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 struct rcu_string *name; 253 u64 devid = BTRFS_DEV_REPLACE_DEVID; 254 int ret = 0; 255 256 *device_out = NULL; 257 if (srcdev->fs_devices->seeding) { 258 btrfs_err(fs_info, "the filesystem is a seed filesystem!"); 259 return -EINVAL; 260 } 261 262 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 263 fs_info->bdev_holder); 264 if (IS_ERR(bdev)) { 265 btrfs_err(fs_info, "target device %s is invalid!", device_path); 266 return PTR_ERR(bdev); 267 } 268 269 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 270 btrfs_err(fs_info, 271 "dev-replace: zoned type of target device mismatch with filesystem"); 272 ret = -EINVAL; 273 goto error; 274 } 275 276 sync_blockdev(bdev); 277 278 list_for_each_entry(device, &fs_devices->devices, dev_list) { 279 if (device->bdev == bdev) { 280 btrfs_err(fs_info, 281 "target device is in the filesystem!"); 282 ret = -EEXIST; 283 goto error; 284 } 285 } 286 287 288 if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(srcdev)) { 289 btrfs_err(fs_info, 290 "target device is smaller than source device!"); 291 ret = -EINVAL; 292 goto error; 293 } 294 295 296 device = btrfs_alloc_device(NULL, &devid, NULL); 297 if (IS_ERR(device)) { 298 ret = PTR_ERR(device); 299 goto error; 300 } 301 302 name = rcu_string_strdup(device_path, GFP_KERNEL); 303 if (!name) { 304 btrfs_free_device(device); 305 ret = -ENOMEM; 306 goto error; 307 } 308 rcu_assign_pointer(device->name, name); 309 ret = lookup_bdev(device_path, &device->devt); 310 if (ret) 311 goto error; 312 313 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 314 device->generation = 0; 315 device->io_width = fs_info->sectorsize; 316 device->io_align = fs_info->sectorsize; 317 device->sector_size = fs_info->sectorsize; 318 device->total_bytes = btrfs_device_get_total_bytes(srcdev); 319 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); 320 device->bytes_used = btrfs_device_get_bytes_used(srcdev); 321 device->commit_total_bytes = srcdev->commit_total_bytes; 322 device->commit_bytes_used = device->bytes_used; 323 device->fs_info = fs_info; 324 device->bdev = bdev; 325 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 326 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 327 device->mode = FMODE_EXCL; 328 device->dev_stats_valid = 1; 329 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 330 device->fs_devices = fs_devices; 331 332 ret = btrfs_get_dev_zone_info(device, false); 333 if (ret) 334 goto error; 335 336 mutex_lock(&fs_devices->device_list_mutex); 337 list_add(&device->dev_list, &fs_devices->devices); 338 fs_devices->num_devices++; 339 fs_devices->open_devices++; 340 mutex_unlock(&fs_devices->device_list_mutex); 341 342 *device_out = device; 343 return 0; 344 345 error: 346 blkdev_put(bdev, FMODE_EXCL); 347 return ret; 348 } 349 350 /* 351 * called from commit_transaction. Writes changed device replace state to 352 * disk. 353 */ 354 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans) 355 { 356 struct btrfs_fs_info *fs_info = trans->fs_info; 357 int ret; 358 struct btrfs_root *dev_root = fs_info->dev_root; 359 struct btrfs_path *path; 360 struct btrfs_key key; 361 struct extent_buffer *eb; 362 struct btrfs_dev_replace_item *ptr; 363 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 364 365 down_read(&dev_replace->rwsem); 366 if (!dev_replace->is_valid || 367 !dev_replace->item_needs_writeback) { 368 up_read(&dev_replace->rwsem); 369 return 0; 370 } 371 up_read(&dev_replace->rwsem); 372 373 key.objectid = 0; 374 key.type = BTRFS_DEV_REPLACE_KEY; 375 key.offset = 0; 376 377 path = btrfs_alloc_path(); 378 if (!path) { 379 ret = -ENOMEM; 380 goto out; 381 } 382 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 383 if (ret < 0) { 384 btrfs_warn(fs_info, 385 "error %d while searching for dev_replace item!", 386 ret); 387 goto out; 388 } 389 390 if (ret == 0 && 391 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 392 /* 393 * need to delete old one and insert a new one. 394 * Since no attempt is made to recover any old state, if the 395 * dev_replace state is 'running', the data on the target 396 * drive is lost. 397 * It would be possible to recover the state: just make sure 398 * that the beginning of the item is never changed and always 399 * contains all the essential information. Then read this 400 * minimal set of information and use it as a base for the 401 * new state. 