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                         dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
 176                         btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
 177                                 dev_replace->tgtdev);
 178                 }
 179                 break;
 180         }
 181
 182 out:
 183         btrfs_free_path(path);
 184         return ret;
 185 }
 186
 187 /*
 188  * called from commit_transaction. Writes changed device replace state to
 189  * disk.
 190  */
 191 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 192                           struct btrfs_fs_info *fs_info)
 193 {
 194         int ret;
 195         struct btrfs_root *dev_root = fs_info->dev_root;
 196         struct btrfs_path *path;
 197         struct btrfs_key key;
 198         struct extent_buffer *eb;
 199         struct btrfs_dev_replace_item *ptr;
 200         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 201
 202         btrfs_dev_replace_lock(dev_replace, 0);
 203         if (!dev_replace->is_valid ||
 204             !dev_replace->item_needs_writeback) {
 205                 btrfs_dev_replace_unlock(dev_replace, 0);
 206                 return 0;
 207         }
 208         btrfs_dev_replace_unlock(dev_replace, 0);
 209
 210         key.objectid = 0;
 211         key.type = BTRFS_DEV_REPLACE_KEY;
 212         key.offset = 0;
 213
 214         path = btrfs_alloc_path();
 215         if (!path) {
 216                 ret = -ENOMEM;
 217                 goto out;
 218         }
 219         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 220         if (ret < 0) {
 221                 btrfs_warn(fs_info,
 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 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 308                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 309                 int read_src)
 310 {
 311         struct btrfs_root *root = fs_info->dev_root;
 312         struct btrfs_trans_handle *trans;
 313         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 314         int ret;
 315         struct btrfs_device *tgt_device = NULL;
 316         struct btrfs_device *src_device = NULL;
 317
 318         /* the disk copy procedure reuses the scrub code */
 319         mutex_lock(&fs_info->volume_mutex);
 320         ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
 321                                             srcdev_name, &src_device);
 322         if (ret) {
 323                 mutex_unlock(&fs_info->volume_mutex);
 324                 return ret;
 325         }
 326
 327         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 328                                             src_device, &tgt_device);
 329         mutex_unlock(&fs_info->volume_mutex);
 330         if (ret)
 331                 return ret;
 332
 333         /*
 334          * Here we commit the transaction to make sure commit_total_bytes
 335          * of all the devices are updated.
 336          */
 337         trans = btrfs_attach_transaction(root);
 338         if (!IS_ERR(trans)) {
 339                 ret = btrfs_commit_transaction(trans);
 340                 if (ret)
 341                         return ret;
 342         } else if (PTR_ERR(trans) != -ENOENT) {
 343                 return PTR_ERR(trans);
 344         }
 345
 346         btrfs_dev_replace_lock(dev_replace, 1);
 347         switch (dev_replace->replace_state) {
 348         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 349         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 350         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 351                 break;
 352         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 353         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 354                 ASSERT(0);
 355                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 356                 goto leave;
 357         }
 358
 359         dev_replace->cont_reading_from_srcdev_mode = read_src;
 360         WARN_ON(!src_device);
 361         dev_replace->srcdev = src_device;
 362         WARN_ON(!tgt_device);
 363         dev_replace->tgtdev = tgt_device;
 364
 365         btrfs_info_in_rcu(fs_info,
 366                       "dev_replace from %s (devid %llu) to %s started",
 367                       src_device->missing ? "<missing disk>" :
 368                         rcu_str_deref(src_device->name),
 369                       src_device->devid,
 370                       rcu_str_deref(tgt_device->name));
 371
 372         /*
 373          * from now on, the writes to the srcdev are all duplicated to
 374          * go to the tgtdev as well (refer to btrfs_map_block()).
