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