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 <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43
  44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45                                 struct btrfs_root *root,
  46                                 struct btrfs_device *device);
  47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
  50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  51
  52 static DEFINE_MUTEX(uuid_mutex);
  53 static LIST_HEAD(fs_uuids);
  54
  55 static void lock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_lock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void unlock_chunks(struct btrfs_root *root)
  61 {
  62         mutex_unlock(&root->fs_info->chunk_mutex);
  63 }
  64
  65 static struct btrfs_fs_devices *__alloc_fs_devices(void)
  66 {
  67         struct btrfs_fs_devices *fs_devs;
  68
  69         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
  70         if (!fs_devs)
  71                 return ERR_PTR(-ENOMEM);
  72
  73         mutex_init(&fs_devs->device_list_mutex);
  74
  75         INIT_LIST_HEAD(&fs_devs->devices);
  76         INIT_LIST_HEAD(&fs_devs->alloc_list);
  77         INIT_LIST_HEAD(&fs_devs->list);
  78
  79         return fs_devs;
  80 }
  81
  82 /**
  83  * alloc_fs_devices - allocate struct btrfs_fs_devices
  84  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
  85  *              generated.
  86  *
  87  * Return: a pointer to a new &struct btrfs_fs_devices on success;
  88  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
  89  * can be destroyed with kfree() right away.
  90  */
  91 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
  92 {
  93         struct btrfs_fs_devices *fs_devs;
  94
  95         fs_devs = __alloc_fs_devices();
  96         if (IS_ERR(fs_devs))
  97                 return fs_devs;
  98
  99         if (fsid)
 100                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 101         else
 102                 generate_random_uuid(fs_devs->fsid);
 103
 104         return fs_devs;
 105 }
 106
 107 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 108 {
 109         struct btrfs_device *device;
 110         WARN_ON(fs_devices->opened);
 111         while (!list_empty(&fs_devices->devices)) {
 112                 device = list_entry(fs_devices->devices.next,
 113                                     struct btrfs_device, dev_list);
 114                 list_del(&device->dev_list);
 115                 rcu_string_free(device->name);
 116                 kfree(device);
 117         }
 118         kfree(fs_devices);
 119 }
 120
 121 static void btrfs_kobject_uevent(struct block_device *bdev,
 122                                  enum kobject_action action)
 123 {
 124         int ret;
 125
 126         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 127         if (ret)
 128                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 129                         action,
 130                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 131                         &disk_to_dev(bdev->bd_disk)->kobj);
 132 }
 133
 134 void btrfs_cleanup_fs_uuids(void)
 135 {
 136         struct btrfs_fs_devices *fs_devices;
 137
 138         while (!list_empty(&fs_uuids)) {
 139                 fs_devices = list_entry(fs_uuids.next,
 140                                         struct btrfs_fs_devices, list);
 141                 list_del(&fs_devices->list);
 142                 free_fs_devices(fs_devices);
 143         }
 144 }
 145
 146 static struct btrfs_device *__alloc_device(void)
 147 {
 148         struct btrfs_device *dev;
 149
 150         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 151         if (!dev)
 152                 return ERR_PTR(-ENOMEM);
 153
 154         INIT_LIST_HEAD(&dev->dev_list);
 155         INIT_LIST_HEAD(&dev->dev_alloc_list);
 156
 157         spin_lock_init(&dev->io_lock);
 158
 159         spin_lock_init(&dev->reada_lock);
 160         atomic_set(&dev->reada_in_flight, 0);
 161         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 162         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 163
 164         return dev;
 165 }
 166
 167 static noinline struct btrfs_device *__find_device(struct list_head *head,
 168                                                    u64 devid, u8 *uuid)
 169 {
 170         struct btrfs_device *dev;
 171
 172         list_for_each_entry(dev, head, dev_list) {
 173                 if (dev->devid == devid &&
 174                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 175                         return dev;
 176                 }
 177         }
 178         return NULL;
 179 }
 180
 181 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 182 {
 183         struct btrfs_fs_devices *fs_devices;
 184
 185         list_for_each_entry(fs_devices, &fs_uuids, list) {
 186                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 187                         return fs_devices;
 188         }
 189         return NULL;
 190 }
 191
 192 static int
 193 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 194                       int flush, struct block_device **bdev,
 195                       struct buffer_head **bh)
 196 {
 197         int ret;
 198
 199         *bdev = blkdev_get_by_path(device_path, flags, holder);
 200
 201         if (IS_ERR(*bdev)) {
 202                 ret = PTR_ERR(*bdev);
 203                 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
 204                 goto error;
 205         }
 206
 207         if (flush)
 208                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 209         ret = set_blocksize(*bdev, 4096);
 210         if (ret) {
 211                 blkdev_put(*bdev, flags);
 212                 goto error;
 213         }
 214         invalidate_bdev(*bdev);
 215         *bh = btrfs_read_dev_super(*bdev);
 216         if (!*bh) {
 217                 ret = -EINVAL;
 218                 blkdev_put(*bdev, flags);
 219                 goto error;
 220         }
 221
 222         return 0;
 223
 224 error:
 225         *bdev = NULL;
 226         *bh = NULL;
 227         return ret;
 228 }
 229
 230 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 231                         struct bio *head, struct bio *tail)
 232 {
 233
 234         struct bio *old_head;
 235
 236         old_head = pending_bios->head;
 237         pending_bios->head = head;
 238         if (pending_bios->tail)
 239                 tail->bi_next = old_head;
 240         else
 241                 pending_bios->tail = tail;
 242 }
 243
 244 /*
 245  * we try to collect pending bios for a device so we don't get a large
 246  * number of procs sending bios down to the same device.  This greatly
 247  * improves the schedulers ability to collect and merge the bios.
 248  *
 249  * But, it also turns into a long list of bios to process and that is sure
 250  * to eventually make the worker thread block.  The solution here is to
 251  * make some progress and then put this work struct back at the end of
 252  * the list if the block device is congested.  This way, multiple devices
 253  * can make progress from a single worker thread.
 254  */
 255 static noinline void run_scheduled_bios(struct btrfs_device *device)
 256 {
 257         struct bio *pending;
 258         struct backing_dev_info *bdi;
 259         struct btrfs_fs_info *fs_info;
 260         struct btrfs_pending_bios *pending_bios;
 261         struct bio *tail;
 262         struct bio *cur;
 263         int again = 0;
 264         unsigned long num_run;
 265         unsigned long batch_run = 0;
 266         unsigned long limit;
 267         unsigned long last_waited = 0;
 268         int force_reg = 0;
 269         int sync_pending = 0;
 270         struct blk_plug plug;
 271
 272         /*
 273          * this function runs all the bios we've collected for
 274          * a particular device.  We don't want to wander off to
 275          * another device without first sending all of these down.
 276          * So, setup a plug here and finish it off before we return
 277          */
 278         blk_start_plug(&plug);
 279
 280         bdi = blk_get_backing_dev_info(device->bdev);
 281         fs_info = device->dev_root->fs_info;
 282         limit = btrfs_async_submit_limit(fs_info);
 283         limit = limit * 2 / 3;
 284
 285 loop:
 286         spin_lock(&device->io_lock);
 287
 288 loop_lock:
 289         num_run = 0;
 290
 291         /* take all the bios off the list at once and process them
 292          * later on (without the lock held).  But, remember the
 293          * tail and other pointers so the bios can be properly reinserted
 294          * into the list if we hit congestion
 295          */
 296         if (!force_reg && device->pending_sync_bios.head) {
 297                 pending_bios = &device->pending_sync_bios;
 298                 force_reg = 1;
 299         } else {
 300                 pending_bios = &device->pending_bios;
 301                 force_reg = 0;
 302         }
 303
 304         pending = pending_bios->head;
 305         tail = pending_bios->tail;
 306         WARN_ON(pending && !tail);
 307
 308         /*
 309          * if pending was null this time around, no bios need processing
 310          * at all and we can stop.  Otherwise it'll loop back up again
 311          * and do an additional check so no bios are missed.
 312          *
 313          * device->running_pending is used to synchronize with the
 314          * schedule_bio code.
 315          */
 316         if (device->pending_sync_bios.head == NULL &&
 317             device->pending_bios.head == NULL) {
 318                 again = 0;
 319                 device->running_pending = 0;
 320         } else {
 321                 again = 1;
 322                 device->running_pending = 1;
 323         }
 324
 325         pending_bios->head = NULL;
 326         pending_bios->tail = NULL;
 327
 328         spin_unlock(&device->io_lock);
 329
 330         while (pending) {
 331
 332                 rmb();
 333                 /* we want to work on both lists, but do more bios on the
 334                  * sync list than the regular list
 335                  */
 336                 if ((num_run > 32 &&
 337                     pending_bios != &device->pending_sync_bios &&
 338                     device->pending_sync_bios.head) ||
 339                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 340                     device->pending_bios.head)) {
 341                         spin_lock(&device->io_lock);
 342                         requeue_list(pending_bios, pending, tail);
 343                         goto loop_lock;
 344                 }
 345
 346                 cur = pending;
 347                 pending = pending->bi_next;
 348                 cur->bi_next = NULL;
 349
 350                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 351                     waitqueue_active(&fs_info->async_submit_wait))
 352                         wake_up(&fs_info->async_submit_wait);
 353
 354                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 355
 356                 /*
 357                  * if we're doing the sync list, record that our
 358                  * plug has some sync requests on it
 359                  *
 360                  * If we're doing the regular list and there are
 361                  * sync requests sitting around, unplug before
 362                  * we add more
 363                  */
 364                 if (pending_bios == &device->pending_sync_bios) {
 365                         sync_pending = 1;
 366                 } else if (sync_pending) {
 367                         blk_finish_plug(&plug);
 368                         blk_start_plug(&plug);
 369                         sync_pending = 0;
 370                 }
 371
 372                 btrfsic_submit_bio(cur->bi_rw, cur);
 373                 num_run++;
 374                 batch_run++;
 375                 if (need_resched())
 376                         cond_resched();
 377
 378                 /*
 379                  * we made progress, there is more work to do and the bdi
 380                  * is now congested.  Back off and let other work structs
 381                  * run instead
 382                  */
 383                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 384                     fs_info->fs_devices->open_devices > 1) {
 385                         struct io_context *ioc;
 386
 387                         ioc = current->io_context;
 388
 389                         /*
 390                          * the main goal here is that we don't want to
 391                          * block if we're going to be able to submit
 392                          * more requests without blocking.
 393                          *
 394                          * This code does two great things, it pokes into
 395                          * the elevator code from a filesystem _and_
 396                          * it makes assumptions about how batching works.
 397                          */
 398                         if (ioc && ioc->nr_batch_requests > 0 &&
 399                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 400                             (last_waited == 0 ||
 401                              ioc->last_waited == last_waited)) {
 402                                 /*
 403                                  * we want to go through our batch of
 404                                  * requests and stop.  So, we copy out
 405                                  * the ioc->last_waited time and test
 406                                  * against it before looping
 407                                  */
 408                                 last_waited = ioc->last_waited;
 409                                 if (need_resched())
 410                                         cond_resched();
 411                                 continue;
 412                         }
 413                         spin_lock(&device->io_lock);
 414                         requeue_list(pending_bios, pending, tail);
 415                         device->running_pending = 1;
 416
 417                         spin_unlock(&device->io_lock);
 418                         btrfs_queue_work(fs_info->submit_workers,
 419                                          &device->work);
 420                         goto done;
 421                 }
 422                 /* unplug every 64 requests just for good measure */
 423                 if (batch_run % 64 == 0) {
 424                         blk_finish_plug(&plug);
 425                         blk_start_plug(&plug);
 426                         sync_pending = 0;
 427                 }
 428         }
 429
 430         cond_resched();
 431         if (again)
 432                 goto loop;
 433
 434         spin_lock(&device->io_lock);
 435         if (device->pending_bios.head || device->pending_sync_bios.head)
 436                 goto loop_lock;
 437         spin_unlock(&device->io_lock);
 438
 439 done:
 440         blk_finish_plug(&plug);
 441 }
 442
 443 static void pending_bios_fn(struct btrfs_work *work)
 444 {
 445         struct btrfs_device *device;
 446
 447         device = container_of(work, struct btrfs_device, work);
 448         run_scheduled_bios(device);
 449 }
 450
 451 /*
 452  * Add new device to list of registered devices
 453  *
 454  * Returns:
 455  * 1   - first time device is seen
 456  * 0   - device already known
 457  * < 0 - error
 458  */
 459 static noinline int device_list_add(const char *path,
 460                            struct btrfs_super_block *disk_super,
 461                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 462 {
 463         struct btrfs_device *device;
 464         struct btrfs_fs_devices *fs_devices;
 465         struct rcu_string *name;
 466         int ret = 0;
 467         u64 found_transid = btrfs_super_generation(disk_super);
 468
 469         fs_devices = find_fsid(disk_super->fsid);
 470         if (!fs_devices) {
 471                 fs_devices = alloc_fs_devices(disk_super->fsid);
 472                 if (IS_ERR(fs_devices))
 473                         return PTR_ERR(fs_devices);
 474
 475                 list_add(&fs_devices->list, &fs_uuids);
 476                 fs_devices->latest_devid = devid;
 477                 fs_devices->latest_trans = found_transid;
 478
 479                 device = NULL;
 480         } else {
 481                 device = __find_device(&fs_devices->devices, devid,
 482                                        disk_super->dev_item.uuid);
 483         }
 484         if (!device) {
 485                 if (fs_devices->opened)
 486                         return -EBUSY;
 487
 488                 device = btrfs_alloc_device(NULL, &devid,
 489                                             disk_super->dev_item.uuid);
 490                 if (IS_ERR(device)) {
 491                         /* we can safely leave the fs_devices entry around */
 492                         return PTR_ERR(device);
 493                 }
 494
 495                 name = rcu_string_strdup(path, GFP_NOFS);
 496                 if (!name) {
 497                         kfree(device);
 498                         return -ENOMEM;
 499                 }
 500                 rcu_assign_pointer(device->name, name);
 501
 502                 mutex_lock(&fs_devices->device_list_mutex);
 503                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 504                 fs_devices->num_devices++;
 505                 mutex_unlock(&fs_devices->device_list_mutex);
 506
 507                 ret = 1;
 508                 device->fs_devices = fs_devices;
 509         } else if (!device->name || strcmp(device->name->str, path)) {
 510                 name = rcu_string_strdup(path, GFP_NOFS);
 511                 if (!name)
 512                         return -ENOMEM;
 513                 rcu_string_free(device->name);
 514                 rcu_assign_pointer(device->name, name);
 515                 if (device->missing) {
 516                         fs_devices->missing_devices--;
 517                         device->missing = 0;
 518                 }
 519         }
 520
 521         if (found_transid > fs_devices->latest_trans) {
 522                 fs_devices->latest_devid = devid;
 523                 fs_devices->latest_trans = found_transid;
 524         }
 525         *fs_devices_ret = fs_devices;
 526
 527         return ret;
 528 }
 529
 530 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 531 {
 532         struct btrfs_fs_devices *fs_devices;
 533         struct btrfs_device *device;
 534         struct btrfs_device *orig_dev;
 535
 536         fs_devices = alloc_fs_devices(orig->fsid);
 537         if (IS_ERR(fs_devices))
 538                 return fs_devices;
 539
 540         fs_devices->latest_devid = orig->latest_devid;
 541         fs_devices->latest_trans = orig->latest_trans;
 542         fs_devices->total_devices = orig->total_devices;
 543
 544         /* We have held the volume lock, it is safe to get the devices. */
 545         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 546                 struct rcu_string *name;
 547
 548                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 549                                             orig_dev->uuid);
 550                 if (IS_ERR(device))
 551                         goto error;
 552
 553                 /*
 554                  * This is ok to do without rcu read locked because we hold the
 555                  * uuid mutex so nothing we touch in here is going to disappear.
 556                  */
 557                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 558                 if (!name) {
 559                         kfree(device);
 560                         goto error;
 561                 }
 562                 rcu_assign_pointer(device->name, name);
 563
 564                 list_add(&device->dev_list, &fs_devices->devices);
 565                 device->fs_devices = fs_devices;
 566                 fs_devices->num_devices++;
 567         }
 568         return fs_devices;
 569 error:
 570         free_fs_devices(fs_devices);
 571         return ERR_PTR(-ENOMEM);
 572 }
 573
 574 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 575                                struct btrfs_fs_devices *fs_devices, int step)
 576 {
 577         struct btrfs_device *device, *next;
 578
 579         struct block_device *latest_bdev = NULL;
 580         u64 latest_devid = 0;
 581         u64 latest_transid = 0;
 582
 583         mutex_lock(&uuid_mutex);
 584 again:
 585         /* This is the initialized path, it is safe to release the devices. */
 586         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 587                 if (device->in_fs_metadata) {
 588                         if (!device->is_tgtdev_for_dev_replace &&
 589                             (!latest_transid ||
 590                              device->generation > latest_transid)) {
 591                                 latest_devid = device->devid;
 592                                 latest_transid = device->generation;
 593                                 latest_bdev = device->bdev;
 594                         }
 595                         continue;
 596                 }
 597
 598                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 599                         /*
 600                          * In the first step, keep the device which has
 601                          * the correct fsid and the devid that is used
 602                          * for the dev_replace procedure.
 603                          * In the second step, the dev_replace state is
 604                          * read from the device tree and it is known
 605                          * whether the procedure is really active or
 606                          * not, which means whether this device is
 607                          * used or whether it should be removed.
 608                          */
 609                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 610                                 continue;
 611                         }
 612                 }
 613                 if (device->bdev) {
 614                         blkdev_put(device->bdev, device->mode);
 615                         device->bdev = NULL;
 616                         fs_devices->open_devices--;
 617                 }
 618                 if (device->writeable) {
 619                         list_del_init(&device->dev_alloc_list);
 620                         device->writeable = 0;
 621                         if (!device->is_tgtdev_for_dev_replace)
 622                                 fs_devices->rw_devices--;
 623                 }
 624                 list_del_init(&device->dev_list);
 625                 fs_devices->num_devices--;
 626                 rcu_string_free(device->name);
 627                 kfree(device);
 628         }
 629
 630         if (fs_devices->seed) {
 631                 fs_devices = fs_devices->seed;
 632                 goto again;
 633         }
 634
 635         fs_devices->latest_bdev = latest_bdev;
 636         fs_devices->latest_devid = latest_devid;
 637         fs_devices->latest_trans = latest_transid;
 638
 639         mutex_unlock(&uuid_mutex);
 640 }
 641
 642 static void __free_device(struct work_struct *work)
 643 {
 644         struct btrfs_device *device;
 645
 646         device = container_of(work, struct btrfs_device, rcu_work);
 647
 648         if (device->bdev)
 649                 blkdev_put(device->bdev, device->mode);
 650
 651         rcu_string_free(device->name);
 652         kfree(device);
 653 }
 654
 655 static void free_device(struct rcu_head *head)
 656 {
 657         struct btrfs_device *device;
 658
 659         device = container_of(head, struct btrfs_device, rcu);
 660
 661         INIT_WORK(&device->rcu_work, __free_device);
 662         schedule_work(&device->rcu_work);
 663 }
 664
 665 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 666 {
 667         struct btrfs_device *device;
 668
 669         if (--fs_devices->opened > 0)
 670                 return 0;
 671
 672         mutex_lock(&fs_devices->device_list_mutex);
 673         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 674                 struct btrfs_device *new_device;
 675                 struct rcu_string *name;
 676
 677                 if (device->bdev)
 678                         fs_devices->open_devices--;
 679
 680                 if (device->writeable &&
 681                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 682                         list_del_init(&device->dev_alloc_list);
 683                         fs_devices->rw_devices--;
 684                 }
 685
 686                 if (device->can_discard)
 687                         fs_devices->num_can_discard--;
 688                 if (device->missing)
 689                         fs_devices->missing_devices--;
 690
 691                 new_device = btrfs_alloc_device(NULL, &device->devid,
 692                                                 device->uuid);
 693                 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 694
 695                 /* Safe because we are under uuid_mutex */
 696                 if (device->name) {
 697                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 698                         BUG_ON(!name); /* -ENOMEM */
 699                         rcu_assign_pointer(new_device->name, name);
 700                 }
 701
 702                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 703                 new_device->fs_devices = device->fs_devices;
 704
 705                 call_rcu(&device->rcu, free_device);
 706         }
 707         mutex_unlock(&fs_devices->device_list_mutex);
 708
 709         WARN_ON(fs_devices->open_devices);
 710         WARN_ON(fs_devices->rw_devices);
 711         fs_devices->opened = 0;
 712         fs_devices->seeding = 0;
 713
 714         return 0;
 715 }
 716
 717 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 718 {
 719         struct btrfs_fs_devices *seed_devices = NULL;
 720         int ret;
 721
 722         mutex_lock(&uuid_mutex);
 723         ret = __btrfs_close_devices(fs_devices);
 724         if (!fs_devices->opened) {
 725                 seed_devices = fs_devices->seed;
 726                 fs_devices->seed = NULL;
 727         }
 728         mutex_unlock(&uuid_mutex);
 729
 730         while (seed_devices) {
 731                 fs_devices = seed_devices;
 732                 seed_devices = fs_devices->seed;
 733                 __btrfs_close_devices(fs_devices);
 734                 free_fs_devices(fs_devices);
 735         }
 736         /*
 737          * Wait for rcu kworkers under __btrfs_close_devices
 738          * to finish all blkdev_puts so device is really
 739          * free when umount is done.
 740          */
 741         rcu_barrier();
 742         return ret;
 743 }
 744
 745 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 746                                 fmode_t flags, void *holder)
 747 {
 748         struct request_queue *q;
 749         struct block_device *bdev;
 750         struct list_head *head = &fs_devices->devices;
 751         struct btrfs_device *device;
 752         struct block_device *latest_bdev = NULL;
 753         struct buffer_head *bh;
 754         struct btrfs_super_block *disk_super;
 755         u64 latest_devid = 0;
 756         u64 latest_transid = 0;
 757         u64 devid;
 758         int seeding = 1;
 759         int ret = 0;
 760
 761         flags |= FMODE_EXCL;
 762
 763         list_for_each_entry(device, head, dev_list) {
 764                 if (device->bdev)
 765                         continue;
 766                 if (!device->name)
 767                         continue;
 768
 769                 /* Just open everything we can; ignore failures here */
 770                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 771                                             &bdev, &bh))
 772                         continue;
 773
 774                 disk_super = (struct btrfs_super_block *)bh->b_data;
 775                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 776                 if (devid != device->devid)
 777                         goto error_brelse;
 778
 779                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 780                            BTRFS_UUID_SIZE))
 781                         goto error_brelse;
 782
 783                 device->generation = btrfs_super_generation(disk_super);
 784                 if (!latest_transid || device->generation > latest_transid) {
 785                         latest_devid = devid;
 786                         latest_transid = device->generation;
 787                         latest_bdev = bdev;
 788                 }
 789
 790                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 791                         device->writeable = 0;
 792                 } else {
 793                         device->writeable = !bdev_read_only(bdev);
 794                         seeding = 0;
 795                 }
 796
 797                 q = bdev_get_queue(bdev);
 798                 if (blk_queue_discard(q)) {
 799                         device->can_discard = 1;
 800                         fs_devices->num_can_discard++;
 801                 }
 802
 803                 device->bdev = bdev;
 804                 device->in_fs_metadata = 0;
 805                 device->mode = flags;
 806
 807                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 808                         fs_devices->rotating = 1;
 809
 810                 fs_devices->open_devices++;
 811                 if (device->writeable &&
 812                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 813                         fs_devices->rw_devices++;
 814                         list_add(&device->dev_alloc_list,
 815                                  &fs_devices->alloc_list);
 816                 }
 817                 brelse(bh);
 818                 continue;
 819
 820 error_brelse:
 821                 brelse(bh);
 822                 blkdev_put(bdev, flags);
 823                 continue;
 824         }
 825         if (fs_devices->open_devices == 0) {
 826                 ret = -EINVAL;
 827                 goto out;
 828         }
 829         fs_devices->seeding = seeding;
 830         fs_devices->opened = 1;
 831         fs_devices->latest_bdev = latest_bdev;
 832         fs_devices->latest_devid = latest_devid;
 833         fs_devices->latest_trans = latest_transid;
 834         fs_devices->total_rw_bytes = 0;
 835 out:
 836         return ret;
 837 }
 838
 839 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 840                        fmode_t flags, void *holder)
 841 {
 842         int ret;
 843
 844         mutex_lock(&uuid_mutex);
 845         if (fs_devices->opened) {
 846                 fs_devices->opened++;
 847                 ret = 0;
 848         } else {
 849                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 850         }
 851         mutex_unlock(&uuid_mutex);
 852         return ret;
 853 }
 854
 855 /*
 856  * Look for a btrfs signature on a device. This may be called out of the mount path
 857  * and we are not allowed to call set_blocksize during the scan. The superblock
 858  * is read via pagecache
 859  */
 860 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 861                           struct btrfs_fs_devices **fs_devices_ret)
 862 {
 863         struct btrfs_super_block *disk_super;
 864         struct block_device *bdev;
 865         struct page *page;
 866         void *p;
 867         int ret = -EINVAL;
 868         u64 devid;
 869         u64 transid;
 870         u64 total_devices;
 871         u64 bytenr;
 872         pgoff_t index;
 873
 874         /*
 875          * we would like to check all the supers, but that would make
 876          * a btrfs mount succeed after a mkfs from a different FS.
