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         if (path)
 187                 btrfs_free_path(path);
 188         return ret;
 189 }
 190
 191 /*
 192  * called from commit_transaction. Writes changed device replace state to
 193  * disk.
 194  */
 195 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 196                           struct btrfs_fs_info *fs_info)
 197 {
 198         int ret;
 199         struct btrfs_root *dev_root = fs_info->dev_root;
 200         struct btrfs_path *path;
 201         struct btrfs_key key;
 202         struct extent_buffer *eb;
 203         struct btrfs_dev_replace_item *ptr;
 204         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 205
 206         btrfs_dev_replace_lock(dev_replace);
 207         if (!dev_replace->is_valid ||
 208             !dev_replace->item_needs_writeback) {
 209                 btrfs_dev_replace_unlock(dev_replace);
 210                 return 0;
 211         }
 212         btrfs_dev_replace_unlock(dev_replace);
 213
 214         key.objectid = 0;
 215         key.type = BTRFS_DEV_REPLACE_KEY;
 216         key.offset = 0;
 217
 218         path = btrfs_alloc_path();
 219         if (!path) {
 220                 ret = -ENOMEM;
 221                 goto out;
 222         }
 223         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 224         if (ret < 0) {
 225                 btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
 226                         ret);
 227                 goto out;
 228         }
 229
 230         if (ret == 0 &&
 231             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 232                 /*
 233                  * need to delete old one and insert a new one.
 234                  * Since no attempt is made to recover any old state, if the
 235                  * dev_replace state is 'running', the data on the target
 236                  * drive is lost.
 237                  * It would be possible to recover the state: just make sure
 238                  * that the beginning of the item is never changed and always
 239                  * contains all the essential information. Then read this
 240                  * minimal set of information and use it as a base for the
 241                  * new state.
 242                  */
 243                 ret = btrfs_del_item(trans, dev_root, path);
 244                 if (ret != 0) {
 245                         btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
 246                                 ret);
 247                         goto out;
 248                 }
 249                 ret = 1;
 250         }
 251
 252         if (ret == 1) {
 253                 /* need to insert a new item */
 254                 btrfs_release_path(path);
 255                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 256                                               &key, sizeof(*ptr));
 257                 if (ret < 0) {
 258                         btrfs_warn(fs_info, "insert dev_replace item failed %d!",
 259                                 ret);
 260                         goto out;
 261                 }
 262         }
 263
 264         eb = path->nodes[0];
 265         ptr = btrfs_item_ptr(eb, path->slots[0],
 266                              struct btrfs_dev_replace_item);
 267
 268         btrfs_dev_replace_lock(dev_replace);
 269         if (dev_replace->srcdev)
 270                 btrfs_set_dev_replace_src_devid(eb, ptr,
 271                         dev_replace->srcdev->devid);
 272         else
 273                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 274         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 275                 dev_replace->cont_reading_from_srcdev_mode);
 276         btrfs_set_dev_replace_replace_state(eb, ptr,
 277                 dev_replace->replace_state);
 278         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 279         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 280         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 281                 atomic64_read(&dev_replace->num_write_errors));
 282         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 283                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 284         dev_replace->cursor_left_last_write_of_item =
 285                 dev_replace->cursor_left;
 286         btrfs_set_dev_replace_cursor_left(eb, ptr,
 287                 dev_replace->cursor_left_last_write_of_item);
 288         btrfs_set_dev_replace_cursor_right(eb, ptr,
 289                 dev_replace->cursor_right);
 290         dev_replace->item_needs_writeback = 0;
 291         btrfs_dev_replace_unlock(dev_replace);
 292
 293         btrfs_mark_buffer_dirty(eb);
 294
 295 out:
 296         btrfs_free_path(path);
 297
 298         return ret;
 299 }
 300
 301 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 302 {
 303         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 304
 305         dev_replace->committed_cursor_left =
 306                 dev_replace->cursor_left_last_write_of_item;
 307 }
 308
 309 int btrfs_dev_replace_start(struct btrfs_root *root,
 310                             struct btrfs_ioctl_dev_replace_args *args)
 311 {
 312         struct btrfs_trans_handle *trans;
 313         struct btrfs_fs_info *fs_info = root->fs_info;
 314         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 315         int ret;
 316         struct btrfs_device *tgt_device = NULL;
 317         struct btrfs_device *src_device = NULL;
 318
 319         if (btrfs_fs_incompat(fs_info, RAID56)) {
 320                 btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6");
 321                 return -EOPNOTSUPP;
 322         }
 323
 324         switch (args->start.cont_reading_from_srcdev_mode) {
 325         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 326         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 327                 break;
 328         default:
 329                 return -EINVAL;
 330         }
 331
 332         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 333             args->start.tgtdev_name[0] == '\0')
 334                 return -EINVAL;
 335
 336         /*
 337          * Here we commit the transaction to make sure commit_total_bytes
 338          * of all the devices are updated.