402 */ 403 ret = btrfs_del_item(trans, dev_root, path); 404 if (ret != 0) { 405 btrfs_warn(fs_info, 406 "delete too small dev_replace item failed %d!", 407 ret); 408 goto out; 409 } 410 ret = 1; 411 } 412 413 if (ret == 1) { 414 /* need to insert a new item */ 415 btrfs_release_path(path); 416 ret = btrfs_insert_empty_item(trans, dev_root, path, 417 &key, sizeof(*ptr)); 418 if (ret < 0) { 419 btrfs_warn(fs_info, 420 "insert dev_replace item failed %d!", ret); 421 goto out; 422 } 423 } 424 425 eb = path->nodes[0]; 426 ptr = btrfs_item_ptr(eb, path->slots[0], 427 struct btrfs_dev_replace_item); 428 429 down_write(&dev_replace->rwsem); 430 if (dev_replace->srcdev) 431 btrfs_set_dev_replace_src_devid(eb, ptr, 432 dev_replace->srcdev->devid); 433 else 434 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); 435 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, 436 dev_replace->cont_reading_from_srcdev_mode); 437 btrfs_set_dev_replace_replace_state(eb, ptr, 438 dev_replace->replace_state); 439 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); 440 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); 441 btrfs_set_dev_replace_num_write_errors(eb, ptr, 442 atomic64_read(&dev_replace->num_write_errors)); 443 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, 444 atomic64_read(&dev_replace->num_uncorrectable_read_errors)); 445 dev_replace->cursor_left_last_write_of_item = 446 dev_replace->cursor_left; 447 btrfs_set_dev_replace_cursor_left(eb, ptr, 448 dev_replace->cursor_left_last_write_of_item); 449 btrfs_set_dev_replace_cursor_right(eb, ptr, 450 dev_replace->cursor_right); 451 dev_replace->item_needs_writeback = 0; 452 up_write(&dev_replace->rwsem); 453 454 btrfs_mark_buffer_dirty(eb); 455 456 out: 457 btrfs_free_path(path); 458 459 return ret; 460 } 461 462 static char* btrfs_dev_name(struct btrfs_device *device) 463 { 464 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 465 return "<missing disk>"; 466 else 467 return rcu_str_deref(device->name); 468 } 469 470 static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info, 471 struct btrfs_device *src_dev) 472 { 473 struct btrfs_path *path; 474 struct btrfs_key key; 475 struct btrfs_key found_key; 476 struct btrfs_root *root = fs_info->dev_root; 477 struct btrfs_dev_extent *dev_extent = NULL; 478 struct btrfs_block_group *cache; 479 struct btrfs_trans_handle *trans; 480 int iter_ret = 0; 481 int ret = 0; 482 u64 chunk_offset; 483 484 /* Do not use "to_copy" on non zoned filesystem for now */ 485 if (!btrfs_is_zoned(fs_info)) 486 return 0; 487 488 mutex_lock(&fs_info->chunk_mutex); 489 490 /* Ensure we don't have pending new block group */ 491 spin_lock(&fs_info->trans_lock); 492 while (fs_info->running_transaction && 493 !list_empty(&fs_info->running_transaction->dev_update_list)) { 494 spin_unlock(&fs_info->trans_lock); 495 mutex_unlock(&fs_info->chunk_mutex); 496 trans = btrfs_attach_transaction(root); 497 if (IS_ERR(trans)) { 498 ret = PTR_ERR(trans); 499 mutex_lock(&fs_info->chunk_mutex); 500 if (ret == -ENOENT) { 501 spin_lock(&fs_info->trans_lock); 502 continue; 503 } else { 504 goto unlock; 505 } 506 } 507 508 ret = btrfs_commit_transaction(trans); 509 mutex_lock(&fs_info->chunk_mutex); 510 if (ret) 511 goto unlock; 512 513 spin_lock(&fs_info->trans_lock); 514 } 515 spin_unlock(&fs_info->trans_lock); 516 517 path = btrfs_alloc_path(); 518 if (!path) { 519 ret = -ENOMEM; 520 goto unlock; 521 } 522 523 path->reada = READA_FORWARD; 524 path->search_commit_root = 1; 525 path->skip_locking = 1; 526 527 key.objectid = src_dev->devid; 528 key.type = BTRFS_DEV_EXTENT_KEY; 529 key.offset = 0; 530 531 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 532 struct extent_buffer *leaf = path->nodes[0]; 533 534 if (found_key.objectid != src_dev->devid) 535 break; 536 537 if (found_key.type != BTRFS_DEV_EXTENT_KEY) 538 break; 539 540 if (found_key.offset < key.offset) 541 break; 542 