1 /* 2 * Copyright (C) STRATO AG 2012. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 #include <linux/sched.h> 19 #include <linux/bio.h> 20 #include <linux/slab.h> 21 #include <linux/buffer_head.h> 22 #include <linux/blkdev.h> 23 #include <linux/random.h> 24 #include <linux/iocontext.h> 25 #include <linux/capability.h> 26 #include <linux/kthread.h> 27 #include <linux/math64.h> 28 #include <asm/div64.h> 29 #include "ctree.h" 30 #include "extent_map.h" 31 #include "disk-io.h" 32 #include "transaction.h" 33 #include "print-tree.h" 34 #include "volumes.h" 35 #include "async-thread.h" 36 #include "check-integrity.h" 37 #include "rcu-string.h" 38 #include "dev-replace.h" 39 #include "sysfs.h" 40 41 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 42 int scrub_ret); 43 static void btrfs_dev_replace_update_device_in_mapping_tree( 44 struct btrfs_fs_info *fs_info, 45 struct btrfs_device *srcdev, 46 struct btrfs_device *tgtdev); 47 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info); 48 static int btrfs_dev_replace_kthread(void *data); 49 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info); 50 51 52 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) 53 { 54 struct btrfs_key key; 55 struct btrfs_root *dev_root = fs_info->dev_root; 56 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 57 struct extent_buffer *eb; 58 int slot; 59 int ret = 0; 60 struct btrfs_path *path = NULL; 61 int item_size; 62 struct btrfs_dev_replace_item *ptr; 63 u64 src_devid; 64 65 path = btrfs_alloc_path(); 66 if (!path) { 67 ret = -ENOMEM; 68 goto out; 69 } 70 71 key.objectid = 0; 72 key.type = BTRFS_DEV_REPLACE_KEY; 73 key.offset = 0; 74 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); 75 if (ret) { 76 no_valid_dev_replace_entry_found: 77 ret = 0; 78 dev_replace->replace_state = 79 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED; 80 dev_replace->cont_reading_from_srcdev_mode = 81 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; 82 dev_replace->replace_state = 0; 83 dev_replace->time_started = 0; 84 dev_replace->time_stopped = 0; 85 atomic64_set(&dev_replace->num_write_errors, 0); 86 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 87 dev_replace->cursor_left = 0; 88 dev_replace->committed_cursor_left = 0; 89 dev_replace->cursor_left_last_write_of_item = 0; 90 dev_replace->cursor_right = 0; 91 dev_replace->srcdev = NULL; 92 dev_replace->tgtdev = NULL; 93 dev_replace->is_valid = 0; 94 dev_replace->item_needs_writeback = 0; 95 goto out; 96 } 97 slot = path->slots[0]; 98 eb = path->nodes[0]; 99 item_size = btrfs_item_size_nr(eb, slot); 100 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); 101 102 if (item_size != sizeof(struct btrfs_dev_replace_item)) { 103 btrfs_warn(fs_info, 104 "dev_replace entry found has unexpected size, ignore entry"); 105 goto no_valid_dev_replace_entry_found; 106 } 107 108 src_devid = btrfs_dev_replace_src_devid(eb, ptr); 109 dev_replace->cont_reading_from_srcdev_mode = 110 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); 111 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); 112 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); 113 dev_replace->time_stopped = 114 btrfs_dev_replace_time_stopped(eb, ptr); 115 atomic64_set(&dev_replace->num_write_errors, 116 btrfs_dev_replace_num_write_errors(eb, ptr)); 117 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 118 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); 119 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); 120 dev_replace->committed_cursor_left = dev_replace->cursor_left; 121 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; 122 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); 123 dev_replace->is_valid = 1; 124 125 dev_replace->item_needs_writeback = 0; 126 switch (dev_replace->replace_state) { 127 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 128 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 129 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 130 dev_replace->srcdev = NULL; 131 dev_replace->tgtdev = NULL; 132 break; 133 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 134 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 135 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid, 136 NULL, NULL); 137 dev_replace->tgtdev = btrfs_find_device(fs_info, 138 BTRFS_DEV_REPLACE_DEVID, 139 NULL, NULL); 