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