fs/btrfs/dev-replace.c

   1 /*
   2  * Copyright (C) STRATO AG 2012.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/kthread.h>
  27 #include <linux/math64.h>
  28 #include <asm/div64.h>
  29 #include "ctree.h"
  30 #include "extent_map.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "print-tree.h"
  34 #include "volumes.h"
  35 #include "async-thread.h"
  36 #include "check-integrity.h"
  37 #include "rcu-string.h"
  38 #include "dev-replace.h"
  39 #include "sysfs.h"
  40
  41 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
  42                                        int scrub_ret);
  43 static void btrfs_dev_replace_update_device_in_mapping_tree(
  44                                                 struct btrfs_fs_info *fs_info,
  45                                                 struct btrfs_device *srcdev,
  46                                                 struct btrfs_device *tgtdev);
  47 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
  48 static int btrfs_dev_replace_kthread(void *data);
  49 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
  50
  51
  52 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  53 {
  54         struct btrfs_key key;
  55         struct btrfs_root *dev_root = fs_info->dev_root;
  56         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  57         struct extent_buffer *eb;
  58         int slot;
  59         int ret = 0;
  60         struct btrfs_path *path = NULL;
  61         int item_size;
  62         struct btrfs_dev_replace_item *ptr;
  63         u64 src_devid;
  64
  65         path = btrfs_alloc_path();
  66         if (!path) {
  67                 ret = -ENOMEM;
  68                 goto out;
  69         }
  70
  71         key.objectid = 0;
  72         key.type = BTRFS_DEV_REPLACE_KEY;
  73         key.offset = 0;
  74         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  75         if (ret) {
  76 no_valid_dev_replace_entry_found:
  77                 ret = 0;
  78                 dev_replace->replace_state =
  79                         BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
  80                 dev_replace->cont_reading_from_srcdev_mode =
  81                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
  82                 dev_replace->replace_state = 0;
  83                 dev_replace->time_started = 0;
  84                 dev_replace->time_stopped = 0;
  85                 atomic64_set(&dev_replace->num_write_errors, 0);
  86                 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
  87                 dev_replace->cursor_left = 0;
  88                 dev_replace->committed_cursor_left = 0;
  89                 dev_replace->cursor_left_last_write_of_item = 0;
  90                 dev_replace->cursor_right = 0;
  91                 dev_replace->srcdev = NULL;
  92                 dev_replace->tgtdev = NULL;
  93                 dev_replace->is_valid = 0;
  94                 dev_replace->item_needs_writeback = 0;
  95                 goto out;
  96         }
  97         slot = path->slots[0];
  98         eb = path->nodes[0];
  99         item_size = btrfs_item_size_nr(eb, slot);
 100         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
 101
 102         if (item_size != sizeof(struct btrfs_dev_replace_item)) {
 103                 btrfs_warn(fs_info,
 104                         "dev_replace entry found has unexpected size, ignore entry");
 105                 goto no_valid_dev_replace_entry_found;
 106         }
 107
 108         src_devid = btrfs_dev_replace_src_devid(eb, ptr);
 109         dev_replace->cont_reading_from_srcdev_mode =
 110                 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
 111         dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
 112         dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
 113         dev_replace->time_stopped =
 114                 btrfs_dev_replace_time_stopped(eb, ptr);
 115         atomic64_set(&dev_replace->num_write_errors,
 116                      btrfs_dev_replace_num_write_errors(eb, ptr));
 117         atomic64_set(&dev_replace->num_uncorrectable_read_errors,
 118                      btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
 119         dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
 120         dev_replace->committed_cursor_left = dev_replace->cursor_left;
 121         dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
 122         dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
 123         dev_replace->is_valid = 1;
 124
 125         dev_replace->item_needs_writeback = 0;
 126         switch (dev_replace->replace_state) {
 127         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 128         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 129         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 130                 dev_replace->srcdev = NULL;
 131                 dev_replace->tgtdev = NULL;
 132                 break;
 133         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 134         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 135                 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
 136                                                         NULL, NULL);
