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 "ctree.h" 13 #include "extent_map.h" 14 #include "disk-io.h" 15 #include "transaction.h" 16 #include "print-tree.h" 17 #include "volumes.h" 18 #include "async-thread.h" 19 #include "check-integrity.h" 20 #include "rcu-string.h" 21 #include "dev-replace.h" 22 #include "sysfs.h" 23 24 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 25 int scrub_ret); 26 static void btrfs_dev_replace_update_device_in_mapping_tree( 27 struct btrfs_fs_info *fs_info, 28 struct btrfs_device *srcdev, 29 struct btrfs_device *tgtdev); 30 static int btrfs_dev_replace_kthread(void *data); 31 32 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) 33 { 34 struct btrfs_key key; 35 struct btrfs_root *dev_root = fs_info->dev_root; 36 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 37 struct extent_buffer *eb; 38 int slot; 39 int ret = 0; 40 struct btrfs_path *path = NULL; 41 int item_size; 42 struct btrfs_dev_replace_item *ptr; 43 u64 src_devid; 44 45 path = btrfs_alloc_path(); 46 if (!path) { 47 ret = -ENOMEM; 48 goto out; 49 } 50 51 key.objectid = 0; 52 key.type = BTRFS_DEV_REPLACE_KEY; 53 key.offset = 0; 54 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); 55 if (ret) { 56 no_valid_dev_replace_entry_found: 57 ret = 0; 58 dev_replace->replace_state = 59 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED; 60 dev_replace->cont_reading_from_srcdev_mode = 61 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; 62 dev_replace->time_started = 0; 63 dev_replace->time_stopped = 0; 64 atomic64_set(&dev_replace->num_write_errors, 0); 65 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 66 dev_replace->cursor_left = 0; 67 dev_replace->committed_cursor_left = 0; 68 dev_replace->cursor_left_last_write_of_item = 0; 69 dev_replace->cursor_right = 0; 70 dev_replace->srcdev = NULL; 71 dev_replace->tgtdev = NULL; 72 dev_replace->is_valid = 0; 73 dev_replace->item_needs_writeback = 0; 74 goto out; 75 } 76 slot = path->slots[0]; 77 eb = path->nodes[0]; 78 item_size = btrfs_item_size_nr(eb, slot); 79 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); 80 81 if (item_size != sizeof(struct btrfs_dev_replace_item)) { 82 btrfs_warn(fs_info, 83 "dev_replace entry found has unexpected size, ignore entry"); 84 goto no_valid_dev_replace_entry_found; 85 } 86 87 src_devid = btrfs_dev_replace_src_devid(eb, ptr); 88 dev_replace->cont_reading_from_srcdev_mode = 89 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); 90 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); 91 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); 92 dev_replace->time_stopped = 93 btrfs_dev_replace_time_stopped(eb, ptr); 94 atomic64_set(&dev_replace->num_write_errors, 95 btrfs_dev_replace_num_write_errors(eb, ptr)); 96 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 97 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); 98 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); 99 dev_replace->committed_cursor_left = dev_replace->cursor_left; 100 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; 101 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); 102 dev_replace->is_valid = 1; 103 104 dev_replace->item_needs_writeback = 0; 105 switch (dev_replace->replace_state) { 106 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 107 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 108 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 109 dev_replace->srcdev = NULL; 110 dev_replace->tgtdev = NULL; 111 break; 112 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 113 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 114 dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices, 115 src_devid, NULL, NULL, true); 116 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices, 117 BTRFS_DEV_REPLACE_DEVID, 118 NULL, NULL, true); 119 /* 120 * allow 'btrfs dev replace_cancel' if src/tgt device is 121 * missing 122 */ 123 if (!dev_replace->srcdev && 124 !btrfs_test_opt(fs_info, DEGRADED)) { 125 ret = -EIO; 126 btrfs_warn(fs_info, 127 "cannot mount because device replace operation is ongoing and"); 128 btrfs_warn(fs_info, 129 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 130 src_devid); 131 } 132 if (!dev_replace->tgtdev && 133 !btrfs_test_opt(fs_info, DEGRADED)) { 134 ret = -EIO; 135 btrfs_warn(fs_info, 136 "cannot mount because device replace operation is ongoing and"); 137 btrfs_warn(fs_info, 138 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 139 BTRFS_DEV_REPLACE_DEVID); 140 } 141 if (dev_replace->tgtdev) { 142 if (dev_replace->srcdev) { 143 dev_replace->tgtdev->total_bytes = 144 dev_replace->srcdev->total_bytes; 145 dev_replace->tgtdev->disk_total_bytes = 146 dev_replace->srcdev->disk_total_bytes; 147 dev_replace->tgtdev->commit_total_bytes = 148 dev_replace->srcdev->commit_total_bytes; 149 dev_replace->tgtdev->bytes_used = 150 dev_replace->srcdev->bytes_used; 151 dev_replace->tgtdev->commit_bytes_used = 152 dev_replace->srcdev->commit_bytes_used; 153 } 154 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, 155 &dev_replace->tgtdev->dev_state); 156 157 WARN_ON(fs_info->fs_devices->rw_devices == 0); 158 dev_replace->tgtdev->io_width = fs_info->sectorsize; 159 dev_replace->tgtdev->io_align = fs_info->sectorsize; 160 dev_replace->tgtdev->sector_size = fs_info->sectorsize; 161 dev_replace->tgtdev->fs_info = fs_info; 162 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 163 &dev_replace->tgtdev->dev_state); 164 } 165 break; 166 } 167 168 out: 169 btrfs_free_path(path); 170 return ret; 171 } 172 173 /* 174 * Initialize a new device for device replace target from a given source dev 175 * and path. 176 * 177 * Return 0 and new device in @device_out, otherwise return < 0 178 */ 179 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, 180 const char *device_path, 181 struct btrfs_device *srcdev, 182 struct btrfs_device **device_out) 183 { 184 struct btrfs_device *device; 185 struct block_device *bdev; 186 struct list_head *devices; 187 struct rcu_string *name; 188 u64 devid = BTRFS_DEV_REPLACE_DEVID; 189 int ret = 0; 190 191 *device_out = NULL; 192 if (fs_info->fs_devices->seeding) { 193 btrfs_err(fs_info, "the filesystem is a seed filesystem!"); 194 return -EINVAL; 195 } 196 197 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 198 fs_info->bdev_holder); 199 if (IS_ERR(bdev)) { 200 btrfs_err(fs_info, "target device %s is invalid!", device_path); 201 return PTR_ERR(bdev); 202 } 203 204 sync_blockdev(bdev); 205 206 devices = &fs_info->fs_devices->devices; 207 list_for_each_entry(device, devices, dev_list) { 208 if (device->bdev == bdev) { 209 btrfs_err(fs_info, 210 "target device is in the filesystem!"); 211 ret = -EEXIST; 212 goto error; 213 } 214 } 215 216 217 if (i_size_read(bdev->bd_inode) < 218 btrfs_device_get_total_bytes(srcdev)) { 219 btrfs_err(fs_info, 220 "target device is smaller than source device!"); 221 ret = -EINVAL; 222 goto error; 223 } 224 225 226 device = btrfs_alloc_device(NULL, &devid, NULL); 227 if (IS_ERR(device)) { 228 ret = PTR_ERR(device); 229 goto error; 230 } 231 232 name = rcu_string_strdup(device_path, GFP_KERNEL); 233 if (!name) { 234 btrfs_free_device(device); 235 ret = -ENOMEM; 236 goto error; 237 } 238 rcu_assign_pointer(device->name, name); 239 240 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 241 device->generation = 0; 242 device->io_width = fs_info->sectorsize; 243 device->io_align = fs_info->sectorsize; 244 device->sector_size = fs_info->sectorsize; 245 device->total_bytes = btrfs_device_get_total_bytes(srcdev); 246 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); 247 device->bytes_used = btrfs_device_get_bytes_used(srcdev); 248 device->commit_total_bytes = srcdev->commit_total_bytes; 249 device->commit_bytes_used = device->bytes_used; 250 device->fs_info = fs_info; 251 device->bdev = bdev; 252 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 253 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 254 device->mode = FMODE_EXCL; 255 device->dev_stats_valid = 1; 256 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 257 device->fs_devices = fs_info->fs_devices; 258 259 mutex_lock(&fs_info->fs_devices->device_list_mutex); 260 list_add(&device->dev_list, &fs_info->fs_devices->devices); 261 fs_info->fs_devices->num_devices++; 262 fs_info->fs_devices->open_devices++; 263 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 264 265 *device_out = device; 266 return 0; 267 268 error: 269 