fs/btrfs/dev-replace.c

   1 /*
   2  * Copyright (C) STRATO AG 2012.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/kthread.h>
  27 #include <linux/math64.h>
  28 #include <asm/div64.h>
  29 #include "ctree.h"
  30 #include "extent_map.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "print-tree.h"
  34 #include "volumes.h"
  35 #include "async-thread.h"
  36 #include "check-integrity.h"
  37 #include "rcu-string.h"
  38 #include "dev-replace.h"
  39 #include "sysfs.h"
  40
  41 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
  42                                        int scrub_ret);
  43 static void btrfs_dev_replace_update_device_in_mapping_tree(
  44                                                 struct btrfs_fs_info *fs_info,
  45                                                 struct btrfs_device *srcdev,
  46                                                 struct btrfs_device *tgtdev);
  47 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
  48                                          char *srcdev_name,
  49                                          struct btrfs_device **device);
  50 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
  51 static int btrfs_dev_replace_kthread(void *data);
  52 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
  53
  54
  55 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  56 {
  57         struct btrfs_key key;
  58         struct btrfs_root *dev_root = fs_info->dev_root;
  59         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  60         struct extent_buffer *eb;
  61         int slot;
  62         int ret = 0;
  63         struct btrfs_path *path = NULL;
  64         int item_size;
  65         struct btrfs_dev_replace_item *ptr;
  66         u64 src_devid;
  67
  68         path = btrfs_alloc_path();
  69         if (!path) {
  70                 ret = -ENOMEM;
  71                 goto out;
  72         }
  73
  74         key.objectid = 0;
  75         key.type = BTRFS_DEV_REPLACE_KEY;
  76         key.offset = 0;
  77         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  78         if (ret) {
  79 no_valid_dev_replace_entry_found:
  80                 ret = 0;
  81                 dev_replace->replace_state =
  82                         BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
  83                 dev_replace->cont_reading_from_srcdev_mode =
  84                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
  85                 dev_replace->replace_state = 0;
  86                 dev_replace->time_started = 0;
  87                 dev_replace->time_stopped = 0;
  88                 atomic64_set(&dev_replace->num_write_errors, 0);
  89                 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
  90                 dev_replace->cursor_left = 0;
  91                 dev_replace->committed_cursor_left = 0;
  92                 dev_replace->cursor_left_last_write_of_item = 0;
  93                 dev_replace->cursor_right = 0;
  94                 dev_replace->srcdev = NULL;
  95                 dev_replace->tgtdev = NULL;
  96                 dev_replace->is_valid = 0;
  97                 dev_replace->item_needs_writeback = 0;
  98                 goto out;
  99         }
 100         slot = path->slots[0];
 101         eb = path->nodes[0];
 102         item_size = btrfs_item_size_nr(eb, slot);
 103         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
 104
 105         if (item_size != sizeof(struct btrfs_dev_replace_item)) {
 106                 btrfs_warn(fs_info,
 107                         "dev_replace entry found has unexpected size, ignore entry");
 108                 goto no_valid_dev_replace_entry_found;
 109         }
 110
 111         src_devid = btrfs_dev_replace_src_devid(eb, ptr);
 112         dev_replace->cont_reading_from_srcdev_mode =
 113                 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
 114         dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
 115         dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
 116         dev_replace->time_stopped =
 117                 btrfs_dev_replace_time_stopped(eb, ptr);
 118         atomic64_set(&dev_replace->num_write_errors,
 119                      btrfs_dev_replace_num_write_errors(eb, ptr));
 120         atomic64_set(&dev_replace->num_uncorrectable_read_errors,
 121                      btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
 122         dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
 123         dev_replace->committed_cursor_left = dev_replace->cursor_left;
 124         dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
 125         dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
 126         dev_replace->is_valid = 1;
 127
 128         dev_replace->item_needs_writeback = 0;
 129         switch (dev_replace->replace_state) {
