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 int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, 48 char *srcdev_name, 49 struct btrfs_device **device); 50 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info); 51 static int btrfs_dev_replace_kthread(void *data); 52 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info); 53 54 55 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) 56 { 57 struct btrfs_key key; 58 struct btrfs_root *dev_root = fs_info->dev_root; 59 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 60 struct extent_buffer *eb; 61 int slot; 62 int ret = 0; 63 struct btrfs_path *path = NULL; 64 int item_size; 65 struct btrfs_dev_replace_item *ptr; 66 u64 src_devid; 67 68 path = btrfs_alloc_path(); 69 if (!path) { 70 ret = -ENOMEM; 71 goto out; 72 } 73 74 key.objectid = 0; 75 key.type = BTRFS_DEV_REPLACE_KEY; 76 key.offset = 0; 77 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); 78 if (ret) { 79 no_valid_dev_replace_entry_found: 80 ret = 0; 81 dev_replace->replace_state = 82 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED; 83 dev_replace->cont_reading_from_srcdev_mode = 84 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; 85 dev_replace->replace_state = 0; 86 dev_replace->time_started = 0; 87 dev_replace->time_stopped = 0; 88 atomic64_set(&dev_replace->num_write_errors, 0); 89 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 90 dev_replace->cursor_left = 0; 91 dev_replace->committed_cursor_left = 0; 92 dev_replace->cursor_left_last_write_of_item = 0; 93 dev_replace->cursor_right = 0; 94 dev_replace->srcdev = NULL; 95 dev_replace->tgtdev = NULL; 96 dev_replace->is_valid = 0; 97 dev_replace->item_needs_writeback = 0; 98 goto out; 99 } 100 slot = path->slots[0]; 101 eb = path->nodes[0]; 102 item_size = btrfs_item_size_nr(eb, slot); 103 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); 104 105 if (item_size != sizeof(struct btrfs_dev_replace_item)) { 106 btrfs_warn(fs_info, 107 "dev_replace entry found has unexpected size, ignore entry"); 108 goto no_valid_dev_replace_entry_found; 109 } 110 111 src_devid = btrfs_dev_replace_src_devid(eb, ptr); 112 dev_replace->cont_reading_from_srcdev_mode = 113 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); 114 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); 115 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); 116 dev_replace->time_stopped = 117 btrfs_dev_replace_time_stopped(eb, ptr); 118 atomic64_set(&dev_replace->num_write_errors, 119 btrfs_dev_replace_num_write_errors(eb, ptr)); 120 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 121 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); 122 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); 123 dev_replace->committed_cursor_left = dev_replace->cursor_left; 124 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; 125 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); 126 dev_replace->is_valid = 1; 127 128 dev_replace->item_needs_writeback = 0; 129 switch (dev_replace->replace_state) { 130 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 131 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 132 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 133 dev_replace->srcdev = NULL; 134 dev_replace->tgtdev = NULL; 135 break; 136 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 137 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 138 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid, 139 NULL, NULL); 140 dev_replace->tgtdev = btrfs_find_device(fs_info, 141 BTRFS_DEV_REPLACE_DEVID, 142 NULL, NULL); 143 /* 144 * allow 'btrfs dev replace_cancel' if src/tgt device is 145 * missing 146 */ 147 if (!dev_replace->srcdev && 148 !btrfs_test_opt(dev_root, DEGRADED)) { 149 ret = -EIO; 150 btrfs_warn(fs_info, 151 "cannot mount because device replace operation is ongoing and"); 152 btrfs_warn(fs_info, 153 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 154 src_devid); 155 } 156 if (!dev_replace->tgtdev && 157 !btrfs_test_opt(dev_root, DEGRADED)) { 158 ret = -EIO; 159 btrfs_warn(fs_info, 160 "cannot mount because device replace operation is ongoing and"); 161 btrfs_warn(fs_info, 162 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", 