 877          * So, we need to add a special mount option to scan for
 878          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 879          */
 880         bytenr = btrfs_sb_offset(0);
 881         flags |= FMODE_EXCL;
 882         mutex_lock(&uuid_mutex);
 883
 884         bdev = blkdev_get_by_path(path, flags, holder);
 885
 886         if (IS_ERR(bdev)) {
 887                 ret = PTR_ERR(bdev);
 888                 goto error;
 889         }
 890
 891         /* make sure our super fits in the device */
 892         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 893                 goto error_bdev_put;
 894
 895         /* make sure our super fits in the page */
 896         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 897                 goto error_bdev_put;
 898
 899         /* make sure our super doesn't straddle pages on disk */
 900         index = bytenr >> PAGE_CACHE_SHIFT;
 901         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 902                 goto error_bdev_put;
 903
 904         /* pull in the page with our super */
 905         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 906                                    index, GFP_NOFS);
 907
 908         if (IS_ERR_OR_NULL(page))
 909                 goto error_bdev_put;
 910
 911         p = kmap(page);
 912
 913         /* align our pointer to the offset of the super block */
 914         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 915
 916         if (btrfs_super_bytenr(disk_super) != bytenr ||
 917             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
 918                 goto error_unmap;
 919
 920         devid = btrfs_stack_device_id(&disk_super->dev_item);
 921         transid = btrfs_super_generation(disk_super);
 922         total_devices = btrfs_super_num_devices(disk_super);
 923
 924         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 925         if (ret > 0) {
 926                 if (disk_super->label[0]) {
 927                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 928                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 929                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
 930                 } else {
 931                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
 932                 }
 933
 934                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
 935                 ret = 0;
 936         }
 937         if (!ret && fs_devices_ret)
 938                 (*fs_devices_ret)->total_devices = total_devices;
 939
 940 error_unmap:
 941         kunmap(page);
 942         page_cache_release(page);
 943
 944 error_bdev_put:
 945         blkdev_put(bdev, flags);
 946 error:
 947         mutex_unlock(&uuid_mutex);
 948         return ret;
 949 }
 950
 951 /* helper to account the used device space in the range */
 952 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 953                                    u64 end, u64 *length)
 954 {
 955         struct btrfs_key key;
 956         struct btrfs_root *root = device->dev_root;
 957         struct btrfs_dev_extent *dev_extent;
 958         struct btrfs_path *path;
 959         u64 extent_end;
 960         int ret;
 961         int slot;
 962         struct extent_buffer *l;
 963
 964         *length = 0;
 965
 966         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 967                 return 0;
 968
 969         path = btrfs_alloc_path();
 970         if (!path)
 971                 return -ENOMEM;
 972         path->reada = 2;
 973
 974         key.objectid = device->devid;
 975         key.offset = start;
 976         key.type = BTRFS_DEV_EXTENT_KEY;
 977
 978         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 979         if (ret < 0)
 980                 goto out;
 981         if (ret > 0) {
 982                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 983                 if (ret < 0)
 984                         goto out;
 985         }
 986
 987         while (1) {
 988                 l = path->nodes[0];
 989                 slot = path->slots[0];
 990                 if (slot >= btrfs_header_nritems(l)) {
 991                         ret = btrfs_next_leaf(root, path);
 992                         if (ret == 0)
 993                                 continue;
 994                         if (ret < 0)
 995                                 goto out;
 996
 997                         break;
 998                 }
 999                 btrfs_item_key_to_cpu(l, &key, slot);
1000
1001                 if (key.objectid < device->devid)
1002                         goto next;
1003
1004                 if (key.objectid > device->devid)
1005                         break;
1006
1007                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1008                         goto next;
1009
1010                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1011                 extent_end = key.offset + btrfs_dev_extent_length(l,
1012                                                                   dev_extent);
1013                 if (key.offset <= start && extent_end > end) {
1014                         *length = end - start + 1;
1015                         break;
1016                 } else if (key.offset <= start && extent_end > start)
1017                         *length += extent_end - start;
1018                 else if (key.offset > start && extent_end <= end)
1019                         *length += extent_end - key.offset;
1020                 else if (key.offset > start && key.offset <= end) {
1021                         *length += end - key.offset + 1;
1022                         break;
1023                 } else if (key.offset > end)
1024                         break;
1025
1026 next:
1027                 path->slots[0]++;
1028         }
1029         ret = 0;
1030 out:
1031         btrfs_free_path(path);
1032         return ret;
1033 }
1034
1035 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1036                                    struct btrfs_device *device,
1037                                    u64 *start, u64 len)
1038 {
1039         struct extent_map *em;
1040         int ret = 0;
1041
1042         list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1043                 struct map_lookup *map;
1044                 int i;
1045
1046                 map = (struct map_lookup *)em->bdev;
1047                 for (i = 0; i < map->num_stripes; i++) {
1048                         if (map->stripes[i].dev != device)
1049                                 continue;
1050                         if (map->stripes[i].physical >= *start + len ||
1051                             map->stripes[i].physical + em->orig_block_len <=
1052                             *start)
1053                                 continue;
1054                         *start = map->stripes[i].physical +
1055                                 em->orig_block_len;
1056                         ret = 1;
1057                 }
1058         }
1059
1060         return ret;
1061 }
1062
1063
1064 /*
1065  * find_free_dev_extent - find free space in the specified device
1066  * @device:     the device which we search the free space in
1067  * @num_bytes:  the size of the free space that we need
1068  * @start:      store the start of the free space.
1069  * @len:        the size of the free space. that we find, or the size of the max
1070  *              free space if we don't find suitable free space
1071  *
1072  * this uses a pretty simple search, the expectation is that it is
1073  * called very infrequently and that a given device has a small number
1074  * of extents
1075  *
1076  * @start is used to store the start of the free space if we find. But if we
1077  * don't find suitable free space, it will be used to store the start position
1078  * of the max free space.
1079  *
1080  * @len is used to store the size of the free space that we find.
1081  * But if we don't find suitable free space, it is used to store the size of
1082  * the max free space.
1083  */
1084 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1085                          struct btrfs_device *device, u64 num_bytes,
1086                          u64 *start, u64 *len)
1087 {
1088         struct btrfs_key key;
1089         struct btrfs_root *root = device->dev_root;
1090         struct btrfs_dev_extent *dev_extent;
1091         struct btrfs_path *path;
1092         u64 hole_size;
1093         u64 max_hole_start;
1094         u64 max_hole_size;
1095         u64 extent_end;
1096         u64 search_start;
1097         u64 search_end = device->total_bytes;
1098         int ret;
1099         int slot;
1100         struct extent_buffer *l;
1101
1102         /* FIXME use last free of some kind */
1103
1104         /* we don't want to overwrite the superblock on the drive,
1105          * so we make sure to start at an offset of at least 1MB
1106          */
1107         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1108
1109         path = btrfs_alloc_path();
1110         if (!path)
1111                 return -ENOMEM;
1112 again:
1113         max_hole_start = search_start;
1114         max_hole_size = 0;
1115         hole_size = 0;
1116
1117         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1118                 ret = -ENOSPC;
1119                 goto out;
1120         }
1121
1122         path->reada = 2;
1123         path->search_commit_root = 1;
1124         path->skip_locking = 1;
1125
1126         key.objectid = device->devid;
1127         key.offset = search_start;
1128         key.type = BTRFS_DEV_EXTENT_KEY;
1129
1130         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1131         if (ret < 0)
1132                 goto out;
1133         if (ret > 0) {
1134                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1135                 if (ret < 0)
1136                         goto out;
1137         }
1138
1139         while (1) {
1140                 l = path->nodes[0];
1141                 slot = path->slots[0];
1142                 if (slot >= btrfs_header_nritems(l)) {
1143                         ret = btrfs_next_leaf(root, path);
1144                         if (ret == 0)
1145                                 continue;
1146                         if (ret < 0)
1147                                 goto out;
1148
1149                         break;
1150                 }
1151                 btrfs_item_key_to_cpu(l, &key, slot);
1152
1153                 if (key.objectid < device->devid)
1154                         goto next;
1155
1156                 if (key.objectid > device->devid)
1157                         break;
1158
1159                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1160                         goto next;
1161
1162                 if (key.offset > search_start) {
1163                         hole_size = key.offset - search_start;
1164
1165                         /*
1166                          * Have to check before we set max_hole_start, otherwise
1167                          * we could end up sending back this offset anyway.
1168                          */
1169                         if (contains_pending_extent(trans, device,
1170                                                     &search_start,
1171                                                     hole_size))
1172                                 hole_size = 0;
1173
1174                         if (hole_size > max_hole_size) {
1175                                 max_hole_start = search_start;
1176                                 max_hole_size = hole_size;
1177                         }
1178
1179                         /*
1180                          * If this free space is greater than which we need,
1181                          * it must be the max free space that we have found
1182                          * until now, so max_hole_start must point to the start
1183                          * of this free space and the length of this free space
1184                          * is stored in max_hole_size. Thus, we return
1185                          * max_hole_start and max_hole_size and go back to the
1186                          * caller.
1187                          */
1188                         if (hole_size >= num_bytes) {
1189                                 ret = 0;
1190                                 goto out;
1191                         }
1192                 }
1193
1194                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1195                 extent_end = key.offset + btrfs_dev_extent_length(l,
1196                                                                   dev_extent);
1197                 if (extent_end > search_start)
1198                         search_start = extent_end;
1199 next:
1200                 path->slots[0]++;
1201                 cond_resched();
1202         }
1203
1204         /*
1205          * At this point, search_start should be the end of
1206          * allocated dev extents, and when shrinking the device,
1207          * search_end may be smaller than search_start.
1208          */
1209         if (search_end > search_start)
1210                 hole_size = search_end - search_start;
1211
1212         if (hole_size > max_hole_size) {
1213                 max_hole_start = search_start;
1214                 max_hole_size = hole_size;
1215         }
1216
1217         if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1218                 btrfs_release_path(path);
1219                 goto again;
1220         }
1221
1222         /* See above. */
1223         if (hole_size < num_bytes)
1224                 ret = -ENOSPC;
1225         else
1226                 ret = 0;
1227
1228 out:
1229         btrfs_free_path(path);
1230         *start = max_hole_start;
1231         if (len)
1232                 *len = max_hole_size;
1233         return ret;
1234 }
1235
1236 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1237                           struct btrfs_device *device,
1238                           u64 start)
1239 {
1240         int ret;
1241         struct btrfs_path *path;
1242         struct btrfs_root *root = device->dev_root;
1243         struct btrfs_key key;
1244         struct btrfs_key found_key;
1245         struct extent_buffer *leaf = NULL;
1246         struct btrfs_dev_extent *extent = NULL;
1247
1248         path = btrfs_alloc_path();
1249         if (!path)
1250                 return -ENOMEM;
1251
1252         key.objectid = device->devid;
1253         key.offset = start;
1254         key.type = BTRFS_DEV_EXTENT_KEY;
1255 again:
1256         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1257         if (ret > 0) {
1258                 ret = btrfs_previous_item(root, path, key.objectid,
1259                                           BTRFS_DEV_EXTENT_KEY);
1260                 if (ret)
1261                         goto out;
1262                 leaf = path->nodes[0];
1263                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1264                 extent = btrfs_item_ptr(leaf, path->slots[0],
1265                                         struct btrfs_dev_extent);
1266                 BUG_ON(found_key.offset > start || found_key.offset +
1267                        btrfs_dev_extent_length(leaf, extent) < start);
1268                 key = found_key;
1269                 btrfs_release_path(path);
1270                 goto again;
1271         } else if (ret == 0) {
1272                 leaf = path->nodes[0];
1273                 extent = btrfs_item_ptr(leaf, path->slots[0],
1274                                         struct btrfs_dev_extent);
1275         } else {
1276                 btrfs_error(root->fs_info, ret, "Slot search failed");
1277                 goto out;
1278         }
1279
1280         if (device->bytes_used > 0) {
1281                 u64 len = btrfs_dev_extent_length(leaf, extent);
1282                 device->bytes_used -= len;
1283                 spin_lock(&root->fs_info->free_chunk_lock);
1284                 root->fs_info->free_chunk_space += len;
1285                 spin_unlock(&root->fs_info->free_chunk_lock);
1286         }
1287         ret = btrfs_del_item(trans, root, path);
1288         if (ret) {
1289                 btrfs_error(root->fs_info, ret,
1290                             "Failed to remove dev extent item");
1291         }
1292 out:
1293         btrfs_free_path(path);
1294         return ret;
1295 }
1296
1297 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1298                                   struct btrfs_device *device,
1299                                   u64 chunk_tree, u64 chunk_objectid,
1300                                   u64 chunk_offset, u64 start, u64 num_bytes)
1301 {
1302         int ret;
1303         struct btrfs_path *path;
1304         struct btrfs_root *root = device->dev_root;
1305         struct btrfs_dev_extent *extent;
1306         struct extent_buffer *leaf;
1307         struct btrfs_key key;
1308
1309         WARN_ON(!device->in_fs_metadata);
1310         WARN_ON(device->is_tgtdev_for_dev_replace);
1311         path = btrfs_alloc_path();
1312         if (!path)
1313                 return -ENOMEM;
1314
1315         key.objectid = device->devid;
1316         key.offset = start;
1317         key.type = BTRFS_DEV_EXTENT_KEY;
1318         ret = btrfs_insert_empty_item(trans, root, path, &key,
1319                                       sizeof(*extent));
1320         if (ret)
1321                 goto out;
1322
1323         leaf = path->nodes[0];
1324         extent = btrfs_item_ptr(leaf, path->slots[0],
1325                                 struct btrfs_dev_extent);
1326         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1327         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1328         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1329
1330         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1331                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1332
1333         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1334         btrfs_mark_buffer_dirty(leaf);
1335 out:
1336         btrfs_free_path(path);
1337         return ret;
1338 }
1339
1340 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1341 {
1342         struct extent_map_tree *em_tree;
1343         struct extent_map *em;
1344         struct rb_node *n;
1345         u64 ret = 0;
1346
1347         em_tree = &fs_info->mapping_tree.map_tree;
1348         read_lock(&em_tree->lock);
1349         n = rb_last(&em_tree->map);
1350         if (n) {
1351                 em = rb_entry(n, struct extent_map, rb_node);
1352                 ret = em->start + em->len;
1353         }
1354         read_unlock(&em_tree->lock);
1355
1356         return ret;
1357 }
1358
1359 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1360                                     u64 *devid_ret)
1361 {
1362         int ret;
1363         struct btrfs_key key;
1364         struct btrfs_key found_key;
1365         struct btrfs_path *path;
1366
1367         path = btrfs_alloc_path();
1368         if (!path)
1369                 return -ENOMEM;
1370
1371         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1372         key.type = BTRFS_DEV_ITEM_KEY;
1373         key.offset = (u64)-1;
1374
1375         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1376         if (ret < 0)
1377                 goto error;
1378
1379         BUG_ON(ret == 0); /* Corruption */
1380
1381         ret = btrfs_previous_item(fs_info->chunk_root, path,
1382                                   BTRFS_DEV_ITEMS_OBJECTID,
1383                                   BTRFS_DEV_ITEM_KEY);
1384         if (ret) {
1385                 *devid_ret = 1;
1386         } else {
1387                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1388                                       path->slots[0]);
1389                 *devid_ret = found_key.offset + 1;
1390         }
1391         ret = 0;
1392 error:
1393         btrfs_free_path(path);
1394         return ret;
1395 }
1396
1397 /*
1398  * the device information is stored in the chunk root
1399  * the btrfs_device struct should be fully filled in
1400  */
1401 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1402                             struct btrfs_root *root,
1403                             struct btrfs_device *device)
1404 {
1405         int ret;
1406         struct btrfs_path *path;
1407         struct btrfs_dev_item *dev_item;
1408         struct extent_buffer *leaf;
1409         struct btrfs_key key;
1410         unsigned long ptr;
1411
1412         root = root->fs_info->chunk_root;
1413
1414         path = btrfs_alloc_path();
1415         if (!path)
1416                 return -ENOMEM;
1417
1418         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1419         key.type = BTRFS_DEV_ITEM_KEY;
1420         key.offset = device->devid;
1421
1422         ret = btrfs_insert_empty_item(trans, root, path, &key,
1423                                       sizeof(*dev_item));
1424         if (ret)
1425                 goto out;
1426
1427         leaf = path->nodes[0];
1428         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1429
1430         btrfs_set_device_id(leaf, dev_item, device->devid);
1431         btrfs_set_device_generation(leaf, dev_item, 0);
1432         btrfs_set_device_type(leaf, dev_item, device->type);
1433         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1434         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1435         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1436         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1437         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1438         btrfs_set_device_group(leaf, dev_item, 0);
1439         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1440         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1441         btrfs_set_device_start_offset(leaf, dev_item, 0);
1442
1443         ptr = btrfs_device_uuid(dev_item);
1444         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1445         ptr = btrfs_device_fsid(dev_item);
1446         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1447         btrfs_mark_buffer_dirty(leaf);
1448
1449         ret = 0;
1450 out:
1451         btrfs_free_path(path);
1452         return ret;
1453 }
1454
1455 /*
1456  * Function to update ctime/mtime for a given device path.
1457  * Mainly used for ctime/mtime based probe like libblkid.
1458  */
1459 static void update_dev_time(char *path_name)
1460 {
1461         struct file *filp;
1462
1463         filp = filp_open(path_name, O_RDWR, 0);
1464         if (!filp)
1465                 return;
1466         file_update_time(filp);
1467         filp_close(filp, NULL);
1468         return;
1469 }
1470
1471 static int btrfs_rm_dev_item(struct btrfs_root *root,
1472                              struct btrfs_device *device)
1473 {
1474         int ret;
1475         struct btrfs_path *path;
1476         struct btrfs_key key;
1477         struct btrfs_trans_handle *trans;
1478
1479         root = root->fs_info->chunk_root;
1480
1481         path = btrfs_alloc_path();
1482         if (!path)
1483                 return -ENOMEM;
1484
1485         trans = btrfs_start_transaction(root, 0);
1486         if (IS_ERR(trans)) {
1487                 btrfs_free_path(path);
1488                 return PTR_ERR(trans);
1489         }
1490         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1491         key.type = BTRFS_DEV_ITEM_KEY;
1492         key.offset = device->devid;
1493         lock_chunks(root);
1494
1495         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1496         if (ret < 0)
1497                 goto out;
1498
1499         if (ret > 0) {
1500                 ret = -ENOENT;
1501                 goto out;
1502         }
1503
1504         ret = btrfs_del_item(trans, root, path);
1505         if (ret)
1506                 goto out;
1507 out:
1508         btrfs_free_path(path);
1509         unlock_chunks(root);
1510         btrfs_commit_transaction(trans, root);
1511         return ret;
1512 }
1513
1514 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1515 {
1516         struct btrfs_device *device;
1517         struct btrfs_device *next_device;
1518         struct block_device *bdev;
1519         struct buffer_head *bh = NULL;
1520         struct btrfs_super_block *disk_super;
1521         struct btrfs_fs_devices *cur_devices;
1522         u64 all_avail;
1523         u64 devid;
1524         u64 num_devices;
1525         u8 *dev_uuid;
1526         unsigned seq;
1527         int ret = 0;
1528         bool clear_super = false;
1529
1530         mutex_lock(&uuid_mutex);
1531
1532         do {
1533                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1534
1535                 all_avail = root->fs_info->avail_data_alloc_bits |
1536                             root->fs_info->avail_system_alloc_bits |
1537                             root->fs_info->avail_metadata_alloc_bits;
1538         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1539
1540         num_devices = root->fs_info->fs_devices->num_devices;
1541         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1542         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1543                 WARN_ON(num_devices < 1);
1544                 num_devices--;
1545         }
1546         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1547
1548         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1549                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1550                 goto out;
1551         }
1552
1553         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1554                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1555                 goto out;
1556         }
1557
1558         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1559             root->fs_info->fs_devices->rw_devices <= 2) {
1560                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1561                 goto out;
1562         }
1563         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1564             root->fs_info->fs_devices->rw_devices <= 3) {
1565                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1566                 goto out;
1567         }
1568
1569         if (strcmp(device_path, "missing") == 0) {
1570                 struct list_head *devices;
1571                 struct btrfs_device *tmp;
1572
1573                 device = NULL;
1574                 devices = &root->fs_info->fs_devices->devices;
1575                 /*
1576                  * It is safe to read the devices since the volume_mutex
1577                  * is held.
1578                  */
1579                 list_for_each_entry(tmp, devices, dev_list) {
1580                         if (tmp->in_fs_metadata &&
1581                             !tmp->is_tgtdev_for_dev_replace &&
1582                             !tmp->bdev) {
1583                                 device = tmp;
1584                                 break;
1585                         }
1586                 }
1587                 bdev = NULL;
1588                 bh = NULL;
1589                 disk_super = NULL;
1590                 if (!device) {
1591                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1592                         goto out;
1593                 }
1594         } else {
1595                 ret = btrfs_get_bdev_and_sb(device_path,
1596                                             FMODE_WRITE | FMODE_EXCL,
1597                                             root->fs_info->bdev_holder, 0,
1598                                             &bdev, &bh);
1599                 if (ret)
1600                         goto out;
1601                 disk_super = (struct btrfs_super_block *)bh->b_data;
1602                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1603                 dev_uuid = disk_super->dev_item.uuid;
1604                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1605                                            disk_super->fsid);
1606                 if (!device) {
1607                         ret = -ENOENT;
1608                         goto error_brelse;
1609                 }
1610         }
1611
1612         if (device->is_tgtdev_for_dev_replace) {
1613                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1614                 goto error_brelse;
1615         }
1616
1617         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1618                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1619                 goto error_brelse;
1620         }
1621
1622         if (device->writeable) {
1623                 lock_chunks(root);
1624                 list_del_init(&device->dev_alloc_list);
1625                 unlock_chunks(root);
1626                 root->fs_info->fs_devices->rw_devices--;
1627                 clear_super = true;
1628         }
1629
1630         mutex_unlock(&uuid_mutex);
1631         ret = btrfs_shrink_device(device, 0);
1632         mutex_lock(&uuid_mutex);
1633         if (ret)
1634                 goto error_undo;
1635
1636         /*
1637          * TODO: the superblock still includes this device in its num_devices
1638          * counter although write_all_supers() is not locked out. This
1639          * could give a filesystem state which requires a degraded mount.
1640          */
1641         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1642         if (ret)
1643                 goto error_undo;
1644
1645         spin_lock(&root->fs_info->free_chunk_lock);
1646         root->fs_info->free_chunk_space = device->total_bytes -
1647                 device->bytes_used;
1648         spin_unlock(&root->fs_info->free_chunk_lock);
1649
1650         device->in_fs_metadata = 0;
1651         btrfs_scrub_cancel_dev(root->fs_info, device);
1652
1653         /*
1654          * the device list mutex makes sure that we don't change
1655          * the device list while someone else is writing out all
1656          * the device supers. Whoever is writing all supers, should
1657          * lock the device list mutex before getting the number of
1658          * devices in the super block (super_copy). Conversely,
1659          * whoever updates the number of devices in the super block
1660          * (super_copy) should hold the device list mutex.