140 /* 141 * allow 'btrfs dev replace_cancel' if src/tgt device is 142 * missing 143 */ 144 if (!dev_replace->srcdev && 145 !btrfs_test_opt(dev_root, DEGRADED)) { 146 ret = -EIO; 147 btrfs_warn(fs_info, 148 "cannot mount because device replace operation is ongoing and"); 149 btrfs_warn(fs_info, 150 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 151 src_devid); 152 } 153 if (!dev_replace->tgtdev && 154 !btrfs_test_opt(dev_root, DEGRADED)) { 155 ret = -EIO; 156 btrfs_warn(fs_info, 157 "cannot mount because device replace operation is ongoing and"); 158 btrfs_warn(fs_info, 159 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 160 BTRFS_DEV_REPLACE_DEVID); 161 } 162 if (dev_replace->tgtdev) { 163 if (dev_replace->srcdev) { 164 dev_replace->tgtdev->total_bytes = 165 dev_replace->srcdev->total_bytes; 166 dev_replace->tgtdev->disk_total_bytes = 167 dev_replace->srcdev->disk_total_bytes; 168 dev_replace->tgtdev->commit_total_bytes = 169 dev_replace->srcdev->commit_total_bytes; 170 dev_replace->tgtdev->bytes_used = 171 dev_replace->srcdev->bytes_used; 172 dev_replace->tgtdev->commit_bytes_used = 173 dev_replace->srcdev->commit_bytes_used; 174 } 175 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1; 176 btrfs_init_dev_replace_tgtdev_for_resume(fs_info, 177 dev_replace->tgtdev); 178 } 179 break; 180 } 181 182 out: 183 btrfs_free_path(path); 184 return ret; 185 } 186 187 /* 188 * called from commit_transaction. Writes changed device replace state to 189 * disk. 190 */ 191 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans, 192 struct btrfs_fs_info *fs_info) 193 { 194 int ret; 195 struct btrfs_root *dev_root = fs_info->dev_root; 196 struct btrfs_path *path; 197 struct btrfs_key key; 198 struct extent_buffer *eb; 199 struct btrfs_dev_replace_item *ptr; 200 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 201 202 btrfs_dev_replace_lock(dev_replace, 0); 203 if (!dev_replace->is_valid || 204 !dev_replace->item_needs_writeback) { 205 btrfs_dev_replace_unlock(dev_replace, 0); 206 return 0; 207 } 208 btrfs_dev_replace_unlock(dev_replace, 0); 209 210 key.objectid = 0; 211 key.type = BTRFS_DEV_REPLACE_KEY; 212 key.offset = 0; 213 214 path = btrfs_alloc_path(); 215 if (!path) { 216 ret = -ENOMEM; 217 goto out; 218 } 219 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 220 if (ret < 0) { 221 btrfs_warn(fs_info, "error %d while searching for dev_replace item!", 222 ret); 223 goto out; 224 } 225 226 if (ret == 0 && 227 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 228 /* 229 * need to delete old one and insert a new one. 230 * Since no attempt is made to recover any old state, if the 231 * dev_replace state is 'running', the data on the target 232 * drive is lost. 233 * It would be possible to recover the state: just make sure 234 * that the beginning of the item is never changed and always 235 * contains all the essential information. Then read this 236 * minimal set of information and use it as a base for the 237 * new state. 238 */ 239 ret = btrfs_del_item(trans, dev_root, path); 240 if (ret != 0) { 241 btrfs_warn(fs_info, "delete too small dev_replace item failed %d!", 242 ret); 243 goto out; 244 } 245 ret = 1; 246 } 247 248 if (ret == 1) { 249 /* need to insert a new item */ 250 btrfs_release_path(path); 251 ret = btrfs_insert_empty_item(trans, dev_root, path, 252 &key, sizeof(*ptr)); 253 if (ret < 0) { 254 btrfs_warn(fs_info, "insert dev_replace item failed %d!", 255 ret); 256 goto out; 257 } 258 } 259 260 eb = path->nodes[0]; 261 ptr = btrfs_item_ptr(eb, path->slots[0], 262 struct btrfs_dev_replace_item); 263 264 btrfs_dev_replace_lock(dev_replace, 1); 265 if (dev_replace->srcdev) 266 btrfs_set_dev_replace_src_devid(eb, ptr, 267 dev_replace->srcdev->devid); 268 else 269 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); 270 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, 271 dev_replace->cont_reading_from_srcdev_mode); 272 btrfs_set_dev_replace_replace_state(eb, ptr, 273 dev_replace->replace_state); 274 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); 275 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); 276 btrfs_set_dev_replace_num_write_errors(eb, ptr, 277 atomic64_read(&dev_replace->num_write_errors)); 278 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, 279 atomic64_read(&dev_replace->num_uncorrectable_read_errors)); 