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_kthread(void *data);
  48 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
  49
  50
  51 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  52 {
  53         struct btrfs_key key;
  54         struct btrfs_root *dev_root = fs_info->dev_root;
  55         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  56         struct extent_buffer *eb;
  57         int slot;
  58         int ret = 0;
  59         struct btrfs_path *path = NULL;
  60         int item_size;
  61         struct btrfs_dev_replace_item *ptr;
  62         u64 src_devid;
  63
  64         path = btrfs_alloc_path();
  65         if (!path) {
  66                 ret = -ENOMEM;
  67                 goto out;
  68         }
  69
  70         key.objectid = 0;
  71         key.type = BTRFS_DEV_REPLACE_KEY;
  72         key.offset = 0;
  73         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  74         if (ret) {
  75 no_valid_dev_replace_entry_found:
  76                 ret = 0;
  77                 dev_replace->replace_state =
  78                         BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
  79                 dev_replace->cont_reading_from_srcdev_mode =
  80                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
  81                 dev_replace->replace_state = 0;
  82                 dev_replace->time_started = 0;
  83                 dev_replace->time_stopped = 0;
  84                 atomic64_set(&dev_replace->num_write_errors, 0);
  85                 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
  86                 dev_replace->cursor_left = 0;
  87                 dev_replace->committed_cursor_left = 0;
  88                 dev_replace->cursor_left_last_write_of_item = 0;
  89                 dev_replace->cursor_right = 0;
  90                 dev_replace->srcdev = NULL;
  91                 dev_replace->tgtdev = NULL;
  92                 dev_replace->is_valid = 0;
  93                 dev_replace->item_needs_writeback = 0;
  94                 goto out;
  95         }
  96         slot = path->slots[0];
  97         eb = path->nodes[0];
  98         item_size = btrfs_item_size_nr(eb, slot);
  99         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
 100
 101         if (item_size != sizeof(struct btrfs_dev_replace_item)) {
 102                 btrfs_warn(fs_info,
 103                         "dev_replace entry found has unexpected size, ignore entry");
 104                 goto no_valid_dev_replace_entry_found;
 105         }
 106
 107         src_devid = btrfs_dev_replace_src_devid(eb, ptr);
 108         dev_replace->cont_reading_from_srcdev_mode =
 109                 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
 110         dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
 111         dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
 112         dev_replace->time_stopped =
 113                 btrfs_dev_replace_time_stopped(eb, ptr);
 114         atomic64_set(&dev_replace->num_write_errors,
 115                      btrfs_dev_replace_num_write_errors(eb, ptr));
 116         atomic64_set(&dev_replace->num_uncorrectable_read_errors,
 117                      btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
 118         dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
 119         dev_replace->committed_cursor_left = dev_replace->cursor_left;
 120         dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
 121         dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
 122         dev_replace->is_valid = 1;
 123
 124         dev_replace->item_needs_writeback = 0;
 125         switch (dev_replace->replace_state) {
 126         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 127         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 128         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 129                 dev_replace->srcdev = NULL;
 130                 dev_replace->tgtdev = NULL;
 131                 break;
 132         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 133         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 134                 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
 135                                                         NULL, NULL);
 136                 dev_replace->tgtdev = btrfs_find_device(fs_info,
 137                                                         BTRFS_DEV_REPLACE_DEVID,
 138                                                         NULL, NULL);
 139                 /*
 140                  * allow 'btrfs dev replace_cancel' if src/tgt device is
 141                  * missing
 142                  */
 143                 if (!dev_replace->srcdev &&
 144                     !btrfs_test_opt(fs_info, DEGRADED)) {
 145                         ret = -EIO;
 146                         btrfs_warn(fs_info,
 147                            "cannot mount because device replace operation is ongoing and");
 148                         btrfs_warn(fs_info,
 149                            "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 150                            src_devid);
 151                 }
 152                 if (!dev_replace->tgtdev &&
 153                     !btrfs_test_opt(fs_info, DEGRADED)) {
 154                         ret = -EIO;
 155                         btrfs_warn(fs_info,
 156                            "cannot mount because device replace operation is ongoing and");
 157                         btrfs_warn(fs_info,
 158                            "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 159                                 BTRFS_DEV_REPLACE_DEVID);
 160                 }