 339          */
 340         trans = btrfs_attach_transaction(root);
 341         if (!IS_ERR(trans)) {
 342                 ret = btrfs_commit_transaction(trans, root);
 343                 if (ret)
 344                         return ret;
 345         } else if (PTR_ERR(trans) != -ENOENT) {
 346                 return PTR_ERR(trans);
 347         }
 348
 349         /* the disk copy procedure reuses the scrub code */
 350         mutex_lock(&fs_info->volume_mutex);
 351         ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
 352                                             args->start.srcdev_name,
 353                                             &src_device);
 354         if (ret) {
 355                 mutex_unlock(&fs_info->volume_mutex);
 356                 return ret;
 357         }
 358
 359         ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
 360                                             src_device, &tgt_device);
 361         mutex_unlock(&fs_info->volume_mutex);
 362         if (ret)
 363                 return ret;
 364
 365         btrfs_dev_replace_lock(dev_replace);
 366         switch (dev_replace->replace_state) {
 367         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 368         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 369         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 370                 break;
 371         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 372         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 373                 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 374                 goto leave;
 375         }
 376
 377         dev_replace->cont_reading_from_srcdev_mode =
 378                 args->start.cont_reading_from_srcdev_mode;
 379         WARN_ON(!src_device);
 380         dev_replace->srcdev = src_device;
 381         WARN_ON(!tgt_device);
 382         dev_replace->tgtdev = tgt_device;
 383
 384         printk_in_rcu(KERN_INFO
 385                       "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
 386                       src_device->missing ? "<missing disk>" :
 387                         rcu_str_deref(src_device->name),
 388                       src_device->devid,
 389                       rcu_str_deref(tgt_device->name));
 390
 391         /*
 392          * from now on, the writes to the srcdev are all duplicated to
 393          * go to the tgtdev as well (refer to btrfs_map_block()).
 394          */
 395         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 396         dev_replace->time_started = get_seconds();
 397         dev_replace->cursor_left = 0;
 398         dev_replace->committed_cursor_left = 0;
 399         dev_replace->cursor_left_last_write_of_item = 0;
 400         dev_replace->cursor_right = 0;
 401         dev_replace->is_valid = 1;
 402         dev_replace->item_needs_writeback = 1;
 403         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 404         btrfs_dev_replace_unlock(dev_replace);
 405
 406         btrfs_wait_ordered_roots(root->fs_info, -1);
 407
 408         /* force writing the updated state information to disk */
 409         trans = btrfs_start_transaction(root, 0);
 410         if (IS_ERR(trans)) {
 411                 ret = PTR_ERR(trans);
 412                 btrfs_dev_replace_lock(dev_replace);
 413                 goto leave;
 414         }
 415
 416         ret = btrfs_commit_transaction(trans, root);
 417         WARN_ON(ret);
 418
 419         /* the disk copy procedure reuses the scrub code */
 420         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 421                               btrfs_device_get_total_bytes(src_device),
 422                               &dev_replace->scrub_progress, 0, 1);
 423
 424         ret = btrfs_dev_replace_finishing(root->fs_info, ret);
 425         /* don't warn if EINPROGRESS, someone else might be running scrub */
 426         if (ret == -EINPROGRESS) {
 427                 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 428                 ret = 0;
 429         } else {
 430                 WARN_ON(ret);
 431         }
 432
 433         return ret;
 434
 435 leave:
 436         dev_replace->srcdev = NULL;
 437         dev_replace->tgtdev = NULL;
 438         btrfs_dev_replace_unlock(dev_replace);
 439         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 440         return ret;
 441 }
 442
 443 /*
 444  * blocked until all flighting bios are finished.
