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->replace_state = 0; 63 dev_replace->time_started = 0; 64 dev_replace->time_stopped = 0; 65 atomic64_set(&dev_replace->num_write_errors, 0); 66 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 67 dev_replace->cursor_left = 0; 68 dev_replace->committed_cursor_left = 0; 69 dev_replace->cursor_left_last_write_of_item = 0; 70 dev_replace->cursor_right = 0; 71 dev_replace->srcdev = NULL; 72 dev_replace->tgtdev = NULL; 73 dev_replace->is_valid = 0; 74 dev_replace->item_needs_writeback = 0; 75 goto out; 76 } 77 slot = path->slots[0]; 78 eb = path->nodes[0]; 79 item_size = btrfs_item_size_nr(eb, slot); 80 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); 81 82 if (item_size != sizeof(struct btrfs_dev_replace_item)) { 83 btrfs_warn(fs_info, 84 "dev_replace entry found has unexpected size, ignore entry"); 85 goto no_valid_dev_replace_entry_found; 86 } 87 88 src_devid = btrfs_dev_replace_src_devid(eb, ptr); 89 dev_replace->cont_reading_from_srcdev_mode = 90 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); 91 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); 92 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); 93 dev_replace->time_stopped = 94 btrfs_dev_replace_time_stopped(eb, ptr); 95 atomic64_set(&dev_replace->num_write_errors, 96 btrfs_dev_replace_num_write_errors(eb, ptr)); 97 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 98 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); 99 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); 100 dev_replace->committed_cursor_left = dev_replace->cursor_left; 101 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; 102 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); 103 dev_replace->is_valid = 1; 104 105 dev_replace->item_needs_writeback = 0; 106 switch (dev_replace->replace_state) { 107 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 108 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 109 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 110 dev_replace->srcdev = NULL; 111 dev_replace->tgtdev = NULL; 112 break; 113 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 114 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 115 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid, 116 NULL, NULL); 117 dev_replace->tgtdev = btrfs_find_device(fs_info, 118 BTRFS_DEV_REPLACE_DEVID, 119 NULL, NULL); 120 /* 121 * allow 'btrfs dev replace_cancel' if src/tgt device is 122 * missing 123 */ 124 if (!dev_replace->srcdev && 125 !btrfs_test_opt(fs_info, DEGRADED)) { 126 ret = -EIO; 127 btrfs_warn(fs_info, 128 "cannot mount because device replace operation is ongoing and"); 129 btrfs_warn(fs_info, 130 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 131 src_devid); 132 } 133 if (!dev_replace->tgtdev && 134 !btrfs_test_opt(fs_info, DEGRADED)) { 135 ret = -EIO; 136 btrfs_warn(fs_info, 137 "cannot mount because device replace operation is ongoing and"); 138 btrfs_warn(fs_info, 139 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 140 BTRFS_DEV_REPLACE_DEVID); 141 } 142 if (dev_replace->tgtdev) { 143 if (dev_replace->srcdev) { 144 dev_replace->tgtdev->total_bytes = 145 dev_replace->srcdev->total_bytes; 146 dev_replace->tgtdev->disk_total_bytes = 147 dev_replace->srcdev->disk_total_bytes; 148 dev_replace->tgtdev->commit_total_bytes = 149 dev_replace->srcdev->commit_total_bytes; 150 dev_replace->tgtdev->bytes_used = 151 dev_replace->srcdev->bytes_used; 152 dev_replace->tgtdev->commit_bytes_used = 153 dev_replace->srcdev->commit_bytes_used; 154 } 155 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, 156 &dev_replace->tgtdev->dev_state); 157 158 WARN_ON(fs_info->fs_devices->rw_devices == 0); 159 dev_replace->tgtdev->io_width = fs_info->sectorsize; 160 dev_replace->tgtdev->io_align = fs_info->sectorsize; 161 dev_replace->tgtdev->sector_size = fs_info->sectorsize; 162 dev_replace->tgtdev->fs_info = fs_info; 163 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 164 &dev_replace->tgtdev->dev_state); 165 } 166 break; 167 } 168 169 out: 170 btrfs_free_path(path); 171 return ret; 172 } 173 174 /* 175 * Initialize a new device for device replace target from a given source dev 176 * and path. 177 * 178 * Return 0 and new device in @device_out, otherwise return < 0 179 */ 180 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, 181 const char *device_path, 182 struct btrfs_device *srcdev, 183 struct btrfs_device **device_out) 184 { 185 struct btrfs_device *device; 186 struct block_device *bdev; 187 struct list_head *devices; 188 struct rcu_string *name; 189 u64 devid = BTRFS_DEV_REPLACE_DEVID; 190 int ret = 0; 191 