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 "ctree.h"
  13 #include "extent_map.h"
  14 #include "disk-io.h"
  15 #include "transaction.h"
  16 #include "print-tree.h"
  17 #include "volumes.h"
  18 #include "async-thread.h"
  19 #include "check-integrity.h"
  20 #include "rcu-string.h"
  21 #include "dev-replace.h"
  22 #include "sysfs.h"
  23
  24 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
  25                                        int scrub_ret);
  26 static void btrfs_dev_replace_update_device_in_mapping_tree(
  27                                                 struct btrfs_fs_info *fs_info,
  28                                                 struct btrfs_device *srcdev,
  29                                                 struct btrfs_device *tgtdev);
  30 static int btrfs_dev_replace_kthread(void *data);
  31
  32 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  33 {
  34         struct btrfs_key key;
  35         struct btrfs_root *dev_root = fs_info->dev_root;
  36         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  37         struct extent_buffer *eb;
  38         int slot;
  39         int ret = 0;
  40         struct btrfs_path *path = NULL;
  41         int item_size;
  42         struct btrfs_dev_replace_item *ptr;
  43         u64 src_devid;
  44
  45         path = btrfs_alloc_path();
  46         if (!path) {
  47                 ret = -ENOMEM;
  48                 goto out;
  49         }
  50
  51         key.objectid = 0;
  52         key.type = BTRFS_DEV_REPLACE_KEY;
  53         key.offset = 0;
  54         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  55         if (ret) {
  56 no_valid_dev_replace_entry_found:
  57                 /*
  58                  * We don't have a replace item or it's corrupted.  If there is
  59                  * a replace target, fail the mount.
  60                  */
  61                 if (btrfs_find_device(fs_info->fs_devices,
  62                                       BTRFS_DEV_REPLACE_DEVID, NULL, NULL, false)) {
  63                         btrfs_err(fs_info,
  64                         "found replace target device without a valid replace item");
  65                         ret = -EUCLEAN;
  66                         goto out;
  67                 }
  68                 ret = 0;
  69                 dev_replace->replace_state =
  70                         BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
  71                 dev_replace->cont_reading_from_srcdev_mode =
  72                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
  73                 dev_replace->replace_state = 0;
  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 devid present without an active replace item");
 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         filemap_write_and_wait(bdev->bd_inode->i_mapping);
 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         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 264         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 265         device->generation = 0;
 266         device->io_width = fs_info->sectorsize;
 267         device->io_align = fs_info->sectorsize;
 268         device->sector_size = fs_info->sectorsize;
 269         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
 270         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
 271         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
 272         device->commit_total_bytes = srcdev->commit_total_bytes;
 273         device->commit_bytes_used = device->bytes_used;
 274         device->fs_info = fs_info;
 275         device->bdev = bdev;
 276         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
 277         set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
 278         device->mode = FMODE_EXCL;
 279         device->dev_stats_valid = 1;
 280         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
 281         device->fs_devices = fs_info->fs_devices;
 282         list_add(&device->dev_list, &fs_info->fs_devices->devices);
 283         fs_info->fs_devices->num_devices++;
 284         fs_info->fs_devices->open_devices++;
 285         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 286
 287         *device_out = device;
 288         return 0;
 289
 290 error:
 291         blkdev_put(bdev, FMODE_EXCL);
 292         return ret;
 293 }
 294
 295 /*
 296  * called from commit_transaction. Writes changed device replace state to
 297  * disk.
 298  */
 299 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 300                           struct btrfs_fs_info *fs_info)
 301 {
 302         int ret;
 303         struct btrfs_root *dev_root = fs_info->dev_root;
 304         struct btrfs_path *path;
 305         struct btrfs_key key;
 306         struct extent_buffer *eb;
 307         struct btrfs_dev_replace_item *ptr;
 308         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 309
 310         btrfs_dev_replace_read_lock(dev_replace);
 311         if (!dev_replace->is_valid ||
 312             !dev_replace->item_needs_writeback) {
 313                 btrfs_dev_replace_read_unlock(dev_replace);
 314                 return 0;
 315         }
 316         btrfs_dev_replace_read_unlock(dev_replace);
 317
 318         key.objectid = 0;
 319         key.type = BTRFS_DEV_REPLACE_KEY;
 320         key.offset = 0;
 321
 322         path = btrfs_alloc_path();
 323         if (!path) {
 324                 ret = -ENOMEM;
 325                 goto out;
 326         }
 327         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 328         if (ret < 0) {
 329                 btrfs_warn(fs_info,
 330                            "error %d while searching for dev_replace item!",
 331                            ret);
 332                 goto out;
 333         }
 334
 335         if (ret == 0 &&
 336             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 337                 /*
 338                  * need to delete old one and insert a new one.
