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
 177                         WARN_ON(fs_info->fs_devices->rw_devices == 0);
 178                         dev_replace->tgtdev->io_width = fs_info->sectorsize;
 179                         dev_replace->tgtdev->io_align = fs_info->sectorsize;
 180                         dev_replace->tgtdev->sector_size = fs_info->sectorsize;
 181                         dev_replace->tgtdev->fs_info = fs_info;
 182                         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
 183                                 &dev_replace->tgtdev->dev_state);
 184                 }
 185                 break;
 186         }
 187
 188 out:
 189         btrfs_free_path(path);
 190         return ret;
 191 }
 192
 193 /*
 194  * called from commit_transaction. Writes changed device replace state to
 195  * disk.
 196  */
 197 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 198                           struct btrfs_fs_info *fs_info)
 199 {
 200         int ret;
 201         struct btrfs_root *dev_root = fs_info->dev_root;
 202         struct btrfs_path *path;
 203         struct btrfs_key key;
 204         struct extent_buffer *eb;
 205         struct btrfs_dev_replace_item *ptr;
 206         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 207
 208         btrfs_dev_replace_lock(dev_replace, 0);
 209         if (!dev_replace->is_valid ||
 210             !dev_replace->item_needs_writeback) {
 211                 btrfs_dev_replace_unlock(dev_replace, 0);
 212                 return 0;
 213         }
 214         btrfs_dev_replace_unlock(dev_replace, 0);
 215
 216         key.objectid = 0;
 217         key.type = BTRFS_DEV_REPLACE_KEY;
 218         key.offset = 0;
 219
 220         path = btrfs_alloc_path();
 221         if (!path) {
 222                 ret = -ENOMEM;
 223                 goto out;
 224         }
 225         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 226         if (ret < 0) {
 227                 btrfs_warn(fs_info,
 228                            "error %d while searching for dev_replace item!",
 229                            ret);
 230                 goto out;
 231         }
 232
 233         if (ret == 0 &&
 234             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 235                 /*
 236                  * need to delete old one and insert a new one.
 237                  * Since no attempt is made to recover any old state, if the
 238                  * dev_replace state is 'running', the data on the target
 239                  * drive is lost.
 240                  * It would be possible to recover the state: just make sure
 241                  * that the beginning of the item is never changed and always
 242                  * contains all the essential information. Then read this
 243                  * minimal set of information and use it as a base for the
 244                  * new state.
 245                  */
 246                 ret = btrfs_del_item(trans, dev_root, path);
 247                 if (ret != 0) {
 248                         btrfs_warn(fs_info,
 249                                    "delete too small dev_replace item failed %d!",
 250                                    ret);
 251                         goto out;
 252                 }
 253                 ret = 1;
 254         }
 255
 256         if (ret == 1) {
 257                 /* need to insert a new item */
 258                 btrfs_release_path(path);
 259                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 260                                               &key, sizeof(*ptr));
 261                 if (ret < 0) {
 262                         btrfs_warn(fs_info,
 263                                    "insert dev_replace item failed %d!", ret);
 264                         goto out;
 265                 }
 266         }
 267
 268         eb = path->nodes[0];
 269         ptr = btrfs_item_ptr(eb, path->slots[0],
 270                              struct btrfs_dev_replace_item);
 271
 272         btrfs_dev_replace_lock(dev_replace, 1);
 273         if (dev_replace->srcdev)
 274                 btrfs_set_dev_replace_src_devid(eb, ptr,
 275                         dev_replace->srcdev->devid);
 276         else
 277                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 278         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 279                 dev_replace->cont_reading_from_srcdev_mode);