 161                 if (dev_replace->tgtdev) {
 162                         if (dev_replace->srcdev) {
 163                                 dev_replace->tgtdev->total_bytes =
 164                                         dev_replace->srcdev->total_bytes;
 165                                 dev_replace->tgtdev->disk_total_bytes =
 166                                         dev_replace->srcdev->disk_total_bytes;
 167                                 dev_replace->tgtdev->commit_total_bytes =
 168                                         dev_replace->srcdev->commit_total_bytes;
 169                                 dev_replace->tgtdev->bytes_used =
 170                                         dev_replace->srcdev->bytes_used;
 171                                 dev_replace->tgtdev->commit_bytes_used =
 172                                         dev_replace->srcdev->commit_bytes_used;
 173                         }
 174                         set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 175                                 &dev_replace->tgtdev->dev_state);
 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,
 222                            "error %d while searching for dev_replace item!",
 223                            ret);
 224                 goto out;
 225         }
 226
 227         if (ret == 0 &&
 228             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 229                 /*
 230                  * need to delete old one and insert a new one.
 231                  * Since no attempt is made to recover any old state, if the
 232                  * dev_replace state is 'running', the data on the target
 233                  * drive is lost.
 234                  * It would be possible to recover the state: just make sure
 235                  * that the beginning of the item is never changed and always
 236                  * contains all the essential information. Then read this
 237                  * minimal set of information and use it as a base for the
 238                  * new state.
 239                  */
 240                 ret = btrfs_del_item(trans, dev_root, path);
 241                 if (ret != 0) {
 242                         btrfs_warn(fs_info,
 243                                    "delete too small dev_replace item failed %d!",
 244                                    ret);
 245                         goto out;
 246                 }
 247                 ret = 1;
 248         }
 249
 250         if (ret == 1) {
 251                 /* need to insert a new item */
 252                 btrfs_release_path(path);
 253                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 254                                               &key, sizeof(*ptr));
 255                 if (ret < 0) {
 256                         btrfs_warn(fs_info,
 257                                    "insert dev_replace item failed %d!", ret);
 258                         goto out;
 259                 }
 260         }
 261
 262         eb = path->nodes[0];
 263         ptr = btrfs_item_ptr(eb, path->slots[0],
 264                              struct btrfs_dev_replace_item);
 265
 266         btrfs_dev_replace_lock(dev_replace, 1);
 267         if (dev_replace->srcdev)
 268                 btrfs_set_dev_replace_src_devid(eb, ptr,
 269                         dev_replace->srcdev->devid);
 270         else
 271                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 272         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 273                 dev_replace->cont_reading_from_srcdev_mode);
 274         btrfs_set_dev_replace_replace_state(eb, ptr,
 275                 dev_replace->replace_state);
 276         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 277         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 278         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 279                 atomic64_read(&dev_replace->num_write_errors));
 280         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 281                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 282         dev_replace->cursor_left_last_write_of_item =
 283                 dev_replace->cursor_left;
 284         btrfs_set_dev_replace_cursor_left(eb, ptr,
 285                 dev_replace->cursor_left_last_write_of_item);
 286         btrfs_set_dev_replace_cursor_right(eb, ptr,
 287                 dev_replace->cursor_right);
 288         dev_replace->item_needs_writeback = 0;
 289         btrfs_dev_replace_unlock(dev_replace, 1);
 290
 291         btrfs_mark_buffer_dirty(eb);
 292
 293 out:
 294         btrfs_free_path(path);
 295
 296         return ret;
 297 }
 298
 299 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 300 {
 301         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 302
 303         dev_replace->committed_cursor_left =
 304                 dev_replace->cursor_left_last_write_of_item;
 305 }
 306
 307 static char* btrfs_dev_name(struct btrfs_device *device)
 308 {
 309         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 310                 return "<missing disk>";
 311         else
 312                 return rcu_str_deref(device->name);
 313 }
 314
 315 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 316                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 317                 int read_src)
 318 {
 319         struct btrfs_root *root = fs_info->dev_root;
 320         struct btrfs_trans_handle *trans;
 321         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 322         int ret;
 323         struct btrfs_device *tgt_device = NULL;
 324         struct btrfs_device *src_device = NULL;
 325