 339                  * Since no attempt is made to recover any old state, if the
 340                  * dev_replace state is 'running', the data on the target
 341                  * drive is lost.
 342                  * It would be possible to recover the state: just make sure
 343                  * that the beginning of the item is never changed and always
 344                  * contains all the essential information. Then read this
 345                  * minimal set of information and use it as a base for the
 346                  * new state.
 347                  */
 348                 ret = btrfs_del_item(trans, dev_root, path);
 349                 if (ret != 0) {
 350                         btrfs_warn(fs_info,
 351                                    "delete too small dev_replace item failed %d!",
 352                                    ret);
 353                         goto out;
 354                 }
 355                 ret = 1;
 356         }
 357
 358         if (ret == 1) {
 359                 /* need to insert a new item */
 360                 btrfs_release_path(path);
 361                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 362                                               &key, sizeof(*ptr));
 363                 if (ret < 0) {
 364                         btrfs_warn(fs_info,
 365                                    "insert dev_replace item failed %d!", ret);
 366                         goto out;
 367                 }
 368         }
 369
 370         eb = path->nodes[0];
 371         ptr = btrfs_item_ptr(eb, path->slots[0],
 372                              struct btrfs_dev_replace_item);
 373
 374         btrfs_dev_replace_write_lock(dev_replace);
 375         if (dev_replace->srcdev)
 376                 btrfs_set_dev_replace_src_devid(eb, ptr,
 377                         dev_replace->srcdev->devid);
 378         else
 379                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 380         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 381                 dev_replace->cont_reading_from_srcdev_mode);
 382         btrfs_set_dev_replace_replace_state(eb, ptr,
 383                 dev_replace->replace_state);
 384         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 385         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 386         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 387                 atomic64_read(&dev_replace->num_write_errors));
 388         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 389                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 390         dev_replace->cursor_left_last_write_of_item =
 391                 dev_replace->cursor_left;
 392         btrfs_set_dev_replace_cursor_left(eb, ptr,
 393                 dev_replace->cursor_left_last_write_of_item);
 394         btrfs_set_dev_replace_cursor_right(eb, ptr,
 395                 dev_replace->cursor_right);
 396         dev_replace->item_needs_writeback = 0;
 397         btrfs_dev_replace_write_unlock(dev_replace);
 398
 399         btrfs_mark_buffer_dirty(eb);
 400
 401 out:
 402         btrfs_free_path(path);
 403
 404         return ret;
 405 }
 406
 407 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 408 {
 409         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 410
 411         dev_replace->committed_cursor_left =
 412                 dev_replace->cursor_left_last_write_of_item;
 413 }
 414
 415 static char* btrfs_dev_name(struct btrfs_device *device)
 416 {
 417         if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 418                 return "<missing disk>";
 419         else
 420                 return rcu_str_deref(device->name);
 421 }
 422
 423 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 424                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 425                 int read_src)
 426 {
 427         struct btrfs_root *root = fs_info->dev_root;
 428         struct btrfs_trans_handle *trans;
 429         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 430         int ret;
 431         struct btrfs_device *tgt_device = NULL;
 432         struct btrfs_device *src_device = NULL;
 433
 434         ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
 435                                             srcdev_name, &src_device);
 436         if (ret)
 437                 return ret;
 438
 439         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 440                                             src_device, &tgt_device);
 441         if (ret)
 442                 return ret;
 443
 444         /*
 445          * Here we commit the transaction to make sure commit_total_bytes
 446          * of all the devices are updated.
