fs/btrfs/dev-replace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) STRATO AG 2012.  All rights reserved.
   4  */
   5
   6 #include <linux/sched.h>
   7 #include <linux/bio.h>
   8 #include <linux/slab.h>
   9 #include <linux/blkdev.h>
  10 #include <linux/kthread.h>
  11 #include <linux/math64.h>
  12 #include "misc.h"
  13 #include "ctree.h"
  14 #include "extent_map.h"
  15 #include "disk-io.h"
  16 #include "transaction.h"
  17 #include "print-tree.h"
  18 #include "volumes.h"
  19 #include "async-thread.h"
  20 #include "check-integrity.h"
  21 #include "rcu-string.h"
  22 #include "dev-replace.h"
  23 #include "sysfs.h"
  24
  25 /*
  26  * Device replace overview
  27  *
  28  * [Objective]
  29  * To copy all extents (both new and on-disk) from source device to target
  30  * device, while still keeping the filesystem read-write.
  31  *
  32  * [Method]
  33  * There are two main methods involved:
  34  *
  35  * - Write duplication
  36  *
  37  *   All new writes will be written to both target and source devices, so even
  38  *   if replace gets canceled, sources device still contans up-to-date data.
  39  *
  40  *   Location:          handle_ops_on_dev_replace() from __btrfs_map_block()
  41  *   Start:             btrfs_dev_replace_start()
  42  *   End:               btrfs_dev_replace_finishing()
  43  *   Content:           Latest data/metadata
  44  *
  45  * - Copy existing extents
  46  *
  47  *   This happens by re-using scrub facility, as scrub also iterates through
  48  *   existing extents from commit root.
  49  *
  50  *   Location:          scrub_write_block_to_dev_replace() from
  51  *                      scrub_block_complete()
  52  *   Content:           Data/meta from commit root.
  53  *
  54  * Due to the content difference, we need to avoid nocow write when dev-replace
  55  * is happening.  This is done by marking the block group read-only and waiting
  56  * for NOCOW writes.
  57  *
  58  * After replace is done, the finishing part is done by swapping the target and
  59  * source devices.
  60  *
  61  *   Location:          btrfs_dev_replace_update_device_in_mapping_tree() from
  62  *                      btrfs_dev_replace_finishing()
  63  */
  64
  65 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
  66                                        int scrub_ret);
  67 static int btrfs_dev_replace_kthread(void *data);
  68
  69 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  70 {
  71         struct btrfs_key key;
  72         struct btrfs_root *dev_root = fs_info->dev_root;
  73         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  74         struct extent_buffer *eb;
  75         int slot;
  76         int ret = 0;
  77         struct btrfs_path *path = NULL;
  78         int item_size;
  79         struct btrfs_dev_replace_item *ptr;
  80         u64 src_devid;
  81
  82         path = btrfs_alloc_path();
  83         if (!path) {
  84                 ret = -ENOMEM;
  85                 goto out;
  86         }
  87
  88         key.objectid = 0;
  89         key.type = BTRFS_DEV_REPLACE_KEY;
  90         key.offset = 0;
  91         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  92         if (ret) {
  93 no_valid_dev_replace_entry_found:
  94                 ret = 0;
  95                 dev_replace->replace_state =
  96                         BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
  97                 dev_replace->cont_reading_from_srcdev_mode =
  98                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
  99                 dev_replace->time_started = 0;
 100                 dev_replace->time_stopped = 0;
 101                 atomic64_set(&dev_replace->num_write_errors, 0);
 102                 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 103                 dev_replace->cursor_left = 0;
 104                 dev_replace->committed_cursor_left = 0;
 105                 dev_replace->cursor_left_last_write_of_item = 0;
 106                 dev_replace->cursor_right = 0;
 107                 dev_replace->srcdev = NULL;
 108                 dev_replace->tgtdev = NULL;
 109                 dev_replace->is_valid = 0;
 110                 dev_replace->item_needs_writeback = 0;
 111                 goto out;
 112         }
 113         slot = path->slots[0];
 114         eb = path->nodes[0];
 115         item_size = btrfs_item_size_nr(eb, slot);
 116         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
 117
 118         if (item_size != sizeof(struct btrfs_dev_replace_item)) {
 119                 btrfs_warn(fs_info,
 120                         "dev_replace entry found has unexpected size, ignore entry");
 121                 goto no_valid_dev_replace_entry_found;
 122         }
 123
 124         src_devid = btrfs_dev_replace_src_devid(eb, ptr);
 125         dev_replace->cont_reading_from_srcdev_mode =