 326         /* the disk copy procedure reuses the scrub code */
 327         mutex_lock(&fs_info->volume_mutex);
 328         ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
 329                                             srcdev_name, &src_device);
 330         if (ret) {
 331                 mutex_unlock(&fs_info->volume_mutex);
 332                 return ret;
 333         }
 334
 335         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 336                                             src_device, &tgt_device);
 337         mutex_unlock(&fs_info->volume_mutex);
 338         if (ret)
 339                 return ret;
 340
 341         /*
 342          * Here we commit the transaction to make sure commit_total_bytes
 343          * of all the devices are updated.
 344          */
 345         trans = btrfs_attach_transaction(root);
 346         if (!IS_ERR(trans)) {
 347                 ret = btrfs_commit_transaction(trans);
 348                 if (ret)
 349                         return ret;
 350         } else if (PTR_ERR(trans) != -ENOENT) {
 351                 return PTR_ERR(trans);
 352         }
 353
 354         btrfs_dev_replace_lock(dev_replace, 1);
 355         switch (dev_replace->replace_state) {
 356         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 357         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 358         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 359                 break;
 360         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 361         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 362                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 363                 goto leave;
 364         }
 365
 366         dev_replace->cont_reading_from_srcdev_mode = read_src;
 367         WARN_ON(!src_device);
 368         dev_replace->srcdev = src_device;
 369         WARN_ON(!tgt_device);
 370         dev_replace->tgtdev = tgt_device;
 371
 372         btrfs_info_in_rcu(fs_info,
 373                       "dev_replace from %s (devid %llu) to %s started",
 374                       btrfs_dev_name(src_device),
 375                       src_device->devid,
 376                       rcu_str_deref(tgt_device->name));
 377
 378         /*
 379          * from now on, the writes to the srcdev are all duplicated to
 380          * go to the tgtdev as well (refer to btrfs_map_block()).
 381          */
 382         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 383         dev_replace->time_started = get_seconds();
 384         dev_replace->cursor_left = 0;
 385         dev_replace->committed_cursor_left = 0;
 386         dev_replace->cursor_left_last_write_of_item = 0;
 387         dev_replace->cursor_right = 0;
 388         dev_replace->is_valid = 1;
 389         dev_replace->item_needs_writeback = 1;
 390         atomic64_set(&dev_replace->num_write_errors, 0);
 391         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 392         btrfs_dev_replace_unlock(dev_replace, 1);
 393
 394         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 395         if (ret)
 396                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
 397
 398         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 399
 400         /* force writing the updated state information to disk */
 401         trans = btrfs_start_transaction(root, 0);
 402         if (IS_ERR(trans)) {
 403                 ret = PTR_ERR(trans);
 404                 btrfs_dev_replace_lock(dev_replace, 1);
 405                 goto leave;
 406         }
 407
 408         ret = btrfs_commit_transaction(trans);
 409         WARN_ON(ret);
 410
 411         /* the disk copy procedure reuses the scrub code */
 412         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 413                               btrfs_device_get_total_bytes(src_device),
 414                               &dev_replace->scrub_progress, 0, 1);
 415
 416         ret = btrfs_dev_replace_finishing(fs_info, ret);
 417         if (ret == -EINPROGRESS) {
 418                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 419         } else {
 420                 WARN_ON(ret);
 421         }
 422
 423         return ret;
 424
 425 leave:
 426         dev_replace->srcdev = NULL;
 427         dev_replace->tgtdev = NULL;
 428         btrfs_dev_replace_unlock(dev_replace, 1);
 429         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 430         return ret;
 431 }
 432
 433 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 434                             struct btrfs_ioctl_dev_replace_args *args)
 435 {
 436         int ret;
 437
 438         switch (args->start.cont_reading_from_srcdev_mode) {
 439         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 440         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 441                 break;
 442         default:
 443                 return -EINVAL;
 444         }
 445
 446         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 447             args->start.tgtdev_name[0] == '\0')
 448                 return -EINVAL;
 449
 450         ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
 451                                         args->start.srcdevid,
 452                                         args->start.srcdev_name,
 453                                         args->start.cont_reading_from_srcdev_mode);
 454         args->result = ret;
 455         /* don't warn if EINPROGRESS, someone else might be running scrub */
 456         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
 457                 ret = 0;
 458
 459         return ret;
 460 }
 461
 462 /*
 463  * blocked until all in-flight bios operations are finished.