543 dev_extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); 544 545 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent); 546 547 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 548 if (!cache) 549 continue; 550 551 set_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 552 btrfs_put_block_group(cache); 553 } 554 if (iter_ret < 0) 555 ret = iter_ret; 556 557 btrfs_free_path(path); 558 unlock: 559 mutex_unlock(&fs_info->chunk_mutex); 560 561 return ret; 562 } 563 564 bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev, 565 struct btrfs_block_group *cache, 566 u64 physical) 567 { 568 struct btrfs_fs_info *fs_info = cache->fs_info; 569 struct extent_map *em; 570 struct map_lookup *map; 571 u64 chunk_offset = cache->start; 572 int num_extents, cur_extent; 573 int i; 574 575 /* Do not use "to_copy" on non zoned filesystem for now */ 576 if (!btrfs_is_zoned(fs_info)) 577 return true; 578 579 spin_lock(&cache->lock); 580 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags)) { 581 spin_unlock(&cache->lock); 582 return true; 583 } 584 spin_unlock(&cache->lock); 585 586 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 587 ASSERT(!IS_ERR(em)); 588 map = em->map_lookup; 589 590 num_extents = 0; 591 cur_extent = 0; 592 for (i = 0; i < map->num_stripes; i++) { 593 /* We have more device extent to copy */ 594 if (srcdev != map->stripes[i].dev) 595 continue; 596 597 num_extents++; 598 if (physical == map->stripes[i].physical) 599 cur_extent = i; 600 } 601 602 free_extent_map(em); 603 604 if (num_extents > 1 && cur_extent < num_extents - 1) { 605 /* 606 * Has more stripes on this device. Keep this block group 607 * readonly until we finish all the stripes. 608 */ 609 return false; 610 } 611 612 /* Last stripe on this device */ 613 clear_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 614 615 return true; 616 } 617 618 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info, 619 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name, 620 int read_src) 621 { 622 struct btrfs_root *root = fs_info->dev_root; 623 struct btrfs_trans_handle *trans; 624 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 625 int ret; 626 struct btrfs_device *tgt_device = NULL; 627 struct btrfs_device *src_device = NULL; 628 629 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid, 630 srcdev_name); 631 if (IS_ERR(src_device)) 632 return PTR_ERR(src_device); 633 634 if (btrfs_pinned_by_swapfile(fs_info, src_device)) { 635 btrfs_warn_in_rcu(fs_info, 636 "cannot replace device %s (devid %llu) due to active swapfile", 637 btrfs_dev_name(src_device), src_device->devid); 638 return -ETXTBSY; 639 } 640 641 /* 642 * Here we commit the transaction to make sure commit_total_bytes 643 * of all the devices are updated. 644 */ 645 trans = btrfs_attach_transaction(root); 646 if (!IS_ERR(trans)) { 647 ret = btrfs_commit_transaction(trans); 648 if (ret) 649 return ret; 650 } else if (PTR_ERR(trans) != -ENOENT) { 651 return PTR_ERR(trans); 652 } 653 654 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name, 655 src_device, &tgt_device); 656 if (ret) 657 return ret; 658 659 ret = mark_block_group_to_copy(fs_info, src_device); 660 if (ret) 661 return ret; 662 663 down_write(&dev_replace->rwsem); 664 switch (dev_replace->replace_state) { 665 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 666 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 667 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 668 break; 669 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 670 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 671 ASSERT(0); 672 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 673 up_write(&dev_replace->rwsem); 674 goto leave; 675 } 676 677 dev_replace->cont_reading_from_srcdev_mode = read_src; 678 dev_replace->srcdev = src_device; 679 dev_replace->tgtdev = tgt_device; 680 681 btrfs_info_in_rcu(fs_info, 682 "dev_replace from %s (devid %llu) to %s started", 683 btrfs_dev_name(src_device), 684 src_device->devid, 685 rcu_str_deref(tgt_device->name)); 686 687 /* 688 * from now on, the writes to the srcdev are all duplicated to 689 * go to the tgtdev as well (refer to btrfs_map_block()). 