280 dev_replace->cursor_left_last_write_of_item = 281 dev_replace->cursor_left; 282 btrfs_set_dev_replace_cursor_left(eb, ptr, 283 dev_replace->cursor_left_last_write_of_item); 284 btrfs_set_dev_replace_cursor_right(eb, ptr, 285 dev_replace->cursor_right); 286 dev_replace->item_needs_writeback = 0; 287 btrfs_dev_replace_unlock(dev_replace, 1); 288 289 btrfs_mark_buffer_dirty(eb); 290 291 out: 292 btrfs_free_path(path); 293 294 return ret; 295 } 296 297 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info) 298 { 299 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 300 301 dev_replace->committed_cursor_left = 302 dev_replace->cursor_left_last_write_of_item; 303 } 304 305 int btrfs_dev_replace_start(struct btrfs_root *root, 306 struct btrfs_ioctl_dev_replace_args *args) 307 { 308 struct btrfs_trans_handle *trans; 309 struct btrfs_fs_info *fs_info = root->fs_info; 310 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 311 int ret; 312 struct btrfs_device *tgt_device = NULL; 313 struct btrfs_device *src_device = NULL; 314 315 switch (args->start.cont_reading_from_srcdev_mode) { 316 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 317 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 318 break; 319 default: 320 return -EINVAL; 321 } 322 323 /* the disk copy procedure reuses the scrub code */ 324 mutex_lock(&fs_info->volume_mutex); 325 ret = btrfs_find_device_by_devspec(root, args->start.srcdevid, 326 args->start.srcdev_name, 327 &src_device); 328 if (ret) { 329 mutex_unlock(&fs_info->volume_mutex); 330 return ret; 331 } 332 333 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name, 334 src_device, &tgt_device); 335 mutex_unlock(&fs_info->volume_mutex); 336 if (ret) 337 return ret; 338 339 /* 340 * Here we commit the transaction to make sure commit_total_bytes 341 * of all the devices are updated. 342 */ 343 trans = btrfs_attach_transaction(root); 344 if (!IS_ERR(trans)) { 345 ret = btrfs_commit_transaction(trans, root); 346 if (ret) 347 return ret; 348 } else if (PTR_ERR(trans) != -ENOENT) { 349 return PTR_ERR(trans); 350 } 351 352 btrfs_dev_replace_lock(dev_replace, 1); 353 switch (dev_replace->replace_state) { 354 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 355 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 356 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 357 break; 358 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 359 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 360 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 361 goto leave; 362 } 363 364 dev_replace->cont_reading_from_srcdev_mode = 365 args->start.cont_reading_from_srcdev_mode; 366 WARN_ON(!src_device); 367 dev_replace->srcdev = src_device; 368 WARN_ON(!tgt_device); 369 dev_replace->tgtdev = tgt_device; 370 371 btrfs_info_in_rcu(root->fs_info, 372 "dev_replace from %s (devid %llu) to %s started", 373 src_device->missing ? "<missing disk>" : 374 rcu_str_deref(src_device->name), 375 src_device->devid, 376 rcu_str_deref(tgt_device->name)); 377 378 /* 379 * from now on, the writes to the srcdev are all duplicated to 380 * go to the tgtdev as well (refer to btrfs_map_block()). 381 */ 382 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 383 dev_replace->time_started = get_seconds(); 384 dev_replace->cursor_left = 0; 385 dev_replace->committed_cursor_left = 0; 386 dev_replace->cursor_left_last_write_of_item = 0; 387 dev_replace->cursor_right = 0; 388 dev_replace->is_valid = 1; 389 dev_replace->item_needs_writeback = 1; 390 atomic64_set(&dev_replace->num_write_errors, 0); 391 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 392 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 393 btrfs_dev_replace_unlock(dev_replace, 1); 394 395 ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device); 396 if (ret) 397 btrfs_err(root->fs_info, "kobj add dev failed %d\n", ret); 398 399 btrfs_wait_ordered_roots(root->fs_info, -1); 400 401 /* force writing the updated state information to disk */ 402 trans = btrfs_start_transaction(root, 0); 403 if (IS_ERR(trans)) { 404 ret = PTR_ERR(trans); 405 btrfs_dev_replace_lock(dev_replace, 1); 406 goto leave; 407 } 408 409 ret = btrfs_commit_transaction(trans, root); 410 WARN_ON(ret); 411 412 /* the disk copy procedure reuses the scrub code */ 413 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 414 btrfs_device_get_total_bytes(src_device), 415 &dev_replace->scrub_progress, 0, 1); 416 417 ret = btrfs_dev_replace_finishing(root->fs_info, ret); 418 /* don't warn if EINPROGRESS, someone else might be running scrub */ 419 if (ret == -EINPROGRESS) { 420 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS; 421 ret = 0; 422 } else { 423 WARN_ON(ret); 424 } 425 426 return ret; 427 428 leave: 429 dev_replace->srcdev = NULL; 430 dev_replace->tgtdev = NULL; 431 btrfs_dev_replace_unlock(dev_replace, 1); 432 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 433 return ret; 434 } 435 436 /* 437 * blocked until all flighting bios are finished. 