163 BTRFS_DEV_REPLACE_DEVID); 164 } 165 if (dev_replace->tgtdev) { 166 if (dev_replace->srcdev) { 167 dev_replace->tgtdev->total_bytes = 168 dev_replace->srcdev->total_bytes; 169 dev_replace->tgtdev->disk_total_bytes = 170 dev_replace->srcdev->disk_total_bytes; 171 dev_replace->tgtdev->bytes_used = 172 dev_replace->srcdev->bytes_used; 173 } 174 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1; 175 btrfs_init_dev_replace_tgtdev_for_resume(fs_info, 176 dev_replace->tgtdev); 177 } 178 break; 179 } 180 181 out: 182 if (path) 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); 203 if (!dev_replace->is_valid || 204 !dev_replace->item_needs_writeback) { 205 btrfs_dev_replace_unlock(dev_replace); 206 return 0; 207 } 208 btrfs_dev_replace_unlock(dev_replace); 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); 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); 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 if (btrfs_fs_incompat(fs_info, RAID56)) { 316 btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6"); 317 return -EOPNOTSUPP; 318 } 319 320 switch (args->start.cont_reading_from_srcdev_mode) { 321 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 322 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 323 break; 324 default: 325 return -EINVAL; 326 } 327 328 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 329 args->start.tgtdev_name[0] == '\0') 330 return -EINVAL; 331 332 mutex_lock(&fs_info->volume_mutex); 333 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid, 334 args->start.srcdev_name, 335 &src_device); 336 if (ret) { 337 mutex_unlock(&fs_info->volume_mutex); 338 return ret; 339 } 340 341 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name, 342 src_device, &tgt_device); 343 mutex_unlock(&fs_info->volume_mutex); 344 if (ret) 345 return ret; 346 347 btrfs_dev_replace_lock(dev_replace); 348 switch (dev_replace->replace_state) { 349 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 350 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 351 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 352 break; 353 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 354 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 355 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 356 goto leave; 357 } 358 359 dev_replace->cont_reading_from_srcdev_mode = 360 args->start.cont_reading_from_srcdev_mode; 361 WARN_ON(!src_device); 362 dev_replace->srcdev = src_device; 363 WARN_ON(!tgt_device); 364 dev_replace->tgtdev = tgt_device; 365 366 printk_in_rcu(KERN_INFO 367 "BTRFS: dev_replace from %s (devid %llu) to %s started\n", 368 src_device->missing ? "<missing disk>" : 369 rcu_str_deref(src_device->name), 370 src_device->devid, 371 rcu_str_deref(tgt_device->name)); 372 373 /* 374 * from now on, the writes to the srcdev are all duplicated to 375 * go to the tgtdev as well (refer to btrfs_map_block()). 376 */ 377 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 378 dev_replace->time_started = get_seconds(); 379 dev_replace->cursor_left = 0; 380 dev_replace->committed_cursor_left = 0; 381 dev_replace->cursor_left_last_write_of_item = 0; 382 dev_replace->cursor_right = 0; 383 dev_replace->is_valid = 1; 384 dev_replace->item_needs_writeback = 1; 385 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 386 btrfs_dev_replace_unlock(dev_replace); 387 388 btrfs_wait_ordered_roots(root->fs_info, -1); 389 390 /* force writing the updated state information to disk */ 391 trans = btrfs_start_transaction(root, 0); 392 if (IS_ERR(trans)) { 393 ret = PTR_ERR(trans); 394 btrfs_dev_replace_lock(dev_replace); 395 goto leave; 396 } 397 398 ret = btrfs_commit_transaction(trans, root); 399 WARN_ON(ret); 400 401 /* the disk copy procedure reuses the scrub code */ 402 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 403 src_device->total_bytes, 404 &dev_replace->scrub_progress, 0, 1); 405 406 ret = btrfs_dev_replace_finishing(root->fs_info, ret); 407 WARN_ON(ret); 408 409 return 0; 410 411 leave: 412 dev_replace->srcdev = NULL; 413 dev_replace->tgtdev = NULL; 414 btrfs_dev_replace_unlock(dev_replace); 415 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 416 return ret; 417 } 418 419 /* 420 * blocked until all flighting bios are finished. 