 280         btrfs_set_dev_replace_replace_state(eb, ptr,
 281                 dev_replace->replace_state);
 282         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 283         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 284         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 285                 atomic64_read(&dev_replace->num_write_errors));
 286         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 287                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 288         dev_replace->cursor_left_last_write_of_item =
 289                 dev_replace->cursor_left;
 290         btrfs_set_dev_replace_cursor_left(eb, ptr,
 291                 dev_replace->cursor_left_last_write_of_item);
 292         btrfs_set_dev_replace_cursor_right(eb, ptr,
 293                 dev_replace->cursor_right);
 294         dev_replace->item_needs_writeback = 0;
 295         btrfs_dev_replace_unlock(dev_replace, 1);
 296
 297         btrfs_mark_buffer_dirty(eb);
 298
 299 out:
 300         btrfs_free_path(path);
 301
 302         return ret;
 303 }
 304
 305 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 306 {
 307         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 308
 309         dev_replace->committed_cursor_left =
 310                 dev_replace->cursor_left_last_write_of_item;
 311 }
 312
 313 static char* btrfs_dev_name(struct btrfs_device *device)
 314 {
 315         if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 316                 return "<missing disk>";
 317         else
 318                 return rcu_str_deref(device->name);
 319 }
 320
 321 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 322                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 323                 int read_src)
 324 {
 325         struct btrfs_root *root = fs_info->dev_root;
 326         struct btrfs_trans_handle *trans;
 327         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 328         int ret;
 329         struct btrfs_device *tgt_device = NULL;
 330         struct btrfs_device *src_device = NULL;
 331
 332         /* the disk copy procedure reuses the scrub code */
 333         mutex_lock(&fs_info->volume_mutex);
 334         ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
 335                                             srcdev_name, &src_device);
 336         if (ret) {
 337                 mutex_unlock(&fs_info->volume_mutex);
 338                 return ret;
 339         }
 340
 341         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 342                                             src_device, &tgt_device);
 343         mutex_unlock(&fs_info->volume_mutex);
 344         if (ret)
 345                 return ret;
 346
 347         /*
 348          * Here we commit the transaction to make sure commit_total_bytes
 349          * of all the devices are updated.
 350          */
 351         trans = btrfs_attach_transaction(root);
 352         if (!IS_ERR(trans)) {
 353                 ret = btrfs_commit_transaction(trans);
 354                 if (ret)
 355                         return ret;
 356         } else if (PTR_ERR(trans) != -ENOENT) {
 357                 return PTR_ERR(trans);
 358         }
 359
 360         btrfs_dev_replace_lock(dev_replace, 1);
 361         switch (dev_replace->replace_state) {
 362         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 363         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 364         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 365                 break;
 366         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 367         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 368                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 369                 goto leave;
 370         }
 371
 372         dev_replace->cont_reading_from_srcdev_mode = read_src;
 373         WARN_ON(!src_device);
 374         dev_replace->srcdev = src_device;
 375         WARN_ON(!tgt_device);
 376         dev_replace->tgtdev = tgt_device;
 377
 378         btrfs_info_in_rcu(fs_info,
 379                       "dev_replace from %s (devid %llu) to %s started",
 380                       btrfs_dev_name(src_device),
 381                       src_device->devid,
 382                       rcu_str_deref(tgt_device->name));
 383
 384         /*
 385          * from now on, the writes to the srcdev are all duplicated to
 386          * go to the tgtdev as well (refer to btrfs_map_block()).