 130         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 131         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 132         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 133                 dev_replace->srcdev = NULL;
 134                 dev_replace->tgtdev = NULL;
 135                 break;
 136         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 137         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 138                 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
 139                                                         NULL, NULL);
 140                 dev_replace->tgtdev = btrfs_find_device(fs_info,
 141                                                         BTRFS_DEV_REPLACE_DEVID,
 142                                                         NULL, NULL);
 143                 /*
 144                  * allow 'btrfs dev replace_cancel' if src/tgt device is
 145                  * missing
 146                  */
 147                 if (!dev_replace->srcdev &&
 148                     !btrfs_test_opt(dev_root, DEGRADED)) {
 149                         ret = -EIO;
 150                         btrfs_warn(fs_info,
 151                            "cannot mount because device replace operation is ongoing and");
 152                         btrfs_warn(fs_info,
 153                            "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 154                            src_devid);
 155                 }
 156                 if (!dev_replace->tgtdev &&
 157                     !btrfs_test_opt(dev_root, DEGRADED)) {
 158                         ret = -EIO;
 159                         btrfs_warn(fs_info,
 160                            "cannot mount because device replace operation is ongoing and");
 161                         btrfs_warn(fs_info,
 162                            "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 163                                 BTRFS_DEV_REPLACE_DEVID);
 164                 }
 165                 if (dev_replace->tgtdev) {
 166                         if (dev_replace->srcdev) {
 167                                 dev_replace->tgtdev->total_bytes =
 168                                         dev_replace->srcdev->total_bytes;
 169                                 dev_replace->tgtdev->disk_total_bytes =
 170                                         dev_replace->srcdev->disk_total_bytes;
 171                                 dev_replace->tgtdev->commit_total_bytes =
 172                                         dev_replace->srcdev->commit_total_bytes;
 173                                 dev_replace->tgtdev->bytes_used =
 174                                         dev_replace->srcdev->bytes_used;
 175                                 dev_replace->tgtdev->commit_bytes_used =
 176                                         dev_replace->srcdev->commit_bytes_used;
 177                         }
 178                         dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
 179                         btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
 180                                 dev_replace->tgtdev);
 181                 }
 182                 break;
 183         }
 184
 185 out:
 186         btrfs_free_path(path);
 187         return ret;
 188 }
 189
 190 /*
 191  * called from commit_transaction. Writes changed device replace state to
 192  * disk.
 193  */
 194 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 195                           struct btrfs_fs_info *fs_info)
 196 {
 197         int ret;
 198         struct btrfs_root *dev_root = fs_info->dev_root;
 199         struct btrfs_path *path;
 200         struct btrfs_key key;
 201         struct extent_buffer *eb;
 202         struct btrfs_dev_replace_item *ptr;
 203         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 204
 205         btrfs_dev_replace_lock(dev_replace);
 206         if (!dev_replace->is_valid ||
 207             !dev_replace->item_needs_writeback) {
 208                 btrfs_dev_replace_unlock(dev_replace);
 209                 return 0;
 210         }
 211         btrfs_dev_replace_unlock(dev_replace);
 212
 213         key.objectid = 0;
 214         key.type = BTRFS_DEV_REPLACE_KEY;
 215         key.offset = 0;
 216
 217         path = btrfs_alloc_path();
 218         if (!path) {
 219                 ret = -ENOMEM;
 220                 goto out;
 221         }
 222         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 223         if (ret < 0) {
 224                 btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
 225                         ret);
 226                 goto out;
 227         }
 228
 229         if (ret == 0 &&
 230             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 231                 /*
 232                  * need to delete old one and insert a new one.
 233                  * Since no attempt is made to recover any old state, if the
 234                  * dev_replace state is 'running', the data on the target
 235                  * drive is lost.
