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->commit_total_bytes = 172 dev_replace->srcdev->commit_total_bytes; 173 dev_replace->tgtdev->bytes_used = 174 dev_replace->srcdev->bytes_used; 175 } 176 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1; 177 btrfs_init_dev_replace_tgtdev_for_resume(fs_info, 178 dev_replace->tgtdev); 179 } 180 break; 181 } 182 183 out: 184 if (path) 185 btrfs_free_path(path); 186 return ret; 187 } 188 189 /* 190 * called from commit_transaction. Writes changed device replace state to 191 * disk. 192 */ 193 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans, 194 struct btrfs_fs_info *fs_info) 195 { 196 int ret; 197 struct btrfs_root *dev_root = fs_info->dev_root; 198 struct btrfs_path *path; 199 struct btrfs_key key; 200 struct extent_buffer *eb; 201 struct btrfs_dev_replace_item *ptr; 202 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 203 204 btrfs_dev_replace_lock(dev_replace); 205 if (!dev_replace->is_valid || 206 !dev_replace->item_needs_writeback) { 207 btrfs_dev_replace_unlock(dev_replace); 208 return 0; 209 } 210 btrfs_dev_replace_unlock(dev_replace); 211 212 key.objectid = 0; 213 key.type = BTRFS_DEV_REPLACE_KEY; 214 key.offset = 0; 215 216 path = btrfs_alloc_path(); 217 if (!path) { 218 ret = -ENOMEM; 219 goto out; 220 } 221 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 222 if (ret < 0) { 223 btrfs_warn(fs_info, "error %d while searching for dev_replace item!", 224 ret); 225 goto out; 226 } 227 228 if (ret == 0 && 229 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 230 /* 231 * need to delete old one and insert a new one. 232 * Since no attempt is made to recover any old state, if the 233 * dev_replace state is 'running', the data on the target 234 * drive is lost. 235 * It would be possible to recover the state: just make sure 236 * that the beginning of the item is never changed and always 237 * contains all the essential information. Then read this 238 * minimal set of information and use it as a base for the 239 * new state. 240 */ 241 ret = btrfs_del_item(trans, dev_root, path); 242 if (ret != 0) { 243 btrfs_warn(fs_info, "delete too small dev_replace item failed %d!", 244 ret); 245 goto out; 246 } 247 ret = 1; 248 } 249 250 if (ret == 1) { 251 /* need to insert a new item */ 252 btrfs_release_path(path); 253 ret = btrfs_insert_empty_item(trans, dev_root, path, 254 &key, sizeof(*ptr)); 255 if (ret < 0) { 256 btrfs_warn(fs_info, "insert dev_replace item failed %d!", 257 ret); 258 goto out; 259 } 260 } 261 262 eb = path->nodes[0]; 263 ptr = btrfs_item_ptr(eb, path->slots[0], 264 struct btrfs_dev_replace_item); 265 266 btrfs_dev_replace_lock(dev_replace); 267 if (dev_replace->srcdev) 268 btrfs_set_dev_replace_src_devid(eb, ptr, 269 dev_replace->srcdev->devid); 270 else 271 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); 272 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, 273 dev_replace->cont_reading_from_srcdev_mode); 274 btrfs_set_dev_replace_replace_state(eb, ptr, 275 dev_replace->replace_state); 276 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); 277 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); 278 btrfs_set_dev_replace_num_write_errors(eb, ptr, 279 atomic64_read(&dev_replace->num_write_errors)); 280 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, 281 atomic64_read(&dev_replace->num_uncorrectable_read_errors)); 282 dev_replace->cursor_left_last_write_of_item = 283 dev_replace->cursor_left; 284 btrfs_set_dev_replace_cursor_left(eb, ptr, 285 dev_replace->cursor_left_last_write_of_item); 286 btrfs_set_dev_replace_cursor_right(eb, ptr, 287 dev_replace->cursor_right); 288 dev_replace->item_needs_writeback = 0; 289 btrfs_dev_replace_unlock(dev_replace); 290 291 btrfs_mark_buffer_dirty(eb); 292 293 out: 294 btrfs_free_path(path); 295 296 return ret; 297 } 298 299 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info) 300 { 301 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 302 303 dev_replace->committed_cursor_left = 304 dev_replace->cursor_left_last_write_of_item; 305 } 306 307 int btrfs_dev_replace_start(struct btrfs_root *root, 308 struct btrfs_ioctl_dev_replace_args *args) 