 464  */
 465 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 466 {
 467         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 468         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 469                    &fs_info->bio_counter));
 470 }
 471
 472 /*
 473  * we have removed target device, it is safe to allow new bios request.
 474  */
 475 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 476 {
 477         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 478         wake_up(&fs_info->replace_wait);
 479 }
 480
 481 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 482                                        int scrub_ret)
 483 {
 484         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 485         struct btrfs_device *tgt_device;
 486         struct btrfs_device *src_device;
 487         struct btrfs_root *root = fs_info->tree_root;
 488         u8 uuid_tmp[BTRFS_UUID_SIZE];
 489         struct btrfs_trans_handle *trans;
 490         int ret = 0;
 491
 492         /* don't allow cancel or unmount to disturb the finishing procedure */
 493         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 494
 495         btrfs_dev_replace_lock(dev_replace, 0);
 496         /* was the operation canceled, or is it finished? */
 497         if (dev_replace->replace_state !=
 498             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 499                 btrfs_dev_replace_unlock(dev_replace, 0);
 500                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 501                 return 0;
 502         }
 503
 504         tgt_device = dev_replace->tgtdev;
 505         src_device = dev_replace->srcdev;
 506         btrfs_dev_replace_unlock(dev_replace, 0);
 507
 508         /*
 509          * flush all outstanding I/O and inode extent mappings before the
 510          * copy operation is declared as being finished
 511          */
 512         ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
 513         if (ret) {
 514                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 515                 return ret;
 516         }
 517         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 518
 519         trans = btrfs_start_transaction(root, 0);
 520         if (IS_ERR(trans)) {
 521                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 522                 return PTR_ERR(trans);
 523         }
 524         ret = btrfs_commit_transaction(trans);
 525         WARN_ON(ret);
 526
 527         mutex_lock(&uuid_mutex);
 528         /* keep away write_all_supers() during the finishing procedure */
 529         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 530         mutex_lock(&fs_info->chunk_mutex);
 531         btrfs_dev_replace_lock(dev_replace, 1);
 532         dev_replace->replace_state =
 533                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 534                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 535         dev_replace->tgtdev = NULL;
 536         dev_replace->srcdev = NULL;
 537         dev_replace->time_stopped = get_seconds();
 538         dev_replace->item_needs_writeback = 1;
 539
 540         /* replace old device with new one in mapping tree */
 541         if (!scrub_ret) {
 542                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 543                                                                 src_device,
 544                                                                 tgt_device);
 545         } else {
 546                 btrfs_err_in_rcu(fs_info,
 547                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 548                                  btrfs_dev_name(src_device),
 549                                  src_device->devid,
 550                                  rcu_str_deref(tgt_device->name), scrub_ret);
 551                 btrfs_dev_replace_unlock(dev_replace, 1);
 552                 mutex_unlock(&fs_info->chunk_mutex);
 553                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 554                 mutex_unlock(&uuid_mutex);
 555                 btrfs_rm_dev_replace_blocked(fs_info);
 556                 if (tgt_device)
 557                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 558                 btrfs_rm_dev_replace_unblocked(fs_info);
 559                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 560
 561                 return scrub_ret;
 562         }
 563
 564         btrfs_info_in_rcu(fs_info,
 565                           "dev_replace from %s (devid %llu) to %s finished",
 566                           btrfs_dev_name(src_device),
 567                           src_device->devid,
 568                           rcu_str_deref(tgt_device->name));
 569         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 570         tgt_device->devid = src_device->devid;
 571         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 572         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 573         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 574         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 575         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 576         btrfs_device_set_disk_total_bytes(tgt_device,
