++;
2699         }
2700
2701         index = 0;
2702         stripe = &chunk->stripe;
2703         while (index < map->num_stripes) {
2704                 device = map->stripes[index].dev;
2705                 dev_offset = map->stripes[index].physical;
2706
2707                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2708                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2709                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2710                 stripe++;
2711                 index++;
2712         }
2713
2714         btrfs_set_stack_chunk_length(chunk, chunk_size);
2715         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2716         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2717         btrfs_set_stack_chunk_type(chunk, map->type);
2718         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2719         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2720         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2721         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2722         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2723
2724         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2725         key.type = BTRFS_CHUNK_ITEM_KEY;
2726         key.offset = chunk_offset;
2727
2728         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2729         BUG_ON(ret);
2730
2731         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2732                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2733                                              item_size);
2734                 BUG_ON(ret);
2735         }
2736         kfree(chunk);
2737         return 0;
2738 }
2739
2740 /*
2741  * Chunk allocation falls into two parts. The first part does works
2742  * that make the new allocated chunk useable, but not do any operation
2743  * that modifies the chunk tree. The second part does the works that
2744  * require modifying the chunk tree. This division is important for the
2745  * bootstrap process of adding storage to a seed btrfs.
2746  */
2747 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2748                       struct btrfs_root *extent_root, u64 type)
2749 {
2750         u64 chunk_offset;
2751         u64 chunk_size;
2752         u64 stripe_size;
2753         struct map_lookup *map;
2754         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2755         int ret;
2756
2757         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2758                               &chunk_offset);
2759         if (ret)
2760                 return ret;
2761
2762         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2763                                   &stripe_size, chunk_offset, type);
2764         if (ret)
2765                 return ret;
2766
2767         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2768                                    chunk_size, stripe_size);
2769         BUG_ON(ret);
2770         return 0;
2771 }
2772
2773 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2774                                          struct btrfs_root *root,
2775                                          struct btrfs_device *device)
2776 {
2777         u64 chunk_offset;
2778         u64 sys_chunk_offset;
2779         u64 chunk_size;
2780         u64 sys_chunk_size;
2781         u64 stripe_size;
2782         u64 sys_stripe_size;
2783         u64 alloc_profile;
2784         struct map_lookup *map;
2785         struct map_lookup *sys_map;
2786         struct btrfs_fs_info *fs_info = root->fs_info;
2787         struct btrfs_root *extent_root = fs_info->extent_root;
2788         int ret;
2789
2790         ret = find_next_chunk(fs_info->chunk_root,
2791                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2792         BUG_ON(ret);
2793
2794         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2795                         (fs_info->metadata_alloc_profile &
2796                          fs_info->avail_metadata_alloc_bits);
2797         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2798
2799         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2800                                   &stripe_size, chunk_offset, alloc_profile);
2801         BUG_ON(ret);
2802
2803         sys_chunk_offset = chunk_offset + chunk_size;
2804
2805         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2806                         (fs_info->system_alloc_profile &
2807                          fs_info->avail_system_alloc_bits);
2808         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2809
2810         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2811                                   &sys_chunk_size, &sys_stripe_size,
2812                                   sys_chunk_offset, alloc_profile);
2813         BUG_ON(ret);
2814
2815         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2816         BUG_ON(ret);
2817
2818         /*
2819          * Modifying chunk tree needs allocating new blocks from both
2820          * system block group and metadata block group. So we only can
2821          * do operations require modifying the chunk tree after both
2822          * block groups were created.
