fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <asm/div64.h>
  27 #include "compat.h"
  28 #include "ctree.h"
  29 #include "extent_map.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32 #include "print-tree.h"
  33 #include "volumes.h"
  34 #include "async-thread.h"
  35
  36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  37                                 struct btrfs_root *root,
  38                                 struct btrfs_device *device);
  39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  40
  41 static DEFINE_MUTEX(uuid_mutex);
  42 static LIST_HEAD(fs_uuids);
  43
  44 static void lock_chunks(struct btrfs_root *root)
  45 {
  46         mutex_lock(&root->fs_info->chunk_mutex);
  47 }
  48
  49 static void unlock_chunks(struct btrfs_root *root)
  50 {
  51         mutex_unlock(&root->fs_info->chunk_mutex);
  52 }
  53
  54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  55 {
  56         struct btrfs_device *device;
  57         WARN_ON(fs_devices->opened);
  58         while (!list_empty(&fs_devices->devices)) {
  59                 device = list_entry(fs_devices->devices.next,
  60                                     struct btrfs_device, dev_list);
  61                 list_del(&device->dev_list);
  62                 kfree(device->name);
  63                 kfree(device);
  64         }
  65         kfree(fs_devices);
  66 }
  67
  68 int btrfs_cleanup_fs_uuids(void)
  69 {
  70         struct btrfs_fs_devices *fs_devices;
  71
  72         while (!list_empty(&fs_uuids)) {
  73                 fs_devices = list_entry(fs_uuids.next,
  74                                         struct btrfs_fs_devices, list);
  75                 list_del(&fs_devices->list);
  76                 free_fs_devices(fs_devices);
  77         }
  78         return 0;
  79 }
  80
  81 static noinline struct btrfs_device *__find_device(struct list_head *head,
  82                                                    u64 devid, u8 *uuid)
  83 {
  84         struct btrfs_device *dev;
  85
  86         list_for_each_entry(dev, head, dev_list) {
  87                 if (dev->devid == devid &&
  88                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  89                         return dev;
  90                 }
  91         }
  92         return NULL;
  93 }
  94
  95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
  96 {
  97         struct btrfs_fs_devices *fs_devices;
  98
  99         list_for_each_entry(fs_devices, &fs_uuids, list) {
 100                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 101                         return fs_devices;
 102         }
 103         return NULL;
 104 }
 105
 106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 107                         struct bio *head, struct bio *tail)
 108 {
 109
 110         struct bio *old_head;
 111
 112         old_head = pending_bios->head;
 113         pending_bios->head = head;
 114         if (pending_bios->tail)
 115                 tail->bi_next = old_head;
 116         else
 117                 pending_bios->tail = tail;
 118 }
 119
 120 /*
 121  * we try to collect pending bios for a device so we don't get a large
 122  * number of procs sending bios down to the same device.  This greatly
 123  * improves the schedulers ability to collect and merge the bios.
 124  *
 125  * But, it also turns into a long list of bios to process and that is sure
 126  * to eventually make the worker thread block.  The solution here is to
 127  * make some progress and then put this work struct back at the end of
 128  * the list if the block device is congested.  This way, multiple devices
 129  * can make progress from a single worker thread.
 130  */
 131 static noinline int run_scheduled_bios(struct btrfs_device *device)
 132 {
 133         struct bio *pending;
 134         struct backing_dev_info *bdi;
 135         struct btrfs_fs_info *fs_info;
 136         struct btrfs_pending_bios *pending_bios;
 137         struct bio *tail;
 138         struct bio *cur;
 139         int again = 0;
 140         unsigned long num_run;
 141         unsigned long batch_run = 0;
 142         unsigned long limit;
 143         unsigned long last_waited = 0;
 144         int force_reg = 0;
 145         int sync_pending = 0;
 146         struct blk_plug plug;
 147
 148         /*
 149          * this function runs all the bios we've collected for
 150          * a particular device.  We don't want to wander off to
 151          * another device without first sending all of these down.
 152          * So, setup a plug here and finish it off before we return
 153          */
 154         blk_start_plug(&plug);
 155
 156         bdi = blk_get_backing_dev_info(device->bdev);
 157         fs_info = device->dev_root->fs_info;
 158         limit = btrfs_async_submit_limit(fs_info);
 159         limit = limit * 2 / 3;
 160
 161 loop:
 162         spin_lock(&device->io_lock);
 163
 164 loop_lock:
 165         num_run = 0;
 166
 167         /* take all the bios off the list at once and process them
 168          * later on (without the lock held).  But, remember the
 169          * tail and other pointers so the bios can be properly reinserted
 170          * into the list if we hit congestion
 171          */
 172         if (!force_reg && device->pending_sync_bios.head) {
 173                 pending_bios = &device->pending_sync_bios;
 174                 force_reg = 1;
 175         } else {
 176                 pending_bios = &device->pending_bios;
 177                 force_reg = 0;
 178         }
 179
 180         pending = pending_bios->head;
 181         tail = pending_bios->tail;
 182         WARN_ON(pending && !tail);
 183
 184         /*
 185          * if pending was null this time around, no bios need processing
 186          * at all and we can stop.  Otherwise it'll loop back up again
 187          * and do an additional check so no bios are missed.
 188          *
 189          * device->running_pending is used to synchronize with the
 190          * schedule_bio code.
 191          */
 192         if (device->pending_sync_bios.head == NULL &&
 193             device->pending_bios.head == NULL) {
 194                 again = 0;
 195                 device->running_pending = 0;
 196         } else {
 197                 again = 1;
 198                 device->running_pending = 1;
 199         }
 200
 201         pending_bios->head = NULL;
 202         pending_bios->tail = NULL;
 203
 204         spin_unlock(&device->io_lock);
 205
 206         while (pending) {
 207
 208                 rmb();
 209                 /* we want to work on both lists, but do more bios on the
 210                  * sync list than the regular list
 211                  */
 212                 if ((num_run > 32 &&
 213                     pending_bios != &device->pending_sync_bios &&
 214                     device->pending_sync_bios.head) ||
 215                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 216                     device->pending_bios.head)) {
 217                         spin_lock(&device->io_lock);
 218                         requeue_list(pending_bios, pending, tail);
 219                         goto loop_lock;
 220                 }
 221
 222                 cur = pending;
 223                 pending = pending->bi_next;
 224                 cur->bi_next = NULL;
 225                 atomic_dec(&fs_info->nr_async_bios);
 226
 227                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
 228                     waitqueue_active(&fs_info->async_submit_wait))
 229                         wake_up(&fs_info->async_submit_wait);
 230
 231                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 232
 233                 /*
 234                  * if we're doing the sync list, record that our
 235                  * plug has some sync requests on it
 236                  *
 237                  * If we're doing the regular list and there are
 238                  * sync requests sitting around, unplug before
 239                  * we add more
 240                  */
 241                 if (pending_bios == &device->pending_sync_bios) {
 242                         sync_pending = 1;
 243                 } else if (sync_pending) {
 244                         blk_finish_plug(&plug);
 245                         blk_start_plug(&plug);
 246                         sync_pending = 0;
 247                 }
 248
 249                 submit_bio(cur->bi_rw, cur);
 250                 num_run++;
 251                 batch_run++;
 252                 if (need_resched())
 253                         cond_resched();
 254
 255                 /*
 256                  * we made progress, there is more work to do and the bdi
 257                  * is now congested.  Back off and let other work structs
 258                  * run instead
 259                  */
 260                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 261                     fs_info->fs_devices->open_devices > 1) {
 262                         struct io_context *ioc;
 263
 264                         ioc = current->io_context;
 265
 266                         /*
 267                          * the main goal here is that we don't want to
 268                          * block if we're going to be able to submit
 269                          * more requests without blocking.
 270                          *
 271                          * This code does two great things, it pokes into
 272                          * the elevator code from a filesystem _and_
 273                          * it makes assumptions about how batching works.
 274                          */
 275                         if (ioc && ioc->nr_batch_requests > 0 &&
 276                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 277                             (last_waited == 0 ||
 278                              ioc->last_waited == last_waited)) {
 279                                 /*
 280                                  * we want to go through our batch of
 281                                  * requests and stop.  So, we copy out
 282                                  * the ioc->last_waited time and test
 283                                  * against it before looping
 284                                  */
 285                                 last_waited = ioc->last_waited;
 286                                 if (need_resched())
 287                                         cond_resched();
 288                                 continue;
 289                         }
 290                         spin_lock(&device->io_lock);
 291                         requeue_list(pending_bios, pending, tail);
 292                         device->running_pending = 1;
 293
 294                         spin_unlock(&device->io_lock);
 295                         btrfs_requeue_work(&device->work);
 296                         goto done;
 297                 }
 298         }
 299
 300         cond_resched();
 301         if (again)
 302                 goto loop;
 303
 304         spin_lock(&device->io_lock);
 305         if (device->pending_bios.head || device->pending_sync_bios.head)
 306                 goto loop_lock;
 307         spin_unlock(&device->io_lock);
 308
 309 done:
 310         blk_finish_plug(&plug);
 311         return 0;
 312 }
 313
 314 static void pending_bios_fn(struct btrfs_work *work)
 315 {
 316         struct btrfs_device *device;
 317
 318         device = container_of(work, struct btrfs_device, work);
 319         run_scheduled_bios(device);
 320 }
 321
 322 static noinline int device_list_add(const char *path,
 323                            struct btrfs_super_block *disk_super,
 324                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 325 {
 326         struct btrfs_device *device;
 327         struct btrfs_fs_devices *fs_devices;
 328         u64 found_transid = btrfs_super_generation(disk_super);
 329         char *name;
 330
 331         fs_devices = find_fsid(disk_super->fsid);
 332         if (!fs_devices) {
 333                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 334                 if (!fs_devices)
 335                         return -ENOMEM;
 336                 INIT_LIST_HEAD(&fs_devices->devices);
 337                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 338                 list_add(&fs_devices->list, &fs_uuids);
 339                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 340                 fs_devices->latest_devid = devid;
 341                 fs_devices->latest_trans = found_transid;
 342                 mutex_init(&fs_devices->device_list_mutex);
 343                 device = NULL;
 344         } else {
 345                 device = __find_device(&fs_devices->devices, devid,
 346                                        disk_super->dev_item.uuid);
 347         }
 348         if (!device) {
 349                 if (fs_devices->opened)
 350                         return -EBUSY;
 351
 352                 device = kzalloc(sizeof(*device), GFP_NOFS);
 353                 if (!device) {
 354                         /* we can safely leave the fs_devices entry around */
 355                         return -ENOMEM;
 356                 }
 357                 device->devid = devid;
 358                 device->work.func = pending_bios_fn;
 359                 memcpy(device->uuid, disk_super->dev_item.uuid,
 360                        BTRFS_UUID_SIZE);
 361                 spin_lock_init(&device->io_lock);
 362                 device->name = kstrdup(path, GFP_NOFS);
 363                 if (!device->name) {
 364                         kfree(device);
 365                         return -ENOMEM;
 366                 }
 367                 INIT_LIST_HEAD(&device->dev_alloc_list);
 368
 369                 mutex_lock(&fs_devices->device_list_mutex);
 370                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 371                 mutex_unlock(&fs_devices->device_list_mutex);
 372
 373                 device->fs_devices = fs_devices;
 374                 fs_devices->num_devices++;
 375         } else if (!device->name || strcmp(device->name, path)) {
 376                 name = kstrdup(path, GFP_NOFS);
 377                 if (!name)
 378                         return -ENOMEM;
 379                 kfree(device->name);
 380                 device->name = name;
 381                 if (device->missing) {
 382                         fs_devices->missing_devices--;
 383                         device->missing = 0;
 384                 }
 385         }
 386
 387         if (found_transid > fs_devices->latest_trans) {
 388                 fs_devices->latest_devid = devid;
 389                 fs_devices->latest_trans = found_transid;
 390         }
 391         *fs_devices_ret = fs_devices;
 392         return 0;
 393 }
 394
 395 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 396 {
 397         struct btrfs_fs_devices *fs_devices;
 398         struct btrfs_device *device;
 399         struct btrfs_device *orig_dev;
 400
 401         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 402         if (!fs_devices)
 403                 return ERR_PTR(-ENOMEM);
 404
 405         INIT_LIST_HEAD(&fs_devices->devices);
 406         INIT_LIST_HEAD(&fs_devices->alloc_list);
 407         INIT_LIST_HEAD(&fs_devices->list);
 408         mutex_init(&fs_devices->device_list_mutex);
 409         fs_devices->latest_devid = orig->latest_devid;
 410         fs_devices->latest_trans = orig->latest_trans;
 411         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 412
 413         /* We have held the volume lock, it is safe to get the devices. */
 414         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 415                 device = kzalloc(sizeof(*device), GFP_NOFS);
 416                 if (!device)
 417                         goto error;
 418
 419                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
 420                 if (!device->name) {
 421                         kfree(device);
 422                         goto error;
 423                 }
 424
 425                 device->devid = orig_dev->devid;
 426                 device->work.func = pending_bios_fn;
 427                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 428                 spin_lock_init(&device->io_lock);
 429                 INIT_LIST_HEAD(&device->dev_list);
 430                 INIT_LIST_HEAD(&device->dev_alloc_list);
 431
 432                 list_add(&device->dev_list, &fs_devices->devices);
 433                 device->fs_devices = fs_devices;
 434                 fs_devices->num_devices++;
 435         }
 436         return fs_devices;
 437 error:
 438         free_fs_devices(fs_devices);
 439         return ERR_PTR(-ENOMEM);
 440 }
 441
 442 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 443 {
 444         struct btrfs_device *device, *next;
 445
 446         mutex_lock(&uuid_mutex);
 447 again:
 448         /* This is the initialized path, it is safe to release the devices. */
 449         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 450                 if (device->in_fs_metadata)
 451                         continue;
 452
 453                 if (device->bdev) {
 454                         blkdev_put(device->bdev, device->mode);
 455                         device->bdev = NULL;
 456                         fs_devices->open_devices--;
 457                 }
 458                 if (device->writeable) {
 459                         list_del_init(&device->dev_alloc_list);
 460                         device->writeable = 0;
 461                         fs_devices->rw_devices--;
 462                 }
 463                 list_del_init(&device->dev_list);
 464                 fs_devices->num_devices--;
 465                 kfree(device->name);
 466                 kfree(device);
 467         }
 468
 469         if (fs_devices->seed) {
 470                 fs_devices = fs_devices->seed;