 236                  * It would be possible to recover the state: just make sure
 237                  * that the beginning of the item is never changed and always
 238                  * contains all the essential information. Then read this
 239                  * minimal set of information and use it as a base for the
 240                  * new state.
 241                  */
 242                 ret = btrfs_del_item(trans, dev_root, path);
 243                 if (ret != 0) {
 244                         btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
 245                                 ret);
 246                         goto out;
 247                 }
 248                 ret = 1;
 249         }
 250
 251         if (ret == 1) {
 252                 /* need to insert a new item */
 253                 btrfs_release_path(path);
 254                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 255                                               &key, sizeof(*ptr));
 256                 if (ret < 0) {
 257                         btrfs_warn(fs_info, "insert dev_replace item failed %d!",
 258                                 ret);
 259                         goto out;
 260                 }
 261         }
 262
 263         eb = path->nodes[0];
 264         ptr = btrfs_item_ptr(eb, path->slots[0],
 265                              struct btrfs_dev_replace_item);
 266
 267         btrfs_dev_replace_lock(dev_replace);
 268         if (dev_replace->srcdev)
 269                 btrfs_set_dev_replace_src_devid(eb, ptr,
 270                         dev_replace->srcdev->devid);
 271         else
 272                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 273         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 274                 dev_replace->cont_reading_from_srcdev_mode);
 275         btrfs_set_dev_replace_replace_state(eb, ptr,
 276                 dev_replace->replace_state);
 277         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 278         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 279         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 280                 atomic64_read(&dev_replace->num_write_errors));
 281         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 282                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 283         dev_replace->cursor_left_last_write_of_item =
 284                 dev_replace->cursor_left;
 285         btrfs_set_dev_replace_cursor_left(eb, ptr,
 286                 dev_replace->cursor_left_last_write_of_item);
 287         btrfs_set_dev_replace_cursor_right(eb, ptr,
 288                 dev_replace->cursor_right);
 289         dev_replace->item_needs_writeback = 0;
 290         btrfs_dev_replace_unlock(dev_replace);
 291
 292         btrfs_mark_buffer_dirty(eb);
 293
 294 out:
 295         btrfs_free_path(path);
 296
 297         return ret;
 298 }
 299
 300 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 301 {
 302         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 303
 304         dev_replace->committed_cursor_left =
 305                 dev_replace->cursor_left_last_write_of_item;
 306 }
 307
 308 int btrfs_dev_replace_start(struct btrfs_root *root,
 309                             struct btrfs_ioctl_dev_replace_args *args)
 310 {
 311         struct btrfs_trans_handle *trans;
 312         struct btrfs_fs_info *fs_info = root->fs_info;
 313         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 314         int ret;
 315         struct btrfs_device *tgt_device = NULL;
 316         struct btrfs_device *src_device = NULL;
 317
 318         switch (args->start.cont_reading_from_srcdev_mode) {
 319         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 320         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 321                 break;
 322         default:
 323                 return -EINVAL;
 324         }
 325
 326         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 327             args->start.tgtdev_name[0] == '\0')
 328                 return -EINVAL;
 329
 330         /* the disk copy procedure reuses the scrub code */
 331         mutex_lock(&fs_info->volume_mutex);
 332         ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
 333                                             args->start.srcdev_name,
 334                                             &src_device);
 335         if (ret) {
 336                 mutex_unlock(&fs_info->volume_mutex);
 337                 return ret;
 338         }
 339
 340         ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
 341                                             src_device, &tgt_device);
 342         mutex_unlock(&fs_info->volume_mutex);
 343         if (ret)
 344                 return ret;
 345
 346         /*
 347          * Here we commit the transaction to make sure commit_total_bytes
 348          * of all the devices are updated.