2823          */
2824         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2825                                    chunk_size, stripe_size);
2826         BUG_ON(ret);
2827
2828         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2829                                    sys_chunk_offset, sys_chunk_size,
2830                                    sys_stripe_size);
2831         BUG_ON(ret);
2832         return 0;
2833 }
2834
2835 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2836 {
2837         struct extent_map *em;
2838         struct map_lookup *map;
2839         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2840         int readonly = 0;
2841         int i;
2842
2843         read_lock(&map_tree->map_tree.lock);
2844         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2845         read_unlock(&map_tree->map_tree.lock);
2846         if (!em)
2847                 return 1;
2848
2849         if (btrfs_test_opt(root, DEGRADED)) {
2850                 free_extent_map(em);
2851                 return 0;
2852         }
2853
2854         map = (struct map_lookup *)em->bdev;
2855         for (i = 0; i < map->num_stripes; i++) {
2856                 if (!map->stripes[i].dev->writeable) {
2857                         readonly = 1;
2858                         break;
2859                 }
2860         }
2861         free_extent_map(em);
2862         return readonly;
2863 }
2864
2865 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2866 {
2867         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2868 }
2869
2870 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2871 {
2872         struct extent_map *em;
2873
2874         while (1) {
2875                 write_lock(&tree->map_tree.lock);
2876                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2877                 if (em)
2878                         remove_extent_mapping(&tree->map_tree, em);
2879                 write_unlock(&tree->map_tree.lock);
2880                 if (!em)
2881                         break;
2882                 kfree(em->bdev);
2883                 /* once for us */
2884                 free_extent_map(em);
2885                 /* once for the tree */
2886                 free_extent_map(em);
2887         }
2888 }
2889
2890 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2891 {
2892         struct extent_map *em;
2893         struct map_lookup *map;
2894         struct extent_map_tree *em_tree = &map_tree->map_tree;
2895         int ret;
2896
2897         read_lock(&em_tree->lock);
2898         em = lookup_extent_mapping(em_tree, logical, len);
2899         read_unlock(&em_tree->lock);
2900         BUG_ON(!em);
2901
2902         BUG_ON(em->start > logical || em->start + em->len < logical);
2903         map = (struct map_lookup *)em->bdev;
2904         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2905                 ret = map->num_stripes;
2906         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2907                 ret = map->sub_stripes;
2908         else
2909                 ret = 1;
2910         free_extent_map(em);
2911         return ret;
2912 }
2913
2914 static int find_live_mirror(struct map_lookup *map, int first, int num,
2915                             int optimal)
2916 {
2917         int i;
2918         if (map->stripes[optimal].dev->bdev)
2919                 return optimal;
2920         for (i = first; i < first + num; i++) {
2921                 if (map->stripes[i].dev->bdev)
2922                         return i;
2923         }
2924         /* we couldn't find one that doesn't fail.  Just return something
2925          * and the io error handling code will clean up eventually
2926          */
2927         return optimal;
2928 }
2929
2930 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2931                              u64 logical, u64 *length,
2932                              struct btrfs_multi_bio **multi_ret,
2933                              int mirror_num, struct page *unplug_page)
2934 {
2935         struct extent_map *em;
2936         struct map_lookup *map;
2937         struct extent_map_tree *em_tree = &map_tree->map_tree;
2938         u64 offset;
2939         u64 stripe_offset;
2940         u64 stripe_nr;
2941         int stripes_allocated = 8;
2942         int stripes_required = 1;
2943         int stripe_index;
2944         int i;
2945         int num_stripes;
2946         int max_errors = 0;
2947         struct btrfs_multi_bio *multi = NULL;
2948
2949         if (multi_ret && !(rw & REQ_WRITE))
2950                 stripes_allocated = 1;
2951 again:
2952         if (multi_ret) {
2953                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2954                                 GFP_NOFS);
2955                 if (!multi)
2956                         return -ENOMEM;
2957
2958                 atomic_set(&multi->error, 0);
2959         }
2960
2961         read_lock(&em_tree->lock);
2962         em = lookup_extent_mapping(em_tree, logical, *length);
2963         read_unlock(&em_tree->lock);
2964
2965         if (!em && unplug_page) {
2966                 kfree(multi);
2967                 return 0;
2968         }
2969
2970         if (!em) {
2971                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2972                        (unsigned long long)logical,