 445  */
 446 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 447 {
 448         s64 writers;
 449         DEFINE_WAIT(wait);
 450
 451         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 452         do {
 453                 prepare_to_wait(&fs_info->replace_wait, &wait,
 454                                 TASK_UNINTERRUPTIBLE);
 455                 writers = percpu_counter_sum(&fs_info->bio_counter);
 456                 if (writers)
 457                         schedule();
 458                 finish_wait(&fs_info->replace_wait, &wait);
 459         } while (writers);
 460 }
 461
 462 /*
 463  * we have removed target device, it is safe to allow new bios request.
 464  */
 465 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 466 {
 467         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 468         if (waitqueue_active(&fs_info->replace_wait))
 469                 wake_up(&fs_info->replace_wait);
 470 }
 471
 472 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 473                                        int scrub_ret)
 474 {
 475         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 476         struct btrfs_device *tgt_device;
 477         struct btrfs_device *src_device;
 478         struct btrfs_root *root = fs_info->tree_root;
 479         u8 uuid_tmp[BTRFS_UUID_SIZE];
 480         struct btrfs_trans_handle *trans;
 481         int ret = 0;
 482
 483         /* don't allow cancel or unmount to disturb the finishing procedure */
 484         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 485
 486         btrfs_dev_replace_lock(dev_replace);
 487         /* was the operation canceled, or is it finished? */
 488         if (dev_replace->replace_state !=
 489             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 490                 btrfs_dev_replace_unlock(dev_replace);
 491                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 492                 return 0;
 493         }
 494
 495         tgt_device = dev_replace->tgtdev;
 496         src_device = dev_replace->srcdev;
 497         btrfs_dev_replace_unlock(dev_replace);
 498
 499         /*
 500          * flush all outstanding I/O and inode extent mappings before the
 501          * copy operation is declared as being finished
 502          */
 503         ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
 504         if (ret) {
 505                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 506                 return ret;
 507         }
 508         btrfs_wait_ordered_roots(root->fs_info, -1);
 509
 510         trans = btrfs_start_transaction(root, 0);
 511         if (IS_ERR(trans)) {
 512                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 513                 return PTR_ERR(trans);
 514         }
 515         ret = btrfs_commit_transaction(trans, root);
 516         WARN_ON(ret);
 517
 518         mutex_lock(&uuid_mutex);
 519         /* keep away write_all_supers() during the finishing procedure */
 520         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 521         mutex_lock(&root->fs_info->chunk_mutex);
 522         btrfs_dev_replace_lock(dev_replace);
 523         dev_replace->replace_state =
 524                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 525                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 526         dev_replace->tgtdev = NULL;
 527         dev_replace->srcdev = NULL;
 528         dev_replace->time_stopped = get_seconds();
 529         dev_replace->item_needs_writeback = 1;
 530
 531         /* replace old device with new one in mapping tree */
 532         if (!scrub_ret) {
 533                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 534                                                                 src_device,
 535                                                                 tgt_device);
 536         } else {
 537                 printk_in_rcu(KERN_ERR
 538                               "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
 539                               src_device->missing ? "<missing disk>" :
 540                                 rcu_str_deref(src_device->name),
 541                               src_device->devid,
 542                               rcu_str_deref(tgt_device->name), scrub_ret);
 543                 btrfs_dev_replace_unlock(dev_replace);
 544                 mutex_unlock(&root->fs_info->chunk_mutex);
 545                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 546                 mutex_unlock(&uuid_mutex);
 547                 if (tgt_device)
 548                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 549                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 550
 551                 return scrub_ret;
 552         }
 553
 554         printk_in_rcu(KERN_INFO
 555                       "BTRFS: dev_replace from %s (devid %llu) to %s finished\n",
 556                       src_device->missing ? "<missing disk>" :
 557                         rcu_str_deref(src_device->name),
 558                       src_device->devid,
 559                       rcu_str_deref(tgt_device->name));
 560         tgt_device->is_tgtdev_for_dev_replace = 0;
 561         tgt_device->devid = src_device->devid;
 562         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 563         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 564         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 565         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 566         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 567         btrfs_device_set_disk_total_bytes(tgt_device,
 568                                           src_device->disk_total_bytes);
 569         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 570         ASSERT(list_empty(&src_device->resized_list));