421 */ 422 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) 423 { 424 s64 writers; 425 DEFINE_WAIT(wait); 426 427 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 428 do { 429 prepare_to_wait(&fs_info->replace_wait, &wait, 430 TASK_UNINTERRUPTIBLE); 431 writers = percpu_counter_sum(&fs_info->bio_counter); 432 if (writers) 433 schedule(); 434 finish_wait(&fs_info->replace_wait, &wait); 435 } while (writers); 436 } 437 438 /* 439 * we have removed target device, it is safe to allow new bios request. 440 */ 441 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) 442 { 443 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 444 if (waitqueue_active(&fs_info->replace_wait)) 445 wake_up(&fs_info->replace_wait); 446 } 447 448 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 449 int scrub_ret) 450 { 451 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 452 struct btrfs_device *tgt_device; 453 struct btrfs_device *src_device; 454 struct btrfs_root *root = fs_info->tree_root; 455 u8 uuid_tmp[BTRFS_UUID_SIZE]; 456 struct btrfs_trans_handle *trans; 457 int ret = 0; 458 459 /* don't allow cancel or unmount to disturb the finishing procedure */ 460 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 461 462 btrfs_dev_replace_lock(dev_replace); 463 /* was the operation canceled, or is it finished? */ 464 if (dev_replace->replace_state != 465 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 466 btrfs_dev_replace_unlock(dev_replace); 467 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 468 return 0; 469 } 470 471 tgt_device = dev_replace->tgtdev; 472 src_device = dev_replace->srcdev; 473 btrfs_dev_replace_unlock(dev_replace); 474 475 /* 476 * flush all outstanding I/O and inode extent mappings before the 477 * copy operation is declared as being finished 478 */ 479 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1); 480 if (ret) { 481 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 482 return ret; 483 } 484 btrfs_wait_ordered_roots(root->fs_info, -1); 485 486 trans = btrfs_start_transaction(root, 0); 487 if (IS_ERR(trans)) { 488 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 489 return PTR_ERR(trans); 490 } 491 ret = btrfs_commit_transaction(trans, root); 492 WARN_ON(ret); 493 494 /* keep away write_all_supers() during the finishing procedure */ 495 mutex_lock(&root->fs_info->chunk_mutex); 496 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 497 btrfs_dev_replace_lock(dev_replace); 498 dev_replace->replace_state = 499 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 500 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 501 dev_replace->tgtdev = NULL; 502 dev_replace->srcdev = NULL; 503 dev_replace->time_stopped = get_seconds(); 504 dev_replace->item_needs_writeback = 1; 505 506 /* replace old device with new one in mapping tree */ 507 if (!scrub_ret) { 508 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 509 src_device, 510 tgt_device); 511 } else { 512 printk_in_rcu(KERN_ERR 513 "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n", 514 src_device->missing ? "<missing disk>" : 515 rcu_str_deref(src_device->name), 516 src_device->devid, 517 rcu_str_deref(tgt_device->name), scrub_ret); 518 btrfs_dev_replace_unlock(dev_replace); 519 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 520 mutex_unlock(&root->fs_info->chunk_mutex); 521 if (tgt_device) 522 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 523 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 524 525 return 0; 526 } 527 528 printk_in_rcu(KERN_INFO 529 "BTRFS: dev_replace from %s (devid %llu) to %s finished\n", 530 src_device->missing ? "<missing disk>" : 531 rcu_str_deref(src_device->name), 532 src_device->devid, 533 rcu_str_deref(tgt_device->name)); 534 tgt_device->is_tgtdev_for_dev_replace = 0; 535 tgt_device->devid = src_device->devid; 536 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 537 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 538 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 539 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 540 tgt_device->total_bytes = src_device->total_bytes; 541 tgt_device->disk_total_bytes = src_device->disk_total_bytes; 542 