 387          */
 388         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 389         dev_replace->time_started = get_seconds();
 390         dev_replace->cursor_left = 0;
 391         dev_replace->committed_cursor_left = 0;
 392         dev_replace->cursor_left_last_write_of_item = 0;
 393         dev_replace->cursor_right = 0;
 394         dev_replace->is_valid = 1;
 395         dev_replace->item_needs_writeback = 1;
 396         atomic64_set(&dev_replace->num_write_errors, 0);
 397         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 398         btrfs_dev_replace_unlock(dev_replace, 1);
 399
 400         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 401         if (ret)
 402                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
 403
 404         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 405
 406         /* force writing the updated state information to disk */
 407         trans = btrfs_start_transaction(root, 0);
 408         if (IS_ERR(trans)) {
 409                 ret = PTR_ERR(trans);
 410                 btrfs_dev_replace_lock(dev_replace, 1);
 411                 goto leave;
 412         }
 413
 414         ret = btrfs_commit_transaction(trans);
 415         WARN_ON(ret);
 416
 417         /* the disk copy procedure reuses the scrub code */
 418         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 419                               btrfs_device_get_total_bytes(src_device),
 420                               &dev_replace->scrub_progress, 0, 1);
 421
 422         ret = btrfs_dev_replace_finishing(fs_info, ret);
 423         if (ret == -EINPROGRESS) {
 424                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 425         } else {
 426                 WARN_ON(ret);
 427         }
 428
 429         return ret;
 430
 431 leave:
 432         dev_replace->srcdev = NULL;
 433         dev_replace->tgtdev = NULL;
 434         btrfs_dev_replace_unlock(dev_replace, 1);
 435         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 436         return ret;
 437 }
 438
 439 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 440                             struct btrfs_ioctl_dev_replace_args *args)
 441 {
 442         int ret;
 443
 444         switch (args->start.cont_reading_from_srcdev_mode) {
 445         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 446         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 447                 break;
 448         default:
 449                 return -EINVAL;
 450         }
 451
 452         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 453             args->start.tgtdev_name[0] == '\0')
 454                 return -EINVAL;
 455
 456         ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
 457                                         args->start.srcdevid,
 458                                         args->start.srcdev_name,
 459                                         args->start.cont_reading_from_srcdev_mode);
 460         args->result = ret;
 461         /* don't warn if EINPROGRESS, someone else might be running scrub */
 462         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
 463                 ret = 0;
 464
 465         return ret;
 466 }
 467
 468 /*
 469  * blocked until all in-flight bios operations are finished.
 470  */
 471 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 472 {
 473         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 474         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 475                    &fs_info->bio_counter));
 476 }
 477
 478 /*
 479  * we have removed target device, it is safe to allow new bios request.
 480  */
 481 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 482 {
 483         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 484         wake_up(&fs_info->replace_wait);
 485 }
 486
 487 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 488                                        int scrub_ret)
 489 {
 490         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 491         struct btrfs_device *tgt_device;
 492         struct btrfs_device *src_device;
 493         struct btrfs_root *root = fs_info->tree_root;
 494         u8 uuid_tmp[BTRFS_UUID_SIZE];
 495         struct btrfs_trans_handle *trans;
 496         int ret = 0;
 497
 498         /* don't allow cancel or unmount to disturb the finishing procedure */
 499         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 500
 501         btrfs_dev_replace_lock(dev_replace, 0);
 502         /* was the operation canceled, or is it finished? */
 503         if (dev_replace->replace_state !=
 504             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 505                 btrfs_dev_replace_unlock(dev_replace, 0);
 506                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 507                 return 0;
 508         }
 509
 510         tgt_device = dev_replace->tgtdev;
 511         src_device = dev_replace->srcdev;
 512         btrfs_dev_replace_unlock(dev_replace, 0);
 513
 514         /*
 515          * flush all outstanding I/O and inode extent mappings before the
 516          * copy operation is declared as being finished
 517          */
 518         ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
 519         if (ret) {
 520                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 521                 return ret;
 522         }
 523         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 524
 525         trans = btrfs_start_transaction(root, 0);
 526         if (IS_ERR(trans)) {
 527                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 528                 return PTR_ERR(trans);
 529         }
 530         ret = btrfs_commit_transaction(trans);
 531         WARN_ON(ret);
 532
 533         mutex_lock(&uuid_mutex);
 534         /* keep away write_all_supers() during the finishing procedure */
 535         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 536         mutex_lock(&fs_info->chunk_mutex);
 537         btrfs_dev_replace_lock(dev_replace, 1);