 137                 dev_replace->tgtdev = btrfs_find_device(fs_info,
 138                                                         BTRFS_DEV_REPLACE_DEVID,
 139                                                         NULL, NULL);
 140                 /*
 141                  * allow 'btrfs dev replace_cancel' if src/tgt device is
 142                  * missing
 143                  */
 144                 if (!dev_replace->srcdev &&
 145                     !btrfs_test_opt(dev_root->fs_info, DEGRADED)) {
 146                         ret = -EIO;
 147                         btrfs_warn(fs_info,
 148                            "cannot mount because device replace operation is ongoing and");
 149                         btrfs_warn(fs_info,
 150                            "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 151                            src_devid);
 152                 }
 153                 if (!dev_replace->tgtdev &&
 154                     !btrfs_test_opt(dev_root->fs_info, DEGRADED)) {
 155                         ret = -EIO;
 156                         btrfs_warn(fs_info,
 157                            "cannot mount because device replace operation is ongoing and");
 158                         btrfs_warn(fs_info,
 159                            "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 160                                 BTRFS_DEV_REPLACE_DEVID);
 161                 }
 162                 if (dev_replace->tgtdev) {
 163                         if (dev_replace->srcdev) {
 164                                 dev_replace->tgtdev->total_bytes =
 165                                         dev_replace->srcdev->total_bytes;
 166                                 dev_replace->tgtdev->disk_total_bytes =
 167                                         dev_replace->srcdev->disk_total_bytes;
 168                                 dev_replace->tgtdev->commit_total_bytes =
 169                                         dev_replace->srcdev->commit_total_bytes;
 170                                 dev_replace->tgtdev->bytes_used =
 171                                         dev_replace->srcdev->bytes_used;
 172                                 dev_replace->tgtdev->commit_bytes_used =
 173                                         dev_replace->srcdev->commit_bytes_used;
 174                         }
 175                         dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
 176                         btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
 177                                 dev_replace->tgtdev);
 178                 }
 179                 break;
 180         }
 181
 182 out:
 183         btrfs_free_path(path);
 184         return ret;
 185 }
 186
 187 /*
 188  * called from commit_transaction. Writes changed device replace state to
 189  * disk.
 190  */
 191 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 192                           struct btrfs_fs_info *fs_info)
 193 {
 194         int ret;
 195         struct btrfs_root *dev_root = fs_info->dev_root;
 196         struct btrfs_path *path;
 197         struct btrfs_key key;
 198         struct extent_buffer *eb;
 199         struct btrfs_dev_replace_item *ptr;
 200         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 201
 202         btrfs_dev_replace_lock(dev_replace, 0);
 203         if (!dev_replace->is_valid ||
 204             !dev_replace->item_needs_writeback) {
 205                 btrfs_dev_replace_unlock(dev_replace, 0);
 206                 return 0;
 207         }
 208         btrfs_dev_replace_unlock(dev_replace, 0);
 209
 210         key.objectid = 0;
 211         key.type = BTRFS_DEV_REPLACE_KEY;
 212         key.offset = 0;
 213
 214         path = btrfs_alloc_path();
 215         if (!path) {
 216                 ret = -ENOMEM;
 217                 goto out;
 218         }
 219         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 220         if (ret < 0) {
 221                 btrfs_warn(fs_info,
 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_root *root, char *tgtdev_name,
 308                                 u64 srcdevid, char *srcdev_name, int read_src)
 309 {
 310         struct btrfs_trans_handle *trans;
 311         struct btrfs_fs_info *fs_info = root->fs_info;
 312         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 313         int ret;
 314         struct btrfs_device *tgt_device = NULL;
 315         struct btrfs_device *src_device = NULL;
 316
 317         /* the disk copy procedure reuses the scrub code */
 318         mutex_lock(&fs_info->volume_mutex);
 319         ret = btrfs_find_device_by_devspec(root, srcdevid,
 320                                             srcdev_name, &src_device);
 321         if (ret) {
 322                 mutex_unlock(&fs_info->volume_mutex);
 323                 return ret;
 324         }
 325
 326         ret = btrfs_init_dev_replace_tgtdev(root, tgtdev_name,
 327                                             src_device, &tgt_device);
 328         mutex_unlock(&fs_info->volume_mutex);
 329         if (ret)
 330                 return ret;
 331
 332         /*
 333          * Here we commit the transaction to make sure commit_total_bytes
 334          * of all the devices are updated.