 126                 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
 127         dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
 128         dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
 129         dev_replace->time_stopped =
 130                 btrfs_dev_replace_time_stopped(eb, ptr);
 131         atomic64_set(&dev_replace->num_write_errors,
 132                      btrfs_dev_replace_num_write_errors(eb, ptr));
 133         atomic64_set(&dev_replace->num_uncorrectable_read_errors,
 134                      btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
 135         dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
 136         dev_replace->committed_cursor_left = dev_replace->cursor_left;
 137         dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
 138         dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
 139         dev_replace->is_valid = 1;
 140
 141         dev_replace->item_needs_writeback = 0;
 142         switch (dev_replace->replace_state) {
 143         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 144         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 145         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 146                 dev_replace->srcdev = NULL;
 147                 dev_replace->tgtdev = NULL;
 148                 break;
 149         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 150         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 151                 dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices,
 152                                                 src_devid, NULL, NULL, true);
 153                 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
 154                                                         BTRFS_DEV_REPLACE_DEVID,
 155                                                         NULL, NULL, true);
 156                 /*
 157                  * allow 'btrfs dev replace_cancel' if src/tgt device is
 158                  * missing
 159                  */
 160                 if (!dev_replace->srcdev &&
 161                     !btrfs_test_opt(fs_info, DEGRADED)) {
 162                         ret = -EIO;
 163                         btrfs_warn(fs_info,
 164                            "cannot mount because device replace operation is ongoing and");
 165                         btrfs_warn(fs_info,
 166                            "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 167                            src_devid);
 168                 }
 169                 if (!dev_replace->tgtdev &&
 170                     !btrfs_test_opt(fs_info, DEGRADED)) {
 171                         ret = -EIO;
 172                         btrfs_warn(fs_info,
 173                            "cannot mount because device replace operation is ongoing and");
 174                         btrfs_warn(fs_info,
 175                            "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 176                                 BTRFS_DEV_REPLACE_DEVID);
 177                 }
 178                 if (dev_replace->tgtdev) {
 179                         if (dev_replace->srcdev) {
 180                                 dev_replace->tgtdev->total_bytes =
 181                                         dev_replace->srcdev->total_bytes;
 182                                 dev_replace->tgtdev->disk_total_bytes =
 183                                         dev_replace->srcdev->disk_total_bytes;
 184                                 dev_replace->tgtdev->commit_total_bytes =
 185                                         dev_replace->srcdev->commit_total_bytes;
 186                                 dev_replace->tgtdev->bytes_used =
 187                                         dev_replace->srcdev->bytes_used;
 188                                 dev_replace->tgtdev->commit_bytes_used =
 189                                         dev_replace->srcdev->commit_bytes_used;
 190                         }
 191                         set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 192                                 &dev_replace->tgtdev->dev_state);
 193
 194                         WARN_ON(fs_info->fs_devices->rw_devices == 0);
 195                         dev_replace->tgtdev->io_width = fs_info->sectorsize;
 196                         dev_replace->tgtdev->io_align = fs_info->sectorsize;
 197                         dev_replace->tgtdev->sector_size = fs_info->sectorsize;
 198                         dev_replace->tgtdev->fs_info = fs_info;
 199                         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
 200                                 &dev_replace->tgtdev->dev_state);
 201                 }
 202                 break;
 203         }
 204
 205 out:
 206         btrfs_free_path(path);
 207         return ret;
 208 }
 209
 210 /*
 211  * Initialize a new device for device replace target from a given source dev
 212  * and path.