2973                        (unsigned long long)*length);
2974                 BUG();
2975         }
2976
2977         BUG_ON(em->start > logical || em->start + em->len < logical);
2978         map = (struct map_lookup *)em->bdev;
2979         offset = logical - em->start;
2980
2981         if (mirror_num > map->num_stripes)
2982                 mirror_num = 0;
2983
2984         /* if our multi bio struct is too small, back off and try again */
2985         if (rw & REQ_WRITE) {
2986                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2987                                  BTRFS_BLOCK_GROUP_DUP)) {
2988                         stripes_required = map->num_stripes;
2989                         max_errors = 1;
2990                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2991                         stripes_required = map->sub_stripes;
2992                         max_errors = 1;
2993                 }
2994         }
2995         if (multi_ret && (rw & REQ_WRITE) &&
2996             stripes_allocated < stripes_required) {
2997                 stripes_allocated = map->num_stripes;
2998                 free_extent_map(em);
2999                 kfree(multi);
3000                 goto again;
3001         }
3002         stripe_nr = offset;
3003         /*
3004          * stripe_nr counts the total number of stripes we have to stride
3005          * to get to this block
3006          */
3007         do_div(stripe_nr, map->stripe_len);
3008
3009         stripe_offset = stripe_nr * map->stripe_len;
3010         BUG_ON(offset < stripe_offset);
3011
3012         /* stripe_offset is the offset of this block in its stripe*/
3013         stripe_offset = offset - stripe_offset;
3014
3015         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3016                          BTRFS_BLOCK_GROUP_RAID10 |
3017                          BTRFS_BLOCK_GROUP_DUP)) {
3018                 /* we limit the length of each bio to what fits in a stripe */
3019                 *length = min_t(u64, em->len - offset,
3020                               map->stripe_len - stripe_offset);
3021         } else {
3022                 *length = em->len - offset;
3023         }
3024
3025         if (!multi_ret && !unplug_page)
3026                 goto out;
3027
3028         num_stripes = 1;
3029         stripe_index = 0;
3030         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3031                 if (unplug_page || (rw & REQ_WRITE))
3032                         num_stripes = map->num_stripes;
3033                 else if (mirror_num)
3034                         stripe_index = mirror_num - 1;
3035                 else {
3036                         stripe_index = find_live_mirror(map, 0,
3037                                             map->num_stripes,
3038                                             current->pid % map->num_stripes);
3039                 }
3040
3041         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3042                 if (rw & REQ_WRITE)
3043                         num_stripes = map->num_stripes;
3044                 else if (mirror_num)
3045                         stripe_index = mirror_num - 1;
3046
3047         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3048                 int factor = map->num_stripes / map->sub_stripes;
3049
3050                 stripe_index = do_div(stripe_nr, factor);
3051                 stripe_index *= map->sub_stripes;
3052
3053                 if (unplug_page || (rw & REQ_WRITE))
3054                         num_stripes = map->sub_stripes;
3055                 else if (mirror_num)
3056                         stripe_index += mirror_num - 1;
3057                 else {
3058                         stripe_index = find_live_mirror(map, stripe_index,
3059                                               map->sub_stripes, stripe_index +
3060                                               current->pid % map->sub_stripes);
3061                 }
3062         } else {
3063                 /*
3064                  * after this do_div call, stripe_nr is the number of stripes
3065                  * on this device we have to walk to find the data, and
3066                  * stripe_index is the number of our device in the stripe array
3067                  */
3068                 stripe_index = do_div(stripe_nr, map->num_stripes);
3069         }
3070         BUG_ON(stripe_index >= map->num_stripes);
3071
3072         for (i = 0; i < num_stripes; i++) {
3073                 if (unplug_page) {
3074                         struct btrfs_device *device;
3075                         struct backing_dev_info *bdi;
3076
3077                         device = map->stripes[stripe_index].dev;
3078                         if (device->bdev) {
3079                                 bdi = blk_get_backing_dev_info(device->bdev);
3080                                 if (bdi->unplug_io_fn)
3081                                         bdi->unplug_io_fn(bdi, unplug_page);
3082                         }
3083                 } else {
3084                         multi->stripes[i].physical =
3085                                 map->stripes[stripe_index].physical +
3086                                 stripe_offset + stripe_nr * map->stripe_len;
3087                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3088                 }
3089                 stripe_index++;