309 { 310 struct btrfs_trans_handle *trans; 311 struct btrfs_fs_info *fs_info = root->fs_info; 312 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 313 int ret; 314 struct btrfs_device *tgt_device = NULL; 315 struct btrfs_device *src_device = NULL; 316 317 if (btrfs_fs_incompat(fs_info, RAID56)) { 318 btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6"); 319 return -EOPNOTSUPP; 320 } 321 322 switch (args->start.cont_reading_from_srcdev_mode) { 323 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 324 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 325 break; 326 default: 327 return -EINVAL; 328 } 329 330 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 331 args->start.tgtdev_name[0] == '\0') 332 return -EINVAL; 333 334 /* 335 * Here we commit the transaction to make sure commit_total_bytes 336 * of all the devices are updated. 337 */ 338 trans = btrfs_attach_transaction(root); 339 if (!IS_ERR(trans)) { 340 ret = btrfs_commit_transaction(trans, root); 341 if (ret) 342 return ret; 343 } else if (PTR_ERR(trans) != -ENOENT) { 344 return PTR_ERR(trans); 345 } 346 347 /* the disk copy procedure reuses the scrub code */ 348 mutex_lock(&fs_info->volume_mutex); 349 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid, 350 args->start.srcdev_name, 351 &src_device); 352 if (ret) { 353 mutex_unlock(&fs_info->volume_mutex); 354 return ret; 355 } 356 357 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name, 358 src_device, &tgt_device); 359 mutex_unlock(&fs_info->volume_mutex); 360 if (ret) 361 return ret; 362 363 btrfs_dev_replace_lock(dev_replace); 364 switch (dev_replace->replace_state) { 365 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 366 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 367 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 368 break; 369 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 370 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 371 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 372 goto leave; 373 } 374 375 dev_replace->cont_reading_from_srcdev_mode = 376 args->start.cont_reading_from_srcdev_mode; 377 WARN_ON(!src_device); 378 dev_replace->srcdev = src_device; 379 WARN_ON(!tgt_device); 380 dev_replace->tgtdev = tgt_device; 381 382 printk_in_rcu(KERN_INFO 383 "BTRFS: dev_replace from %s (devid %llu) to %s started\n", 384 src_device->missing ? "<missing disk>" : 385 rcu_str_deref(src_device->name), 386 src_device->devid, 387 rcu_str_deref(tgt_device->name)); 388 389 /* 390 * from now on, the writes to the srcdev are all duplicated to 391 * go to the tgtdev as well (refer to btrfs_map_block()). 392 */ 393 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 394 dev_replace->time_started = get_seconds(); 395 dev_replace->cursor_left = 0; 396 dev_replace->committed_cursor_left = 0; 397 dev_replace->cursor_left_last_write_of_item = 0; 398 dev_replace->cursor_right = 0; 399 dev_replace->is_valid = 1; 400 dev_replace->item_needs_writeback = 1; 401 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 402 btrfs_dev_replace_unlock(dev_replace); 403 404 btrfs_wait_ordered_roots(root->fs_info, -1); 405 406 /* force writing the updated state information to disk */ 407 trans = btrfs_start_transaction(root, 0); 408 if (IS_ERR(trans)) { 409 ret = PTR_ERR(trans); 410 btrfs_dev_replace_lock(dev_replace); 411 goto leave; 412 } 413 414 ret = btrfs_commit_transaction(trans, root); 415 WARN_ON(ret); 416 417 /* the disk copy procedure reuses the scrub code */ 418 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 419 src_device->total_bytes, 420 &dev_replace->scrub_progress, 0, 1); 421 422 ret = btrfs_dev_replace_finishing(root->fs_info, ret); 423 WARN_ON(ret); 424 425 return 0; 426 427 leave: 428 dev_replace->srcdev = NULL; 429 dev_replace->tgtdev = NULL; 430 btrfs_dev_replace_unlock(dev_replace); 431 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 432 return ret; 433 } 434 435 /* 436 * blocked until all flighting bios are finished. 