 447          */
 448         trans = btrfs_attach_transaction(root);
 449         if (!IS_ERR(trans)) {
 450                 ret = btrfs_commit_transaction(trans);
 451                 if (ret)
 452                         return ret;
 453         } else if (PTR_ERR(trans) != -ENOENT) {
 454                 return PTR_ERR(trans);
 455         }
 456
 457         btrfs_dev_replace_write_lock(dev_replace);
 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                 goto leave;
 468         }
 469
 470         dev_replace->cont_reading_from_srcdev_mode = read_src;
 471         WARN_ON(!src_device);
 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         btrfs_dev_replace_write_unlock(dev_replace);
 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         /* force writing the updated state information to disk */
 504         trans = btrfs_start_transaction(root, 0);
 505         if (IS_ERR(trans)) {
 506                 ret = PTR_ERR(trans);
 507                 btrfs_dev_replace_write_lock(dev_replace);
 508                 dev_replace->replace_state =
 509                         BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
 510                 dev_replace->srcdev = NULL;
 511                 dev_replace->tgtdev = NULL;
 512                 goto leave;
 513         }
 514
 515         ret = btrfs_commit_transaction(trans);
 516         WARN_ON(ret);
 517
 518         /* the disk copy procedure reuses the scrub code */
 519         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 520                               btrfs_device_get_total_bytes(src_device),
 521                               &dev_replace->scrub_progress, 0, 1);
 522
 523         ret = btrfs_dev_replace_finishing(fs_info, ret);
 524         if (ret == -EINPROGRESS) {
 525                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 526         } else {
 527                 WARN_ON(ret);
 528         }
 529
 530         return ret;
 531
 532 leave:
 533         btrfs_dev_replace_write_unlock(dev_replace);
 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 = 0;
 563
 564         return ret;
 565 }
 566
 567 /*
 568  * blocked until all in-flight bios operations are finished.
 569  */
 570 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 571 {
 572         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 573         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 574                    &fs_info->bio_counter));
 575 }
 576
 577 /*
 578  * we have removed target device, it is safe to allow new bios request.
 579  */
 580 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 581 {
 582         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 583         wake_up(&fs_info->replace_wait);
 584 }
 585
 586 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 587                                        int scrub_ret)
 588 {
 589         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 590         struct btrfs_device *tgt_device;
 591         struct btrfs_device *src_device;
 592         struct btrfs_root *root = fs_info->tree_root;
 593         u8 uuid_tmp[BTRFS_UUID_SIZE];
 594         struct btrfs_trans_handle *trans;
 595         int ret = 0;
 596
 597         /* don't allow cancel or unmount to disturb the finishing procedure */
 598         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 599
 600         btrfs_dev_replace_read_lock(dev_replace);
 601         /* was the operation canceled, or is it finished? */
 602         if (dev_replace->replace_state !=
 603             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 604                 btrfs_dev_replace_read_unlock(dev_replace);
 605                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 606                 return 0;
 607         }
 608
 609         tgt_device = dev_replace->tgtdev;
 610         src_device = dev_replace->srcdev;
 611         btrfs_dev_replace_read_unlock(dev_replace);
 612
 613         /*
 614          * flush all outstanding I/O and inode extent mappings before the
 615          * copy operation is declared as being finished
 616          */
 617         ret = btrfs_start_delalloc_roots(fs_info, -1);
 618         if (ret) {
 619                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 620                 return ret;
 621         }
 622         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 623
 624         while (1) {
 625                 trans = btrfs_start_transaction(root, 0);
 626                 if (IS_ERR(trans)) {
 627                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 628                         return PTR_ERR(trans);
 629                 }
 630                 ret = btrfs_commit_transaction(trans);
 631                 WARN_ON(ret);
 632                 /* keep away write_all_supers() during the finishing procedure */
 633                 mutex_lock(&fs_info->fs_devices->device_list_mutex);
 634                 mutex_lock(&fs_info->chunk_mutex);
 635                 if (src_device->has_pending_chunks) {
 636                         mutex_unlock(&root->fs_info->chunk_mutex);
 637                         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 638                 } else {
 639                         break;
 640                 }
 641         }
 642
 643         btrfs_dev_replace_write_lock(dev_replace);
 644         dev_replace->replace_state =
 645                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 646                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 647         dev_replace->tgtdev = NULL;
 648         dev_replace->srcdev = NULL;
 649         dev_replace->time_stopped = ktime_get_real_seconds();
 650         dev_replace->item_needs_writeback = 1;
 651
 652         /* replace old device with new one in mapping tree */
 653         if (!scrub_ret) {
 654                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 655                                                                 src_device,
 656                                                                 tgt_device);
 657         } else {
 658                 btrfs_err_in_rcu(fs_info,
 659                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 660                                  btrfs_dev_name(src_device),
 661                                  src_device->devid,
 662                                  rcu_str_deref(tgt_device->name), scrub_ret);
 663                 btrfs_dev_replace_write_unlock(dev_replace);
 664                 mutex_unlock(&fs_info->chunk_mutex);
 665                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 666                 btrfs_rm_dev_replace_blocked(fs_info);