 213  *
 214  * Return 0 and new device in @device_out, otherwise return < 0
 215  */
 216 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
 217                                   const char *device_path,
 218                                   struct btrfs_device *srcdev,
 219                                   struct btrfs_device **device_out)
 220 {
 221         struct btrfs_device *device;
 222         struct block_device *bdev;
 223         struct rcu_string *name;
 224         u64 devid = BTRFS_DEV_REPLACE_DEVID;
 225         int ret = 0;
 226
 227         *device_out = NULL;
 228         if (srcdev->fs_devices->seeding) {
 229                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
 230                 return -EINVAL;
 231         }
 232
 233         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
 234                                   fs_info->bdev_holder);
 235         if (IS_ERR(bdev)) {
 236                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
 237                 return PTR_ERR(bdev);
 238         }
 239
 240         sync_blockdev(bdev);
 241
 242         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
 243                 if (device->bdev == bdev) {
 244                         btrfs_err(fs_info,
 245                                   "target device is in the filesystem!");
 246                         ret = -EEXIST;
 247                         goto error;
 248                 }
 249         }
 250
 251
 252         if (i_size_read(bdev->bd_inode) <
 253             btrfs_device_get_total_bytes(srcdev)) {
 254                 btrfs_err(fs_info,
 255                           "target device is smaller than source device!");
 256                 ret = -EINVAL;
 257                 goto error;
 258         }
 259
 260
 261         device = btrfs_alloc_device(NULL, &devid, NULL);
 262         if (IS_ERR(device)) {
 263                 ret = PTR_ERR(device);
 264                 goto error;
 265         }
 266
 267         name = rcu_string_strdup(device_path, GFP_KERNEL);
 268         if (!name) {
 269                 btrfs_free_device(device);
 270                 ret = -ENOMEM;
 271                 goto error;
 272         }
 273         rcu_assign_pointer(device->name, name);
 274
 275         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 276         device->generation = 0;
 277         device->io_width = fs_info->sectorsize;
 278         device->io_align = fs_info->sectorsize;
 279         device->sector_size = fs_info->sectorsize;
 280         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
 281         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
 282         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
 283         device->commit_total_bytes = srcdev->commit_total_bytes;
 284         device->commit_bytes_used = device->bytes_used;
 285         device->fs_info = fs_info;
 286         device->bdev = bdev;
 287         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
 288         set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
 289         device->mode = FMODE_EXCL;
 290         device->dev_stats_valid = 1;
 291         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
 292         device->fs_devices = fs_info->fs_devices;
 293
 294         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 295         list_add(&device->dev_list, &fs_info->fs_devices->devices);
 296         fs_info->fs_devices->num_devices++;
 297         fs_info->fs_devices->open_devices++;
 298         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 299
 300         *device_out = device;
 301         return 0;
 302
 303 error:
 304         blkdev_put(bdev, FMODE_EXCL);
 305         return ret;
 306 }
 307
 308 /*
 309  * called from commit_transaction. Writes changed device replace state to
 310  * disk.
 311  */
 312 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
 313 {
 314         struct btrfs_fs_info *fs_info = trans->fs_info;
 315         int ret;
 316         struct btrfs_root *dev_root = fs_info->dev_root;
 317         struct btrfs_path *path;
 318         struct btrfs_key key;
 319         struct extent_buffer *eb;
 320         struct btrfs_dev_replace_item *ptr;
 321         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 322
 323         down_read(&dev_replace->rwsem);
 324         if (!dev_replace->is_valid ||
 325             !dev_replace->item_needs_writeback) {
 326                 up_read(&dev_replace->rwsem);
 327                 return 0;
 328         }
 329         up_read(&dev_replace->rwsem);
 330
 331         key.objectid = 0;
 332         key.type = BTRFS_DEV_REPLACE_KEY;
 333         key.offset = 0;
 334
 335         path = btrfs_alloc_path();
 336         if (!path) {
 337                 ret = -ENOMEM;
 338                 goto out;
 339         }
 340         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 341         if (ret < 0) {
 342                 btrfs_warn(fs_info,
 343                            "error %d while searching for dev_replace item!",
 344                            ret);
 345                 goto out;
 346         }
 347
 348         if (ret == 0 &&
 349             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 350                 /*
 351                  * need to delete old one and insert a new one.
 352                  * Since no attempt is made to recover any old state, if the
 353                  * dev_replace state is 'running', the data on the target
 354                  * drive is lost.
 355                  * It would be possible to recover the state: just make sure
 356                  * that the beginning of the item is never changed and always
 357                  * contains all the essential information. Then read this
 358                  * minimal set of information and use it as a base for the
 359                  * new state.