 335          */
 336         trans = btrfs_attach_transaction(root);
 337         if (!IS_ERR(trans)) {
 338                 ret = btrfs_commit_transaction(trans, root);
 339                 if (ret)
 340                         return ret;
 341         } else if (PTR_ERR(trans) != -ENOENT) {
 342                 return PTR_ERR(trans);
 343         }
 344
 345         btrfs_dev_replace_lock(dev_replace, 1);
 346         switch (dev_replace->replace_state) {
 347         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 348         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 349         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 350                 break;
 351         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 352         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 353                 ASSERT(0);
 354                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 355                 goto leave;
 356         }
 357
 358         dev_replace->cont_reading_from_srcdev_mode = read_src;
 359         WARN_ON(!src_device);
 360         dev_replace->srcdev = src_device;
 361         WARN_ON(!tgt_device);
 362         dev_replace->tgtdev = tgt_device;
 363
 364         btrfs_info_in_rcu(fs_info,
 365                       "dev_replace from %s (devid %llu) to %s started",
 366                       src_device->missing ? "<missing disk>" :
 367                         rcu_str_deref(src_device->name),
 368                       src_device->devid,
 369                       rcu_str_deref(tgt_device->name));
 370
 371         /*
 372          * from now on, the writes to the srcdev are all duplicated to
 373          * go to the tgtdev as well (refer to btrfs_map_block()).
 374          */
 375         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 376         dev_replace->time_started = get_seconds();
 377         dev_replace->cursor_left = 0;
 378         dev_replace->committed_cursor_left = 0;
 379         dev_replace->cursor_left_last_write_of_item = 0;
 380         dev_replace->cursor_right = 0;
 381         dev_replace->is_valid = 1;
 382         dev_replace->item_needs_writeback = 1;
 383         atomic64_set(&dev_replace->num_write_errors, 0);
 384         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 385         btrfs_dev_replace_unlock(dev_replace, 1);
 386
 387         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 388         if (ret)
 389                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
 390
 391         btrfs_wait_ordered_roots(root->fs_info, -1, 0, (u64)-1);
 392
 393         /* force writing the updated state information to disk */
 394         trans = btrfs_start_transaction(root, 0);
 395         if (IS_ERR(trans)) {
 396                 ret = PTR_ERR(trans);
 397                 btrfs_dev_replace_lock(dev_replace, 1);
 398                 dev_replace->replace_state =
 399                         BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
 400                 dev_replace->srcdev = NULL;
 401                 dev_replace->tgtdev = NULL;
 402                 goto leave;
 403         }
 404
 405         ret = btrfs_commit_transaction(trans, root);
 406         WARN_ON(ret);
 407
 408         /* the disk copy procedure reuses the scrub code */
 409         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 410                               btrfs_device_get_total_bytes(src_device),
 411                               &dev_replace->scrub_progress, 0, 1);
 412
 413         ret = btrfs_dev_replace_finishing(fs_info, ret);
 414         if (ret == -EINPROGRESS) {
 415                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 416         } else {
 417                 WARN_ON(ret);
 418         }
 419
 420         return ret;
 421
 422 leave:
 423         btrfs_dev_replace_unlock(dev_replace, 1);
 424         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 425         return ret;
 426 }
 427
 428 int btrfs_dev_replace_by_ioctl(struct btrfs_root *root,
 429                             struct btrfs_ioctl_dev_replace_args *args)
 430 {
 431         int ret;
 432
 433         switch (args->start.cont_reading_from_srcdev_mode) {
 434         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 435         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 436                 break;
 437         default:
 438                 return -EINVAL;
 439         }
 440
 441         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 442             args->start.tgtdev_name[0] == '\0')
 443                 return -EINVAL;
 444
 445         ret = btrfs_dev_replace_start(root, args->start.tgtdev_name,
 446                                         args->start.srcdevid,
 447                                         args->start.srcdev_name,
 448                                         args->start.cont_reading_from_srcdev_mode);
 449         args->result = ret;
 450         /* don't warn if EINPROGRESS, someone else might be running scrub */
 451         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
 452                 ret = 0;
 453
 454         return ret;
 455 }
 456
 457 /*
 458  * blocked until all in-flight bios operations are finished.