1661          */
1662
1663         cur_devices = device->fs_devices;
1664         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1665         list_del_rcu(&device->dev_list);
1666
1667         device->fs_devices->num_devices--;
1668         device->fs_devices->total_devices--;
1669
1670         if (device->missing)
1671                 root->fs_info->fs_devices->missing_devices--;
1672
1673         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1674                                  struct btrfs_device, dev_list);
1675         if (device->bdev == root->fs_info->sb->s_bdev)
1676                 root->fs_info->sb->s_bdev = next_device->bdev;
1677         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1678                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1679
1680         if (device->bdev)
1681                 device->fs_devices->open_devices--;
1682
1683         call_rcu(&device->rcu, free_device);
1684
1685         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1686         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1687         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1688
1689         if (cur_devices->open_devices == 0) {
1690                 struct btrfs_fs_devices *fs_devices;
1691                 fs_devices = root->fs_info->fs_devices;
1692                 while (fs_devices) {
1693                         if (fs_devices->seed == cur_devices) {
1694                                 fs_devices->seed = cur_devices->seed;
1695                                 break;
1696                         }
1697                         fs_devices = fs_devices->seed;
1698                 }
1699                 cur_devices->seed = NULL;
1700                 lock_chunks(root);
1701                 __btrfs_close_devices(cur_devices);
1702                 unlock_chunks(root);
1703                 free_fs_devices(cur_devices);
1704         }
1705
1706         root->fs_info->num_tolerated_disk_barrier_failures =
1707                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1708
1709         /*
1710          * at this point, the device is zero sized.  We want to
1711          * remove it from the devices list and zero out the old super
1712          */
1713         if (clear_super && disk_super) {
1714                 u64 bytenr;
1715                 int i;
1716
1717                 /* make sure this device isn't detected as part of
1718                  * the FS anymore
1719                  */
1720                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1721                 set_buffer_dirty(bh);
1722                 sync_dirty_buffer(bh);
1723
1724                 /* clear the mirror copies of super block on the disk
1725                  * being removed, 0th copy is been taken care above and
1726                  * the below would take of the rest
1727                  */
1728                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1729                         bytenr = btrfs_sb_offset(i);
1730                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1731                                         i_size_read(bdev->bd_inode))
1732                                 break;
1733
1734                         brelse(bh);
1735                         bh = __bread(bdev, bytenr / 4096,
1736                                         BTRFS_SUPER_INFO_SIZE);
1737                         if (!bh)
1738                                 continue;
1739
1740                         disk_super = (struct btrfs_super_block *)bh->b_data;
1741
1742                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1743                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1744                                 continue;
1745                         }
1746                         memset(&disk_super->magic, 0,
1747                                                 sizeof(disk_super->magic));
1748                         set_buffer_dirty(bh);
1749                         sync_dirty_buffer(bh);
1750                 }
1751         }
1752
1753         ret = 0;
1754
1755         if (bdev) {
1756                 /* Notify udev that device has changed */
1757                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1758
1759                 /* Update ctime/mtime for device path for libblkid */
1760                 update_dev_time(device_path);
1761         }
1762
1763 error_brelse:
1764         brelse(bh);
1765         if (bdev)
1766                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1767 out:
1768         mutex_unlock(&uuid_mutex);
1769         return ret;
1770 error_undo:
1771         if (device->writeable) {
1772                 lock_chunks(root);
1773                 list_add(&device->dev_alloc_list,
1774                          &root->fs_info->fs_devices->alloc_list);
1775                 unlock_chunks(root);
1776                 root->fs_info->fs_devices->rw_devices++;
1777         }
1778         goto error_brelse;
1779 }
1780
1781 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1782                                  struct btrfs_device *srcdev)
1783 {
1784         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1785
1786         list_del_rcu(&srcdev->dev_list);
1787         list_del_rcu(&srcdev->dev_alloc_list);
1788         fs_info->fs_devices->num_devices--;
1789         if (srcdev->missing) {
1790                 fs_info->fs_devices->missing_devices--;
1791                 fs_info->fs_devices->rw_devices++;
1792         }
1793         if (srcdev->can_discard)
1794                 fs_info->fs_devices->num_can_discard--;
1795         if (srcdev->bdev) {
1796                 fs_info->fs_devices->open_devices--;
1797
1798                 /* zero out the old super */
1799                 btrfs_scratch_superblock(srcdev);
1800         }
1801
1802         call_rcu(&srcdev->rcu, free_device);
1803 }
1804
1805 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1806                                       struct btrfs_device *tgtdev)
1807 {
1808         struct btrfs_device *next_device;
1809
1810         WARN_ON(!tgtdev);
1811         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1812         if (tgtdev->bdev) {
1813                 btrfs_scratch_superblock(tgtdev);
1814                 fs_info->fs_devices->open_devices--;
1815         }
1816         fs_info->fs_devices->num_devices--;
1817         if (tgtdev->can_discard)
1818                 fs_info->fs_devices->num_can_discard++;
1819
1820         next_device = list_entry(fs_info->fs_devices->devices.next,
1821                                  struct btrfs_device, dev_list);
1822         if (tgtdev->bdev == fs_info->sb->s_bdev)
1823                 fs_info->sb->s_bdev = next_device->bdev;
1824         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1825                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1826         list_del_rcu(&tgtdev->dev_list);
1827
1828         call_rcu(&tgtdev->rcu, free_device);
1829
1830         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1831 }
1832
1833 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1834                                      struct btrfs_device **device)
1835 {
1836         int ret = 0;
1837         struct btrfs_super_block *disk_super;
1838         u64 devid;
1839         u8 *dev_uuid;
1840         struct block_device *bdev;
1841         struct buffer_head *bh;
1842
1843         *device = NULL;
1844         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1845                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1846         if (ret)
1847                 return ret;
1848         disk_super = (struct btrfs_super_block *)bh->b_data;
1849         devid = btrfs_stack_device_id(&disk_super->dev_item);
1850         dev_uuid = disk_super->dev_item.uuid;
1851         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1852                                     disk_super->fsid);
1853         brelse(bh);
1854         if (!*device)
1855                 ret = -ENOENT;
1856         blkdev_put(bdev, FMODE_READ);
1857         return ret;
1858 }
1859
1860 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1861                                          char *device_path,
1862                                          struct btrfs_device **device)
1863 {
1864         *device = NULL;
1865         if (strcmp(device_path, "missing") == 0) {
1866                 struct list_head *devices;
1867                 struct btrfs_device *tmp;
1868
1869                 devices = &root->fs_info->fs_devices->devices;
1870                 /*
1871                  * It is safe to read the devices since the volume_mutex
1872                  * is held by the caller.
1873                  */
1874                 list_for_each_entry(tmp, devices, dev_list) {
1875                         if (tmp->in_fs_metadata && !tmp->bdev) {
1876                                 *device = tmp;
1877                                 break;
1878                         }
1879                 }
1880
1881                 if (!*device) {
1882                         btrfs_err(root->fs_info, "no missing device found");
1883                         return -ENOENT;
1884                 }
1885
1886                 return 0;
1887         } else {
1888                 return btrfs_find_device_by_path(root, device_path, device);
1889         }
1890 }
1891
1892 /*
1893  * does all the dirty work required for changing file system's UUID.
1894  */
1895 static int btrfs_prepare_sprout(struct btrfs_root *root)
1896 {
1897         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1898         struct btrfs_fs_devices *old_devices;
1899         struct btrfs_fs_devices *seed_devices;
1900         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1901         struct btrfs_device *device;
1902         u64 super_flags;
1903
1904         BUG_ON(!mutex_is_locked(&uuid_mutex));
1905         if (!fs_devices->seeding)
1906                 return -EINVAL;
1907
1908         seed_devices = __alloc_fs_devices();
1909         if (IS_ERR(seed_devices))
1910                 return PTR_ERR(seed_devices);
1911
1912         old_devices = clone_fs_devices(fs_devices);
1913         if (IS_ERR(old_devices)) {
1914                 kfree(seed_devices);
1915                 return PTR_ERR(old_devices);
1916         }
1917
1918         list_add(&old_devices->list, &fs_uuids);
1919
1920         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1921         seed_devices->opened = 1;
1922         INIT_LIST_HEAD(&seed_devices->devices);
1923         INIT_LIST_HEAD(&seed_devices->alloc_list);
1924         mutex_init(&seed_devices->device_list_mutex);
1925
1926         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1927         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1928                               synchronize_rcu);
1929
1930         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1931         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1932                 device->fs_devices = seed_devices;
1933         }
1934
1935         fs_devices->seeding = 0;
1936         fs_devices->num_devices = 0;
1937         fs_devices->open_devices = 0;
1938         fs_devices->seed = seed_devices;
1939
1940         generate_random_uuid(fs_devices->fsid);
1941         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1942         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1943         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1944
1945         super_flags = btrfs_super_flags(disk_super) &
1946                       ~BTRFS_SUPER_FLAG_SEEDING;
1947         btrfs_set_super_flags(disk_super, super_flags);
1948
1949         return 0;
1950 }
1951
1952 /*
1953  * strore the expected generation for seed devices in device items.
1954  */
1955 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1956                                struct btrfs_root *root)
1957 {
1958         struct btrfs_path *path;
1959         struct extent_buffer *leaf;
1960         struct btrfs_dev_item *dev_item;
1961         struct btrfs_device *device;
1962         struct btrfs_key key;
1963         u8 fs_uuid[BTRFS_UUID_SIZE];
1964         u8 dev_uuid[BTRFS_UUID_SIZE];
1965         u64 devid;
1966         int ret;
1967
1968         path = btrfs_alloc_path();
1969         if (!path)
1970                 return -ENOMEM;
1971
1972         root = root->fs_info->chunk_root;
1973         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1974         key.offset = 0;
1975         key.type = BTRFS_DEV_ITEM_KEY;
1976
1977         while (1) {
1978                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1979                 if (ret < 0)
1980                         goto error;
1981
1982                 leaf = path->nodes[0];
1983 next_slot:
1984                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1985                         ret = btrfs_next_leaf(root, path);
1986                         if (ret > 0)
1987                                 break;
1988                         if (ret < 0)
1989                                 goto error;
1990                         leaf = path->nodes[0];
1991                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1992                         btrfs_release_path(path);
1993                         continue;
1994                 }
1995
1996                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1997                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1998                     key.type != BTRFS_DEV_ITEM_KEY)
1999                         break;
2000
2001                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2002                                           struct btrfs_dev_item);
2003                 devid = btrfs_device_id(leaf, dev_item);
2004                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2005                                    BTRFS_UUID_SIZE);
2006                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2007                                    BTRFS_UUID_SIZE);
2008                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2009                                            fs_uuid);
2010                 BUG_ON(!device); /* Logic error */
2011
2012                 if (device->fs_devices->seeding) {
2013                         btrfs_set_device_generation(leaf, dev_item,
2014                                                     device->generation);
2015                         btrfs_mark_buffer_dirty(leaf);
2016                 }
2017
2018                 path->slots[0]++;
2019                 goto next_slot;
2020         }
2021         ret = 0;
2022 error:
2023         btrfs_free_path(path);
2024         return ret;
2025 }
2026
2027 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2028 {
2029         struct request_queue *q;
2030         struct btrfs_trans_handle *trans;
2031         struct btrfs_device *device;
2032         struct block_device *bdev;
2033         struct list_head *devices;
2034         struct super_block *sb = root->fs_info->sb;
2035         struct rcu_string *name;
2036         u64 total_bytes;
2037         int seeding_dev = 0;
2038         int ret = 0;
2039
2040         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2041                 return -EROFS;
2042
2043         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2044                                   root->fs_info->bdev_holder);
2045         if (IS_ERR(bdev))
2046                 return PTR_ERR(bdev);
2047
2048         if (root->fs_info->fs_devices->seeding) {
2049                 seeding_dev = 1;
2050                 down_write(&sb->s_umount);
2051                 mutex_lock(&uuid_mutex);
2052         }
2053
2054         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2055
2056         devices = &root->fs_info->fs_devices->devices;
2057
2058         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2059         list_for_each_entry(device, devices, dev_list) {
2060                 if (device->bdev == bdev) {
2061                         ret = -EEXIST;
2062                         mutex_unlock(
2063                                 &root->fs_info->fs_devices->device_list_mutex);
2064                         goto error;
2065                 }
2066         }
2067         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2068
2069         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2070         if (IS_ERR(device)) {
2071                 /* we can safely leave the fs_devices entry around */
2072                 ret = PTR_ERR(device);
2073                 goto error;
2074         }
2075
2076         name = rcu_string_strdup(device_path, GFP_NOFS);
2077         if (!name) {
2078                 kfree(device);
2079                 ret = -ENOMEM;
2080                 goto error;
2081         }
2082         rcu_assign_pointer(device->name, name);
2083
2084         trans = btrfs_start_transaction(root, 0);
2085         if (IS_ERR(trans)) {
2086                 rcu_string_free(device->name);
2087                 kfree(device);
2088                 ret = PTR_ERR(trans);
2089                 goto error;
2090         }
2091
2092         lock_chunks(root);
2093
2094         q = bdev_get_queue(bdev);
2095         if (blk_queue_discard(q))
2096                 device->can_discard = 1;
2097         device->writeable = 1;
2098         device->generation = trans->transid;
2099         device->io_width = root->sectorsize;
2100         device->io_align = root->sectorsize;
2101         device->sector_size = root->sectorsize;
2102         device->total_bytes = i_size_read(bdev->bd_inode);
2103         device->disk_total_bytes = device->total_bytes;
2104         device->dev_root = root->fs_info->dev_root;
2105         device->bdev = bdev;
2106         device->in_fs_metadata = 1;
2107         device->is_tgtdev_for_dev_replace = 0;
2108         device->mode = FMODE_EXCL;
2109         device->dev_stats_valid = 1;
2110         set_blocksize(device->bdev, 4096);
2111
2112         if (seeding_dev) {
2113                 sb->s_flags &= ~MS_RDONLY;
2114                 ret = btrfs_prepare_sprout(root);
2115                 BUG_ON(ret); /* -ENOMEM */
2116         }
2117
2118         device->fs_devices = root->fs_info->fs_devices;
2119
2120         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2121         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2122         list_add(&device->dev_alloc_list,
2123                  &root->fs_info->fs_devices->alloc_list);
2124         root->fs_info->fs_devices->num_devices++;
2125         root->fs_info->fs_devices->open_devices++;
2126         root->fs_info->fs_devices->rw_devices++;
2127         root->fs_info->fs_devices->total_devices++;
2128         if (device->can_discard)
2129                 root->fs_info->fs_devices->num_can_discard++;
2130         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2131
2132         spin_lock(&root->fs_info->free_chunk_lock);
2133         root->fs_info->free_chunk_space += device->total_bytes;
2134         spin_unlock(&root->fs_info->free_chunk_lock);
2135
2136         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2137                 root->fs_info->fs_devices->rotating = 1;
2138
2139         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2140         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2141                                     total_bytes + device->total_bytes);
2142
2143         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2144         btrfs_set_super_num_devices(root->fs_info->super_copy,
2145                                     total_bytes + 1);
2146         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2147
2148         if (seeding_dev) {
2149                 ret = init_first_rw_device(trans, root, device);
2150                 if (ret) {
2151                         btrfs_abort_transaction(trans, root, ret);
2152                         goto error_trans;
2153                 }
2154                 ret = btrfs_finish_sprout(trans, root);
2155                 if (ret) {
2156                         btrfs_abort_transaction(trans, root, ret);
2157                         goto error_trans;
2158                 }
2159         } else {
2160                 ret = btrfs_add_device(trans, root, device);
2161                 if (ret) {
2162                         btrfs_abort_transaction(trans, root, ret);
2163                         goto error_trans;
2164                 }
2165         }
2166
2167         /*
2168          * we've got more storage, clear any full flags on the space
2169          * infos
2170          */
2171         btrfs_clear_space_info_full(root->fs_info);
2172
2173         unlock_chunks(root);
2174         root->fs_info->num_tolerated_disk_barrier_failures =
2175                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2176         ret = btrfs_commit_transaction(trans, root);
2177
2178         if (seeding_dev) {
2179                 mutex_unlock(&uuid_mutex);
2180                 up_write(&sb->s_umount);
2181
2182                 if (ret) /* transaction commit */
2183                         return ret;
2184
2185                 ret = btrfs_relocate_sys_chunks(root);
2186                 if (ret < 0)
2187                         btrfs_error(root->fs_info, ret,
2188                                     "Failed to relocate sys chunks after "
2189                                     "device initialization. This can be fixed "
2190                                     "using the \"btrfs balance\" command.");
2191                 trans = btrfs_attach_transaction(root);
2192                 if (IS_ERR(trans)) {
2193                         if (PTR_ERR(trans) == -ENOENT)
2194                                 return 0;
2195                         return PTR_ERR(trans);
2196                 }
2197                 ret = btrfs_commit_transaction(trans, root);
2198         }
2199
2200         /* Update ctime/mtime for libblkid */
2201         update_dev_time(device_path);
2202         return ret;
2203
2204 error_trans:
2205         unlock_chunks(root);
2206         btrfs_end_transaction(trans, root);
2207         rcu_string_free(device->name);
2208         kfree(device);
2209 error:
2210         blkdev_put(bdev, FMODE_EXCL);
2211         if (seeding_dev) {
2212                 mutex_unlock(&uuid_mutex);
2213                 up_write(&sb->s_umount);
2214         }
2215         return ret;
2216 }
2217
2218 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2219                                   struct btrfs_device **device_out)
2220 {
2221         struct request_queue *q;
2222         struct btrfs_device *device;
2223         struct block_device *bdev;
2224         struct btrfs_fs_info *fs_info = root->fs_info;
2225         struct list_head *devices;
2226         struct rcu_string *name;
2227         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2228         int ret = 0;
2229
2230         *device_out = NULL;
2231         if (fs_info->fs_devices->seeding)
2232                 return -EINVAL;
2233
2234         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2235                                   fs_info->bdev_holder);
2236         if (IS_ERR(bdev))
2237                 return PTR_ERR(bdev);
2238
2239         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2240
2241         devices = &fs_info->fs_devices->devices;
2242         list_for_each_entry(device, devices, dev_list) {
2243                 if (device->bdev == bdev) {
2244                         ret = -EEXIST;
2245                         goto error;
2246                 }
2247         }
2248
2249         device = btrfs_alloc_device(NULL, &devid, NULL);
2250         if (IS_ERR(device)) {
2251                 ret = PTR_ERR(device);
2252                 goto error;
2253         }
2254
2255         name = rcu_string_strdup(device_path, GFP_NOFS);
2256         if (!name) {
2257                 kfree(device);
2258                 ret = -ENOMEM;
2259                 goto error;
2260         }
2261         rcu_assign_pointer(device->name, name);
2262
2263         q = bdev_get_queue(bdev);
2264         if (blk_queue_discard(q))
2265                 device->can_discard = 1;
2266         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2267         device->writeable = 1;
2268         device->generation = 0;
2269         device->io_width = root->sectorsize;
2270         device->io_align = root->sectorsize;
2271         device->sector_size = root->sectorsize;
2272         device->total_bytes = i_size_read(bdev->bd_inode);
2273         device->disk_total_bytes = device->total_bytes;
2274         device->dev_root = fs_info->dev_root;
2275         device->bdev = bdev;
2276         device->in_fs_metadata = 1;
2277         device->is_tgtdev_for_dev_replace = 1;
2278         device->mode = FMODE_EXCL;
2279         device->dev_stats_valid = 1;
2280         set_blocksize(device->bdev, 4096);
2281         device->fs_devices = fs_info->fs_devices;
2282         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2283         fs_info->fs_devices->num_devices++;
2284         fs_info->fs_devices->open_devices++;
2285         if (device->can_discard)
2286                 fs_info->fs_devices->num_can_discard++;
2287         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2288
2289         *device_out = device;
2290         return ret;
2291
2292 error:
2293         blkdev_put(bdev, FMODE_EXCL);
2294         return ret;
2295 }
2296
2297 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2298                                               struct btrfs_device *tgtdev)
2299 {
2300         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2301         tgtdev->io_width = fs_info->dev_root->sectorsize;
2302         tgtdev->io_align = fs_info->dev_root->sectorsize;
2303         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2304         tgtdev->dev_root = fs_info->dev_root;
2305         tgtdev->in_fs_metadata = 1;
2306 }
2307
2308 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2309                                         struct btrfs_device *device)
2310 {
2311         int ret;
2312         struct btrfs_path *path;
2313         struct btrfs_root *root;
2314         struct btrfs_dev_item *dev_item;
2315         struct extent_buffer *leaf;
2316         struct btrfs_key key;
2317
2318         root = device->dev_root->fs_info->chunk_root;
2319
2320         path = btrfs_alloc_path();
2321         if (!path)
2322                 return -ENOMEM;
2323
2324         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2325         key.type = BTRFS_DEV_ITEM_KEY;
2326         key.offset = device->devid;
2327
2328         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2329         if (ret < 0)
2330                 goto out;
2331
2332         if (ret > 0) {
2333                 ret = -ENOENT;
2334                 goto out;
2335         }
2336
2337         leaf = path->nodes[0];
2338         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2339
2340         btrfs_set_device_id(leaf, dev_item, device->devid);
2341         btrfs_set_device_type(leaf, dev_item, device->type);
2342         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2343         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2344         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2345         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2346         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2347         btrfs_mark_buffer_dirty(leaf);
2348
2349 out:
2350         btrfs_free_path(path);
2351         return ret;
2352 }
2353
2354 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2355                       struct btrfs_device *device, u64 new_size)
2356 {
2357         struct btrfs_super_block *super_copy =
2358                 device->dev_root->fs_info->super_copy;
2359         u64 old_total = btrfs_super_total_bytes(super_copy);
2360         u64 diff = new_size - device->total_bytes;
2361
2362         if (!device->writeable)
2363                 return -EACCES;
2364         if (new_size <= device->total_bytes ||
2365             device->is_tgtdev_for_dev_replace)
2366                 return -EINVAL;
2367
2368         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2369         device->fs_devices->total_rw_bytes += diff;
2370
2371         device->total_bytes = new_size;
2372         device->disk_total_bytes = new_size;
2373         btrfs_clear_space_info_full(device->dev_root->fs_info);
2374
2375         return btrfs_update_device(trans, device);
2376 }
2377
2378 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2379                       struct btrfs_device *device, u64 new_size)
2380 {
2381         int ret;
2382         lock_chunks(device->dev_root);
2383         ret = __btrfs_grow_device(trans, device, new_size);
2384         unlock_chunks(device->dev_root);
2385         return ret;
2386 }
2387
2388 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2389                             struct btrfs_root *root,
2390                             u64 chunk_tree, u64 chunk_objectid,
2391                             u64 chunk_offset)
2392 {
2393         int ret;
2394         struct btrfs_path *path;
2395         struct btrfs_key key;
2396
2397         root = root->fs_info->chunk_root;
2398         path = btrfs_alloc_path();
2399         if (!path)
2400                 return -ENOMEM;
2401
2402         key.objectid = chunk_objectid;
2403         key.offset = chunk_offset;
2404         key.type = BTRFS_CHUNK_ITEM_KEY;
2405
2406         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2407         if (ret < 0)
2408                 goto out;
2409         else if (ret > 0) { /* Logic error or corruption */
2410                 btrfs_error(root->fs_info, -ENOENT,
2411                             "Failed lookup while freeing chunk.");
2412                 ret = -ENOENT;