438 */ 439 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) 440 { 441 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 442 wait_event(fs_info->replace_wait, !percpu_counter_sum( 443 &fs_info->bio_counter)); 444 } 445 446 /* 447 * we have removed target device, it is safe to allow new bios request. 448 */ 449 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) 450 { 451 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 452 wake_up(&fs_info->replace_wait); 453 } 454 455 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 456 int scrub_ret) 457 { 458 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 459 struct btrfs_device *tgt_device; 460 struct btrfs_device *src_device; 461 struct btrfs_root *root = fs_info->tree_root; 462 u8 uuid_tmp[BTRFS_UUID_SIZE]; 463 struct btrfs_trans_handle *trans; 464 int ret = 0; 465 466 /* don't allow cancel or unmount to disturb the finishing procedure */ 467 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 468 469 btrfs_dev_replace_lock(dev_replace, 0); 470 /* was the operation canceled, or is it finished? */ 471 if (dev_replace->replace_state != 472 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 473 btrfs_dev_replace_unlock(dev_replace, 0); 474 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 475 return 0; 476 } 477 478 tgt_device = dev_replace->tgtdev; 479 src_device = dev_replace->srcdev; 480 btrfs_dev_replace_unlock(dev_replace, 0); 481 482 /* 483 * flush all outstanding I/O and inode extent mappings before the 484 * copy operation is declared as being finished 485 */ 486 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1); 487 if (ret) { 488 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 489 return ret; 490 } 491 btrfs_wait_ordered_roots(root->fs_info, -1); 492 493 trans = btrfs_start_transaction(root, 0); 494 if (IS_ERR(trans)) { 495 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 496 return PTR_ERR(trans); 497 } 498 ret = btrfs_commit_transaction(trans, root); 499 WARN_ON(ret); 500 501 mutex_lock(&uuid_mutex); 502 /* keep away write_all_supers() during the finishing procedure */ 503 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 504 mutex_lock(&root->fs_info->chunk_mutex); 505 btrfs_dev_replace_lock(dev_replace, 1); 506 dev_replace->replace_state = 507 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 508 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 509 dev_replace->tgtdev = NULL; 510 dev_replace->srcdev = NULL; 511 dev_replace->time_stopped = get_seconds(); 512 dev_replace->item_needs_writeback = 1; 513 514 /* replace old device with new one in mapping tree */ 515 if (!scrub_ret) { 516 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 517 src_device, 518 tgt_device); 519 } else { 520 btrfs_err_in_rcu(root->fs_info, 521 "btrfs_scrub_dev(%s, %llu, %s) failed %d", 522 src_device->missing ? "<missing disk>" : 523 rcu_str_deref(src_device->name), 524 src_device->devid, 525 rcu_str_deref(tgt_device->name), scrub_ret); 526 btrfs_dev_replace_unlock(dev_replace, 1); 527 mutex_unlock(&root->fs_info->chunk_mutex); 528 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 529 mutex_unlock(&uuid_mutex); 530 if (tgt_device) 531 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 532 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 533 534 return scrub_ret; 535 } 536 537 btrfs_info_in_rcu(root->fs_info, 538 "dev_replace from %s (devid %llu) to %s finished", 539 src_device->missing ? "<missing disk>" : 540 rcu_str_deref(src_device->name), 541 src_device->devid, 542 rcu_str_deref(tgt_device->name)); 543 tgt_device->is_tgtdev_for_dev_replace = 0; 544 tgt_device->devid = src_device->devid; 545 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 