3090         }
3091         if (multi_ret) {
3092                 *multi_ret = multi;
3093                 multi->num_stripes = num_stripes;
3094                 multi->max_errors = max_errors;
3095         }
3096 out:
3097         free_extent_map(em);
3098         return 0;
3099 }
3100
3101 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3102                       u64 logical, u64 *length,
3103                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3104 {
3105         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3106                                  mirror_num, NULL);
3107 }
3108
3109 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3110                      u64 chunk_start, u64 physical, u64 devid,
3111                      u64 **logical, int *naddrs, int *stripe_len)
3112 {
3113         struct extent_map_tree *em_tree = &map_tree->map_tree;
3114         struct extent_map *em;
3115         struct map_lookup *map;
3116         u64 *buf;
3117         u64 bytenr;
3118         u64 length;
3119         u64 stripe_nr;
3120         int i, j, nr = 0;
3121
3122         read_lock(&em_tree->lock);
3123         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3124         read_unlock(&em_tree->lock);
3125
3126         BUG_ON(!em || em->start != chunk_start);
3127         map = (struct map_lookup *)em->bdev;
3128
3129         length = em->len;
3130         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3131                 do_div(length, map->num_stripes / map->sub_stripes);
3132         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3133                 do_div(length, map->num_stripes);
3134
3135         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3136         BUG_ON(!buf);
3137
3138         for (i = 0; i < map->num_stripes; i++) {
3139                 if (devid && map->stripes[i].dev->devid != devid)
3140                         continue;
3141                 if (map->stripes[i].physical > physical ||
3142                     map->stripes[i].physical + length <= physical)
3143                         continue;
3144
3145                 stripe_nr = physical - map->stripes[i].physical;
3146                 do_div(stripe_nr, map->stripe_len);
3147
3148                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3149                         stripe_nr = stripe_nr * map->num_stripes + i;
3150                         do_div(stripe_nr, map->sub_stripes);
3151                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3152                         stripe_nr = stripe_nr * map->num_stripes + i;
3153                 }
3154                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3155                 WARN_ON(nr >= map->num_stripes);
3156                 for (j = 0; j < nr; j++) {
3157                         if (buf[j] == bytenr)
3158                                 break;
3159                 }
3160                 if (j == nr) {
3161                         WARN_ON(nr >= map->num_stripes);
3162                         buf[nr++] = bytenr;
3163                 }
3164         }
3165
3166         *logical = buf;
3167         *naddrs = nr;
3168         *stripe_len = map->stripe_len;
3169
3170         free_extent_map(em);
3171         return 0;
3172 }
3173
3174 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3175                       u64 logical, struct page *page)
3176 {
3177         u64 length = PAGE_CACHE_SIZE;
3178         return __btrfs_map_block(map_tree, READ, logical, &length,
3179                                  NULL, 0, page);
3180 }
3181
3182 static void end_bio_multi_stripe(struct bio *bio, int err)
3183 {
3184         struct btrfs_multi_bio *multi = bio->bi_private;
3185         int is_orig_bio = 0;
3186
3187         if (err)
3188                 atomic_inc(&multi->error);
3189
3190         if (bio == multi->orig_bio)
3191                 is_orig_bio = 1;
3192
3193         if (atomic_dec_and_test(&multi->stripes_pending)) {
3194                 if (!is_orig_bio) {
3195                         bio_put(bio);
3196                         bio = multi->orig_bio;
3197                 }
3198                 bio->bi_private = multi->private;
3199                 bio->bi_end_io = multi->end_io;
3200                 /* only send an error to the higher layers if it is
3201                  * beyond the tolerance of the multi-bio
3202                  */
3203                 if (atomic_read(&multi->error) > multi->max_errors) {
3204                         err = -EIO;
3205                 } else if (err) {
3206                         /*
3207                          * this bio is actually up to date, we didn't
3208                          * go over the max number of errors
3209                          */
3210                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3211                         err = 0;
3212                 }
3213                 kfree(multi);
3214
3215                 bio_endio(bio, err);
3216         } else if (!is_orig_bio) {
3217                 bio_put(bio);
3218         }
3219 }
3220
3221 struct async_sched {
3222         struct bio *bio;
3223         int rw;
3224         struct btrfs_fs_info *info;
3225         struct btrfs_work work;
3226 };
3227
3228 /*
3229  * see run_scheduled_bios for a description of why bios are collected for
3230  * async submit.