tgt_device->bytes_used = src_device->bytes_used; 543 if (fs_info->sb->s_bdev == src_device->bdev) 544 fs_info->sb->s_bdev = tgt_device->bdev; 545 if (fs_info->fs_devices->latest_bdev == src_device->bdev) 546 fs_info->fs_devices->latest_bdev = tgt_device->bdev; 547 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); 548 if (src_device->fs_devices->seeding) 549 fs_info->fs_devices->rw_devices++; 550 551 /* replace the sysfs entry */ 552 btrfs_kobj_rm_device(fs_info, src_device); 553 btrfs_kobj_add_device(fs_info, tgt_device); 554 555 btrfs_dev_replace_unlock(dev_replace); 556 557 btrfs_rm_dev_replace_blocked(fs_info); 558 559 btrfs_rm_dev_replace_srcdev(fs_info, src_device); 560 561 btrfs_rm_dev_replace_unblocked(fs_info); 562 563 /* 564 * this is again a consistent state where no dev_replace procedure 565 * is running, the target device is part of the filesystem, the 566 * source device is not part of the filesystem anymore and its 1st 567 * superblock is scratched out so that it is no longer marked to 568 * belong to this filesystem. 569 */ 570 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 571 mutex_unlock(&root->fs_info->chunk_mutex); 572 573 /* write back the superblocks */ 574 trans = btrfs_start_transaction(root, 0); 575 if (!IS_ERR(trans)) 576 btrfs_commit_transaction(trans, root); 577 578 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 579 580 return 0; 581 } 582 583 static void btrfs_dev_replace_update_device_in_mapping_tree( 584 struct btrfs_fs_info *fs_info, 585 struct btrfs_device *srcdev, 586 struct btrfs_device *tgtdev) 587 { 588 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; 589 struct extent_map *em; 590 struct map_lookup *map; 591 u64 start = 0; 592 int i; 593 594 write_lock(&em_tree->lock); 595 do { 596 em = lookup_extent_mapping(em_tree, start, (u64)-1); 597 if (!em) 598 break; 599 map = (struct map_lookup *)em->bdev; 600 for (i = 0; i < map->num_stripes; i++) 601 if (srcdev == map->stripes[i].dev) 602 map->stripes[i].dev = tgtdev; 603 start = em->start + em->len; 604 free_extent_map(em); 605 } while (start); 606 write_unlock(&em_tree->lock); 607 } 608 609 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, 610 char *srcdev_name, 611 struct btrfs_device **device) 612 { 613 int ret; 614 615 if (srcdevid) { 616 ret = 0; 617 *device = btrfs_find_device(root->fs_info, srcdevid, NULL, 618 NULL); 619 if (!*device) 620 ret = -ENOENT; 621 } else { 622 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name, 623 device); 624 } 625 return ret; 626 } 627 628 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 629 struct btrfs_ioctl_dev_replace_args *args) 630 { 631 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 632 633 btrfs_dev_replace_lock(dev_replace); 634 /* even if !dev_replace_is_valid, the values are good enough for 635 * the replace_status ioctl */ 636 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 637 args->status.replace_state = dev_replace->replace_state; 638 args->status.time_started = dev_replace->time_started; 639 args->status.time_stopped = dev_replace->time_stopped; 640 args->status.num_write_errors = 641 atomic64_read(&dev_replace->num_write_errors); 642 args->status.num_uncorrectable_read_errors = 643 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 644 switch (dev_replace->replace_state) { 645 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 646 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 647 args->status.progress_1000 = 0; 648 break; 649 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 650 args->status.progress_1000 = 1000; 651 break; 652 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 653 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 654 args->status.progress_1000 = div64_u64(dev_replace->cursor_left, 655 div64_u64(dev_replace->srcdev->total_bytes, 1000)); 656 break; 657 } 658 btrfs_dev_replace_unlock(dev_replace); 659 } 660 661 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info, 662 struct btrfs_ioctl_dev_replace_args *args) 663 { 664 args->result = __btrfs_dev_replace_cancel(fs_info); 665 return 0; 666 } 667 668 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 669 { 670 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 671 