1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) STRATO AG 2012. All rights reserved. 4 */ 5 6 #include <linux/sched.h> 7 #include <linux/bio.h> 8 #include <linux/slab.h> 9 #include <linux/blkdev.h> 10 #include <linux/kthread.h> 11 #include <linux/math64.h> 12 #include "misc.h" 13 #include "ctree.h" 14 #include "extent_map.h" 15 #include "disk-io.h" 16 #include "transaction.h" 17 #include "print-tree.h" 18 #include "volumes.h" 19 #include "async-thread.h" 20 #include "check-integrity.h" 21 #include "rcu-string.h" 22 #include "dev-replace.h" 23 #include "sysfs.h" 24 #include "zoned.h" 25 #include "block-group.h" 26 #include "fs.h" 27 #include "accessors.h" 28 #include "scrub.h" 29 30 /* 31 * Device replace overview 32 * 33 * [Objective] 34 * To copy all extents (both new and on-disk) from source device to target 35 * device, while still keeping the filesystem read-write. 36 * 37 * [Method] 38 * There are two main methods involved: 39 * 40 * - Write duplication 41 * 42 * All new writes will be written to both target and source devices, so even 43 * if replace gets canceled, sources device still contains up-to-date data. 44 * 45 * Location: handle_ops_on_dev_replace() from __btrfs_map_block() 46 * Start: btrfs_dev_replace_start() 47 * End: btrfs_dev_replace_finishing() 48 * Content: Latest data/metadata 49 * 50 * - Copy existing extents 51 * 52 * This happens by re-using scrub facility, as scrub also iterates through 53 * existing extents from commit root. 54 * 55 * Location: scrub_write_block_to_dev_replace() from 56 * scrub_block_complete() 57 * Content: Data/meta from commit root. 58 * 59 * Due to the content difference, we need to avoid nocow write when dev-replace 60 * is happening. This is done by marking the block group read-only and waiting 61 * for NOCOW writes. 62 * 63 * After replace is done, the finishing part is done by swapping the target and 64 * source devices. 65 * 66 * Location: btrfs_dev_replace_update_device_in_mapping_tree() from 67 * btrfs_dev_replace_finishing() 68 */ 69 70 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 71 int scrub_ret); 72 static int btrfs_dev_replace_kthread(void *data); 73 74 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) 75 { 76 struct btrfs_dev_lookup_args args = { .devid = BTRFS_DEV_REPLACE_DEVID }; 77 struct btrfs_key key; 78 struct btrfs_root *dev_root = fs_info->dev_root; 79 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 80 struct extent_buffer *eb; 81 int slot; 82 int ret = 0; 83 struct btrfs_path *path = NULL; 84 int item_size; 85 struct btrfs_dev_replace_item *ptr; 86 u64 src_devid; 87 88 if (!dev_root) 89 return 0; 90 91 path = btrfs_alloc_path(); 92 if (!path) { 93 ret = -ENOMEM; 94 goto out; 95 } 96 97 key.objectid = 0; 98 key.type = BTRFS_DEV_REPLACE_KEY; 99 key.offset = 0; 100 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); 101 if (ret) { 102 no_valid_dev_replace_entry_found: 103 /* 104 * We don't have a replace item or it's corrupted. If there is 105 * a replace target, fail the mount. 106 */ 107 if (btrfs_find_device(fs_info->fs_devices, &args)) { 108 btrfs_err(fs_info, 109 "found replace target device without a valid replace item"); 110 ret = -EUCLEAN; 111 goto out; 112 } 113 ret = 0; 114 dev_replace->replace_state = 115 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED; 116 dev_replace->cont_reading_from_srcdev_mode = 117 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; 118 dev_replace->time_started = 0; 119 dev_replace->time_stopped = 0; 120 atomic64_set(&dev_replace->num_write_errors, 0); 121 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 122 dev_replace->cursor_left = 0; 123 dev_replace->committed_cursor_left = 0; 124 dev_replace->cursor_left_last_write_of_item = 0; 125 dev_replace->cursor_right = 0; 126 dev_replace->srcdev = NULL; 127 dev_replace->tgtdev = NULL; 128 dev_replace->is_valid = 0; 129 dev_replace->item_needs_writeback = 0; 130 goto out; 131 } 132 slot = path->slots[0]; 133 eb = path->nodes[0]; 134 item_size = btrfs_item_size(eb, slot); 135 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); 136 137 if (item_size != sizeof(struct btrfs_dev_replace_item)) { 138 btrfs_warn(fs_info, 139 "dev_replace entry found has unexpected size, ignore entry"); 140 goto no_valid_dev_replace_entry_found; 141 } 142 143 src_devid = btrfs_dev_replace_src_devid(eb, ptr); 144 dev_replace->cont_reading_from_srcdev_mode = 145 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); 146 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); 147 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); 148 dev_replace->time_stopped = 149 btrfs_dev_replace_time_stopped(eb, ptr); 150 atomic64_set(&dev_replace->num_write_errors, 151 btrfs_dev_replace_num_write_errors(eb, ptr)); 152 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 153 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); 154 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); 155 dev_replace->committed_cursor_left = dev_replace->cursor_left; 156 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; 157 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); 158 dev_replace->is_valid = 1; 159 160 dev_replace->item_needs_writeback = 0; 161 switch (dev_replace->replace_state) { 162 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 163 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 164 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 165 /* 166 * We don't have an active replace item but if there is a 167 * replace target, fail the mount. 