546 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 547 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 548 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 549 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes); 550 btrfs_device_set_disk_total_bytes(tgt_device, 551 src_device->disk_total_bytes); 552 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used); 553 ASSERT(list_empty(&src_device->resized_list)); 554 tgt_device->commit_total_bytes = src_device->commit_total_bytes; 555 tgt_device->commit_bytes_used = src_device->bytes_used; 556 if (fs_info->sb->s_bdev == src_device->bdev) 557 fs_info->sb->s_bdev = tgt_device->bdev; 558 if (fs_info->fs_devices->latest_bdev == src_device->bdev) 559 fs_info->fs_devices->latest_bdev = tgt_device->bdev; 560 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); 561 fs_info->fs_devices->rw_devices++; 562 563 btrfs_dev_replace_unlock(dev_replace, 1); 564 565 btrfs_rm_dev_replace_blocked(fs_info); 566 567 btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device); 568 569 btrfs_rm_dev_replace_unblocked(fs_info); 570 571 /* 572 * this is again a consistent state where no dev_replace procedure 573 * is running, the target device is part of the filesystem, the 574 * source device is not part of the filesystem anymore and its 1st 575 * superblock is scratched out so that it is no longer marked to 576 * belong to this filesystem. 577 */ 578 mutex_unlock(&root->fs_info->chunk_mutex); 579 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 580 mutex_unlock(&uuid_mutex); 581 582 /* replace the sysfs entry */ 583 btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device); 584 btrfs_rm_dev_replace_free_srcdev(fs_info, src_device); 585 586 /* write back the superblocks */ 587 trans = btrfs_start_transaction(root, 0); 588 if (!IS_ERR(trans)) 589 btrfs_commit_transaction(trans, root); 590 591 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 592 593 return 0; 594 } 595 596 static void btrfs_dev_replace_update_device_in_mapping_tree( 597 struct btrfs_fs_info *fs_info, 598 struct btrfs_device *srcdev, 599 struct btrfs_device *tgtdev) 600 { 601 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; 602 struct extent_map *em; 603 struct map_lookup *map; 604 u64 start = 0; 605 int i; 606 607 write_lock(&em_tree->lock); 608 do { 609 em = lookup_extent_mapping(em_tree, start, (u64)-1); 610 if (!em) 611 break; 612 map = em->map_lookup; 613 for (i = 0; i < map->num_stripes; i++) 614 if (srcdev == map->stripes[i].dev) 615 map->stripes[i].dev = tgtdev; 616 start = em->start + em->len; 617 free_extent_map(em); 618 } while (start); 619 write_unlock(&em_tree->lock); 620 } 621 622 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 623 struct btrfs_ioctl_dev_replace_args *args) 624 { 625 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 626 struct btrfs_device *srcdev; 627 628 btrfs_dev_replace_lock(dev_replace, 0); 629 /* even if !dev_replace_is_valid, the values are good enough for 630 * the replace_status ioctl */ 631 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 632 args->status.replace_state = dev_replace->replace_state; 633 args->status.time_started = dev_replace->time_started; 634 args->status.time_stopped = dev_replace->time_stopped; 635 args->status.num_write_errors = 636 atomic64_read(&dev_replace->num_write_errors); 637 args->status.num_uncorrectable_read_errors = 638 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 639 switch (dev_replace->replace_state) { 640 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 641 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 642 args->status.progress_1000 = 0; 643 break; 644 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 645 args->status.progress_1000 = 1000; 646 break; 647 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 648 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 649 srcdev = dev_replace->srcdev; 650 args->status.progress_1000 = div_u64(dev_replace->cursor_left, 651 div_u64(btrfs_device_get_total_bytes(srcdev), 1000)); 652 break; 653 } 654 btrfs_dev_replace_unlock(dev_replace, 0); 655 } 656 657 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info, 658 struct btrfs_ioctl_dev_replace_args *args) 659 { 660 args->result = __btrfs_dev_replace_cancel(fs_info); 661 return 0; 662 } 663 664 