struct btrfs_device *tgt_device = NULL; 672 struct btrfs_trans_handle *trans; 673 struct btrfs_root *root = fs_info->tree_root; 674 u64 result; 675 int ret; 676 677 if (fs_info->sb->s_flags & MS_RDONLY) 678 return -EROFS; 679 680 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 681 btrfs_dev_replace_lock(dev_replace); 682 switch (dev_replace->replace_state) { 683 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 684 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 685 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 686 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 687 btrfs_dev_replace_unlock(dev_replace); 688 goto leave; 689 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 690 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 691 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 692 tgt_device = dev_replace->tgtdev; 693 dev_replace->tgtdev = NULL; 694 dev_replace->srcdev = NULL; 695 break; 696 } 697 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 698 dev_replace->time_stopped = get_seconds(); 699 dev_replace->item_needs_writeback = 1; 700 btrfs_dev_replace_unlock(dev_replace); 701 btrfs_scrub_cancel(fs_info); 702 703 trans = btrfs_start_transaction(root, 0); 704 if (IS_ERR(trans)) { 705 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 706 return PTR_ERR(trans); 707 } 708 ret = btrfs_commit_transaction(trans, root); 709 WARN_ON(ret); 710 if (tgt_device) 711 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 712 713 leave: 714 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 715 return result; 716 } 717 718 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 719 { 720 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 721 722 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 723 btrfs_dev_replace_lock(dev_replace); 724 switch (dev_replace->replace_state) { 725 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 726 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 727 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 728 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 729 break; 730 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 731 dev_replace->replace_state = 732 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 733 dev_replace->time_stopped = get_seconds(); 734 dev_replace->item_needs_writeback = 1; 735 btrfs_info(fs_info, "suspending dev_replace for unmount"); 736 break; 737 } 738 739 btrfs_dev_replace_unlock(dev_replace); 740 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 741 } 742 743 /* resume dev_replace procedure that was interrupted by unmount */ 744 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 745 { 746 struct task_struct *task; 747 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 748 749 btrfs_dev_replace_lock(dev_replace); 750 switch (dev_replace->replace_state) { 751 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 752 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 753 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 754 btrfs_dev_replace_unlock(dev_replace); 755 return 0; 756 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 757 break; 758 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 759 dev_replace->replace_state = 760 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 761 break; 762 } 763 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 764 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing"); 765 btrfs_info(fs_info, 766 "you may cancel the operation after 'mount -o degraded'"); 767 btrfs_dev_replace_unlock(dev_replace); 768 return 0; 769 } 770 btrfs_dev_replace_unlock(dev_replace); 771 772 WARN_ON(atomic_xchg( 773 &fs_info->mutually_exclusive_operation_running, 1)); 774 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 775 return PTR_ERR_OR_ZERO(task); 776 } 777 778 static int btrfs_dev_replace_kthread(void *data) 779 { 780 struct btrfs_fs_info *fs_info = data; 781 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 782 struct btrfs_ioctl_dev_replace_args *status_args; 783 u64 progress; 784 785 status_args = kzalloc(sizeof(*status_args), GFP_NOFS); 786 if (status_args) { 787 btrfs_dev_replace_status(fs_info, status_args); 788 progress = status_args->status.progress_1000; 789 kfree(status_args); 790 do_div(progress, 10); 791 printk_in_rcu(KERN_INFO 792 "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n", 793 dev_replace->srcdev->missing ? "<missing disk>" : 794 rcu_str_deref(dev_replace->srcdev->name), 795 dev_replace->srcdev->devid, 796 dev_replace->tgtdev ? 