blkdev_put(bdev, FMODE_EXCL); 270 return ret; 271 } 272 273 /* 274 * called from commit_transaction. Writes changed device replace state to 275 * disk. 276 */ 277 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans) 278 { 279 struct btrfs_fs_info *fs_info = trans->fs_info; 280 int ret; 281 struct btrfs_root *dev_root = fs_info->dev_root; 282 struct btrfs_path *path; 283 struct btrfs_key key; 284 struct extent_buffer *eb; 285 struct btrfs_dev_replace_item *ptr; 286 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 287 288 down_read(&dev_replace->rwsem); 289 if (!dev_replace->is_valid || 290 !dev_replace->item_needs_writeback) { 291 up_read(&dev_replace->rwsem); 292 return 0; 293 } 294 up_read(&dev_replace->rwsem); 295 296 key.objectid = 0; 297 key.type = BTRFS_DEV_REPLACE_KEY; 298 key.offset = 0; 299 300 path = btrfs_alloc_path(); 301 if (!path) { 302 ret = -ENOMEM; 303 goto out; 304 } 305 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 306 if (ret < 0) { 307 btrfs_warn(fs_info, 308 "error %d while searching for dev_replace item!", 309 ret); 310 goto out; 311 } 312 313 if (ret == 0 && 314 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 315 /* 316 * need to delete old one and insert a new one. 317 * Since no attempt is made to recover any old state, if the 318 * dev_replace state is 'running', the data on the target 319 * drive is lost. 320 * It would be possible to recover the state: just make sure 321 * that the beginning of the item is never changed and always 322 * contains all the essential information. Then read this 323 * minimal set of information and use it as a base for the 324 * new state. 325 */ 326 ret = btrfs_del_item(trans, dev_root, path); 327 if (ret != 0) { 328 btrfs_warn(fs_info, 329 "delete too small dev_replace item failed %d!", 330 ret); 331 goto out; 332 } 333 ret = 1; 334 } 335 336 if (ret == 1) { 337 /* need to insert a new item */ 338 btrfs_release_path(path); 339 ret = btrfs_insert_empty_item(trans, dev_root, path, 340 &key, sizeof(*ptr)); 341 if (ret < 0) { 342 btrfs_warn(fs_info, 343 "insert dev_replace item failed %d!", ret); 344 goto out; 345 } 346 } 347 348 eb = path->nodes[0]; 349 ptr = btrfs_item_ptr(eb, path->slots[0], 350 struct btrfs_dev_replace_item); 351 352 down_write(&dev_replace->rwsem); 353 if (dev_replace->srcdev) 354 btrfs_set_dev_replace_src_devid(eb, ptr, 355 dev_replace->srcdev->devid); 356 else 357 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); 358 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, 359 dev_replace->cont_reading_from_srcdev_mode); 360 btrfs_set_dev_replace_replace_state(eb, ptr, 361 dev_replace->replace_state); 362 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); 363 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); 364 btrfs_set_dev_replace_num_write_errors(eb, ptr, 365 atomic64_read(&dev_replace->num_write_errors)); 366 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, 367 atomic64_read(&dev_replace->num_uncorrectable_read_errors)); 368 dev_replace->cursor_left_last_write_of_item = 369 dev_replace->cursor_left; 370 btrfs_set_dev_replace_cursor_left(eb, ptr, 371 dev_replace->cursor_left_last_write_of_item); 372 btrfs_set_dev_replace_cursor_right(eb, ptr, 373 dev_replace->cursor_right); 374 dev_replace->item_needs_writeback = 0; 375 up_write(&dev_replace->rwsem); 376 377 btrfs_mark_buffer_dirty(eb); 378 379 out: 380 btrfs_free_path(path); 381 382 return ret; 383 } 384 385 static char* btrfs_dev_name(struct btrfs_device *device) 386 { 387 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 388 return "<missing disk>"; 389 else 390 return rcu_str_deref(device->name); 391 } 392 393 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info, 394 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name, 395 int read_src) 396 { 397 struct btrfs_root *root = fs_info->dev_root; 398 struct btrfs_trans_handle *trans; 399 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 400 int ret; 401 struct btrfs_device *tgt_device = NULL; 402 struct btrfs_device *src_device = NULL; 403 bool need_unlock; 404 405 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid, 406 srcdev_name); 407 if (IS_ERR(src_device)) 408 return PTR_ERR(src_device); 409 410 if (btrfs_pinned_by_swapfile(fs_info, src_device)) { 411 btrfs_warn_in_rcu(fs_info, 412 "cannot replace device %s (devid %llu) due to active swapfile", 413 btrfs_dev_name(src_device), src_device->devid); 414 return -ETXTBSY; 415 } 416 417 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name, 418 src_device, &tgt_device); 419 if (ret) 420 return ret; 421 422 /* 423 * Here we commit the transaction to make sure commit_total_bytes 424 * of all the devices are updated. 