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