 667                 if (tgt_device)
 668                         btrfs_destroy_dev_replace_tgtdev(tgt_device);
 669                 btrfs_rm_dev_replace_unblocked(fs_info);
 670                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 671
 672                 return scrub_ret;
 673         }
 674
 675         btrfs_info_in_rcu(fs_info,
 676                           "dev_replace from %s (devid %llu) to %s finished",
 677                           btrfs_dev_name(src_device),
 678                           src_device->devid,
 679                           rcu_str_deref(tgt_device->name));
 680         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 681         tgt_device->devid = src_device->devid;
 682         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 683         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 684         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 685         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 686         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 687         btrfs_device_set_disk_total_bytes(tgt_device,
 688                                           src_device->disk_total_bytes);
 689         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 690         ASSERT(list_empty(&src_device->resized_list));
 691         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 692         tgt_device->commit_bytes_used = src_device->bytes_used;
 693
 694         btrfs_assign_next_active_device(src_device, tgt_device);
 695
 696         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 697         fs_info->fs_devices->rw_devices++;
 698
 699         btrfs_dev_replace_write_unlock(dev_replace);
 700
 701         btrfs_rm_dev_replace_blocked(fs_info);
 702
 703         btrfs_rm_dev_replace_remove_srcdev(src_device);
 704
 705         btrfs_rm_dev_replace_unblocked(fs_info);
 706
 707         /*
 708          * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
 709          * update on-disk dev stats value during commit transaction
 710          */
 711         atomic_inc(&tgt_device->dev_stats_ccnt);
 712
 713         /*
 714          * this is again a consistent state where no dev_replace procedure
 715          * is running, the target device is part of the filesystem, the
 716          * source device is not part of the filesystem anymore and its 1st
 717          * superblock is scratched out so that it is no longer marked to
 718          * belong to this filesystem.
 719          */
 720         mutex_unlock(&fs_info->chunk_mutex);
 721         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 722
 723         /* replace the sysfs entry */
 724         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 725         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 726
 727         /* write back the superblocks */
 728         trans = btrfs_start_transaction(root, 0);
 729         if (!IS_ERR(trans))
 730                 btrfs_commit_transaction(trans);
 731
 732         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 733
 734         return 0;
 735 }
 736
 737 static void btrfs_dev_replace_update_device_in_mapping_tree(
 738                                                 struct btrfs_fs_info *fs_info,
 739                                                 struct btrfs_device *srcdev,
 740                                                 struct btrfs_device *tgtdev)
 741 {
 742         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 743         struct extent_map *em;
 744         struct map_lookup *map;
 745         u64 start = 0;
 746         int i;
 747
 748         write_lock(&em_tree->lock);
 749         do {
 750                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 751                 if (!em)
 752                         break;
 753                 map = em->map_lookup;
 754                 for (i = 0; i < map->num_stripes; i++)
 755                         if (srcdev == map->stripes[i].dev)
 756                                 map->stripes[i].dev = tgtdev;
 757                 start = em->start + em->len;
 758                 free_extent_map(em);
 759         } while (start);
 760         write_unlock(&em_tree->lock);
 761 }
 762
 763 /*
 764  * Read progress of device replace status according to the state and last
 765  * stored position. The value format is the same as for
 766  * btrfs_dev_replace::progress_1000
 767  */
 768 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 769 {
 770         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 771         u64 ret = 0;
 772
 773         switch (dev_replace->replace_state) {
 774         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 775         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 776                 ret = 0;
 777                 break;
 778         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 779                 ret = 1000;
 780                 break;
 781         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 782         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 783                 ret = div64_u64(dev_replace->cursor_left,
 784                                 div_u64(btrfs_device_get_total_bytes(
 785                                                 dev_replace->srcdev), 1000));
 786                 break;
 787         }
 788
 789         return ret;
 790 }
 791
 792 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 793                               struct btrfs_ioctl_dev_replace_args *args)
 794 {
 795         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 796
 797         btrfs_dev_replace_read_lock(dev_replace);
 798         /* even if !dev_replace_is_valid, the values are good enough for
 799          * the replace_status ioctl */
 800         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 801         args->status.replace_state = dev_replace->replace_state;
 802         args->status.time_started = dev_replace->time_started;
 803         args->status.time_stopped = dev_replace->time_stopped;
 804         args->status.num_write_errors =
 805                 atomic64_read(&dev_replace->num_write_errors);
 806         args->status.num_uncorrectable_read_errors =
 807                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 808         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 809         btrfs_dev_replace_read_unlock(dev_replace);
 810 }
 811
 812 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 813 {
 814         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 815         struct btrfs_device *tgt_device = NULL;
 816         struct btrfs_device *src_device = NULL;
 817         struct btrfs_trans_handle *trans;