192 *device_out = NULL; 193 if (fs_info->fs_devices->seeding) { 194 btrfs_err(fs_info, "the filesystem is a seed filesystem!"); 195 return -EINVAL; 196 } 197 198 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 199 fs_info->bdev_holder); 200 if (IS_ERR(bdev)) { 201 btrfs_err(fs_info, "target device %s is invalid!", device_path); 202 return PTR_ERR(bdev); 203 } 204 205 filemap_write_and_wait(bdev->bd_inode->i_mapping); 206 207 devices = &fs_info->fs_devices->devices; 208 list_for_each_entry(device, devices, dev_list) { 209 if (device->bdev == bdev) { 210 btrfs_err(fs_info, 211 "target device is in the filesystem!"); 212 ret = -EEXIST; 213 goto error; 214 } 215 } 216 217 218 if (i_size_read(bdev->bd_inode) < 219 btrfs_device_get_total_bytes(srcdev)) { 220 btrfs_err(fs_info, 221 "target device is smaller than source device!"); 222 ret = -EINVAL; 223 goto error; 224 } 225 226 227 device = btrfs_alloc_device(NULL, &devid, NULL); 228 if (IS_ERR(device)) { 229 ret = PTR_ERR(device); 230 goto error; 231 } 232 233 name = rcu_string_strdup(device_path, GFP_KERNEL); 234 if (!name) { 235 btrfs_free_device(device); 236 ret = -ENOMEM; 237 goto error; 238 } 239 rcu_assign_pointer(device->name, name); 240 241 mutex_lock(&fs_info->fs_devices->device_list_mutex); 242 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 243 device->generation = 0; 244 device->io_width = fs_info->sectorsize; 245 device->io_align = fs_info->sectorsize; 246 device->sector_size = fs_info->sectorsize; 247 device->total_bytes = btrfs_device_get_total_bytes(srcdev); 248 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); 249 device->bytes_used = btrfs_device_get_bytes_used(srcdev); 250 device->commit_total_bytes = srcdev->commit_total_bytes; 251 device->commit_bytes_used = device->bytes_used; 252 device->fs_info = fs_info; 253 device->bdev = bdev; 254 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 255 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 256 device->mode = FMODE_EXCL; 257 device->dev_stats_valid = 1; 258 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 259 device->fs_devices = fs_info->fs_devices; 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 struct btrfs_fs_info *fs_info) 279 { 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 btrfs_dev_replace_read_lock(dev_replace); 289 if (!dev_replace->is_valid || 290 !dev_replace->item_needs_writeback) { 291 btrfs_dev_replace_read_unlock(dev_replace); 292 return 0; 293 } 294 btrfs_dev_replace_read_unlock(dev_replace); 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 btrfs_dev_replace_write_lock(dev_replace); 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 btrfs_dev_replace_write_unlock(dev_replace); 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 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; 403 404 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid, 405 srcdev_name); 406 if (IS_ERR(src_device)) 407 return PTR_ERR(src_device); 408 409 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name, 410 src_device, &tgt_device); 411 if (ret) 412 return ret; 413 414 /* 415 * Here we commit the transaction to make sure commit_total_bytes 416 * of all the devices are updated. 417 */ 418 trans = btrfs_attach_transaction(root); 419 if (!IS_ERR(trans)) { 420 ret = btrfs_commit_transaction(trans); 421 if (ret) 422 return ret; 423 } else if (PTR_ERR(trans) != -ENOENT) { 424 return PTR_ERR(trans); 425 } 426 427 btrfs_dev_replace_write_lock(dev_replace); 428 switch (dev_replace->replace_state) { 429 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 430 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 431 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 432 break; 433 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 434 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 435 ASSERT(0); 436 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 437 goto leave; 438 } 439 440 dev_replace->cont_reading_from_srcdev_mode = read_src; 441 WARN_ON(!src_device); 442 dev_replace->srcdev = src_device; 443 dev_replace->tgtdev = tgt_device; 444 445 btrfs_info_in_rcu(fs_info, 446 "dev_replace from %s (devid %llu) to %s started", 447 btrfs_dev_name(src_device), 448 src_device->devid, 449 rcu_str_deref(tgt_device->name)); 450 451 /* 452 * from now on, the writes to the srcdev are all duplicated to 453 * go to the tgtdev as well (refer to btrfs_map_block()). 