 360                  */
 361                 ret = btrfs_del_item(trans, dev_root, path);
 362                 if (ret != 0) {
 363                         btrfs_warn(fs_info,
 364                                    "delete too small dev_replace item failed %d!",
 365                                    ret);
 366                         goto out;
 367                 }
 368                 ret = 1;
 369         }
 370
 371         if (ret == 1) {
 372                 /* need to insert a new item */
 373                 btrfs_release_path(path);
 374                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 375                                               &key, sizeof(*ptr));
 376                 if (ret < 0) {
 377                         btrfs_warn(fs_info,
 378                                    "insert dev_replace item failed %d!", ret);
 379                         goto out;
 380                 }
 381         }
 382
 383         eb = path->nodes[0];
 384         ptr = btrfs_item_ptr(eb, path->slots[0],
 385                              struct btrfs_dev_replace_item);
 386
 387         down_write(&dev_replace->rwsem);
 388         if (dev_replace->srcdev)
 389                 btrfs_set_dev_replace_src_devid(eb, ptr,
 390                         dev_replace->srcdev->devid);
 391         else
 392                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 393         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 394                 dev_replace->cont_reading_from_srcdev_mode);
 395         btrfs_set_dev_replace_replace_state(eb, ptr,
 396                 dev_replace->replace_state);
 397         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 398         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 399         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 400                 atomic64_read(&dev_replace->num_write_errors));
 401         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 402                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 403         dev_replace->cursor_left_last_write_of_item =
 404                 dev_replace->cursor_left;
 405         btrfs_set_dev_replace_cursor_left(eb, ptr,
 406                 dev_replace->cursor_left_last_write_of_item);
 407         btrfs_set_dev_replace_cursor_right(eb, ptr,
 408                 dev_replace->cursor_right);
 409         dev_replace->item_needs_writeback = 0;
 410         up_write(&dev_replace->rwsem);
 411
 412         btrfs_mark_buffer_dirty(eb);
 413
 414 out:
 415         btrfs_free_path(path);
 416
 417         return ret;
 418 }
 419
 420 static char* btrfs_dev_name(struct btrfs_device *device)
 421 {
 422         if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 423                 return "<missing disk>";
 424         else
 425                 return rcu_str_deref(device->name);
 426 }
 427
 428 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 429                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 430                 int read_src)
 431 {
 432         struct btrfs_root *root = fs_info->dev_root;
 433         struct btrfs_trans_handle *trans;
 434         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 435         int ret;
 436         struct btrfs_device *tgt_device = NULL;
 437         struct btrfs_device *src_device = NULL;
 438
 439         src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
 440                                                   srcdev_name);
 441         if (IS_ERR(src_device))
 442                 return PTR_ERR(src_device);
 443
 444         if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
 445                 btrfs_warn_in_rcu(fs_info,
 446           "cannot replace device %s (devid %llu) due to active swapfile",
 447                         btrfs_dev_name(src_device), src_device->devid);
 448                 return -ETXTBSY;
 449         }
 450
 451         /*
 452          * Here we commit the transaction to make sure commit_total_bytes
 453          * of all the devices are updated.
 454          */
 455         trans = btrfs_attach_transaction(root);
 456         if (!IS_ERR(trans)) {
 457                 ret = btrfs_commit_transaction(trans);
 458                 if (ret)
 459                         return ret;
 460         } else if (PTR_ERR(trans) != -ENOENT) {
 461                 return PTR_ERR(trans);
 462         }
 463
 464         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 465                                             src_device, &tgt_device);
 466         if (ret)
 467                 return ret;
 468
 469         down_write(&dev_replace->rwsem);
 470         switch (dev_replace->replace_state) {
 471         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 472         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 473         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 474                 break;
 475         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 476         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 477                 ASSERT(0);
 478                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 479                 up_write(&dev_replace->rwsem);
 480                 goto leave;
 481         }
 482
 483         dev_replace->cont_reading_from_srcdev_mode = read_src;
 484         dev_replace->srcdev = src_device;
 485         dev_replace->tgtdev = tgt_device;
 486
 487         btrfs_info_in_rcu(fs_info,
 488                       "dev_replace from %s (devid %llu) to %s started",
 489                       btrfs_dev_name(src_device),
 490                       src_device->devid,
 491                       rcu_str_deref(tgt_device->name));
 492
 493         /*
 494          * from now on, the writes to the srcdev are all duplicated to
 495          * go to the tgtdev as well (refer to btrfs_map_block()).
