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(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(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                         set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 176                                 &dev_replace->tgtdev->dev_state);
 177                         btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
 178                                 dev_replace->tgtdev);
 179                 }
 180                 break;
 181         }
 182
 183 out:
 184         btrfs_free_path(path);
 185         return ret;
 186 }
 187
 188 /*
 189  * called from commit_transaction. Writes changed device replace state to
 190  * disk.
 191  */
 192 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 193                           struct btrfs_fs_info *fs_info)
 194 {
 195         int ret;
 196         struct btrfs_root *dev_root = fs_info->dev_root;
 197         struct btrfs_path *path;
 198         struct btrfs_key key;
 199         struct extent_buffer *eb;
 200         struct btrfs_dev_replace_item *ptr;
 201         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 202
 203         btrfs_dev_replace_lock(dev_replace, 0);
 204         if (!dev_replace->is_valid ||
 205             !dev_replace->item_needs_writeback) {
 206                 btrfs_dev_replace_unlock(dev_replace, 0);
 207                 return 0;
 208         }
 209         btrfs_dev_replace_unlock(dev_replace, 0);
 210
 211         key.objectid = 0;
 212         key.type = BTRFS_DEV_REPLACE_KEY;
 213         key.offset = 0;
 214
 215         path = btrfs_alloc_path();
 216         if (!path) {
 217                 ret = -ENOMEM;
 218                 goto out;
 219         }
 220         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 221         if (ret < 0) {
 222                 btrfs_warn(fs_info,
 223                            "error %d while searching for dev_replace item!",
 224                            ret);
 225                 goto out;
 226         }
 227
 228         if (ret == 0 &&
 229             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 230                 /*
 231                  * need to delete old one and insert a new one.
 232                  * Since no attempt is made to recover any old state, if the
 233                  * dev_replace state is 'running', the data on the target
 234                  * drive is lost.
 235                  * It would be possible to recover the state: just make sure
 236                  * that the beginning of the item is never changed and always
 237                  * contains all the essential information. Then read this
 238                  * minimal set of information and use it as a base for the
 239                  * new state.
 240                  */
 241                 ret = btrfs_del_item(trans, dev_root, path);
 242                 if (ret != 0) {
 243                         btrfs_warn(fs_info,
 244                                    "delete too small dev_replace item failed %d!",
 245                                    ret);
 246                         goto out;
 247                 }
 248                 ret = 1;
 249         }
 250
 251         if (ret == 1) {
 252                 /* need to insert a new item */
 253                 btrfs_release_path(path);
 254                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 255                                               &key, sizeof(*ptr));
 256                 if (ret < 0) {
 257                         btrfs_warn(fs_info,
 258                                    "insert dev_replace item failed %d!", ret);
 259                         goto out;
 260                 }
 261         }
 262
 263         eb = path->nodes[0];
 264         ptr = btrfs_item_ptr(eb, path->slots[0],
 265                              struct btrfs_dev_replace_item);
 266
 267         btrfs_dev_replace_lock(dev_replace, 1);
 268         if (dev_replace->srcdev)
 269                 btrfs_set_dev_replace_src_devid(eb, ptr,
 270                         dev_replace->srcdev->devid);
 271         else
 272                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 273         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 274                 dev_replace->cont_reading_from_srcdev_mode);
 275         btrfs_set_dev_replace_replace_state(eb, ptr,
 276                 dev_replace->replace_state);
 277         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 278         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 279         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 280                 atomic64_read(&dev_replace->num_write_errors));
 281         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 282                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 283         dev_replace->cursor_left_last_write_of_item =
 284                 dev_replace->cursor_left;
 285         btrfs_set_dev_replace_cursor_left(eb, ptr,
 286                 dev_replace->cursor_left_last_write_of_item);
 287         btrfs_set_dev_replace_cursor_right(eb, ptr,
 288                 dev_replace->cursor_right);
 289         dev_replace->item_needs_writeback = 0;
 290         btrfs_dev_replace_unlock(dev_replace, 1);
 291
 292         btrfs_mark_buffer_dirty(eb);
 293
 294 out:
 295         btrfs_free_path(path);
 296
 297         return ret;
 298 }
 299
 300 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 301 {
 302         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 303
 304         dev_replace->committed_cursor_left =
 305                 dev_replace->cursor_left_last_write_of_item;
 306 }
 307
 308 static char* btrfs_dev_name(struct btrfs_device *device)
 309 {
 310         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 311                 return "<missing disk>";
 312         else
 313                 return rcu_str_deref(device->name);
 314 }
 315
 316 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 317                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 318                 int read_src)
 319 {
 320         struct btrfs_root *root = fs_info->dev_root;
 321         struct btrfs_trans_handle *trans;
 322         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 323         int ret;
 324         struct btrfs_device *tgt_device = NULL;
 325         struct btrfs_device *src_device = NULL;
 326
 327         /* the disk copy procedure reuses the scrub code */
 328         mutex_lock(&fs_info->volume_mutex);
 329         ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
 330                                             srcdev_name, &src_device);
 331         if (ret) {
 332                 mutex_unlock(&fs_info->volume_mutex);
 333                 return ret;
 334         }
 335
 336         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 337                                             src_device, &tgt_device);
 338         mutex_unlock(&fs_info->volume_mutex);
 339         if (ret)
 340                 return ret;
 341
 342         /*
 343          * Here we commit the transaction to make sure commit_total_bytes
 344          * of all the devices are updated.