 459  */
 460 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 461 {
 462         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 463         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 464                    &fs_info->bio_counter));
 465 }
 466
 467 /*
 468  * we have removed target device, it is safe to allow new bios request.
 469  */
 470 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 471 {
 472         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 473         wake_up(&fs_info->replace_wait);
 474 }
 475
 476 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 477                                        int scrub_ret)
 478 {
 479         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 480         struct btrfs_device *tgt_device;
 481         struct btrfs_device *src_device;
 482         struct btrfs_root *root = fs_info->tree_root;
 483         u8 uuid_tmp[BTRFS_UUID_SIZE];
 484         struct btrfs_trans_handle *trans;
 485         int ret = 0;
 486
 487         /* don't allow cancel or unmount to disturb the finishing procedure */
 488         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 489
 490         btrfs_dev_replace_lock(dev_replace, 0);
 491         /* was the operation canceled, or is it finished? */
 492         if (dev_replace->replace_state !=
 493             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 494                 btrfs_dev_replace_unlock(dev_replace, 0);
 495                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 496                 return 0;
 497         }
 498
 499         tgt_device = dev_replace->tgtdev;
 500         src_device = dev_replace->srcdev;
 501         btrfs_dev_replace_unlock(dev_replace, 0);
 502
 503         /*
 504          * flush all outstanding I/O and inode extent mappings before the
 505          * copy operation is declared as being finished
 506          */
 507         ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
 508         if (ret) {
 509                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 510                 return ret;
 511         }
 512         btrfs_wait_ordered_roots(root->fs_info, -1, 0, (u64)-1);
 513
 514         while (1) {
 515                 trans = btrfs_start_transaction(root, 0);
 516                 if (IS_ERR(trans)) {
 517                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 518                         return PTR_ERR(trans);
 519                 }
 520                 ret = btrfs_commit_transaction(trans, root);
 521                 WARN_ON(ret);
 522                 mutex_lock(&uuid_mutex);
 523                 /* keep away write_all_supers() during the finishing procedure */
 524                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 525                 mutex_lock(&root->fs_info->chunk_mutex);
 526                 if (src_device->has_pending_chunks) {
 527                         mutex_unlock(&root->fs_info->chunk_mutex);
 528                         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 529                         mutex_unlock(&uuid_mutex);
 530                 } else {
 531                         break;
 532                 }
 533         }
 534
 535         btrfs_dev_replace_lock(dev_replace, 1);
 536         dev_replace->replace_state =
 537                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 538                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 539         dev_replace->tgtdev = NULL;
 540         dev_replace->srcdev = NULL;
 541         dev_replace->time_stopped = get_seconds();
 542         dev_replace->item_needs_writeback = 1;
 543
 544         /* replace old device with new one in mapping tree */
 545         if (!scrub_ret) {
 546                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 547                                                                 src_device,
 548                                                                 tgt_device);
 549         } else {
 550                 btrfs_err_in_rcu(root->fs_info,
 551                               "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 552                               src_device->missing ? "<missing disk>" :
 553                                 rcu_str_deref(src_device->name),
 554                               src_device->devid,
 555                               rcu_str_deref(tgt_device->name), scrub_ret);
 556                 btrfs_dev_replace_unlock(dev_replace, 1);
 557                 mutex_unlock(&root->fs_info->chunk_mutex);
 558                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 559                 mutex_unlock(&uuid_mutex);
 560                 if (tgt_device)
 561                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 562                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 563
 564                 return scrub_ret;
 565         }
 566
 567         btrfs_info_in_rcu(root->fs_info,
 568                       "dev_replace from %s (devid %llu) to %s finished",
 569                       src_device->missing ? "<missing disk>" :
 570                         rcu_str_deref(src_device->name),
 571                       src_device->devid,
 572                       rcu_str_deref(tgt_device->name));
 573         tgt_device->is_tgtdev_for_dev_replace = 0;
 574         tgt_device->devid = src_device->devid;
 575         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 576         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 577         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 578         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 579         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 580         btrfs_device_set_disk_total_bytes(tgt_device,