 496          */
 497         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 498         dev_replace->time_started = ktime_get_real_seconds();
 499         dev_replace->cursor_left = 0;
 500         dev_replace->committed_cursor_left = 0;
 501         dev_replace->cursor_left_last_write_of_item = 0;
 502         dev_replace->cursor_right = 0;
 503         dev_replace->is_valid = 1;
 504         dev_replace->item_needs_writeback = 1;
 505         atomic64_set(&dev_replace->num_write_errors, 0);
 506         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 507         up_write(&dev_replace->rwsem);
 508
 509         ret = btrfs_sysfs_add_device(tgt_device);
 510         if (ret)
 511                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
 512
 513         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 514
 515         /* Commit dev_replace state and reserve 1 item for it. */
 516         trans = btrfs_start_transaction(root, 1);
 517         if (IS_ERR(trans)) {
 518                 ret = PTR_ERR(trans);
 519                 down_write(&dev_replace->rwsem);
 520                 dev_replace->replace_state =
 521                         BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
 522                 dev_replace->srcdev = NULL;
 523                 dev_replace->tgtdev = NULL;
 524                 up_write(&dev_replace->rwsem);
 525                 goto leave;
 526         }
 527
 528         ret = btrfs_commit_transaction(trans);
 529         WARN_ON(ret);
 530
 531         /* the disk copy procedure reuses the scrub code */
 532         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 533                               btrfs_device_get_total_bytes(src_device),
 534                               &dev_replace->scrub_progress, 0, 1);
 535
 536         ret = btrfs_dev_replace_finishing(fs_info, ret);
 537         if (ret == -EINPROGRESS)
 538                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 539
 540         return ret;
 541
 542 leave:
 543         btrfs_destroy_dev_replace_tgtdev(tgt_device);
 544         return ret;
 545 }
 546
 547 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 548                             struct btrfs_ioctl_dev_replace_args *args)
 549 {
 550         int ret;
 551
 552         switch (args->start.cont_reading_from_srcdev_mode) {
 553         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 554         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 555                 break;
 556         default:
 557                 return -EINVAL;
 558         }
 559
 560         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 561             args->start.tgtdev_name[0] == '\0')
 562                 return -EINVAL;
 563
 564         ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
 565                                         args->start.srcdevid,
 566                                         args->start.srcdev_name,
 567                                         args->start.cont_reading_from_srcdev_mode);
 568         args->result = ret;
 569         /* don't warn if EINPROGRESS, someone else might be running scrub */
 570         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
 571             ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
 572                 return 0;
 573
 574         return ret;
 575 }
 576
 577 /*
 578  * blocked until all in-flight bios operations are finished.
 579  */
 580 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 581 {
 582         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 583         wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
 584                    &fs_info->dev_replace.bio_counter));
 585 }
 586
 587 /*
 588  * we have removed target device, it is safe to allow new bios request.
 589  */
 590 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 591 {
 592         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 593         wake_up(&fs_info->dev_replace.replace_wait);
 594 }
 595
 596 /*
 597  * When finishing the device replace, before swapping the source device with the
 598  * target device we must update the chunk allocation state in the target device,
 599  * as it is empty because replace works by directly copying the chunks and not
 600  * through the normal chunk allocation path.
 601  */
 602 static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
 603                                         struct btrfs_device *tgtdev)
 604 {
 605         struct extent_state *cached_state = NULL;
 606         u64 start = 0;
 607         u64 found_start;
 608         u64 found_end;
 609         int ret = 0;
 610
 611         lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
 612
 613         while (!find_first_extent_bit(&srcdev->alloc_state, start,
 614                                       &found_start, &found_end,
 615                                       CHUNK_ALLOCATED, &cached_state)) {
 616                 ret = set_extent_bits(&tgtdev->alloc_state, found_start,
 617                                       found_end, CHUNK_ALLOCATED);
 618                 if (ret)
 619                         break;
 620                 start = found_end + 1;
 621         }
 622
 623         free_extent_state(cached_state);
 624         return ret;
 625 }
 626
 627 static void btrfs_dev_replace_update_device_in_mapping_tree(
 628                                                 struct btrfs_fs_info *fs_info,
 629                                                 struct btrfs_device *srcdev,
 630                                                 struct btrfs_device *tgtdev)
 631 {
 632         struct extent_map_tree *em_tree = &fs_info->mapping_tree;
 633         struct extent_map *em;
 634         struct map_lookup *map;
 635         u64 start = 0;
 636         int i;
 637
 638         write_lock(&em_tree->lock);
 639         do {
 640                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 641                 if (!em)
 642                         break;
 643                 map = em->map_lookup;
 644                 for (i = 0; i < map->num_stripes; i++)
 645                         if (srcdev == map->stripes[i].dev)
 646                                 map->stripes[i].dev = tgtdev;
 647                 start = em->start + em->len;
 648                 free_extent_map(em);
 649         } while (start);
 650         write_unlock(&em_tree->lock);
 651 }
 652
 653 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 654                                        int scrub_ret)
 655 {
 656         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 657         struct btrfs_device *tgt_device;
 658         struct btrfs_device *src_device;
 659         struct btrfs_root *root = fs_info->tree_root;
 660         u8 uuid_tmp[BTRFS_UUID_SIZE];
 661         struct btrfs_trans_handle *trans;
 662         int ret = 0;
 663
 664         /* don't allow cancel or unmount to disturb the finishing procedure */
 665         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 666
 667         down_read(&dev_replace->rwsem);
 668         /* was the operation canceled, or is it finished? */
 669         if (dev_replace->replace_state !=
 670             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 671                 up_read(&dev_replace->rwsem);
 672                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 673                 return 0;
 674         }
 675
 676         tgt_device = dev_replace->tgtdev;
 677         src_device = dev_replace->srcdev;
 678         up_read(&dev_replace->rwsem);
 679
 680         /*
 681          * flush all outstanding I/O and inode extent mappings before the
 682          * copy operation is declared as being finished
 683          */
 684         ret = btrfs_start_delalloc_roots(fs_info, U64_MAX);
 685         if (ret) {
 686                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 687                 return ret;
 688         }
 689         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 690
 691         if (!scrub_ret)
 692                 btrfs_reada_remove_dev(src_device);
 693
 694         /*
 695          * We have to use this loop approach because at this point src_device
 696          * has to be available for transaction commit to complete, yet new
 697          * chunks shouldn't be allocated on the device.