 349          */
 350         trans = btrfs_attach_transaction(root);
 351         if (!IS_ERR(trans)) {
 352                 ret = btrfs_commit_transaction(trans, root);
 353                 if (ret)
 354                         return ret;
 355         } else if (PTR_ERR(trans) != -ENOENT) {
 356                 return PTR_ERR(trans);
 357         }
 358
 359         btrfs_dev_replace_lock(dev_replace);
 360         switch (dev_replace->replace_state) {
 361         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 362         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 363         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 364                 break;
 365         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 366         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 367                 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 368                 goto leave;
 369         }
 370
 371         dev_replace->cont_reading_from_srcdev_mode =
 372                 args->start.cont_reading_from_srcdev_mode;
 373         WARN_ON(!src_device);
 374         dev_replace->srcdev = src_device;
 375         WARN_ON(!tgt_device);
 376         dev_replace->tgtdev = tgt_device;
 377
 378         btrfs_info_in_rcu(root->fs_info,
 379                       "dev_replace from %s (devid %llu) to %s started",
 380                       src_device->missing ? "<missing disk>" :
 381                         rcu_str_deref(src_device->name),
 382                       src_device->devid,
 383                       rcu_str_deref(tgt_device->name));
 384
 385         /*
 386          * from now on, the writes to the srcdev are all duplicated to
 387          * go to the tgtdev as well (refer to btrfs_map_block()).
 388          */
 389         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 390         dev_replace->time_started = get_seconds();
 391         dev_replace->cursor_left = 0;
 392         dev_replace->committed_cursor_left = 0;
 393         dev_replace->cursor_left_last_write_of_item = 0;
 394         dev_replace->cursor_right = 0;
 395         dev_replace->is_valid = 1;
 396         dev_replace->item_needs_writeback = 1;
 397         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 398         btrfs_dev_replace_unlock(dev_replace);
 399
 400         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 401         if (ret)
 402                 btrfs_err(root->fs_info, "kobj add dev failed %d\n", ret);
 403
 404         btrfs_wait_ordered_roots(root->fs_info, -1);
 405
 406         /* force writing the updated state information to disk */
 407         trans = btrfs_start_transaction(root, 0);
 408         if (IS_ERR(trans)) {
 409                 ret = PTR_ERR(trans);
 410                 btrfs_dev_replace_lock(dev_replace);
 411                 goto leave;
 412         }
 413
 414         ret = btrfs_commit_transaction(trans, root);
 415         WARN_ON(ret);
 416
 417         /* the disk copy procedure reuses the scrub code */
 418         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 419                               btrfs_device_get_total_bytes(src_device),
 420                               &dev_replace->scrub_progress, 0, 1);
 421
 422         ret = btrfs_dev_replace_finishing(root->fs_info, ret);
 423         /* don't warn if EINPROGRESS, someone else might be running scrub */
 424         if (ret == -EINPROGRESS) {
 425                 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 426                 ret = 0;
 427         } else {
 428                 WARN_ON(ret);
 429         }
 430
 431         return ret;
 432
 433 leave:
 434         dev_replace->srcdev = NULL;
 435         dev_replace->tgtdev = NULL;
 436         btrfs_dev_replace_unlock(dev_replace);
 437         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 438         return ret;
 439 }
 440
 441 /*
 442  * blocked until all flighting bios are finished.
 443  */
 444 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 445 {
 446         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 447         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 448                    &fs_info->bio_counter));
 449 }
 450
 451 /*
 452  * we have removed target device, it is safe to allow new bios request.