425 */ 426 trans = btrfs_attach_transaction(root); 427 if (!IS_ERR(trans)) { 428 ret = btrfs_commit_transaction(trans); 429 if (ret) 430 return ret; 431 } else if (PTR_ERR(trans) != -ENOENT) { 432 return PTR_ERR(trans); 433 } 434 435 need_unlock = true; 436 down_write(&dev_replace->rwsem); 437 switch (dev_replace->replace_state) { 438 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 439 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 440 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 441 break; 442 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 443 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 444 ASSERT(0); 445 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 446 goto leave; 447 } 448 449 dev_replace->cont_reading_from_srcdev_mode = read_src; 450 WARN_ON(!src_device); 451 dev_replace->srcdev = src_device; 452 dev_replace->tgtdev = tgt_device; 453 454 btrfs_info_in_rcu(fs_info, 455 "dev_replace from %s (devid %llu) to %s started", 456 btrfs_dev_name(src_device), 457 src_device->devid, 458 rcu_str_deref(tgt_device->name)); 459 460 /* 461 * from now on, the writes to the srcdev are all duplicated to 462 * go to the tgtdev as well (refer to btrfs_map_block()). 463 */ 464 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 465 dev_replace->time_started = ktime_get_real_seconds(); 466 dev_replace->cursor_left = 0; 467 dev_replace->committed_cursor_left = 0; 468 dev_replace->cursor_left_last_write_of_item = 0; 469 dev_replace->cursor_right = 0; 470 dev_replace->is_valid = 1; 471 dev_replace->item_needs_writeback = 1; 472 atomic64_set(&dev_replace->num_write_errors, 0); 473 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 474 up_write(&dev_replace->rwsem); 475 need_unlock = false; 476 477 ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device); 478 if (ret) 479 btrfs_err(fs_info, "kobj add dev failed %d", ret); 480 481 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 482 483 /* force writing the updated state information to disk */ 484 trans = btrfs_start_transaction(root, 0); 485 if (IS_ERR(trans)) { 486 ret = PTR_ERR(trans); 487 need_unlock = true; 488 down_write(&dev_replace->rwsem); 489 dev_replace->replace_state = 490 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED; 491 dev_replace->srcdev = NULL; 492 dev_replace->tgtdev = NULL; 493 goto leave; 494 } 495 496 ret = btrfs_commit_transaction(trans); 497 WARN_ON(ret); 498 499 /* the disk copy procedure reuses the scrub code */ 500 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 501 btrfs_device_get_total_bytes(src_device), 502 &dev_replace->scrub_progress, 0, 1); 503 504 ret = btrfs_dev_replace_finishing(fs_info, ret); 505 if (ret == -EINPROGRESS) { 506 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS; 507 } else if (ret != -ECANCELED) { 508 WARN_ON(ret); 509 } 510 511 return ret; 512 513 leave: 514 if (need_unlock) 515 up_write(&dev_replace->rwsem); 516 btrfs_destroy_dev_replace_tgtdev(tgt_device); 517 return ret; 518 } 519 520 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info, 521 struct btrfs_ioctl_dev_replace_args *args) 522 { 523 int ret; 524 525 switch (args->start.cont_reading_from_srcdev_mode) { 526 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 527 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 528 break; 529 default: 530 return -EINVAL; 531 } 532 533 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 534 args->start.tgtdev_name[0] == '\0') 535 return -EINVAL; 536 537 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name, 538 args->start.srcdevid, 539 args->start.srcdev_name, 540 args->start.cont_reading_from_srcdev_mode); 541 args->result = ret; 542 /* don't warn if EINPROGRESS, someone else might be running scrub */ 543 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS || 544 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR) 545 return 0; 546 547 return ret; 548 } 549 550 /* 551 * blocked until all in-flight bios operations are finished. 