 698          */
 699         while (1) {
 700                 trans = btrfs_start_transaction(root, 0);
 701                 if (IS_ERR(trans)) {
 702                         btrfs_reada_undo_remove_dev(src_device);
 703                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 704                         return PTR_ERR(trans);
 705                 }
 706                 ret = btrfs_commit_transaction(trans);
 707                 WARN_ON(ret);
 708
 709                 /* Prevent write_all_supers() during the finishing procedure */
 710                 mutex_lock(&fs_info->fs_devices->device_list_mutex);
 711                 /* Prevent new chunks being allocated on the source device */
 712                 mutex_lock(&fs_info->chunk_mutex);
 713
 714                 if (!list_empty(&src_device->post_commit_list)) {
 715                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 716                         mutex_unlock(&fs_info->chunk_mutex);
 717                 } else {
 718                         break;
 719                 }
 720         }
 721
 722         down_write(&dev_replace->rwsem);
 723         dev_replace->replace_state =
 724                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 725                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 726         dev_replace->tgtdev = NULL;
 727         dev_replace->srcdev = NULL;
 728         dev_replace->time_stopped = ktime_get_real_seconds();
 729         dev_replace->item_needs_writeback = 1;
 730
 731         /*
 732          * Update allocation state in the new device and replace the old device
 733          * with the new one in the mapping tree.
 734          */
 735         if (!scrub_ret) {
 736                 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device);
 737                 if (scrub_ret)
 738                         goto error;
 739                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 740                                                                 src_device,
 741                                                                 tgt_device);
 742         } else {
 743                 if (scrub_ret != -ECANCELED)
 744                         btrfs_err_in_rcu(fs_info,
 745                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 746                                  btrfs_dev_name(src_device),
 747                                  src_device->devid,
 748                                  rcu_str_deref(tgt_device->name), scrub_ret);
 749 error:
 750                 up_write(&dev_replace->rwsem);
 751                 mutex_unlock(&fs_info->chunk_mutex);
 752                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 753                 btrfs_reada_undo_remove_dev(src_device);
 754                 btrfs_rm_dev_replace_blocked(fs_info);
 755                 if (tgt_device)
 756                         btrfs_destroy_dev_replace_tgtdev(tgt_device);
 757                 btrfs_rm_dev_replace_unblocked(fs_info);
 758                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 759
 760                 return scrub_ret;
 761         }
 762
 763         btrfs_info_in_rcu(fs_info,
 764                           "dev_replace from %s (devid %llu) to %s finished",
 765                           btrfs_dev_name(src_device),
 766                           src_device->devid,
 767                           rcu_str_deref(tgt_device->name));
 768         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 769         tgt_device->devid = src_device->devid;
 770         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 771         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 772         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 773         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 774         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 775         btrfs_device_set_disk_total_bytes(tgt_device,
 776                                           src_device->disk_total_bytes);
 777         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 778         tgt_device->commit_bytes_used = src_device->bytes_used;
 779
 780         btrfs_assign_next_active_device(src_device, tgt_device);
 781
 782         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 783         fs_info->fs_devices->rw_devices++;
 784
 785         up_write(&dev_replace->rwsem);
 786         btrfs_rm_dev_replace_blocked(fs_info);
 787
 788         btrfs_rm_dev_replace_remove_srcdev(src_device);
 789
 790         btrfs_rm_dev_replace_unblocked(fs_info);
 791
 792         /*
 793          * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
 794          * update on-disk dev stats value during commit transaction
 795          */
 796         atomic_inc(&tgt_device->dev_stats_ccnt);
 797
 798         /*
 799          * this is again a consistent state where no dev_replace procedure
 800          * is running, the target device is part of the filesystem, the
 801          * source device is not part of the filesystem anymore and its 1st
 802          * superblock is scratched out so that it is no longer marked to
 803          * belong to this filesystem.