797 rcu_str_deref(dev_replace->tgtdev->name) : 798 "<missing target disk>", 799 (unsigned int)progress); 800 } 801 btrfs_dev_replace_continue_on_mount(fs_info); 802 atomic_set(&fs_info->mutually_exclusive_operation_running, 0); 803 804 return 0; 805 } 806 807 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info) 808 { 809 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 810 int ret; 811 812 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 813 dev_replace->committed_cursor_left, 814 dev_replace->srcdev->total_bytes, 815 &dev_replace->scrub_progress, 0, 1); 816 ret = btrfs_dev_replace_finishing(fs_info, ret); 817 WARN_ON(ret); 818 return 0; 819 } 820 821 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 822 { 823 if (!dev_replace->is_valid) 824 return 0; 825 826 switch (dev_replace->replace_state) { 827 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 828 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 829 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 830 return 0; 831 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 832 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 833 /* 834 * return true even if tgtdev is missing (this is 835 * something that can happen if the dev_replace 836 * procedure is suspended by an umount and then 837 * the tgtdev is missing (or "btrfs dev scan") was 838 * not called and the the filesystem is remounted 839 * in degraded state. This does not stop the 840 * dev_replace procedure. It needs to be canceled 841 * manually if the cancelation is wanted. 842 */ 843 break; 844 } 845 return 1; 846 } 847 848 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace) 849 { 850 /* the beginning is just an optimization for the typical case */ 851 if (atomic_read(&dev_replace->nesting_level) == 0) { 852 acquire_lock: 853 /* this is not a nested case where the same thread 854 * is trying to acqurire the same lock twice */ 855 mutex_lock(&dev_replace->lock); 856 mutex_lock(&dev_replace->lock_management_lock); 857 dev_replace->lock_owner = current->pid; 858 atomic_inc(&dev_replace->nesting_level); 859 mutex_unlock(&dev_replace->lock_management_lock); 860 return; 861 } 862 863 mutex_lock(&dev_replace->lock_management_lock); 864 if (atomic_read(&dev_replace->nesting_level) > 0 && 865 dev_replace->lock_owner == current->pid) { 866 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 867 atomic_inc(&dev_replace->nesting_level); 868 mutex_unlock(&dev_replace->lock_management_lock); 869 return; 870 } 871 872 mutex_unlock(&dev_replace->lock_management_lock); 873 goto acquire_lock; 874 } 875 876 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace) 877 { 878 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 879 mutex_lock(&dev_replace->lock_management_lock); 880 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1); 881 WARN_ON(dev_replace->lock_owner != current->pid); 882 atomic_dec(&dev_replace->nesting_level); 883 if (atomic_read(&dev_replace->nesting_level) == 0) { 884 dev_replace->lock_owner = 0; 885 mutex_unlock(&dev_replace->lock_management_lock); 886 mutex_unlock(&dev_replace->lock); 887 } else { 888 mutex_unlock(&dev_replace->lock_management_lock); 889 } 890 } 891 892 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) 893 { 894 percpu_counter_inc(&fs_info->bio_counter); 895 } 896 897 void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info) 898 { 899 percpu_counter_dec(&fs_info->bio_counter); 900 901 if (waitqueue_active(&fs_info->replace_wait)) 902 wake_up(&fs_info->replace_wait); 903 } 904 905 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) 906 { 907 DEFINE_WAIT(wait); 908 again: 909 percpu_counter_inc(&fs_info->bio_counter); 910 if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) { 911 btrfs_bio_counter_dec(fs_info); 912 wait_event(fs_info->replace_wait, 913 !test_bit(BTRFS_FS_STATE_DEV_REPLACING, 914 &fs_info->fs_state)); 915 goto again; 916 } 917 918 } 919