454 */ 455 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 456 dev_replace->time_started = ktime_get_real_seconds(); 457 dev_replace->cursor_left = 0; 458 dev_replace->committed_cursor_left = 0; 459 dev_replace->cursor_left_last_write_of_item = 0; 460 dev_replace->cursor_right = 0; 461 dev_replace->is_valid = 1; 462 dev_replace->item_needs_writeback = 1; 463 atomic64_set(&dev_replace->num_write_errors, 0); 464 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 465 btrfs_dev_replace_write_unlock(dev_replace); 466 467 ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device); 468 if (ret) 469 btrfs_err(fs_info, "kobj add dev failed %d", ret); 470 471 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 472 473 /* force writing the updated state information to disk */ 474 trans = btrfs_start_transaction(root, 0); 475 if (IS_ERR(trans)) { 476 ret = PTR_ERR(trans); 477 btrfs_dev_replace_write_lock(dev_replace); 478 dev_replace->replace_state = 479 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED; 480 dev_replace->srcdev = NULL; 481 dev_replace->tgtdev = NULL; 482 goto leave; 483 } 484 485 ret = btrfs_commit_transaction(trans); 486 WARN_ON(ret); 487 488 /* the disk copy procedure reuses the scrub code */ 489 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 490 btrfs_device_get_total_bytes(src_device), 491 &dev_replace->scrub_progress, 0, 1); 492 493 ret = btrfs_dev_replace_finishing(fs_info, ret); 494 if (ret == -EINPROGRESS) { 495 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS; 496 } else { 497 WARN_ON(ret); 498 } 499 500 return ret; 501 502 leave: 503 btrfs_dev_replace_write_unlock(dev_replace); 504 btrfs_destroy_dev_replace_tgtdev(tgt_device); 505 return ret; 506 } 507 508 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info, 509 struct btrfs_ioctl_dev_replace_args *args) 510 { 511 int ret; 512 513 switch (args->start.cont_reading_from_srcdev_mode) { 514 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 515 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 516 break; 517 default: 518 return -EINVAL; 519 } 520 521 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 522 args->start.tgtdev_name[0] == '\0') 523 return -EINVAL; 524 525 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name, 526 args->start.srcdevid, 527 args->start.srcdev_name, 528 args->start.cont_reading_from_srcdev_mode); 529 args->result = ret; 530 /* don't warn if EINPROGRESS, someone else might be running scrub */ 531 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS) 532 ret = 0; 533 534 return ret; 535 } 536 537 /* 538 * blocked until all in-flight bios operations are finished. 539 */ 540 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) 541 { 542 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 543 wait_event(fs_info->replace_wait, !percpu_counter_sum( 544 &fs_info->bio_counter)); 545 } 546 547 /* 548 * we have removed target device, it is safe to allow new bios request. 549 */ 550 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) 551 { 552 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 553 wake_up(&fs_info->replace_wait); 554 } 555 556 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 557 int scrub_ret) 558 { 559 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 560 struct btrfs_device *tgt_device; 561 struct btrfs_device *src_device; 562 struct btrfs_root *root = fs_info->tree_root; 563 u8 uuid_tmp[BTRFS_UUID_SIZE]; 564 struct btrfs_trans_handle *trans; 565 int ret = 0; 566 567 /* don't allow cancel or unmount to disturb the finishing procedure */ 568 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 569 570 btrfs_dev_replace_read_lock(dev_replace); 571 /* was the operation canceled, or is it finished? */ 572 if (dev_replace->replace_state != 573 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 574 btrfs_dev_replace_read_unlock(dev_replace); 575 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 576 return 0; 577 } 578 579 tgt_device = dev_replace->tgtdev; 580 src_device = dev_replace->srcdev; 581 btrfs_dev_replace_read_unlock(dev_replace); 582 583 /* 584 * flush all outstanding I/O and inode extent mappings before the 585 * copy operation is declared as being finished 586 */ 587 ret = btrfs_start_delalloc_roots(fs_info, -1); 588 if (ret) { 589 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 590 return ret; 591 } 592 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 593 594 trans = btrfs_start_transaction(root, 0); 595 if (IS_ERR(trans)) { 596 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 597 return PTR_ERR(trans); 598 } 599 ret = btrfs_commit_transaction(trans); 600 WARN_ON(ret); 601 602 /* keep away write_all_supers() during the finishing procedure */ 603 mutex_lock(&fs_info->fs_devices->device_list_mutex); 604 mutex_lock(&fs_info->chunk_mutex); 605 btrfs_dev_replace_write_lock(dev_replace); 606 dev_replace->replace_state = 607 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 608 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 609 dev_replace->tgtdev = NULL; 610 dev_replace->srcdev = NULL; 611 dev_replace->time_stopped = ktime_get_real_seconds(); 612 dev_replace->item_needs_writeback = 1; 613 614 /* replace old device with new one in mapping tree */ 615 if (!scrub_ret) { 616 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 617 src_device, 618 tgt_device); 619 } else { 620 btrfs_err_in_rcu(fs_info, 621 "btrfs_scrub_dev(%s, %llu, %s) failed %d", 622 btrfs_dev_name(src_device), 623 src_device->devid, 624 rcu_str_deref(tgt_device->name), scrub_ret); 625 btrfs_dev_replace_write_unlock(dev_replace); 626 mutex_unlock(&fs_info->chunk_mutex); 627 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 628 btrfs_rm_dev_replace_blocked(fs_info); 629 if (tgt_device) 630 btrfs_destroy_dev_replace_tgtdev(tgt_device); 631 btrfs_rm_dev_replace_unblocked(fs_info); 632 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 633 634 return scrub_ret; 635 } 636 637 btrfs_info_in_rcu(fs_info, 638 "dev_replace from %s (devid %llu) to %s finished", 639 btrfs_dev_name(src_device), 640 src_device->devid, 641 rcu_str_deref(tgt_device->name)); 642 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state); 643 tgt_device->devid = src_device->devid; 644 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 645 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 646 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 647 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 648 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes); 649 btrfs_device_set_disk_total_bytes(tgt_device, 650 src_device->disk_total_bytes); 651 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used); 652 ASSERT(list_empty(&src_device->resized_list)); 653 tgt_device->commit_total_bytes = src_device->commit_total_bytes; 654 tgt_device->commit_bytes_used = src_device->bytes_used; 655 656 btrfs_assign_next_active_device(src_device, tgt_device); 657 658 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); 659 fs_info->fs_devices->rw_devices++; 660 661 btrfs_dev_replace_write_unlock(dev_replace); 662 663 btrfs_rm_dev_replace_blocked(fs_info); 664 665 btrfs_rm_dev_replace_remove_srcdev(src_device); 666 667 btrfs_rm_dev_replace_unblocked(fs_info); 668 669 /* 670 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will 671 * update on-disk dev stats value during commit transaction 672 */ 673 atomic_inc(&tgt_device->dev_stats_ccnt); 674 675 /* 676 * this is again a consistent state where no dev_replace procedure 677 * is running, the target device is part of the filesystem, the 678 * source device is not part of the filesystem anymore and its 1st 679 * superblock is scratched out so that it is no longer marked to 680 * belong to this filesystem. 681 */ 682 mutex_unlock(&fs_info->chunk_mutex); 683 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 684 685 /* replace the sysfs entry */ 686 btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device); 687 btrfs_rm_dev_replace_free_srcdev(fs_info, src_device); 688 689 /* write back the superblocks */ 690 trans = btrfs_start_transaction(root, 0); 691 if (!IS_ERR(trans)) 692 btrfs_commit_transaction(trans); 693 694 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 695 696 return 0; 697 } 698 699 static void btrfs_dev_replace_update_device_in_mapping_tree( 700 struct btrfs_fs_info *fs_info, 701 struct btrfs_device *srcdev, 702 struct btrfs_device *tgtdev) 703 { 704 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; 705 struct extent_map *em; 706 struct map_lookup *map; 707 u64 start = 0; 708 int i; 709 710 write_lock(&em_tree->lock); 711 do { 712 em = lookup_extent_mapping(em_tree, start, (u64)-1); 713 if (!em) 714 break; 715 map = em->map_lookup; 716 for (i = 0; i < map->num_stripes; i++) 717 if (srcdev == map->stripes[i].dev) 718 map->stripes[i].dev = tgtdev; 719 start = em->start + em->len; 720 free_extent_map(em); 721 } while (start); 722 write_unlock(&em_tree->lock); 723 } 724 725 /* 726 * Read progress of device replace status according to the state and last 727 * stored position. The value format is the same as for 728 * btrfs_dev_replace::progress_1000 729 */ 730 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info) 731 { 732 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 733 u64 ret = 0; 734 735 switch (dev_replace->replace_state) { 736 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 737 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 738 ret = 0; 739 break; 740 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 741 