168 */ 169 if (btrfs_find_device(fs_info->fs_devices, &args)) { 170 btrfs_err(fs_info, 171 "replace without active item, run 'device scan --forget' on the target device"); 172 ret = -EUCLEAN; 173 } else { 174 dev_replace->srcdev = NULL; 175 dev_replace->tgtdev = NULL; 176 } 177 break; 178 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 179 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 180 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices, &args); 181 args.devid = src_devid; 182 dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices, &args); 183 184 /* 185 * allow 'btrfs dev replace_cancel' if src/tgt device is 186 * missing 187 */ 188 if (!dev_replace->srcdev && 189 !btrfs_test_opt(fs_info, DEGRADED)) { 190 ret = -EIO; 191 btrfs_warn(fs_info, 192 "cannot mount because device replace operation is ongoing and"); 193 btrfs_warn(fs_info, 194 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 195 src_devid); 196 } 197 if (!dev_replace->tgtdev && 198 !btrfs_test_opt(fs_info, DEGRADED)) { 199 ret = -EIO; 200 btrfs_warn(fs_info, 201 "cannot mount because device replace operation is ongoing and"); 202 btrfs_warn(fs_info, 203 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 204 BTRFS_DEV_REPLACE_DEVID); 205 } 206 if (dev_replace->tgtdev) { 207 if (dev_replace->srcdev) { 208 dev_replace->tgtdev->total_bytes = 209 dev_replace->srcdev->total_bytes; 210 dev_replace->tgtdev->disk_total_bytes = 211 dev_replace->srcdev->disk_total_bytes; 212 dev_replace->tgtdev->commit_total_bytes = 213 dev_replace->srcdev->commit_total_bytes; 214 dev_replace->tgtdev->bytes_used = 215 dev_replace->srcdev->bytes_used; 216 dev_replace->tgtdev->commit_bytes_used = 217 dev_replace->srcdev->commit_bytes_used; 218 } 219 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, 220 &dev_replace->tgtdev->dev_state); 221 222 WARN_ON(fs_info->fs_devices->rw_devices == 0); 223 dev_replace->tgtdev->io_width = fs_info->sectorsize; 224 dev_replace->tgtdev->io_align = fs_info->sectorsize; 225 dev_replace->tgtdev->sector_size = fs_info->sectorsize; 226 dev_replace->tgtdev->fs_info = fs_info; 227 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 228 &dev_replace->tgtdev->dev_state); 229 } 230 break; 231 } 232 233 out: 234 btrfs_free_path(path); 235 return ret; 236 } 237 238 /* 239 * Initialize a new device for device replace target from a given source dev 240 * and path. 241 * 242 * Return 0 and new device in @device_out, otherwise return < 0 243 */ 244 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, 245 const char *device_path, 246 struct btrfs_device *srcdev, 247 struct btrfs_device **device_out) 248 { 249 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 250 struct btrfs_device *device; 251 struct block_device *bdev; 252 u64 devid = BTRFS_DEV_REPLACE_DEVID; 253 int ret = 0; 254 255 *device_out = NULL; 256 if (srcdev->fs_devices->seeding) { 257 btrfs_err(fs_info, "the filesystem is a seed filesystem!"); 258 return -EINVAL; 259 } 260 261 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 262 fs_info->bdev_holder); 263 if (IS_ERR(bdev)) { 264 btrfs_err(fs_info, "target device %s is invalid!", device_path); 265 return PTR_ERR(bdev); 266 } 267 268 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 269 btrfs_err(fs_info, 270 "dev-replace: zoned type of target device mismatch with filesystem"); 271 ret = -EINVAL; 272 goto error; 273 } 274 275 sync_blockdev(bdev); 276 277 list_for_each_entry(device, &fs_devices->devices, dev_list) { 278 if (device->bdev == bdev) { 279 btrfs_err(fs_info, 280 "target device is in the filesystem!"); 281 ret = -EEXIST; 282 goto error; 283 } 284 } 285 286 287 if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(srcdev)) { 288 btrfs_err(fs_info, 289 "target device is smaller than source device!"); 290 ret = -EINVAL; 291 goto error; 292 } 293 294 295 device = btrfs_alloc_device(NULL, &devid, NULL, device_path); 296 if (IS_ERR(device)) { 297 ret = PTR_ERR(device); 298 goto error; 299 } 300 301 ret = lookup_bdev(device_path, &device->devt); 302 if (ret) 303 goto error; 304 305 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 306 device->generation = 0; 307 device->io_width = fs_info->sectorsize; 308 device->io_align = fs_info->sectorsize; 309 device->sector_size = fs_info->sectorsize; 310 device->total_bytes = btrfs_device_get_total_bytes(srcdev); 311 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); 312 device->bytes_used = btrfs_device_get_bytes_used(srcdev); 313 device->commit_total_bytes = srcdev->commit_total_bytes; 314 device->commit_bytes_used = device->bytes_used; 315 device->fs_info = fs_info; 316 device->bdev = bdev; 317 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 318 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 319 device->mode = FMODE_EXCL; 320 device->dev_stats_valid = 1; 321 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 322 device->fs_devices = fs_devices; 323 324 ret = btrfs_get_dev_zone_info(device, false); 325 if (ret) 326 goto error; 327 328 mutex_lock(&fs_devices->device_list_mutex); 329 list_add(&device->dev_list, &fs_devices->devices); 330 fs_devices->num_devices++; 331 fs_devices->open_devices++; 332 mutex_unlock(&fs_devices->device_list_mutex); 333 334 *device_out = device; 335 return 0; 336 337 error: 338 blkdev_put(bdev, FMODE_EXCL); 339 return ret; 340 } 341 342 /* 343 * called from commit_transaction. Writes changed device replace state to 344 * disk. 345 */ 346 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans) 347 { 348 struct btrfs_fs_info *fs_info = trans->fs_info; 349 int ret; 350 struct btrfs_root *dev_root = fs_info->dev_root; 351 struct btrfs_path *path; 352 struct btrfs_key key; 353 struct extent_buffer *eb; 354 struct btrfs_dev_replace_item *ptr; 355 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 356 357 down_read(&dev_replace->rwsem); 358 if (!dev_replace->is_valid || 359 !dev_replace->item_needs_writeback) { 360 up_read(&dev_replace->rwsem); 361 return 0; 362 } 363 up_read(&dev_replace->rwsem); 364 365 key.objectid = 0; 366 key.type = BTRFS_DEV_REPLACE_KEY; 367 key.offset = 0; 368 369 path = btrfs_alloc_path(); 370 if (!path) { 371 ret = -ENOMEM; 372 goto out; 373 } 374 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 375 if (ret < 0) { 376 btrfs_warn(fs_info, 377 "error %d while searching for dev_replace item!", 378 ret); 379 goto out; 380 } 381 382 if (ret == 0 && 383 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 384 /* 385 * need to delete old one and insert a new one. 386 * Since no attempt is made to recover any old state, if the 387 * dev_replace state is 'running', the data on the target 388 * drive is lost. 389 * It would be possible to recover the state: just make sure 390 * that the beginning of the item is never changed and always 391 * contains all the essential information. Then read this 392 * minimal set of information and use it as a base for the 393 * new state. 394 */ 395 ret = btrfs_del_item(trans, dev_root, path); 396 if (ret != 0) { 397 btrfs_warn(fs_info, 398 "delete too small dev_replace item failed %d!", 399 ret); 400 goto out; 401 } 402 ret = 1; 403 } 404 405 if (ret == 1) { 406 /* need to insert a new item */ 407 btrfs_release_path(path); 408 ret = btrfs_insert_empty_item(trans, dev_root, path, 409 &key, sizeof(*ptr)); 410 if (ret < 0) { 411 btrfs_warn(fs_info, 412 "insert dev_replace item failed %d!", ret); 413 goto out; 414 } 415 } 416 417 eb = path->nodes[0]; 418 ptr = btrfs_item_ptr(eb, path->slots[0], 419 struct btrfs_dev_replace_item); 420 421 down_write(&dev_replace->rwsem); 422 if (dev_replace->srcdev) 423 btrfs_set_dev_replace_src_devid(eb, ptr, 424 dev_replace->srcdev->devid); 425 else 426 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); 427 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, 428 dev_replace->cont_reading_from_srcdev_mode); 429 btrfs_set_dev_replace_replace_state(eb, ptr, 430 dev_replace->replace_state); 431 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); 432 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); 433 btrfs_set_dev_replace_num_write_errors(eb, ptr, 434 atomic64_read(&dev_replace->num_write_errors)); 435 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, 436 atomic64_read(&dev_replace->num_uncorrectable_read_errors)); 437 dev_replace->cursor_left_last_write_of_item = 438 dev_replace->cursor_left; 439 btrfs_set_dev_replace_cursor_left(eb, ptr, 440 dev_replace->cursor_left_last_write_of_item); 441 btrfs_set_dev_replace_cursor_right(eb, ptr, 442 dev_replace->cursor_right); 443 dev_replace->item_needs_writeback = 0; 444 up_write(&dev_replace->rwsem); 445 446 btrfs_mark_buffer_dirty(eb); 447 448 out: 449 btrfs_free_path(path); 450 451 return ret; 452 } 453 454 static char* btrfs_dev_name(struct btrfs_device *device) 455 { 456 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 457 return "<missing disk>"; 458 else 459 return rcu_str_deref(device->name); 460 } 461 462 static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info, 463 struct btrfs_device *src_dev) 464 { 465 struct btrfs_path *path; 466 struct btrfs_key key; 467 struct btrfs_key found_key; 468 struct btrfs_root *root = fs_info->dev_root; 469 struct btrfs_dev_extent *dev_extent = NULL; 470 struct btrfs_block_group *cache; 471 struct btrfs_trans_handle *trans; 472 int iter_ret = 0; 473 int ret = 0; 474 u64 chunk_offset; 475 476 /* Do not use "to_copy" on non zoned filesystem for now */ 477 if (!btrfs_is_zoned(fs_info)) 478 return 0; 479 480 mutex_lock(&fs_info->chunk_mutex); 481 482 /* Ensure we don't have pending new block group */ 483 spin_lock(&fs_info->trans_lock); 484 while (fs_info->running_transaction && 485 !list_empty(&fs_info->running_transaction->dev_update_list)) { 486 spin_unlock(&fs_info->trans_lock); 487 mutex_unlock(&fs_info->chunk_mutex); 488 trans = btrfs_attach_transaction(root); 489 if (IS_ERR(trans)) { 490 ret = PTR_ERR(trans); 491 mutex_lock(&fs_info->chunk_mutex); 492 if (ret == -ENOENT) { 493 spin_lock(&fs_info->trans_lock); 