437 */ 438 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) 439 { 440 s64 writers; 441 DEFINE_WAIT(wait); 442 443 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 444 do { 445 prepare_to_wait(&fs_info->replace_wait, &wait, 446 TASK_UNINTERRUPTIBLE); 447 writers = percpu_counter_sum(&fs_info->bio_counter); 448 if (writers) 449 schedule(); 450 finish_wait(&fs_info->replace_wait, &wait); 451 } while (writers); 452 } 453 454 /* 455 * we have removed target device, it is safe to allow new bios request. 456 */ 457 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) 458 { 459 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); 460 if (waitqueue_active(&fs_info->replace_wait)) 461 wake_up(&fs_info->replace_wait); 462 } 463 464 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 465 int scrub_ret) 466 { 467 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 468 struct btrfs_device *tgt_device; 469 struct btrfs_device *src_device; 470 struct btrfs_root *root = fs_info->tree_root; 471 u8 uuid_tmp[BTRFS_UUID_SIZE]; 472 struct btrfs_trans_handle *trans; 473 int ret = 0; 474 475 /* don't allow cancel or unmount to disturb the finishing procedure */ 476 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 477 478 btrfs_dev_replace_lock(dev_replace); 479 /* was the operation canceled, or is it finished? */ 480 if (dev_replace->replace_state != 481 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 482 btrfs_dev_replace_unlock(dev_replace); 483 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 484 return 0; 485 } 486 487 tgt_device = dev_replace->tgtdev; 488 src_device = dev_replace->srcdev; 489 btrfs_dev_replace_unlock(dev_replace); 490 491 /* 492 * flush all outstanding I/O and inode extent mappings before the 493 * copy operation is declared as being finished 494 */ 495 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1); 496 if (ret) { 497 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 498 return ret; 499 } 500 btrfs_wait_ordered_roots(root->fs_info, -1); 501 502 trans = btrfs_start_transaction(root, 0); 503 if (IS_ERR(trans)) { 504 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 505 return PTR_ERR(trans); 506 } 507 ret = btrfs_commit_transaction(trans, root); 508 WARN_ON(ret); 509 510 /* keep away write_all_supers() during the finishing procedure */ 511 mutex_lock(&root->fs_info->chunk_mutex); 512 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 513 btrfs_dev_replace_lock(dev_replace); 514 dev_replace->replace_state = 515 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 516 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 517 dev_replace->tgtdev = NULL; 518 dev_replace->srcdev = NULL; 519 dev_replace->time_stopped = get_seconds(); 520 dev_replace->item_needs_writeback = 1; 521 522 /* replace old device with new one in mapping tree */ 523 if (!scrub_ret) { 524 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 525 src_device, 526 tgt_device); 527 } else { 528 printk_in_rcu(KERN_ERR 529 "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\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), scrub_ret); 534 btrfs_dev_replace_unlock(dev_replace); 535 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 536 mutex_unlock(&root->fs_info->chunk_mutex); 537 if (tgt_device) 538 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 539 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 540 541 return 0; 542 } 543 544 printk_in_rcu(KERN_INFO 545 "BTRFS: dev_replace from %s (devid %llu) to %s finished\n", 546 src_device->missing ? "<missing disk>" : 547 rcu_str_deref(src_device->name), 548 src_device->devid, 549 rcu_str_deref(tgt_device->name)); 550 tgt_device->is_tgtdev_for_dev_replace = 0; 551 tgt_device->devid = src_device->devid; 552 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 553 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 554 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 555 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 556 tgt_device->total_bytes = src_device->total_bytes; 557 tgt_device->disk_total_bytes = src_device->disk_total_bytes; 558 ASSERT(list_empty(&src_device->resized_list)); 559 tgt_device->commit_total_bytes = src_device->commit_total_bytes; 560 tgt_device->bytes_used = src_device->bytes_used; 561 if (fs_info->sb->s_bdev == src_device->bdev) 562 fs_info->sb->s_bdev = tgt_device->bdev; 563 