 818         struct btrfs_root *root = fs_info->tree_root;
 819         int result;
 820         int ret;
 821
 822         if (sb_rdonly(fs_info->sb))
 823                 return -EROFS;
 824
 825         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 826         btrfs_dev_replace_write_lock(dev_replace);
 827         switch (dev_replace->replace_state) {
 828         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 829         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 830         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 831                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 832                 btrfs_dev_replace_write_unlock(dev_replace);
 833                 break;
 834         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 835                 tgt_device = dev_replace->tgtdev;
 836                 src_device = dev_replace->srcdev;
 837                 btrfs_dev_replace_write_unlock(dev_replace);
 838                 ret = btrfs_scrub_cancel(fs_info);
 839                 if (ret < 0) {
 840                         result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 841                 } else {
 842                         result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 843                         /*
 844                          * btrfs_dev_replace_finishing() will handle the
 845                          * cleanup part
 846                          */
 847                         btrfs_info_in_rcu(fs_info,
 848                                 "dev_replace from %s (devid %llu) to %s canceled",
 849                                 btrfs_dev_name(src_device), src_device->devid,
 850                                 btrfs_dev_name(tgt_device));
 851                 }
 852                 break;
 853         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 854                 /*
 855                  * Scrub doing the replace isn't running so we need to do the
 856                  * cleanup step of btrfs_dev_replace_finishing() here
 857                  */
 858                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 859                 tgt_device = dev_replace->tgtdev;
 860                 src_device = dev_replace->srcdev;
 861                 dev_replace->tgtdev = NULL;
 862                 dev_replace->srcdev = NULL;
 863                 dev_replace->replace_state =
 864                                 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 865                 dev_replace->time_stopped = ktime_get_real_seconds();
 866                 dev_replace->item_needs_writeback = 1;
 867
 868                 btrfs_dev_replace_write_unlock(dev_replace);
 869
 870                 btrfs_scrub_cancel(fs_info);
 871
 872                 trans = btrfs_start_transaction(root, 0);
 873                 if (IS_ERR(trans)) {
 874                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 875                         return PTR_ERR(trans);
 876                 }
 877                 ret = btrfs_commit_transaction(trans);
 878                 WARN_ON(ret);
 879
 880                 btrfs_info_in_rcu(fs_info,
 881                 "suspended dev_replace from %s (devid %llu) to %s canceled",
 882                         btrfs_dev_name(src_device), src_device->devid,
 883                         btrfs_dev_name(tgt_device));
 884
 885                 if (tgt_device)
 886                         btrfs_destroy_dev_replace_tgtdev(tgt_device);
 887                 break;
 888         default:
 889                 result = -EINVAL;
 890         }
 891
 892         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 893         return result;
 894 }
 895
 896 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 897 {
 898         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 899
 900         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 901         btrfs_dev_replace_write_lock(dev_replace);
 902         switch (dev_replace->replace_state) {
 903         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 904         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 905         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 906         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 907                 break;
 908         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 909                 dev_replace->replace_state =
 910                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 911                 dev_replace->time_stopped = ktime_get_real_seconds();
 912                 dev_replace->item_needs_writeback = 1;
 913                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 914                 break;
 915         }
 916
 917         btrfs_dev_replace_write_unlock(dev_replace);
 918         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 919 }
 920
 921 /* resume dev_replace procedure that was interrupted by unmount */
 922 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 923 {
 924         struct task_struct *task;
 925         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 926
 927         btrfs_dev_replace_write_lock(dev_replace);
 928         switch (dev_replace->replace_state) {
 929         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 930         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 931         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 932                 btrfs_dev_replace_write_unlock(dev_replace);
 933                 return 0;
 934         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 935                 break;
 936         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 937                 dev_replace->replace_state =
 938                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 939                 break;
 940         }
 941         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 942                 btrfs_info(fs_info,
 943                            "cannot continue dev_replace, tgtdev is missing");
 944                 btrfs_info(fs_info,
 945                            "you may cancel the operation after 'mount -o degraded'");
 946                 dev_replace->replace_state =
 947                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 948                 btrfs_dev_replace_write_unlock(dev_replace);
 949                 return 0;
 950         }
 951         btrfs_dev_replace_write_unlock(dev_replace);
 952
 953         /*
 954          * This could collide with a paused balance, but the exclusive op logic
 955          * should never allow both to start and pause. We don't want to allow
 956          * dev-replace to start anyway.
