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