ret = 1000; 742 break; 743 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 744 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 745 ret = div64_u64(dev_replace->cursor_left, 746 div_u64(btrfs_device_get_total_bytes( 747 dev_replace->srcdev), 1000)); 748 break; 749 } 750 751 return ret; 752 } 753 754 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 755 struct btrfs_ioctl_dev_replace_args *args) 756 { 757 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 758 759 btrfs_dev_replace_read_lock(dev_replace); 760 /* even if !dev_replace_is_valid, the values are good enough for 761 * the replace_status ioctl */ 762 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 763 args->status.replace_state = dev_replace->replace_state; 764 args->status.time_started = dev_replace->time_started; 765 args->status.time_stopped = dev_replace->time_stopped; 766 args->status.num_write_errors = 767 atomic64_read(&dev_replace->num_write_errors); 768 args->status.num_uncorrectable_read_errors = 769 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 770 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info); 771 btrfs_dev_replace_read_unlock(dev_replace); 772 } 773 774 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 775 { 776 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 777 struct btrfs_device *tgt_device = NULL; 778 struct btrfs_device *src_device = NULL; 779 struct btrfs_trans_handle *trans; 780 struct btrfs_root *root = fs_info->tree_root; 781 int result; 782 int ret; 783 784 if (sb_rdonly(fs_info->sb)) 785 return -EROFS; 786 787 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 788 btrfs_dev_replace_write_lock(dev_replace); 789 switch (dev_replace->replace_state) { 790 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 791 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 792 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 793 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 794 btrfs_dev_replace_write_unlock(dev_replace); 795 goto leave; 796 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 797 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 798 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 799 tgt_device = dev_replace->tgtdev; 800 src_device = dev_replace->srcdev; 801 dev_replace->tgtdev = NULL; 802 dev_replace->srcdev = NULL; 803 break; 804 } 805 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 806 dev_replace->time_stopped = ktime_get_real_seconds(); 807 dev_replace->item_needs_writeback = 1; 808 btrfs_dev_replace_write_unlock(dev_replace); 809 btrfs_scrub_cancel(fs_info); 810 811 trans = btrfs_start_transaction(root, 0); 812 if (IS_ERR(trans)) { 813 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 814 return PTR_ERR(trans); 815 } 816 ret = btrfs_commit_transaction(trans); 817 WARN_ON(ret); 818 819 btrfs_info_in_rcu(fs_info, 820 "dev_replace from %s (devid %llu) to %s canceled", 821 btrfs_dev_name(src_device), src_device->devid, 822 btrfs_dev_name(tgt_device)); 823 824 if (tgt_device) 825 btrfs_destroy_dev_replace_tgtdev(tgt_device); 826 827 leave: 828 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 829 return result; 830 } 831 832 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 833 { 834 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 835 836 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 837 btrfs_dev_replace_write_lock(dev_replace); 838 switch (dev_replace->replace_state) { 839 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 840 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 841 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 842 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 843 break; 844 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 845 dev_replace->replace_state = 846 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 847 dev_replace->time_stopped = ktime_get_real_seconds(); 848 dev_replace->item_needs_writeback = 1; 849 btrfs_info(fs_info, "suspending dev_replace for unmount"); 850 break; 851 } 852 853 btrfs_dev_replace_write_unlock(dev_replace); 854 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 855 } 856 857 /* resume dev_replace procedure that was interrupted by unmount */ 858 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 859 { 860 struct task_struct *task; 861 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 862 863 btrfs_dev_replace_write_lock(dev_replace); 864 switch (dev_replace->replace_state) { 865 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 866 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 867 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 868 