2413                 goto out;
2414         }
2415
2416         ret = btrfs_del_item(trans, root, path);
2417         if (ret < 0)
2418                 btrfs_error(root->fs_info, ret,
2419                             "Failed to delete chunk item.");
2420 out:
2421         btrfs_free_path(path);
2422         return ret;
2423 }
2424
2425 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2426                         chunk_offset)
2427 {
2428         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2429         struct btrfs_disk_key *disk_key;
2430         struct btrfs_chunk *chunk;
2431         u8 *ptr;
2432         int ret = 0;
2433         u32 num_stripes;
2434         u32 array_size;
2435         u32 len = 0;
2436         u32 cur;
2437         struct btrfs_key key;
2438
2439         array_size = btrfs_super_sys_array_size(super_copy);
2440
2441         ptr = super_copy->sys_chunk_array;
2442         cur = 0;
2443
2444         while (cur < array_size) {
2445                 disk_key = (struct btrfs_disk_key *)ptr;
2446                 btrfs_disk_key_to_cpu(&key, disk_key);
2447
2448                 len = sizeof(*disk_key);
2449
2450                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2451                         chunk = (struct btrfs_chunk *)(ptr + len);
2452                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2453                         len += btrfs_chunk_item_size(num_stripes);
2454                 } else {
2455                         ret = -EIO;
2456                         break;
2457                 }
2458                 if (key.objectid == chunk_objectid &&
2459                     key.offset == chunk_offset) {
2460                         memmove(ptr, ptr + len, array_size - (cur + len));
2461                         array_size -= len;
2462                         btrfs_set_super_sys_array_size(super_copy, array_size);
2463                 } else {
2464                         ptr += len;
2465                         cur += len;
2466                 }
2467         }
2468         return ret;
2469 }
2470
2471 static int btrfs_relocate_chunk(struct btrfs_root *root,
2472                          u64 chunk_tree, u64 chunk_objectid,
2473                          u64 chunk_offset)
2474 {
2475         struct extent_map_tree *em_tree;
2476         struct btrfs_root *extent_root;
2477         struct btrfs_trans_handle *trans;
2478         struct extent_map *em;
2479         struct map_lookup *map;
2480         int ret;
2481         int i;
2482
2483         root = root->fs_info->chunk_root;
2484         extent_root = root->fs_info->extent_root;
2485         em_tree = &root->fs_info->mapping_tree.map_tree;
2486
2487         ret = btrfs_can_relocate(extent_root, chunk_offset);
2488         if (ret)
2489                 return -ENOSPC;
2490
2491         /* step one, relocate all the extents inside this chunk */
2492         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2493         if (ret)
2494                 return ret;
2495
2496         trans = btrfs_start_transaction(root, 0);
2497         if (IS_ERR(trans)) {
2498                 ret = PTR_ERR(trans);
2499                 btrfs_std_error(root->fs_info, ret);
2500                 return ret;
2501         }
2502
2503         lock_chunks(root);
2504
2505         /*
2506          * step two, delete the device extents and the
2507          * chunk tree entries
2508          */
2509         read_lock(&em_tree->lock);
2510         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2511         read_unlock(&em_tree->lock);
2512
2513         BUG_ON(!em || em->start > chunk_offset ||
2514                em->start + em->len < chunk_offset);
2515         map = (struct map_lookup *)em->bdev;
2516
2517         for (i = 0; i < map->num_stripes; i++) {
2518                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2519                                             map->stripes[i].physical);
2520                 BUG_ON(ret);
2521
2522                 if (map->stripes[i].dev) {
2523                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2524                         BUG_ON(ret);
2525                 }
2526         }
2527         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2528                                chunk_offset);
2529
2530         BUG_ON(ret);
2531
2532         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2533
2534         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2535                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2536                 BUG_ON(ret);
2537         }
2538
2539         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2540         BUG_ON(ret);
2541
2542         write_lock(&em_tree->lock);
2543         remove_extent_mapping(em_tree, em);
2544         write_unlock(&em_tree->lock);
2545
2546         kfree(map);
2547         em->bdev = NULL;
2548
2549         /* once for the tree */
2550         free_extent_map(em);
2551         /* once for us */
2552         free_extent_map(em);
2553
2554         unlock_chunks(root);
2555         btrfs_end_transaction(trans, root);
2556         return 0;
2557 }
2558
2559 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2560 {
2561         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2562         struct btrfs_path *path;
2563         struct extent_buffer *leaf;
2564         struct btrfs_chunk *chunk;
2565         struct btrfs_key key;
2566         struct btrfs_key found_key;
2567         u64 chunk_tree = chunk_root->root_key.objectid;
2568         u64 chunk_type;
2569         bool retried = false;
2570         int failed = 0;
2571         int ret;
2572
2573         path = btrfs_alloc_path();
2574         if (!path)
2575                 return -ENOMEM;
2576
2577 again:
2578         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2579         key.offset = (u64)-1;
2580         key.type = BTRFS_CHUNK_ITEM_KEY;
2581
2582         while (1) {
2583                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2584                 if (ret < 0)
2585                         goto error;
2586                 BUG_ON(ret == 0); /* Corruption */
2587
2588                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2589                                           key.type);
2590                 if (ret < 0)
2591                         goto error;
2592                 if (ret > 0)
2593                         break;
2594
2595                 leaf = path->nodes[0];
2596                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2597
2598                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2599                                        struct btrfs_chunk);
2600                 chunk_type = btrfs_chunk_type(leaf, chunk);
2601                 btrfs_release_path(path);
2602
2603                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2604                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2605                                                    found_key.objectid,
2606                                                    found_key.offset);
2607                         if (ret == -ENOSPC)
2608                                 failed++;
2609                         else if (ret)
2610                                 BUG();
2611                 }
2612
2613                 if (found_key.offset == 0)
2614                         break;
2615                 key.offset = found_key.offset - 1;
2616         }
2617         ret = 0;
2618         if (failed && !retried) {
2619                 failed = 0;
2620                 retried = true;
2621                 goto again;
2622         } else if (WARN_ON(failed && retried)) {
2623                 ret = -ENOSPC;
2624         }
2625 error:
2626         btrfs_free_path(path);
2627         return ret;
2628 }
2629
2630 static int insert_balance_item(struct btrfs_root *root,
2631                                struct btrfs_balance_control *bctl)
2632 {
2633         struct btrfs_trans_handle *trans;
2634         struct btrfs_balance_item *item;
2635         struct btrfs_disk_balance_args disk_bargs;
2636         struct btrfs_path *path;
2637         struct extent_buffer *leaf;
2638         struct btrfs_key key;
2639         int ret, err;
2640
2641         path = btrfs_alloc_path();
2642         if (!path)
2643                 return -ENOMEM;
2644
2645         trans = btrfs_start_transaction(root, 0);
2646         if (IS_ERR(trans)) {
2647                 btrfs_free_path(path);
2648                 return PTR_ERR(trans);
2649         }
2650
2651         key.objectid = BTRFS_BALANCE_OBJECTID;
2652         key.type = BTRFS_BALANCE_ITEM_KEY;
2653         key.offset = 0;
2654
2655         ret = btrfs_insert_empty_item(trans, root, path, &key,
2656                                       sizeof(*item));
2657         if (ret)
2658                 goto out;
2659
2660         leaf = path->nodes[0];
2661         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2662
2663         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2664
2665         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2666         btrfs_set_balance_data(leaf, item, &disk_bargs);
2667         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2668         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2669         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2670         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2671
2672         btrfs_set_balance_flags(leaf, item, bctl->flags);
2673
2674         btrfs_mark_buffer_dirty(leaf);
2675 out:
2676         btrfs_free_path(path);
2677         err = btrfs_commit_transaction(trans, root);
2678         if (err && !ret)
2679                 ret = err;
2680         return ret;
2681 }
2682
2683 static int del_balance_item(struct btrfs_root *root)
2684 {
2685         struct btrfs_trans_handle *trans;
2686         struct btrfs_path *path;
2687         struct btrfs_key key;
2688         int ret, err;
2689
2690         path = btrfs_alloc_path();
2691         if (!path)
2692                 return -ENOMEM;
2693
2694         trans = btrfs_start_transaction(root, 0);
2695         if (IS_ERR(trans)) {
2696                 btrfs_free_path(path);
2697                 return PTR_ERR(trans);
2698         }
2699
2700         key.objectid = BTRFS_BALANCE_OBJECTID;
2701         key.type = BTRFS_BALANCE_ITEM_KEY;
2702         key.offset = 0;
2703
2704         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2705         if (ret < 0)
2706                 goto out;
2707         if (ret > 0) {
2708                 ret = -ENOENT;
2709                 goto out;
2710         }
2711
2712         ret = btrfs_del_item(trans, root, path);
2713 out:
2714         btrfs_free_path(path);
2715         err = btrfs_commit_transaction(trans, root);
2716         if (err && !ret)
2717                 ret = err;
2718         return ret;
2719 }
2720
2721 /*
2722  * This is a heuristic used to reduce the number of chunks balanced on
2723  * resume after balance was interrupted.
2724  */
2725 static void update_balance_args(struct btrfs_balance_control *bctl)
2726 {
2727         /*
2728          * Turn on soft mode for chunk types that were being converted.
2729          */
2730         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2731                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2732         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2733                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2734         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2735                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2736
2737         /*
2738          * Turn on usage filter if is not already used.  The idea is
2739          * that chunks that we have already balanced should be
2740          * reasonably full.  Don't do it for chunks that are being
2741          * converted - that will keep us from relocating unconverted
2742          * (albeit full) chunks.
2743          */
2744         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2745             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2746                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2747                 bctl->data.usage = 90;
2748         }
2749         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2750             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2751                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2752                 bctl->sys.usage = 90;
2753         }
2754         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2755             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2756                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2757                 bctl->meta.usage = 90;
2758         }
2759 }
2760
2761 /*
2762  * Should be called with both balance and volume mutexes held to
2763  * serialize other volume operations (add_dev/rm_dev/resize) with
2764  * restriper.  Same goes for unset_balance_control.
2765  */
2766 static void set_balance_control(struct btrfs_balance_control *bctl)
2767 {
2768         struct btrfs_fs_info *fs_info = bctl->fs_info;
2769
2770         BUG_ON(fs_info->balance_ctl);
2771
2772         spin_lock(&fs_info->balance_lock);
2773         fs_info->balance_ctl = bctl;
2774         spin_unlock(&fs_info->balance_lock);
2775 }
2776
2777 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2778 {
2779         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2780
2781         BUG_ON(!fs_info->balance_ctl);
2782
2783         spin_lock(&fs_info->balance_lock);
2784         fs_info->balance_ctl = NULL;
2785         spin_unlock(&fs_info->balance_lock);
2786
2787         kfree(bctl);
2788 }
2789
2790 /*
2791  * Balance filters.  Return 1 if chunk should be filtered out
2792  * (should not be balanced).
2793  */
2794 static int chunk_profiles_filter(u64 chunk_type,
2795                                  struct btrfs_balance_args *bargs)
2796 {
2797         chunk_type = chunk_to_extended(chunk_type) &
2798                                 BTRFS_EXTENDED_PROFILE_MASK;
2799
2800         if (bargs->profiles & chunk_type)
2801                 return 0;
2802
2803         return 1;
2804 }
2805
2806 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2807                               struct btrfs_balance_args *bargs)
2808 {
2809         struct btrfs_block_group_cache *cache;
2810         u64 chunk_used, user_thresh;
2811         int ret = 1;
2812
2813         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2814         chunk_used = btrfs_block_group_used(&cache->item);
2815
2816         if (bargs->usage == 0)
2817                 user_thresh = 1;
2818         else if (bargs->usage > 100)
2819                 user_thresh = cache->key.offset;
2820         else
2821                 user_thresh = div_factor_fine(cache->key.offset,
2822                                               bargs->usage);
2823
2824         if (chunk_used < user_thresh)
2825                 ret = 0;
2826
2827         btrfs_put_block_group(cache);
2828         return ret;
2829 }
2830
2831 static int chunk_devid_filter(struct extent_buffer *leaf,
2832                               struct btrfs_chunk *chunk,
2833                               struct btrfs_balance_args *bargs)
2834 {
2835         struct btrfs_stripe *stripe;
2836         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2837         int i;
2838
2839         for (i = 0; i < num_stripes; i++) {
2840                 stripe = btrfs_stripe_nr(chunk, i);
2841                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2842                         return 0;
2843         }
2844
2845         return 1;
2846 }
2847
2848 /* [pstart, pend) */
2849 static int chunk_drange_filter(struct extent_buffer *leaf,
2850                                struct btrfs_chunk *chunk,
2851                                u64 chunk_offset,
2852                                struct btrfs_balance_args *bargs)
2853 {
2854         struct btrfs_stripe *stripe;
2855         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2856         u64 stripe_offset;
2857         u64 stripe_length;
2858         int factor;
2859         int i;
2860
2861         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2862                 return 0;
2863
2864         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2865              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2866                 factor = num_stripes / 2;
2867         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2868                 factor = num_stripes - 1;
2869         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2870                 factor = num_stripes - 2;
2871         } else {
2872                 factor = num_stripes;
2873         }
2874
2875         for (i = 0; i < num_stripes; i++) {
2876                 stripe = btrfs_stripe_nr(chunk, i);
2877                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2878                         continue;
2879
2880                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2881                 stripe_length = btrfs_chunk_length(leaf, chunk);
2882                 do_div(stripe_length, factor);
2883
2884                 if (stripe_offset < bargs->pend &&
2885                     stripe_offset + stripe_length > bargs->pstart)
2886                         return 0;
2887         }
2888
2889         return 1;
2890 }
2891
2892 /* [vstart, vend) */
2893 static int chunk_vrange_filter(struct extent_buffer *leaf,
2894                                struct btrfs_chunk *chunk,
2895                                u64 chunk_offset,
2896                                struct btrfs_balance_args *bargs)
2897 {
2898         if (chunk_offset < bargs->vend &&
2899             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2900                 /* at least part of the chunk is inside this vrange */
2901                 return 0;
2902
2903         return 1;
2904 }
2905
2906 static int chunk_soft_convert_filter(u64 chunk_type,
2907                                      struct btrfs_balance_args *bargs)
2908 {
2909         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2910                 return 0;
2911
2912         chunk_type = chunk_to_extended(chunk_type) &
2913                                 BTRFS_EXTENDED_PROFILE_MASK;
2914
2915         if (bargs->target == chunk_type)
2916                 return 1;
2917
2918         return 0;
2919 }
2920
2921 static int should_balance_chunk(struct btrfs_root *root,
2922                                 struct extent_buffer *leaf,
2923                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2924 {
2925         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2926         struct btrfs_balance_args *bargs = NULL;
2927         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2928
2929         /* type filter */
2930         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2931               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2932                 return 0;
2933         }
2934
2935         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2936                 bargs = &bctl->data;
2937         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2938                 bargs = &bctl->sys;
2939         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2940                 bargs = &bctl->meta;
2941
2942         /* profiles filter */
2943         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2944             chunk_profiles_filter(chunk_type, bargs)) {
2945                 return 0;
2946         }
2947
2948         /* usage filter */
2949         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2950             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2951                 return 0;
2952         }
2953
2954         /* devid filter */
2955         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2956             chunk_devid_filter(leaf, chunk, bargs)) {
2957                 return 0;
2958         }
2959
2960         /* drange filter, makes sense only with devid filter */
2961         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2962             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2963                 return 0;
2964         }
2965
2966         /* vrange filter */
2967         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2968             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2969                 return 0;
2970         }
2971
2972         /* soft profile changing mode */
2973         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2974             chunk_soft_convert_filter(chunk_type, bargs)) {
2975                 return 0;
2976         }
2977
2978         /*
2979          * limited by count, must be the last filter
2980          */
2981         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
2982                 if (bargs->limit == 0)
2983                         return 0;
2984                 else
2985                         bargs->limit--;
2986         }
2987
2988         return 1;
2989 }
2990
2991 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2992 {
2993         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2994         struct btrfs_root *chunk_root = fs_info->chunk_root;
2995         struct btrfs_root *dev_root = fs_info->dev_root;
2996         struct list_head *devices;
2997         struct btrfs_device *device;
2998         u64 old_size;
2999         u64 size_to_free;
3000         struct btrfs_chunk *chunk;
3001         struct btrfs_path *path;
3002         struct btrfs_key key;
3003         struct btrfs_key found_key;
3004         struct btrfs_trans_handle *trans;
3005         struct extent_buffer *leaf;
3006         int slot;
3007         int ret;
3008         int enospc_errors = 0;
3009         bool counting = true;
3010         u64 limit_data = bctl->data.limit;
3011         u64 limit_meta = bctl->meta.limit;
3012         u64 limit_sys = bctl->sys.limit;
3013
3014         /* step one make some room on all the devices */
3015         devices = &fs_info->fs_devices->devices;
3016         list_for_each_entry(device, devices, dev_list) {
3017                 old_size = device->total_bytes;
3018                 size_to_free = div_factor(old_size, 1);
3019                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3020                 if (!device->writeable ||
3021                     device->total_bytes - device->bytes_used > size_to_free ||
3022                     device->is_tgtdev_for_dev_replace)
3023                         continue;
3024
3025                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3026                 if (ret == -ENOSPC)
3027                         break;
3028                 BUG_ON(ret);
3029
3030                 trans = btrfs_start_transaction(dev_root, 0);
3031                 BUG_ON(IS_ERR(trans));
3032
3033                 ret = btrfs_grow_device(trans, device, old_size);
3034                 BUG_ON(ret);
3035
3036                 btrfs_end_transaction(trans, dev_root);
3037         }
3038
3039         /* step two, relocate all the chunks */
3040         path = btrfs_alloc_path();
3041         if (!path) {
3042                 ret = -ENOMEM;
3043                 goto error;
3044         }
3045
3046         /* zero out stat counters */
3047         spin_lock(&fs_info->balance_lock);
3048         memset(&bctl->stat, 0, sizeof(bctl->stat));
3049         spin_unlock(&fs_info->balance_lock);
3050 again:
3051         if (!counting) {
3052                 bctl->data.limit = limit_data;
3053                 bctl->meta.limit = limit_meta;
3054                 bctl->sys.limit = limit_sys;
3055         }
3056         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3057         key.offset = (u64)-1;
3058         key.type = BTRFS_CHUNK_ITEM_KEY;
3059
3060         while (1) {
3061                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3062                     atomic_read(&fs_info->balance_cancel_req)) {
3063                         ret = -ECANCELED;
3064                         goto error;
3065                 }
3066
3067                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3068                 if (ret < 0)
3069                         goto error;
3070
3071                 /*
3072                  * this shouldn't happen, it means the last relocate
3073                  * failed
3074                  */
3075                 if (ret == 0)
3076                         BUG(); /* FIXME break ? */
3077
3078                 ret = btrfs_previous_item(chunk_root, path, 0,
3079                                           BTRFS_CHUNK_ITEM_KEY);
3080                 if (ret) {
3081                         ret = 0;
3082                         break;
3083                 }
3084
3085                 leaf = path->nodes[0];
3086                 slot = path->slots[0];
3087                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3088
3089                 if (found_key.objectid != key.objectid)
3090                         break;
3091
3092                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3093
3094                 if (!counting) {
3095                         spin_lock(&fs_info->balance_lock);
3096                         bctl->stat.considered++;
3097                         spin_unlock(&fs_info->balance_lock);
3098                 }
3099
3100                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3101                                            found_key.offset);
3102                 btrfs_release_path(path);
3103                 if (!ret)
3104                         goto loop;
3105
3106                 if (counting) {
3107                         spin_lock(&fs_info->balance_lock);
3108                         bctl->stat.expected++;
3109                         spin_unlock(&fs_info->balance_lock);
3110                         goto loop;
3111                 }
3112
3113                 ret = btrfs_relocate_chunk(chunk_root,
3114                                            chunk_root->root_key.objectid,
3115                                            found_key.objectid,
3116                                            found_key.offset);
3117                 if (ret && ret != -ENOSPC)
3118                         goto error;
3119                 if (ret == -ENOSPC) {
3120                         enospc_errors++;
3121                 } else {
3122                         spin_lock(&fs_info->balance_lock);
3123                         bctl->stat.completed++;
3124                         spin_unlock(&fs_info->balance_lock);
3125                 }
3126 loop:
3127                 if (found_key.offset == 0)
3128                         break;
3129                 key.offset = found_key.offset - 1;
3130         }
3131
3132         if (counting) {
3133                 btrfs_release_path(path);
3134                 counting = false;
3135                 goto again;
3136         }
3137 error:
3138         btrfs_free_path(path);
3139         if (enospc_errors) {
3140                 btrfs_info(fs_info, "%d enospc errors during balance",
3141                        enospc_errors);
3142                 if (!ret)
3143                         ret = -ENOSPC;
3144         }
3145
3146         return ret;
3147 }
3148
3149 /**
3150  * alloc_profile_is_valid - see if a given profile is valid and reduced
3151  * @flags: profile to validate
3152  * @extended: if true @flags is treated as an extended profile
3153  */
3154 static int alloc_profile_is_valid(u64 flags, int extended)
3155 {
3156         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3157                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3158
3159         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3160
3161         /* 1) check that all other bits are zeroed */
3162         if (flags & ~mask)
3163                 return 0;
3164
3165         /* 2) see if profile is reduced */
3166         if (flags == 0)
3167                 return !extended; /* "0" is valid for usual profiles */
3168
3169         /* true if exactly one bit set */
3170         return (flags & (flags - 1)) == 0;
3171 }
3172
3173 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3174 {
3175         /* cancel requested || normal exit path */
3176         return atomic_read(&fs_info->balance_cancel_req) ||
3177                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3178                  atomic_read(&fs_info->balance_cancel_req) == 0);
3179 }
3180
3181 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3182 {
3183         int ret;
3184
3185         unset_balance_control(fs_info);
3186         ret = del_balance_item(fs_info->tree_root);
3187         if (ret)
3188                 btrfs_std_error(fs_info, ret);
3189
3190         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3191 }
3192
3193 /*
3194  * Should be called with both balance and volume mutexes held
3195  */
3196 int btrfs_balance(struct btrfs_balance_control *bctl,
3197                   struct btrfs_ioctl_balance_args *bargs)
3198 {
3199         struct btrfs_fs_info *fs_info = bctl->fs_info;
3200         u64 allowed;
3201         int mixed = 0;
3202         int ret;
3203         u64 num_devices;
3204         unsigned seq;
3205
3206         if (btrfs_fs_closing(fs_info) ||
3207             atomic_read(&fs_info->balance_pause_req) ||
3208             atomic_read(&fs_info->balance_cancel_req)) {
3209                 ret = -EINVAL;
3210                 goto out;
3211         }
3212
3213         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3214         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3215                 mixed = 1;
3216
3217         /*
3218          * In case of mixed groups both data and meta should be picked,
3219          * and identical options should be given for both of them.