 581                                           src_device->disk_total_bytes);
 582         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 583         ASSERT(list_empty(&src_device->resized_list));
 584         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 585         tgt_device->commit_bytes_used = src_device->bytes_used;
 586
 587         btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
 588
 589         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 590         fs_info->fs_devices->rw_devices++;
 591
 592         btrfs_dev_replace_unlock(dev_replace, 1);
 593
 594         btrfs_rm_dev_replace_blocked(fs_info);
 595
 596         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 597
 598         btrfs_rm_dev_replace_unblocked(fs_info);
 599
 600         /*
 601          * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
 602          * update on-disk dev stats value during commit transaction
 603          */
 604         atomic_inc(&tgt_device->dev_stats_ccnt);
 605
 606         /*
 607          * this is again a consistent state where no dev_replace procedure
 608          * is running, the target device is part of the filesystem, the
 609          * source device is not part of the filesystem anymore and its 1st
 610          * superblock is scratched out so that it is no longer marked to
 611          * belong to this filesystem.
 612          */
 613         mutex_unlock(&root->fs_info->chunk_mutex);
 614         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 615         mutex_unlock(&uuid_mutex);
 616
 617         /* replace the sysfs entry */
 618         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 619         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 620
 621         /* write back the superblocks */
 622         trans = btrfs_start_transaction(root, 0);
 623         if (!IS_ERR(trans))
 624                 btrfs_commit_transaction(trans, root);
 625
 626         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 627
 628         return 0;
 629 }
 630
 631 static void btrfs_dev_replace_update_device_in_mapping_tree(
 632                                                 struct btrfs_fs_info *fs_info,
 633                                                 struct btrfs_device *srcdev,
 634                                                 struct btrfs_device *tgtdev)
 635 {
 636         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 637         struct extent_map *em;
 638         struct map_lookup *map;
 639         u64 start = 0;
 640         int i;
 641
 642         write_lock(&em_tree->lock);
 643         do {
 644                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 645                 if (!em)
 646                         break;
 647                 map = em->map_lookup;
 648                 for (i = 0; i < map->num_stripes; i++)
 649                         if (srcdev == map->stripes[i].dev)
 650                                 map->stripes[i].dev = tgtdev;
 651                 start = em->start + em->len;
 652                 free_extent_map(em);
 653         } while (start);
 654         write_unlock(&em_tree->lock);
 655 }
 656
 657 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 658                               struct btrfs_ioctl_dev_replace_args *args)
 659 {
 660         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 661         struct btrfs_device *srcdev;
 662
 663         btrfs_dev_replace_lock(dev_replace, 0);
 664         /* even if !dev_replace_is_valid, the values are good enough for
 665          * the replace_status ioctl */
 666         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 667         args->status.replace_state = dev_replace->replace_state;
 668         args->status.time_started = dev_replace->time_started;
 669         args->status.time_stopped = dev_replace->time_stopped;
 670         args->status.num_write_errors =
 671                 atomic64_read(&dev_replace->num_write_errors);
 672         args->status.num_uncorrectable_read_errors =
 673                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 674         switch (dev_replace->replace_state) {
 675         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 676         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 677                 args->status.progress_1000 = 0;
 678                 break;
 679         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 680                 args->status.progress_1000 = 1000;
 681                 break;
 682         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 683         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 684                 srcdev = dev_replace->srcdev;
 685                 args->status.progress_1000 = div_u64(dev_replace->cursor_left,
 686                         div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
 687                 break;
 688         }
 689         btrfs_dev_replace_unlock(dev_replace, 0);
 690 }
 691
 692 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
 693                              struct btrfs_ioctl_dev_replace_args *args)
 694 {
 695         args->result = __btrfs_dev_replace_cancel(fs_info);
 696         return 0;
 697 }
 698
 699 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 700 {
 701         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 702         struct btrfs_device *tgt_device = NULL;
 703         struct btrfs_trans_handle *trans;