btrfs_dev_replace_write_unlock(dev_replace); 869 return 0; 870 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 871 break; 872 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 873 dev_replace->replace_state = 874 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 875 break; 876 } 877 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 878 btrfs_info(fs_info, 879 "cannot continue dev_replace, tgtdev is missing"); 880 btrfs_info(fs_info, 881 "you may cancel the operation after 'mount -o degraded'"); 882 btrfs_dev_replace_write_unlock(dev_replace); 883 return 0; 884 } 885 btrfs_dev_replace_write_unlock(dev_replace); 886 887 /* 888 * This could collide with a paused balance, but the exclusive op logic 889 * should never allow both to start and pause. We don't want to allow 890 * dev-replace to start anyway. 891 */ 892 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { 893 btrfs_info(fs_info, 894 "cannot resume dev-replace, other exclusive operation running"); 895 return 0; 896 } 897 898 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 899 return PTR_ERR_OR_ZERO(task); 900 } 901 902 static int btrfs_dev_replace_kthread(void *data) 903 { 904 struct btrfs_fs_info *fs_info = data; 905 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 906 u64 progress; 907 int ret; 908 909 progress = btrfs_dev_replace_progress(fs_info); 910 progress = div_u64(progress, 10); 911 btrfs_info_in_rcu(fs_info, 912 "continuing dev_replace from %s (devid %llu) to target %s @%u%%", 913 btrfs_dev_name(dev_replace->srcdev), 914 dev_replace->srcdev->devid, 915 btrfs_dev_name(dev_replace->tgtdev), 916 (unsigned int)progress); 917 918 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 919 dev_replace->committed_cursor_left, 920 btrfs_device_get_total_bytes(dev_replace->srcdev), 921 &dev_replace->scrub_progress, 0, 1); 922 ret = btrfs_dev_replace_finishing(fs_info, ret); 923 WARN_ON(ret); 924 925 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 926 return 0; 927 } 928 929 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 930 { 931 if (!dev_replace->is_valid) 932 return 0; 933 934 switch (dev_replace->replace_state) { 935 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 936 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 937 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 938 return 0; 939 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 940 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 941 /* 942 * return true even if tgtdev is missing (this is 943 * something that can happen if the dev_replace 944 * procedure is suspended by an umount and then 945 * the tgtdev is missing (or "btrfs dev scan") was 946 * not called and the the filesystem is remounted 947 * in degraded state. This does not stop the 948 * dev_replace procedure. It needs to be canceled 949 * manually if the cancellation is wanted. 950 */ 951 break; 952 } 953 return 1; 954 } 955 956 void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace) 957 { 958 read_lock(&dev_replace->lock); 959 } 960 961 void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace) 962 { 963 read_unlock(&dev_replace->lock); 964 } 965 966 void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace) 967 { 968 again: 969 wait_event(dev_replace->read_lock_wq, 970 atomic_read(&dev_replace->blocking_readers) == 0); 971 write_lock(&dev_replace->lock); 972 if (atomic_read(&dev_replace->blocking_readers)) { 973 write_unlock(&dev_replace->lock); 974 goto again; 975 } 976 } 977 978 void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace) 979 { 980 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0); 981 write_unlock(&dev_replace->lock); 982 } 983 984 /* inc blocking cnt and release read lock */ 985 void btrfs_dev_replace_set_lock_blocking( 986 struct btrfs_dev_replace *dev_replace) 987 { 988 /* only set blocking for read lock */ 989 atomic_inc(&dev_replace->blocking_readers); 990 read_unlock(&dev_replace->lock); 991 } 992 993 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) 994 { 995 percpu_counter_inc(&fs_info->bio_counter); 996 } 997 998 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount) 999 { 1000 percpu_counter_sub(&fs_info->bio_counter, amount); 1001 cond_wake_up_nomb(&fs_info->replace_wait); 1002 } 1003 1004 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) 1005 { 1006 while (1) { 1007 percpu_counter_inc(&fs_info->bio_counter); 1008 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1009 &fs_info->fs_state))) 1010 break; 1011 1012 btrfs_bio_counter_dec(fs_info); 1013 wait_event(fs_info->replace_wait, 1014 !test_bit(BTRFS_FS_STATE_DEV_REPLACING, 1015 &fs_info->fs_state)); 1016 } 1017 } 1018