690 */ 691 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 692 dev_replace->time_started = ktime_get_real_seconds(); 693 dev_replace->cursor_left = 0; 694 dev_replace->committed_cursor_left = 0; 695 dev_replace->cursor_left_last_write_of_item = 0; 696 dev_replace->cursor_right = 0; 697 dev_replace->is_valid = 1; 698 dev_replace->item_needs_writeback = 1; 699 atomic64_set(&dev_replace->num_write_errors, 0); 700 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 701 up_write(&dev_replace->rwsem); 702 703 ret = btrfs_sysfs_add_device(tgt_device); 704 if (ret) 705 btrfs_err(fs_info, "kobj add dev failed %d", ret); 706 707 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 708 709 /* 710 * Commit dev_replace state and reserve 1 item for it. 711 * This is crucial to ensure we won't miss copying extents for new block 712 * groups that are allocated after we started the device replace, and 713 * must be done after setting up the device replace state. 714 */ 715 trans = btrfs_start_transaction(root, 1); 716 if (IS_ERR(trans)) { 717 ret = PTR_ERR(trans); 718 down_write(&dev_replace->rwsem); 719 dev_replace->replace_state = 720 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED; 721 dev_replace->srcdev = NULL; 722 dev_replace->tgtdev = NULL; 723 up_write(&dev_replace->rwsem); 724 goto leave; 725 } 726 727 ret = btrfs_commit_transaction(trans); 728 WARN_ON(ret); 729 730 /* the disk copy procedure reuses the scrub code */ 731 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 732 btrfs_device_get_total_bytes(src_device), 733 &dev_replace->scrub_progress, 0, 1); 734 735 ret = btrfs_dev_replace_finishing(fs_info, ret); 736 if (ret == -EINPROGRESS) 737 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS; 738 739 return ret; 740 741 leave: 742 btrfs_destroy_dev_replace_tgtdev(tgt_device); 743 return ret; 744 } 745 746 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info, 747 struct btrfs_ioctl_dev_replace_args *args) 748 { 749 int ret; 750 751 switch (args->start.cont_reading_from_srcdev_mode) { 752 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 753 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 754 break; 755 default: 756 return -EINVAL; 757 } 758 759 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 760 args->start.tgtdev_name[0] == '\0') 761 return -EINVAL; 762 763 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name, 764 args->start.srcdevid, 765 args->start.srcdev_name, 766 args->start.cont_reading_from_srcdev_mode); 767 args->result = ret; 768 /* don't warn if EINPROGRESS, someone else might be running scrub */ 769 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS || 770 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR) 771 return 0; 772 773 return ret; 774 } 775 776 /* 777 * blocked until all in-flight bios operations are finished. 778 */ 779 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) 780 { 781 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 782 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum( 783 &fs_info->dev_replace.bio_counter)); 784 } 785 786 /* 787 * we have removed target device, it is safe to allow new bios request. 788 */ 789 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) 790 { 791 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 792 wake_up(&fs_info->dev_replace.replace_wait); 793 } 794 795 /* 796 * When finishing the device replace, before swapping the source device with the 797 * target device we must update the chunk allocation state in the target device, 798 * as it is empty because replace works by directly copying the chunks and not 799 * through the normal chunk allocation path. 800 */ 801 static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev, 802 struct btrfs_device *tgtdev) 803 { 804 struct extent_state *cached_state = NULL; 805 u64 start = 0; 806 u64 found_start; 807 u64 found_end; 808 int ret = 0; 809 810 lockdep_assert_held(&srcdev->fs_info->chunk_mutex); 811 812 while (!find_first_extent_bit(&srcdev->alloc_state, start, 813 &found_start, &found_end, 814 CHUNK_ALLOCATED, &cached_state)) { 815 ret = set_extent_bits(&tgtdev->alloc_state, found_start, 816 found_end, CHUNK_ALLOCATED); 