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