3231  *
3232  * This will add one bio to the pending list for a device and make sure
3233  * the work struct is scheduled.
3234  */
3235 static noinline int schedule_bio(struct btrfs_root *root,
3236                                  struct btrfs_device *device,
3237                                  int rw, struct bio *bio)
3238 {
3239         int should_queue = 1;
3240         struct btrfs_pending_bios *pending_bios;
3241
3242         /* don't bother with additional async steps for reads, right now */
3243         if (!(rw & REQ_WRITE)) {
3244                 bio_get(bio);
3245                 submit_bio(rw, bio);
3246                 bio_put(bio);
3247                 return 0;
3248         }
3249
3250         /*
3251          * nr_async_bios allows us to reliably return congestion to the
3252          * higher layers.  Otherwise, the async bio makes it appear we have
3253          * made progress against dirty pages when we've really just put it
3254          * on a queue for later
3255          */
3256         atomic_inc(&root->fs_info->nr_async_bios);
3257         WARN_ON(bio->bi_next);
3258         bio->bi_next = NULL;
3259         bio->bi_rw |= rw;
3260
3261         spin_lock(&device->io_lock);
3262         if (bio->bi_rw & REQ_SYNC)
3263                 pending_bios = &device->pending_sync_bios;
3264         else
3265                 pending_bios = &device->pending_bios;
3266
3267         if (pending_bios->tail)
3268                 pending_bios->tail->bi_next = bio;
3269
3270         pending_bios->tail = bio;
3271         if (!pending_bios->head)
3272                 pending_bios->head = bio;
3273         if (device->running_pending)
3274                 should_queue = 0;
3275
3276         spin_unlock(&device->io_lock);
3277
3278         if (should_queue)
3279                 btrfs_queue_worker(&root->fs_info->submit_workers,
3280                                    &device->work);
3281         return 0;
3282 }
3283
3284 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3285                   int mirror_num, int async_submit)
3286 {
3287         struct btrfs_mapping_tree *map_tree;
3288         struct btrfs_device *dev;
3289         struct bio *first_bio = bio;
3290         u64 logical = (u64)bio->bi_sector << 9;
3291         u64 length = 0;
3292         u64 map_length;
3293         struct btrfs_multi_bio *multi = NULL;
3294         int ret;
3295         int dev_nr = 0;
3296         int total_devs = 1;
3297
3298         length = bio->bi_size;
3299         map_tree = &root->fs_info->mapping_tree;
3300         map_length = length;
3301
3302         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3303                               mirror_num);
3304         BUG_ON(ret);
3305
3306         total_devs = multi->num_stripes;
3307         if (map_length < length) {
3308                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3309                        "len %llu\n", (unsigned long long)logical,
3310                        (unsigned long long)length,
3311                        (unsigned long long)map_length);
3312                 BUG();
3313         }
3314         multi->end_io = first_bio->bi_end_io;
3315         multi->private = first_bio->bi_private;
3316         multi->orig_bio = first_bio;
3317         atomic_set(&multi->stripes_pending, multi->num_stripes);
3318
3319         while (dev_nr < total_devs) {
3320                 if (total_devs > 1) {
3321                         if (dev_nr < total_devs - 1) {
3322                                 bio = bio_clone(first_bio, GFP_NOFS);
3323                                 BUG_ON(!bio);
3324                         } else {
3325                                 bio = first_bio;
3326                         }
3327                         bio->bi_private = multi;
3328                         bio->bi_end_io = end_bio_multi_stripe;
3329                 }
3330                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3331                 dev = multi->stripes[dev_nr].dev;
3332                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3333                         bio->bi_bdev = dev->bdev;
3334                         if (async_submit)
3335                                 schedule_bio(root, dev, rw, bio);
3336                         else
3337                                 submit_bio(rw, bio);
3338                 } else {
3339                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3340                         bio->bi_sector = logical >> 9;
3341                         bio_endio(bio, -EIO);
3342                 }
3343                 dev_nr++;
3344         }
3345         if (total_devs == 1)
3346                 kfree(multi);
3347         return 0;
3348 }
3349
3350 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3351                                        u8 *uuid, u8 *fsid)