 453  */
 454 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 455 {
 456         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 457         wake_up(&fs_info->replace_wait);
 458 }
 459
 460 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 461                                        int scrub_ret)
 462 {
 463         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 464         struct btrfs_device *tgt_device;
 465         struct btrfs_device *src_device;
 466         struct btrfs_root *root = fs_info->tree_root;
 467         u8 uuid_tmp[BTRFS_UUID_SIZE];
 468         struct btrfs_trans_handle *trans;
 469         int ret = 0;
 470
 471         /* don't allow cancel or unmount to disturb the finishing procedure */
 472         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 473
 474         btrfs_dev_replace_lock(dev_replace);
 475         /* was the operation canceled, or is it finished? */
 476         if (dev_replace->replace_state !=
 477             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 478                 btrfs_dev_replace_unlock(dev_replace);
 479                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 480                 return 0;
 481         }
 482
 483         tgt_device = dev_replace->tgtdev;
 484         src_device = dev_replace->srcdev;
 485         btrfs_dev_replace_unlock(dev_replace);
 486
 487         /*
 488          * flush all outstanding I/O and inode extent mappings before the
 489          * copy operation is declared as being finished
 490          */
 491         ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
 492         if (ret) {
 493                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 494                 return ret;
 495         }
 496         btrfs_wait_ordered_roots(root->fs_info, -1);
 497
 498         while (1) {
 499                 trans = btrfs_start_transaction(root, 0);
 500                 if (IS_ERR(trans)) {
 501                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 502                         return PTR_ERR(trans);
 503                 }
 504                 ret = btrfs_commit_transaction(trans, root);
 505                 WARN_ON(ret);
 506                 mutex_lock(&uuid_mutex);
 507                 /* keep away write_all_supers() during the finishing procedure */
 508                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 509                 mutex_lock(&root->fs_info->chunk_mutex);
 510                 if (src_device->has_pending_chunks) {
 511                         mutex_unlock(&root->fs_info->chunk_mutex);
 512                         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 513                         mutex_unlock(&uuid_mutex);
 514                 } else {
 515                         break;
 516                 }
 517         }
 518
 519         btrfs_dev_replace_lock(dev_replace);
 520         dev_replace->replace_state =
 521                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 522                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 523         dev_replace->tgtdev = NULL;
 524         dev_replace->srcdev = NULL;
 525         dev_replace->time_stopped = get_seconds();
 526         dev_replace->item_needs_writeback = 1;
 527
 528         /* replace old device with new one in mapping tree */
 529         if (!scrub_ret) {
 530                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 531                                                                 src_device,
 532                                                                 tgt_device);
 533         } else {
 534                 btrfs_err_in_rcu(root->fs_info,
 535                               "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 536                               src_device->missing ? "<missing disk>" :
 537                                 rcu_str_deref(src_device->name),
 538                               src_device->devid,
 539                               rcu_str_deref(tgt_device->name), scrub_ret);
 540                 btrfs_dev_replace_unlock(dev_replace);
 541                 mutex_unlock(&root->fs_info->chunk_mutex);
 542                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 543                 mutex_unlock(&uuid_mutex);
 544                 if (tgt_device)
 545                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 546                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 547
 548                 return scrub_ret;
 549         }
 550
 551         btrfs_info_in_rcu(root->fs_info,
 552                       "dev_replace from %s (devid %llu) to %s finished",
 553                       src_device->missing ? "<missing disk>" :
 554                         rcu_str_deref(src_device->name),
 555                       src_device->devid,
 556                       rcu_str_deref(tgt_device->name));
 557         tgt_device->is_tgtdev_for_dev_replace = 0;
 558         tgt_device->devid = src_device->devid;
 559         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 560         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 561         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 562         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 563         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 564         btrfs_device_set_disk_total_bytes(tgt_device,
 565                                           src_device->disk_total_bytes);
 566         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 567         ASSERT(list_empty(&src_device->resized_list));
 568         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 569         tgt_device->commit_bytes_used = src_device->bytes_used;
 570         if (fs_info->sb->s_bdev == src_device->bdev)
 571                 fs_info->sb->s_bdev = tgt_device->bdev;
 572         if (fs_info->fs_devices->latest_bdev == src_device->bdev)
 573                 fs_info->fs_devices->latest_bdev = tgt_device->bdev;
 574         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 575         fs_info->fs_devices->rw_devices++;
 576
 577         btrfs_dev_replace_unlock(dev_replace);
 578
 579         btrfs_rm_dev_replace_blocked(fs_info);
 580
 581         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 582
 583         btrfs_rm_dev_replace_unblocked(fs_info);
 584
 585         /*
 586          * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
 587          * update on-disk dev stats value during commit transaction
 588          */
 589         atomic_inc(&tgt_device->dev_stats_ccnt);
 590
 591         /*
 592          * this is again a consistent state where no dev_replace procedure
 593          * is running, the target device is part of the filesystem, the
 594          * source device is not part of the filesystem anymore and its 1st
 595          * superblock is scratched out so that it is no longer marked to
 596          * belong to this filesystem.