 538         dev_replace->replace_state =
 539                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 540                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 541         dev_replace->tgtdev = NULL;
 542         dev_replace->srcdev = NULL;
 543         dev_replace->time_stopped = get_seconds();
 544         dev_replace->item_needs_writeback = 1;
 545
 546         /* replace old device with new one in mapping tree */
 547         if (!scrub_ret) {
 548                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 549                                                                 src_device,
 550                                                                 tgt_device);
 551         } else {
 552                 btrfs_err_in_rcu(fs_info,
 553                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 554                                  btrfs_dev_name(src_device),
 555                                  src_device->devid,
 556                                  rcu_str_deref(tgt_device->name), scrub_ret);
 557                 btrfs_dev_replace_unlock(dev_replace, 1);
 558                 mutex_unlock(&fs_info->chunk_mutex);
 559                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 560                 mutex_unlock(&uuid_mutex);
 561                 btrfs_rm_dev_replace_blocked(fs_info);
 562                 if (tgt_device)
 563                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 564                 btrfs_rm_dev_replace_unblocked(fs_info);
 565                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 566
 567                 return scrub_ret;
 568         }
 569
 570         btrfs_info_in_rcu(fs_info,
 571                           "dev_replace from %s (devid %llu) to %s finished",
 572                           btrfs_dev_name(src_device),
 573                           src_device->devid,
 574                           rcu_str_deref(tgt_device->name));
 575         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 576         tgt_device->devid = src_device->devid;
 577         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 578         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 579         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 580         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 581         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 582         btrfs_device_set_disk_total_bytes(tgt_device,
 583                                           src_device->disk_total_bytes);
 584         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 585         ASSERT(list_empty(&src_device->resized_list));
 586         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 587         tgt_device->commit_bytes_used = src_device->bytes_used;
 588
 589         btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
 590
 591         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 592         fs_info->fs_devices->rw_devices++;
 593
 594         btrfs_dev_replace_unlock(dev_replace, 1);
 595
 596         btrfs_rm_dev_replace_blocked(fs_info);
 597
 598         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 599
 600         btrfs_rm_dev_replace_unblocked(fs_info);
 601
 602         /*
 603          * this is again a consistent state where no dev_replace procedure
 604          * is running, the target device is part of the filesystem, the
 605          * source device is not part of the filesystem anymore and its 1st
 606          * superblock is scratched out so that it is no longer marked to
 607          * belong to this filesystem.
 608          */
 609         mutex_unlock(&fs_info->chunk_mutex);
 610         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 611         mutex_unlock(&uuid_mutex);
 612
 613         /* replace the sysfs entry */
 614         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 615         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 616
 617         /* write back the superblocks */
 618         trans = btrfs_start_transaction(root, 0);
 619         if (!IS_ERR(trans))
 620                 btrfs_commit_transaction(trans);
 621
 622         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 623
 624         return 0;
 625 }
 626
 627 static void btrfs_dev_replace_update_device_in_mapping_tree(
 628                                                 struct btrfs_fs_info *fs_info,
 629                                                 struct btrfs_device *srcdev,
 630                                                 struct btrfs_device *tgtdev)
 631 {
 632         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 633         struct extent_map *em;
 634         struct map_lookup *map;
 635         u64 start = 0;
 636         int i;
 637
 638         write_lock(&em_tree->lock);
 639         do {
 640                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 641                 if (!em)
 642                         break;
 643                 map = em->map_lookup;
 644                 for (i = 0; i < map->num_stripes; i++)
 645                         if (srcdev == map->stripes[i].dev)
 646                                 map->stripes[i].dev = tgtdev;
 647                 start = em->start + em->len;
 648                 free_extent_map(em);
 649         } while (start);
 650         write_unlock(&em_tree->lock);
 651 }
 652
 653 /*
 654  * Read progress of device replace status according to the state and last
 655  * stored position. The value format is the same as for
 656  * btrfs_dev_replace::progress_1000
 657  */
 658 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 659 {
 660         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 661         u64 ret = 0;
 662
 663         switch (dev_replace->replace_state) {
 664         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 665         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 666                 ret = 0;
 667                 break;
 668         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 669                 ret = 1000;
 670                 break;