3352 {
3353         struct btrfs_device *device;
3354         struct btrfs_fs_devices *cur_devices;
3355
3356         cur_devices = root->fs_info->fs_devices;
3357         while (cur_devices) {
3358                 if (!fsid ||
3359                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3360                         device = __find_device(&cur_devices->devices,
3361                                                devid, uuid);
3362                         if (device)
3363                                 return device;
3364                 }
3365                 cur_devices = cur_devices->seed;
3366         }
3367         return NULL;
3368 }
3369
3370 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3371                                             u64 devid, u8 *dev_uuid)
3372 {
3373         struct btrfs_device *device;
3374         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3375
3376         device = kzalloc(sizeof(*device), GFP_NOFS);
3377         if (!device)
3378                 return NULL;
3379         list_add(&device->dev_list,
3380                  &fs_devices->devices);
3381         device->barriers = 1;
3382         device->dev_root = root->fs_info->dev_root;
3383         device->devid = devid;
3384         device->work.func = pending_bios_fn;
3385         device->fs_devices = fs_devices;
3386         device->missing = 1;
3387         fs_devices->num_devices++;
3388         fs_devices->missing_devices++;
3389         spin_lock_init(&device->io_lock);
3390         INIT_LIST_HEAD(&device->dev_alloc_list);
3391         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3392         return device;
3393 }
3394
3395 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3396                           struct extent_buffer *leaf,
3397                           struct btrfs_chunk *chunk)
3398 {
3399         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3400         struct map_lookup *map;
3401         struct extent_map *em;
3402         u64 logical;
3403         u64 length;
3404         u64 devid;
3405         u8 uuid[BTRFS_UUID_SIZE];
3406         int num_stripes;
3407         int ret;
3408         int i;
3409
3410         logical = key->offset;
3411         length = btrfs_chunk_length(leaf, chunk);
3412
3413         read_lock(&map_tree->map_tree.lock);
3414         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3415         read_unlock(&map_tree->map_tree.lock);
3416
3417         /* already mapped? */
3418         if (em && em->start <= logical && em->start + em->len > logical) {
3419                 free_extent_map(em);
3420                 return 0;
3421         } else if (em) {
3422                 free_extent_map(em);
3423         }
3424
3425         em = alloc_extent_map(GFP_NOFS);
3426         if (!em)
3427                 return -ENOMEM;
3428         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3429         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3430         if (!map) {
3431                 free_extent_map(em);
3432                 return -ENOMEM;
3433         }
3434
3435         em->bdev = (struct block_device *)map;
3436         em->start = logical;
3437         em->len = length;
3438         em->block_start = 0;
3439         em->block_len = em->len;
3440
3441         map->num_stripes = num_stripes;
3442         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3443         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3444         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3445         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3446         map->type = btrfs_chunk_type(leaf, chunk);
3447         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3448         for (i = 0; i < num_stripes; i++) {
3449                 map->stripes[i].physical =
3450                         btrfs_stripe_offset_nr(leaf, chunk, i);
3451                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3452                 read_extent_buffer(leaf, uuid, (unsigned long)
3453                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3454                                    BTRFS_UUID_SIZE);
3455                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3456                                                         NULL);
3457                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3458                         kfree(map);
3459                         free_extent_map(em);
3460                         return -EIO;
3461                 }
3462                 if (!map->stripes[i].dev) {
3463                         map->stripes[i].dev =
3464                                 add_missing_dev(root, devid, uuid);
3465                         if (!map->stripes[i].dev) {
3466                                 kfree(map);