552 */ 553 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) 554 { 555 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 556 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum( 557 &fs_info->dev_replace.bio_counter)); 558 } 559 560 /* 561 * we have removed target device, it is safe to allow new bios request. 562 */ 563 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) 564 { 565 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 566 wake_up(&fs_info->dev_replace.replace_wait); 567 } 568 569 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 570 int scrub_ret) 571 { 572 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 573 struct btrfs_device *tgt_device; 574 struct btrfs_device *src_device; 575 struct btrfs_root *root = fs_info->tree_root; 576 u8 uuid_tmp[BTRFS_UUID_SIZE]; 577 struct btrfs_trans_handle *trans; 578 int ret = 0; 579 580 /* don't allow cancel or unmount to disturb the finishing procedure */ 581 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 582 583 down_read(&dev_replace->rwsem); 584 /* was the operation canceled, or is it finished? */ 585 if (dev_replace->replace_state != 586 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 587 up_read(&dev_replace->rwsem); 588 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 589 return 0; 590 } 591 592 tgt_device = dev_replace->tgtdev; 593 src_device = dev_replace->srcdev; 594 up_read(&dev_replace->rwsem); 595 596 /* 597 * flush all outstanding I/O and inode extent mappings before the 598 * copy operation is declared as being finished 599 */ 600 ret = btrfs_start_delalloc_roots(fs_info, -1); 601 if (ret) { 602 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 603 return ret; 604 } 605 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 606 607 /* 608 * We have to use this loop approach because at this point src_device 609 * has to be available for transaction commit to complete, yet new 610 * chunks shouldn't be allocated on the device. 611 */ 612 while (1) { 613 trans = btrfs_start_transaction(root, 0); 614 if (IS_ERR(trans)) { 615 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 616 return PTR_ERR(trans); 617 } 618 ret = btrfs_commit_transaction(trans); 619 WARN_ON(ret); 620 621 /* Prevent write_all_supers() during the finishing procedure */ 622 mutex_lock(&fs_info->fs_devices->device_list_mutex); 623 /* Prevent new chunks being allocated on the source device */ 624 mutex_lock(&fs_info->chunk_mutex); 625 626 if (!list_empty(&src_device->post_commit_list)) { 627 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 628 mutex_unlock(&fs_info->chunk_mutex); 629 } else { 630 break; 631 } 632 } 633 634 down_write(&dev_replace->rwsem); 635 dev_replace->replace_state = 636 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 637 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 638 dev_replace->tgtdev = NULL; 639 dev_replace->srcdev = NULL; 640 dev_replace->time_stopped = ktime_get_real_seconds(); 641 dev_replace->item_needs_writeback = 1; 642 643 /* replace old device with new one in mapping tree */ 644 if (!scrub_ret) { 645 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 646 src_device, 647 tgt_device); 648 } else { 649 if (scrub_ret != -ECANCELED) 650 btrfs_err_in_rcu(fs_info, 651 "btrfs_scrub_dev(%s, %llu, %s) failed %d", 652 btrfs_dev_name(src_device), 653 src_device->devid, 654 rcu_str_deref(tgt_device->name), scrub_ret); 655 up_write(&dev_replace->rwsem); 656 mutex_unlock(&fs_info->chunk_mutex); 657 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 658 btrfs_rm_dev_replace_blocked(fs_info); 659 if (tgt_device) 660 btrfs_destroy_dev_replace_tgtdev(tgt_device); 661 btrfs_rm_dev_replace_unblocked(fs_info); 662 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 663 664 return scrub_ret; 665 } 666 667 btrfs_info_in_rcu(fs_info, 668 "dev_replace from %s (devid %llu) to %s finished", 669 btrfs_dev_name(src_device), 670 src_device->devid, 671 rcu_str_deref(tgt_device->name)); 672 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state); 