 671         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 672         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 673                 ret = div64_u64(dev_replace->cursor_left,
 674                                 div_u64(btrfs_device_get_total_bytes(
 675                                                 dev_replace->srcdev), 1000));
 676                 break;
 677         }
 678
 679         return ret;
 680 }
 681
 682 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 683                               struct btrfs_ioctl_dev_replace_args *args)
 684 {
 685         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 686
 687         btrfs_dev_replace_lock(dev_replace, 0);
 688         /* even if !dev_replace_is_valid, the values are good enough for
 689          * the replace_status ioctl */
 690         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 691         args->status.replace_state = dev_replace->replace_state;
 692         args->status.time_started = dev_replace->time_started;
 693         args->status.time_stopped = dev_replace->time_stopped;
 694         args->status.num_write_errors =
 695                 atomic64_read(&dev_replace->num_write_errors);
 696         args->status.num_uncorrectable_read_errors =
 697                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 698         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 699         btrfs_dev_replace_unlock(dev_replace, 0);
 700 }
 701
 702 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 703 {
 704         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 705         struct btrfs_device *tgt_device = NULL;
 706         struct btrfs_trans_handle *trans;
 707         struct btrfs_root *root = fs_info->tree_root;
 708         int result;
 709         int ret;
 710
 711         if (sb_rdonly(fs_info->sb))
 712                 return -EROFS;
 713
 714         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 715         btrfs_dev_replace_lock(dev_replace, 1);
 716         switch (dev_replace->replace_state) {
 717         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 718         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 719         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 720                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 721                 btrfs_dev_replace_unlock(dev_replace, 1);
 722                 goto leave;
 723         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 724         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 725                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 726                 tgt_device = dev_replace->tgtdev;
 727                 dev_replace->tgtdev = NULL;
 728                 dev_replace->srcdev = NULL;
 729                 break;
 730         }
 731         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 732         dev_replace->time_stopped = get_seconds();
 733         dev_replace->item_needs_writeback = 1;
 734         btrfs_dev_replace_unlock(dev_replace, 1);
 735         btrfs_scrub_cancel(fs_info);
 736
 737         trans = btrfs_start_transaction(root, 0);
 738         if (IS_ERR(trans)) {
 739                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 740                 return PTR_ERR(trans);
 741         }
 742         ret = btrfs_commit_transaction(trans);
 743         WARN_ON(ret);
 744         if (tgt_device)
 745                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 746
 747 leave:
 748         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 749         return result;
 750 }
 751
 752 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 753 {
 754         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 755
 756         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 757         btrfs_dev_replace_lock(dev_replace, 1);
 758         switch (dev_replace->replace_state) {
 759         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 760         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 761         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 762         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 763                 break;
 764         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 765                 dev_replace->replace_state =
 766                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 767                 dev_replace->time_stopped = get_seconds();
 768                 dev_replace->item_needs_writeback = 1;
 769                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 770                 break;
 771         }
 772
 773         btrfs_dev_replace_unlock(dev_replace, 1);
 774         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 775 }
 776
 777 /* resume dev_replace procedure that was interrupted by unmount */
 778 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 779 {
 780         struct task_struct *task;
 781         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 782
 783         btrfs_dev_replace_lock(dev_replace, 1);
 784         switch (dev_replace->replace_state) {
 785         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 786         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 787         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 788                 btrfs_dev_replace_unlock(dev_replace, 1);
 789                 return 0;
 790         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 791                 break;
 792         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 793                 dev_replace->replace_state =
 794                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 795                 break;
 796         }
 797         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 798                 btrfs_info(fs_info,
 799                            "cannot continue dev_replace, tgtdev is missing");
 800                 btrfs_info(fs_info,
 801                            "you may cancel the operation after 'mount -o degraded'");
 802                 btrfs_dev_replace_unlock(dev_replace, 1);