 704         struct btrfs_root *root = fs_info->tree_root;
 705         u64 result;
 706         int ret;
 707
 708         if (fs_info->sb->s_flags & MS_RDONLY)
 709                 return -EROFS;
 710
 711         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 712         btrfs_dev_replace_lock(dev_replace, 1);
 713         switch (dev_replace->replace_state) {
 714         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 715         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 716         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 717                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 718                 btrfs_dev_replace_unlock(dev_replace, 1);
 719                 goto leave;
 720         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 721         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 722                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 723                 tgt_device = dev_replace->tgtdev;
 724                 dev_replace->tgtdev = NULL;
 725                 dev_replace->srcdev = NULL;
 726                 break;
 727         }
 728         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 729         dev_replace->time_stopped = get_seconds();
 730         dev_replace->item_needs_writeback = 1;
 731         btrfs_dev_replace_unlock(dev_replace, 1);
 732         btrfs_scrub_cancel(fs_info);
 733
 734         trans = btrfs_start_transaction(root, 0);
 735         if (IS_ERR(trans)) {
 736                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 737                 return PTR_ERR(trans);
 738         }
 739         ret = btrfs_commit_transaction(trans, root);
 740         WARN_ON(ret);
 741         if (tgt_device)
 742                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 743
 744 leave:
 745         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 746         return result;
 747 }
 748
 749 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 750 {
 751         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 752
 753         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 754         btrfs_dev_replace_lock(dev_replace, 1);
 755         switch (dev_replace->replace_state) {
 756         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 757         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 758         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 759         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 760                 break;
 761         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 762                 dev_replace->replace_state =
 763                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 764                 dev_replace->time_stopped = get_seconds();
 765                 dev_replace->item_needs_writeback = 1;
 766                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 767                 break;
 768         }
 769
 770         btrfs_dev_replace_unlock(dev_replace, 1);
 771         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 772 }
 773
 774 /* resume dev_replace procedure that was interrupted by unmount */
 775 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 776 {
 777         struct task_struct *task;
 778         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 779
 780         btrfs_dev_replace_lock(dev_replace, 1);
 781         switch (dev_replace->replace_state) {
 782         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 783         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 784         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 785                 btrfs_dev_replace_unlock(dev_replace, 1);
 786                 return 0;
 787         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 788                 break;
 789         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 790                 dev_replace->replace_state =
 791                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 792                 break;
 793         }
 794         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 795                 btrfs_info(fs_info,
 796                            "cannot continue dev_replace, tgtdev is missing");
 797                 btrfs_info(fs_info,
 798                            "you may cancel the operation after 'mount -o degraded'");
 799                 dev_replace->replace_state =
 800                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 801                 btrfs_dev_replace_unlock(dev_replace, 1);
 802                 return 0;
 803         }
 804         btrfs_dev_replace_unlock(dev_replace, 1);
 805
 806         WARN_ON(atomic_xchg(
 807                 &fs_info->mutually_exclusive_operation_running, 1));
 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         struct btrfs_ioctl_dev_replace_args *status_args;
 817         u64 progress;
 818
 819         status_args = kzalloc(sizeof(*status_args), GFP_KERNEL);
 820         if (status_args) {
 821                 btrfs_dev_replace_status(fs_info, status_args);
 822                 progress = status_args->status.progress_1000;
 823                 kfree(status_args);
 824                 progress = div_u64(progress, 10);
 825                 btrfs_info_in_rcu(fs_info,
 826                         "continuing dev_replace from %s (devid %llu) to %s @%u%%",
 827                         dev_replace->srcdev->missing ? "<missing disk>" :
 828                         rcu_str_deref(dev_replace->srcdev->name),
 829                         dev_replace->srcdev->devid,
 830                         dev_replace->tgtdev ?