673 tgt_device->devid = src_device->devid; 674 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 675 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 676 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 677 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 678 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes); 679 btrfs_device_set_disk_total_bytes(tgt_device, 680 src_device->disk_total_bytes); 681 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used); 682 tgt_device->commit_total_bytes = src_device->commit_total_bytes; 683 tgt_device->commit_bytes_used = src_device->bytes_used; 684 685 btrfs_assign_next_active_device(src_device, tgt_device); 686 687 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); 688 fs_info->fs_devices->rw_devices++; 689 690 up_write(&dev_replace->rwsem); 691 btrfs_rm_dev_replace_blocked(fs_info); 692 693 btrfs_rm_dev_replace_remove_srcdev(src_device); 694 695 btrfs_rm_dev_replace_unblocked(fs_info); 696 697 /* 698 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will 699 * update on-disk dev stats value during commit transaction 700 */ 701 atomic_inc(&tgt_device->dev_stats_ccnt); 702 703 /* 704 * this is again a consistent state where no dev_replace procedure 705 * is running, the target device is part of the filesystem, the 706 * source device is not part of the filesystem anymore and its 1st 707 * superblock is scratched out so that it is no longer marked to 708 * belong to this filesystem. 709 */ 710 mutex_unlock(&fs_info->chunk_mutex); 711 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 712 713 /* replace the sysfs entry */ 714 btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device); 715 btrfs_rm_dev_replace_free_srcdev(src_device); 716 717 /* write back the superblocks */ 718 trans = btrfs_start_transaction(root, 0); 719 if (!IS_ERR(trans)) 720 btrfs_commit_transaction(trans); 721 722 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 723 724 return 0; 725 } 726 727 static void btrfs_dev_replace_update_device_in_mapping_tree( 728 struct btrfs_fs_info *fs_info, 729 struct btrfs_device *srcdev, 730 struct btrfs_device *tgtdev) 731 { 732 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; 733 struct extent_map *em; 734 struct map_lookup *map; 735 u64 start = 0; 736 int i; 737 738 write_lock(&em_tree->lock); 739 do { 740 em = lookup_extent_mapping(em_tree, start, (u64)-1); 741 if (!em) 742 break; 743 map = em->map_lookup; 744 for (i = 0; i < map->num_stripes; i++) 745 if (srcdev == map->stripes[i].dev) 746 map->stripes[i].dev = tgtdev; 747 start = em->start + em->len; 748 free_extent_map(em); 749 } while (start); 750 write_unlock(&em_tree->lock); 751 } 752 753 /* 754 * Read progress of device replace status according to the state and last 755 * stored position. The value format is the same as for 756 * btrfs_dev_replace::progress_1000 757 */ 758 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info) 759 { 760 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 761 u64 ret = 0; 762 763 switch (dev_replace->replace_state) { 764 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 765 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 766 ret = 0; 767 break; 768 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 769 ret = 1000; 770 break; 771 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 772 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 773 ret = div64_u64(dev_replace->cursor_left, 774 div_u64(btrfs_device_get_total_bytes( 775 dev_replace->srcdev), 1000)); 776 break; 777 } 778 779 return ret; 780 } 781 782 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 783 struct btrfs_ioctl_dev_replace_args *args) 784 { 785 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 786 787 down_read(&dev_replace->rwsem); 788 /* even if !dev_replace_is_valid, the values are good enough for 789 * the replace_status ioctl */ 790 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 791 args->status.replace_state = dev_replace->replace_state; 792 args->status.time_started = dev_replace->time_started; 793 args->status.time_stopped = dev_replace->time_stopped; 794 args->status.num_write_errors = 795 atomic64_read(&dev_replace->num_write_errors); 796 