 803                 return 0;
 804         }
 805         btrfs_dev_replace_unlock(dev_replace, 1);
 806
 807         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
 808         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 809         return PTR_ERR_OR_ZERO(task);
 810 }
 811
 812 static int btrfs_dev_replace_kthread(void *data)
 813 {
 814         struct btrfs_fs_info *fs_info = data;
 815         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 816         u64 progress;
 817
 818         progress = btrfs_dev_replace_progress(fs_info);
 819         progress = div_u64(progress, 10);
 820         btrfs_info_in_rcu(fs_info,
 821                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
 822                 btrfs_dev_name(dev_replace->srcdev),
 823                 dev_replace->srcdev->devid,
 824                 btrfs_dev_name(dev_replace->tgtdev),
 825                 (unsigned int)progress);
 826
 827         btrfs_dev_replace_continue_on_mount(fs_info);
 828         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 829
 830         return 0;
 831 }
 832
 833 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
 834 {
 835         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 836         int ret;
 837
 838         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 839                               dev_replace->committed_cursor_left,
 840                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 841                               &dev_replace->scrub_progress, 0, 1);
 842         ret = btrfs_dev_replace_finishing(fs_info, ret);
 843         WARN_ON(ret);
 844         return 0;
 845 }
 846
 847 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 848 {
 849         if (!dev_replace->is_valid)
 850                 return 0;
 851
 852         switch (dev_replace->replace_state) {
 853         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 854         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 855         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 856                 return 0;
 857         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 858         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 859                 /*
 860                  * return true even if tgtdev is missing (this is
 861                  * something that can happen if the dev_replace
 862                  * procedure is suspended by an umount and then
 863                  * the tgtdev is missing (or "btrfs dev scan") was
 864                  * not called and the the filesystem is remounted
 865                  * in degraded state. This does not stop the
 866                  * dev_replace procedure. It needs to be canceled
 867                  * manually if the cancellation is wanted.
 868                  */
 869                 break;
 870         }
 871         return 1;
 872 }
 873
 874 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
 875 {
 876         if (rw == 1) {
 877                 /* write */
 878 again:
 879                 wait_event(dev_replace->read_lock_wq,
 880                            atomic_read(&dev_replace->blocking_readers) == 0);
 881                 write_lock(&dev_replace->lock);
 882                 if (atomic_read(&dev_replace->blocking_readers)) {
 883                         write_unlock(&dev_replace->lock);
 884                         goto again;
 885                 }
 886         } else {
 887                 read_lock(&dev_replace->lock);
 888                 atomic_inc(&dev_replace->read_locks);
 889         }
 890 }
 891
 892 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
 893 {
 894         if (rw == 1) {
 895                 /* write */
 896                 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
 897                 write_unlock(&dev_replace->lock);
 898         } else {
 899                 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 900                 atomic_dec(&dev_replace->read_locks);
 901                 read_unlock(&dev_replace->lock);
 902         }
 903 }
 904
 905 /* inc blocking cnt and release read lock */
 906 void btrfs_dev_replace_set_lock_blocking(
 907                                         struct btrfs_dev_replace *dev_replace)
 908 {
 909         /* only set blocking for read lock */
 910         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 911         atomic_inc(&dev_replace->blocking_readers);
 912         read_unlock(&dev_replace->lock);
 913 }
 914
 915 /* acquire read lock and dec blocking cnt */
 916 void btrfs_dev_replace_clear_lock_blocking(
 917                                         struct btrfs_dev_replace *dev_replace)
 918 {
 919         /* only set blocking for read lock */
 920         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 921         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
 922         read_lock(&dev_replace->lock);
 923         if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
 924             waitqueue_active(&dev_replace->read_lock_wq))
 925                 wake_up(&dev_replace->read_lock_wq);
 926 }
 927
 928 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 929 {
 930         percpu_counter_inc(&fs_info->bio_counter);
 931 }
 932
 933 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
 934 {
 935         percpu_counter_sub(&fs_info->bio_counter, amount);
 936
 937         if (waitqueue_active(&fs_info->replace_wait))
 938                 wake_up(&fs_info->replace_wait);
 939 }
 940
 941 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
 942 {
 943         while (1) {
 944                 percpu_counter_inc(&fs_info->bio_counter);
 945                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 946                                      &fs_info->fs_state)))
 947                         break;
 948
 949                 btrfs_bio_counter_dec(fs_info);
 950                 wait_event(fs_info->replace_wait,
 951                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 952                                      &fs_info->fs_state));
 953         }
 954 }