 375          */
 376         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 377         dev_replace->time_started = get_seconds();
 378         dev_replace->cursor_left = 0;
 379         dev_replace->committed_cursor_left = 0;
 380         dev_replace->cursor_left_last_write_of_item = 0;
 381         dev_replace->cursor_right = 0;
 382         dev_replace->is_valid = 1;
 383         dev_replace->item_needs_writeback = 1;
 384         atomic64_set(&dev_replace->num_write_errors, 0);
 385         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 386         btrfs_dev_replace_unlock(dev_replace, 1);
 387
 388         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 389         if (ret)
 390                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
 391
 392         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 393
 394         /* force writing the updated state information to disk */
 395         trans = btrfs_start_transaction(root, 0);
 396         if (IS_ERR(trans)) {
 397                 ret = PTR_ERR(trans);
 398                 btrfs_dev_replace_lock(dev_replace, 1);
 399                 dev_replace->replace_state =
 400                         BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
 401                 dev_replace->srcdev = NULL;
 402                 dev_replace->tgtdev = NULL;
 403                 goto leave;
 404         }
 405
 406         ret = btrfs_commit_transaction(trans);
 407         WARN_ON(ret);
 408
 409         /* the disk copy procedure reuses the scrub code */
 410         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 411                               btrfs_device_get_total_bytes(src_device),
 412                               &dev_replace->scrub_progress, 0, 1);
 413
 414         ret = btrfs_dev_replace_finishing(fs_info, ret);
 415         if (ret == -EINPROGRESS) {
 416                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 417         } else {
 418                 WARN_ON(ret);
 419         }
 420
 421         return ret;
 422
 423 leave:
 424         btrfs_dev_replace_unlock(dev_replace, 1);
 425         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 426         return ret;
 427 }
 428
 429 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 430                             struct btrfs_ioctl_dev_replace_args *args)
 431 {
 432         int ret;
 433
 434         switch (args->start.cont_reading_from_srcdev_mode) {
 435         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 436         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 437                 break;
 438         default:
 439                 return -EINVAL;
 440         }
 441
 442         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 443             args->start.tgtdev_name[0] == '\0')
 444                 return -EINVAL;
 445
 446         ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
 447                                         args->start.srcdevid,
 448                                         args->start.srcdev_name,
 449                                         args->start.cont_reading_from_srcdev_mode);
 450         args->result = ret;
 451         /* don't warn if EINPROGRESS, someone else might be running scrub */
 452         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
 453                 ret = 0;
 454
 455         return ret;
 456 }
 457
 458 /*
 459  * blocked until all in-flight bios operations are finished.
 460  */
 461 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 462 {
 463         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 464         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 465                    &fs_info->bio_counter));
 466 }
 467
 468 /*
 469  * we have removed target device, it is safe to allow new bios request.
 470  */
 471 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 472 {
 473         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 474         wake_up(&fs_info->replace_wait);
 475 }
 476
 477 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 478                                        int scrub_ret)
 479 {
 480         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 481         struct btrfs_device *tgt_device;
 482         struct btrfs_device *src_device;
 483         struct btrfs_root *root = fs_info->tree_root;
 484         u8 uuid_tmp[BTRFS_UUID_SIZE];
 485         struct btrfs_trans_handle *trans;
 486         int ret = 0;
 487
 488         /* don't allow cancel or unmount to disturb the finishing procedure */
 489         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 490
 491         btrfs_dev_replace_lock(dev_replace, 0);
 492         /* was the operation canceled, or is it finished? */
 493         if (dev_replace->replace_state !=
 494             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 495                 btrfs_dev_replace_unlock(dev_replace, 0);
 496                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 497                 return 0;
 498         }
 499
 500         tgt_device = dev_replace->tgtdev;
 501         src_device = dev_replace->srcdev;
 502         btrfs_dev_replace_unlock(dev_replace, 0);
 503
 504         /*
 505          * flush all outstanding I/O and inode extent mappings before the
 506          * copy operation is declared as being finished
 507          */
 508         ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
 509         if (ret) {
 510                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 511                 return ret;
 512         }
 513         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 514
 515         while (1) {
 516                 trans = btrfs_start_transaction(root, 0);
 517                 if (IS_ERR(trans)) {
 518                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 519                         return PTR_ERR(trans);
 520                 }
 521                 ret = btrfs_commit_transaction(trans);
 522                 WARN_ON(ret);
 523                 mutex_lock(&uuid_mutex);
 524                 /* keep away write_all_supers() during the finishing procedure */
 525                 mutex_lock(&fs_info->fs_devices->device_list_mutex);
 526                 mutex_lock(&fs_info->chunk_mutex);