817 if (ret) 818 break; 819 start = found_end + 1; 820 } 821 822 free_extent_state(cached_state); 823 return ret; 824 } 825 826 static void btrfs_dev_replace_update_device_in_mapping_tree( 827 struct btrfs_fs_info *fs_info, 828 struct btrfs_device *srcdev, 829 struct btrfs_device *tgtdev) 830 { 831 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 832 struct extent_map *em; 833 struct map_lookup *map; 834 u64 start = 0; 835 int i; 836 837 write_lock(&em_tree->lock); 838 do { 839 em = lookup_extent_mapping(em_tree, start, (u64)-1); 840 if (!em) 841 break; 842 map = em->map_lookup; 843 for (i = 0; i < map->num_stripes; i++) 844 if (srcdev == map->stripes[i].dev) 845 map->stripes[i].dev = tgtdev; 846 start = em->start + em->len; 847 free_extent_map(em); 848 } while (start); 849 write_unlock(&em_tree->lock); 850 } 851 852 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 853 int scrub_ret) 854 { 855 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 856 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 857 struct btrfs_device *tgt_device; 858 struct btrfs_device *src_device; 859 struct btrfs_root *root = fs_info->tree_root; 860 u8 uuid_tmp[BTRFS_UUID_SIZE]; 861 struct btrfs_trans_handle *trans; 862 int ret = 0; 863 864 /* don't allow cancel or unmount to disturb the finishing procedure */ 865 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 866 867 down_read(&dev_replace->rwsem); 868 /* was the operation canceled, or is it finished? */ 869 if (dev_replace->replace_state != 870 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 871 up_read(&dev_replace->rwsem); 872 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 873 return 0; 874 } 875 876 tgt_device = dev_replace->tgtdev; 877 src_device = dev_replace->srcdev; 878 up_read(&dev_replace->rwsem); 879 880 /* 881 * flush all outstanding I/O and inode extent mappings before the 882 * copy operation is declared as being finished 883 */ 884 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false); 885 if (ret) { 886 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 887 return ret; 888 } 889 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 890 891 /* 892 * We have to use this loop approach because at this point src_device 893 * has to be available for transaction commit to complete, yet new 894 * chunks shouldn't be allocated on the device. 895 */ 896 while (1) { 897 trans = btrfs_start_transaction(root, 0); 898 if (IS_ERR(trans)) { 899 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 900 return PTR_ERR(trans); 901 } 902 ret = btrfs_commit_transaction(trans); 903 WARN_ON(ret); 904 905 /* Prevent write_all_supers() during the finishing procedure */ 906 mutex_lock(&fs_devices->device_list_mutex); 907 /* Prevent new chunks being allocated on the source device */ 908 mutex_lock(&fs_info->chunk_mutex); 909 910 if (!list_empty(&src_device->post_commit_list)) { 911 mutex_unlock(&fs_devices->device_list_mutex); 912 mutex_unlock(&fs_info->chunk_mutex); 913 } else { 914 break; 915 } 916 } 917 918 down_write(&dev_replace->rwsem); 919 dev_replace->replace_state = 920 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 921 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 922 dev_replace->tgtdev = NULL; 923 dev_replace->srcdev = NULL; 924 dev_replace->time_stopped = ktime_get_real_seconds(); 925 dev_replace->item_needs_writeback = 1; 926 927 /* 928 * Update allocation state in the new device and replace the old device 929 * with the new one in the mapping tree. 930 */ 931 if (!scrub_ret) { 932 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device); 933 if (scrub_ret) 934 goto error; 935 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 936 src_device, 937 tgt_device); 938 } else { 939 if (scrub_ret != -ECANCELED) 940 btrfs_err_in_rcu(fs_info, 941 "btrfs_scrub_dev(%s, %llu, %s) failed %d", 942 btrfs_dev_name(src_device), 943 src_device->devid, 944 rcu_str_deref(tgt_device->name), scrub_ret); 945 error: 946 up_write(&dev_replace->rwsem); 947 mutex_unlock(&fs_info->chunk_mutex); 948 mutex_unlock(&fs_devices->device_list_mutex); 949 btrfs_rm_dev_replace_blocked(fs_info); 950 if (tgt_device) 951 