 571         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 572         tgt_device->commit_bytes_used = src_device->bytes_used;
 573         if (fs_info->sb->s_bdev == src_device->bdev)
 574                 fs_info->sb->s_bdev = tgt_device->bdev;
 575         if (fs_info->fs_devices->latest_bdev == src_device->bdev)
 576                 fs_info->fs_devices->latest_bdev = tgt_device->bdev;
 577         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 578         fs_info->fs_devices->rw_devices++;
 579
 580         btrfs_dev_replace_unlock(dev_replace);
 581
 582         btrfs_rm_dev_replace_blocked(fs_info);
 583
 584         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 585
 586         btrfs_rm_dev_replace_unblocked(fs_info);
 587
 588         /*
 589          * this is again a consistent state where no dev_replace procedure
 590          * is running, the target device is part of the filesystem, the
 591          * source device is not part of the filesystem anymore and its 1st
 592          * superblock is scratched out so that it is no longer marked to
 593          * belong to this filesystem.
 594          */
 595         mutex_unlock(&root->fs_info->chunk_mutex);
 596         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 597         mutex_unlock(&uuid_mutex);
 598
 599         /* replace the sysfs entry */
 600         btrfs_kobj_rm_device(fs_info, src_device);
 601         btrfs_kobj_add_device(fs_info, tgt_device);
 602         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 603
 604         /* write back the superblocks */
 605         trans = btrfs_start_transaction(root, 0);
 606         if (!IS_ERR(trans))
 607                 btrfs_commit_transaction(trans, root);
 608
 609         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 610
 611         return 0;
 612 }
 613
 614 static void btrfs_dev_replace_update_device_in_mapping_tree(
 615                                                 struct btrfs_fs_info *fs_info,
 616                                                 struct btrfs_device *srcdev,
 617                                                 struct btrfs_device *tgtdev)
 618 {
 619         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 620         struct extent_map *em;
 621         struct map_lookup *map;
 622         u64 start = 0;
 623         int i;
 624
 625         write_lock(&em_tree->lock);
 626         do {
 627                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 628                 if (!em)
 629                         break;
 630                 map = (struct map_lookup *)em->bdev;
 631                 for (i = 0; i < map->num_stripes; i++)
 632                         if (srcdev == map->stripes[i].dev)
 633                                 map->stripes[i].dev = tgtdev;
 634                 start = em->start + em->len;
 635                 free_extent_map(em);
 636         } while (start);
 637         write_unlock(&em_tree->lock);
 638 }
 639
 640 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
 641                                          char *srcdev_name,
 642                                          struct btrfs_device **device)
 643 {
 644         int ret;
 645
 646         if (srcdevid) {
 647                 ret = 0;
 648                 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
 649                                             NULL);
 650                 if (!*device)
 651                         ret = -ENOENT;
 652         } else {
 653                 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
 654                                                            device);
 655         }
 656         return ret;
 657 }
 658
 659 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 660                               struct btrfs_ioctl_dev_replace_args *args)
 661 {
 662         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 663         struct btrfs_device *srcdev;
 664
 665         btrfs_dev_replace_lock(dev_replace);
 666         /* even if !dev_replace_is_valid, the values are good enough for
 667          * the replace_status ioctl */
 668         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 669         args->status.replace_state = dev_replace->replace_state;
 670         args->status.time_started = dev_replace->time_started;
 671         args->status.time_stopped = dev_replace->time_stopped;
 672         args->status.num_write_errors =
 673                 atomic64_read(&dev_replace->num_write_errors);
 674         args->status.num_uncorrectable_read_errors =
 675                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 676         switch (dev_replace->replace_state) {
 677         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 678         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 679                 args->status.progress_1000 = 0;
 680                 break;
 681         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 682                 args->status.progress_1000 = 1000;
 683                 break;
 684         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 685         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 686                 srcdev = dev_replace->srcdev;
 687                 args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
 688                         div64_u64(btrfs_device_get_total_bytes(srcdev), 1000));
 689                 break;
 690         }
 691         btrfs_dev_replace_unlock(dev_replace);
 692 }
 693
 694 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
 695                              struct btrfs_ioctl_dev_replace_args *args)
 696 {
 697         args->result = __btrfs_dev_replace_cancel(fs_info);
 698         return 0;
 699 }
 700
 701 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 702 {