if (fs_info->fs_devices->latest_bdev == src_device->bdev) 564 fs_info->fs_devices->latest_bdev = tgt_device->bdev; 565 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); 566 if (src_device->fs_devices->seeding) 567 fs_info->fs_devices->rw_devices++; 568 569 /* replace the sysfs entry */ 570 btrfs_kobj_rm_device(fs_info, src_device); 571 btrfs_kobj_add_device(fs_info, tgt_device); 572 573 btrfs_dev_replace_unlock(dev_replace); 574 575 btrfs_rm_dev_replace_blocked(fs_info); 576 577 btrfs_rm_dev_replace_srcdev(fs_info, src_device); 578 579 btrfs_rm_dev_replace_unblocked(fs_info); 580 581 /* 582 * this is again a consistent state where no dev_replace procedure 583 * is running, the target device is part of the filesystem, the 584 * source device is not part of the filesystem anymore and its 1st 585 * superblock is scratched out so that it is no longer marked to 586 * belong to this filesystem. 587 */ 588 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 589 mutex_unlock(&root->fs_info->chunk_mutex); 590 591 /* write back the superblocks */ 592 trans = btrfs_start_transaction(root, 0); 593 if (!IS_ERR(trans)) 594 btrfs_commit_transaction(trans, root); 595 596 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 597 598 return 0; 599 } 600 601 static void btrfs_dev_replace_update_device_in_mapping_tree( 602 struct btrfs_fs_info *fs_info, 603 struct btrfs_device *srcdev, 604 struct btrfs_device *tgtdev) 605 { 606 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; 607 struct extent_map *em; 608 struct map_lookup *map; 609 u64 start = 0; 610 int i; 611 612 write_lock(&em_tree->lock); 613 do { 614 em = lookup_extent_mapping(em_tree, start, (u64)-1); 615 if (!em) 616 break; 617 map = (struct map_lookup *)em->bdev; 618 for (i = 0; i < map->num_stripes; i++) 619 if (srcdev == map->stripes[i].dev) 620 map->stripes[i].dev = tgtdev; 621 start = em->start + em->len; 622 free_extent_map(em); 623 } while (start); 624 write_unlock(&em_tree->lock); 625 } 626 627 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, 628 char *srcdev_name, 629 struct btrfs_device **device) 630 { 631 int ret; 632 633 if (srcdevid) { 634 ret = 0; 635 *device = btrfs_find_device(root->fs_info, srcdevid, NULL, 636 NULL); 637 if (!*device) 638 ret = -ENOENT; 639 } else { 640 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name, 641 device); 642 } 643 return ret; 644 } 645 646 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 647 struct btrfs_ioctl_dev_replace_args *args) 648 { 649 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 650 651 btrfs_dev_replace_lock(dev_replace); 652 /* even if !dev_replace_is_valid, the values are good enough for 653 * the replace_status ioctl */ 654 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 655 args->status.replace_state = dev_replace->replace_state; 656 args->status.time_started = dev_replace->time_started; 657 args->status.time_stopped = dev_replace->time_stopped; 658 args->status.num_write_errors = 659 atomic64_read(&dev_replace->num_write_errors); 660 args->status.num_uncorrectable_read_errors = 661 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 662 switch (dev_replace->replace_state) { 663 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 664 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 665 args->status.progress_1000 = 0; 666 break; 667 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 668 args->status.progress_1000 = 1000; 669 break; 670 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 671 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 672 args->status.progress_1000 = div64_u64(dev_replace->cursor_left, 673 div64_u64(dev_replace->srcdev->total_bytes, 1000)); 674 break; 675 } 676 btrfs_dev_replace_unlock(dev_replace); 677 } 678 679 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info, 680 struct btrfs_ioctl_dev_replace_args *args) 681 { 682 args->result = __btrfs_dev_replace_cancel(fs_info); 683 return 0; 684 } 685 686 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 687 { 688 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 689 struct btrfs_device *tgt_device = NULL; 690 struct btrfs_trans_handle *trans; 691 struct