btrfs_destroy_dev_replace_tgtdev(tgt_device); 952 btrfs_rm_dev_replace_unblocked(fs_info); 953 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 954 955 return scrub_ret; 956 } 957 958 btrfs_info_in_rcu(fs_info, 959 "dev_replace from %s (devid %llu) to %s finished", 960 btrfs_dev_name(src_device), 961 src_device->devid, 962 rcu_str_deref(tgt_device->name)); 963 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state); 964 tgt_device->devid = src_device->devid; 965 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 966 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 967 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 968 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 969 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes); 970 btrfs_device_set_disk_total_bytes(tgt_device, 971 src_device->disk_total_bytes); 972 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used); 973 tgt_device->commit_bytes_used = src_device->bytes_used; 974 975 btrfs_assign_next_active_device(src_device, tgt_device); 976 977 list_add(&tgt_device->dev_alloc_list, &fs_devices->alloc_list); 978 fs_devices->rw_devices++; 979 980 up_write(&dev_replace->rwsem); 981 btrfs_rm_dev_replace_blocked(fs_info); 982 983 btrfs_rm_dev_replace_remove_srcdev(src_device); 984 985 btrfs_rm_dev_replace_unblocked(fs_info); 986 987 /* 988 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will 989 * update on-disk dev stats value during commit transaction 990 */ 991 atomic_inc(&tgt_device->dev_stats_ccnt); 992 993 /* 994 * this is again a consistent state where no dev_replace procedure 995 * is running, the target device is part of the filesystem, the 996 * source device is not part of the filesystem anymore and its 1st 997 * superblock is scratched out so that it is no longer marked to 998 * belong to this filesystem. 999 */ 1000 mutex_unlock(&fs_info->chunk_mutex); 1001 mutex_unlock(&fs_devices->device_list_mutex); 1002 1003 /* replace the sysfs entry */ 1004 btrfs_sysfs_remove_device(src_device); 1005 btrfs_sysfs_update_devid(tgt_device); 1006 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state)) 1007 btrfs_scratch_superblocks(fs_info, src_device->bdev, 1008 src_device->name->str); 1009 1010 /* write back the superblocks */ 1011 trans = btrfs_start_transaction(root, 0); 1012 if (!IS_ERR(trans)) 1013 btrfs_commit_transaction(trans); 1014 1015 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1016 1017 btrfs_rm_dev_replace_free_srcdev(src_device); 1018 1019 return 0; 1020 } 1021 1022 /* 1023 * Read progress of device replace status according to the state and last 1024 * stored position. The value format is the same as for 1025 * btrfs_dev_replace::progress_1000 1026 */ 1027 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info) 1028 { 1029 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1030 u64 ret = 0; 1031 1032 switch (dev_replace->replace_state) { 1033 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1034 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1035 ret = 0; 1036 break; 1037 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1038 ret = 1000; 1039 break; 1040 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1041 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1042 ret = div64_u64(dev_replace->cursor_left, 1043 div_u64(btrfs_device_get_total_bytes( 1044 dev_replace->srcdev), 1000)); 1045 break; 1046 } 1047 1048 return ret; 1049 } 1050 1051 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 1052 struct btrfs_ioctl_dev_replace_args *args) 1053 { 1054 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1055 1056 down_read(&dev_replace->rwsem); 1057 /* even if !dev_replace_is_valid, the values are good enough for 1058 * the replace_status ioctl */ 1059 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 1060 args->status.replace_state = dev_replace->replace_state; 1061 args->status.time_started = dev_replace->time_started; 1062 args->status.time_stopped = dev_replace->time_stopped; 1063 args->status.num_write_errors = 1064 atomic64_read(&dev_replace->num_write_errors); 1065 args->status.num_uncorrectable_read_errors = 1066 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 