 471                 goto again;
 472         }
 473
 474         mutex_unlock(&uuid_mutex);
 475         return 0;
 476 }
 477
 478 static void __free_device(struct work_struct *work)
 479 {
 480         struct btrfs_device *device;
 481
 482         device = container_of(work, struct btrfs_device, rcu_work);
 483
 484         if (device->bdev)
 485                 blkdev_put(device->bdev, device->mode);
 486
 487         kfree(device->name);
 488         kfree(device);
 489 }
 490
 491 static void free_device(struct rcu_head *head)
 492 {
 493         struct btrfs_device *device;
 494
 495         device = container_of(head, struct btrfs_device, rcu);
 496
 497         INIT_WORK(&device->rcu_work, __free_device);
 498         schedule_work(&device->rcu_work);
 499 }
 500
 501 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 502 {
 503         struct btrfs_device *device;
 504
 505         if (--fs_devices->opened > 0)
 506                 return 0;
 507
 508         mutex_lock(&fs_devices->device_list_mutex);
 509         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 510                 struct btrfs_device *new_device;
 511
 512                 if (device->bdev)
 513                         fs_devices->open_devices--;
 514
 515                 if (device->writeable) {
 516                         list_del_init(&device->dev_alloc_list);
 517                         fs_devices->rw_devices--;
 518                 }
 519
 520                 if (device->can_discard)
 521                         fs_devices->num_can_discard--;
 522
 523                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 524                 BUG_ON(!new_device);
 525                 memcpy(new_device, device, sizeof(*new_device));
 526                 new_device->name = kstrdup(device->name, GFP_NOFS);
 527                 BUG_ON(device->name && !new_device->name);
 528                 new_device->bdev = NULL;
 529                 new_device->writeable = 0;
 530                 new_device->in_fs_metadata = 0;
 531                 new_device->can_discard = 0;
 532                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 533
 534                 call_rcu(&device->rcu, free_device);
 535         }
 536         mutex_unlock(&fs_devices->device_list_mutex);
 537
 538         WARN_ON(fs_devices->open_devices);
 539         WARN_ON(fs_devices->rw_devices);
 540         fs_devices->opened = 0;
 541         fs_devices->seeding = 0;
 542
 543         return 0;
 544 }
 545
 546 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 547 {
 548         struct btrfs_fs_devices *seed_devices = NULL;
 549         int ret;
 550
 551         mutex_lock(&uuid_mutex);
 552         ret = __btrfs_close_devices(fs_devices);
 553         if (!fs_devices->opened) {
 554                 seed_devices = fs_devices->seed;
 555                 fs_devices->seed = NULL;
 556         }
 557         mutex_unlock(&uuid_mutex);
 558
 559         while (seed_devices) {
 560                 fs_devices = seed_devices;
 561                 seed_devices = fs_devices->seed;
 562                 __btrfs_close_devices(fs_devices);
 563                 free_fs_devices(fs_devices);
 564         }
 565         return ret;
 566 }
 567
 568 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 569                                 fmode_t flags, void *holder)
 570 {
 571         struct request_queue *q;
 572         struct block_device *bdev;
 573         struct list_head *head = &fs_devices->devices;
 574         struct btrfs_device *device;
 575         struct block_device *latest_bdev = NULL;
 576         struct buffer_head *bh;
 577         struct btrfs_super_block *disk_super;
 578         u64 latest_devid = 0;
 579         u64 latest_transid = 0;
 580         u64 devid;
 581         int seeding = 1;
 582         int ret = 0;
 583
 584         flags |= FMODE_EXCL;
 585
 586         list_for_each_entry(device, head, dev_list) {
 587                 if (device->bdev)
 588                         continue;
 589                 if (!device->name)
 590                         continue;
 591
 592                 bdev = blkdev_get_by_path(device->name, flags, holder);
 593                 if (IS_ERR(bdev)) {
 594                         printk(KERN_INFO "open %s failed\n", device->name);
 595                         goto error;
 596                 }
 597                 set_blocksize(bdev, 4096);
 598
 599                 bh = btrfs_read_dev_super(bdev);
 600                 if (!bh) {
 601                         ret = -EINVAL;
 602                         goto error_close;
 603                 }
 604
 605                 disk_super = (struct btrfs_super_block *)bh->b_data;
 606                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 607                 if (devid != device->devid)
 608                         goto error_brelse;
 609
 610                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 611                            BTRFS_UUID_SIZE))
 612                         goto error_brelse;
 613
 614                 device->generation = btrfs_super_generation(disk_super);
 615                 if (!latest_transid || device->generation > latest_transid) {
 616                         latest_devid = devid;
 617                         latest_transid = device->generation;
 618                         latest_bdev = bdev;
 619                 }
 620
 621                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 622                         device->writeable = 0;
 623                 } else {
 624                         device->writeable = !bdev_read_only(bdev);
 625                         seeding = 0;
 626                 }
 627
 628                 q = bdev_get_queue(bdev);
 629                 if (blk_queue_discard(q)) {
 630                         device->can_discard = 1;
 631                         fs_devices->num_can_discard++;
 632                 }
 633
 634                 device->bdev = bdev;
 635                 device->in_fs_metadata = 0;
 636                 device->mode = flags;
 637
 638                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 639                         fs_devices->rotating = 1;
 640
 641                 fs_devices->open_devices++;
 642                 if (device->writeable) {
 643                         fs_devices->rw_devices++;
 644                         list_add(&device->dev_alloc_list,
 645                                  &fs_devices->alloc_list);
 646                 }
 647                 brelse(bh);
 648                 continue;
 649
 650 error_brelse:
 651                 brelse(bh);
 652 error_close:
 653                 blkdev_put(bdev, flags);
 654 error:
 655                 continue;
 656         }
 657         if (fs_devices->open_devices == 0) {
 658                 ret = -EIO;
 659                 goto out;
 660         }
 661         fs_devices->seeding = seeding;
 662         fs_devices->opened = 1;
 663         fs_devices->latest_bdev = latest_bdev;
 664         fs_devices->latest_devid = latest_devid;
 665         fs_devices->latest_trans = latest_transid;
 666         fs_devices->total_rw_bytes = 0;
 667 out:
 668         return ret;
 669 }
 670
 671 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 672                        fmode_t flags, void *holder)
 673 {
 674         int ret;
 675
 676         mutex_lock(&uuid_mutex);
 677         if (fs_devices->opened) {
 678                 fs_devices->opened++;
 679                 ret = 0;
 680         } else {
 681                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 682         }
 683         mutex_unlock(&uuid_mutex);
 684         return ret;
 685 }
 686
 687 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 688                           struct btrfs_fs_devices **fs_devices_ret)
 689 {
 690         struct btrfs_super_block *disk_super;
 691         struct block_device *bdev;
 692         struct buffer_head *bh;
 693         int ret;
 694         u64 devid;
 695         u64 transid;
 696
 697         mutex_lock(&uuid_mutex);
 698
 699         flags |= FMODE_EXCL;
 700         bdev = blkdev_get_by_path(path, flags, holder);
 701
 702         if (IS_ERR(bdev)) {
 703                 ret = PTR_ERR(bdev);
 704                 goto error;
 705         }
 706
 707         ret = set_blocksize(bdev, 4096);
 708         if (ret)
 709                 goto error_close;
 710         bh = btrfs_read_dev_super(bdev);
 711         if (!bh) {
 712                 ret = -EINVAL;
 713                 goto error_close;
 714         }
 715         disk_super = (struct btrfs_super_block *)bh->b_data;
 716         devid = btrfs_stack_device_id(&disk_super->dev_item);
 717         transid = btrfs_super_generation(disk_super);
 718         if (disk_super->label[0])
 719                 printk(KERN_INFO "device label %s ", disk_super->label);
 720         else
 721                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 722         printk(KERN_CONT "devid %llu transid %llu %s\n",
 723                (unsigned long long)devid, (unsigned long long)transid, path);
 724         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 725
 726         brelse(bh);
 727 error_close:
 728         blkdev_put(bdev, flags);
 729 error:
 730         mutex_unlock(&uuid_mutex);
 731         return ret;
 732 }
 733
 734 /* helper to account the used device space in the range */
 735 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 736                                    u64 end, u64 *length)
 737 {
 738         struct btrfs_key key;
 739         struct btrfs_root *root = device->dev_root;
 740         struct btrfs_dev_extent *dev_extent;
 741         struct btrfs_path *path;
 742         u64 extent_end;
 743         int ret;
 744         int slot;
 745         struct extent_buffer *l;
 746
 747         *length = 0;
 748
 749         if (start >= device->total_bytes)
 750                 return 0;
 751
 752         path = btrfs_alloc_path();
 753         if (!path)
 754                 return -ENOMEM;
 755         path->reada = 2;
 756
 757         key.objectid = device->devid;
 758         key.offset = start;
 759         key.type = BTRFS_DEV_EXTENT_KEY;
 760
 761         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 762         if (ret < 0)
 763                 goto out;
 764         if (ret > 0) {
 765                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 766                 if (ret < 0)
 767                         goto out;
 768         }
 769
 770         while (1) {
 771                 l = path->nodes[0];
 772                 slot = path->slots[0];
 773                 if (slot >= btrfs_header_nritems(l)) {
 774                         ret = btrfs_next_leaf(root, path);
 775                         if (ret == 0)
 776                                 continue;
 777                         if (ret < 0)
 778                                 goto out;
 779
 780                         break;
 781                 }
 782                 btrfs_item_key_to_cpu(l, &key, slot);
 783
 784                 if (key.objectid < device->devid)
 785                         goto next;
 786
 787                 if (key.objectid > device->devid)
 788                         break;
 789
 790                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 791                         goto next;
 792
 793                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 794                 extent_end = key.offset + btrfs_dev_extent_length(l,
 795                                                                   dev_extent);
 796                 if (key.offset <= start && extent_end > end) {
 797                         *length = end - start + 1;
 798                         break;
 799                 } else if (key.offset <= start && extent_end > start)
 800                         *length += extent_end - start;
 801                 else if (key.offset > start && extent_end <= end)
 802                         *length += extent_end - key.offset;
 803                 else if (key.offset > start && key.offset <= end) {
 804                         *length += end - key.offset + 1;
 805                         break;
 806                 } else if (key.offset > end)
 807                         break;
 808
 809 next:
 810                 path->slots[0]++;
 811         }
 812         ret = 0;
 813 out:
 814         btrfs_free_path(path);
 815         return ret;
 816 }
 817
 818 /*
 819  * find_free_dev_extent - find free space in the specified device
 820  * @trans:      transaction handler
 821  * @device:     the device which we search the free space in
 822  * @num_bytes:  the size of the free space that we need
 823  * @start:      store the start of the free space.
 824  * @len:        the size of the free space. that we find, or the size of the max
 825  *              free space if we don't find suitable free space
 826  *
 827  * this uses a pretty simple search, the expectation is that it is
 828  * called very infrequently and that a given device has a small number
 829  * of extents
 830  *
 831  * @start is used to store the start of the free space if we find. But if we
 832  * don't find suitable free space, it will be used to store the start position
 833  * of the max free space.
 834  *
 835  * @len is used to store the size of the free space that we find.
 836  * But if we don't find suitable free space, it is used to store the size of
 837  * the max free space.
 838  */
 839 int find_free_dev_extent(struct btrfs_trans_handle *trans,
 840                          struct btrfs_device *device, u64 num_bytes,
 841                          u64 *start, u64 *len)
 842 {
 843         struct btrfs_key key;
 844         struct btrfs_root *root = device->dev_root;
 845         struct btrfs_dev_extent *dev_extent;
 846         struct btrfs_path *path;
 847         u64 hole_size;
 848         u64 max_hole_start;
 849         u64 max_hole_size;
 850         u64 extent_end;
 851         u64 search_start;
 852         u64 search_end = device->total_bytes;
 853         int ret;
 854         int slot;
 855         struct extent_buffer *l;
 856
 857         /* FIXME use last free of some kind */
 858
 859         /* we don't want to overwrite the superblock on the drive,
 860          * so we make sure to start at an offset of at least 1MB
 861          */
 862         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
 863
 864         max_hole_start = search_start;
 865         max_hole_size = 0;
 866         hole_size = 0;
 867
 868         if (search_start >= search_end) {
 869                 ret = -ENOSPC;
 870                 goto error;
 871         }
 872
 873         path = btrfs_alloc_path();
 874         if (!path) {
 875                 ret = -ENOMEM;
 876                 goto error;
 877         }
 878         path->reada = 2;
 879
 880         key.objectid = device->devid;
 881         key.offset = search_start;
 882         key.type = BTRFS_DEV_EXTENT_KEY;
 883
 884         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
 885         if (ret < 0)
 886                 goto out;
 887         if (ret > 0) {
 888                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 889                 if (ret < 0)
 890                         goto out;
 891         }
 892
 893         while (1) {
 894                 l = path->nodes[0];
 895                 slot = path->slots[0];
 896                 if (slot >= btrfs_header_nritems(l)) {
 897                         ret = btrfs_next_leaf(root, path);
 898                         if (ret == 0)
 899                                 continue;
 900                         if (ret < 0)
 901                                 goto out;
 902
 903                         break;
 904                 }
 905                 btrfs_item_key_to_cpu(l, &key, slot);
 906
 907                 if (key.objectid < device->devid)
 908                         goto next;
 909
 910                 if (key.objectid > device->devid)
 911                         break;
 912
 913                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 914                         goto next;
 915
 916                 if (key.offset > search_start) {
 917                         hole_size = key.offset - search_start;
 918
 919                         if (hole_size > max_hole_size) {
 920                                 max_hole_start = search_start;
 921                                 max_hole_size = hole_size;
 922                         }
 923
 924                         /*
 925                          * If this free space is greater than which we need,
 926                          * it must be the max free space that we have found
 927                          * until now, so max_hole_start must point to the start
 928                          * of this free space and the length of this free space
 929                          * is stored in max_hole_size. Thus, we return
 930                          * max_hole_start and max_hole_size and go back to the
 931                          * caller.
 932                          */
 933                         if (hole_size >= num_bytes) {
 934                                 ret = 0;
 935                                 goto out;
 936                         }
 937                 }
 938
 939                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 940                 extent_end = key.offset + btrfs_dev_extent_length(l,
 941                                                                   dev_extent);
 942                 if (extent_end > search_start)
 943                         search_start = extent_end;
 944 next:
 945                 path->slots[0]++;
 946                 cond_resched();
 947         }
 948
 949         /*
 950          * At this point, search_start should be the end of
 951          * allocated dev extents, and when shrinking the device,
 952          * search_end may be smaller than search_start.