 703         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 704         struct btrfs_device *tgt_device = NULL;
 705         struct btrfs_trans_handle *trans;
 706         struct btrfs_root *root = fs_info->tree_root;
 707         u64 result;
 708         int ret;
 709
 710         if (fs_info->sb->s_flags & MS_RDONLY)
 711                 return -EROFS;
 712
 713         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 714         btrfs_dev_replace_lock(dev_replace);
 715         switch (dev_replace->replace_state) {
 716         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 717         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 718         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 719                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 720                 btrfs_dev_replace_unlock(dev_replace);
 721                 goto leave;
 722         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 723         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 724                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 725                 tgt_device = dev_replace->tgtdev;
 726                 dev_replace->tgtdev = NULL;
 727                 dev_replace->srcdev = NULL;
 728                 break;
 729         }
 730         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 731         dev_replace->time_stopped = get_seconds();
 732         dev_replace->item_needs_writeback = 1;
 733         btrfs_dev_replace_unlock(dev_replace);
 734         btrfs_scrub_cancel(fs_info);
 735
 736         trans = btrfs_start_transaction(root, 0);
 737         if (IS_ERR(trans)) {
 738                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 739                 return PTR_ERR(trans);
 740         }
 741         ret = btrfs_commit_transaction(trans, root);
 742         WARN_ON(ret);
 743         if (tgt_device)
 744                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 745
 746 leave:
 747         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 748         return result;
 749 }
 750
 751 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 752 {
 753         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 754
 755         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 756         btrfs_dev_replace_lock(dev_replace);
 757         switch (dev_replace->replace_state) {
 758         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 759         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 760         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 761         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 762                 break;
 763         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 764                 dev_replace->replace_state =
 765                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 766                 dev_replace->time_stopped = get_seconds();
 767                 dev_replace->item_needs_writeback = 1;
 768                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 769                 break;
 770         }
 771
 772         btrfs_dev_replace_unlock(dev_replace);
 773         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 774 }
 775
 776 /* resume dev_replace procedure that was interrupted by unmount */
 777 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 778 {
 779         struct task_struct *task;
 780         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 781
 782         btrfs_dev_replace_lock(dev_replace);
 783         switch (dev_replace->replace_state) {
 784         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 785         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 786         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 787                 btrfs_dev_replace_unlock(dev_replace);
 788                 return 0;
 789         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 790                 break;
 791         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 792                 dev_replace->replace_state =
 793                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 794                 break;
 795         }
 796         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 797                 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
 798                 btrfs_info(fs_info,
 799                         "you may cancel the operation after 'mount -o degraded'");
 800                 btrfs_dev_replace_unlock(dev_replace);
 801                 return 0;
 802         }
 803         btrfs_dev_replace_unlock(dev_replace);
 804
 805         WARN_ON(atomic_xchg(
 806                 &fs_info->mutually_exclusive_operation_running, 1));
 807         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 808         return PTR_ERR_OR_ZERO(task);
 809 }
 810
 811 static int btrfs_dev_replace_kthread(void *data)
 812 {
 813         struct btrfs_fs_info *fs_info = data;
 814         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 815         struct btrfs_ioctl_dev_replace_args *status_args;
 816         u64 progress;
 817
 818         status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
 819         if (status_args) {
 820                 btrfs_dev_replace_status(fs_info, status_args);
 821                 progress = status_args->status.progress_1000;
 822                 kfree(status_args);
 823                 do_div(progress, 10);
 824                 printk_in_rcu(KERN_INFO
 825                         "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
 826                         dev_replace->srcdev->missing ? "<missing disk>" :
 827                         rcu_str_deref(dev_replace->srcdev->name),
 828                         dev_replace->srcdev->devid,
 829                         dev_replace->tgtdev ?