btrfs_root *root = fs_info->tree_root; 692 u64 result; 693 int ret; 694 695 if (fs_info->sb->s_flags & MS_RDONLY) 696 return -EROFS; 697 698 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 699 btrfs_dev_replace_lock(dev_replace); 700 switch (dev_replace->replace_state) { 701 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 702 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 703 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 704 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 705 btrfs_dev_replace_unlock(dev_replace); 706 goto leave; 707 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 708 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 709 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 710 tgt_device = dev_replace->tgtdev; 711 dev_replace->tgtdev = NULL; 712 dev_replace->srcdev = NULL; 713 break; 714 } 715 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 716 dev_replace->time_stopped = get_seconds(); 717 dev_replace->item_needs_writeback = 1; 718 btrfs_dev_replace_unlock(dev_replace); 719 btrfs_scrub_cancel(fs_info); 720 721 trans = btrfs_start_transaction(root, 0); 722 if (IS_ERR(trans)) { 723 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 724 return PTR_ERR(trans); 725 } 726 ret = btrfs_commit_transaction(trans, root); 727 WARN_ON(ret); 728 if (tgt_device) 729 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 730 731 leave: 732 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 733 return result; 734 } 735 736 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 737 { 738 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 739 740 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 741 btrfs_dev_replace_lock(dev_replace); 742 switch (dev_replace->replace_state) { 743 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 744 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 745 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 746 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 747 break; 748 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 749 dev_replace->replace_state = 750 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 751 dev_replace->time_stopped = get_seconds(); 752 dev_replace->item_needs_writeback = 1; 753 btrfs_info(fs_info, "suspending dev_replace for unmount"); 754 break; 755 } 756 757 btrfs_dev_replace_unlock(dev_replace); 758 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 759 } 760 761 /* resume dev_replace procedure that was interrupted by unmount */ 762 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 763 { 764 struct task_struct *task; 765 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 766 767 btrfs_dev_replace_lock(dev_replace); 768 switch (dev_replace->replace_state) { 769 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 770 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 771 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 772 btrfs_dev_replace_unlock(dev_replace); 773 return 0; 774 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 775 break; 776 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 777 dev_replace->replace_state = 778 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 779 break; 780 } 781 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 782 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing"); 783 btrfs_info(fs_info, 784 "you may cancel the operation after 'mount -o degraded'"); 785 btrfs_dev_replace_unlock(dev_replace); 786 return 0; 787 } 788 btrfs_dev_replace_unlock(dev_replace); 789 790 WARN_ON(atomic_xchg( 791 &fs_info->mutually_exclusive_operation_running, 1)); 792 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 793 return PTR_ERR_OR_ZERO(task); 794 } 795 796 static int btrfs_dev_replace_kthread(void *data) 797 { 798 struct btrfs_fs_info *fs_info = data; 799 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 800 struct btrfs_ioctl_dev_replace_args *status_args; 801 u64 progress; 802 803 status_args = kzalloc(sizeof(*status_args), GFP_NOFS); 804 if (status_args) { 805 btrfs_dev_replace_status(fs_info, status_args); 806 progress = status_args->status.progress_1000; 807 kfree(status_args); 808 do_div(progress, 10); 809 printk_in_rcu(KERN_INFO 810 "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n", 811 dev_replace->srcdev->missing ? "<missing disk>" : 812 rcu_str_deref(dev_replace->srcdev->name), 813 dev_replace->srcdev->devid, 814 dev_replace->tgtdev ? 