1067 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info); 1068 up_read(&dev_replace->rwsem); 1069 } 1070 1071 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 1072 { 1073 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1074 struct btrfs_device *tgt_device = NULL; 1075 struct btrfs_device *src_device = NULL; 1076 struct btrfs_trans_handle *trans; 1077 struct btrfs_root *root = fs_info->tree_root; 1078 int result; 1079 int ret; 1080 1081 if (sb_rdonly(fs_info->sb)) 1082 return -EROFS; 1083 1084 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 1085 down_write(&dev_replace->rwsem); 1086 switch (dev_replace->replace_state) { 1087 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1088 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1089 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1090 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 1091 up_write(&dev_replace->rwsem); 1092 break; 1093 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1094 tgt_device = dev_replace->tgtdev; 1095 src_device = dev_replace->srcdev; 1096 up_write(&dev_replace->rwsem); 1097 ret = btrfs_scrub_cancel(fs_info); 1098 if (ret < 0) { 1099 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 1100 } else { 1101 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 1102 /* 1103 * btrfs_dev_replace_finishing() will handle the 1104 * cleanup part 1105 */ 1106 btrfs_info_in_rcu(fs_info, 1107 "dev_replace from %s (devid %llu) to %s canceled", 1108 btrfs_dev_name(src_device), src_device->devid, 1109 btrfs_dev_name(tgt_device)); 1110 } 1111 break; 1112 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1113 /* 1114 * Scrub doing the replace isn't running so we need to do the 1115 * cleanup step of btrfs_dev_replace_finishing() here 1116 */ 1117 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 1118 tgt_device = dev_replace->tgtdev; 1119 src_device = dev_replace->srcdev; 1120 dev_replace->tgtdev = NULL; 1121 dev_replace->srcdev = NULL; 1122 dev_replace->replace_state = 1123 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 1124 dev_replace->time_stopped = ktime_get_real_seconds(); 1125 dev_replace->item_needs_writeback = 1; 1126 1127 up_write(&dev_replace->rwsem); 1128 1129 /* Scrub for replace must not be running in suspended state */ 1130 btrfs_scrub_cancel(fs_info); 1131 1132 trans = btrfs_start_transaction(root, 0); 1133 if (IS_ERR(trans)) { 1134 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1135 return PTR_ERR(trans); 1136 } 1137 ret = btrfs_commit_transaction(trans); 1138 WARN_ON(ret); 1139 1140 btrfs_info_in_rcu(fs_info, 1141 "suspended dev_replace from %s (devid %llu) to %s canceled", 1142 btrfs_dev_name(src_device), src_device->devid, 1143 btrfs_dev_name(tgt_device)); 1144 1145 if (tgt_device) 1146 btrfs_destroy_dev_replace_tgtdev(tgt_device); 1147 break; 1148 default: 1149 up_write(&dev_replace->rwsem); 1150 result = -EINVAL; 1151 } 1152 1153 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1154 return result; 1155 } 1156 1157 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 1158 { 1159 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1160 1161 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 1162 down_write(&dev_replace->rwsem); 1163 1164 switch (dev_replace->replace_state) { 1165 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1166 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1167 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1168 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1169 break; 1170 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1171 dev_replace->replace_state = 1172 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 1173 dev_replace->time_stopped = ktime_get_real_seconds(); 1174 dev_replace->item_needs_writeback = 1; 1175 btrfs_info(fs_info, "suspending dev_replace for unmount"); 1176 break; 1177 } 1178 1179 up_write(&dev_replace->rwsem); 1180 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 1181 } 1182 1183 /* resume dev_replace procedure that was interrupted by unmount */ 1184 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 1185 { 1186 struct task_struct *task; 1187 