 577                                           src_device->disk_total_bytes);
 578         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 579         ASSERT(list_empty(&src_device->resized_list));
 580         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 581         tgt_device->commit_bytes_used = src_device->bytes_used;
 582
 583         btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
 584
 585         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 586         fs_info->fs_devices->rw_devices++;
 587
 588         btrfs_dev_replace_unlock(dev_replace, 1);
 589
 590         btrfs_rm_dev_replace_blocked(fs_info);
 591
 592         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 593
 594         btrfs_rm_dev_replace_unblocked(fs_info);
 595
 596         /*
 597          * this is again a consistent state where no dev_replace procedure
 598          * is running, the target device is part of the filesystem, the
 599          * source device is not part of the filesystem anymore and its 1st
 600          * superblock is scratched out so that it is no longer marked to
 601          * belong to this filesystem.
 602          */
 603         mutex_unlock(&fs_info->chunk_mutex);
 604         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 605         mutex_unlock(&uuid_mutex);
 606
 607         /* replace the sysfs entry */
 608         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 609         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 610
 611         /* write back the superblocks */
 612         trans = btrfs_start_transaction(root, 0);
 613         if (!IS_ERR(trans))
 614                 btrfs_commit_transaction(trans);
 615
 616         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 617
 618         return 0;
 619 }
 620
 621 static void btrfs_dev_replace_update_device_in_mapping_tree(
 622                                                 struct btrfs_fs_info *fs_info,
 623                                                 struct btrfs_device *srcdev,
 624                                                 struct btrfs_device *tgtdev)
 625 {
 626         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 627         struct extent_map *em;
 628         struct map_lookup *map;
 629         u64 start = 0;
 630         int i;
 631
 632         write_lock(&em_tree->lock);
 633         do {
 634                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 635                 if (!em)
 636                         break;
 637                 map = em->map_lookup;
 638                 for (i = 0; i < map->num_stripes; i++)
 639                         if (srcdev == map->stripes[i].dev)
 640                                 map->stripes[i].dev = tgtdev;
 641                 start = em->start + em->len;
 642                 free_extent_map(em);
 643         } while (start);
 644         write_unlock(&em_tree->lock);
 645 }
 646
 647 /*
 648  * Read progress of device replace status according to the state and last
 649  * stored position. The value format is the same as for
 650  * btrfs_dev_replace::progress_1000
 651  */
 652 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 653 {
 654         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 655         u64 ret = 0;
 656
 657         switch (dev_replace->replace_state) {
 658         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 659         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 660                 ret = 0;
 661                 break;
 662         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 663                 ret = 1000;
 664                 break;
 665         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 666         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 667                 ret = div64_u64(dev_replace->cursor_left,
 668                                 div_u64(btrfs_device_get_total_bytes(
 669                                                 dev_replace->srcdev), 1000));
 670                 break;
 671         }
 672
 673         return ret;
 674 }
 675
 676 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 677                               struct btrfs_ioctl_dev_replace_args *args)
 678 {
 679         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 680
 681         btrfs_dev_replace_lock(dev_replace, 0);
 682         /* even if !dev_replace_is_valid, the values are good enough for
 683          * the replace_status ioctl */
 684         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 685         args->status.replace_state = dev_replace->replace_state;
 686         args->status.time_started = dev_replace->time_started;
 687         args->status.time_stopped = dev_replace->time_stopped;
 688         args->status.num_write_errors =
 689                 atomic64_read(&dev_replace->num_write_errors);
 690         args->status.num_uncorrectable_read_errors =
 691                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 692         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 693         btrfs_dev_replace_unlock(dev_replace, 0);
 694 }
 695
 696 u64 btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 697 {
 698         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 699         struct btrfs_device *tgt_device = NULL;