 953          */
 954         if (search_end > search_start)
 955                 hole_size = search_end - search_start;
 956
 957         if (hole_size > max_hole_size) {
 958                 max_hole_start = search_start;
 959                 max_hole_size = hole_size;
 960         }
 961
 962         /* See above. */
 963         if (hole_size < num_bytes)
 964                 ret = -ENOSPC;
 965         else
 966                 ret = 0;
 967
 968 out:
 969         btrfs_free_path(path);
 970 error:
 971         *start = max_hole_start;
 972         if (len)
 973                 *len = max_hole_size;
 974         return ret;
 975 }
 976
 977 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
 978                           struct btrfs_device *device,
 979                           u64 start)
 980 {
 981         int ret;
 982         struct btrfs_path *path;
 983         struct btrfs_root *root = device->dev_root;
 984         struct btrfs_key key;
 985         struct btrfs_key found_key;
 986         struct extent_buffer *leaf = NULL;
 987         struct btrfs_dev_extent *extent = NULL;
 988
 989         path = btrfs_alloc_path();
 990         if (!path)
 991                 return -ENOMEM;
 992
 993         key.objectid = device->devid;
 994         key.offset = start;
 995         key.type = BTRFS_DEV_EXTENT_KEY;
 996
 997         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 998         if (ret > 0) {
 999                 ret = btrfs_previous_item(root, path, key.objectid,
1000                                           BTRFS_DEV_EXTENT_KEY);
1001                 if (ret)
1002                         goto out;
1003                 leaf = path->nodes[0];
1004                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1005                 extent = btrfs_item_ptr(leaf, path->slots[0],
1006                                         struct btrfs_dev_extent);
1007                 BUG_ON(found_key.offset > start || found_key.offset +
1008                        btrfs_dev_extent_length(leaf, extent) < start);
1009         } else if (ret == 0) {
1010                 leaf = path->nodes[0];
1011                 extent = btrfs_item_ptr(leaf, path->slots[0],
1012                                         struct btrfs_dev_extent);
1013         }
1014         BUG_ON(ret);
1015
1016         if (device->bytes_used > 0) {
1017                 u64 len = btrfs_dev_extent_length(leaf, extent);
1018                 device->bytes_used -= len;
1019                 spin_lock(&root->fs_info->free_chunk_lock);
1020                 root->fs_info->free_chunk_space += len;
1021                 spin_unlock(&root->fs_info->free_chunk_lock);
1022         }
1023         ret = btrfs_del_item(trans, root, path);
1024
1025 out:
1026         btrfs_free_path(path);
1027         return ret;
1028 }
1029
1030 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1031                            struct btrfs_device *device,
1032                            u64 chunk_tree, u64 chunk_objectid,
1033                            u64 chunk_offset, u64 start, u64 num_bytes)
1034 {
1035         int ret;
1036         struct btrfs_path *path;
1037         struct btrfs_root *root = device->dev_root;
1038         struct btrfs_dev_extent *extent;
1039         struct extent_buffer *leaf;
1040         struct btrfs_key key;
1041
1042         WARN_ON(!device->in_fs_metadata);
1043         path = btrfs_alloc_path();
1044         if (!path)
1045                 return -ENOMEM;
1046
1047         key.objectid = device->devid;
1048         key.offset = start;
1049         key.type = BTRFS_DEV_EXTENT_KEY;
1050         ret = btrfs_insert_empty_item(trans, root, path, &key,
1051                                       sizeof(*extent));
1052         BUG_ON(ret);
1053
1054         leaf = path->nodes[0];
1055         extent = btrfs_item_ptr(leaf, path->slots[0],
1056                                 struct btrfs_dev_extent);
1057         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1058         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1059         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1060
1061         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1062                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1063                     BTRFS_UUID_SIZE);
1064
1065         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1066         btrfs_mark_buffer_dirty(leaf);
1067         btrfs_free_path(path);
1068         return ret;
1069 }
1070
1071 static noinline int find_next_chunk(struct btrfs_root *root,
1072                                     u64 objectid, u64 *offset)
1073 {
1074         struct btrfs_path *path;
1075         int ret;
1076         struct btrfs_key key;
1077         struct btrfs_chunk *chunk;
1078         struct btrfs_key found_key;
1079
1080         path = btrfs_alloc_path();
1081         if (!path)
1082                 return -ENOMEM;
1083
1084         key.objectid = objectid;
1085         key.offset = (u64)-1;
1086         key.type = BTRFS_CHUNK_ITEM_KEY;
1087
1088         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1089         if (ret < 0)
1090                 goto error;
1091
1092         BUG_ON(ret == 0);
1093
1094         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1095         if (ret) {
1096                 *offset = 0;
1097         } else {
1098                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1099                                       path->slots[0]);
1100                 if (found_key.objectid != objectid)
1101                         *offset = 0;
1102                 else {
1103                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1104                                                struct btrfs_chunk);
1105                         *offset = found_key.offset +
1106                                 btrfs_chunk_length(path->nodes[0], chunk);
1107                 }
1108         }
1109         ret = 0;
1110 error:
1111         btrfs_free_path(path);
1112         return ret;
1113 }
1114
1115 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1116 {
1117         int ret;
1118         struct btrfs_key key;
1119         struct btrfs_key found_key;
1120         struct btrfs_path *path;
1121
1122         root = root->fs_info->chunk_root;
1123
1124         path = btrfs_alloc_path();
1125         if (!path)
1126                 return -ENOMEM;
1127
1128         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1129         key.type = BTRFS_DEV_ITEM_KEY;
1130         key.offset = (u64)-1;
1131
1132         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1133         if (ret < 0)
1134                 goto error;
1135
1136         BUG_ON(ret == 0);
1137
1138         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1139                                   BTRFS_DEV_ITEM_KEY);
1140         if (ret) {
1141                 *objectid = 1;
1142         } else {
1143                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1144                                       path->slots[0]);
1145                 *objectid = found_key.offset + 1;
1146         }
1147         ret = 0;
1148 error:
1149         btrfs_free_path(path);
1150         return ret;
1151 }
1152
1153 /*
1154  * the device information is stored in the chunk root
1155  * the btrfs_device struct should be fully filled in
1156  */
1157 int btrfs_add_device(struct btrfs_trans_handle *trans,
1158                      struct btrfs_root *root,
1159                      struct btrfs_device *device)
1160 {
1161         int ret;
1162         struct btrfs_path *path;
1163         struct btrfs_dev_item *dev_item;
1164         struct extent_buffer *leaf;
1165         struct btrfs_key key;
1166         unsigned long ptr;
1167
1168         root = root->fs_info->chunk_root;
1169
1170         path = btrfs_alloc_path();
1171         if (!path)
1172                 return -ENOMEM;
1173
1174         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1175         key.type = BTRFS_DEV_ITEM_KEY;
1176         key.offset = device->devid;
1177
1178         ret = btrfs_insert_empty_item(trans, root, path, &key,
1179                                       sizeof(*dev_item));
1180         if (ret)
1181                 goto out;
1182
1183         leaf = path->nodes[0];
1184         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1185
1186         btrfs_set_device_id(leaf, dev_item, device->devid);
1187         btrfs_set_device_generation(leaf, dev_item, 0);
1188         btrfs_set_device_type(leaf, dev_item, device->type);
1189         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1190         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1191         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1192         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1193         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1194         btrfs_set_device_group(leaf, dev_item, 0);
1195         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1196         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1197         btrfs_set_device_start_offset(leaf, dev_item, 0);
1198
1199         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1200         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1201         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1202         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1203         btrfs_mark_buffer_dirty(leaf);
1204
1205         ret = 0;
1206 out:
1207         btrfs_free_path(path);
1208         return ret;
1209 }
1210
1211 static int btrfs_rm_dev_item(struct btrfs_root *root,
1212                              struct btrfs_device *device)
1213 {
1214         int ret;
1215         struct btrfs_path *path;
1216         struct btrfs_key key;
1217         struct btrfs_trans_handle *trans;
1218
1219         root = root->fs_info->chunk_root;
1220
1221         path = btrfs_alloc_path();
1222         if (!path)
1223                 return -ENOMEM;
1224
1225         trans = btrfs_start_transaction(root, 0);
1226         if (IS_ERR(trans)) {
1227                 btrfs_free_path(path);
1228                 return PTR_ERR(trans);
1229         }
1230         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1231         key.type = BTRFS_DEV_ITEM_KEY;
1232         key.offset = device->devid;
1233         lock_chunks(root);
1234
1235         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1236         if (ret < 0)
1237                 goto out;
1238
1239         if (ret > 0) {
1240                 ret = -ENOENT;
1241                 goto out;
1242         }
1243
1244         ret = btrfs_del_item(trans, root, path);
1245         if (ret)
1246                 goto out;
1247 out:
1248         btrfs_free_path(path);
1249         unlock_chunks(root);
1250         btrfs_commit_transaction(trans, root);
1251         return ret;
1252 }
1253
1254 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1255 {
1256         struct btrfs_device *device;
1257         struct btrfs_device *next_device;
1258         struct block_device *bdev;
1259         struct buffer_head *bh = NULL;
1260         struct btrfs_super_block *disk_super;
1261         struct btrfs_fs_devices *cur_devices;
1262         u64 all_avail;
1263         u64 devid;
1264         u64 num_devices;
1265         u8 *dev_uuid;
1266         int ret = 0;
1267         bool clear_super = false;
1268
1269         mutex_lock(&uuid_mutex);
1270         mutex_lock(&root->fs_info->volume_mutex);
1271
1272         all_avail = root->fs_info->avail_data_alloc_bits |
1273                 root->fs_info->avail_system_alloc_bits |
1274                 root->fs_info->avail_metadata_alloc_bits;
1275
1276         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1277             root->fs_info->fs_devices->num_devices <= 4) {
1278                 printk(KERN_ERR "btrfs: unable to go below four devices "
1279                        "on raid10\n");
1280                 ret = -EINVAL;
1281                 goto out;
1282         }
1283
1284         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1285             root->fs_info->fs_devices->num_devices <= 2) {
1286                 printk(KERN_ERR "btrfs: unable to go below two "
1287                        "devices on raid1\n");
1288                 ret = -EINVAL;
1289                 goto out;
1290         }
1291
1292         if (strcmp(device_path, "missing") == 0) {
1293                 struct list_head *devices;
1294                 struct btrfs_device *tmp;
1295
1296                 device = NULL;
1297                 devices = &root->fs_info->fs_devices->devices;
1298                 /*
1299                  * It is safe to read the devices since the volume_mutex
1300                  * is held.
1301                  */
1302                 list_for_each_entry(tmp, devices, dev_list) {
1303                         if (tmp->in_fs_metadata && !tmp->bdev) {
1304                                 device = tmp;
1305                                 break;
1306                         }
1307                 }
1308                 bdev = NULL;
1309                 bh = NULL;
1310                 disk_super = NULL;
1311                 if (!device) {
1312                         printk(KERN_ERR "btrfs: no missing devices found to "
1313                                "remove\n");
1314                         goto out;
1315                 }
1316         } else {
1317                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1318                                           root->fs_info->bdev_holder);
1319                 if (IS_ERR(bdev)) {
1320                         ret = PTR_ERR(bdev);
1321                         goto out;
1322                 }
1323
1324                 set_blocksize(bdev, 4096);
1325                 bh = btrfs_read_dev_super(bdev);
1326                 if (!bh) {
1327                         ret = -EINVAL;
1328                         goto error_close;
1329                 }
1330                 disk_super = (struct btrfs_super_block *)bh->b_data;
1331                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1332                 dev_uuid = disk_super->dev_item.uuid;
1333                 device = btrfs_find_device(root, devid, dev_uuid,
1334                                            disk_super->fsid);
1335                 if (!device) {
1336                         ret = -ENOENT;
1337                         goto error_brelse;
1338                 }
1339         }
1340
1341         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1342                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1343                        "device\n");
1344                 ret = -EINVAL;
1345                 goto error_brelse;
1346         }
1347
1348         if (device->writeable) {
1349                 lock_chunks(root);
1350                 list_del_init(&device->dev_alloc_list);
1351                 unlock_chunks(root);
1352                 root->fs_info->fs_devices->rw_devices--;
1353                 clear_super = true;
1354         }
1355
1356         ret = btrfs_shrink_device(device, 0);
1357         if (ret)
1358                 goto error_undo;
1359
1360         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1361         if (ret)
1362                 goto error_undo;
1363
1364         spin_lock(&root->fs_info->free_chunk_lock);
1365         root->fs_info->free_chunk_space = device->total_bytes -
1366                 device->bytes_used;
1367         spin_unlock(&root->fs_info->free_chunk_lock);
1368
1369         device->in_fs_metadata = 0;
1370         btrfs_scrub_cancel_dev(root, device);
1371
1372         /*
1373          * the device list mutex makes sure that we don't change
1374          * the device list while someone else is writing out all
1375          * the device supers.
1376          */
1377
1378         cur_devices = device->fs_devices;
1379         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1380         list_del_rcu(&device->dev_list);
1381
1382         device->fs_devices->num_devices--;
1383
1384         if (device->missing)
1385                 root->fs_info->fs_devices->missing_devices--;
1386
1387         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1388                                  struct btrfs_device, dev_list);
1389         if (device->bdev == root->fs_info->sb->s_bdev)
1390                 root->fs_info->sb->s_bdev = next_device->bdev;
1391         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1392                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1393
1394         if (device->bdev)
1395                 device->fs_devices->open_devices--;
1396
1397         call_rcu(&device->rcu, free_device);
1398         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1399
1400         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1401         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1402
1403         if (cur_devices->open_devices == 0) {
1404                 struct btrfs_fs_devices *fs_devices;
1405                 fs_devices = root->fs_info->fs_devices;
1406                 while (fs_devices) {
1407                         if (fs_devices->seed == cur_devices)
1408                                 break;
1409                         fs_devices = fs_devices->seed;
1410                 }
1411                 fs_devices->seed = cur_devices->seed;
1412                 cur_devices->seed = NULL;
1413                 lock_chunks(root);
1414                 __btrfs_close_devices(cur_devices);
1415                 unlock_chunks(root);
1416                 free_fs_devices(cur_devices);
1417         }
1418
1419         /*
1420          * at this point, the device is zero sized.  We want to
1421          * remove it from the devices list and zero out the old super
1422          */
1423         if (clear_super) {
1424                 /* make sure this device isn't detected as part of
1425                  * the FS anymore
1426                  */
1427                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1428                 set_buffer_dirty(bh);
1429                 sync_dirty_buffer(bh);
1430         }
1431
1432         ret = 0;
1433
1434 error_brelse:
1435         brelse(bh);
1436 error_close:
1437         if (bdev)
1438                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1439 out:
1440         mutex_unlock(&root->fs_info->volume_mutex);
1441         mutex_unlock(&uuid_mutex);
1442         return ret;
1443 error_undo:
1444         if (device->writeable) {
1445                 lock_chunks(root);
1446                 list_add(&device->dev_alloc_list,
1447                          &root->fs_info->fs_devices->alloc_list);
1448                 unlock_chunks(root);
1449                 root->fs_info->fs_devices->rw_devices++;
1450         }
1451         goto error_brelse;
1452 }
1453
1454 /*
1455  * does all the dirty work required for changing file system's UUID.