 527                 if (src_device->has_pending_chunks) {
 528                         mutex_unlock(&root->fs_info->chunk_mutex);
 529                         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 530                         mutex_unlock(&uuid_mutex);
 531                 } else {
 532                         break;
 533                 }
 534         }
 535
 536         btrfs_dev_replace_lock(dev_replace, 1);
 537         dev_replace->replace_state =
 538                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 539                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 540         dev_replace->tgtdev = NULL;
 541         dev_replace->srcdev = NULL;
 542         dev_replace->time_stopped = get_seconds();
 543         dev_replace->item_needs_writeback = 1;
 544
 545         /* replace old device with new one in mapping tree */
 546         if (!scrub_ret) {
 547                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 548                                                                 src_device,
 549                                                                 tgt_device);
 550         } else {
 551                 btrfs_err_in_rcu(fs_info,
 552                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 553                                  src_device->missing ? "<missing disk>" :
 554                                  rcu_str_deref(src_device->name),
 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                           src_device->missing ? "<missing disk>" :
 573                           rcu_str_deref(src_device->name),
 574                           src_device->devid,
 575                           rcu_str_deref(tgt_device->name));
 576         tgt_device->is_tgtdev_for_dev_replace = 0;
 577         tgt_device->devid = src_device->devid;
 578         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 579         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 580         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 581         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 582         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 583         btrfs_device_set_disk_total_bytes(tgt_device,
 584                                           src_device->disk_total_bytes);
 585         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 586         ASSERT(list_empty(&src_device->resized_list));
 587         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 588         tgt_device->commit_bytes_used = src_device->bytes_used;
 589
 590         btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
 591
 592         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 593         fs_info->fs_devices->rw_devices++;
 594
 595         btrfs_dev_replace_unlock(dev_replace, 1);
 596
 597         btrfs_rm_dev_replace_blocked(fs_info);
 598
 599         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 600
 601         btrfs_rm_dev_replace_unblocked(fs_info);
 602
 603         /*
 604          * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
 605          * update on-disk dev stats value during commit transaction
 606          */
 607         atomic_inc(&tgt_device->dev_stats_ccnt);
 608
 609         /*
 610          * this is again a consistent state where no dev_replace procedure
 611          * is running, the target device is part of the filesystem, the
 612          * source device is not part of the filesystem anymore and its 1st
 613          * superblock is scratched out so that it is no longer marked to
 614          * belong to this filesystem.
 615          */
 616         mutex_unlock(&fs_info->chunk_mutex);
 617         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 618         mutex_unlock(&uuid_mutex);
 619
 620         /* replace the sysfs entry */
 621         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 622         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 623
 624         /* write back the superblocks */
 625         trans = btrfs_start_transaction(root, 0);
 626         if (!IS_ERR(trans))
 627                 btrfs_commit_transaction(trans);
 628
 629         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 630
 631         return 0;
 632 }
 633
 634 static void btrfs_dev_replace_update_device_in_mapping_tree(
 635                                                 struct btrfs_fs_info *fs_info,
 636                                                 struct btrfs_device *srcdev,
 637                                                 struct btrfs_device *tgtdev)
 638 {
 639         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 640         struct extent_map *em;
 641         struct map_lookup *map;
 642         u64 start = 0;
 643         int i;
 644
 645         write_lock(&em_tree->lock);
 646         do {
 647                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 648                 if (!em)
 649                         break;
 650                 map = em->map_lookup;
 651                 for (i = 0; i < map->num_stripes; i++)
 652                         if (srcdev == map->stripes[i].dev)
 653                                 map->stripes[i].dev = tgtdev;
 654                 start = em->start + em->len;
 655                 free_extent_map(em);
 656         } while (start);
 657         write_unlock(&em_tree->lock);
 658 }
 659
 660 /*
 661  * Read progress of device replace status according to the state and last
 662  * stored position. The value format is the same as for
 663  * btrfs_dev_replace::progress_1000
 664  */
 665 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 666 {
 667         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 668         u64 ret = 0;
 669
 670         switch (dev_replace->replace_state) {
 671         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 672         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 673                 ret = 0;
 674                 break;
 675         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 676                 ret = 1000;
 677                 break;
 678         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 679         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 680                 ret = div64_u64(dev_replace->cursor_left,