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 665 { 666 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 667 struct btrfs_device *tgt_device = NULL; 668 struct btrfs_trans_handle *trans; 669 struct btrfs_root *root = fs_info->tree_root; 670 u64 result; 671 int ret; 672 673 if (fs_info->sb->s_flags & MS_RDONLY) 674 return -EROFS; 675 676 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 677 btrfs_dev_replace_lock(dev_replace, 1); 678 switch (dev_replace->replace_state) { 679 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 680 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 681 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 682 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 683 btrfs_dev_replace_unlock(dev_replace, 1); 684 goto leave; 685 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 686 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 687 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 688 tgt_device = dev_replace->tgtdev; 689 dev_replace->tgtdev = NULL; 690 dev_replace->srcdev = NULL; 691 break; 692 } 693 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 694 dev_replace->time_stopped = get_seconds(); 695 dev_replace->item_needs_writeback = 1; 696 btrfs_dev_replace_unlock(dev_replace, 1); 697 btrfs_scrub_cancel(fs_info); 698 699 trans = btrfs_start_transaction(root, 0); 700 if (IS_ERR(trans)) { 701 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 702 return PTR_ERR(trans); 703 } 704 ret = btrfs_commit_transaction(trans, root); 705 WARN_ON(ret); 706 if (tgt_device) 707 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 708 709 leave: 710 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 711 return result; 712 } 713 714 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 715 { 716 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 717 718 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 719 btrfs_dev_replace_lock(dev_replace, 1); 720 switch (dev_replace->replace_state) { 721 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 722 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 723 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 724 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 725 break; 726 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 727 dev_replace->replace_state = 728 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 729 dev_replace->time_stopped = get_seconds(); 730 dev_replace->item_needs_writeback = 1; 731 btrfs_info(fs_info, "suspending dev_replace for unmount"); 732 break; 733 } 734 735 btrfs_dev_replace_unlock(dev_replace, 1); 736 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 737 } 738 739 /* resume dev_replace procedure that was interrupted by unmount */ 740 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 741 { 742 struct task_struct *task; 743 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 744 745 btrfs_dev_replace_lock(dev_replace, 1); 746 switch (dev_replace->replace_state) { 747 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 748 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 749 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 750 btrfs_dev_replace_unlock(dev_replace, 1); 751 return 0; 752 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 753 break; 754 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 755 dev_replace->replace_state = 756 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 757 break; 758 } 759 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 760 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing"); 761 btrfs_info(fs_info, 762 "you may cancel the operation after 'mount -o degraded'"); 763 btrfs_dev_replace_unlock(dev_replace, 1); 764 return 0; 765 } 766 btrfs_dev_replace_unlock(dev_replace, 1); 767 768 WARN_ON(atomic_xchg( 769 &fs_info->mutually_exclusive_operation_running, 1)); 770 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 771 return PTR_ERR_OR_ZERO(task); 772 } 773 774 static int btrfs_dev_replace_kthread(void *data) 775 { 776 struct btrfs_fs_info *fs_info = data; 777 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 778 struct btrfs_ioctl_dev_replace_args *status_args; 779 u64 progress; 780 781 status_args = kzalloc(sizeof(*status_args), GFP_KERNEL); 782 if (status_args) { 783 btrfs_dev_replace_status(fs_info, status_args); 784 progress = status_args->status.progress_1000; 785 kfree(status_args); 786 progress = div_u64(progress, 10); 787 btrfs_info_in_rcu(fs_info, 788 "continuing dev_replace from %s (devid %llu) to %s @%u%%", 789 dev_replace->srcdev->missing ? "<missing disk>" : 790 rcu_str_deref(dev_replace->srcdev->name), 791 dev_replace->srcdev->devid, 792 dev_replace->tgtdev ? 