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