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1188 1189 down_write(&dev_replace->rwsem); 1190 1191 switch (dev_replace->replace_state) { 1192 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1193 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1194 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1195 up_write(&dev_replace->rwsem); 1196 return 0; 1197 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1198 break; 1199 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1200 dev_replace->replace_state = 1201 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 1202 break; 1203 } 1204 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 1205 btrfs_info(fs_info, 1206 "cannot continue dev_replace, tgtdev is missing"); 1207 btrfs_info(fs_info, 1208 "you may cancel the operation after 'mount -o degraded'"); 1209 dev_replace->replace_state = 1210 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 1211 up_write(&dev_replace->rwsem); 1212 return 0; 1213 } 1214 up_write(&dev_replace->rwsem); 1215 1216 /* 1217 * This could collide with a paused balance, but the exclusive op logic 1218 * should never allow both to start and pause. We don't want to allow 1219 * dev-replace to start anyway. 1220 */ 1221 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) { 1222 down_write(&dev_replace->rwsem); 1223 dev_replace->replace_state = 1224 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 1225 up_write(&dev_replace->rwsem); 1226 btrfs_info(fs_info, 1227 "cannot resume dev-replace, other exclusive operation running"); 1228 return 0; 1229 } 1230 1231 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 1232 return PTR_ERR_OR_ZERO(task); 1233 } 1234 1235 static int btrfs_dev_replace_kthread(void *data) 1236 { 1237 struct btrfs_fs_info *fs_info = data; 1238 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1239 u64 progress; 1240 int ret; 1241 1242 progress = btrfs_dev_replace_progress(fs_info); 1243 progress = div_u64(progress, 10); 1244 btrfs_info_in_rcu(fs_info, 1245 "continuing dev_replace from %s (devid %llu) to target %s @%u%%", 1246 btrfs_dev_name(dev_replace->srcdev), 1247 dev_replace->srcdev->devid, 1248 btrfs_dev_name(dev_replace->tgtdev), 1249 (unsigned int)progress); 1250 1251 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 1252 dev_replace->committed_cursor_left, 1253 btrfs_device_get_total_bytes(dev_replace->srcdev), 1254 &dev_replace->scrub_progress, 0, 1); 1255 ret = btrfs_dev_replace_finishing(fs_info, ret); 1256 WARN_ON(ret && ret != -ECANCELED); 1257 1258 btrfs_exclop_finish(fs_info); 1259 return 0; 1260 } 1261 1262 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 1263 { 1264 if (!dev_replace->is_valid) 1265 return 0; 1266 1267 switch (dev_replace->replace_state) { 1268 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1269 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1270 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1271 return 0; 1272 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1273 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1274 /* 1275 * return true even if tgtdev is missing (this is 1276 * something that can happen if the dev_replace 1277 * procedure is suspended by an umount and then 1278 * the tgtdev is missing (or "btrfs dev scan") was 1279 * not called and the filesystem is remounted 1280 * in degraded state. This does not stop the 1281 * dev_replace procedure. It needs to be canceled 1282 * manually if the cancellation is wanted. 1283 */ 1284 break; 1285 } 1286 return 1; 1287 } 1288 1289 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount) 1290 { 1291 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount); 1292 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait); 1293 } 1294 1295 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) 1296 { 1297 while (1) { 1298 percpu_counter_inc(&fs_info->dev_replace.bio_counter); 1299 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1300 &fs_info->fs_state))) 1301 break; 1302 1303 btrfs_bio_counter_dec(fs_info); 1304 wait_event(fs_info->dev_replace.replace_wait, 1305 !test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1306 &fs_info->fs_state)); 1307 } 1308 } 1309