fs/btrfs/dev-replace.c

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