 700         struct btrfs_trans_handle *trans;
 701         struct btrfs_root *root = fs_info->tree_root;
 702         u64 result;
 703         int ret;
 704
 705         if (sb_rdonly(fs_info->sb))
 706                 return -EROFS;
 707
 708         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 709         btrfs_dev_replace_lock(dev_replace, 1);
 710         switch (dev_replace->replace_state) {
 711         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 712         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 713         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 714                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 715                 btrfs_dev_replace_unlock(dev_replace, 1);
 716                 goto leave;
 717         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 718         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 719                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 720                 tgt_device = dev_replace->tgtdev;
 721                 dev_replace->tgtdev = NULL;
 722                 dev_replace->srcdev = NULL;
 723                 break;
 724         }
 725         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 726         dev_replace->time_stopped = get_seconds();
 727         dev_replace->item_needs_writeback = 1;
 728         btrfs_dev_replace_unlock(dev_replace, 1);
 729         btrfs_scrub_cancel(fs_info);
 730
 731         trans = btrfs_start_transaction(root, 0);
 732         if (IS_ERR(trans)) {
 733                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 734                 return PTR_ERR(trans);
 735         }
 736         ret = btrfs_commit_transaction(trans);
 737         WARN_ON(ret);
 738         if (tgt_device)
 739                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 740
 741 leave:
 742         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 743         return result;
 744 }
 745
 746 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 747 {
 748         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 749
 750         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 751         btrfs_dev_replace_lock(dev_replace, 1);
 752         switch (dev_replace->replace_state) {
 753         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 754         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 755         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 756         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 757                 break;
 758         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 759                 dev_replace->replace_state =
 760                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 761                 dev_replace->time_stopped = get_seconds();
 762                 dev_replace->item_needs_writeback = 1;
 763                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 764                 break;
 765         }
 766
 767         btrfs_dev_replace_unlock(dev_replace, 1);
 768         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 769 }
 770
 771 /* resume dev_replace procedure that was interrupted by unmount */
 772 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 773 {
 774         struct task_struct *task;
 775         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 776
 777         btrfs_dev_replace_lock(dev_replace, 1);
 778         switch (dev_replace->replace_state) {
 779         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 780         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 781         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 782                 btrfs_dev_replace_unlock(dev_replace, 1);
 783                 return 0;
 784         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 785                 break;
 786         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 787                 dev_replace->replace_state =
 788                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 789                 break;
 790         }
 791         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 792                 btrfs_info(fs_info,
 793                            "cannot continue dev_replace, tgtdev is missing");
 794                 btrfs_info(fs_info,
 795                            "you may cancel the operation after 'mount -o degraded'");
 796                 btrfs_dev_replace_unlock(dev_replace, 1);
 797                 return 0;
 798         }
 799         btrfs_dev_replace_unlock(dev_replace, 1);
 800
 801         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
 802         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 803         return PTR_ERR_OR_ZERO(task);
 804 }
 805
 806 static int btrfs_dev_replace_kthread(void *data)
 807 {
 808         struct btrfs_fs_info *fs_info = data;
 809         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 810         u64 progress;
 811
 812         progress = btrfs_dev_replace_progress(fs_info);
 813         progress = div_u64(progress, 10);
 814         btrfs_info_in_rcu(fs_info,
 815                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
 816                 btrfs_dev_name(dev_replace->srcdev),
 817                 dev_replace->srcdev->devid,
