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