3467                                 free_extent_map(em);
3468                                 return -EIO;
3469                         }
3470                 }
3471                 map->stripes[i].dev->in_fs_metadata = 1;
3472         }
3473
3474         write_lock(&map_tree->map_tree.lock);
3475         ret = add_extent_mapping(&map_tree->map_tree, em);
3476         write_unlock(&map_tree->map_tree.lock);
3477         BUG_ON(ret);
3478         free_extent_map(em);
3479
3480         return 0;
3481 }
3482
3483 static int fill_device_from_item(struct extent_buffer *leaf,
3484                                  struct btrfs_dev_item *dev_item,
3485                                  struct btrfs_device *device)
3486 {
3487         unsigned long ptr;
3488
3489         device->devid = btrfs_device_id(leaf, dev_item);
3490         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3491         device->total_bytes = device->disk_total_bytes;
3492         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3493         device->type = btrfs_device_type(leaf, dev_item);
3494         device->io_align = btrfs_device_io_align(leaf, dev_item);
3495         device->io_width = btrfs_device_io_width(leaf, dev_item);
3496         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3497
3498         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3499         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3500
3501         return 0;
3502 }
3503
3504 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3505 {
3506         struct btrfs_fs_devices *fs_devices;
3507         int ret;
3508
3509         mutex_lock(&uuid_mutex);
3510
3511         fs_devices = root->fs_info->fs_devices->seed;
3512         while (fs_devices) {
3513                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3514                         ret = 0;
3515                         goto out;
3516                 }
3517                 fs_devices = fs_devices->seed;
3518         }
3519
3520         fs_devices = find_fsid(fsid);
3521         if (!fs_devices) {
3522                 ret = -ENOENT;
3523                 goto out;
3524         }
3525
3526         fs_devices = clone_fs_devices(fs_devices);
3527         if (IS_ERR(fs_devices)) {
3528                 ret = PTR_ERR(fs_devices);
3529                 goto out;
3530         }
3531
3532         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3533                                    root->fs_info->bdev_holder);
3534         if (ret)
3535                 goto out;
3536
3537         if (!fs_devices->seeding) {
3538                 __btrfs_close_devices(fs_devices);
3539                 free_fs_devices(fs_devices);
3540                 ret = -EINVAL;
3541                 goto out;
3542         }
3543
3544         fs_devices->seed = root->fs_info->fs_devices->seed;
3545         root->fs_info->fs_devices->seed = fs_devices;
3546 out:
3547         mutex_unlock(&uuid_mutex);
3548         return ret;
3549 }
3550
3551 static int read_one_dev(struct btrfs_root *root,
3552                         struct extent_buffer *leaf,
3553                         struct btrfs_dev_item *dev_item)
3554 {
3555         struct btrfs_device *device;
3556         u64 devid;
3557         int ret;
3558         u8 fs_uuid[BTRFS_UUID_SIZE];
3559         u8 dev_uuid[BTRFS_UUID_SIZE];
3560
3561         devid = btrfs_device_id(leaf, dev_item);
3562         read_extent_buffer(leaf, dev_uuid,
3563                            (unsigned long)btrfs_device_uuid(dev_item),
3564                            BTRFS_UUID_SIZE);
3565         read_extent_buffer(leaf, fs_uuid,
3566                            (unsigned long)btrfs_device_fsid(dev_item),
3567                            BTRFS_UUID_SIZE);
3568
3569         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3570                 ret = open_seed_devices(root, fs_uuid);
3571                 if (ret && !btrfs_test_opt(root, DEGRADED))
3572                         return ret;
3573         }
3574
3575         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3576         if (!device || !device->bdev) {
3577                 if (!btrfs_test_opt(root, DEGRADED))
3578                         return -EIO;
3579
3580                 if (!device) {
3581                         printk(KERN_WARNING "warning devid %llu missing\n",
3582                                (unsigned long long)devid);
3583                         device = add_missing_dev(root, devid, dev_uuid);
3584                         if (!device)
3585                                 return -ENOMEM;
3586                 } else if (!device->missing) {
3587                         /*
3588                          * this happens when a device that was properly setup
3589                          * in the device info lists suddenly goes bad.