815 rcu_str_deref(dev_replace->tgtdev->name) : 816 "<missing target disk>", 817 (unsigned int)progress); 818 } 819 btrfs_dev_replace_continue_on_mount(fs_info); 820 atomic_set(&fs_info->mutually_exclusive_operation_running, 0); 821 822 return 0; 823 } 824 825 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info) 826 { 827 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 828 int ret; 829 830 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 831 dev_replace->committed_cursor_left, 832 dev_replace->srcdev->total_bytes, 833 &dev_replace->scrub_progress, 0, 1); 834 ret = btrfs_dev_replace_finishing(fs_info, ret); 835 WARN_ON(ret); 836 return 0; 837 } 838 839 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 840 { 841 if (!dev_replace->is_valid) 842 return 0; 843 844 switch (dev_replace->replace_state) { 845 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 846 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 847 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 848 return 0; 849 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 850 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 851 /* 852 * return true even if tgtdev is missing (this is 853 * something that can happen if the dev_replace 854 * procedure is suspended by an umount and then 855 * the tgtdev is missing (or "btrfs dev scan") was 856 * not called and the the filesystem is remounted 857 * in degraded state. This does not stop the 858 * dev_replace procedure. It needs to be canceled 859 * manually if the cancelation is wanted. 860 */ 861 break; 862 } 863 return 1; 864 } 865 866 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace) 867 { 868 /* the beginning is just an optimization for the typical case */ 869 if (atomic_read(&dev_replace->nesting_level) == 0) { 870 acquire_lock: 871 /* this is not a nested case where the same thread 872 * is trying to acqurire the same lock twice */ 873 mutex_lock(&dev_replace->lock); 874 mutex_lock(&dev_replace->lock_management_lock); 875 dev_replace->lock_owner = current->pid; 876 atomic_inc(&dev_replace->nesting_level); 877 mutex_unlock(&dev_replace->lock_management_lock); 878 return; 879 } 880 881 mutex_lock(&dev_replace->lock_management_lock); 882 if (atomic_read(&dev_replace->nesting_level) > 0 && 883 dev_replace->lock_owner == current->pid) { 884 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 885 atomic_inc(&dev_replace->nesting_level); 886 mutex_unlock(&dev_replace->lock_management_lock); 887 return; 888 } 889 890 mutex_unlock(&dev_replace->lock_management_lock); 891 goto acquire_lock; 892 } 893 894 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace) 895 { 896 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 897 mutex_lock(&dev_replace->lock_management_lock); 898 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1); 899 WARN_ON(dev_replace->lock_owner != current->pid); 900 atomic_dec(&dev_replace->nesting_level); 901 if (atomic_read(&dev_replace->nesting_level) == 0) { 902 dev_replace->lock_owner = 0; 903 mutex_unlock(&dev_replace->lock_management_lock); 904 mutex_unlock(&dev_replace->lock); 905 } else { 906 mutex_unlock(&dev_replace->lock_management_lock); 907 } 908 } 909 910 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) 911 { 912 percpu_counter_inc(&fs_info->bio_counter); 913 } 914 915 void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info) 916 { 917 percpu_counter_dec(&fs_info->bio_counter); 918 919 if (waitqueue_active(&fs_info->replace_wait)) 920 wake_up(&fs_info->replace_wait); 921 } 922 923 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) 924 { 925 DEFINE_WAIT(wait); 926 again: 927 percpu_counter_inc(&fs_info->bio_counter); 928 if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) { 929 btrfs_bio_counter_dec(fs_info); 930 wait_event(fs_info->replace_wait, 931 !test_bit(BTRFS_FS_STATE_DEV_REPLACING, 932 &fs_info->fs_state)); 933 goto again; 934 } 935 936 } 937