1456  */
1457 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1458                                 struct btrfs_root *root)
1459 {
1460         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1461         struct btrfs_fs_devices *old_devices;
1462         struct btrfs_fs_devices *seed_devices;
1463         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1464         struct btrfs_device *device;
1465         u64 super_flags;
1466
1467         BUG_ON(!mutex_is_locked(&uuid_mutex));
1468         if (!fs_devices->seeding)
1469                 return -EINVAL;
1470
1471         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1472         if (!seed_devices)
1473                 return -ENOMEM;
1474
1475         old_devices = clone_fs_devices(fs_devices);
1476         if (IS_ERR(old_devices)) {
1477                 kfree(seed_devices);
1478                 return PTR_ERR(old_devices);
1479         }
1480
1481         list_add(&old_devices->list, &fs_uuids);
1482
1483         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1484         seed_devices->opened = 1;
1485         INIT_LIST_HEAD(&seed_devices->devices);
1486         INIT_LIST_HEAD(&seed_devices->alloc_list);
1487         mutex_init(&seed_devices->device_list_mutex);
1488
1489         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1490         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1491                               synchronize_rcu);
1492         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1493
1494         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1495         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1496                 device->fs_devices = seed_devices;
1497         }
1498
1499         fs_devices->seeding = 0;
1500         fs_devices->num_devices = 0;
1501         fs_devices->open_devices = 0;
1502         fs_devices->seed = seed_devices;
1503
1504         generate_random_uuid(fs_devices->fsid);
1505         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1506         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1507         super_flags = btrfs_super_flags(disk_super) &
1508                       ~BTRFS_SUPER_FLAG_SEEDING;
1509         btrfs_set_super_flags(disk_super, super_flags);
1510
1511         return 0;
1512 }
1513
1514 /*
1515  * strore the expected generation for seed devices in device items.
1516  */
1517 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1518                                struct btrfs_root *root)
1519 {
1520         struct btrfs_path *path;
1521         struct extent_buffer *leaf;
1522         struct btrfs_dev_item *dev_item;
1523         struct btrfs_device *device;
1524         struct btrfs_key key;
1525         u8 fs_uuid[BTRFS_UUID_SIZE];
1526         u8 dev_uuid[BTRFS_UUID_SIZE];
1527         u64 devid;
1528         int ret;
1529
1530         path = btrfs_alloc_path();
1531         if (!path)
1532                 return -ENOMEM;
1533
1534         root = root->fs_info->chunk_root;
1535         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1536         key.offset = 0;
1537         key.type = BTRFS_DEV_ITEM_KEY;
1538
1539         while (1) {
1540                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1541                 if (ret < 0)
1542                         goto error;
1543
1544                 leaf = path->nodes[0];
1545 next_slot:
1546                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1547                         ret = btrfs_next_leaf(root, path);
1548                         if (ret > 0)
1549                                 break;
1550                         if (ret < 0)
1551                                 goto error;
1552                         leaf = path->nodes[0];
1553                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1554                         btrfs_release_path(path);
1555                         continue;
1556                 }
1557
1558                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1559                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1560                     key.type != BTRFS_DEV_ITEM_KEY)
1561                         break;
1562
1563                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1564                                           struct btrfs_dev_item);
1565                 devid = btrfs_device_id(leaf, dev_item);
1566                 read_extent_buffer(leaf, dev_uuid,
1567                                    (unsigned long)btrfs_device_uuid(dev_item),
1568                                    BTRFS_UUID_SIZE);
1569                 read_extent_buffer(leaf, fs_uuid,
1570                                    (unsigned long)btrfs_device_fsid(dev_item),
1571                                    BTRFS_UUID_SIZE);
1572                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1573                 BUG_ON(!device);
1574
1575                 if (device->fs_devices->seeding) {
1576                         btrfs_set_device_generation(leaf, dev_item,
1577                                                     device->generation);
1578                         btrfs_mark_buffer_dirty(leaf);
1579                 }
1580
1581                 path->slots[0]++;
1582                 goto next_slot;
1583         }
1584         ret = 0;
1585 error:
1586         btrfs_free_path(path);
1587         return ret;
1588 }
1589
1590 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1591 {
1592         struct request_queue *q;
1593         struct btrfs_trans_handle *trans;
1594         struct btrfs_device *device;
1595         struct block_device *bdev;
1596         struct list_head *devices;
1597         struct super_block *sb = root->fs_info->sb;
1598         u64 total_bytes;
1599         int seeding_dev = 0;
1600         int ret = 0;
1601
1602         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1603                 return -EINVAL;
1604
1605         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1606                                   root->fs_info->bdev_holder);
1607         if (IS_ERR(bdev))
1608                 return PTR_ERR(bdev);
1609
1610         if (root->fs_info->fs_devices->seeding) {
1611                 seeding_dev = 1;
1612                 down_write(&sb->s_umount);
1613                 mutex_lock(&uuid_mutex);
1614         }
1615
1616         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1617         mutex_lock(&root->fs_info->volume_mutex);
1618
1619         devices = &root->fs_info->fs_devices->devices;
1620         /*
1621          * we have the volume lock, so we don't need the extra
1622          * device list mutex while reading the list here.
1623          */
1624         list_for_each_entry(device, devices, dev_list) {
1625                 if (device->bdev == bdev) {
1626                         ret = -EEXIST;
1627                         goto error;
1628                 }
1629         }
1630
1631         device = kzalloc(sizeof(*device), GFP_NOFS);
1632         if (!device) {
1633                 /* we can safely leave the fs_devices entry around */
1634                 ret = -ENOMEM;
1635                 goto error;
1636         }
1637
1638         device->name = kstrdup(device_path, GFP_NOFS);
1639         if (!device->name) {
1640                 kfree(device);
1641                 ret = -ENOMEM;
1642                 goto error;
1643         }
1644
1645         ret = find_next_devid(root, &device->devid);
1646         if (ret) {
1647                 kfree(device->name);
1648                 kfree(device);
1649                 goto error;
1650         }
1651
1652         trans = btrfs_start_transaction(root, 0);
1653         if (IS_ERR(trans)) {
1654                 kfree(device->name);
1655                 kfree(device);
1656                 ret = PTR_ERR(trans);
1657                 goto error;
1658         }
1659
1660         lock_chunks(root);
1661
1662         q = bdev_get_queue(bdev);
1663         if (blk_queue_discard(q))
1664                 device->can_discard = 1;
1665         device->writeable = 1;
1666         device->work.func = pending_bios_fn;
1667         generate_random_uuid(device->uuid);
1668         spin_lock_init(&device->io_lock);
1669         device->generation = trans->transid;
1670         device->io_width = root->sectorsize;
1671         device->io_align = root->sectorsize;
1672         device->sector_size = root->sectorsize;
1673         device->total_bytes = i_size_read(bdev->bd_inode);
1674         device->disk_total_bytes = device->total_bytes;
1675         device->dev_root = root->fs_info->dev_root;
1676         device->bdev = bdev;
1677         device->in_fs_metadata = 1;
1678         device->mode = FMODE_EXCL;
1679         set_blocksize(device->bdev, 4096);
1680
1681         if (seeding_dev) {
1682                 sb->s_flags &= ~MS_RDONLY;
1683                 ret = btrfs_prepare_sprout(trans, root);
1684                 BUG_ON(ret);
1685         }
1686
1687         device->fs_devices = root->fs_info->fs_devices;
1688
1689         /*
1690          * we don't want write_supers to jump in here with our device
1691          * half setup
1692          */
1693         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1694         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1695         list_add(&device->dev_alloc_list,
1696                  &root->fs_info->fs_devices->alloc_list);
1697         root->fs_info->fs_devices->num_devices++;
1698         root->fs_info->fs_devices->open_devices++;
1699         root->fs_info->fs_devices->rw_devices++;
1700         if (device->can_discard)
1701                 root->fs_info->fs_devices->num_can_discard++;
1702         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1703
1704         spin_lock(&root->fs_info->free_chunk_lock);
1705         root->fs_info->free_chunk_space += device->total_bytes;
1706         spin_unlock(&root->fs_info->free_chunk_lock);
1707
1708         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1709                 root->fs_info->fs_devices->rotating = 1;
1710
1711         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1712         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1713                                     total_bytes + device->total_bytes);
1714
1715         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1716         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1717                                     total_bytes + 1);
1718         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1719
1720         if (seeding_dev) {
1721                 ret = init_first_rw_device(trans, root, device);
1722                 BUG_ON(ret);
1723                 ret = btrfs_finish_sprout(trans, root);
1724                 BUG_ON(ret);
1725         } else {
1726                 ret = btrfs_add_device(trans, root, device);
1727         }
1728
1729         /*
1730          * we've got more storage, clear any full flags on the space
1731          * infos
1732          */
1733         btrfs_clear_space_info_full(root->fs_info);
1734
1735         unlock_chunks(root);
1736         btrfs_commit_transaction(trans, root);
1737
1738         if (seeding_dev) {
1739                 mutex_unlock(&uuid_mutex);
1740                 up_write(&sb->s_umount);
1741
1742                 ret = btrfs_relocate_sys_chunks(root);
1743                 BUG_ON(ret);
1744         }
1745 out:
1746         mutex_unlock(&root->fs_info->volume_mutex);
1747         return ret;
1748 error:
1749         blkdev_put(bdev, FMODE_EXCL);
1750         if (seeding_dev) {
1751                 mutex_unlock(&uuid_mutex);
1752                 up_write(&sb->s_umount);
1753         }
1754         goto out;
1755 }
1756
1757 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1758                                         struct btrfs_device *device)
1759 {
1760         int ret;
1761         struct btrfs_path *path;
1762         struct btrfs_root *root;
1763         struct btrfs_dev_item *dev_item;
1764         struct extent_buffer *leaf;
1765         struct btrfs_key key;
1766
1767         root = device->dev_root->fs_info->chunk_root;
1768
1769         path = btrfs_alloc_path();
1770         if (!path)
1771                 return -ENOMEM;
1772
1773         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1774         key.type = BTRFS_DEV_ITEM_KEY;
1775         key.offset = device->devid;
1776
1777         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1778         if (ret < 0)
1779                 goto out;
1780
1781         if (ret > 0) {
1782                 ret = -ENOENT;
1783                 goto out;
1784         }
1785
1786         leaf = path->nodes[0];
1787         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1788
1789         btrfs_set_device_id(leaf, dev_item, device->devid);
1790         btrfs_set_device_type(leaf, dev_item, device->type);
1791         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1792         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1793         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1794         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1795         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1796         btrfs_mark_buffer_dirty(leaf);
1797
1798 out:
1799         btrfs_free_path(path);
1800         return ret;
1801 }
1802
1803 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1804                       struct btrfs_device *device, u64 new_size)
1805 {
1806         struct btrfs_super_block *super_copy =
1807                 &device->dev_root->fs_info->super_copy;
1808         u64 old_total = btrfs_super_total_bytes(super_copy);
1809         u64 diff = new_size - device->total_bytes;
1810
1811         if (!device->writeable)
1812                 return -EACCES;
1813         if (new_size <= device->total_bytes)
1814                 return -EINVAL;
1815
1816         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1817         device->fs_devices->total_rw_bytes += diff;
1818
1819         device->total_bytes = new_size;
1820         device->disk_total_bytes = new_size;
1821         btrfs_clear_space_info_full(device->dev_root->fs_info);
1822
1823         return btrfs_update_device(trans, device);
1824 }
1825
1826 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1827                       struct btrfs_device *device, u64 new_size)
1828 {
1829         int ret;
1830         lock_chunks(device->dev_root);
1831         ret = __btrfs_grow_device(trans, device, new_size);
1832         unlock_chunks(device->dev_root);
1833         return ret;
1834 }
1835
1836 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1837                             struct btrfs_root *root,
1838                             u64 chunk_tree, u64 chunk_objectid,
1839                             u64 chunk_offset)
1840 {
1841         int ret;
1842         struct btrfs_path *path;
1843         struct btrfs_key key;
1844
1845         root = root->fs_info->chunk_root;
1846         path = btrfs_alloc_path();
1847         if (!path)
1848                 return -ENOMEM;
1849
1850         key.objectid = chunk_objectid;
1851         key.offset = chunk_offset;
1852         key.type = BTRFS_CHUNK_ITEM_KEY;
1853
1854         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1855         BUG_ON(ret);
1856
1857         ret = btrfs_del_item(trans, root, path);
1858
1859         btrfs_free_path(path);
1860         return ret;
1861 }
1862
1863 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1864                         chunk_offset)
1865 {
1866         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1867         struct btrfs_disk_key *disk_key;
1868         struct btrfs_chunk *chunk;
1869         u8 *ptr;
1870         int ret = 0;
1871         u32 num_stripes;
1872         u32 array_size;
1873         u32 len = 0;
1874         u32 cur;
1875         struct btrfs_key key;
1876
1877         array_size = btrfs_super_sys_array_size(super_copy);
1878
1879         ptr = super_copy->sys_chunk_array;
1880         cur = 0;
1881
1882         while (cur < array_size) {
1883                 disk_key = (struct btrfs_disk_key *)ptr;
1884                 btrfs_disk_key_to_cpu(&key, disk_key);
1885
1886                 len = sizeof(*disk_key);
1887
1888                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1889                         chunk = (struct btrfs_chunk *)(ptr + len);
1890                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1891                         len += btrfs_chunk_item_size(num_stripes);
1892                 } else {
1893                         ret = -EIO;
1894                         break;
1895                 }
1896                 if (key.objectid == chunk_objectid &&
1897                     key.offset == chunk_offset) {
1898                         memmove(ptr, ptr + len, array_size - (cur + len));
1899                         array_size -= len;
1900                         btrfs_set_super_sys_array_size(super_copy, array_size);
1901                 } else {
1902                         ptr += len;
1903                         cur += len;
1904                 }
1905         }
1906         return ret;
1907 }
1908
1909 static int btrfs_relocate_chunk(struct btrfs_root *root,
1910                          u64 chunk_tree, u64 chunk_objectid,
1911                          u64 chunk_offset)
1912 {
1913         struct extent_map_tree *em_tree;
1914         struct btrfs_root *extent_root;
1915         struct btrfs_trans_handle *trans;
1916         struct extent_map *em;
1917         struct map_lookup *map;
1918         int ret;
1919         int i;
1920
1921         root = root->fs_info->chunk_root;
1922         extent_root = root->fs_info->extent_root;
1923         em_tree = &root->fs_info->mapping_tree.map_tree;
1924
1925         ret = btrfs_can_relocate(extent_root, chunk_offset);