 597          */
 598         mutex_unlock(&root->fs_info->chunk_mutex);
 599         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 600         mutex_unlock(&uuid_mutex);
 601
 602         /* replace the sysfs entry */
 603         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 604         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 605
 606         /* write back the superblocks */
 607         trans = btrfs_start_transaction(root, 0);
 608         if (!IS_ERR(trans))
 609                 btrfs_commit_transaction(trans, root);
 610
 611         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 612
 613         return 0;
 614 }
 615
 616 static void btrfs_dev_replace_update_device_in_mapping_tree(
 617                                                 struct btrfs_fs_info *fs_info,
 618                                                 struct btrfs_device *srcdev,
 619                                                 struct btrfs_device *tgtdev)
 620 {
 621         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 622         struct extent_map *em;
 623         struct map_lookup *map;
 624         u64 start = 0;
 625         int i;
 626
 627         write_lock(&em_tree->lock);
 628         do {
 629                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 630                 if (!em)
 631                         break;
 632                 map = em->map_lookup;
 633                 for (i = 0; i < map->num_stripes; i++)
 634                         if (srcdev == map->stripes[i].dev)
 635                                 map->stripes[i].dev = tgtdev;
 636                 start = em->start + em->len;
 637                 free_extent_map(em);
 638         } while (start);
 639         write_unlock(&em_tree->lock);
 640 }
 641
 642 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
 643                                          char *srcdev_name,
 644                                          struct btrfs_device **device)
 645 {
 646         int ret;
 647
 648         if (srcdevid) {
 649                 ret = 0;
 650                 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
 651                                             NULL);
 652                 if (!*device)
 653                         ret = -ENOENT;
 654         } else {
 655                 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
 656                                                            device);
 657         }
 658         return ret;
 659 }
 660
 661 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 662                               struct btrfs_ioctl_dev_replace_args *args)
 663 {
 664         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 665         struct btrfs_device *srcdev;
 666
 667         btrfs_dev_replace_lock(dev_replace);
 668         /* even if !dev_replace_is_valid, the values are good enough for
 669          * the replace_status ioctl */
 670         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 671         args->status.replace_state = dev_replace->replace_state;
 672         args->status.time_started = dev_replace->time_started;
 673         args->status.time_stopped = dev_replace->time_stopped;
 674         args->status.num_write_errors =
 675                 atomic64_read(&dev_replace->num_write_errors);
 676         args->status.num_uncorrectable_read_errors =
 677                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 678         switch (dev_replace->replace_state) {
 679         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 680         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 681                 args->status.progress_1000 = 0;
 682                 break;
 683         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 684                 args->status.progress_1000 = 1000;
 685                 break;
 686         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 687         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 688                 srcdev = dev_replace->srcdev;
 689                 args->status.progress_1000 = div_u64(dev_replace->cursor_left,
 690                         div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
 691                 break;
 692         }
 693         btrfs_dev_replace_unlock(dev_replace);
 694 }
 695
 696 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
 697                              struct btrfs_ioctl_dev_replace_args *args)
 698 {
 699         args->result = __btrfs_dev_replace_cancel(fs_info);
 700         return 0;
 701 }
 702
 703 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 704 {
 705         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 706         struct btrfs_device *tgt_device = NULL;
 707         struct btrfs_trans_handle *trans;
 708         struct btrfs_root *root = fs_info->tree_root;
 709         u64 result;
 710         int ret;
 711
 712         if (fs_info->sb->s_flags & MS_RDONLY)
 713                 return -EROFS;
 714
 715         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 716         btrfs_dev_replace_lock(dev_replace);
 717         switch (dev_replace->replace_state) {