494 continue; 495 } else { 496 goto unlock; 497 } 498 } 499 500 ret = btrfs_commit_transaction(trans); 501 mutex_lock(&fs_info->chunk_mutex); 502 if (ret) 503 goto unlock; 504 505 spin_lock(&fs_info->trans_lock); 506 } 507 spin_unlock(&fs_info->trans_lock); 508 509 path = btrfs_alloc_path(); 510 if (!path) { 511 ret = -ENOMEM; 512 goto unlock; 513 } 514 515 path->reada = READA_FORWARD; 516 path->search_commit_root = 1; 517 path->skip_locking = 1; 518 519 key.objectid = src_dev->devid; 520 key.type = BTRFS_DEV_EXTENT_KEY; 521 key.offset = 0; 522 523 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 524 struct extent_buffer *leaf = path->nodes[0]; 525 526 if (found_key.objectid != src_dev->devid) 527 break; 528 529 if (found_key.type != BTRFS_DEV_EXTENT_KEY) 530 break; 531 532 if (found_key.offset < key.offset) 533 break; 534 535 dev_extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); 536 537 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent); 538 539 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 540 if (!cache) 541 continue; 542 543 set_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 544 btrfs_put_block_group(cache); 545 } 546 if (iter_ret < 0) 547 ret = iter_ret; 548 549 btrfs_free_path(path); 550 unlock: 551 mutex_unlock(&fs_info->chunk_mutex); 552 553 return ret; 554 } 555 556 bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev, 557 struct btrfs_block_group *cache, 558 u64 physical) 559 { 560 struct btrfs_fs_info *fs_info = cache->fs_info; 561 struct extent_map *em; 562 struct map_lookup *map; 563 u64 chunk_offset = cache->start; 564 int num_extents, cur_extent; 565 int i; 566 567 /* Do not use "to_copy" on non zoned filesystem for now */ 568 if (!btrfs_is_zoned(fs_info)) 569 return true; 570 571 spin_lock(&cache->lock); 572 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags)) { 573 spin_unlock(&cache->lock); 574 return true; 575 } 576 spin_unlock(&cache->lock); 577 578 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 579 ASSERT(!IS_ERR(em)); 580 map = em->map_lookup; 581 582 num_extents = 0; 583 cur_extent = 0; 584 for (i = 0; i < map->num_stripes; i++) { 585 /* We have more device extent to copy */ 586 if (srcdev != map->stripes[i].dev) 587 continue; 588 589 num_extents++; 590 if (physical == map->stripes[i].physical) 591 cur_extent = i; 592 } 593 594 free_extent_map(em); 595 596 if (num_extents > 1 && cur_extent < num_extents - 1) { 597 /* 598 * Has more stripes on this device. Keep this block group 599 * readonly until we finish all the stripes. 600 */ 601 return false; 602 } 603 604 /* Last stripe on this device */ 605 clear_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 606 607 return true; 608 } 609 610 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info, 611 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name, 612 int read_src) 613 { 614 struct btrfs_root *root = fs_info->dev_root; 615 struct btrfs_trans_handle *trans; 616 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 617 int ret; 618 struct btrfs_device *tgt_device = NULL; 619 struct btrfs_device *src_device = NULL; 620 621 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid, 622 srcdev_name); 623 if (IS_ERR(src_device)) 624 return PTR_ERR(src_device); 625 626 if (btrfs_pinned_by_swapfile(fs_info, src_device)) { 627 btrfs_warn_in_rcu(fs_info, 628 "cannot replace device %s (devid %llu) due to active swapfile", 629 btrfs_dev_name(src_device), src_device->devid); 630 return -ETXTBSY; 631 } 632 633 /* 634 * Here we commit the transaction to make sure commit_total_bytes 635 * of all the devices are updated. 636 */ 637 trans = btrfs_attach_transaction(root); 638 if (!IS_ERR(trans)) { 639 ret = btrfs_commit_transaction(trans); 640 if (ret) 641 return ret; 642 } else if (PTR_ERR(trans) != -ENOENT) { 643 return PTR_ERR(trans); 644 } 645 646 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name, 647 src_device, &tgt_device); 648 if (ret) 649 return ret; 650 651 ret = mark_block_group_to_copy(fs_info, src_device); 652 if (ret) 653 return ret; 654 655 down_write(&dev_replace->rwsem); 656 switch (dev_replace->replace_state) { 657 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 658 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 659 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 660 break; 661 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 662 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 663 ASSERT(0); 664 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 665 up_write(&dev_replace->rwsem); 666 goto leave; 667 } 668 669 dev_replace->cont_reading_from_srcdev_mode = read_src; 670 dev_replace->srcdev = src_device; 671 dev_replace->tgtdev = tgt_device; 672 673 btrfs_info_in_rcu(fs_info, 674 "dev_replace from %s (devid %llu) to %s started", 675 btrfs_dev_name(src_device), 676 src_device->devid, 677 rcu_str_deref(tgt_device->name)); 678 679 /* 680 * from now on, the writes to the srcdev are all duplicated to 681 * go to the tgtdev as well (refer to btrfs_map_block()). 