 345          */
 346         trans = btrfs_attach_transaction(root);
 347         if (!IS_ERR(trans)) {
 348                 ret = btrfs_commit_transaction(trans);
 349                 if (ret)
 350                         return ret;
 351         } else if (PTR_ERR(trans) != -ENOENT) {
 352                 return PTR_ERR(trans);
 353         }
 354
 355         btrfs_dev_replace_lock(dev_replace, 1);
 356         switch (dev_replace->replace_state) {
 357         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 358         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 359         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 360                 break;
 361         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 362         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 363                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 364                 goto leave;
 365         }
 366
 367         dev_replace->cont_reading_from_srcdev_mode = read_src;
 368         WARN_ON(!src_device);
 369         dev_replace->srcdev = src_device;
 370         WARN_ON(!tgt_device);
 371         dev_replace->tgtdev = tgt_device;
 372
 373         btrfs_info_in_rcu(fs_info,
 374                       "dev_replace from %s (devid %llu) to %s started",
 375                       btrfs_dev_name(src_device),
 376                       src_device->devid,
 377                       rcu_str_deref(tgt_device->name));
 378
 379         /*
 380          * from now on, the writes to the srcdev are all duplicated to
 381          * go to the tgtdev as well (refer to btrfs_map_block()).
 382          */
 383         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 384         dev_replace->time_started = get_seconds();
 385         dev_replace->cursor_left = 0;
 386         dev_replace->committed_cursor_left = 0;
 387         dev_replace->cursor_left_last_write_of_item = 0;
 388         dev_replace->cursor_right = 0;
 389         dev_replace->is_valid = 1;
 390         dev_replace->item_needs_writeback = 1;
 391         atomic64_set(&dev_replace->num_write_errors, 0);
 392         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 393         btrfs_dev_replace_unlock(dev_replace, 1);
 394
 395         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 396         if (ret)
 397                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
 398
 399         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 400
 401         /* force writing the updated state information to disk */
 402         trans = btrfs_start_transaction(root, 0);
 403         if (IS_ERR(trans)) {
 404                 ret = PTR_ERR(trans);
 405                 btrfs_dev_replace_lock(dev_replace, 1);
 406                 goto leave;
 407         }
 408
 409         ret = btrfs_commit_transaction(trans);
 410         WARN_ON(ret);
 411
 412         /* the disk copy procedure reuses the scrub code */
 413         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 414                               btrfs_device_get_total_bytes(src_device),
 415                               &dev_replace->scrub_progress, 0, 1);
 416
 417         ret = btrfs_dev_replace_finishing(fs_info, ret);
 418         if (ret == -EINPROGRESS) {
 419                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 420         } else {
 421                 WARN_ON(ret);
 422         }
 423
 424         return ret;
 425
 426 leave:
 427         dev_replace->srcdev = NULL;
 428         dev_replace->tgtdev = NULL;
 429         btrfs_dev_replace_unlock(dev_replace, 1);
 430         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 431         return ret;
 432 }
 433
 434 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 435                             struct btrfs_ioctl_dev_replace_args *args)
 436 {
 437         int ret;
 438
 439         switch (args->start.cont_reading_from_srcdev_mode) {
 440         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 441         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 442                 break;
 443         default:
 444                 return -EINVAL;
 445         }
 446
 447         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 448             args->start.tgtdev_name[0] == '\0')
 449                 return -EINVAL;
 450
 451         ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
 452                                         args->start.srcdevid,
 453                                         args->start.srcdev_name,
 454                                         args->start.cont_reading_from_srcdev_mode);
 455         args->result = ret;
 456         /* don't warn if EINPROGRESS, someone else might be running scrub */
 457         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
 458                 ret = 0;
 459
 460         return ret;
 461 }
 462
 463 /*
 464  * blocked until all in-flight bios operations are finished.