 718         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 719         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 720         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 721                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 722                 btrfs_dev_replace_unlock(dev_replace);
 723                 goto leave;
 724         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 725         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 726                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 727                 tgt_device = dev_replace->tgtdev;
 728                 dev_replace->tgtdev = NULL;
 729                 dev_replace->srcdev = NULL;
 730                 break;
 731         }
 732         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 733         dev_replace->time_stopped = get_seconds();
 734         dev_replace->item_needs_writeback = 1;
 735         btrfs_dev_replace_unlock(dev_replace);
 736         btrfs_scrub_cancel(fs_info);
 737
 738         trans = btrfs_start_transaction(root, 0);
 739         if (IS_ERR(trans)) {
 740                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 741                 return PTR_ERR(trans);
 742         }
 743         ret = btrfs_commit_transaction(trans, root);
 744         WARN_ON(ret);
 745         if (tgt_device)
 746                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 747
 748 leave:
 749         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 750         return result;
 751 }
 752
 753 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 754 {
 755         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 756
 757         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 758         btrfs_dev_replace_lock(dev_replace);
 759         switch (dev_replace->replace_state) {
 760         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 761         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 762         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 763         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 764                 break;
 765         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 766                 dev_replace->replace_state =
 767                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 768                 dev_replace->time_stopped = get_seconds();
 769                 dev_replace->item_needs_writeback = 1;
 770                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 771                 break;
 772         }
 773
 774         btrfs_dev_replace_unlock(dev_replace);
 775         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 776 }
 777
 778 /* resume dev_replace procedure that was interrupted by unmount */
 779 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 780 {
 781         struct task_struct *task;
 782         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 783
 784         btrfs_dev_replace_lock(dev_replace);
 785         switch (dev_replace->replace_state) {
 786         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 787         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 788         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 789                 btrfs_dev_replace_unlock(dev_replace);
 790                 return 0;
 791         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 792                 break;
 793         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 794                 dev_replace->replace_state =
 795                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 796                 break;
 797         }
 798         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 799                 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
 800                 btrfs_info(fs_info,
 801                         "you may cancel the operation after 'mount -o degraded'");
 802                 btrfs_dev_replace_unlock(dev_replace);
 803                 return 0;
 804         }
 805         btrfs_dev_replace_unlock(dev_replace);
 806
 807         WARN_ON(atomic_xchg(
 808                 &fs_info->mutually_exclusive_operation_running, 1));
 809         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 810         return PTR_ERR_OR_ZERO(task);
 811 }
 812
 813 static int btrfs_dev_replace_kthread(void *data)
 814 {
 815         struct btrfs_fs_info *fs_info = data;
 816         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 817         struct btrfs_ioctl_dev_replace_args *status_args;
 818         u64 progress;
 819
 820         status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
 821         if (status_args) {
 822                 btrfs_dev_replace_status(fs_info, status_args);
 823                 progress = status_args->status.progress_1000;
 824                 kfree(status_args);
 825                 progress = div_u64(progress, 10);
 826                 btrfs_info_in_rcu(fs_info,
 827                         "continuing dev_replace from %s (devid %llu) to %s @%u%%",
 828                         dev_replace->srcdev->missing ? "<missing disk>" :
 829                         rcu_str_deref(dev_replace->srcdev->name),
 830                         dev_replace->srcdev->devid,
 831                         dev_replace->tgtdev ?