1926         if (ret)
1927                 return -ENOSPC;
1928
1929         /* step one, relocate all the extents inside this chunk */
1930         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1931         if (ret)
1932                 return ret;
1933
1934         trans = btrfs_start_transaction(root, 0);
1935         BUG_ON(IS_ERR(trans));
1936
1937         lock_chunks(root);
1938
1939         /*
1940          * step two, delete the device extents and the
1941          * chunk tree entries
1942          */
1943         read_lock(&em_tree->lock);
1944         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1945         read_unlock(&em_tree->lock);
1946
1947         BUG_ON(em->start > chunk_offset ||
1948                em->start + em->len < chunk_offset);
1949         map = (struct map_lookup *)em->bdev;
1950
1951         for (i = 0; i < map->num_stripes; i++) {
1952                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1953                                             map->stripes[i].physical);
1954                 BUG_ON(ret);
1955
1956                 if (map->stripes[i].dev) {
1957                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1958                         BUG_ON(ret);
1959                 }
1960         }
1961         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1962                                chunk_offset);
1963
1964         BUG_ON(ret);
1965
1966         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1967
1968         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1969                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1970                 BUG_ON(ret);
1971         }
1972
1973         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1974         BUG_ON(ret);
1975
1976         write_lock(&em_tree->lock);
1977         remove_extent_mapping(em_tree, em);
1978         write_unlock(&em_tree->lock);
1979
1980         kfree(map);
1981         em->bdev = NULL;
1982
1983         /* once for the tree */
1984         free_extent_map(em);
1985         /* once for us */
1986         free_extent_map(em);
1987
1988         unlock_chunks(root);
1989         btrfs_end_transaction(trans, root);
1990         return 0;
1991 }
1992
1993 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1994 {
1995         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1996         struct btrfs_path *path;
1997         struct extent_buffer *leaf;
1998         struct btrfs_chunk *chunk;
1999         struct btrfs_key key;
2000         struct btrfs_key found_key;
2001         u64 chunk_tree = chunk_root->root_key.objectid;
2002         u64 chunk_type;
2003         bool retried = false;
2004         int failed = 0;
2005         int ret;
2006
2007         path = btrfs_alloc_path();
2008         if (!path)
2009                 return -ENOMEM;
2010
2011 again:
2012         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2013         key.offset = (u64)-1;
2014         key.type = BTRFS_CHUNK_ITEM_KEY;
2015
2016         while (1) {
2017                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2018                 if (ret < 0)
2019                         goto error;
2020                 BUG_ON(ret == 0);
2021
2022                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2023                                           key.type);
2024                 if (ret < 0)
2025                         goto error;
2026                 if (ret > 0)
2027                         break;
2028
2029                 leaf = path->nodes[0];
2030                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2031
2032                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2033                                        struct btrfs_chunk);
2034                 chunk_type = btrfs_chunk_type(leaf, chunk);
2035                 btrfs_release_path(path);
2036
2037                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2038                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2039                                                    found_key.objectid,
2040                                                    found_key.offset);
2041                         if (ret == -ENOSPC)
2042                                 failed++;
2043                         else if (ret)
2044                                 BUG();
2045                 }
2046
2047                 if (found_key.offset == 0)
2048                         break;
2049                 key.offset = found_key.offset - 1;
2050         }
2051         ret = 0;
2052         if (failed && !retried) {
2053                 failed = 0;
2054                 retried = true;
2055                 goto again;
2056         } else if (failed && retried) {
2057                 WARN_ON(1);
2058                 ret = -ENOSPC;
2059         }
2060 error:
2061         btrfs_free_path(path);
2062         return ret;
2063 }
2064
2065 static u64 div_factor(u64 num, int factor)
2066 {
2067         if (factor == 10)
2068                 return num;
2069         num *= factor;
2070         do_div(num, 10);
2071         return num;
2072 }
2073
2074 int btrfs_balance(struct btrfs_root *dev_root)
2075 {
2076         int ret;
2077         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2078         struct btrfs_device *device;
2079         u64 old_size;
2080         u64 size_to_free;
2081         struct btrfs_path *path;
2082         struct btrfs_key key;
2083         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2084         struct btrfs_trans_handle *trans;
2085         struct btrfs_key found_key;
2086
2087         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2088                 return -EROFS;
2089
2090         if (!capable(CAP_SYS_ADMIN))
2091                 return -EPERM;
2092
2093         mutex_lock(&dev_root->fs_info->volume_mutex);
2094         dev_root = dev_root->fs_info->dev_root;
2095
2096         /* step one make some room on all the devices */
2097         list_for_each_entry(device, devices, dev_list) {
2098                 old_size = device->total_bytes;
2099                 size_to_free = div_factor(old_size, 1);
2100                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2101                 if (!device->writeable ||
2102                     device->total_bytes - device->bytes_used > size_to_free)
2103                         continue;
2104
2105                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2106                 if (ret == -ENOSPC)
2107                         break;
2108                 BUG_ON(ret);
2109
2110                 trans = btrfs_start_transaction(dev_root, 0);
2111                 BUG_ON(IS_ERR(trans));
2112
2113                 ret = btrfs_grow_device(trans, device, old_size);
2114                 BUG_ON(ret);
2115
2116                 btrfs_end_transaction(trans, dev_root);
2117         }
2118
2119         /* step two, relocate all the chunks */
2120         path = btrfs_alloc_path();
2121         if (!path) {
2122                 ret = -ENOMEM;
2123                 goto error;
2124         }
2125         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2126         key.offset = (u64)-1;
2127         key.type = BTRFS_CHUNK_ITEM_KEY;
2128
2129         while (1) {
2130                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2131                 if (ret < 0)
2132                         goto error;
2133
2134                 /*
2135                  * this shouldn't happen, it means the last relocate
2136                  * failed
2137                  */
2138                 if (ret == 0)
2139                         break;
2140
2141                 ret = btrfs_previous_item(chunk_root, path, 0,
2142                                           BTRFS_CHUNK_ITEM_KEY);
2143                 if (ret)
2144                         break;
2145
2146                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2147                                       path->slots[0]);
2148                 if (found_key.objectid != key.objectid)
2149                         break;
2150
2151                 /* chunk zero is special */
2152                 if (found_key.offset == 0)
2153                         break;
2154
2155                 btrfs_release_path(path);
2156                 ret = btrfs_relocate_chunk(chunk_root,
2157                                            chunk_root->root_key.objectid,
2158                                            found_key.objectid,
2159                                            found_key.offset);
2160                 if (ret && ret != -ENOSPC)
2161                         goto error;
2162                 key.offset = found_key.offset - 1;
2163         }
2164         ret = 0;
2165 error:
2166         btrfs_free_path(path);
2167         mutex_unlock(&dev_root->fs_info->volume_mutex);
2168         return ret;
2169 }
2170
2171 /*
2172  * shrinking a device means finding all of the device extents past
2173  * the new size, and then following the back refs to the chunks.
2174  * The chunk relocation code actually frees the device extent
2175  */
2176 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2177 {
2178         struct btrfs_trans_handle *trans;
2179         struct btrfs_root *root = device->dev_root;
2180         struct btrfs_dev_extent *dev_extent = NULL;
2181         struct btrfs_path *path;
2182         u64 length;
2183         u64 chunk_tree;
2184         u64 chunk_objectid;
2185         u64 chunk_offset;
2186         int ret;
2187         int slot;
2188         int failed = 0;
2189         bool retried = false;
2190         struct extent_buffer *l;
2191         struct btrfs_key key;
2192         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2193         u64 old_total = btrfs_super_total_bytes(super_copy);
2194         u64 old_size = device->total_bytes;
2195         u64 diff = device->total_bytes - new_size;
2196
2197         if (new_size >= device->total_bytes)
2198                 return -EINVAL;
2199
2200         path = btrfs_alloc_path();
2201         if (!path)
2202                 return -ENOMEM;
2203
2204         path->reada = 2;
2205
2206         lock_chunks(root);
2207
2208         device->total_bytes = new_size;
2209         if (device->writeable) {
2210                 device->fs_devices->total_rw_bytes -= diff;
2211                 spin_lock(&root->fs_info->free_chunk_lock);
2212                 root->fs_info->free_chunk_space -= diff;
2213                 spin_unlock(&root->fs_info->free_chunk_lock);
2214         }
2215         unlock_chunks(root);
2216
2217 again:
2218         key.objectid = device->devid;
2219         key.offset = (u64)-1;
2220         key.type = BTRFS_DEV_EXTENT_KEY;
2221
2222         while (1) {
2223                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2224                 if (ret < 0)
2225                         goto done;
2226
2227                 ret = btrfs_previous_item(root, path, 0, key.type);
2228                 if (ret < 0)
2229                         goto done;
2230                 if (ret) {
2231                         ret = 0;
2232                         btrfs_release_path(path);
2233                         break;
2234                 }
2235
2236                 l = path->nodes[0];
2237                 slot = path->slots[0];
2238                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2239
2240                 if (key.objectid != device->devid) {
2241                         btrfs_release_path(path);
2242                         break;
2243                 }
2244
2245                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2246                 length = btrfs_dev_extent_length(l, dev_extent);
2247
2248                 if (key.offset + length <= new_size) {
2249                         btrfs_release_path(path);
2250                         break;
2251                 }
2252
2253                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2254                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2255                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2256                 btrfs_release_path(path);
2257
2258                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2259                                            chunk_offset);
2260                 if (ret && ret != -ENOSPC)
2261                         goto done;
2262                 if (ret == -ENOSPC)
2263                         failed++;
2264                 key.offset -= 1;
2265         }
2266
2267         if (failed && !retried) {
2268                 failed = 0;
2269                 retried = true;
2270                 goto again;
2271         } else if (failed && retried) {
2272                 ret = -ENOSPC;
2273                 lock_chunks(root);
2274
2275                 device->total_bytes = old_size;
2276                 if (device->writeable)
2277                         device->fs_devices->total_rw_bytes += diff;
2278                 spin_lock(&root->fs_info->free_chunk_lock);
2279                 root->fs_info->free_chunk_space += diff;
2280                 spin_unlock(&root->fs_info->free_chunk_lock);
2281                 unlock_chunks(root);
2282                 goto done;
2283         }
2284
2285         /* Shrinking succeeded, else we would be at "done". */
2286         trans = btrfs_start_transaction(root, 0);
2287         if (IS_ERR(trans)) {
2288                 ret = PTR_ERR(trans);
2289                 goto done;
2290         }
2291
2292         lock_chunks(root);
2293
2294         device->disk_total_bytes = new_size;
2295         /* Now btrfs_update_device() will change the on-disk size. */
2296         ret = btrfs_update_device(trans, device);
2297         if (ret) {
2298                 unlock_chunks(root);
2299                 btrfs_end_transaction(trans, root);
2300                 goto done;
2301         }
2302         WARN_ON(diff > old_total);
2303         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2304         unlock_chunks(root);
2305         btrfs_end_transaction(trans, root);
2306 done:
2307         btrfs_free_path(path);
2308         return ret;
2309 }
2310
2311 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2312                            struct btrfs_root *root,
2313                            struct btrfs_key *key,
2314                            struct btrfs_chunk *chunk, int item_size)
2315 {
2316         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2317         struct btrfs_disk_key disk_key;
2318         u32 array_size;
2319         u8 *ptr;
2320
2321         array_size = btrfs_super_sys_array_size(super_copy);
2322         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2323                 return -EFBIG;
2324
2325         ptr = super_copy->sys_chunk_array + array_size;
2326         btrfs_cpu_key_to_disk(&disk_key, key);
2327         memcpy(ptr, &disk_key, sizeof(disk_key));
2328         ptr += sizeof(disk_key);
2329         memcpy(ptr, chunk, item_size);
2330         item_size += sizeof(disk_key);
2331         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2332         return 0;
2333 }
2334
2335 /*
2336  * sort the devices in descending order by max_avail, total_avail
2337  */
2338 static int btrfs_cmp_device_info(const void *a, const void *b)
2339 {
2340         const struct btrfs_device_info *di_a = a;
2341         const struct btrfs_device_info *di_b = b;
2342
2343         if (di_a->max_avail > di_b->max_avail)
2344                 return -1;
2345         if (di_a->max_avail < di_b->max_avail)
2346                 return 1;
2347         if (di_a->total_avail > di_b->total_avail)
2348                 return -1;
2349         if (di_a->total_avail < di_b->total_avail)
2350                 return 1;
2351         return 0;
2352 }
2353
2354 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2355                                struct btrfs_root *extent_root,
2356                                struct map_lookup **map_ret,
2357                                u64 *num_bytes_out, u64 *stripe_size_out,
2358                                u64 start, u64 type)
2359 {
2360         struct btrfs_fs_info *info = extent_root->fs_info;
2361         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2362         struct list_head *cur;
2363         struct map_lookup *map = NULL;
2364         struct extent_map_tree *em_tree;
2365         struct extent_map *em;
2366         struct btrfs_device_info *devices_info = NULL;
2367         u64 total_avail;
2368         int num_stripes;        /* total number of stripes to allocate */
2369         int sub_stripes;        /* sub_stripes info for map */
2370         int dev_stripes;        /* stripes per dev */
2371         int devs_max;           /* max devs to use */
2372         int devs_min;           /* min devs needed */
2373         int devs_increment;     /* ndevs has to be a multiple of this */
2374         int ncopies;            /* how many copies to data has */
2375         int ret;
2376         u64 max_stripe_size;
2377         u64 max_chunk_size;
2378         u64 stripe_size;
2379         u64 num_bytes;
2380         int ndevs;
2381         int i;
2382         int j;
2383
2384         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2385             (type & BTRFS_BLOCK_GROUP_DUP)) {
2386                 WARN_ON(1);
2387                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2388         }
2389
2390         if (list_empty(&fs_devices->alloc_list))
2391                 return -ENOSPC;
2392
2393         sub_stripes = 1;
2394         dev_stripes = 1;
2395         devs_increment = 1;
2396         ncopies = 1;
2397         devs_max = 0;   /* 0 == as many as possible */
2398         devs_min = 1;
2399
2400         /*
2401          * define the properties of each RAID type.