 465  */
 466 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 467 {
 468         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 469         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 470                    &fs_info->bio_counter));
 471 }
 472
 473 /*
 474  * we have removed target device, it is safe to allow new bios request.
 475  */
 476 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 477 {
 478         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 479         wake_up(&fs_info->replace_wait);
 480 }
 481
 482 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 483                                        int scrub_ret)
 484 {
 485         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 486         struct btrfs_device *tgt_device;
 487         struct btrfs_device *src_device;
 488         struct btrfs_root *root = fs_info->tree_root;
 489         u8 uuid_tmp[BTRFS_UUID_SIZE];
 490         struct btrfs_trans_handle *trans;
 491         int ret = 0;
 492
 493         /* don't allow cancel or unmount to disturb the finishing procedure */
 494         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 495
 496         btrfs_dev_replace_lock(dev_replace, 0);
 497         /* was the operation canceled, or is it finished? */
 498         if (dev_replace->replace_state !=
 499             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 500                 btrfs_dev_replace_unlock(dev_replace, 0);
 501                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 502                 return 0;
 503         }
 504
 505         tgt_device = dev_replace->tgtdev;
 506         src_device = dev_replace->srcdev;
 507         btrfs_dev_replace_unlock(dev_replace, 0);
 508
 509         /*
 510          * flush all outstanding I/O and inode extent mappings before the
 511          * copy operation is declared as being finished
 512          */
 513         ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
 514         if (ret) {
 515                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 516                 return ret;
 517         }
 518         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 519
 520         trans = btrfs_start_transaction(root, 0);
 521         if (IS_ERR(trans)) {
 522                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 523                 return PTR_ERR(trans);
 524         }
 525         ret = btrfs_commit_transaction(trans);
 526         WARN_ON(ret);
 527
 528         mutex_lock(&uuid_mutex);
 529         /* keep away write_all_supers() during the finishing procedure */
 530         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 531         mutex_lock(&fs_info->chunk_mutex);
 532         btrfs_dev_replace_lock(dev_replace, 1);
 533         dev_replace->replace_state =
 534                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 535                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 536         dev_replace->tgtdev = NULL;
 537         dev_replace->srcdev = NULL;
 538         dev_replace->time_stopped = get_seconds();
 539         dev_replace->item_needs_writeback = 1;
 540
 541         /* replace old device with new one in mapping tree */
 542         if (!scrub_ret) {
 543                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 544                                                                 src_device,
 545                                                                 tgt_device);
 546         } else {
 547                 btrfs_err_in_rcu(fs_info,
 548                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 549                                  btrfs_dev_name(src_device),
 550                                  src_device->devid,
 551                                  rcu_str_deref(tgt_device->name), scrub_ret);
 552                 btrfs_dev_replace_unlock(dev_replace, 1);
 553                 mutex_unlock(&fs_info->chunk_mutex);
 554                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 555                 mutex_unlock(&uuid_mutex);
 556                 btrfs_rm_dev_replace_blocked(fs_info);
 557                 if (tgt_device)
 558                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 559                 btrfs_rm_dev_replace_unblocked(fs_info);
 560                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 561
 562                 return scrub_ret;
 563         }
 564
 565         btrfs_info_in_rcu(fs_info,
 566                           "dev_replace from %s (devid %llu) to %s finished",
 567                           btrfs_dev_name(src_device),
 568                           src_device->devid,
 569                           rcu_str_deref(tgt_device->name));
 570         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 571         tgt_device->devid = src_device->devid;
 572         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 573         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 574         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 575         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 576         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 577         btrfs_device_set_disk_total_bytes(tgt_device,
 578                                           src_device->disk_total_bytes);
 579         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 580         ASSERT(list_empty(&src_device->resized_list));
 581         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 582         tgt_device->commit_bytes_used = src_device->bytes_used;
 583
 584         btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
 585
 586         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 587         fs_info->fs_devices->rw_devices++;
 588
 589         btrfs_dev_replace_unlock(dev_replace, 1);
 590
 591         btrfs_rm_dev_replace_blocked(fs_info);
 592
 593         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 594
 595         btrfs_rm_dev_replace_unblocked(fs_info);
 596
 597         /*
 598          * this is again a consistent state where no dev_replace procedure
 599          * is running, the target device is part of the filesystem, the
 600          * source device is not part of the filesystem anymore and its 1st
 601          * superblock is scratched out so that it is no longer marked to
 602          * belong to this filesystem.