3220          */
3221         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3222         if (mixed && (bctl->flags & allowed)) {
3223                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3224                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3225                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3226                         btrfs_err(fs_info, "with mixed groups data and "
3227                                    "metadata balance options must be the same");
3228                         ret = -EINVAL;
3229                         goto out;
3230                 }
3231         }
3232
3233         num_devices = fs_info->fs_devices->num_devices;
3234         btrfs_dev_replace_lock(&fs_info->dev_replace);
3235         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3236                 BUG_ON(num_devices < 1);
3237                 num_devices--;
3238         }
3239         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3240         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3241         if (num_devices == 1)
3242                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3243         else if (num_devices > 1)
3244                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3245         if (num_devices > 2)
3246                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3247         if (num_devices > 3)
3248                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3249                             BTRFS_BLOCK_GROUP_RAID6);
3250         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3251             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3252              (bctl->data.target & ~allowed))) {
3253                 btrfs_err(fs_info, "unable to start balance with target "
3254                            "data profile %llu",
3255                        bctl->data.target);
3256                 ret = -EINVAL;
3257                 goto out;
3258         }
3259         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3260             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3261              (bctl->meta.target & ~allowed))) {
3262                 btrfs_err(fs_info,
3263                            "unable to start balance with target metadata profile %llu",
3264                        bctl->meta.target);
3265                 ret = -EINVAL;
3266                 goto out;
3267         }
3268         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3269             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3270              (bctl->sys.target & ~allowed))) {
3271                 btrfs_err(fs_info,
3272                            "unable to start balance with target system profile %llu",
3273                        bctl->sys.target);
3274                 ret = -EINVAL;
3275                 goto out;
3276         }
3277
3278         /* allow dup'ed data chunks only in mixed mode */
3279         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3280             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3281                 btrfs_err(fs_info, "dup for data is not allowed");
3282                 ret = -EINVAL;
3283                 goto out;
3284         }
3285
3286         /* allow to reduce meta or sys integrity only if force set */
3287         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3288                         BTRFS_BLOCK_GROUP_RAID10 |
3289                         BTRFS_BLOCK_GROUP_RAID5 |
3290                         BTRFS_BLOCK_GROUP_RAID6;
3291         do {
3292                 seq = read_seqbegin(&fs_info->profiles_lock);
3293
3294                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3295                      (fs_info->avail_system_alloc_bits & allowed) &&
3296                      !(bctl->sys.target & allowed)) ||
3297                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3298                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3299                      !(bctl->meta.target & allowed))) {
3300                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3301                                 btrfs_info(fs_info, "force reducing metadata integrity");
3302                         } else {
3303                                 btrfs_err(fs_info, "balance will reduce metadata "
3304                                            "integrity, use force if you want this");
3305                                 ret = -EINVAL;
3306                                 goto out;
3307                         }
3308                 }
3309         } while (read_seqretry(&fs_info->profiles_lock, seq));
3310
3311         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3312                 int num_tolerated_disk_barrier_failures;
3313                 u64 target = bctl->sys.target;
3314
3315                 num_tolerated_disk_barrier_failures =
3316                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3317                 if (num_tolerated_disk_barrier_failures > 0 &&
3318                     (target &
3319                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3320                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3321                         num_tolerated_disk_barrier_failures = 0;
3322                 else if (num_tolerated_disk_barrier_failures > 1 &&
3323                          (target &
3324                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3325                         num_tolerated_disk_barrier_failures = 1;
3326
3327                 fs_info->num_tolerated_disk_barrier_failures =
3328                         num_tolerated_disk_barrier_failures;
3329         }
3330
3331         ret = insert_balance_item(fs_info->tree_root, bctl);
3332         if (ret && ret != -EEXIST)
3333                 goto out;
3334
3335         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3336                 BUG_ON(ret == -EEXIST);
3337                 set_balance_control(bctl);
3338         } else {
3339                 BUG_ON(ret != -EEXIST);
3340                 spin_lock(&fs_info->balance_lock);
3341                 update_balance_args(bctl);
3342                 spin_unlock(&fs_info->balance_lock);
3343         }
3344
3345         atomic_inc(&fs_info->balance_running);
3346         mutex_unlock(&fs_info->balance_mutex);
3347
3348         ret = __btrfs_balance(fs_info);
3349
3350         mutex_lock(&fs_info->balance_mutex);
3351         atomic_dec(&fs_info->balance_running);
3352
3353         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3354                 fs_info->num_tolerated_disk_barrier_failures =
3355                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3356         }
3357
3358         if (bargs) {
3359                 memset(bargs, 0, sizeof(*bargs));
3360                 update_ioctl_balance_args(fs_info, 0, bargs);
3361         }
3362
3363         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3364             balance_need_close(fs_info)) {
3365                 __cancel_balance(fs_info);
3366         }
3367
3368         wake_up(&fs_info->balance_wait_q);
3369
3370         return ret;
3371 out:
3372         if (bctl->flags & BTRFS_BALANCE_RESUME)
3373                 __cancel_balance(fs_info);
3374         else {
3375                 kfree(bctl);
3376                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3377         }
3378         return ret;
3379 }
3380
3381 static int balance_kthread(void *data)
3382 {
3383         struct btrfs_fs_info *fs_info = data;
3384         int ret = 0;
3385
3386         mutex_lock(&fs_info->volume_mutex);
3387         mutex_lock(&fs_info->balance_mutex);
3388
3389         if (fs_info->balance_ctl) {
3390                 btrfs_info(fs_info, "continuing balance");
3391                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3392         }
3393
3394         mutex_unlock(&fs_info->balance_mutex);
3395         mutex_unlock(&fs_info->volume_mutex);
3396
3397         return ret;
3398 }
3399
3400 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3401 {
3402         struct task_struct *tsk;
3403
3404         spin_lock(&fs_info->balance_lock);
3405         if (!fs_info->balance_ctl) {
3406                 spin_unlock(&fs_info->balance_lock);
3407                 return 0;
3408         }
3409         spin_unlock(&fs_info->balance_lock);
3410
3411         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3412                 btrfs_info(fs_info, "force skipping balance");
3413                 return 0;
3414         }
3415
3416         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3417         return PTR_ERR_OR_ZERO(tsk);
3418 }
3419
3420 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3421 {
3422         struct btrfs_balance_control *bctl;
3423         struct btrfs_balance_item *item;
3424         struct btrfs_disk_balance_args disk_bargs;
3425         struct btrfs_path *path;
3426         struct extent_buffer *leaf;
3427         struct btrfs_key key;
3428         int ret;
3429
3430         path = btrfs_alloc_path();
3431         if (!path)
3432                 return -ENOMEM;
3433
3434         key.objectid = BTRFS_BALANCE_OBJECTID;
3435         key.type = BTRFS_BALANCE_ITEM_KEY;
3436         key.offset = 0;
3437
3438         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3439         if (ret < 0)
3440                 goto out;
3441         if (ret > 0) { /* ret = -ENOENT; */
3442                 ret = 0;
3443                 goto out;
3444         }
3445
3446         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3447         if (!bctl) {
3448                 ret = -ENOMEM;
3449                 goto out;
3450         }
3451
3452         leaf = path->nodes[0];
3453         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3454
3455         bctl->fs_info = fs_info;
3456         bctl->flags = btrfs_balance_flags(leaf, item);
3457         bctl->flags |= BTRFS_BALANCE_RESUME;
3458
3459         btrfs_balance_data(leaf, item, &disk_bargs);
3460         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3461         btrfs_balance_meta(leaf, item, &disk_bargs);
3462         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3463         btrfs_balance_sys(leaf, item, &disk_bargs);
3464         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3465
3466         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3467
3468         mutex_lock(&fs_info->volume_mutex);
3469         mutex_lock(&fs_info->balance_mutex);
3470
3471         set_balance_control(bctl);
3472
3473         mutex_unlock(&fs_info->balance_mutex);
3474         mutex_unlock(&fs_info->volume_mutex);
3475 out:
3476         btrfs_free_path(path);
3477         return ret;
3478 }
3479
3480 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3481 {
3482         int ret = 0;
3483
3484         mutex_lock(&fs_info->balance_mutex);
3485         if (!fs_info->balance_ctl) {
3486                 mutex_unlock(&fs_info->balance_mutex);
3487                 return -ENOTCONN;
3488         }
3489
3490         if (atomic_read(&fs_info->balance_running)) {
3491                 atomic_inc(&fs_info->balance_pause_req);
3492                 mutex_unlock(&fs_info->balance_mutex);
3493
3494                 wait_event(fs_info->balance_wait_q,
3495                            atomic_read(&fs_info->balance_running) == 0);
3496
3497                 mutex_lock(&fs_info->balance_mutex);
3498                 /* we are good with balance_ctl ripped off from under us */
3499                 BUG_ON(atomic_read(&fs_info->balance_running));
3500                 atomic_dec(&fs_info->balance_pause_req);
3501         } else {
3502                 ret = -ENOTCONN;
3503         }
3504
3505         mutex_unlock(&fs_info->balance_mutex);
3506         return ret;
3507 }
3508
3509 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3510 {
3511         if (fs_info->sb->s_flags & MS_RDONLY)
3512                 return -EROFS;
3513
3514         mutex_lock(&fs_info->balance_mutex);
3515         if (!fs_info->balance_ctl) {
3516                 mutex_unlock(&fs_info->balance_mutex);
3517                 return -ENOTCONN;
3518         }
3519
3520         atomic_inc(&fs_info->balance_cancel_req);
3521         /*
3522          * if we are running just wait and return, balance item is
3523          * deleted in btrfs_balance in this case
3524          */
3525         if (atomic_read(&fs_info->balance_running)) {
3526                 mutex_unlock(&fs_info->balance_mutex);
3527                 wait_event(fs_info->balance_wait_q,
3528                            atomic_read(&fs_info->balance_running) == 0);
3529                 mutex_lock(&fs_info->balance_mutex);
3530         } else {
3531                 /* __cancel_balance needs volume_mutex */
3532                 mutex_unlock(&fs_info->balance_mutex);
3533                 mutex_lock(&fs_info->volume_mutex);
3534                 mutex_lock(&fs_info->balance_mutex);
3535
3536                 if (fs_info->balance_ctl)
3537                         __cancel_balance(fs_info);
3538
3539                 mutex_unlock(&fs_info->volume_mutex);
3540         }
3541
3542         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3543         atomic_dec(&fs_info->balance_cancel_req);
3544         mutex_unlock(&fs_info->balance_mutex);
3545         return 0;
3546 }
3547
3548 static int btrfs_uuid_scan_kthread(void *data)
3549 {
3550         struct btrfs_fs_info *fs_info = data;
3551         struct btrfs_root *root = fs_info->tree_root;
3552         struct btrfs_key key;
3553         struct btrfs_key max_key;
3554         struct btrfs_path *path = NULL;
3555         int ret = 0;
3556         struct extent_buffer *eb;
3557         int slot;
3558         struct btrfs_root_item root_item;
3559         u32 item_size;
3560         struct btrfs_trans_handle *trans = NULL;
3561
3562         path = btrfs_alloc_path();
3563         if (!path) {
3564                 ret = -ENOMEM;
3565                 goto out;
3566         }
3567
3568         key.objectid = 0;
3569         key.type = BTRFS_ROOT_ITEM_KEY;
3570         key.offset = 0;
3571
3572         max_key.objectid = (u64)-1;
3573         max_key.type = BTRFS_ROOT_ITEM_KEY;
3574         max_key.offset = (u64)-1;
3575
3576         path->keep_locks = 1;
3577
3578         while (1) {
3579                 ret = btrfs_search_forward(root, &key, path, 0);
3580                 if (ret) {
3581                         if (ret > 0)
3582                                 ret = 0;
3583                         break;
3584                 }
3585
3586                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3587                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3588                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3589                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3590                         goto skip;
3591
3592                 eb = path->nodes[0];
3593                 slot = path->slots[0];
3594                 item_size = btrfs_item_size_nr(eb, slot);
3595                 if (item_size < sizeof(root_item))
3596                         goto skip;
3597
3598                 read_extent_buffer(eb, &root_item,
3599                                    btrfs_item_ptr_offset(eb, slot),
3600                                    (int)sizeof(root_item));
3601                 if (btrfs_root_refs(&root_item) == 0)
3602                         goto skip;
3603
3604                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3605                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3606                         if (trans)
3607                                 goto update_tree;
3608
3609                         btrfs_release_path(path);
3610                         /*
3611                          * 1 - subvol uuid item
3612                          * 1 - received_subvol uuid item
3613                          */
3614                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3615                         if (IS_ERR(trans)) {
3616                                 ret = PTR_ERR(trans);
3617                                 break;
3618                         }
3619                         continue;
3620                 } else {
3621                         goto skip;
3622                 }
3623 update_tree:
3624                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3625                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3626                                                   root_item.uuid,
3627                                                   BTRFS_UUID_KEY_SUBVOL,
3628                                                   key.objectid);
3629                         if (ret < 0) {
3630                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3631                                         ret);
3632                                 break;
3633                         }
3634                 }
3635
3636                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3637                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3638                                                   root_item.received_uuid,
3639                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3640                                                   key.objectid);
3641                         if (ret < 0) {
3642                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3643                                         ret);
3644                                 break;
3645                         }
3646                 }
3647
3648 skip:
3649                 if (trans) {
3650                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3651                         trans = NULL;
3652                         if (ret)
3653                                 break;
3654                 }
3655
3656                 btrfs_release_path(path);
3657                 if (key.offset < (u64)-1) {
3658                         key.offset++;
3659                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3660                         key.offset = 0;
3661                         key.type = BTRFS_ROOT_ITEM_KEY;
3662                 } else if (key.objectid < (u64)-1) {
3663                         key.offset = 0;
3664                         key.type = BTRFS_ROOT_ITEM_KEY;
3665                         key.objectid++;
3666                 } else {
3667                         break;
3668                 }
3669                 cond_resched();
3670         }
3671
3672 out:
3673         btrfs_free_path(path);
3674         if (trans && !IS_ERR(trans))
3675                 btrfs_end_transaction(trans, fs_info->uuid_root);
3676         if (ret)
3677                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3678         else
3679                 fs_info->update_uuid_tree_gen = 1;
3680         up(&fs_info->uuid_tree_rescan_sem);
3681         return 0;
3682 }
3683
3684 /*
3685  * Callback for btrfs_uuid_tree_iterate().
3686  * returns:
3687  * 0    check succeeded, the entry is not outdated.
3688  * < 0  if an error occured.
3689  * > 0  if the check failed, which means the caller shall remove the entry.
3690  */
3691 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3692                                        u8 *uuid, u8 type, u64 subid)
3693 {
3694         struct btrfs_key key;
3695         int ret = 0;
3696         struct btrfs_root *subvol_root;
3697
3698         if (type != BTRFS_UUID_KEY_SUBVOL &&
3699             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3700                 goto out;
3701
3702         key.objectid = subid;
3703         key.type = BTRFS_ROOT_ITEM_KEY;
3704         key.offset = (u64)-1;
3705         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3706         if (IS_ERR(subvol_root)) {
3707                 ret = PTR_ERR(subvol_root);
3708                 if (ret == -ENOENT)
3709                         ret = 1;
3710                 goto out;
3711         }
3712
3713         switch (type) {
3714         case BTRFS_UUID_KEY_SUBVOL:
3715                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3716                         ret = 1;
3717                 break;
3718         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3719                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3720                            BTRFS_UUID_SIZE))
3721                         ret = 1;
3722                 break;
3723         }
3724
3725 out:
3726         return ret;
3727 }
3728
3729 static int btrfs_uuid_rescan_kthread(void *data)
3730 {
3731         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3732         int ret;
3733
3734         /*
3735          * 1st step is to iterate through the existing UUID tree and
3736          * to delete all entries that contain outdated data.
3737          * 2nd step is to add all missing entries to the UUID tree.
3738          */
3739         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3740         if (ret < 0) {
3741                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3742                 up(&fs_info->uuid_tree_rescan_sem);
3743                 return ret;
3744         }
3745         return btrfs_uuid_scan_kthread(data);
3746 }
3747
3748 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3749 {
3750         struct btrfs_trans_handle *trans;
3751         struct btrfs_root *tree_root = fs_info->tree_root;
3752         struct btrfs_root *uuid_root;
3753         struct task_struct *task;
3754         int ret;
3755
3756         /*
3757          * 1 - root node
3758          * 1 - root item
3759          */
3760         trans = btrfs_start_transaction(tree_root, 2);
3761         if (IS_ERR(trans))
3762                 return PTR_ERR(trans);
3763
3764         uuid_root = btrfs_create_tree(trans, fs_info,
3765                                       BTRFS_UUID_TREE_OBJECTID);
3766         if (IS_ERR(uuid_root)) {
3767                 btrfs_abort_transaction(trans, tree_root,
3768                                         PTR_ERR(uuid_root));
3769                 return PTR_ERR(uuid_root);
3770         }
3771
3772         fs_info->uuid_root = uuid_root;
3773
3774         ret = btrfs_commit_transaction(trans, tree_root);
3775         if (ret)
3776                 return ret;
3777
3778         down(&fs_info->uuid_tree_rescan_sem);
3779         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3780         if (IS_ERR(task)) {
3781                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3782                 btrfs_warn(fs_info, "failed to start uuid_scan task");
3783                 up(&fs_info->uuid_tree_rescan_sem);
3784                 return PTR_ERR(task);
3785         }
3786
3787         return 0;
3788 }
3789
3790 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3791 {
3792         struct task_struct *task;
3793
3794         down(&fs_info->uuid_tree_rescan_sem);
3795         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3796         if (IS_ERR(task)) {
3797                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3798                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3799                 up(&fs_info->uuid_tree_rescan_sem);
3800                 return PTR_ERR(task);
3801         }
3802
3803         return 0;
3804 }
3805
3806 /*
3807  * shrinking a device means finding all of the device extents past
3808  * the new size, and then following the back refs to the chunks.
3809  * The chunk relocation code actually frees the device extent
3810  */
3811 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3812 {
3813         struct btrfs_trans_handle *trans;
3814         struct btrfs_root *root = device->dev_root;
3815         struct btrfs_dev_extent *dev_extent = NULL;
3816         struct btrfs_path *path;
3817         u64 length;
3818         u64 chunk_tree;
3819         u64 chunk_objectid;
3820         u64 chunk_offset;
3821         int ret;
3822         int slot;
3823         int failed = 0;
3824         bool retried = false;
3825         struct extent_buffer *l;
3826         struct btrfs_key key;
3827         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3828         u64 old_total = btrfs_super_total_bytes(super_copy);
3829         u64 old_size = device->total_bytes;
3830         u64 diff = device->total_bytes - new_size;
3831
3832         if (device->is_tgtdev_for_dev_replace)
3833                 return -EINVAL;
3834
3835         path = btrfs_alloc_path();
3836         if (!path)
3837                 return -ENOMEM;
3838
3839         path->reada = 2;
3840
3841         lock_chunks(root);
3842
3843         device->total_bytes = new_size;
3844         if (device->writeable) {
3845                 device->fs_devices->total_rw_bytes -= diff;
3846                 spin_lock(&root->fs_info->free_chunk_lock);
3847                 root->fs_info->free_chunk_space -= diff;
3848                 spin_unlock(&root->fs_info->free_chunk_lock);
3849         }
3850         unlock_chunks(root);
3851
3852 again:
3853         key.objectid = device->devid;
3854         key.offset = (u64)-1;
3855         key.type = BTRFS_DEV_EXTENT_KEY;
3856
3857         do {
3858                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3859                 if (ret < 0)
3860                         goto done;
3861
3862                 ret = btrfs_previous_item(root, path, 0, key.type);
3863                 if (ret < 0)
3864                         goto done;
3865                 if (ret) {
3866                         ret = 0;
3867                         btrfs_release_path(path);
3868                         break;
3869                 }
3870
3871                 l = path->nodes[0];
3872                 slot = path->slots[0];
3873                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3874
3875                 if (key.objectid != device->devid) {
3876                         btrfs_release_path(path);
3877                         break;
3878                 }
3879
3880                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3881                 length = btrfs_dev_extent_length(l, dev_extent);
3882
3883                 if (key.offset + length <= new_size) {
3884                         btrfs_release_path(path);
3885                         break;
3886                 }
3887
3888                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3889                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3890                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3891                 btrfs_release_path(path);
3892
3893                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3894                                            chunk_offset);
3895                 if (ret && ret != -ENOSPC)
3896                         goto done;
3897                 if (ret == -ENOSPC)
3898                         failed++;
3899         } while (key.offset-- > 0);
3900
3901         if (failed && !retried) {
3902                 failed = 0;
3903                 retried = true;
3904                 goto again;
3905         } else if (failed && retried) {
3906                 ret = -ENOSPC;
3907                 lock_chunks(root);
3908
3909                 device->total_bytes = old_size;
3910                 if (device->writeable)
3911                         device->fs_devices->total_rw_bytes += diff;
3912                 spin_lock(&root->fs_info->free_chunk_lock);
3913                 root->fs_info->free_chunk_space += diff;
3914                 spin_unlock(&root->fs_info->free_chunk_lock);
3915                 unlock_chunks(root);
3916                 goto done;
3917         }
3918
3919         /* Shrinking succeeded, else we would be at "done". */
3920         trans = btrfs_start_transaction(root, 0);
3921         if (IS_ERR(trans)) {
3922                 ret = PTR_ERR(trans);
3923                 goto done;
3924         }
3925
3926         lock_chunks(root);
3927
3928         device->disk_total_bytes = new_size;
3929         /* Now btrfs_update_device() will change the on-disk size. */
3930         ret = btrfs_update_device(trans, device);
3931         if (ret) {
3932                 unlock_chunks(root);
3933                 btrfs_end_transaction(trans, root);
3934                 goto done;
3935         }
3936         WARN_ON(diff > old_total);
3937         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3938         unlock_chunks(root);
3939         btrfs_end_transaction(trans, root);
3940 done:
3941         btrfs_free_path(path);
3942         return ret;
3943 }
3944
3945 static int btrfs_add_system_chunk(struct btrfs_root *root,
3946                            struct btrfs_key *key,
3947                            struct btrfs_chunk *chunk, int item_size)
3948 {
3949         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3950         struct btrfs_disk_key disk_key;
3951         u32 array_size;
3952         u8 *ptr;
3953
3954         array_size = btrfs_super_sys_array_size(super_copy);
3955         if (array_size + item_size + sizeof(disk_key)
3956                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3957                 return -EFBIG;
3958
3959         ptr = super_copy->sys_chunk_array + array_size;
3960         btrfs_cpu_key_to_disk(&disk_key, key);
3961         memcpy(ptr, &disk_key, sizeof(disk_key));
3962         ptr += sizeof(disk_key);
3963         memcpy(ptr, chunk, item_size);
3964         item_size += sizeof(disk_key);
3965         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3966         return 0;
3967 }
3968
3969 /*
3970  * sort the devices in descending order by max_avail, total_avail
3971  */
3972 static int btrfs_cmp_device_info(const void *a, const void *b)
3973 {
3974         const struct btrfs_device_info *di_a = a;
3975         const struct btrfs_device_info *di_b = b;
3976
3977         if (di_a->max_avail > di_b->max_avail)
3978                 return -1;
3979         if (di_a->max_avail < di_b->max_avail)
3980                 return 1;
3981         if (di_a->total_avail > di_b->total_avail)
3982                 return -1;
3983         if (di_a->total_avail < di_b->total_avail)
3984                 return 1;
3985         return 0;
3986 }
3987
3988 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3989         [BTRFS_RAID_RAID10] = {
3990                 .sub_stripes    = 2,
3991                 .dev_stripes    = 1,
3992                 .devs_max       = 0,    /* 0 == as many as possible */
3993                 .devs_min       = 4,
3994                 .devs_increment = 2,
3995                 .ncopies        = 2,
3996         },
3997         [BTRFS_RAID_RAID1] = {
3998                 .sub_stripes    = 1,
3999                 .dev_stripes    = 1,
4000                 .devs_max       = 2,
4001                 .devs_min       = 2,
4002                 .devs_increment = 2,
4003                 .ncopies        = 2,
4004         },
4005         [BTRFS_RAID_DUP] = {
4006                 .sub_stripes    = 1,
4007                 .dev_stripes    = 2,
4008                 .devs_max       = 1,
4009                 .devs_min       = 1,
4010                 .devs_increment = 1,
4011                 .ncopies        = 2,
4012         },
4013         [BTRFS_RAID_RAID0] = {
4014                 .sub_stripes    = 1,
4015                 .dev_stripes    = 1,
4016                 .devs_max       = 0,
4017                 .devs_min       = 2,
4018                 .devs_increment = 1,
4019                 .ncopies        = 1,
4020         },
4021         [BTRFS_RAID_SINGLE] = {
4022                 .sub_stripes    = 1,
4023                 .dev_stripes    = 1,
4024                 .devs_max       = 1,
4025                 .devs_min       = 1,
4026                 .devs_increment = 1,
4027                 .ncopies        = 1,
4028         },
4029         [BTRFS_RAID_RAID5] = {
4030                 .sub_stripes    = 1,
4031                 .dev_stripes    = 1,
4032                 .devs_max       = 0,
4033                 .devs_min       = 2,
4034                 .devs_increment = 1,
4035                 .ncopies        = 2,
4036         },
4037         [BTRFS_RAID_RAID6] = {
4038                 .sub_stripes    = 1,
4039                 .dev_stripes    = 1,
4040                 .devs_max       = 0,
4041                 .devs_min       = 3,
4042                 .devs_increment = 1,
4043                 .ncopies        = 3,
4044         },
4045 };
4046
4047 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4048 {
4049         /* TODO allow them to set a preferred stripe size */
4050         return 64 * 1024;
4051 }
4052
4053 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4054 {
4055         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4056                 return;
4057
4058         btrfs_set_fs_incompat(info, RAID56);
4059 }
4060
4061 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4062                         - sizeof(struct btrfs_item)             \
4063                         - sizeof(struct btrfs_chunk))           \
4064                         / sizeof(struct btrfs_stripe) + 1)
4065
4066 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4067                                 - 2 * sizeof(struct btrfs_disk_key)     \
4068                                 - 2 * sizeof(struct btrfs_chunk))       \
4069                                 / sizeof(struct btrfs_stripe) + 1)
4070
4071 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4072                                struct btrfs_root *extent_root, u64 start,
4073                                u64 type)
4074 {
4075         struct btrfs_fs_info *info = extent_root->fs_info;
4076         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4077         struct list_head *cur;
4078         struct map_lookup *map = NULL;
4079         struct extent_map_tree *em_tree;
4080         struct extent_map *em;
4081         struct btrfs_device_info *devices_info = NULL;
4082         u64 total_avail;
4083         int num_stripes;        /* total number of stripes to allocate */
4084         int data_stripes;       /* number of stripes that count for
4085                                    block group size */
4086         int sub_stripes;        /* sub_stripes info for map */
4087         int dev_stripes;        /* stripes per dev */
4088         int devs_max;           /* max devs to use */
4089         int devs_min;           /* min devs needed */
4090         int devs_increment;     /* ndevs has to be a multiple of this */
4091         int ncopies;            /* how many copies to data has */
4092         int ret;
4093         u64 max_stripe_size;
4094         u64 max_chunk_size;
4095         u64 stripe_size;
4096         u64 num_bytes;
4097         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4098         int ndevs;
4099         int i;
4100         int j;
4101         int index;
4102
4103         BUG_ON(!alloc_profile_is_valid(type, 0));
4104
4105         if (list_empty(&fs_devices->alloc_list))
4106                 return -ENOSPC;
4107
4108         index = __get_raid_index(type);
4109
4110         sub_stripes = btrfs_raid_array[index].sub_stripes;
4111         dev_stripes = btrfs_raid_array[index].dev_stripes;
4112         devs_max = btrfs_raid_array[index].devs_max;
4113         devs_min = btrfs_raid_array[index].devs_min;
4114         devs_increment = btrfs_raid_array[index].devs_increment;
4115         ncopies = btrfs_raid_array[index].ncopies;
4116
4117         if (type & BTRFS_BLOCK_GROUP_DATA) {
4118                 max_stripe_size = 1024 * 1024 * 1024;
4119                 max_chunk_size = 10 * max_stripe_size;
4120                 if (!devs_max)
4121                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4122         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4123                 /* for larger filesystems, use larger metadata chunks */
4124                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4125                         max_stripe_size = 1024 * 1024 * 1024;
4126                 else
4127                         max_stripe_size = 256 * 1024 * 1024;
4128                 max_chunk_size = max_stripe_size;
4129                 if (!devs_max)
4130                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4131         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4132                 max_stripe_size = 32 * 1024 * 1024;
4133                 max_chunk_size = 2 * max_stripe_size;
4134                 if (!devs_max)
4135                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4136         } else {
4137                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4138                        type);
4139                 BUG_ON(1);
4140         }
4141
4142         /* we don't want a chunk larger than 10% of writeable space */
4143         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4144                              max_chunk_size);
4145
4146         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4147                                GFP_NOFS);
4148         if (!devices_info)
4149                 return -ENOMEM;
4150
4151         cur = fs_devices->alloc_list.next;
4152
4153         /*
4154          * in the first pass through the devices list, we gather information
4155          * about the available holes on each device.