args->status.num_uncorrectable_read_errors = 797 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 798 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info); 799 up_read(&dev_replace->rwsem); 800 } 801 802 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 803 { 804 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 805 struct btrfs_device *tgt_device = NULL; 806 struct btrfs_device *src_device = NULL; 807 struct btrfs_trans_handle *trans; 808 struct btrfs_root *root = fs_info->tree_root; 809 int result; 810 int ret; 811 812 if (sb_rdonly(fs_info->sb)) 813 return -EROFS; 814 815 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 816 down_write(&dev_replace->rwsem); 817 switch (dev_replace->replace_state) { 818 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 819 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 820 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 821 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 822 up_write(&dev_replace->rwsem); 823 break; 824 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 825 tgt_device = dev_replace->tgtdev; 826 src_device = dev_replace->srcdev; 827 up_write(&dev_replace->rwsem); 828 ret = btrfs_scrub_cancel(fs_info); 829 if (ret < 0) { 830 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 831 } else { 832 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 833 /* 834 * btrfs_dev_replace_finishing() will handle the 835 * cleanup part 836 */ 837 btrfs_info_in_rcu(fs_info, 838 "dev_replace from %s (devid %llu) to %s canceled", 839 btrfs_dev_name(src_device), src_device->devid, 840 btrfs_dev_name(tgt_device)); 841 } 842 break; 843 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 844 /* 845 * Scrub doing the replace isn't running so we need to do the 846 * cleanup step of btrfs_dev_replace_finishing() here 847 */ 848 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 849 tgt_device = dev_replace->tgtdev; 850 src_device = dev_replace->srcdev; 851 dev_replace->tgtdev = NULL; 852 dev_replace->srcdev = NULL; 853 dev_replace->replace_state = 854 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 855 dev_replace->time_stopped = ktime_get_real_seconds(); 856 dev_replace->item_needs_writeback = 1; 857 858 up_write(&dev_replace->rwsem); 859 860 /* Scrub for replace must not be running in suspended state */ 861 ret = btrfs_scrub_cancel(fs_info); 862 ASSERT(ret != -ENOTCONN); 863 864 trans = btrfs_start_transaction(root, 0); 865 if (IS_ERR(trans)) { 866 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 867 return PTR_ERR(trans); 868 } 869 ret = btrfs_commit_transaction(trans); 870 WARN_ON(ret); 871 872 btrfs_info_in_rcu(fs_info, 873 "suspended dev_replace from %s (devid %llu) to %s canceled", 874 btrfs_dev_name(src_device), src_device->devid, 875 btrfs_dev_name(tgt_device)); 876 877 if (tgt_device) 878 btrfs_destroy_dev_replace_tgtdev(tgt_device); 879 break; 880 default: 881 up_write(&dev_replace->rwsem); 882 result = -EINVAL; 883 } 884 885 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 886 return result; 887 } 888 889 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 890 { 891 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 892 893 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 894 down_write(&dev_replace->rwsem); 895 896 switch (dev_replace->replace_state) { 897 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 898 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 899 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 900 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 901 break; 902 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 903 dev_replace->replace_state = 904 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 905 dev_replace->time_stopped = ktime_get_real_seconds(); 906 dev_replace->item_needs_writeback = 1; 907 btrfs_info(fs_info, "suspending dev_replace for unmount"); 908 break; 909 } 910 911 up_write(&dev_replace->rwsem); 912 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 913 } 914 915 /* resume dev_replace procedure that was interrupted by unmount */ 916 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 917 { 918 struct task_struct *task; 919 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 