 830                         rcu_str_deref(dev_replace->tgtdev->name) :
 831                         "<missing target disk>",
 832                         (unsigned int)progress);
 833         }
 834         btrfs_dev_replace_continue_on_mount(fs_info);
 835         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
 836
 837         return 0;
 838 }
 839
 840 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
 841 {
 842         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 843         int ret;
 844
 845         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 846                               dev_replace->committed_cursor_left,
 847                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 848                               &dev_replace->scrub_progress, 0, 1);
 849         ret = btrfs_dev_replace_finishing(fs_info, ret);
 850         WARN_ON(ret);
 851         return 0;
 852 }
 853
 854 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 855 {
 856         if (!dev_replace->is_valid)
 857                 return 0;
 858
 859         switch (dev_replace->replace_state) {
 860         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 861         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 862         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 863                 return 0;
 864         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 865         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 866                 /*
 867                  * return true even if tgtdev is missing (this is
 868                  * something that can happen if the dev_replace
 869                  * procedure is suspended by an umount and then
 870                  * the tgtdev is missing (or "btrfs dev scan") was
 871                  * not called and the the filesystem is remounted
 872                  * in degraded state. This does not stop the
 873                  * dev_replace procedure. It needs to be canceled
 874                  * manually if the cancelation is wanted.
 875                  */
 876                 break;
 877         }
 878         return 1;
 879 }
 880
 881 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
 882 {
 883         /* the beginning is just an optimization for the typical case */
 884         if (atomic_read(&dev_replace->nesting_level) == 0) {
 885 acquire_lock:
 886                 /* this is not a nested case where the same thread
 887                  * is trying to acqurire the same lock twice */
 888                 mutex_lock(&dev_replace->lock);
 889                 mutex_lock(&dev_replace->lock_management_lock);
 890                 dev_replace->lock_owner = current->pid;
 891                 atomic_inc(&dev_replace->nesting_level);
 892                 mutex_unlock(&dev_replace->lock_management_lock);
 893                 return;
 894         }
 895
 896         mutex_lock(&dev_replace->lock_management_lock);
 897         if (atomic_read(&dev_replace->nesting_level) > 0 &&
 898             dev_replace->lock_owner == current->pid) {
 899                 WARN_ON(!mutex_is_locked(&dev_replace->lock));
 900                 atomic_inc(&dev_replace->nesting_level);
 901                 mutex_unlock(&dev_replace->lock_management_lock);
 902                 return;
 903         }
 904
 905         mutex_unlock(&dev_replace->lock_management_lock);
 906         goto acquire_lock;
 907 }
 908
 909 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
 910 {
 911         WARN_ON(!mutex_is_locked(&dev_replace->lock));
 912         mutex_lock(&dev_replace->lock_management_lock);
 913         WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
 914         WARN_ON(dev_replace->lock_owner != current->pid);
 915         atomic_dec(&dev_replace->nesting_level);
 916         if (atomic_read(&dev_replace->nesting_level) == 0) {
 917                 dev_replace->lock_owner = 0;
 918                 mutex_unlock(&dev_replace->lock_management_lock);
 919                 mutex_unlock(&dev_replace->lock);
 920         } else {
 921                 mutex_unlock(&dev_replace->lock_management_lock);
 922         }
 923 }
 924
 925 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 926 {
 927         percpu_counter_inc(&fs_info->bio_counter);
 928 }
 929
 930 void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info)
 931 {
 932         percpu_counter_dec(&fs_info->bio_counter);
 933
 934         if (waitqueue_active(&fs_info->replace_wait))
 935                 wake_up(&fs_info->replace_wait);
 936 }
 937
 938 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
 939 {
 940         DEFINE_WAIT(wait);
 941 again:
 942         percpu_counter_inc(&fs_info->bio_counter);
 943         if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) {
 944                 btrfs_bio_counter_dec(fs_info);
 945                 wait_event(fs_info->replace_wait,
 946                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 947                                      &fs_info->fs_state));
 948                 goto again;
 949         }
 950
 951 }