3590                          * device->bdev is NULL, and so we have to set
3591                          * device->missing to one here
3592                          */
3593                         root->fs_info->fs_devices->missing_devices++;
3594                         device->missing = 1;
3595                 }
3596         }
3597
3598         if (device->fs_devices != root->fs_info->fs_devices) {
3599                 BUG_ON(device->writeable);
3600                 if (device->generation !=
3601                     btrfs_device_generation(leaf, dev_item))
3602                         return -EINVAL;
3603         }
3604
3605         fill_device_from_item(leaf, dev_item, device);
3606         device->dev_root = root->fs_info->dev_root;
3607         device->in_fs_metadata = 1;
3608         if (device->writeable)
3609                 device->fs_devices->total_rw_bytes += device->total_bytes;
3610         ret = 0;
3611         return ret;
3612 }
3613
3614 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3615 {
3616         struct btrfs_dev_item *dev_item;
3617
3618         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3619                                                      dev_item);
3620         return read_one_dev(root, buf, dev_item);
3621 }
3622
3623 int btrfs_read_sys_array(struct btrfs_root *root)
3624 {
3625         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3626         struct extent_buffer *sb;
3627         struct btrfs_disk_key *disk_key;
3628         struct btrfs_chunk *chunk;
3629         u8 *ptr;
3630         unsigned long sb_ptr;
3631         int ret = 0;
3632         u32 num_stripes;
3633         u32 array_size;
3634         u32 len = 0;
3635         u32 cur;
3636         struct btrfs_key key;
3637
3638         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3639                                           BTRFS_SUPER_INFO_SIZE);
3640         if (!sb)
3641                 return -ENOMEM;
3642         btrfs_set_buffer_uptodate(sb);
3643         btrfs_set_buffer_lockdep_class(sb, 0);
3644
3645         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3646         array_size = btrfs_super_sys_array_size(super_copy);
3647
3648         ptr = super_copy->sys_chunk_array;
3649         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3650         cur = 0;
3651
3652         while (cur < array_size) {
3653                 disk_key = (struct btrfs_disk_key *)ptr;
3654                 btrfs_disk_key_to_cpu(&key, disk_key);
3655
3656                 len = sizeof(*disk_key); ptr += len;
3657                 sb_ptr += len;
3658                 cur += len;
3659
3660                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3661                         chunk = (struct btrfs_chunk *)sb_ptr;
3662                         ret = read_one_chunk(root, &key, sb, chunk);
3663                         if (ret)
3664                                 break;
3665                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3666                         len = btrfs_chunk_item_size(num_stripes);
3667                 } else {
3668                         ret = -EIO;
3669                         break;
3670                 }
3671                 ptr += len;
3672                 sb_ptr += len;
3673                 cur += len;
3674         }
3675         free_extent_buffer(sb);
3676         return ret;
3677 }
3678
3679 int btrfs_read_chunk_tree(struct btrfs_root *root)
3680 {
3681         struct btrfs_path *path;
3682         struct extent_buffer *leaf;
3683         struct btrfs_key key;
3684         struct btrfs_key found_key;
3685         int ret;
3686         int slot;
3687
3688         root = root->fs_info->chunk_root;
3689
3690         path = btrfs_alloc_path();
3691         if (!path)
3692                 return -ENOMEM;
3693
3694         /* first we search for all of the device items, and then we
3695          * read in all of the chunk items.  This way we can create chunk
3696          * mappings that reference all of the devices that are afound
3697          */
3698         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3699         key.offset = 0;
3700         key.type = 0;
3701 again:
3702         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3703         if (ret < 0)
3704                 goto error;
3705         while (1) {
3706                 leaf = path->nodes[0];
3707                 slot = path->slots[0];
3708                 if (slot >= btrfs_header_nritems(leaf)) {
3709                         ret = btrfs_next_leaf(root, path);
3710                         if (ret == 0)
3711                                 continue;
3712                         if (ret < 0)
3713                                 goto error;
3714                         break;
3715                 }
3716                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3717                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3718                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3719                                 break;
3720                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3721                                 struct btrfs_dev_item *dev_item;
3722                                 dev_item = btrfs_item_ptr(leaf, slot,
3723                                                   struct btrfs_dev_item);
3724                                 ret = read_one_dev(root, leaf, dev_item);
3725                                 if (ret)
3726                                         goto error;
3727                         }
3728                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3729                         struct btrfs_chunk *chunk;
3730                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3731                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3732                         if (ret)
3733                                 goto error;
3734                 }
3735                 path->slots[0]++;
3736         }
3737         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3738                 key.objectid = 0;
3739                 btrfs_release_path(root, path);
3740                 goto again;
3741         }
3742         ret = 0;
3743 error:
3744         btrfs_free_path(path);
3745         return ret;
3746 }