920 921 down_write(&dev_replace->rwsem); 922 923 switch (dev_replace->replace_state) { 924 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 925 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 926 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 927 up_write(&dev_replace->rwsem); 928 return 0; 929 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 930 break; 931 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 932 dev_replace->replace_state = 933 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 934 break; 935 } 936 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 937 btrfs_info(fs_info, 938 "cannot continue dev_replace, tgtdev is missing"); 939 btrfs_info(fs_info, 940 "you may cancel the operation after 'mount -o degraded'"); 941 dev_replace->replace_state = 942 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 943 up_write(&dev_replace->rwsem); 944 return 0; 945 } 946 up_write(&dev_replace->rwsem); 947 948 /* 949 * This could collide with a paused balance, but the exclusive op logic 950 * should never allow both to start and pause. We don't want to allow 951 * dev-replace to start anyway. 952 */ 953 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { 954 down_write(&dev_replace->rwsem); 955 dev_replace->replace_state = 956 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 957 up_write(&dev_replace->rwsem); 958 btrfs_info(fs_info, 959 "cannot resume dev-replace, other exclusive operation running"); 960 return 0; 961 } 962 963 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 964 return PTR_ERR_OR_ZERO(task); 965 } 966 967 static int btrfs_dev_replace_kthread(void *data) 968 { 969 struct btrfs_fs_info *fs_info = data; 970 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 971 u64 progress; 972 int ret; 973 974 progress = btrfs_dev_replace_progress(fs_info); 975 progress = div_u64(progress, 10); 976 btrfs_info_in_rcu(fs_info, 977 "continuing dev_replace from %s (devid %llu) to target %s @%u%%", 978 btrfs_dev_name(dev_replace->srcdev), 979 dev_replace->srcdev->devid, 980 btrfs_dev_name(dev_replace->tgtdev), 981 (unsigned int)progress); 982 983 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 984 dev_replace->committed_cursor_left, 985 btrfs_device_get_total_bytes(dev_replace->srcdev), 986 &dev_replace->scrub_progress, 0, 1); 987 ret = btrfs_dev_replace_finishing(fs_info, ret); 988 WARN_ON(ret && ret != -ECANCELED); 989 990 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 991 return 0; 992 } 993 994 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 995 { 996 if (!dev_replace->is_valid) 997 return 0; 998 999 switch (dev_replace->replace_state) { 1000 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 1001 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 1002 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 1003 return 0; 1004 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 1005 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 1006 /* 1007 * return true even if tgtdev is missing (this is 1008 * something that can happen if the dev_replace 1009 * procedure is suspended by an umount and then 1010 * the tgtdev is missing (or "btrfs dev scan") was 1011 * not called and the filesystem is remounted 1012 * in degraded state. This does not stop the 1013 * dev_replace procedure. It needs to be canceled 1014 * manually if the cancellation is wanted. 1015 */ 1016 break; 1017 } 1018 return 1; 1019 } 1020 1021 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) 1022 { 1023 percpu_counter_inc(&fs_info->dev_replace.bio_counter); 1024 } 1025 1026 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount) 1027 { 1028 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount); 1029 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait); 1030 } 1031 1032 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) 1033 { 1034 while (1) { 1035 percpu_counter_inc(&fs_info->dev_replace.bio_counter); 1036 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1037 &fs_info->fs_state))) 1038 break; 1039 1040 btrfs_bio_counter_dec(fs_info); 1041 wait_event(fs_info->dev_replace.replace_wait, 1042 !test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1043 &fs_info->fs_state)); 1044 } 1045 } 1046