793 rcu_str_deref(dev_replace->tgtdev->name) : 794 "<missing target disk>", 795 (unsigned int)progress); 796 } 797 btrfs_dev_replace_continue_on_mount(fs_info); 798 atomic_set(&fs_info->mutually_exclusive_operation_running, 0); 799 800 return 0; 801 } 802 803 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info) 804 { 805 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 806 int ret; 807 808 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 809 dev_replace->committed_cursor_left, 810 btrfs_device_get_total_bytes(dev_replace->srcdev), 811 &dev_replace->scrub_progress, 0, 1); 812 ret = btrfs_dev_replace_finishing(fs_info, ret); 813 WARN_ON(ret); 814 return 0; 815 } 816 817 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 818 { 819 if (!dev_replace->is_valid) 820 return 0; 821 822 switch (dev_replace->replace_state) { 823 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 824 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 825 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 826 return 0; 827 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 828 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 829 /* 830 * return true even if tgtdev is missing (this is 831 * something that can happen if the dev_replace 832 * procedure is suspended by an umount and then 833 * the tgtdev is missing (or "btrfs dev scan") was 834 * not called and the the filesystem is remounted 835 * in degraded state. This does not stop the 836 * dev_replace procedure. It needs to be canceled 837 * manually if the cancellation is wanted. 838 */ 839 break; 840 } 841 return 1; 842 } 843 844 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw) 845 { 846 if (rw == 1) { 847 /* write */ 848 again: 849 wait_event(dev_replace->read_lock_wq, 850 atomic_read(&dev_replace->blocking_readers) == 0); 851 write_lock(&dev_replace->lock); 852 if (atomic_read(&dev_replace->blocking_readers)) { 853 write_unlock(&dev_replace->lock); 854 goto again; 855 } 856 } else { 857 read_lock(&dev_replace->lock); 858 atomic_inc(&dev_replace->read_locks); 859 } 860 } 861 862 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw) 863 { 864 if (rw == 1) { 865 /* write */ 866 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0); 867 write_unlock(&dev_replace->lock); 868 } else { 869 ASSERT(atomic_read(&dev_replace->read_locks) > 0); 870 atomic_dec(&dev_replace->read_locks); 871 read_unlock(&dev_replace->lock); 872 } 873 } 874 875 /* inc blocking cnt and release read lock */ 876 void btrfs_dev_replace_set_lock_blocking( 877 struct btrfs_dev_replace *dev_replace) 878 { 879 /* only set blocking for read lock */ 880 ASSERT(atomic_read(&dev_replace->read_locks) > 0); 881 atomic_inc(&dev_replace->blocking_readers); 882 read_unlock(&dev_replace->lock); 883 } 884 885 /* acquire read lock and dec blocking cnt */ 886 void btrfs_dev_replace_clear_lock_blocking( 887 struct btrfs_dev_replace *dev_replace) 888 { 889 /* only set blocking for read lock */ 890 ASSERT(atomic_read(&dev_replace->read_locks) > 0); 891 ASSERT(atomic_read(&dev_replace->blocking_readers) > 0); 892 read_lock(&dev_replace->lock); 893 if (atomic_dec_and_test(&dev_replace->blocking_readers) && 894 waitqueue_active(&dev_replace->read_lock_wq)) 895 wake_up(&dev_replace->read_lock_wq); 896 } 897 898 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) 899 { 900 percpu_counter_inc(&fs_info->bio_counter); 901 } 902 903 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount) 904 { 905 percpu_counter_sub(&fs_info->bio_counter, amount); 906 907 if (waitqueue_active(&fs_info->replace_wait)) 908 wake_up(&fs_info->replace_wait); 909 } 910 911 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) 912 { 913 while (1) { 914 percpu_counter_inc(&fs_info->bio_counter); 915 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING, 916 &fs_info->fs_state))) 917 break; 918 919 btrfs_bio_counter_dec(fs_info); 920 wait_event(fs_info->replace_wait, 921 !test_bit(BTRFS_FS_STATE_DEV_REPLACING, 922 &fs_info->fs_state)); 923 } 924 } 925