2402          * FIXME: move this to a global table and use it in all RAID
2403          * calculation code
2404          */
2405         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2406                 dev_stripes = 2;
2407                 ncopies = 2;
2408                 devs_max = 1;
2409         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2410                 devs_min = 2;
2411         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2412                 devs_increment = 2;
2413                 ncopies = 2;
2414                 devs_max = 2;
2415                 devs_min = 2;
2416         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2417                 sub_stripes = 2;
2418                 devs_increment = 2;
2419                 ncopies = 2;
2420                 devs_min = 4;
2421         } else {
2422                 devs_max = 1;
2423         }
2424
2425         if (type & BTRFS_BLOCK_GROUP_DATA) {
2426                 max_stripe_size = 1024 * 1024 * 1024;
2427                 max_chunk_size = 10 * max_stripe_size;
2428         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2429                 max_stripe_size = 256 * 1024 * 1024;
2430                 max_chunk_size = max_stripe_size;
2431         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2432                 max_stripe_size = 8 * 1024 * 1024;
2433                 max_chunk_size = 2 * max_stripe_size;
2434         } else {
2435                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2436                        type);
2437                 BUG_ON(1);
2438         }
2439
2440         /* we don't want a chunk larger than 10% of writeable space */
2441         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2442                              max_chunk_size);
2443
2444         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2445                                GFP_NOFS);
2446         if (!devices_info)
2447                 return -ENOMEM;
2448
2449         cur = fs_devices->alloc_list.next;
2450
2451         /*
2452          * in the first pass through the devices list, we gather information
2453          * about the available holes on each device.
2454          */
2455         ndevs = 0;
2456         while (cur != &fs_devices->alloc_list) {
2457                 struct btrfs_device *device;
2458                 u64 max_avail;
2459                 u64 dev_offset;
2460
2461                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2462
2463                 cur = cur->next;
2464
2465                 if (!device->writeable) {
2466                         printk(KERN_ERR
2467                                "btrfs: read-only device in alloc_list\n");
2468                         WARN_ON(1);
2469                         continue;
2470                 }
2471
2472                 if (!device->in_fs_metadata)
2473                         continue;
2474
2475                 if (device->total_bytes > device->bytes_used)
2476                         total_avail = device->total_bytes - device->bytes_used;
2477                 else
2478                         total_avail = 0;
2479
2480                 /* If there is no space on this device, skip it. */
2481                 if (total_avail == 0)
2482                         continue;
2483
2484                 ret = find_free_dev_extent(trans, device,
2485                                            max_stripe_size * dev_stripes,
2486                                            &dev_offset, &max_avail);
2487                 if (ret && ret != -ENOSPC)
2488                         goto error;
2489
2490                 if (ret == 0)
2491                         max_avail = max_stripe_size * dev_stripes;
2492
2493                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2494                         continue;
2495
2496                 devices_info[ndevs].dev_offset = dev_offset;
2497                 devices_info[ndevs].max_avail = max_avail;
2498                 devices_info[ndevs].total_avail = total_avail;
2499                 devices_info[ndevs].dev = device;
2500                 ++ndevs;
2501         }
2502
2503         /*
2504          * now sort the devices by hole size / available space
2505          */
2506         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2507              btrfs_cmp_device_info, NULL);
2508
2509         /* round down to number of usable stripes */
2510         ndevs -= ndevs % devs_increment;
2511
2512         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2513                 ret = -ENOSPC;
2514                 goto error;
2515         }
2516
2517         if (devs_max && ndevs > devs_max)
2518                 ndevs = devs_max;
2519         /*
2520          * the primary goal is to maximize the number of stripes, so use as many
2521          * devices as possible, even if the stripes are not maximum sized.
2522          */
2523         stripe_size = devices_info[ndevs-1].max_avail;
2524         num_stripes = ndevs * dev_stripes;
2525
2526         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2527                 stripe_size = max_chunk_size * ncopies;
2528                 do_div(stripe_size, num_stripes);
2529         }
2530
2531         do_div(stripe_size, dev_stripes);
2532         do_div(stripe_size, BTRFS_STRIPE_LEN);
2533         stripe_size *= BTRFS_STRIPE_LEN;
2534
2535         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2536         if (!map) {
2537                 ret = -ENOMEM;
2538                 goto error;
2539         }
2540         map->num_stripes = num_stripes;
2541
2542         for (i = 0; i < ndevs; ++i) {
2543                 for (j = 0; j < dev_stripes; ++j) {
2544                         int s = i * dev_stripes + j;
2545                         map->stripes[s].dev = devices_info[i].dev;
2546                         map->stripes[s].physical = devices_info[i].dev_offset +
2547                                                    j * stripe_size;
2548                 }
2549         }
2550         map->sector_size = extent_root->sectorsize;
2551         map->stripe_len = BTRFS_STRIPE_LEN;
2552         map->io_align = BTRFS_STRIPE_LEN;
2553         map->io_width = BTRFS_STRIPE_LEN;
2554         map->type = type;
2555         map->sub_stripes = sub_stripes;
2556
2557         *map_ret = map;
2558         num_bytes = stripe_size * (num_stripes / ncopies);
2559
2560         *stripe_size_out = stripe_size;
2561         *num_bytes_out = num_bytes;
2562
2563         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2564
2565         em = alloc_extent_map();
2566         if (!em) {
2567                 ret = -ENOMEM;
2568                 goto error;
2569         }
2570         em->bdev = (struct block_device *)map;
2571         em->start = start;
2572         em->len = num_bytes;
2573         em->block_start = 0;
2574         em->block_len = em->len;
2575
2576         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2577         write_lock(&em_tree->lock);
2578         ret = add_extent_mapping(em_tree, em);
2579         write_unlock(&em_tree->lock);
2580         BUG_ON(ret);
2581         free_extent_map(em);
2582
2583         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2584                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2585                                      start, num_bytes);
2586         BUG_ON(ret);
2587
2588         for (i = 0; i < map->num_stripes; ++i) {
2589                 struct btrfs_device *device;
2590                 u64 dev_offset;
2591
2592                 device = map->stripes[i].dev;
2593                 dev_offset = map->stripes[i].physical;
2594
2595                 ret = btrfs_alloc_dev_extent(trans, device,
2596                                 info->chunk_root->root_key.objectid,
2597                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2598                                 start, dev_offset, stripe_size);
2599                 BUG_ON(ret);
2600         }
2601
2602         kfree(devices_info);
2603         return 0;
2604
2605 error:
2606         kfree(map);
2607         kfree(devices_info);
2608         return ret;
2609 }
2610
2611 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2612                                 struct btrfs_root *extent_root,
2613                                 struct map_lookup *map, u64 chunk_offset,
2614                                 u64 chunk_size, u64 stripe_size)
2615 {
2616         u64 dev_offset;
2617         struct btrfs_key key;
2618         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2619         struct btrfs_device *device;
2620         struct btrfs_chunk *chunk;
2621         struct btrfs_stripe *stripe;
2622         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2623         int index = 0;
2624         int ret;
2625
2626         chunk = kzalloc(item_size, GFP_NOFS);
2627         if (!chunk)
2628                 return -ENOMEM;
2629
2630         index = 0;
2631         while (index < map->num_stripes) {
2632                 device = map->stripes[index].dev;
2633                 device->bytes_used += stripe_size;
2634                 ret = btrfs_update_device(trans, device);
2635                 BUG_ON(ret);
2636                 index++;
2637         }
2638
2639         spin_lock(&extent_root->fs_info->free_chunk_lock);
2640         extent_root->fs_info->free_chunk_space -= (stripe_size *
2641                                                    map->num_stripes);
2642         spin_unlock(&extent_root->fs_info->free_chunk_lock);
2643
2644         index = 0;
2645         stripe = &chunk->stripe;
2646         while (index < map->num_stripes) {
2647                 device = map->stripes[index].dev;
2648                 dev_offset = map->stripes[index].physical;
2649
2650                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2651                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2652                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2653                 stripe++;
2654                 index++;
2655         }
2656
2657         btrfs_set_stack_chunk_length(chunk, chunk_size);
2658         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2659         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2660         btrfs_set_stack_chunk_type(chunk, map->type);
2661         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2662         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2663         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2664         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2665         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2666
2667         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2668         key.type = BTRFS_CHUNK_ITEM_KEY;
2669         key.offset = chunk_offset;
2670
2671         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2672         BUG_ON(ret);
2673
2674         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2675                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2676                                              item_size);
2677                 BUG_ON(ret);
2678         }
2679
2680         kfree(chunk);
2681         return 0;
2682 }
2683
2684 /*
2685  * Chunk allocation falls into two parts. The first part does works
2686  * that make the new allocated chunk useable, but not do any operation
2687  * that modifies the chunk tree. The second part does the works that
2688  * require modifying the chunk tree. This division is important for the
2689  * bootstrap process of adding storage to a seed btrfs.
2690  */
2691 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2692                       struct btrfs_root *extent_root, u64 type)
2693 {
2694         u64 chunk_offset;
2695         u64 chunk_size;
2696         u64 stripe_size;
2697         struct map_lookup *map;
2698         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2699         int ret;
2700
2701         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2702                               &chunk_offset);
2703         if (ret)
2704                 return ret;
2705
2706         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2707                                   &stripe_size, chunk_offset, type);
2708         if (ret)
2709                 return ret;
2710
2711         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2712                                    chunk_size, stripe_size);
2713         BUG_ON(ret);
2714         return 0;
2715 }
2716
2717 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2718                                          struct btrfs_root *root,
2719                                          struct btrfs_device *device)
2720 {
2721         u64 chunk_offset;
2722         u64 sys_chunk_offset;
2723         u64 chunk_size;
2724         u64 sys_chunk_size;
2725         u64 stripe_size;
2726         u64 sys_stripe_size;
2727         u64 alloc_profile;
2728         struct map_lookup *map;
2729         struct map_lookup *sys_map;
2730         struct btrfs_fs_info *fs_info = root->fs_info;
2731         struct btrfs_root *extent_root = fs_info->extent_root;
2732         int ret;
2733
2734         ret = find_next_chunk(fs_info->chunk_root,
2735                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2736         if (ret)
2737                 return ret;
2738
2739         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2740                         (fs_info->metadata_alloc_profile &
2741                          fs_info->avail_metadata_alloc_bits);
2742         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2743
2744         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2745                                   &stripe_size, chunk_offset, alloc_profile);
2746         BUG_ON(ret);
2747
2748         sys_chunk_offset = chunk_offset + chunk_size;
2749
2750         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2751                         (fs_info->system_alloc_profile &
2752                          fs_info->avail_system_alloc_bits);
2753         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2754
2755         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2756                                   &sys_chunk_size, &sys_stripe_size,
2757                                   sys_chunk_offset, alloc_profile);
2758         BUG_ON(ret);
2759
2760         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2761         BUG_ON(ret);
2762
2763         /*
2764          * Modifying chunk tree needs allocating new blocks from both
2765          * system block group and metadata block group. So we only can
2766          * do operations require modifying the chunk tree after both
2767          * block groups were created.