 681                                 div_u64(btrfs_device_get_total_bytes(
 682                                                 dev_replace->srcdev), 1000));
 683                 break;
 684         }
 685
 686         return ret;
 687 }
 688
 689 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 690                               struct btrfs_ioctl_dev_replace_args *args)
 691 {
 692         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 693
 694         btrfs_dev_replace_lock(dev_replace, 0);
 695         /* even if !dev_replace_is_valid, the values are good enough for
 696          * the replace_status ioctl */
 697         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 698         args->status.replace_state = dev_replace->replace_state;
 699         args->status.time_started = dev_replace->time_started;
 700         args->status.time_stopped = dev_replace->time_stopped;
 701         args->status.num_write_errors =
 702                 atomic64_read(&dev_replace->num_write_errors);
 703         args->status.num_uncorrectable_read_errors =
 704                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 705         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 706         btrfs_dev_replace_unlock(dev_replace, 0);
 707 }
 708
 709 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
 710                              struct btrfs_ioctl_dev_replace_args *args)
 711 {
 712         args->result = __btrfs_dev_replace_cancel(fs_info);
 713         return 0;
 714 }
 715
 716 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 717 {
 718         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 719         struct btrfs_device *tgt_device = NULL;
 720         struct btrfs_trans_handle *trans;
 721         struct btrfs_root *root = fs_info->tree_root;
 722         u64 result;
 723         int ret;
 724
 725         if (sb_rdonly(fs_info->sb))
 726                 return -EROFS;
 727
 728         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 729         btrfs_dev_replace_lock(dev_replace, 1);
 730         switch (dev_replace->replace_state) {
 731         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 732         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 733         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 734                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 735                 btrfs_dev_replace_unlock(dev_replace, 1);
 736                 goto leave;
 737         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 738         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 739                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 740                 tgt_device = dev_replace->tgtdev;
 741                 dev_replace->tgtdev = NULL;
 742                 dev_replace->srcdev = NULL;
 743                 break;
 744         }
 745         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 746         dev_replace->time_stopped = get_seconds();
 747         dev_replace->item_needs_writeback = 1;
 748         btrfs_dev_replace_unlock(dev_replace, 1);
 749         btrfs_scrub_cancel(fs_info);
 750
 751         trans = btrfs_start_transaction(root, 0);
 752         if (IS_ERR(trans)) {
 753                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 754                 return PTR_ERR(trans);
 755         }
 756         ret = btrfs_commit_transaction(trans);
 757         WARN_ON(ret);
 758         if (tgt_device)
 759                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 760
 761 leave:
 762         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 763         return result;
 764 }
 765
 766 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 767 {
 768         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 769
 770         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 771         btrfs_dev_replace_lock(dev_replace, 1);
 772         switch (dev_replace->replace_state) {
 773         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 774         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 775         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 776         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 777                 break;
 778         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 779                 dev_replace->replace_state =
 780                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 781                 dev_replace->time_stopped = get_seconds();
 782                 dev_replace->item_needs_writeback = 1;
 783                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 784                 break;
 785         }
 786
 787         btrfs_dev_replace_unlock(dev_replace, 1);
 788         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 789 }
 790
 791 /* resume dev_replace procedure that was interrupted by unmount */
 792 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 793 {
 794         struct task_struct *task;
 795         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 796
 797         btrfs_dev_replace_lock(dev_replace, 1);
 798         switch (dev_replace->replace_state) {
 799         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 800         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 801         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 802                 btrfs_dev_replace_unlock(dev_replace, 1);
 803                 return 0;
 804         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 805                 break;
 806         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 807                 dev_replace->replace_state =
 808                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 809                 break;
 810         }
 811         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 812                 btrfs_info(fs_info,
 813                            "cannot continue dev_replace, tgtdev is missing");
 814                 btrfs_info(fs_info,
 815                            "you may cancel the operation after 'mount -o degraded'");
 816                 dev_replace->replace_state =
 817                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 818                 btrfs_dev_replace_unlock(dev_replace, 1);