682 */ 683 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 684 dev_replace->time_started = ktime_get_real_seconds(); 685 dev_replace->cursor_left = 0; 686 dev_replace->committed_cursor_left = 0; 687 dev_replace->cursor_left_last_write_of_item = 0; 688 dev_replace->cursor_right = 0; 689 dev_replace->is_valid = 1; 690 dev_replace->item_needs_writeback = 1; 691 atomic64_set(&dev_replace->num_write_errors, 0); 692 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 693 up_write(&dev_replace->rwsem); 694 695 ret = btrfs_sysfs_add_device(tgt_device); 696 if (ret) 697 btrfs_err(fs_info, "kobj add dev failed %d", ret); 698 699 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 700 701 /* 702 * Commit dev_replace state and reserve 1 item for it. 703 * This is crucial to ensure we won't miss copying extents for new block 704 * groups that are allocated after we started the device replace, and 705 * must be done after setting up the device replace state. 706 */ 707 trans = btrfs_start_transaction(root, 1); 708 if (IS_ERR(trans)) { 709 ret = PTR_ERR(trans); 710 down_write(&dev_replace->rwsem); 711 dev_replace->replace_state = 712 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED; 713 dev_replace->srcdev = NULL; 714 dev_replace->tgtdev = NULL; 715 up_write(&dev_replace->rwsem); 716 goto leave; 717 } 718 719 ret = btrfs_commit_transaction(trans); 720 WARN_ON(ret); 721 722 /* the disk copy procedure reuses the scrub code */ 723 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 724 btrfs_device_get_total_bytes(src_device), 725 &dev_replace->scrub_progress, 0, 1); 726 727 ret = btrfs_dev_replace_finishing(fs_info, ret); 728 if (ret == -EINPROGRESS) 729 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS; 730 731 return ret; 732 733 leave: 734 btrfs_destroy_dev_replace_tgtdev(tgt_device); 735 return ret; 736 } 737 738 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info, 739 struct btrfs_ioctl_dev_replace_args *args) 740 { 741 int ret; 742 743 switch (args->start.cont_reading_from_srcdev_mode) { 744 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 745 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 746 break; 747 default: 748 return -EINVAL; 749 } 750 751 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 752 args->start.tgtdev_name[0] == '\0') 753 return -EINVAL; 754 755 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name, 756 args->start.srcdevid, 757 args->start.srcdev_name, 758 args->start.cont_reading_from_srcdev_mode); 759 args->result = ret; 760 /* don't warn if EINPROGRESS, someone else might be running scrub */ 761 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS || 762 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR) 763 return 0; 764 765 return ret; 766 } 767 768 /* 769 * blocked until all in-flight bios operations are finished. 770 */ 771 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) 772 { 773 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 774 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum( 775 &fs_info->dev_replace.bio_counter)); 776 } 777 778 /* 779 * we have removed target device, it is safe to allow new bios request. 780 */ 781 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) 782 { 783 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 784 wake_up(&fs_info->dev_replace.replace_wait); 785 } 786 787 /* 788 * When finishing the device replace, before swapping the source device with the 789 * target device we must update the chunk allocation state in the target device, 790 * as it is empty because replace works by directly copying the chunks and not 791 * through the normal chunk allocation path. 792 */ 793 static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev, 794 struct btrfs_device *tgtdev) 795 { 796 struct extent_state *cached_state = NULL; 797 u64 start = 0; 798 u64 found_start; 799 u64 found_end; 800 int ret = 0; 801 802 lockdep_assert_held(&srcdev->fs_info->chunk_mutex); 803 804 while (!find_first_extent_bit(&srcdev->alloc_state, start, 805 &found_start, &found_end, 806 CHUNK_ALLOCATED, &cached_state)) { 807 ret = set_extent_bits(&tgtdev->alloc_state, found_start, 808 found_end, CHUNK_ALLOCATED); 809 if (ret) 810 break; 811 start = found_end + 1; 812 } 813 814 free_extent_state(cached_state); 815 return ret; 816 } 817 818 static void btrfs_dev_replace_update_device_in_mapping_tree( 819 struct btrfs_fs_info *fs_info, 820 struct btrfs_device *srcdev, 821 struct btrfs_device *tgtdev) 822 { 823 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 824 struct extent_map *em; 825 struct map_lookup *map; 826 u64 start = 0; 827 int i; 828 829 write_lock(&em_tree->lock); 830 do { 831 em = lookup_extent_mapping(em_tree, start, (u64)-1); 832 if (!em) 833 break; 834 map = em->map_lookup; 835 for (i = 0; i < map->num_stripes; i++) 836 if (srcdev == map->stripes[i].dev) 837 map->stripes[i].dev = tgtdev; 838 start = em->start + em->len; 839 free_extent_map(em); 840 } while (start); 841 write_unlock(&em_tree->lock); 842 } 843 844 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 845 int scrub_ret) 846 { 847 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 848 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 849 struct btrfs_device *tgt_device; 850 struct btrfs_device *src_device; 851 struct btrfs_root *root = fs_info->tree_root; 852 u8 uuid_tmp[BTRFS_UUID_SIZE]; 853 struct btrfs_trans_handle *trans; 854 int ret = 0; 855 856 /* don't allow cancel or unmount to disturb the finishing procedure */ 857 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 858 859 down_read(&dev_replace->rwsem); 860 /* was the operation canceled, or is it finished? */ 861 if (dev_replace->replace_state != 862 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 863 up_read(&dev_replace->rwsem); 864 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 865 return 0; 866 } 867 868 tgt_device = dev_replace->tgtdev; 869 src_device = dev_replace->srcdev; 870 up_read(&dev_replace->rwsem); 871 872 /* 873 * flush all outstanding I/O and inode extent mappings before the 874 * copy operation is declared as being finished 875 */ 876 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false); 877 if (ret) { 878 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 879 return ret; 880 } 881 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 882 883 /* 884 * We have to use this loop approach because at this point src_device 885 * has to be available for transaction commit to complete, yet new 886 * chunks shouldn't be allocated on the device. 887 */ 888 while (1) { 889 trans = btrfs_start_transaction(root, 0); 890 if (IS_ERR(trans)) { 891 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 892 return PTR_ERR(trans); 893 } 894 ret = btrfs_commit_transaction(trans); 895 WARN_ON(ret); 896 897 /* Prevent write_all_supers() during the finishing procedure */ 898 mutex_lock(&fs_devices->device_list_mutex); 899 /* Prevent new chunks being allocated on the source device */ 900 mutex_lock(&fs_info->chunk_mutex); 901 902 if (!list_empty(&src_device->post_commit_list)) { 903 mutex_unlock(&fs_devices->device_list_mutex); 904 mutex_unlock(&fs_info->chunk_mutex); 905 } else { 906 break; 907 } 908 } 909 910 down_write(&dev_replace->rwsem); 911 dev_replace->replace_state = 912 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 913 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 914 dev_replace->tgtdev = NULL; 915 dev_replace->srcdev = NULL; 916 dev_replace->time_stopped = ktime_get_real_seconds(); 917 dev_replace->item_needs_writeback = 1; 918 919 /* 920 * Update allocation state in the new device and replace the old device 921 * with the new one in the mapping tree. 922 */ 923 if (!scrub_ret) { 924 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device); 925 if (scrub_ret) 926 goto error; 927 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 928 src_device, 929 tgt_device); 930 } else { 931 if (scrub_ret != -ECANCELED) 932 btrfs_err_in_rcu(fs_info, 933 "btrfs_scrub_dev(%s, %llu, %s) failed %d", 934 btrfs_dev_name(src_device), 935 src_device->devid, 936 rcu_str_deref(tgt_device->name), scrub_ret); 937 error: 938 up_write(&dev_replace->rwsem); 939 mutex_unlock(&fs_info->chunk_mutex); 940 mutex_unlock(&fs_devices->device_list_mutex); 941 btrfs_rm_dev_replace_blocked(fs_info); 942 if (tgt_device) 943 btrfs_destroy_dev_replace_tgtdev(tgt_device); 944 btrfs_rm_dev_replace_unblocked(fs_info); 945 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 946 947 return scrub_ret; 948 } 949 950 btrfs_info_in_rcu(fs_info, 951 "dev_replace from %s (devid %llu) to %s finished", 952 btrfs_dev_name(src_device), 953 src_device->devid, 954 rcu_str_deref(tgt_device->name)); 955 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state); 956 tgt_device->devid = src_device->devid; 957 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 958 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 959 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 960 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 961 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes); 962 btrfs_device_set_disk_total_bytes(tgt_device, 963 src_device->disk_total_bytes); 964 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used); 965 tgt_device->commit_bytes_used = src_device->bytes_used; 966 967 btrfs_assign_next_active_device(src_device, tgt_device); 968 969 list_add(&tgt_device->dev_alloc_list, &fs_devices->alloc_list); 970 fs_devices->rw_devices++; 971 972 up_write(&dev_replace->rwsem); 973 btrfs_rm_dev_replace_blocked(fs_info); 974 975 btrfs_rm_dev_replace_remove_srcdev(src_device); 976 977 btrfs_rm_dev_replace_unblocked(fs_info); 978 979 /* 980 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will 981 * update on-disk dev stats value during commit transaction 982 */ 983 atomic_inc(&tgt_device->dev_stats_ccnt); 984 985 /* 986 * this is again a consistent state where no dev_replace procedure 987 * is running, the target device is part of the filesystem, the 988 * source device is not part of the filesystem anymore and its 1st 989 * superblock is scratched out so that it is no longer marked to 990 * belong to this filesystem. 