4156          */
4157         ndevs = 0;
4158         while (cur != &fs_devices->alloc_list) {
4159                 struct btrfs_device *device;
4160                 u64 max_avail;
4161                 u64 dev_offset;
4162
4163                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4164
4165                 cur = cur->next;
4166
4167                 if (!device->writeable) {
4168                         WARN(1, KERN_ERR
4169                                "BTRFS: read-only device in alloc_list\n");
4170                         continue;
4171                 }
4172
4173                 if (!device->in_fs_metadata ||
4174                     device->is_tgtdev_for_dev_replace)
4175                         continue;
4176
4177                 if (device->total_bytes > device->bytes_used)
4178                         total_avail = device->total_bytes - device->bytes_used;
4179                 else
4180                         total_avail = 0;
4181
4182                 /* If there is no space on this device, skip it. */
4183                 if (total_avail == 0)
4184                         continue;
4185
4186                 ret = find_free_dev_extent(trans, device,
4187                                            max_stripe_size * dev_stripes,
4188                                            &dev_offset, &max_avail);
4189                 if (ret && ret != -ENOSPC)
4190                         goto error;
4191
4192                 if (ret == 0)
4193                         max_avail = max_stripe_size * dev_stripes;
4194
4195                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4196                         continue;
4197
4198                 if (ndevs == fs_devices->rw_devices) {
4199                         WARN(1, "%s: found more than %llu devices\n",
4200                              __func__, fs_devices->rw_devices);
4201                         break;
4202                 }
4203                 devices_info[ndevs].dev_offset = dev_offset;
4204                 devices_info[ndevs].max_avail = max_avail;
4205                 devices_info[ndevs].total_avail = total_avail;
4206                 devices_info[ndevs].dev = device;
4207                 ++ndevs;
4208         }
4209
4210         /*
4211          * now sort the devices by hole size / available space
4212          */
4213         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4214              btrfs_cmp_device_info, NULL);
4215
4216         /* round down to number of usable stripes */
4217         ndevs -= ndevs % devs_increment;
4218
4219         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4220                 ret = -ENOSPC;
4221                 goto error;
4222         }
4223
4224         if (devs_max && ndevs > devs_max)
4225                 ndevs = devs_max;
4226         /*
4227          * the primary goal is to maximize the number of stripes, so use as many
4228          * devices as possible, even if the stripes are not maximum sized.
4229          */
4230         stripe_size = devices_info[ndevs-1].max_avail;
4231         num_stripes = ndevs * dev_stripes;
4232
4233         /*
4234          * this will have to be fixed for RAID1 and RAID10 over
4235          * more drives
4236          */
4237         data_stripes = num_stripes / ncopies;
4238
4239         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4240                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4241                                  btrfs_super_stripesize(info->super_copy));
4242                 data_stripes = num_stripes - 1;
4243         }
4244         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4245                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4246                                  btrfs_super_stripesize(info->super_copy));
4247                 data_stripes = num_stripes - 2;
4248         }
4249
4250         /*
4251          * Use the number of data stripes to figure out how big this chunk
4252          * is really going to be in terms of logical address space,
4253          * and compare that answer with the max chunk size
4254          */
4255         if (stripe_size * data_stripes > max_chunk_size) {
4256                 u64 mask = (1ULL << 24) - 1;
4257                 stripe_size = max_chunk_size;
4258                 do_div(stripe_size, data_stripes);
4259
4260                 /* bump the answer up to a 16MB boundary */
4261                 stripe_size = (stripe_size + mask) & ~mask;
4262
4263                 /* but don't go higher than the limits we found
4264                  * while searching for free extents
4265                  */
4266                 if (stripe_size > devices_info[ndevs-1].max_avail)
4267                         stripe_size = devices_info[ndevs-1].max_avail;
4268         }
4269
4270         do_div(stripe_size, dev_stripes);
4271
4272         /* align to BTRFS_STRIPE_LEN */
4273         do_div(stripe_size, raid_stripe_len);
4274         stripe_size *= raid_stripe_len;
4275
4276         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4277         if (!map) {
4278                 ret = -ENOMEM;
4279                 goto error;
4280         }
4281         map->num_stripes = num_stripes;
4282
4283         for (i = 0; i < ndevs; ++i) {
4284                 for (j = 0; j < dev_stripes; ++j) {
4285                         int s = i * dev_stripes + j;
4286                         map->stripes[s].dev = devices_info[i].dev;
4287                         map->stripes[s].physical = devices_info[i].dev_offset +
4288                                                    j * stripe_size;
4289                 }
4290         }
4291         map->sector_size = extent_root->sectorsize;
4292         map->stripe_len = raid_stripe_len;
4293         map->io_align = raid_stripe_len;
4294         map->io_width = raid_stripe_len;
4295         map->type = type;
4296         map->sub_stripes = sub_stripes;
4297
4298         num_bytes = stripe_size * data_stripes;
4299
4300         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4301
4302         em = alloc_extent_map();
4303         if (!em) {
4304                 ret = -ENOMEM;
4305                 goto error;
4306         }
4307         em->bdev = (struct block_device *)map;
4308         em->start = start;
4309         em->len = num_bytes;
4310         em->block_start = 0;
4311         em->block_len = em->len;
4312         em->orig_block_len = stripe_size;
4313
4314         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4315         write_lock(&em_tree->lock);
4316         ret = add_extent_mapping(em_tree, em, 0);
4317         if (!ret) {
4318                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4319                 atomic_inc(&em->refs);
4320         }
4321         write_unlock(&em_tree->lock);
4322         if (ret) {
4323                 free_extent_map(em);
4324                 goto error;
4325         }
4326
4327         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4328                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4329                                      start, num_bytes);
4330         if (ret)
4331                 goto error_del_extent;
4332
4333         free_extent_map(em);
4334         check_raid56_incompat_flag(extent_root->fs_info, type);
4335
4336         kfree(devices_info);
4337         return 0;
4338
4339 error_del_extent:
4340         write_lock(&em_tree->lock);
4341         remove_extent_mapping(em_tree, em);
4342         write_unlock(&em_tree->lock);
4343
4344         /* One for our allocation */
4345         free_extent_map(em);
4346         /* One for the tree reference */
4347         free_extent_map(em);
4348 error:
4349         kfree(map);
4350         kfree(devices_info);
4351         return ret;
4352 }
4353
4354 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4355                                 struct btrfs_root *extent_root,
4356                                 u64 chunk_offset, u64 chunk_size)
4357 {
4358         struct btrfs_key key;
4359         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4360         struct btrfs_device *device;
4361         struct btrfs_chunk *chunk;
4362         struct btrfs_stripe *stripe;
4363         struct extent_map_tree *em_tree;
4364         struct extent_map *em;
4365         struct map_lookup *map;
4366         size_t item_size;
4367         u64 dev_offset;
4368         u64 stripe_size;
4369         int i = 0;
4370         int ret;
4371
4372         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4373         read_lock(&em_tree->lock);
4374         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4375         read_unlock(&em_tree->lock);
4376
4377         if (!em) {
4378                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4379                            "%Lu len %Lu", chunk_offset, chunk_size);
4380                 return -EINVAL;
4381         }
4382
4383         if (em->start != chunk_offset || em->len != chunk_size) {
4384                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4385                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4386                           chunk_size, em->start, em->len);
4387                 free_extent_map(em);
4388                 return -EINVAL;
4389         }
4390
4391         map = (struct map_lookup *)em->bdev;
4392         item_size = btrfs_chunk_item_size(map->num_stripes);
4393         stripe_size = em->orig_block_len;
4394
4395         chunk = kzalloc(item_size, GFP_NOFS);
4396         if (!chunk) {
4397                 ret = -ENOMEM;
4398                 goto out;
4399         }
4400
4401         for (i = 0; i < map->num_stripes; i++) {
4402                 device = map->stripes[i].dev;
4403                 dev_offset = map->stripes[i].physical;
4404
4405                 device->bytes_used += stripe_size;
4406                 ret = btrfs_update_device(trans, device);
4407                 if (ret)
4408                         goto out;
4409                 ret = btrfs_alloc_dev_extent(trans, device,
4410                                              chunk_root->root_key.objectid,
4411                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4412                                              chunk_offset, dev_offset,
4413                                              stripe_size);
4414                 if (ret)
4415                         goto out;
4416         }
4417
4418         spin_lock(&extent_root->fs_info->free_chunk_lock);
4419         extent_root->fs_info->free_chunk_space -= (stripe_size *
4420                                                    map->num_stripes);
4421         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4422
4423         stripe = &chunk->stripe;
4424         for (i = 0; i < map->num_stripes; i++) {
4425                 device = map->stripes[i].dev;
4426                 dev_offset = map->stripes[i].physical;
4427
4428                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4429                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4430                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4431                 stripe++;
4432         }
4433
4434         btrfs_set_stack_chunk_length(chunk, chunk_size);
4435         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4436         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4437         btrfs_set_stack_chunk_type(chunk, map->type);
4438         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4439         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4440         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4441         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4442         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4443
4444         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4445         key.type = BTRFS_CHUNK_ITEM_KEY;
4446         key.offset = chunk_offset;
4447
4448         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4449         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4450                 /*
4451                  * TODO: Cleanup of inserted chunk root in case of
4452                  * failure.
4453                  */
4454                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4455                                              item_size);
4456         }
4457
4458 out:
4459         kfree(chunk);
4460         free_extent_map(em);
4461         return ret;
4462 }
4463
4464 /*
4465  * Chunk allocation falls into two parts. The first part does works
4466  * that make the new allocated chunk useable, but not do any operation
4467  * that modifies the chunk tree. The second part does the works that
4468  * require modifying the chunk tree. This division is important for the
4469  * bootstrap process of adding storage to a seed btrfs.
4470  */
4471 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4472                       struct btrfs_root *extent_root, u64 type)
4473 {
4474         u64 chunk_offset;
4475
4476         chunk_offset = find_next_chunk(extent_root->fs_info);
4477         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4478 }
4479
4480 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4481                                          struct btrfs_root *root,
4482                                          struct btrfs_device *device)
4483 {
4484         u64 chunk_offset;
4485         u64 sys_chunk_offset;
4486         u64 alloc_profile;
4487         struct btrfs_fs_info *fs_info = root->fs_info;
4488         struct btrfs_root *extent_root = fs_info->extent_root;
4489         int ret;
4490
4491         chunk_offset = find_next_chunk(fs_info);
4492         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4493         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4494                                   alloc_profile);
4495         if (ret)
4496                 return ret;
4497
4498         sys_chunk_offset = find_next_chunk(root->fs_info);
4499         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4500         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4501                                   alloc_profile);
4502         if (ret) {
4503                 btrfs_abort_transaction(trans, root, ret);
4504                 goto out;
4505         }
4506
4507         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4508         if (ret)
4509                 btrfs_abort_transaction(trans, root, ret);
4510 out:
4511         return ret;
4512 }
4513
4514 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4515 {
4516         struct extent_map *em;
4517         struct map_lookup *map;
4518         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4519         int readonly = 0;
4520         int i;
4521
4522         read_lock(&map_tree->map_tree.lock);
4523         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4524         read_unlock(&map_tree->map_tree.lock);
4525         if (!em)
4526                 return 1;
4527
4528         if (btrfs_test_opt(root, DEGRADED)) {
4529                 free_extent_map(em);
4530                 return 0;
4531         }
4532
4533         map = (struct map_lookup *)em->bdev;
4534         for (i = 0; i < map->num_stripes; i++) {
4535                 if (!map->stripes[i].dev->writeable) {
4536                         readonly = 1;
4537                         break;
4538                 }
4539         }
4540         free_extent_map(em);
4541         return readonly;
4542 }
4543
4544 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4545 {
4546         extent_map_tree_init(&tree->map_tree);
4547 }
4548
4549 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4550 {
4551         struct extent_map *em;
4552
4553         while (1) {
4554                 write_lock(&tree->map_tree.lock);
4555                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4556                 if (em)
4557                         remove_extent_mapping(&tree->map_tree, em);
4558                 write_unlock(&tree->map_tree.lock);
4559                 if (!em)
4560                         break;
4561                 kfree(em->bdev);
4562                 /* once for us */
4563                 free_extent_map(em);
4564                 /* once for the tree */
4565                 free_extent_map(em);
4566         }
4567 }
4568
4569 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4570 {
4571         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4572         struct extent_map *em;
4573         struct map_lookup *map;
4574         struct extent_map_tree *em_tree = &map_tree->map_tree;
4575         int ret;
4576
4577         read_lock(&em_tree->lock);
4578         em = lookup_extent_mapping(em_tree, logical, len);
4579         read_unlock(&em_tree->lock);
4580
4581         /*
4582          * We could return errors for these cases, but that could get ugly and
4583          * we'd probably do the same thing which is just not do anything else
4584          * and exit, so return 1 so the callers don't try to use other copies.
4585          */
4586         if (!em) {
4587                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4588                             logical+len);
4589                 return 1;
4590         }
4591
4592         if (em->start > logical || em->start + em->len < logical) {
4593                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4594                             "%Lu-%Lu", logical, logical+len, em->start,
4595                             em->start + em->len);
4596                 free_extent_map(em);
4597                 return 1;
4598         }
4599
4600         map = (struct map_lookup *)em->bdev;
4601         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4602                 ret = map->num_stripes;
4603         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4604                 ret = map->sub_stripes;
4605         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4606                 ret = 2;
4607         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4608                 ret = 3;
4609         else
4610                 ret = 1;
4611         free_extent_map(em);
4612
4613         btrfs_dev_replace_lock(&fs_info->dev_replace);
4614         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4615                 ret++;
4616         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4617
4618         return ret;
4619 }
4620
4621 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4622                                     struct btrfs_mapping_tree *map_tree,
4623                                     u64 logical)
4624 {
4625         struct extent_map *em;
4626         struct map_lookup *map;
4627         struct extent_map_tree *em_tree = &map_tree->map_tree;
4628         unsigned long len = root->sectorsize;
4629
4630         read_lock(&em_tree->lock);
4631         em = lookup_extent_mapping(em_tree, logical, len);
4632         read_unlock(&em_tree->lock);
4633         BUG_ON(!em);
4634
4635         BUG_ON(em->start > logical || em->start + em->len < logical);
4636         map = (struct map_lookup *)em->bdev;
4637         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4638                          BTRFS_BLOCK_GROUP_RAID6)) {
4639                 len = map->stripe_len * nr_data_stripes(map);
4640         }
4641         free_extent_map(em);
4642         return len;
4643 }
4644
4645 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4646                            u64 logical, u64 len, int mirror_num)
4647 {
4648         struct extent_map *em;
4649         struct map_lookup *map;
4650         struct extent_map_tree *em_tree = &map_tree->map_tree;
4651         int ret = 0;
4652
4653         read_lock(&em_tree->lock);
4654         em = lookup_extent_mapping(em_tree, logical, len);
4655         read_unlock(&em_tree->lock);
4656         BUG_ON(!em);
4657
4658         BUG_ON(em->start > logical || em->start + em->len < logical);
4659         map = (struct map_lookup *)em->bdev;
4660         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4661                          BTRFS_BLOCK_GROUP_RAID6))
4662                 ret = 1;
4663         free_extent_map(em);
4664         return ret;
4665 }
4666
4667 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4668                             struct map_lookup *map, int first, int num,
4669                             int optimal, int dev_replace_is_ongoing)
4670 {
4671         int i;
4672         int tolerance;
4673         struct btrfs_device *srcdev;
4674
4675         if (dev_replace_is_ongoing &&
4676             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4677              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4678                 srcdev = fs_info->dev_replace.srcdev;
4679         else
4680                 srcdev = NULL;
4681
4682         /*
4683          * try to avoid the drive that is the source drive for a
4684          * dev-replace procedure, only choose it if no other non-missing
4685          * mirror is available
4686          */
4687         for (tolerance = 0; tolerance < 2; tolerance++) {
4688                 if (map->stripes[optimal].dev->bdev &&
4689                     (tolerance || map->stripes[optimal].dev != srcdev))
4690                         return optimal;
4691                 for (i = first; i < first + num; i++) {
4692                         if (map->stripes[i].dev->bdev &&
4693                             (tolerance || map->stripes[i].dev != srcdev))
4694                                 return i;
4695                 }
4696         }
4697
4698         /* we couldn't find one that doesn't fail.  Just return something
4699          * and the io error handling code will clean up eventually
4700          */
4701         return optimal;
4702 }
4703
4704 static inline int parity_smaller(u64 a, u64 b)
4705 {
4706         return a > b;
4707 }
4708
4709 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4710 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4711 {
4712         struct btrfs_bio_stripe s;
4713         int i;
4714         u64 l;
4715         int again = 1;
4716
4717         while (again) {
4718                 again = 0;
4719                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4720                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4721                                 s = bbio->stripes[i];
4722                                 l = raid_map[i];
4723                                 bbio->stripes[i] = bbio->stripes[i+1];
4724                                 raid_map[i] = raid_map[i+1];
4725                                 bbio->stripes[i+1] = s;
4726                                 raid_map[i+1] = l;
4727                                 again = 1;
4728                         }
4729                 }
4730         }
4731 }
4732
4733 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4734                              u64 logical, u64 *length,
4735                              struct btrfs_bio **bbio_ret,
4736                              int mirror_num, u64 **raid_map_ret)
4737 {
4738         struct extent_map *em;
4739         struct map_lookup *map;
4740         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4741         struct extent_map_tree *em_tree = &map_tree->map_tree;
4742         u64 offset;
4743         u64 stripe_offset;
4744         u64 stripe_end_offset;
4745         u64 stripe_nr;
4746         u64 stripe_nr_orig;
4747         u64 stripe_nr_end;
4748         u64 stripe_len;
4749         u64 *raid_map = NULL;
4750         int stripe_index;
4751         int i;
4752         int ret = 0;
4753         int num_stripes;
4754         int max_errors = 0;
4755         struct btrfs_bio *bbio = NULL;
4756         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4757         int dev_replace_is_ongoing = 0;
4758         int num_alloc_stripes;
4759         int patch_the_first_stripe_for_dev_replace = 0;
4760         u64 physical_to_patch_in_first_stripe = 0;
4761         u64 raid56_full_stripe_start = (u64)-1;
4762
4763         read_lock(&em_tree->lock);
4764         em = lookup_extent_mapping(em_tree, logical, *length);
4765         read_unlock(&em_tree->lock);
4766
4767         if (!em) {
4768                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4769                         logical, *length);
4770                 return -EINVAL;
4771         }
4772
4773         if (em->start > logical || em->start + em->len < logical) {
4774                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4775                            "found %Lu-%Lu", logical, em->start,
4776                            em->start + em->len);
4777                 free_extent_map(em);
4778                 return -EINVAL;
4779         }
4780
4781         map = (struct map_lookup *)em->bdev;
4782         offset = logical - em->start;
4783
4784         stripe_len = map->stripe_len;
4785         stripe_nr = offset;
4786         /*
4787          * stripe_nr counts the total number of stripes we have to stride
4788          * to get to this block
4789          */
4790         do_div(stripe_nr, stripe_len);
4791
4792         stripe_offset = stripe_nr * stripe_len;
4793         BUG_ON(offset < stripe_offset);
4794
4795         /* stripe_offset is the offset of this block in its stripe*/
4796         stripe_offset = offset - stripe_offset;
4797
4798         /* if we're here for raid56, we need to know the stripe aligned start */
4799         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4800                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4801                 raid56_full_stripe_start = offset;
4802
4803                 /* allow a write of a full stripe, but make sure we don't
4804                  * allow straddling of stripes
4805                  */
4806                 do_div(raid56_full_stripe_start, full_stripe_len);
4807                 raid56_full_stripe_start *= full_stripe_len;
4808         }
4809
4810         if (rw & REQ_DISCARD) {
4811                 /* we don't discard raid56 yet */
4812                 if (map->type &
4813                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4814                         ret = -EOPNOTSUPP;
4815                         goto out;
4816                 }
4817                 *length = min_t(u64, em->len - offset, *length);
4818         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4819                 u64 max_len;
4820                 /* For writes to RAID[56], allow a full stripeset across all disks.
4821                    For other RAID types and for RAID[56] reads, just allow a single
4822                    stripe (on a single disk). */
4823                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4824                     (rw & REQ_WRITE)) {
4825                         max_len = stripe_len * nr_data_stripes(map) -
4826                                 (offset - raid56_full_stripe_start);
4827                 } else {
4828                         /* we limit the length of each bio to what fits in a stripe */
4829                         max_len = stripe_len - stripe_offset;
4830                 }
4831                 *length = min_t(u64, em->len - offset, max_len);
4832         } else {
4833                 *length = em->len - offset;
4834         }
4835
4836         /* This is for when we're called from btrfs_merge_bio_hook() and all
4837            it cares about is the length */
4838         if (!bbio_ret)
4839                 goto out;
4840
4841         btrfs_dev_replace_lock(dev_replace);
4842         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4843         if (!dev_replace_is_ongoing)
4844                 btrfs_dev_replace_unlock(dev_replace);
4845
4846         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4847             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4848             dev_replace->tgtdev != NULL) {
4849                 /*
4850                  * in dev-replace case, for repair case (that's the only
4851                  * case where the mirror is selected explicitly when
4852                  * calling btrfs_map_block), blocks left of the left cursor
4853                  * can also be read from the target drive.
4854                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4855                  * the last one to the array of stripes. For READ, it also
4856                  * needs to be supported using the same mirror number.