 603          */
 604         mutex_unlock(&fs_info->chunk_mutex);
 605         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 606         mutex_unlock(&uuid_mutex);
 607
 608         /* replace the sysfs entry */
 609         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 610         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 611
 612         /* write back the superblocks */
 613         trans = btrfs_start_transaction(root, 0);
 614         if (!IS_ERR(trans))
 615                 btrfs_commit_transaction(trans);
 616
 617         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 618
 619         return 0;
 620 }
 621
 622 static void btrfs_dev_replace_update_device_in_mapping_tree(
 623                                                 struct btrfs_fs_info *fs_info,
 624                                                 struct btrfs_device *srcdev,
 625                                                 struct btrfs_device *tgtdev)
 626 {
 627         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 628         struct extent_map *em;
 629         struct map_lookup *map;
 630         u64 start = 0;
 631         int i;
 632
 633         write_lock(&em_tree->lock);
 634         do {
 635                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 636                 if (!em)
 637                         break;
 638                 map = em->map_lookup;
 639                 for (i = 0; i < map->num_stripes; i++)
 640                         if (srcdev == map->stripes[i].dev)
 641                                 map->stripes[i].dev = tgtdev;
 642                 start = em->start + em->len;
 643                 free_extent_map(em);
 644         } while (start);
 645         write_unlock(&em_tree->lock);
 646 }
 647
 648 /*
 649  * Read progress of device replace status according to the state and last
 650  * stored position. The value format is the same as for
 651  * btrfs_dev_replace::progress_1000
 652  */
 653 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 654 {
 655         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 656         u64 ret = 0;
 657
 658         switch (dev_replace->replace_state) {
 659         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 660         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 661                 ret = 0;
 662                 break;
 663         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 664                 ret = 1000;
 665                 break;
 666         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 667         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 668                 ret = div64_u64(dev_replace->cursor_left,
 669                                 div_u64(btrfs_device_get_total_bytes(
 670                                                 dev_replace->srcdev), 1000));
 671                 break;
 672         }
 673
 674         return ret;
 675 }
 676
 677 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 678                               struct btrfs_ioctl_dev_replace_args *args)
 679 {
 680         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 681
 682         btrfs_dev_replace_lock(dev_replace, 0);
 683         /* even if !dev_replace_is_valid, the values are good enough for
 684          * the replace_status ioctl */
 685         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 686         args->status.replace_state = dev_replace->replace_state;
 687         args->status.time_started = dev_replace->time_started;
 688         args->status.time_stopped = dev_replace->time_stopped;
 689         args->status.num_write_errors =
 690                 atomic64_read(&dev_replace->num_write_errors);
 691         args->status.num_uncorrectable_read_errors =
 692                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 693         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 694         btrfs_dev_replace_unlock(dev_replace, 0);
 695 }
 696
 697 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
 698                              struct btrfs_ioctl_dev_replace_args *args)
 699 {
 700         args->result = __btrfs_dev_replace_cancel(fs_info);
 701         return 0;
 702 }
 703
 704 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 705 {
 706         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 707         struct btrfs_device *tgt_device = NULL;
 708         struct btrfs_trans_handle *trans;
 709         struct btrfs_root *root = fs_info->tree_root;
 710         u64 result;
 711         int ret;
 712
 713         if (sb_rdonly(fs_info->sb))
 714                 return -EROFS;
 715
 716         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 717         btrfs_dev_replace_lock(dev_replace, 1);
 718         switch (dev_replace->replace_state) {
 719         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 720         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 721         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 722                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 723                 btrfs_dev_replace_unlock(dev_replace, 1);
 724                 goto leave;
 725         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 726         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 727                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 728                 tgt_device = dev_replace->tgtdev;
 729                 dev_replace->tgtdev = NULL;
 730                 dev_replace->srcdev = NULL;
 731                 break;
 732         }
 733         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 734         dev_replace->time_stopped = get_seconds();
 735         dev_replace->item_needs_writeback = 1;
 736         btrfs_dev_replace_unlock(dev_replace, 1);
 737         btrfs_scrub_cancel(fs_info);
 738
 739         trans = btrfs_start_transaction(root, 0);
 740         if (IS_ERR(trans)) {
 741                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 742                 return PTR_ERR(trans);
 743         }
 744         ret = btrfs_commit_transaction(trans);
 745         WARN_ON(ret);
 746         if (tgt_device)
 747                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 748
 749 leave:
 750         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 751         return result;
 752 }
 753
 754 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 755 {
 756         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 757
 758         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 759         btrfs_dev_replace_lock(dev_replace, 1);
 760         switch (dev_replace->replace_state) {