 831                         rcu_str_deref(dev_replace->tgtdev->name) :
 832                         "<missing target disk>",
 833                         (unsigned int)progress);
 834         }
 835         btrfs_dev_replace_continue_on_mount(fs_info);
 836         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
 837
 838         return 0;
 839 }
 840
 841 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
 842 {
 843         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 844         int ret;
 845
 846         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 847                               dev_replace->committed_cursor_left,
 848                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 849                               &dev_replace->scrub_progress, 0, 1);
 850         ret = btrfs_dev_replace_finishing(fs_info, ret);
 851         WARN_ON(ret);
 852         return 0;
 853 }
 854
 855 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 856 {
 857         if (!dev_replace->is_valid)
 858                 return 0;
 859
 860         switch (dev_replace->replace_state) {
 861         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 862         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 863         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 864                 return 0;
 865         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 866         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 867                 /*
 868                  * return true even if tgtdev is missing (this is
 869                  * something that can happen if the dev_replace
 870                  * procedure is suspended by an umount and then
 871                  * the tgtdev is missing (or "btrfs dev scan") was
 872                  * not called and the the filesystem is remounted
 873                  * in degraded state. This does not stop the
 874                  * dev_replace procedure. It needs to be canceled
 875                  * manually if the cancellation is wanted.
 876                  */
 877                 break;
 878         }
 879         return 1;
 880 }
 881
 882 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
 883 {
 884         if (rw == 1) {
 885                 /* write */
 886 again:
 887                 wait_event(dev_replace->read_lock_wq,
 888                            atomic_read(&dev_replace->blocking_readers) == 0);
 889                 write_lock(&dev_replace->lock);
 890                 if (atomic_read(&dev_replace->blocking_readers)) {
 891                         write_unlock(&dev_replace->lock);
 892                         goto again;
 893                 }
 894         } else {
 895                 read_lock(&dev_replace->lock);
 896                 atomic_inc(&dev_replace->read_locks);
 897         }
 898 }
 899
 900 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
 901 {
 902         if (rw == 1) {
 903                 /* write */
 904                 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
 905                 write_unlock(&dev_replace->lock);
 906         } else {
 907                 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 908                 atomic_dec(&dev_replace->read_locks);
 909                 read_unlock(&dev_replace->lock);
 910         }
 911 }
 912
 913 /* inc blocking cnt and release read lock */
 914 void btrfs_dev_replace_set_lock_blocking(
 915                                         struct btrfs_dev_replace *dev_replace)
 916 {
 917         /* only set blocking for read lock */
 918         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 919         atomic_inc(&dev_replace->blocking_readers);
 920         read_unlock(&dev_replace->lock);
 921 }
 922
 923 /* acquire read lock and dec blocking cnt */
 924 void btrfs_dev_replace_clear_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         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
 930         read_lock(&dev_replace->lock);
 931         if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
 932             waitqueue_active(&dev_replace->read_lock_wq))
 933                 wake_up(&dev_replace->read_lock_wq);
 934 }
 935
 936 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 937 {
 938         percpu_counter_inc(&fs_info->bio_counter);
 939 }
 940
 941 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
 942 {
 943         percpu_counter_sub(&fs_info->bio_counter, amount);
 944
 945         if (waitqueue_active(&fs_info->replace_wait))
 946                 wake_up(&fs_info->replace_wait);
 947 }
 948
 949 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
 950 {
 951         while (1) {
 952                 percpu_counter_inc(&fs_info->bio_counter);
 953                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 954                                      &fs_info->fs_state)))
 955                         break;
 956
 957                 btrfs_bio_counter_dec(fs_info);
 958                 wait_event(fs_info->replace_wait,
 959                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
 960                                      &fs_info->fs_state));
 961         }
 962 }