4857                  * If the requested block is not left of the left cursor,
4858                  * EIO is returned. This can happen because btrfs_num_copies()
4859                  * returns one more in the dev-replace case.
4860                  */
4861                 u64 tmp_length = *length;
4862                 struct btrfs_bio *tmp_bbio = NULL;
4863                 int tmp_num_stripes;
4864                 u64 srcdev_devid = dev_replace->srcdev->devid;
4865                 int index_srcdev = 0;
4866                 int found = 0;
4867                 u64 physical_of_found = 0;
4868
4869                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4870                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4871                 if (ret) {
4872                         WARN_ON(tmp_bbio != NULL);
4873                         goto out;
4874                 }
4875
4876                 tmp_num_stripes = tmp_bbio->num_stripes;
4877                 if (mirror_num > tmp_num_stripes) {
4878                         /*
4879                          * REQ_GET_READ_MIRRORS does not contain this
4880                          * mirror, that means that the requested area
4881                          * is not left of the left cursor
4882                          */
4883                         ret = -EIO;
4884                         kfree(tmp_bbio);
4885                         goto out;
4886                 }
4887
4888                 /*
4889                  * process the rest of the function using the mirror_num
4890                  * of the source drive. Therefore look it up first.
4891                  * At the end, patch the device pointer to the one of the
4892                  * target drive.
4893                  */
4894                 for (i = 0; i < tmp_num_stripes; i++) {
4895                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4896                                 /*
4897                                  * In case of DUP, in order to keep it
4898                                  * simple, only add the mirror with the
4899                                  * lowest physical address
4900                                  */
4901                                 if (found &&
4902                                     physical_of_found <=
4903                                      tmp_bbio->stripes[i].physical)
4904                                         continue;
4905                                 index_srcdev = i;
4906                                 found = 1;
4907                                 physical_of_found =
4908                                         tmp_bbio->stripes[i].physical;
4909                         }
4910                 }
4911
4912                 if (found) {
4913                         mirror_num = index_srcdev + 1;
4914                         patch_the_first_stripe_for_dev_replace = 1;
4915                         physical_to_patch_in_first_stripe = physical_of_found;
4916                 } else {
4917                         WARN_ON(1);
4918                         ret = -EIO;
4919                         kfree(tmp_bbio);
4920                         goto out;
4921                 }
4922
4923                 kfree(tmp_bbio);
4924         } else if (mirror_num > map->num_stripes) {
4925                 mirror_num = 0;
4926         }
4927
4928         num_stripes = 1;
4929         stripe_index = 0;
4930         stripe_nr_orig = stripe_nr;
4931         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4932         do_div(stripe_nr_end, map->stripe_len);
4933         stripe_end_offset = stripe_nr_end * map->stripe_len -
4934                             (offset + *length);
4935
4936         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4937                 if (rw & REQ_DISCARD)
4938                         num_stripes = min_t(u64, map->num_stripes,
4939                                             stripe_nr_end - stripe_nr_orig);
4940                 stripe_index = do_div(stripe_nr, map->num_stripes);
4941         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4942                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4943                         num_stripes = map->num_stripes;
4944                 else if (mirror_num)
4945                         stripe_index = mirror_num - 1;
4946                 else {
4947                         stripe_index = find_live_mirror(fs_info, map, 0,
4948                                             map->num_stripes,
4949                                             current->pid % map->num_stripes,
4950                                             dev_replace_is_ongoing);
4951                         mirror_num = stripe_index + 1;
4952                 }
4953
4954         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4955                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4956                         num_stripes = map->num_stripes;
4957                 } else if (mirror_num) {
4958                         stripe_index = mirror_num - 1;
4959                 } else {
4960                         mirror_num = 1;
4961                 }
4962
4963         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4964                 int factor = map->num_stripes / map->sub_stripes;
4965
4966                 stripe_index = do_div(stripe_nr, factor);
4967                 stripe_index *= map->sub_stripes;
4968
4969                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4970                         num_stripes = map->sub_stripes;
4971                 else if (rw & REQ_DISCARD)
4972                         num_stripes = min_t(u64, map->sub_stripes *
4973                                             (stripe_nr_end - stripe_nr_orig),
4974                                             map->num_stripes);
4975                 else if (mirror_num)
4976                         stripe_index += mirror_num - 1;
4977                 else {
4978                         int old_stripe_index = stripe_index;
4979                         stripe_index = find_live_mirror(fs_info, map,
4980                                               stripe_index,
4981                                               map->sub_stripes, stripe_index +
4982                                               current->pid % map->sub_stripes,
4983                                               dev_replace_is_ongoing);
4984                         mirror_num = stripe_index - old_stripe_index + 1;
4985                 }
4986
4987         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4988                                 BTRFS_BLOCK_GROUP_RAID6)) {
4989                 u64 tmp;
4990
4991                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4992                     && raid_map_ret) {
4993                         int i, rot;
4994
4995                         /* push stripe_nr back to the start of the full stripe */
4996                         stripe_nr = raid56_full_stripe_start;
4997                         do_div(stripe_nr, stripe_len);
4998
4999                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5000
5001                         /* RAID[56] write or recovery. Return all stripes */
5002                         num_stripes = map->num_stripes;
5003                         max_errors = nr_parity_stripes(map);
5004
5005                         raid_map = kmalloc_array(num_stripes, sizeof(u64),
5006                                            GFP_NOFS);
5007                         if (!raid_map) {
5008                                 ret = -ENOMEM;
5009                                 goto out;
5010                         }
5011
5012                         /* Work out the disk rotation on this stripe-set */
5013                         tmp = stripe_nr;
5014                         rot = do_div(tmp, num_stripes);
5015
5016                         /* Fill in the logical address of each stripe */
5017                         tmp = stripe_nr * nr_data_stripes(map);
5018                         for (i = 0; i < nr_data_stripes(map); i++)
5019                                 raid_map[(i+rot) % num_stripes] =
5020                                         em->start + (tmp + i) * map->stripe_len;
5021
5022                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5023                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5024                                 raid_map[(i+rot+1) % num_stripes] =
5025                                         RAID6_Q_STRIPE;
5026
5027                         *length = map->stripe_len;
5028                         stripe_index = 0;
5029                         stripe_offset = 0;
5030                 } else {
5031                         /*
5032                          * Mirror #0 or #1 means the original data block.
5033                          * Mirror #2 is RAID5 parity block.
5034                          * Mirror #3 is RAID6 Q block.
5035                          */
5036                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5037                         if (mirror_num > 1)
5038                                 stripe_index = nr_data_stripes(map) +
5039                                                 mirror_num - 2;
5040
5041                         /* We distribute the parity blocks across stripes */
5042                         tmp = stripe_nr + stripe_index;
5043                         stripe_index = do_div(tmp, map->num_stripes);
5044                 }
5045         } else {
5046                 /*
5047                  * after this do_div call, stripe_nr is the number of stripes
5048                  * on this device we have to walk to find the data, and
5049                  * stripe_index is the number of our device in the stripe array
5050                  */
5051                 stripe_index = do_div(stripe_nr, map->num_stripes);
5052                 mirror_num = stripe_index + 1;
5053         }
5054         BUG_ON(stripe_index >= map->num_stripes);
5055
5056         num_alloc_stripes = num_stripes;
5057         if (dev_replace_is_ongoing) {
5058                 if (rw & (REQ_WRITE | REQ_DISCARD))
5059                         num_alloc_stripes <<= 1;
5060                 if (rw & REQ_GET_READ_MIRRORS)
5061                         num_alloc_stripes++;
5062         }
5063         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5064         if (!bbio) {
5065                 kfree(raid_map);
5066                 ret = -ENOMEM;
5067                 goto out;
5068         }
5069         atomic_set(&bbio->error, 0);
5070
5071         if (rw & REQ_DISCARD) {
5072                 int factor = 0;
5073                 int sub_stripes = 0;
5074                 u64 stripes_per_dev = 0;
5075                 u32 remaining_stripes = 0;
5076                 u32 last_stripe = 0;
5077
5078                 if (map->type &
5079                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5080                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5081                                 sub_stripes = 1;
5082                         else
5083                                 sub_stripes = map->sub_stripes;
5084
5085                         factor = map->num_stripes / sub_stripes;
5086                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5087                                                       stripe_nr_orig,
5088                                                       factor,
5089                                                       &remaining_stripes);
5090                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5091                         last_stripe *= sub_stripes;
5092                 }
5093
5094                 for (i = 0; i < num_stripes; i++) {
5095                         bbio->stripes[i].physical =
5096                                 map->stripes[stripe_index].physical +
5097                                 stripe_offset + stripe_nr * map->stripe_len;
5098                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5099
5100                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5101                                          BTRFS_BLOCK_GROUP_RAID10)) {
5102                                 bbio->stripes[i].length = stripes_per_dev *
5103                                                           map->stripe_len;
5104
5105                                 if (i / sub_stripes < remaining_stripes)
5106                                         bbio->stripes[i].length +=
5107                                                 map->stripe_len;
5108
5109                                 /*
5110                                  * Special for the first stripe and
5111                                  * the last stripe:
5112                                  *
5113                                  * |-------|...|-------|
5114                                  *     |----------|
5115                                  *    off     end_off
5116                                  */
5117                                 if (i < sub_stripes)
5118                                         bbio->stripes[i].length -=
5119                                                 stripe_offset;
5120
5121                                 if (stripe_index >= last_stripe &&
5122                                     stripe_index <= (last_stripe +
5123                                                      sub_stripes - 1))
5124                                         bbio->stripes[i].length -=
5125                                                 stripe_end_offset;
5126
5127                                 if (i == sub_stripes - 1)
5128                                         stripe_offset = 0;
5129                         } else
5130                                 bbio->stripes[i].length = *length;
5131
5132                         stripe_index++;
5133                         if (stripe_index == map->num_stripes) {
5134                                 /* This could only happen for RAID0/10 */
5135                                 stripe_index = 0;
5136                                 stripe_nr++;
5137                         }
5138                 }
5139         } else {
5140                 for (i = 0; i < num_stripes; i++) {
5141                         bbio->stripes[i].physical =
5142                                 map->stripes[stripe_index].physical +
5143                                 stripe_offset +
5144                                 stripe_nr * map->stripe_len;
5145                         bbio->stripes[i].dev =
5146                                 map->stripes[stripe_index].dev;
5147                         stripe_index++;
5148                 }
5149         }
5150
5151         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5152                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5153                                  BTRFS_BLOCK_GROUP_RAID10 |
5154                                  BTRFS_BLOCK_GROUP_RAID5 |
5155                                  BTRFS_BLOCK_GROUP_DUP)) {
5156                         max_errors = 1;
5157                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5158                         max_errors = 2;
5159                 }
5160         }
5161
5162         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5163             dev_replace->tgtdev != NULL) {
5164                 int index_where_to_add;
5165                 u64 srcdev_devid = dev_replace->srcdev->devid;
5166
5167                 /*
5168                  * duplicate the write operations while the dev replace
5169                  * procedure is running. Since the copying of the old disk
5170                  * to the new disk takes place at run time while the
5171                  * filesystem is mounted writable, the regular write
5172                  * operations to the old disk have to be duplicated to go
5173                  * to the new disk as well.
5174                  * Note that device->missing is handled by the caller, and
5175                  * that the write to the old disk is already set up in the
5176                  * stripes array.
5177                  */
5178                 index_where_to_add = num_stripes;
5179                 for (i = 0; i < num_stripes; i++) {
5180                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5181                                 /* write to new disk, too */
5182                                 struct btrfs_bio_stripe *new =
5183                                         bbio->stripes + index_where_to_add;
5184                                 struct btrfs_bio_stripe *old =
5185                                         bbio->stripes + i;
5186
5187                                 new->physical = old->physical;
5188                                 new->length = old->length;
5189                                 new->dev = dev_replace->tgtdev;
5190                                 index_where_to_add++;
5191                                 max_errors++;
5192                         }
5193                 }
5194                 num_stripes = index_where_to_add;
5195         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5196                    dev_replace->tgtdev != NULL) {
5197                 u64 srcdev_devid = dev_replace->srcdev->devid;
5198                 int index_srcdev = 0;
5199                 int found = 0;
5200                 u64 physical_of_found = 0;
5201
5202                 /*
5203                  * During the dev-replace procedure, the target drive can
5204                  * also be used to read data in case it is needed to repair
5205                  * a corrupt block elsewhere. This is possible if the
5206                  * requested area is left of the left cursor. In this area,
5207                  * the target drive is a full copy of the source drive.
5208                  */
5209                 for (i = 0; i < num_stripes; i++) {
5210                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5211                                 /*
5212                                  * In case of DUP, in order to keep it
5213                                  * simple, only add the mirror with the
5214                                  * lowest physical address
5215                                  */
5216                                 if (found &&
5217                                     physical_of_found <=
5218                                      bbio->stripes[i].physical)
5219                                         continue;
5220                                 index_srcdev = i;
5221                                 found = 1;
5222                                 physical_of_found = bbio->stripes[i].physical;
5223                         }
5224                 }
5225                 if (found) {
5226                         u64 length = map->stripe_len;
5227
5228                         if (physical_of_found + length <=
5229                             dev_replace->cursor_left) {
5230                                 struct btrfs_bio_stripe *tgtdev_stripe =
5231                                         bbio->stripes + num_stripes;
5232
5233                                 tgtdev_stripe->physical = physical_of_found;
5234                                 tgtdev_stripe->length =
5235                                         bbio->stripes[index_srcdev].length;
5236                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5237
5238                                 num_stripes++;
5239                         }
5240                 }
5241         }
5242
5243         *bbio_ret = bbio;
5244         bbio->num_stripes = num_stripes;
5245         bbio->max_errors = max_errors;
5246         bbio->mirror_num = mirror_num;
5247
5248         /*
5249          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5250          * mirror_num == num_stripes + 1 && dev_replace target drive is
5251          * available as a mirror
5252          */
5253         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5254                 WARN_ON(num_stripes > 1);
5255                 bbio->stripes[0].dev = dev_replace->tgtdev;
5256                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5257                 bbio->mirror_num = map->num_stripes + 1;
5258         }
5259         if (raid_map) {
5260                 sort_parity_stripes(bbio, raid_map);
5261                 *raid_map_ret = raid_map;
5262         }
5263 out:
5264         if (dev_replace_is_ongoing)
5265                 btrfs_dev_replace_unlock(dev_replace);
5266         free_extent_map(em);
5267         return ret;
5268 }
5269
5270 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5271                       u64 logical, u64 *length,
5272                       struct btrfs_bio **bbio_ret, int mirror_num)
5273 {
5274         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5275                                  mirror_num, NULL);
5276 }
5277
5278 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5279                      u64 chunk_start, u64 physical, u64 devid,
5280                      u64 **logical, int *naddrs, int *stripe_len)
5281 {
5282         struct extent_map_tree *em_tree = &map_tree->map_tree;
5283         struct extent_map *em;
5284         struct map_lookup *map;
5285         u64 *buf;
5286         u64 bytenr;
5287         u64 length;
5288         u64 stripe_nr;
5289         u64 rmap_len;
5290         int i, j, nr = 0;
5291
5292         read_lock(&em_tree->lock);
5293         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5294         read_unlock(&em_tree->lock);
5295
5296         if (!em) {
5297                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5298                        chunk_start);
5299                 return -EIO;
5300         }
5301
5302         if (em->start != chunk_start) {
5303                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5304                        em->start, chunk_start);
5305                 free_extent_map(em);
5306                 return -EIO;
5307         }
5308         map = (struct map_lookup *)em->bdev;
5309
5310         length = em->len;
5311         rmap_len = map->stripe_len;
5312
5313         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5314                 do_div(length, map->num_stripes / map->sub_stripes);
5315         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5316                 do_div(length, map->num_stripes);
5317         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5318                               BTRFS_BLOCK_GROUP_RAID6)) {
5319                 do_div(length, nr_data_stripes(map));
5320                 rmap_len = map->stripe_len * nr_data_stripes(map);
5321         }
5322
5323         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5324         BUG_ON(!buf); /* -ENOMEM */
5325
5326         for (i = 0; i < map->num_stripes; i++) {
5327                 if (devid && map->stripes[i].dev->devid != devid)
5328                         continue;
5329                 if (map->stripes[i].physical > physical ||
5330                     map->stripes[i].physical + length <= physical)
5331                         continue;
5332
5333                 stripe_nr = physical - map->stripes[i].physical;
5334                 do_div(stripe_nr, map->stripe_len);
5335
5336                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5337                         stripe_nr = stripe_nr * map->num_stripes + i;
5338                         do_div(stripe_nr, map->sub_stripes);
5339                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5340                         stripe_nr = stripe_nr * map->num_stripes + i;
5341                 } /* else if RAID[56], multiply by nr_data_stripes().
5342                    * Alternatively, just use rmap_len below instead of
5343                    * map->stripe_len */
5344
5345                 bytenr = chunk_start + stripe_nr * rmap_len;
5346                 WARN_ON(nr >= map->num_stripes);
5347                 for (j = 0; j < nr; j++) {
5348                         if (buf[j] == bytenr)
5349                                 break;
5350                 }
5351                 if (j == nr) {
5352                         WARN_ON(nr >= map->num_stripes);
5353                         buf[nr++] = bytenr;
5354                 }
5355         }
5356
5357         *logical = buf;
5358         *naddrs = nr;
5359         *stripe_len = rmap_len;
5360
5361         free_extent_map(em);
5362         return 0;
5363 }
5364
5365 static void btrfs_end_bio(struct bio *bio, int err)
5366 {
5367         struct btrfs_bio *bbio = bio->bi_private;
5368         struct btrfs_device *dev = bbio->stripes[0].dev;
5369         int is_orig_bio = 0;
5370
5371         if (err) {
5372                 atomic_inc(&bbio->error);
5373                 if (err == -EIO || err == -EREMOTEIO) {
5374                         unsigned int stripe_index =
5375                                 btrfs_io_bio(bio)->stripe_index;
5376
5377                         BUG_ON(stripe_index >= bbio->num_stripes);
5378                         dev = bbio->stripes[stripe_index].dev;
5379                         if (dev->bdev) {
5380                                 if (bio->bi_rw & WRITE)
5381                                         btrfs_dev_stat_inc(dev,
5382                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5383                                 else
5384                                         btrfs_dev_stat_inc(dev,
5385                                                 BTRFS_DEV_STAT_READ_ERRS);
5386                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5387                                         btrfs_dev_stat_inc(dev,
5388                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5389                                 btrfs_dev_stat_print_on_error(dev);
5390                         }
5391                 }
5392         }
5393
5394         if (bio == bbio->orig_bio)
5395                 is_orig_bio = 1;
5396
5397         btrfs_bio_counter_dec(bbio->fs_info);
5398
5399         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5400                 if (!is_orig_bio) {
5401                         bio_put(bio);
5402                         bio = bbio->orig_bio;
5403                 }
5404
5405                 /*
5406                  * We have original bio now. So increment bi_remaining to
5407                  * account for it in endio
5408                  */
5409                 atomic_inc(&bio->bi_remaining);
5410
5411                 bio->bi_private = bbio->private;
5412                 bio->bi_end_io = bbio->end_io;
5413                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5414                 /* only send an error to the higher layers if it is
5415                  * beyond the tolerance of the btrfs bio
5416                  */
5417                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5418                         err = -EIO;
5419                 } else {
5420                         /*
5421                          * this bio is actually up to date, we didn't
5422                          * go over the max number of errors
5423                          */
5424                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5425                         err = 0;
5426                 }
5427                 kfree(bbio);
5428
5429                 bio_endio(bio, err);
5430         } else if (!is_orig_bio) {
5431                 bio_put(bio);
5432         }
5433 }
5434
5435 /*
5436  * see run_scheduled_bios for a description of why bios are collected for
5437  * async submit.
5438  *
5439  * This will add one bio to the pending list for a device and make sure
5440  * the work struct is scheduled.
5441  */
5442 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5443                                         struct btrfs_device *device,
5444                                         int rw, struct bio *bio)
5445 {
5446         int should_queue = 1;
5447         struct btrfs_pending_bios *pending_bios;
5448
5449         if (device->missing || !device->bdev) {
5450                 bio_endio(bio, -EIO);
5451                 return;
5452         }
5453
5454         /* don't bother with additional async steps for reads, right now */
5455         if (!(rw & REQ_WRITE)) {
5456                 bio_get(bio);
5457                 btrfsic_submit_bio(rw, bio);
5458                 bio_put(bio);
5459                 return;
5460         }
5461
5462         /*
5463          * nr_async_bios allows us to reliably return congestion to the
5464          * higher layers.  Otherwise, the async bio makes it appear we have
5465          * made progress against dirty pages when we've really just put it
5466          * on a queue for later
5467          */
5468         atomic_inc(&root->fs_info->nr_async_bios);
5469         WARN_ON(bio->bi_next);
5470         bio->bi_next = NULL;
5471         bio->bi_rw |= rw;
5472
5473         spin_lock(&device->io_lock);
5474         if (bio->bi_rw & REQ_SYNC)
5475                 pending_bios = &device->pending_sync_bios;
5476         else
5477                 pending_bios = &device->pending_bios;
5478
5479         if (pending_bios->tail)
5480                 pending_bios->tail->bi_next = bio;
5481
5482         pending_bios->tail = bio;
5483         if (!pending_bios->head)
5484                 pending_bios->head = bio;
5485         if (device->running_pending)
5486                 should_queue = 0;
5487
5488         spin_unlock(&device->io_lock);
5489
5490         if (should_queue)
5491                 btrfs_queue_work(root->fs_info->submit_workers,
5492                                  &device->work);
5493 }
5494
5495 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5496                        sector_t sector)
5497 {
5498         struct bio_vec *prev;
5499         struct request_queue *q = bdev_get_queue(bdev);
5500         unsigned int max_sectors = queue_max_sectors(q);
5501         struct bvec_merge_data bvm = {
5502                 .bi_bdev = bdev,
5503                 .bi_sector = sector,
5504                 .bi_rw = bio->bi_rw,
5505         };
5506
5507         if (WARN_ON(bio->bi_vcnt == 0))
5508                 return 1;
5509
5510         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5511         if (bio_sectors(bio) > max_sectors)
5512                 return 0;
5513
5514         if (!q->merge_bvec_fn)
5515                 return 1;
5516
5517         bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5518         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5519                 return 0;
5520         return 1;
5521 }
5522
5523 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5524                               struct bio *bio, u64 physical, int dev_nr,
5525                               int rw, int async)
5526 {
5527         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5528
5529         bio->bi_private = bbio;
5530         btrfs_io_bio(bio)->stripe_index = dev_nr;
5531         bio->bi_end_io = btrfs_end_bio;
5532         bio->bi_iter.bi_sector = physical >> 9;
5533 #ifdef DEBUG
5534         {
5535                 struct rcu_string *name;
5536
5537                 rcu_read_lock();
5538                 name = rcu_dereference(dev->name);
5539                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5540                          "(%s id %llu), size=%u\n", rw,
5541                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5542                          name->str, dev->devid, bio->bi_size);
5543                 rcu_read_unlock();
5544         }
5545 #endif
5546         bio->bi_bdev = dev->bdev;
5547
5548         btrfs_bio_counter_inc_noblocked(root->fs_info);
5549
5550         if (async)
5551                 btrfs_schedule_bio(root, dev, rw, bio);
5552         else
5553                 btrfsic_submit_bio(rw, bio);
5554 }
5555
5556 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5557                               struct bio *first_bio, struct btrfs_device *dev,
5558                               int dev_nr, int rw, int async)
5559 {
5560         struct bio_vec *bvec = first_bio->bi_io_vec;
5561         struct bio *bio;
5562         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5563         u64 physical = bbio->stripes[dev_nr].physical;
5564
5565 again:
5566         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5567         if (!bio)
5568                 return -ENOMEM;
5569
5570         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5571                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5572                                  bvec->bv_offset) < bvec->bv_len) {
5573                         u64 len = bio->bi_iter.bi_size;
5574
5575                         atomic_inc(&bbio->stripes_pending);
5576                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5577                                           rw, async);
5578                         physical += len;
5579                         goto again;
5580                 }
5581                 bvec++;
5582         }
5583
5584         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5585         return 0;
5586 }
5587
5588 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5589 {
5590         atomic_inc(&bbio->error);
5591         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5592                 bio->bi_private = bbio->private;
5593                 bio->bi_end_io = bbio->end_io;
5594                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5595                 bio->bi_iter.bi_sector = logical >> 9;
5596                 kfree(bbio);
5597                 bio_endio(bio, -EIO);
5598         }
5599 }
5600
5601 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5602                   int mirror_num, int async_submit)
5603 {
5604         struct btrfs_device *dev;
5605         struct bio *first_bio = bio;
5606         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5607         u64 length = 0;
5608         u64 map_length;
5609         u64 *raid_map = NULL;
5610         int ret;
5611         int dev_nr = 0;
5612         int total_devs = 1;
5613         struct btrfs_bio *bbio = NULL;
5614
5615         length = bio->bi_iter.bi_size;
5616         map_length = length;
5617
5618         btrfs_bio_counter_inc_blocked(root->fs_info);
5619         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5620                               mirror_num, &raid_map);
5621         if (ret) {
5622                 btrfs_bio_counter_dec(root->fs_info);
5623                 return ret;
5624         }
5625
5626         total_devs = bbio->num_stripes;
5627         bbio->orig_bio = first_bio;
5628         bbio->private = first_bio->bi_private;
5629         bbio->end_io = first_bio->bi_end_io;
5630         bbio->fs_info = root->fs_info;
5631         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5632
5633         if (raid_map) {
5634                 /* In this case, map_length has been set to the length of
5635                    a single stripe; not the whole write */
5636                 if (rw & WRITE) {
5637                         ret = raid56_parity_write(root, bio, bbio,
5638                                                   raid_map, map_length);
5639                 } else {
5640                         ret = raid56_parity_recover(root, bio, bbio,
5641                                                     raid_map, map_length,
5642                                                     mirror_num);
5643                 }
5644                 /*
5645                  * FIXME, replace dosen't support raid56 yet, please fix
5646                  * it in the future.