991 */ 992 mutex_unlock(&fs_info->chunk_mutex); 993 mutex_unlock(&fs_devices->device_list_mutex); 994 995 /* replace the sysfs entry */ 996 btrfs_sysfs_remove_device(src_device); 997 btrfs_sysfs_update_devid(tgt_device); 998 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state)) 999 btrfs_scratch_superblocks(fs_info, src_device->bdev, 1000 src_device->name->str); 1001 1002 /* write back the superblocks */ 1003 trans = btrfs_start_transaction(root, 0); 1004 if (!IS_ERR(trans)) 1005 btrfs_commit_transaction(trans); 1006 1007 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1008 1009 btrfs_rm_dev_replace_free_srcdev(src_device); 1010 1011 return 0; 1012 } 1013 1014 /* 1015 * Read progress of device replace status according to the state and last 1016 * stored position. The value format is the same as for 1017 * btrfs_dev_replace::progress_1000 1018 */ 1019 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info) 1020 { 1021 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1022 u64 ret = 0; 1023 1024 switch (dev_replace->replace_state) { 1025 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1026 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1027 ret = 0; 1028 break; 1029 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1030 ret = 1000; 1031 break; 1032 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1033 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1034 ret = div64_u64(dev_replace->cursor_left, 1035 div_u64(btrfs_device_get_total_bytes( 1036 dev_replace->srcdev), 1000)); 1037 break; 1038 } 1039 1040 return ret; 1041 } 1042 1043 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 1044 struct btrfs_ioctl_dev_replace_args *args) 1045 { 1046 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1047 1048 down_read(&dev_replace->rwsem); 1049 /* even if !dev_replace_is_valid, the values are good enough for 1050 * the replace_status ioctl */ 1051 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 1052 args->status.replace_state = dev_replace->replace_state; 1053 args->status.time_started = dev_replace->time_started; 1054 args->status.time_stopped = dev_replace->time_stopped; 1055 args->status.num_write_errors = 1056 atomic64_read(&dev_replace->num_write_errors); 1057 args->status.num_uncorrectable_read_errors = 1058 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 1059 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info); 1060 up_read(&dev_replace->rwsem); 1061 } 1062 1063 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 1064 { 1065 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1066 struct btrfs_device *tgt_device = NULL; 1067 struct btrfs_device *src_device = NULL; 1068 struct btrfs_trans_handle *trans; 1069 struct btrfs_root *root = fs_info->tree_root; 1070 int result; 1071 int ret; 1072 1073 if (sb_rdonly(fs_info->sb)) 1074 return -EROFS; 1075 1076 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 1077 down_write(&dev_replace->rwsem); 1078 switch (dev_replace->replace_state) { 1079 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1080 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1081 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1082 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 1083 up_write(&dev_replace->rwsem); 1084 break; 1085 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1086 tgt_device = dev_replace->tgtdev; 1087 src_device = dev_replace->srcdev; 1088 up_write(&dev_replace->rwsem); 1089 ret = btrfs_scrub_cancel(fs_info); 1090 if (ret < 0) { 1091 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 1092 } else { 1093 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 1094 /* 1095 * btrfs_dev_replace_finishing() will handle the 1096 * cleanup part 1097 */ 1098 btrfs_info_in_rcu(fs_info, 1099 "dev_replace from %s (devid %llu) to %s canceled", 1100 btrfs_dev_name(src_device), src_device->devid, 1101 btrfs_dev_name(tgt_device)); 1102 } 1103 break; 1104 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1105 /* 1106 * Scrub doing the replace isn't running so we need to do the 1107 * cleanup step of btrfs_dev_replace_finishing() here 1108 */ 1109 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 1110 tgt_device = dev_replace->tgtdev; 1111 src_device = dev_replace->srcdev; 1112 dev_replace->tgtdev = NULL; 1113 dev_replace->srcdev = NULL; 1114 dev_replace->replace_state = 1115 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 1116 dev_replace->time_stopped = ktime_get_real_seconds(); 1117 dev_replace->item_needs_writeback = 1; 1118 1119 up_write(&dev_replace->rwsem); 1120 1121 /* Scrub for replace must not be running in suspended state */ 1122 btrfs_scrub_cancel(fs_info); 1123 1124 trans = btrfs_start_transaction(root, 0); 1125 if (IS_ERR(trans)) { 1126 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1127 return PTR_ERR(trans); 1128 } 1129 ret = btrfs_commit_transaction(trans); 1130 WARN_ON(ret); 1131 1132 btrfs_info_in_rcu(fs_info, 1133 "suspended dev_replace from %s (devid %llu) to %s canceled", 1134 btrfs_dev_name(src_device), src_device->devid, 1135 btrfs_dev_name(tgt_device)); 1136 1137 if (tgt_device) 1138 btrfs_destroy_dev_replace_tgtdev(tgt_device); 1139 break; 1140 default: 1141 up_write(&dev_replace->rwsem); 1142 result = -EINVAL; 1143 } 1144 1145 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1146 return result; 1147 } 1148 1149 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 1150 { 1151 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1152 1153 