2768          */
2769         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2770                                    chunk_size, stripe_size);
2771         BUG_ON(ret);
2772
2773         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2774                                    sys_chunk_offset, sys_chunk_size,
2775                                    sys_stripe_size);
2776         BUG_ON(ret);
2777         return 0;
2778 }
2779
2780 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2781 {
2782         struct extent_map *em;
2783         struct map_lookup *map;
2784         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2785         int readonly = 0;
2786         int i;
2787
2788         read_lock(&map_tree->map_tree.lock);
2789         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2790         read_unlock(&map_tree->map_tree.lock);
2791         if (!em)
2792                 return 1;
2793
2794         if (btrfs_test_opt(root, DEGRADED)) {
2795                 free_extent_map(em);
2796                 return 0;
2797         }
2798
2799         map = (struct map_lookup *)em->bdev;
2800         for (i = 0; i < map->num_stripes; i++) {
2801                 if (!map->stripes[i].dev->writeable) {
2802                         readonly = 1;
2803                         break;
2804                 }
2805         }
2806         free_extent_map(em);
2807         return readonly;
2808 }
2809
2810 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2811 {
2812         extent_map_tree_init(&tree->map_tree);
2813 }
2814
2815 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2816 {
2817         struct extent_map *em;
2818
2819         while (1) {
2820                 write_lock(&tree->map_tree.lock);
2821                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2822                 if (em)
2823                         remove_extent_mapping(&tree->map_tree, em);
2824                 write_unlock(&tree->map_tree.lock);
2825                 if (!em)
2826                         break;
2827                 kfree(em->bdev);
2828                 /* once for us */
2829                 free_extent_map(em);
2830                 /* once for the tree */
2831                 free_extent_map(em);
2832         }
2833 }
2834
2835 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2836 {
2837         struct extent_map *em;
2838         struct map_lookup *map;
2839         struct extent_map_tree *em_tree = &map_tree->map_tree;
2840         int ret;
2841
2842         read_lock(&em_tree->lock);
2843         em = lookup_extent_mapping(em_tree, logical, len);
2844         read_unlock(&em_tree->lock);
2845         BUG_ON(!em);
2846
2847         BUG_ON(em->start > logical || em->start + em->len < logical);
2848         map = (struct map_lookup *)em->bdev;
2849         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2850                 ret = map->num_stripes;
2851         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2852                 ret = map->sub_stripes;
2853         else
2854                 ret = 1;
2855         free_extent_map(em);
2856         return ret;
2857 }
2858
2859 static int find_live_mirror(struct map_lookup *map, int first, int num,
2860                             int optimal)
2861 {
2862         int i;
2863         if (map->stripes[optimal].dev->bdev)
2864                 return optimal;
2865         for (i = first; i < first + num; i++) {
2866                 if (map->stripes[i].dev->bdev)
2867                         return i;
2868         }
2869         /* we couldn't find one that doesn't fail.  Just return something
2870          * and the io error handling code will clean up eventually
2871          */
2872         return optimal;
2873 }
2874
2875 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2876                              u64 logical, u64 *length,
2877                              struct btrfs_multi_bio **multi_ret,
2878                              int mirror_num)
2879 {
2880         struct extent_map *em;
2881         struct map_lookup *map;
2882         struct extent_map_tree *em_tree = &map_tree->map_tree;
2883         u64 offset;
2884         u64 stripe_offset;
2885         u64 stripe_end_offset;
2886         u64 stripe_nr;
2887         u64 stripe_nr_orig;
2888         u64 stripe_nr_end;
2889         int stripes_allocated = 8;
2890         int stripes_required = 1;
2891         int stripe_index;
2892         int i;
2893         int num_stripes;
2894         int max_errors = 0;
2895         struct btrfs_multi_bio *multi = NULL;
2896
2897         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2898                 stripes_allocated = 1;
2899 again:
2900         if (multi_ret) {
2901                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2902                                 GFP_NOFS);
2903                 if (!multi)
2904                         return -ENOMEM;
2905
2906                 atomic_set(&multi->error, 0);
2907         }
2908
2909         read_lock(&em_tree->lock);
2910         em = lookup_extent_mapping(em_tree, logical, *length);
2911         read_unlock(&em_tree->lock);
2912
2913         if (!em) {
2914                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2915                        (unsigned long long)logical,
2916                        (unsigned long long)*length);
2917                 BUG();
2918         }
2919
2920         BUG_ON(em->start > logical || em->start + em->len < logical);
2921         map = (struct map_lookup *)em->bdev;
2922         offset = logical - em->start;
2923
2924         if (mirror_num > map->num_stripes)
2925                 mirror_num = 0;
2926
2927         /* if our multi bio struct is too small, back off and try again */
2928         if (rw & REQ_WRITE) {
2929                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2930                                  BTRFS_BLOCK_GROUP_DUP)) {
2931                         stripes_required = map->num_stripes;
2932                         max_errors = 1;
2933                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2934                         stripes_required = map->sub_stripes;
2935                         max_errors = 1;
2936                 }
2937         }
2938         if (rw & REQ_DISCARD) {
2939                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2940                                  BTRFS_BLOCK_GROUP_RAID1 |
2941                                  BTRFS_BLOCK_GROUP_DUP |
2942                                  BTRFS_BLOCK_GROUP_RAID10)) {
2943                         stripes_required = map->num_stripes;
2944                 }
2945         }
2946         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2947             stripes_allocated < stripes_required) {
2948                 stripes_allocated = map->num_stripes;
2949                 free_extent_map(em);
2950                 kfree(multi);
2951                 goto again;
2952         }
2953         stripe_nr = offset;
2954         /*
2955          * stripe_nr counts the total number of stripes we have to stride
2956          * to get to this block
2957          */
2958         do_div(stripe_nr, map->stripe_len);
2959
2960         stripe_offset = stripe_nr * map->stripe_len;
2961         BUG_ON(offset < stripe_offset);
2962
2963         /* stripe_offset is the offset of this block in its stripe*/
2964         stripe_offset = offset - stripe_offset;
2965
2966         if (rw & REQ_DISCARD)
2967                 *length = min_t(u64, em->len - offset, *length);
2968         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2969                               BTRFS_BLOCK_GROUP_RAID1 |
2970                               BTRFS_BLOCK_GROUP_RAID10 |
2971                               BTRFS_BLOCK_GROUP_DUP)) {
2972                 /* we limit the length of each bio to what fits in a stripe */
2973                 *length = min_t(u64, em->len - offset,
2974                                 map->stripe_len - stripe_offset);
2975         } else {
2976                 *length = em->len - offset;
2977         }
2978
2979         if (!multi_ret)
2980                 goto out;
2981
2982         num_stripes = 1;
2983         stripe_index = 0;
2984         stripe_nr_orig = stripe_nr;
2985         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2986                         (~(map->stripe_len - 1));
2987         do_div(stripe_nr_end, map->stripe_len);
2988         stripe_end_offset = stripe_nr_end * map->stripe_len -
2989                             (offset + *length);
2990         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2991                 if (rw & REQ_DISCARD)
2992                         num_stripes = min_t(u64, map->num_stripes,
2993                                             stripe_nr_end - stripe_nr_orig);
2994                 stripe_index = do_div(stripe_nr, map->num_stripes);
2995         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2996                 if (rw & (REQ_WRITE | REQ_DISCARD))
2997                         num_stripes = map->num_stripes;
2998                 else if (mirror_num)
2999                         stripe_index = mirror_num - 1;
3000                 else {
3001                         stripe_index = find_live_mirror(map, 0,
3002                                             map->num_stripes,
3003                                             current->pid % map->num_stripes);
3004                 }
3005
3006         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3007                 if (rw & (REQ_WRITE | REQ_DISCARD))
3008                         num_stripes = map->num_stripes;
3009                 else if (mirror_num)
3010                         stripe_index = mirror_num - 1;
3011
3012         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3013                 int factor = map->num_stripes / map->sub_stripes;
3014
3015                 stripe_index = do_div(stripe_nr, factor);
3016                 stripe_index *= map->sub_stripes;
3017
3018                 if (rw & REQ_WRITE)
3019                         num_stripes = map->sub_stripes;
3020                 else if (rw & REQ_DISCARD)
3021                         num_stripes = min_t(u64, map->sub_stripes *
3022                                             (stripe_nr_end - stripe_nr_orig),
3023                                             map->num_stripes);
3024                 else if (mirror_num)
3025                         stripe_index += mirror_num - 1;
3026                 else {
3027                         stripe_index = find_live_mirror(map, stripe_index,
3028                                               map->sub_stripes, stripe_index +
3029                                               current->pid % map->sub_stripes);
3030                 }
3031         } else {
3032                 /*
3033                  * after this do_div call, stripe_nr is the number of stripes
3034                  * on this device we have to walk to find the data, and
3035                  * stripe_index is the number of our device in the stripe array
3036                  */
3037                 stripe_index = do_div(stripe_nr, map->num_stripes);
3038         }
3039         BUG_ON(stripe_index >= map->num_stripes);
3040
3041         if (rw & REQ_DISCARD) {
3042                 for (i = 0; i < num_stripes; i++) {
3043                         multi->stripes[i].physical =
3044                                 map->stripes[stripe_index].physical +
3045                                 stripe_offset + stripe_nr * map->stripe_len;
3046                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3047
3048                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3049                                 u64 stripes;
3050                                 u32 last_stripe = 0;
3051                                 int j;
3052
3053                                 div_u64_rem(stripe_nr_end - 1,
3054                                             map->num_stripes,
3055                                             &last_stripe);
3056
3057                                 for (j = 0; j < map->num_stripes; j++) {
3058                                         u32 test;
3059
3060                                         div_u64_rem(stripe_nr_end - 1 - j,
3061                                                     map->num_stripes, &test);
3062                                         if (test == stripe_index)
3063                                                 break;
3064                                 }
3065                                 stripes = stripe_nr_end - 1 - j;
3066                                 do_div(stripes, map->num_stripes);
3067                                 multi->stripes[i].length = map->stripe_len *
3068                                         (stripes - stripe_nr + 1);
3069
3070                                 if (i == 0) {
3071                                         multi->stripes[i].length -=
3072                                                 stripe_offset;
3073                                         stripe_offset = 0;
3074                                 }
3075                                 if (stripe_index == last_stripe)
3076                                         multi->stripes[i].length -=
3077                                                 stripe_end_offset;
3078                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3079                                 u64 stripes;
3080                                 int j;
3081                                 int factor = map->num_stripes /
3082                                              map->sub_stripes;
3083                                 u32 last_stripe = 0;
3084
3085                                 div_u64_rem(stripe_nr_end - 1,
3086                                             factor, &last_stripe);
3087                                 last_stripe *= map->sub_stripes;
3088
3089                                 for (j = 0; j < factor; j++) {
3090                                         u32 test;
3091
3092                                         div_u64_rem(stripe_nr_end - 1 - j,
3093                                                     factor, &test);
3094
3095                                         if (test ==
3096                                             stripe_index / map->sub_stripes)
3097                                                 break;
3098                                 }
3099                                 stripes = stripe_nr_end - 1 - j;
3100                                 do_div(stripes, factor);
3101                                 multi->stripes[i].length = map->stripe_len *
3102                                         (stripes - stripe_nr + 1);
3103
3104                                 if (i < map->sub_stripes) {
3105                                         multi->stripes[i].length -=
3106                                                 stripe_offset;
3107                                         if (i == map->sub_stripes - 1)
3108                                                 stripe_offset = 0;
3109                                 }
3110                                 if (stripe_index >= last_stripe &&
3111                                     stripe_index <= (last_stripe +
3112                                                      map->sub_stripes - 1)) {
3113                                         multi->stripes[i].length -=
3114                                                 stripe_end_offset;
3115                                 }
3116                         } else
3117                                 multi->stripes[i].length = *length;
3118
3119                         stripe_index++;
3120                         if (stripe_index == map->num_stripes) {
3121                                 /* This could only happen for RAID0/10 */
3122                                 stripe_index = 0;
3123                                 stripe_nr++;
3124                         }
3125                 }
3126         } else {
3127                 for (i = 0; i < num_stripes; i++) {
3128                         multi->stripes[i].physical =
3129                                 map->stripes[stripe_index].physical +
3130                                 stripe_offset +
3131                                 stripe_nr * map->stripe_len;
3132                         multi->stripes[i].dev =
3133                                 map->stripes[stripe_index].dev;
3134                         stripe_index++;
3135                 }
3136         }
3137         if (multi_ret) {
3138                 *multi_ret = multi;
3139                 multi->num_stripes = num_stripes;
3140                 multi->max_errors = max_errors;
3141         }
3142 out:
3143         free_extent_map(em);
3144         return 0;
3145 }
3146
3147 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3148                       u64 logical, u64 *length,
3149                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3150 {
3151         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3152                                  mirror_num);
3153 }
3154
3155 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3156                      u64 chunk_start, u64 physical, u64 devid,
3157                      u64 **logical, int *naddrs, int *stripe_len)
3158 {
3159         struct extent_map_tree *em_tree = &map_tree->map_tree;
3160         struct extent_map *em;
3161         struct map_lookup *map;
3162         u64 *buf;
3163         u64 bytenr;
3164         u64 length;
3165         u64 stripe_nr;
3166         int i, j, nr = 0;
3167
3168         read_lock(&em_tree->lock);
3169         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3170         read_unlock(&em_tree->lock);
3171
3172         BUG_ON(!em || em->start != chunk_start);
3173         map = (struct map_lookup *)em->bdev;
3174
3175         length = em->len;
3176         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3177                 do_div(length, map->num_stripes / map->sub_stripes);
3178         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3179                 do_div(length, map->num_stripes);
3180
3181         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3182         BUG_ON(!buf);
3183
3184         for (i = 0; i < map->num_stripes; i++) {
3185                 if (devid && map->stripes[i].dev->devid != devid)
3186                         continue;
3187                 if (map->stripes[i].physical > physical ||
3188                     map->stripes[i].physical + length <= physical)
3189                         continue;
3190
3191                 stripe_nr = physical - map->stripes[i].physical;
3192                 do_div(stripe_nr, map->stripe_len);
3193
3194                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3195                         stripe_nr = stripe_nr * map->num_stripes + i;
3196                         do_div(stripe_nr, map->sub_stripes);
3197                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3198                         stripe_nr = stripe_nr * map->num_stripes + i;
3199                 }
3200                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3201                 WARN_ON(nr >= map->num_stripes);
3202                 for (j = 0; j < nr; j++) {
3203                         if (buf[j] == bytenr)
3204                                 break;
3205                 }
3206                 if (j == nr) {
3207                         WARN_ON(nr >= map->num_stripes);
3208                         buf[nr++] = bytenr;
3209                 }
3210         }
3211
3212         *logical = buf;
3213         *naddrs = nr;
3214         *stripe_len = map->stripe_len;
3215
3216         free_extent_map(em);
3217         return 0;
3218 }
3219
3220 static void end_bio_multi_stripe(struct bio *bio, int err)
3221 {
3222         struct btrfs_multi_bio *multi = bio->bi_private;
3223         int is_orig_bio = 0;
3224
3225         if (err)
3226                 atomic_inc(&multi->error);
3227
3228         if (bio == multi->orig_bio)
3229                 is_orig_bio = 1;
3230
3231         if (atomic_dec_and_test(&multi->stripes_pending)) {
3232                 if (!is_orig_bio) {
3233                         bio_put(bio);
3234                         bio = multi->orig_bio;
3235                 }
3236                 bio->bi_private = multi->private;
3237                 bio->bi_end_io = multi->end_io;
3238                 /* only send an error to the higher layers if it is
3239                  * beyond the tolerance of the multi-bio
3240                  */
3241                 if (atomic_read(&multi->error) > multi->max_errors) {
3242                         err = -EIO;
3243                 } else if (err) {
3244                         /*
3245                          * this bio is actually up to date, we didn't
3246                          * go over the max number of errors
3247                          */
3248                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3249                         err = 0;
3250                 }
3251                 kfree(multi);
3252
3253                 bio_endio(bio, err);
3254         } else if (!is_orig_bio) {
3255                 bio_put(bio);
3256         }
3257 }
3258
3259 struct async_sched {
3260         struct bio *bio;
3261         int rw;
3262         struct btrfs_fs_info *info;
3263         struct btrfs_work work;
3264 };
3265
3266 /*
3267  * see run_scheduled_bios for a description of why bios are collected for
3268  * async submit.
3269  *
3270  * This will add one bio to the pending list for a device and make sure
3271  * the work struct is scheduled.