5647                  */
5648                 btrfs_bio_counter_dec(root->fs_info);
5649                 return ret;
5650         }
5651
5652         if (map_length < length) {
5653                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5654                         logical, length, map_length);
5655                 BUG();
5656         }
5657
5658         while (dev_nr < total_devs) {
5659                 dev = bbio->stripes[dev_nr].dev;
5660                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5661                         bbio_error(bbio, first_bio, logical);
5662                         dev_nr++;
5663                         continue;
5664                 }
5665
5666                 /*
5667                  * Check and see if we're ok with this bio based on it's size
5668                  * and offset with the given device.
5669                  */
5670                 if (!bio_size_ok(dev->bdev, first_bio,
5671                                  bbio->stripes[dev_nr].physical >> 9)) {
5672                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5673                                                  dev_nr, rw, async_submit);
5674                         BUG_ON(ret);
5675                         dev_nr++;
5676                         continue;
5677                 }
5678
5679                 if (dev_nr < total_devs - 1) {
5680                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5681                         BUG_ON(!bio); /* -ENOMEM */
5682                 } else {
5683                         bio = first_bio;
5684                 }
5685
5686                 submit_stripe_bio(root, bbio, bio,
5687                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5688                                   async_submit);
5689                 dev_nr++;
5690         }
5691         btrfs_bio_counter_dec(root->fs_info);
5692         return 0;
5693 }
5694
5695 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5696                                        u8 *uuid, u8 *fsid)
5697 {
5698         struct btrfs_device *device;
5699         struct btrfs_fs_devices *cur_devices;
5700
5701         cur_devices = fs_info->fs_devices;
5702         while (cur_devices) {
5703                 if (!fsid ||
5704                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5705                         device = __find_device(&cur_devices->devices,
5706                                                devid, uuid);
5707                         if (device)
5708                                 return device;
5709                 }
5710                 cur_devices = cur_devices->seed;
5711         }
5712         return NULL;
5713 }
5714
5715 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5716                                             u64 devid, u8 *dev_uuid)
5717 {
5718         struct btrfs_device *device;
5719         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5720
5721         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5722         if (IS_ERR(device))
5723                 return NULL;
5724
5725         list_add(&device->dev_list, &fs_devices->devices);
5726         device->fs_devices = fs_devices;
5727         fs_devices->num_devices++;
5728
5729         device->missing = 1;
5730         fs_devices->missing_devices++;
5731
5732         return device;
5733 }
5734
5735 /**
5736  * btrfs_alloc_device - allocate struct btrfs_device
5737  * @fs_info:    used only for generating a new devid, can be NULL if
5738  *              devid is provided (i.e. @devid != NULL).
5739  * @devid:      a pointer to devid for this device.  If NULL a new devid
5740  *              is generated.
5741  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
5742  *              is generated.
5743  *
5744  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5745  * on error.  Returned struct is not linked onto any lists and can be
5746  * destroyed with kfree() right away.
5747  */
5748 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5749                                         const u64 *devid,
5750                                         const u8 *uuid)
5751 {
5752         struct btrfs_device *dev;
5753         u64 tmp;
5754
5755         if (WARN_ON(!devid && !fs_info))
5756                 return ERR_PTR(-EINVAL);
5757
5758         dev = __alloc_device();
5759         if (IS_ERR(dev))
5760                 return dev;
5761
5762         if (devid)
5763                 tmp = *devid;
5764         else {
5765                 int ret;
5766
5767                 ret = find_next_devid(fs_info, &tmp);
5768                 if (ret) {
5769                         kfree(dev);
5770                         return ERR_PTR(ret);
5771                 }
5772         }
5773         dev->devid = tmp;
5774
5775         if (uuid)
5776                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5777         else
5778                 generate_random_uuid(dev->uuid);
5779
5780         btrfs_init_work(&dev->work, pending_bios_fn, NULL, NULL);
5781
5782         return dev;
5783 }
5784
5785 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5786                           struct extent_buffer *leaf,
5787                           struct btrfs_chunk *chunk)
5788 {
5789         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5790         struct map_lookup *map;
5791         struct extent_map *em;
5792         u64 logical;
5793         u64 length;
5794         u64 devid;
5795         u8 uuid[BTRFS_UUID_SIZE];
5796         int num_stripes;
5797         int ret;
5798         int i;
5799
5800         logical = key->offset;
5801         length = btrfs_chunk_length(leaf, chunk);
5802
5803         read_lock(&map_tree->map_tree.lock);
5804         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5805         read_unlock(&map_tree->map_tree.lock);
5806
5807         /* already mapped? */
5808         if (em && em->start <= logical && em->start + em->len > logical) {
5809                 free_extent_map(em);
5810                 return 0;
5811         } else if (em) {
5812                 free_extent_map(em);
5813         }
5814
5815         em = alloc_extent_map();
5816         if (!em)
5817                 return -ENOMEM;
5818         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5819         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5820         if (!map) {
5821                 free_extent_map(em);
5822                 return -ENOMEM;
5823         }
5824
5825         em->bdev = (struct block_device *)map;
5826         em->start = logical;
5827         em->len = length;
5828         em->orig_start = 0;
5829         em->block_start = 0;
5830         em->block_len = em->len;
5831
5832         map->num_stripes = num_stripes;
5833         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5834         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5835         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5836         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5837         map->type = btrfs_chunk_type(leaf, chunk);
5838         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5839         for (i = 0; i < num_stripes; i++) {
5840                 map->stripes[i].physical =
5841                         btrfs_stripe_offset_nr(leaf, chunk, i);
5842                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5843                 read_extent_buffer(leaf, uuid, (unsigned long)
5844                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5845                                    BTRFS_UUID_SIZE);
5846                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5847                                                         uuid, NULL);
5848                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5849                         kfree(map);
5850                         free_extent_map(em);
5851                         return -EIO;
5852                 }
5853                 if (!map->stripes[i].dev) {
5854                         map->stripes[i].dev =
5855                                 add_missing_dev(root, devid, uuid);
5856                         if (!map->stripes[i].dev) {
5857                                 kfree(map);
5858                                 free_extent_map(em);
5859                                 return -EIO;
5860                         }
5861                 }
5862                 map->stripes[i].dev->in_fs_metadata = 1;
5863         }
5864
5865         write_lock(&map_tree->map_tree.lock);
5866         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5867         write_unlock(&map_tree->map_tree.lock);
5868         BUG_ON(ret); /* Tree corruption */
5869         free_extent_map(em);
5870
5871         return 0;
5872 }
5873
5874 static void fill_device_from_item(struct extent_buffer *leaf,
5875                                  struct btrfs_dev_item *dev_item,
5876                                  struct btrfs_device *device)
5877 {
5878         unsigned long ptr;
5879
5880         device->devid = btrfs_device_id(leaf, dev_item);
5881         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5882         device->total_bytes = device->disk_total_bytes;
5883         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5884         device->type = btrfs_device_type(leaf, dev_item);
5885         device->io_align = btrfs_device_io_align(leaf, dev_item);
5886         device->io_width = btrfs_device_io_width(leaf, dev_item);
5887         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5888         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5889         device->is_tgtdev_for_dev_replace = 0;
5890
5891         ptr = btrfs_device_uuid(dev_item);
5892         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5893 }
5894
5895 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5896 {
5897         struct btrfs_fs_devices *fs_devices;
5898         int ret;
5899
5900         BUG_ON(!mutex_is_locked(&uuid_mutex));
5901
5902         fs_devices = root->fs_info->fs_devices->seed;
5903         while (fs_devices) {
5904                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5905                         ret = 0;
5906                         goto out;
5907                 }
5908                 fs_devices = fs_devices->seed;
5909         }
5910
5911         fs_devices = find_fsid(fsid);
5912         if (!fs_devices) {
5913                 ret = -ENOENT;
5914                 goto out;
5915         }
5916
5917         fs_devices = clone_fs_devices(fs_devices);
5918         if (IS_ERR(fs_devices)) {
5919                 ret = PTR_ERR(fs_devices);
5920                 goto out;
5921         }
5922
5923         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5924                                    root->fs_info->bdev_holder);
5925         if (ret) {
5926                 free_fs_devices(fs_devices);
5927                 goto out;
5928         }
5929
5930         if (!fs_devices->seeding) {
5931                 __btrfs_close_devices(fs_devices);
5932                 free_fs_devices(fs_devices);
5933                 ret = -EINVAL;
5934                 goto out;
5935         }
5936
5937         fs_devices->seed = root->fs_info->fs_devices->seed;
5938         root->fs_info->fs_devices->seed = fs_devices;
5939 out:
5940         return ret;
5941 }
5942
5943 static int read_one_dev(struct btrfs_root *root,
5944                         struct extent_buffer *leaf,
5945                         struct btrfs_dev_item *dev_item)
5946 {
5947         struct btrfs_device *device;
5948         u64 devid;
5949         int ret;
5950         u8 fs_uuid[BTRFS_UUID_SIZE];
5951         u8 dev_uuid[BTRFS_UUID_SIZE];
5952
5953         devid = btrfs_device_id(leaf, dev_item);
5954         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5955                            BTRFS_UUID_SIZE);
5956         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5957                            BTRFS_UUID_SIZE);
5958
5959         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5960                 ret = open_seed_devices(root, fs_uuid);
5961                 if (ret && !btrfs_test_opt(root, DEGRADED))
5962                         return ret;
5963         }
5964
5965         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5966         if (!device || !device->bdev) {
5967                 if (!btrfs_test_opt(root, DEGRADED))
5968                         return -EIO;
5969
5970                 if (!device) {
5971                         btrfs_warn(root->fs_info, "devid %llu missing", devid);
5972                         device = add_missing_dev(root, devid, dev_uuid);
5973                         if (!device)
5974                                 return -ENOMEM;
5975                 } else if (!device->missing) {
5976                         /*
5977                          * this happens when a device that was properly setup
5978                          * in the device info lists suddenly goes bad.
5979                          * device->bdev is NULL, and so we have to set
5980                          * device->missing to one here
5981                          */
5982                         root->fs_info->fs_devices->missing_devices++;
5983                         device->missing = 1;
5984                 }
5985         }
5986
5987         if (device->fs_devices != root->fs_info->fs_devices) {
5988                 BUG_ON(device->writeable);
5989                 if (device->generation !=
5990                     btrfs_device_generation(leaf, dev_item))
5991                         return -EINVAL;
5992         }
5993
5994         fill_device_from_item(leaf, dev_item, device);
5995         device->in_fs_metadata = 1;
5996         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5997                 device->fs_devices->total_rw_bytes += device->total_bytes;
5998                 spin_lock(&root->fs_info->free_chunk_lock);
5999                 root->fs_info->free_chunk_space += device->total_bytes -
6000                         device->bytes_used;
6001                 spin_unlock(&root->fs_info->free_chunk_lock);
6002         }
6003         ret = 0;
6004         return ret;
6005 }
6006
6007 int btrfs_read_sys_array(struct btrfs_root *root)
6008 {
6009         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6010         struct extent_buffer *sb;
6011         struct btrfs_disk_key *disk_key;
6012         struct btrfs_chunk *chunk;
6013         u8 *ptr;
6014         unsigned long sb_ptr;
6015         int ret = 0;
6016         u32 num_stripes;
6017         u32 array_size;
6018         u32 len = 0;
6019         u32 cur;
6020         struct btrfs_key key;
6021
6022         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6023                                           BTRFS_SUPER_INFO_SIZE);
6024         if (!sb)
6025                 return -ENOMEM;
6026         btrfs_set_buffer_uptodate(sb);
6027         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6028         /*
6029          * The sb extent buffer is artifical and just used to read the system array.
6030          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6031          * pages up-to-date when the page is larger: extent does not cover the
6032          * whole page and consequently check_page_uptodate does not find all
6033          * the page's extents up-to-date (the hole beyond sb),
6034          * write_extent_buffer then triggers a WARN_ON.
6035          *
6036          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6037          * but sb spans only this function. Add an explicit SetPageUptodate call
6038          * to silence the warning eg. on PowerPC 64.
6039          */
6040         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6041                 SetPageUptodate(sb->pages[0]);
6042
6043         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6044         array_size = btrfs_super_sys_array_size(super_copy);
6045
6046         ptr = super_copy->sys_chunk_array;
6047         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6048         cur = 0;
6049
6050         while (cur < array_size) {
6051                 disk_key = (struct btrfs_disk_key *)ptr;
6052                 btrfs_disk_key_to_cpu(&key, disk_key);
6053
6054                 len = sizeof(*disk_key); ptr += len;
6055                 sb_ptr += len;
6056                 cur += len;
6057
6058                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6059                         chunk = (struct btrfs_chunk *)sb_ptr;
6060                         ret = read_one_chunk(root, &key, sb, chunk);
6061                         if (ret)
6062                                 break;
6063                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6064                         len = btrfs_chunk_item_size(num_stripes);
6065                 } else {
6066                         ret = -EIO;
6067                         break;
6068                 }
6069                 ptr += len;
6070                 sb_ptr += len;
6071                 cur += len;
6072         }
6073         free_extent_buffer(sb);
6074         return ret;
6075 }
6076
6077 int btrfs_read_chunk_tree(struct btrfs_root *root)
6078 {
6079         struct btrfs_path *path;
6080         struct extent_buffer *leaf;
6081         struct btrfs_key key;
6082         struct btrfs_key found_key;
6083         int ret;
6084         int slot;
6085
6086         root = root->fs_info->chunk_root;
6087
6088         path = btrfs_alloc_path();
6089         if (!path)
6090                 return -ENOMEM;
6091
6092         mutex_lock(&uuid_mutex);
6093         lock_chunks(root);
6094
6095         /*
6096          * Read all device items, and then all the chunk items. All
6097          * device items are found before any chunk item (their object id
6098          * is smaller than the lowest possible object id for a chunk
6099          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6100          */
6101         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6102         key.offset = 0;
6103         key.type = 0;
6104         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6105         if (ret < 0)
6106                 goto error;
6107         while (1) {
6108                 leaf = path->nodes[0];
6109                 slot = path->slots[0];
6110                 if (slot >= btrfs_header_nritems(leaf)) {
6111                         ret = btrfs_next_leaf(root, path);
6112                         if (ret == 0)
6113                                 continue;
6114                         if (ret < 0)
6115                                 goto error;
6116                         break;
6117                 }
6118                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6119                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6120                         struct btrfs_dev_item *dev_item;
6121                         dev_item = btrfs_item_ptr(leaf, slot,
6122                                                   struct btrfs_dev_item);
6123                         ret = read_one_dev(root, leaf, dev_item);
6124                         if (ret)
6125                                 goto error;
6126                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6127                         struct btrfs_chunk *chunk;
6128                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6129                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6130                         if (ret)
6131                                 goto error;
6132                 }
6133                 path->slots[0]++;
6134         }
6135         ret = 0;
6136 error:
6137         unlock_chunks(root);
6138         mutex_unlock(&uuid_mutex);
6139
6140         btrfs_free_path(path);
6141         return ret;
6142 }
6143
6144 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6145 {
6146         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6147         struct btrfs_device *device;
6148
6149         while (fs_devices) {
6150                 mutex_lock(&fs_devices->device_list_mutex);
6151                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6152                         device->dev_root = fs_info->dev_root;
6153                 mutex_unlock(&fs_devices->device_list_mutex);
6154
6155                 fs_devices = fs_devices->seed;
6156         }
6157 }
6158
6159 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6160 {
6161         int i;
6162
6163         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6164                 btrfs_dev_stat_reset(dev, i);
6165 }
6166
6167 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6168 {
6169         struct btrfs_key key;
6170         struct btrfs_key found_key;
6171         struct btrfs_root *dev_root = fs_info->dev_root;
6172         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6173         struct extent_buffer *eb;
6174         int slot;
6175         int ret = 0;
6176         struct btrfs_device *device;
6177         struct btrfs_path *path = NULL;
6178         int i;
6179
6180         path = btrfs_alloc_path();
6181         if (!path) {
6182                 ret = -ENOMEM;
6183                 goto out;
6184         }
6185
6186         mutex_lock(&fs_devices->device_list_mutex);
6187         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6188                 int item_size;
6189                 struct btrfs_dev_stats_item *ptr;
6190
6191                 key.objectid = 0;
6192                 key.type = BTRFS_DEV_STATS_KEY;
6193                 key.offset = device->devid;
6194                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6195                 if (ret) {
6196                         __btrfs_reset_dev_stats(device);
6197                         device->dev_stats_valid = 1;
6198                         btrfs_release_path(path);
6199                         continue;
6200                 }
6201                 slot = path->slots[0];
6202                 eb = path->nodes[0];
6203                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6204                 item_size = btrfs_item_size_nr(eb, slot);
6205
6206                 ptr = btrfs_item_ptr(eb, slot,
6207                                      struct btrfs_dev_stats_item);
6208
6209                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6210                         if (item_size >= (1 + i) * sizeof(__le64))
6211                                 btrfs_dev_stat_set(device, i,
6212                                         btrfs_dev_stats_value(eb, ptr, i));
6213                         else
6214                                 btrfs_dev_stat_reset(device, i);
6215                 }
6216
6217                 device->dev_stats_valid = 1;
6218                 btrfs_dev_stat_print_on_load(device);
6219                 btrfs_release_path(path);
6220         }
6221         mutex_unlock(&fs_devices->device_list_mutex);
6222
6223 out:
6224         btrfs_free_path(path);
6225         return ret < 0 ? ret : 0;
6226 }
6227
6228 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6229                                 struct btrfs_root *dev_root,
6230                                 struct btrfs_device *device)
6231 {
6232         struct btrfs_path *path;
6233         struct btrfs_key key;
6234         struct extent_buffer *eb;
6235         struct btrfs_dev_stats_item *ptr;
6236         int ret;
6237         int i;
6238
6239         key.objectid = 0;
6240         key.type = BTRFS_DEV_STATS_KEY;
6241         key.offset = device->devid;
6242
6243         path = btrfs_alloc_path();
6244         BUG_ON(!path);
6245         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6246         if (ret < 0) {
6247                 printk_in_rcu(KERN_WARNING "BTRFS: "
6248                         "error %d while searching for dev_stats item for device %s!\n",
6249                               ret, rcu_str_deref(device->name));
6250                 goto out;
6251         }
6252
6253         if (ret == 0 &&
6254             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6255                 /* need to delete old one and insert a new one */
6256                 ret = btrfs_del_item(trans, dev_root, path);
6257                 if (ret != 0) {
6258                         printk_in_rcu(KERN_WARNING "BTRFS: "
6259                                 "delete too small dev_stats item for device %s failed %d!\n",
6260                                       rcu_str_deref(device->name), ret);
6261                         goto out;
6262                 }
6263                 ret = 1;
6264         }
6265
6266         if (ret == 1) {
6267                 /* need to insert a new item */
6268                 btrfs_release_path(path);
6269                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6270                                               &key, sizeof(*ptr));
6271                 if (ret < 0) {
6272                         printk_in_rcu(KERN_WARNING "BTRFS: "
6273                                           "insert dev_stats item for device %s failed %d!\n",
6274                                       rcu_str_deref(device->name), ret);
6275                         goto out;
6276                 }
6277         }
6278
6279         eb = path->nodes[0];
6280         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6281         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6282                 btrfs_set_dev_stats_value(eb, ptr, i,
6283                                           btrfs_dev_stat_read(device, i));
6284         btrfs_mark_buffer_dirty(eb);
6285
6286 out:
6287         btrfs_free_path(path);
6288         return ret;
6289 }
6290
6291 /*
6292  * called from commit_transaction. Writes all changed device stats to disk.
6293  */
6294 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6295                         struct btrfs_fs_info *fs_info)
6296 {
6297         struct btrfs_root *dev_root = fs_info->dev_root;
6298         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6299         struct btrfs_device *device;
6300         int ret = 0;
6301
6302         mutex_lock(&fs_devices->device_list_mutex);
6303         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6304                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6305                         continue;
6306
6307                 ret = update_dev_stat_item(trans, dev_root, device);
6308                 if (!ret)
6309                         device->dev_stats_dirty = 0;
6310         }
6311         mutex_unlock(&fs_devices->device_list_mutex);
6312
6313         return ret;
6314 }
6315
6316 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6317 {
6318         btrfs_dev_stat_inc(dev, index);
6319         btrfs_dev_stat_print_on_error(dev);
6320 }
6321
6322 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6323 {
6324         if (!dev->dev_stats_valid)
6325                 return;
6326         printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6327                            "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6328                            rcu_str_deref(dev->name),
6329                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6330                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6331                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6332                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6333                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6334 }
6335
6336 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6337 {
6338         int i;
6339
6340         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6341                 if (btrfs_dev_stat_read(dev, i) != 0)
6342                         break;
6343         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6344                 return; /* all values == 0, suppress message */
6345
6346         printk_in_rcu(KERN_INFO "BTRFS: "
6347                    "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6348                rcu_str_deref(dev->name),
6349                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6350                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6351                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6352                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6353                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6354 }
6355
6356 int btrfs_get_dev_stats(struct btrfs_root *root,
6357                         struct btrfs_ioctl_get_dev_stats *stats)
6358 {
6359         struct btrfs_device *dev;
6360         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6361         int i;
6362
6363         mutex_lock(&fs_devices->device_list_mutex);
6364         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6365         mutex_unlock(&fs_devices->device_list_mutex);
6366
6367         if (!dev) {
6368                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6369                 return -ENODEV;
6370         } else if (!dev->dev_stats_valid) {
6371                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6372                 return -ENODEV;
6373         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6374                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6375                         if (stats->nr_items > i)
6376                                 stats->values[i] =
6377                                         btrfs_dev_stat_read_and_reset(dev, i);
6378                         else
6379                                 btrfs_dev_stat_reset(dev, i);
6380                 }
6381         } else {
6382                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6383                         if (stats->nr_items > i)
6384                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6385         }
6386         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6387                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6388         return 0;
6389 }
6390
6391 int btrfs_scratch_superblock(struct btrfs_device *device)
6392 {
6393         struct buffer_head *bh;
6394         struct btrfs_super_block *disk_super;
6395
6396         bh = btrfs_read_dev_super(device->bdev);
6397         if (!bh)
6398                 return -EINVAL;
6399         disk_super = (struct btrfs_super_block *)bh->b_data;
6400
6401         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6402         set_buffer_dirty(bh);
6403         sync_dirty_buffer(bh);
6404         brelse(bh);
6405
6406         return 0;
6407 }