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 1154 down_write(&dev_replace->rwsem); 1155 1156 switch (dev_replace->replace_state) { 1157 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1158 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1159 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1160 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1161 break; 1162 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1163 dev_replace->replace_state = 1164 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 1165 dev_replace->time_stopped = ktime_get_real_seconds(); 1166 dev_replace->item_needs_writeback = 1; 1167 btrfs_info(fs_info, "suspending dev_replace for unmount"); 1168 break; 1169 } 1170 1171 up_write(&dev_replace->rwsem); 1172 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1173 } 1174 1175 /* resume dev_replace procedure that was interrupted by unmount */ 1176 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 1177 { 1178 struct task_struct *task; 1179 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1180 1181 down_write(&dev_replace->rwsem); 1182 1183 switch (dev_replace->replace_state) { 1184 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1185 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1186 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1187 up_write(&dev_replace->rwsem); 1188 return 0; 1189 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1190 break; 1191 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1192 dev_replace->replace_state = 1193 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 1194 break; 1195 } 1196 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 1197 btrfs_info(fs_info, 1198 "cannot continue dev_replace, tgtdev is missing"); 1199 btrfs_info(fs_info, 1200 "you may cancel the operation after 'mount -o degraded'"); 1201 dev_replace->replace_state = 1202 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 1203 up_write(&dev_replace->rwsem); 1204 return 0; 1205 } 1206 up_write(&dev_replace->rwsem); 1207 1208 /* 1209 * This could collide with a paused balance, but the exclusive op logic 1210 * should never allow both to start and pause. We don't want to allow 1211 * dev-replace to start anyway. 1212 */ 1213 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) { 1214 down_write(&dev_replace->rwsem); 1215 dev_replace->replace_state = 1216 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 1217 up_write(&dev_replace->rwsem); 1218 btrfs_info(fs_info, 1219 "cannot resume dev-replace, other exclusive operation running"); 1220 return 0; 1221 } 1222 1223 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 1224 return PTR_ERR_OR_ZERO(task); 1225 } 1226 1227 static int btrfs_dev_replace_kthread(void *data) 1228 { 1229 struct btrfs_fs_info *fs_info = data; 1230 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1231 u64 progress; 1232 int ret; 1233 1234 progress = btrfs_dev_replace_progress(fs_info); 1235 progress = div_u64(progress, 10); 1236 btrfs_info_in_rcu(fs_info, 1237 "continuing dev_replace from %s (devid %llu) to target %s @%u%%", 1238 btrfs_dev_name(dev_replace->srcdev), 1239 dev_replace->srcdev->devid, 1240 btrfs_dev_name(dev_replace->tgtdev), 1241 (unsigned int)progress); 1242 1243 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 1244 dev_replace->committed_cursor_left, 1245 btrfs_device_get_total_bytes(dev_replace->srcdev), 1246 &dev_replace->scrub_progress, 0, 1); 1247 ret = btrfs_dev_replace_finishing(fs_info, ret); 1248 WARN_ON(ret && ret != -ECANCELED); 1249 1250 btrfs_exclop_finish(fs_info); 1251 return 0; 1252 } 1253 1254 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 1255 { 1256 if (!dev_replace->is_valid) 1257 return 0; 1258 1259 switch (dev_replace->replace_state) { 1260 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1261 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1262 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1263 return 0; 1264 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1265 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1266 /* 1267 * return true even if tgtdev is missing (this is 1268 * something that can happen if the dev_replace 1269 * procedure is suspended by an umount and then 1270 * the tgtdev is missing (or "btrfs dev scan") was 1271 * not called and the filesystem is remounted 1272 * in degraded state. This does not stop the 1273 * dev_replace procedure. It needs to be canceled 1274 * manually if the cancellation is wanted. 1275 */ 1276 break; 1277 } 1278 return 1; 1279 } 1280 1281 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount) 1282 { 1283 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount); 1284 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait); 1285 } 1286 1287 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) 1288 { 1289 while (1) { 1290 percpu_counter_inc(&fs_info->dev_replace.bio_counter); 1291 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1292 &fs_info->fs_state))) 1293 break; 1294 1295 btrfs_bio_counter_dec(fs_info); 1296 wait_event(fs_info->dev_replace.replace_wait, 1297 !test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1298 &fs_info->fs_state)); 1299 } 1300 } 1301