 957          */
 958         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
 959                 btrfs_dev_replace_write_lock(dev_replace);
 960                 dev_replace->replace_state =
 961                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 962                 btrfs_dev_replace_write_unlock(dev_replace);
 963                 btrfs_info(fs_info,
 964                 "cannot resume dev-replace, other exclusive operation running");
 965                 return 0;
 966         }
 967
 968         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 969         return PTR_ERR_OR_ZERO(task);
 970 }
 971
 972 static int btrfs_dev_replace_kthread(void *data)
 973 {
 974         struct btrfs_fs_info *fs_info = data;
 975         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 976         u64 progress;
 977         int ret;
 978
 979         progress = btrfs_dev_replace_progress(fs_info);
 980         progress = div_u64(progress, 10);
 981         btrfs_info_in_rcu(fs_info,
 982                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
 983                 btrfs_dev_name(dev_replace->srcdev),
 984                 dev_replace->srcdev->devid,
 985                 btrfs_dev_name(dev_replace->tgtdev),
 986                 (unsigned int)progress);
 987
 988         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 989                               dev_replace->committed_cursor_left,
 990                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 991                               &dev_replace->scrub_progress, 0, 1);
 992         ret = btrfs_dev_replace_finishing(fs_info, ret);
 993         WARN_ON(ret);
 994
 995         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 996         return 0;
 997 }
 998
 999 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1000 {
1001         if (!dev_replace->is_valid)
1002                 return 0;
1003
1004         switch (dev_replace->replace_state) {
1005         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1006         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1007         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1008                 return 0;
1009         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1010         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1011                 /*
1012                  * return true even if tgtdev is missing (this is
1013                  * something that can happen if the dev_replace
1014                  * procedure is suspended by an umount and then
1015                  * the tgtdev is missing (or "btrfs dev scan") was
1016                  * not called and the the filesystem is remounted
1017                  * in degraded state. This does not stop the
1018                  * dev_replace procedure. It needs to be canceled
1019                  * manually if the cancellation is wanted.
1020                  */
1021                 break;
1022         }
1023         return 1;
1024 }
1025
1026 void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace)
1027 {
1028         read_lock(&dev_replace->lock);
1029         atomic_inc(&dev_replace->read_locks);
1030 }
1031
1032 void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace)
1033 {
1034         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1035         atomic_dec(&dev_replace->read_locks);
1036         read_unlock(&dev_replace->lock);
1037 }
1038
1039 void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace)
1040 {
1041 again:
1042         wait_event(dev_replace->read_lock_wq,
1043                    atomic_read(&dev_replace->blocking_readers) == 0);
1044         write_lock(&dev_replace->lock);
1045         if (atomic_read(&dev_replace->blocking_readers)) {
1046                 write_unlock(&dev_replace->lock);
1047                 goto again;
1048         }
1049 }
1050
1051 void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace)
1052 {
1053         ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
1054         write_unlock(&dev_replace->lock);
1055 }
1056
1057 /* inc blocking cnt and release read lock */
1058 void btrfs_dev_replace_set_lock_blocking(
1059                                         struct btrfs_dev_replace *dev_replace)
1060 {
1061         /* only set blocking for read lock */
1062         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1063         atomic_inc(&dev_replace->blocking_readers);
1064         read_unlock(&dev_replace->lock);
1065 }
1066
1067 /* acquire read lock and dec blocking cnt */
1068 void btrfs_dev_replace_clear_lock_blocking(
1069                                         struct btrfs_dev_replace *dev_replace)
1070 {
1071         /* only set blocking for read lock */
1072         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1073         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
1074         read_lock(&dev_replace->lock);
1075         /* Barrier implied by atomic_dec_and_test */
1076         if (atomic_dec_and_test(&dev_replace->blocking_readers))
1077                 cond_wake_up_nomb(&dev_replace->read_lock_wq);
1078 }
1079
1080 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1081 {
1082         percpu_counter_inc(&fs_info->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->bio_counter, amount);
1088         cond_wake_up_nomb(&fs_info->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->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->replace_wait,
1101                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1102                                      &fs_info->fs_state));
1103         }
1104 }