 761         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 762         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 763         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 764         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 765                 break;
 766         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 767                 dev_replace->replace_state =
 768                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 769                 dev_replace->time_stopped = get_seconds();
 770                 dev_replace->item_needs_writeback = 1;
 771                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 772                 break;
 773         }
 774
 775         btrfs_dev_replace_unlock(dev_replace, 1);
 776         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 777 }
 778
 779 /* resume dev_replace procedure that was interrupted by unmount */
 780 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 781 {
 782         struct task_struct *task;
 783         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 784
 785         btrfs_dev_replace_lock(dev_replace, 1);
 786         switch (dev_replace->replace_state) {
 787         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 788         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 789         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 790                 btrfs_dev_replace_unlock(dev_replace, 1);
 791                 return 0;
 792         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 793                 break;
 794         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 795                 dev_replace->replace_state =
 796                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 797                 break;
 798         }
 799         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 800                 btrfs_info(fs_info,
 801                            "cannot continue dev_replace, tgtdev is missing");
 802                 btrfs_info(fs_info,
 803                            "you may cancel the operation after 'mount -o degraded'");
 804                 btrfs_dev_replace_unlock(dev_replace, 1);
 805                 return 0;
 806         }
 807         btrfs_dev_replace_unlock(dev_replace, 1);
 808
 809         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
 810         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 811         return PTR_ERR_OR_ZERO(task);
 812 }
 813
 814 static int btrfs_dev_replace_kthread(void *data)
 815 {
 816         struct btrfs_fs_info *fs_info = data;
 817         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 818         u64 progress;
 819
 820         progress = btrfs_dev_replace_progress(fs_info);
 821         progress = div_u64(progress, 10);
 822         btrfs_info_in_rcu(fs_info,
 823                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
 824                 btrfs_dev_name(dev_replace->srcdev),
 825                 dev_replace->srcdev->devid,
 826                 btrfs_dev_name(dev_replace->tgtdev),
 827                 (unsigned int)progress);
 828
 829         btrfs_dev_replace_continue_on_mount(fs_info);
 830         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 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 cancellation is wanted.
 870                  */
 871                 break;
 872         }
 873         return 1;
 874 }
 875
 876 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
 877 {
 878         if (rw == 1) {
 879                 /* write */
 880 again:
 881                 wait_event(dev_replace->read_lock_wq,
 882                            atomic_read(&dev_replace->blocking_readers) == 0);
 883                 write_lock(&dev_replace->lock);
 884                 if (atomic_read(&dev_replace->blocking_readers)) {
 885                         write_unlock(&dev_replace->lock);
 886                         goto again;
 887                 }
 888         } else {
 889                 read_lock(&dev_replace->lock);
 890                 atomic_inc(&dev_replace->read_locks);
 891         }
 892 }
 893
 894 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
 895 {
 896         if (rw == 1) {
 897                 /* write */
 898                 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
 899                 write_unlock(&dev_replace->lock);
 900         } else {
 901                 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 902                 atomic_dec(&dev_replace->read_locks);
 903                 read_unlock(&dev_replace->lock);
 904         }
 905 }
 906
 907 /* inc blocking cnt and release read lock */
 908 void btrfs_dev_replace_set_lock_blocking(
 909                                         struct btrfs_dev_replace *dev_replace)
 910 {
 911         /* only set blocking for read lock */
 912         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 913         atomic_inc(&dev_replace->blocking_readers);
 914         read_unlock(&dev_replace->lock);
 915 }
 916
 917 /* acquire read lock and dec blocking cnt */
 918 void btrfs_dev_replace_clear_lock_blocking(
 919                                         struct btrfs_dev_replace *dev_replace)
 920 {
 921         /* only set blocking for read lock */
 922         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 923         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
 924         read_lock(&dev_replace->lock);
 925         if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
 926             waitqueue_active(&dev_replace->read_lock_wq))
 927                 wake_up(&dev_replace->read_lock_wq);
 928 }
 929
 930 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 931 {
 932         percpu_counter_inc(&fs_info->bio_counter);
 933 }
 934
 935 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
 936 {
 937         percpu_counter_sub(&fs_info->bio_counter, amount);
 938
 939         if (waitqueue_active(&fs_info->replace_wait))
 940                 wake_up(&fs_info->replace_wait);
 941 }
 942
 943 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
 944 {
 945         while (1) {
 946                 percpu_counter_inc(&fs_info->bio_counter);
 947                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 948                                      &fs_info->fs_state)))
 949                         break;
 950
 951                 btrfs_bio_counter_dec(fs_info);
 952                 wait_event(fs_info->replace_wait,
 953                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 954                                      &fs_info->fs_state));
 955         }
 956 }