3272  */
3273 static noinline int schedule_bio(struct btrfs_root *root,
3274                                  struct btrfs_device *device,
3275                                  int rw, struct bio *bio)
3276 {
3277         int should_queue = 1;
3278         struct btrfs_pending_bios *pending_bios;
3279
3280         /* don't bother with additional async steps for reads, right now */
3281         if (!(rw & REQ_WRITE)) {
3282                 bio_get(bio);
3283                 submit_bio(rw, bio);
3284                 bio_put(bio);
3285                 return 0;
3286         }
3287
3288         /*
3289          * nr_async_bios allows us to reliably return congestion to the
3290          * higher layers.  Otherwise, the async bio makes it appear we have
3291          * made progress against dirty pages when we've really just put it
3292          * on a queue for later
3293          */
3294         atomic_inc(&root->fs_info->nr_async_bios);
3295         WARN_ON(bio->bi_next);
3296         bio->bi_next = NULL;
3297         bio->bi_rw |= rw;
3298
3299         spin_lock(&device->io_lock);
3300         if (bio->bi_rw & REQ_SYNC)
3301                 pending_bios = &device->pending_sync_bios;
3302         else
3303                 pending_bios = &device->pending_bios;
3304
3305         if (pending_bios->tail)
3306                 pending_bios->tail->bi_next = bio;
3307
3308         pending_bios->tail = bio;
3309         if (!pending_bios->head)
3310                 pending_bios->head = bio;
3311         if (device->running_pending)
3312                 should_queue = 0;
3313
3314         spin_unlock(&device->io_lock);
3315
3316         if (should_queue)
3317                 btrfs_queue_worker(&root->fs_info->submit_workers,
3318                                    &device->work);
3319         return 0;
3320 }
3321
3322 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3323                   int mirror_num, int async_submit)
3324 {
3325         struct btrfs_mapping_tree *map_tree;
3326         struct btrfs_device *dev;
3327         struct bio *first_bio = bio;
3328         u64 logical = (u64)bio->bi_sector << 9;
3329         u64 length = 0;
3330         u64 map_length;
3331         struct btrfs_multi_bio *multi = NULL;
3332         int ret;
3333         int dev_nr = 0;
3334         int total_devs = 1;
3335
3336         length = bio->bi_size;
3337         map_tree = &root->fs_info->mapping_tree;
3338         map_length = length;
3339
3340         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3341                               mirror_num);
3342         BUG_ON(ret);
3343
3344         total_devs = multi->num_stripes;
3345         if (map_length < length) {
3346                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3347                        "len %llu\n", (unsigned long long)logical,
3348                        (unsigned long long)length,
3349                        (unsigned long long)map_length);
3350                 BUG();
3351         }
3352         multi->end_io = first_bio->bi_end_io;
3353         multi->private = first_bio->bi_private;
3354         multi->orig_bio = first_bio;
3355         atomic_set(&multi->stripes_pending, multi->num_stripes);
3356
3357         while (dev_nr < total_devs) {
3358                 if (total_devs > 1) {
3359                         if (dev_nr < total_devs - 1) {
3360                                 bio = bio_clone(first_bio, GFP_NOFS);
3361                                 BUG_ON(!bio);
3362                         } else {
3363                                 bio = first_bio;
3364                         }
3365                         bio->bi_private = multi;
3366                         bio->bi_end_io = end_bio_multi_stripe;
3367                 }
3368                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3369                 dev = multi->stripes[dev_nr].dev;
3370                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3371                         bio->bi_bdev = dev->bdev;
3372                         if (async_submit)
3373                                 schedule_bio(root, dev, rw, bio);
3374                         else
3375                                 submit_bio(rw, bio);
3376                 } else {
3377                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3378                         bio->bi_sector = logical >> 9;
3379                         bio_endio(bio, -EIO);
3380                 }
3381                 dev_nr++;
3382         }
3383         if (total_devs == 1)
3384                 kfree(multi);
3385         return 0;
3386 }
3387
3388 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3389                                        u8 *uuid, u8 *fsid)
3390 {
3391         struct btrfs_device *device;
3392         struct btrfs_fs_devices *cur_devices;
3393
3394         cur_devices = root->fs_info->fs_devices;
3395         while (cur_devices) {
3396                 if (!fsid ||
3397                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3398                         device = __find_device(&cur_devices->devices,
3399                                                devid, uuid);
3400                         if (device)
3401                                 return device;
3402                 }
3403                 cur_devices = cur_devices->seed;
3404         }
3405         return NULL;
3406 }
3407
3408 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3409                                             u64 devid, u8 *dev_uuid)
3410 {
3411         struct btrfs_device *device;
3412         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3413
3414         device = kzalloc(sizeof(*device), GFP_NOFS);
3415         if (!device)
3416                 return NULL;
3417         list_add(&device->dev_list,
3418                  &fs_devices->devices);
3419         device->dev_root = root->fs_info->dev_root;
3420         device->devid = devid;
3421         device->work.func = pending_bios_fn;
3422         device->fs_devices = fs_devices;
3423         device->missing = 1;
3424         fs_devices->num_devices++;
3425         fs_devices->missing_devices++;
3426         spin_lock_init(&device->io_lock);
3427         INIT_LIST_HEAD(&device->dev_alloc_list);
3428         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3429         return device;
3430 }
3431
3432 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3433                           struct extent_buffer *leaf,
3434                           struct btrfs_chunk *chunk)
3435 {
3436         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3437         struct map_lookup *map;
3438         struct extent_map *em;
3439         u64 logical;
3440         u64 length;
3441         u64 devid;
3442         u8 uuid[BTRFS_UUID_SIZE];
3443         int num_stripes;
3444         int ret;
3445         int i;
3446
3447         logical = key->offset;
3448         length = btrfs_chunk_length(leaf, chunk);
3449
3450         read_lock(&map_tree->map_tree.lock);
3451         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3452         read_unlock(&map_tree->map_tree.lock);
3453
3454         /* already mapped? */
3455         if (em && em->start <= logical && em->start + em->len > logical) {
3456                 free_extent_map(em);
3457                 return 0;
3458         } else if (em) {
3459                 free_extent_map(em);
3460         }
3461
3462         em = alloc_extent_map();
3463         if (!em)
3464                 return -ENOMEM;
3465         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3466         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3467         if (!map) {
3468                 free_extent_map(em);
3469                 return -ENOMEM;
3470         }
3471
3472         em->bdev = (struct block_device *)map;
3473         em->start = logical;
3474         em->len = length;
3475         em->block_start = 0;
3476         em->block_len = em->len;
3477
3478         map->num_stripes = num_stripes;
3479         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3480         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3481         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3482         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3483         map->type = btrfs_chunk_type(leaf, chunk);
3484         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3485         for (i = 0; i < num_stripes; i++) {
3486                 map->stripes[i].physical =
3487                         btrfs_stripe_offset_nr(leaf, chunk, i);
3488                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3489                 read_extent_buffer(leaf, uuid, (unsigned long)
3490                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3491                                    BTRFS_UUID_SIZE);
3492                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3493                                                         NULL);
3494                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3495                         kfree(map);
3496                         free_extent_map(em);
3497                         return -EIO;
3498                 }
3499                 if (!map->stripes[i].dev) {
3500                         map->stripes[i].dev =
3501                                 add_missing_dev(root, devid, uuid);
3502                         if (!map->stripes[i].dev) {
3503                                 kfree(map);
3504                                 free_extent_map(em);
3505                                 return -EIO;
3506                         }
3507                 }
3508                 map->stripes[i].dev->in_fs_metadata = 1;
3509         }
3510
3511         write_lock(&map_tree->map_tree.lock);
3512         ret = add_extent_mapping(&map_tree->map_tree, em);
3513         write_unlock(&map_tree->map_tree.lock);
3514         BUG_ON(ret);
3515         free_extent_map(em);
3516
3517         return 0;
3518 }
3519
3520 static int fill_device_from_item(struct extent_buffer *leaf,
3521                                  struct btrfs_dev_item *dev_item,
3522                                  struct btrfs_device *device)
3523 {
3524         unsigned long ptr;
3525
3526         device->devid = btrfs_device_id(leaf, dev_item);
3527         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3528         device->total_bytes = device->disk_total_bytes;
3529         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3530         device->type = btrfs_device_type(leaf, dev_item);
3531         device->io_align = btrfs_device_io_align(leaf, dev_item);
3532         device->io_width = btrfs_device_io_width(leaf, dev_item);
3533         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3534
3535         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3536         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3537
3538         return 0;
3539 }
3540
3541 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3542 {
3543         struct btrfs_fs_devices *fs_devices;
3544         int ret;
3545
3546         mutex_lock(&uuid_mutex);
3547
3548         fs_devices = root->fs_info->fs_devices->seed;
3549         while (fs_devices) {
3550                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3551                         ret = 0;
3552                         goto out;
3553                 }
3554                 fs_devices = fs_devices->seed;
3555         }
3556
3557         fs_devices = find_fsid(fsid);
3558         if (!fs_devices) {
3559                 ret = -ENOENT;
3560                 goto out;
3561         }
3562
3563         fs_devices = clone_fs_devices(fs_devices);
3564         if (IS_ERR(fs_devices)) {
3565                 ret = PTR_ERR(fs_devices);
3566                 goto out;
3567         }
3568
3569         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3570                                    root->fs_info->bdev_holder);
3571         if (ret)
3572                 goto out;
3573
3574         if (!fs_devices->seeding) {
3575                 __btrfs_close_devices(fs_devices);
3576                 free_fs_devices(fs_devices);
3577                 ret = -EINVAL;
3578                 goto out;
3579         }
3580
3581         fs_devices->seed = root->fs_info->fs_devices->seed;
3582         root->fs_info->fs_devices->seed = fs_devices;
3583 out:
3584         mutex_unlock(&uuid_mutex);
3585         return ret;
3586 }
3587
3588 static int read_one_dev(struct btrfs_root *root,
3589                         struct extent_buffer *leaf,
3590                         struct btrfs_dev_item *dev_item)
3591 {
3592         struct btrfs_device *device;
3593         u64 devid;
3594         int ret;
3595         u8 fs_uuid[BTRFS_UUID_SIZE];
3596         u8 dev_uuid[BTRFS_UUID_SIZE];
3597
3598         devid = btrfs_device_id(leaf, dev_item);
3599         read_extent_buffer(leaf, dev_uuid,
3600                            (unsigned long)btrfs_device_uuid(dev_item),
3601                            BTRFS_UUID_SIZE);
3602         read_extent_buffer(leaf, fs_uuid,
3603                            (unsigned long)btrfs_device_fsid(dev_item),
3604                            BTRFS_UUID_SIZE);
3605
3606         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3607                 ret = open_seed_devices(root, fs_uuid);
3608                 if (ret && !btrfs_test_opt(root, DEGRADED))
3609                         return ret;
3610         }
3611
3612         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3613         if (!device || !device->bdev) {
3614                 if (!btrfs_test_opt(root, DEGRADED))
3615                         return -EIO;
3616
3617                 if (!device) {
3618                         printk(KERN_WARNING "warning devid %llu missing\n",
3619                                (unsigned long long)devid);
3620                         device = add_missing_dev(root, devid, dev_uuid);
3621                         if (!device)
3622                                 return -ENOMEM;
3623                 } else if (!device->missing) {
3624                         /*
3625                          * this happens when a device that was properly setup
3626                          * in the device info lists suddenly goes bad.
3627                          * device->bdev is NULL, and so we have to set
3628                          * device->missing to one here
3629                          */
3630                         root->fs_info->fs_devices->missing_devices++;
3631                         device->missing = 1;
3632                 }
3633         }
3634
3635         if (device->fs_devices != root->fs_info->fs_devices) {
3636                 BUG_ON(device->writeable);
3637                 if (device->generation !=
3638                     btrfs_device_generation(leaf, dev_item))
3639                         return -EINVAL;
3640         }
3641
3642         fill_device_from_item(leaf, dev_item, device);
3643         device->dev_root = root->fs_info->dev_root;
3644         device->in_fs_metadata = 1;
3645         if (device->writeable) {
3646                 device->fs_devices->total_rw_bytes += device->total_bytes;
3647                 spin_lock(&root->fs_info->free_chunk_lock);
3648                 root->fs_info->free_chunk_space += device->total_bytes -
3649                         device->bytes_used;
3650                 spin_unlock(&root->fs_info->free_chunk_lock);
3651         }
3652         ret = 0;
3653         return ret;
3654 }
3655
3656 int btrfs_read_sys_array(struct btrfs_root *root)
3657 {
3658         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3659         struct extent_buffer *sb;
3660         struct btrfs_disk_key *disk_key;
3661         struct btrfs_chunk *chunk;
3662         u8 *ptr;
3663         unsigned long sb_ptr;
3664         int ret = 0;
3665         u32 num_stripes;
3666         u32 array_size;
3667         u32 len = 0;
3668         u32 cur;
3669         struct btrfs_key key;
3670
3671         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3672                                           BTRFS_SUPER_INFO_SIZE);
3673         if (!sb)
3674                 return -ENOMEM;
3675         btrfs_set_buffer_uptodate(sb);
3676         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
3677
3678         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3679         array_size = btrfs_super_sys_array_size(super_copy);
3680
3681         ptr = super_copy->sys_chunk_array;
3682         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3683         cur = 0;
3684
3685         while (cur < array_size) {
3686                 disk_key = (struct btrfs_disk_key *)ptr;
3687                 btrfs_disk_key_to_cpu(&key, disk_key);
3688
3689                 len = sizeof(*disk_key); ptr += len;
3690                 sb_ptr += len;
3691                 cur += len;
3692
3693                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3694                         chunk = (struct btrfs_chunk *)sb_ptr;
3695                         ret = read_one_chunk(root, &key, sb, chunk);
3696                         if (ret)
3697                                 break;
3698                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3699                         len = btrfs_chunk_item_size(num_stripes);
3700                 } else {
3701                         ret = -EIO;
3702                         break;
3703                 }
3704                 ptr += len;
3705                 sb_ptr += len;
3706                 cur += len;
3707         }
3708         free_extent_buffer(sb);
3709         return ret;
3710 }
3711
3712 int btrfs_read_chunk_tree(struct btrfs_root *root)
3713 {
3714         struct btrfs_path *path;
3715         struct extent_buffer *leaf;
3716         struct btrfs_key key;
3717         struct btrfs_key found_key;
3718         int ret;
3719         int slot;
3720
3721         root = root->fs_info->chunk_root;
3722
3723         path = btrfs_alloc_path();
3724         if (!path)
3725                 return -ENOMEM;
3726
3727         /* first we search for all of the device items, and then we
3728          * read in all of the chunk items.  This way we can create chunk
3729          * mappings that reference all of the devices that are afound
3730          */
3731         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3732         key.offset = 0;
3733         key.type = 0;
3734 again:
3735         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3736         if (ret < 0)
3737                 goto error;
3738         while (1) {
3739                 leaf = path->nodes[0];
3740                 slot = path->slots[0];
3741                 if (slot >= btrfs_header_nritems(leaf)) {
3742                         ret = btrfs_next_leaf(root, path);
3743                         if (ret == 0)
3744                                 continue;
3745                         if (ret < 0)
3746                                 goto error;
3747                         break;
3748                 }
3749                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3750                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3751                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3752                                 break;
3753                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3754                                 struct btrfs_dev_item *dev_item;
3755                                 dev_item = btrfs_item_ptr(leaf, slot,
3756                                                   struct btrfs_dev_item);
3757                                 ret = read_one_dev(root, leaf, dev_item);
3758                                 if (ret)
3759                                         goto error;
3760                         }
3761                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3762                         struct btrfs_chunk *chunk;
3763                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3764                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3765                         if (ret)
3766                                 goto error;
3767                 }
3768                 path->slots[0]++;
3769         }
3770         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3771                 key.objectid = 0;
3772                 btrfs_release_path(path);
3773                 goto again;
3774         }
3775         ret = 0;
3776 error:
3777         btrfs_free_path(path);
3778         return ret;
3779 }