 819                 return 0;
 820         }
 821         btrfs_dev_replace_unlock(dev_replace, 1);
 822
 823         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
 824         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 825         return PTR_ERR_OR_ZERO(task);
 826 }
 827
 828 static int btrfs_dev_replace_kthread(void *data)
 829 {
 830         struct btrfs_fs_info *fs_info = data;
 831         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 832         u64 progress;
 833
 834         progress = btrfs_dev_replace_progress(fs_info);
 835         progress = div_u64(progress, 10);
 836         btrfs_info_in_rcu(fs_info,
 837                 "continuing dev_replace from %s (devid %llu) to %s @%u%%",
 838                 dev_replace->srcdev->missing ? "<missing disk>"
 839                         : rcu_str_deref(dev_replace->srcdev->name),
 840                 dev_replace->srcdev->devid,
 841                 dev_replace->tgtdev ? rcu_str_deref(dev_replace->tgtdev->name)
 842                         : "<missing target disk>",
 843                 (unsigned int)progress);
 844
 845         btrfs_dev_replace_continue_on_mount(fs_info);
 846         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 847
 848         return 0;
 849 }
 850
 851 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
 852 {
 853         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 854         int ret;
 855
 856         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 857                               dev_replace->committed_cursor_left,
 858                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 859                               &dev_replace->scrub_progress, 0, 1);
 860         ret = btrfs_dev_replace_finishing(fs_info, ret);
 861         WARN_ON(ret);
 862         return 0;
 863 }
 864
 865 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 866 {
 867         if (!dev_replace->is_valid)
 868                 return 0;
 869
 870         switch (dev_replace->replace_state) {
 871         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 872         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 873         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 874                 return 0;
 875         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 876         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 877                 /*
 878                  * return true even if tgtdev is missing (this is
 879                  * something that can happen if the dev_replace
 880                  * procedure is suspended by an umount and then
 881                  * the tgtdev is missing (or "btrfs dev scan") was
 882                  * not called and the the filesystem is remounted
 883                  * in degraded state. This does not stop the
 884                  * dev_replace procedure. It needs to be canceled
 885                  * manually if the cancellation is wanted.
 886                  */
 887                 break;
 888         }
 889         return 1;
 890 }
 891
 892 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
 893 {
 894         if (rw == 1) {
 895                 /* write */
 896 again:
 897                 wait_event(dev_replace->read_lock_wq,
 898                            atomic_read(&dev_replace->blocking_readers) == 0);
 899                 write_lock(&dev_replace->lock);
 900                 if (atomic_read(&dev_replace->blocking_readers)) {
 901                         write_unlock(&dev_replace->lock);
 902                         goto again;
 903                 }
 904         } else {
 905                 read_lock(&dev_replace->lock);
 906                 atomic_inc(&dev_replace->read_locks);
 907         }
 908 }
 909
 910 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
 911 {
 912         if (rw == 1) {
 913                 /* write */
 914                 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
 915                 write_unlock(&dev_replace->lock);
 916         } else {
 917                 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 918                 atomic_dec(&dev_replace->read_locks);
 919                 read_unlock(&dev_replace->lock);
 920         }
 921 }
 922
 923 /* inc blocking cnt and release read lock */
 924 void btrfs_dev_replace_set_lock_blocking(
 925                                         struct btrfs_dev_replace *dev_replace)
 926 {
 927         /* only set blocking for read lock */
 928         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 929         atomic_inc(&dev_replace->blocking_readers);
 930         read_unlock(&dev_replace->lock);
 931 }
 932
 933 /* acquire read lock and dec blocking cnt */
 934 void btrfs_dev_replace_clear_lock_blocking(
 935                                         struct btrfs_dev_replace *dev_replace)
 936 {
 937         /* only set blocking for read lock */
 938         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 939         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
 940         read_lock(&dev_replace->lock);
 941         if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
 942             waitqueue_active(&dev_replace->read_lock_wq))
 943                 wake_up(&dev_replace->read_lock_wq);
 944 }
 945
 946 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 947 {
 948         percpu_counter_inc(&fs_info->bio_counter);
 949 }
 950
 951 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
 952 {
 953         percpu_counter_sub(&fs_info->bio_counter, amount);
 954
 955         if (waitqueue_active(&fs_info->replace_wait))
 956                 wake_up(&fs_info->replace_wait);
 957 }
 958
 959 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
 960 {
 961         while (1) {
 962                 percpu_counter_inc(&fs_info->bio_counter);
 963                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 964                                      &fs_info->fs_state)))
 965                         break;
 966
 967                 btrfs_bio_counter_dec(fs_info);
 968                 wait_event(fs_info->replace_wait,
 969                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 970                                      &fs_info->fs_state));
 971         }
 972 }