 804          */
 805         mutex_unlock(&fs_info->chunk_mutex);
 806         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 807
 808         /* replace the sysfs entry */
 809         btrfs_sysfs_remove_device(src_device);
 810         btrfs_sysfs_update_devid(tgt_device);
 811         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
 812                 btrfs_scratch_superblocks(fs_info, src_device->bdev,
 813                                           src_device->name->str);
 814
 815         /* write back the superblocks */
 816         trans = btrfs_start_transaction(root, 0);
 817         if (!IS_ERR(trans))
 818                 btrfs_commit_transaction(trans);
 819
 820         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 821
 822         btrfs_rm_dev_replace_free_srcdev(src_device);
 823
 824         return 0;
 825 }
 826
 827 /*
 828  * Read progress of device replace status according to the state and last
 829  * stored position. The value format is the same as for
 830  * btrfs_dev_replace::progress_1000
 831  */
 832 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 833 {
 834         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 835         u64 ret = 0;
 836
 837         switch (dev_replace->replace_state) {
 838         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 839         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 840                 ret = 0;
 841                 break;
 842         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 843                 ret = 1000;
 844                 break;
 845         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 846         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 847                 ret = div64_u64(dev_replace->cursor_left,
 848                                 div_u64(btrfs_device_get_total_bytes(
 849                                                 dev_replace->srcdev), 1000));
 850                 break;
 851         }
 852
 853         return ret;
 854 }
 855
 856 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 857                               struct btrfs_ioctl_dev_replace_args *args)
 858 {
 859         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 860
 861         down_read(&dev_replace->rwsem);
 862         /* even if !dev_replace_is_valid, the values are good enough for
 863          * the replace_status ioctl */
 864         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 865         args->status.replace_state = dev_replace->replace_state;
 866         args->status.time_started = dev_replace->time_started;
 867         args->status.time_stopped = dev_replace->time_stopped;
 868         args->status.num_write_errors =
 869                 atomic64_read(&dev_replace->num_write_errors);
 870         args->status.num_uncorrectable_read_errors =
 871                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 872         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 873         up_read(&dev_replace->rwsem);
 874 }
 875
 876 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 877 {
 878         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 879         struct btrfs_device *tgt_device = NULL;
 880         struct btrfs_device *src_device = NULL;
 881         struct btrfs_trans_handle *trans;
 882         struct btrfs_root *root = fs_info->tree_root;
 883         int result;
 884         int ret;
 885
 886         if (sb_rdonly(fs_info->sb))
 887                 return -EROFS;
 888
 889         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 890         down_write(&dev_replace->rwsem);
 891         switch (dev_replace->replace_state) {
 892         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 893         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 894         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 895                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 896                 up_write(&dev_replace->rwsem);
 897                 break;
 898         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 899                 tgt_device = dev_replace->tgtdev;
 900                 src_device = dev_replace->srcdev;
 901                 up_write(&dev_replace->rwsem);
 902                 ret = btrfs_scrub_cancel(fs_info);
 903                 if (ret < 0) {
 904                         result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 905                 } else {
 906                         result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 907                         /*
 908                          * btrfs_dev_replace_finishing() will handle the
 909                          * cleanup part
 910                          */
 911                         btrfs_info_in_rcu(fs_info,
 912                                 "dev_replace from %s (devid %llu) to %s canceled",
 913                                 btrfs_dev_name(src_device), src_device->devid,
 914                                 btrfs_dev_name(tgt_device));
 915                 }
 916                 break;
 917         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 918                 /*
 919                  * Scrub doing the replace isn't running so we need to do the
 920                  * cleanup step of btrfs_dev_replace_finishing() here
 921                  */
 922                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 923                 tgt_device = dev_replace->tgtdev;
 924                 src_device = dev_replace->srcdev;
 925                 dev_replace->tgtdev = NULL;
 926                 dev_replace->srcdev = NULL;
 927                 dev_replace->replace_state =
 928                                 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 929                 dev_replace->time_stopped = ktime_get_real_seconds();
 930                 dev_replace->item_needs_writeback = 1;
 931
 932                 up_write(&dev_replace->rwsem);
 933
 934                 /* Scrub for replace must not be running in suspended state */
 935                 ret = btrfs_scrub_cancel(fs_info);
 936                 ASSERT(ret != -ENOTCONN);
 937
 938                 trans = btrfs_start_transaction(root, 0);
 939                 if (IS_ERR(trans)) {
 940                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 941                         return PTR_ERR(trans);
 942                 }
 943                 ret = btrfs_commit_transaction(trans);
 944                 WARN_ON(ret);
 945
 946                 btrfs_info_in_rcu(fs_info,
 947                 "suspended dev_replace from %s (devid %llu) to %s canceled",
 948                         btrfs_dev_name(src_device), src_device->devid,
 949                         btrfs_dev_name(tgt_device));
 950
 951                 if (tgt_device)
 952                         btrfs_destroy_dev_replace_tgtdev(tgt_device);
 953                 break;
 954         default:
 955                 up_write(&dev_replace->rwsem);
 956                 result = -EINVAL;