 832                         rcu_str_deref(dev_replace->tgtdev->name) :
 833                         "<missing target disk>",
 834                         (unsigned int)progress);
 835         }
 836         btrfs_dev_replace_continue_on_mount(fs_info);
 837         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
 838
 839         return 0;
 840 }
 841
 842 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
 843 {
 844         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 845         int ret;
 846
 847         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 848                               dev_replace->committed_cursor_left,
 849                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 850                               &dev_replace->scrub_progress, 0, 1);
 851         ret = btrfs_dev_replace_finishing(fs_info, ret);
 852         WARN_ON(ret);
 853         return 0;
 854 }
 855
 856 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 857 {
 858         if (!dev_replace->is_valid)
 859                 return 0;
 860
 861         switch (dev_replace->replace_state) {
 862         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 863         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 864         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 865                 return 0;
 866         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 867         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 868                 /*
 869                  * return true even if tgtdev is missing (this is
 870                  * something that can happen if the dev_replace
 871                  * procedure is suspended by an umount and then
 872                  * the tgtdev is missing (or "btrfs dev scan") was
 873                  * not called and the the filesystem is remounted
 874                  * in degraded state. This does not stop the
 875                  * dev_replace procedure. It needs to be canceled
 876                  * manually if the cancelation is wanted.
 877                  */
 878                 break;
 879         }
 880         return 1;
 881 }
 882
 883 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
 884 {
 885         /* the beginning is just an optimization for the typical case */
 886         if (atomic_read(&dev_replace->nesting_level) == 0) {
 887 acquire_lock:
 888                 /* this is not a nested case where the same thread
 889                  * is trying to acqurire the same lock twice */
 890                 mutex_lock(&dev_replace->lock);
 891                 mutex_lock(&dev_replace->lock_management_lock);
 892                 dev_replace->lock_owner = current->pid;
 893                 atomic_inc(&dev_replace->nesting_level);
 894                 mutex_unlock(&dev_replace->lock_management_lock);
 895                 return;
 896         }
 897
 898         mutex_lock(&dev_replace->lock_management_lock);
 899         if (atomic_read(&dev_replace->nesting_level) > 0 &&
 900             dev_replace->lock_owner == current->pid) {
 901                 WARN_ON(!mutex_is_locked(&dev_replace->lock));
 902                 atomic_inc(&dev_replace->nesting_level);
 903                 mutex_unlock(&dev_replace->lock_management_lock);
 904                 return;
 905         }
 906
 907         mutex_unlock(&dev_replace->lock_management_lock);
 908         goto acquire_lock;
 909 }
 910
 911 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
 912 {
 913         WARN_ON(!mutex_is_locked(&dev_replace->lock));
 914         mutex_lock(&dev_replace->lock_management_lock);
 915         WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
 916         WARN_ON(dev_replace->lock_owner != current->pid);
 917         atomic_dec(&dev_replace->nesting_level);
 918         if (atomic_read(&dev_replace->nesting_level) == 0) {
 919                 dev_replace->lock_owner = 0;
 920                 mutex_unlock(&dev_replace->lock_management_lock);
 921                 mutex_unlock(&dev_replace->lock);
 922         } else {
 923                 mutex_unlock(&dev_replace->lock_management_lock);
 924         }
 925 }
 926
 927 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 928 {
 929         percpu_counter_inc(&fs_info->bio_counter);
 930 }
 931
 932 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
 933 {
 934         percpu_counter_sub(&fs_info->bio_counter, amount);
 935
 936         if (waitqueue_active(&fs_info->replace_wait))
 937                 wake_up(&fs_info->replace_wait);
 938 }
 939
 940 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
 941 {
 942         while (1) {
 943                 percpu_counter_inc(&fs_info->bio_counter);
 944                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 945                                      &fs_info->fs_state)))
 946                         break;
 947
 948                 btrfs_bio_counter_dec(fs_info);
 949                 wait_event(fs_info->replace_wait,
 950                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 951                                      &fs_info->fs_state));
 952         }
 953 }