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