 957         }
 958
 959         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 960         return result;
 961 }
 962
 963 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 964 {
 965         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 966
 967         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 968         down_write(&dev_replace->rwsem);
 969
 970         switch (dev_replace->replace_state) {
 971         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 972         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 973         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 974         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 975                 break;
 976         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 977                 dev_replace->replace_state =
 978                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 979                 dev_replace->time_stopped = ktime_get_real_seconds();
 980                 dev_replace->item_needs_writeback = 1;
 981                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 982                 break;
 983         }
 984
 985         up_write(&dev_replace->rwsem);
 986         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 987 }
 988
 989 /* resume dev_replace procedure that was interrupted by unmount */
 990 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 991 {
 992         struct task_struct *task;
 993         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 994
 995         down_write(&dev_replace->rwsem);
 996
 997         switch (dev_replace->replace_state) {
 998         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 999         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1000         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1001                 up_write(&dev_replace->rwsem);
1002                 return 0;
1003         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1004                 break;
1005         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1006                 dev_replace->replace_state =
1007                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1008                 break;
1009         }
1010         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1011                 btrfs_info(fs_info,
1012                            "cannot continue dev_replace, tgtdev is missing");
1013                 btrfs_info(fs_info,
1014                            "you may cancel the operation after 'mount -o degraded'");
1015                 dev_replace->replace_state =
1016                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1017                 up_write(&dev_replace->rwsem);
1018                 return 0;
1019         }
1020         up_write(&dev_replace->rwsem);
1021
1022         /*
1023          * This could collide with a paused balance, but the exclusive op logic
1024          * should never allow both to start and pause. We don't want to allow
1025          * dev-replace to start anyway.
1026          */
1027         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
1028                 down_write(&dev_replace->rwsem);
1029                 dev_replace->replace_state =
1030                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1031                 up_write(&dev_replace->rwsem);
1032                 btrfs_info(fs_info,
1033                 "cannot resume dev-replace, other exclusive operation running");
1034                 return 0;
1035         }
1036
1037         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1038         return PTR_ERR_OR_ZERO(task);
1039 }
1040
1041 static int btrfs_dev_replace_kthread(void *data)
1042 {
1043         struct btrfs_fs_info *fs_info = data;
1044         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1045         u64 progress;
1046         int ret;
1047
1048         progress = btrfs_dev_replace_progress(fs_info);
1049         progress = div_u64(progress, 10);
1050         btrfs_info_in_rcu(fs_info,
1051                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1052                 btrfs_dev_name(dev_replace->srcdev),
1053                 dev_replace->srcdev->devid,
1054                 btrfs_dev_name(dev_replace->tgtdev),
1055                 (unsigned int)progress);
1056
1057         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
1058                               dev_replace->committed_cursor_left,
1059                               btrfs_device_get_total_bytes(dev_replace->srcdev),
1060                               &dev_replace->scrub_progress, 0, 1);
1061         ret = btrfs_dev_replace_finishing(fs_info, ret);
1062         WARN_ON(ret && ret != -ECANCELED);
1063
1064         btrfs_exclop_finish(fs_info);
1065         return 0;
1066 }
1067
1068 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1069 {
1070         if (!dev_replace->is_valid)
1071                 return 0;
1072
1073         switch (dev_replace->replace_state) {
1074         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1075         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1076         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1077                 return 0;
1078         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1079         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1080                 /*
1081                  * return true even if tgtdev is missing (this is
1082                  * something that can happen if the dev_replace
1083                  * procedure is suspended by an umount and then
1084                  * the tgtdev is missing (or "btrfs dev scan") was
1085                  * not called and the filesystem is remounted
1086                  * in degraded state. This does not stop the
1087                  * dev_replace procedure. It needs to be canceled
1088                  * manually if the cancellation is wanted.
1089                  */
1090                 break;
1091         }
1092         return 1;
1093 }
1094
1095 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1096 {
1097         percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1098 }
1099
1100 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1101 {
1102         percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1103         cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1104 }
1105
1106 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1107 {
1108         while (1) {
1109                 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1110                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1111                                      &fs_info->fs_state)))
1112                         break;
1113
1114                 btrfs_bio_counter_dec(fs_info);
1115                 wait_event(fs_info->dev_replace.replace_wait,
1116                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1117                                      &fs_info->fs_state));
1118         }
1119 }