 818                 btrfs_dev_name(dev_replace->tgtdev),
 819                 (unsigned int)progress);
 820
 821         btrfs_dev_replace_continue_on_mount(fs_info);
 822         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 823
 824         return 0;
 825 }
 826
 827 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
 828 {
 829         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 830         int ret;
 831
 832         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 833                               dev_replace->committed_cursor_left,
 834                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 835                               &dev_replace->scrub_progress, 0, 1);
 836         ret = btrfs_dev_replace_finishing(fs_info, ret);
 837         WARN_ON(ret);
 838         return 0;
 839 }
 840
 841 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 842 {
 843         if (!dev_replace->is_valid)
 844                 return 0;
 845
 846         switch (dev_replace->replace_state) {
 847         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 848         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 849         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 850                 return 0;
 851         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 852         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 853                 /*
 854                  * return true even if tgtdev is missing (this is
 855                  * something that can happen if the dev_replace
 856                  * procedure is suspended by an umount and then
 857                  * the tgtdev is missing (or "btrfs dev scan") was
 858                  * not called and the the filesystem is remounted
 859                  * in degraded state. This does not stop the
 860                  * dev_replace procedure. It needs to be canceled
 861                  * manually if the cancellation is wanted.
 862                  */
 863                 break;
 864         }
 865         return 1;
 866 }
 867
 868 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
 869 {
 870         if (rw == 1) {
 871                 /* write */
 872 again:
 873                 wait_event(dev_replace->read_lock_wq,
 874                            atomic_read(&dev_replace->blocking_readers) == 0);
 875                 write_lock(&dev_replace->lock);
 876                 if (atomic_read(&dev_replace->blocking_readers)) {
 877                         write_unlock(&dev_replace->lock);
 878                         goto again;
 879                 }
 880         } else {
 881                 read_lock(&dev_replace->lock);
 882                 atomic_inc(&dev_replace->read_locks);
 883         }
 884 }
 885
 886 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
 887 {
 888         if (rw == 1) {
 889                 /* write */
 890                 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
 891                 write_unlock(&dev_replace->lock);
 892         } else {
 893                 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 894                 atomic_dec(&dev_replace->read_locks);
 895                 read_unlock(&dev_replace->lock);
 896         }
 897 }
 898
 899 /* inc blocking cnt and release read lock */
 900 void btrfs_dev_replace_set_lock_blocking(
 901                                         struct btrfs_dev_replace *dev_replace)
 902 {
 903         /* only set blocking for read lock */
 904         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 905         atomic_inc(&dev_replace->blocking_readers);
 906         read_unlock(&dev_replace->lock);
 907 }
 908
 909 /* acquire read lock and dec blocking cnt */
 910 void btrfs_dev_replace_clear_lock_blocking(
 911                                         struct btrfs_dev_replace *dev_replace)
 912 {
 913         /* only set blocking for read lock */
 914         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 915         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
 916         read_lock(&dev_replace->lock);
 917         if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
 918             waitqueue_active(&dev_replace->read_lock_wq))
 919                 wake_up(&dev_replace->read_lock_wq);
 920 }
 921
 922 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 923 {
 924         percpu_counter_inc(&fs_info->bio_counter);
 925 }
 926
 927 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
 928 {
 929         percpu_counter_sub(&fs_info->bio_counter, amount);
 930
 931         if (waitqueue_active(&fs_info->replace_wait))
 932                 wake_up(&fs_info->replace_wait);
 933 }
 934
 935 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
 936 {
 937         while (1) {
 938                 percpu_counter_inc(&fs_info->bio_counter);
 939                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 940                                      &fs_info->fs_state)))
 941                         break;
 942
 943                 btrfs_bio_counter_dec(fs_info);
 944                 wait_event(fs_info->replace_wait,
 945                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 946                                      &fs_info->fs_state));
 947         }
 948 }