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 "compat.h" 30 #include "ctree.h" 31 #include "extent_map.h" 32 #include "disk-io.h" 33 #include "transaction.h" 34 #include "print-tree.h" 35 #include "volumes.h" 36 #include "async-thread.h" 37 #include "check-integrity.h" 38 #include "rcu-string.h" 39 #include "dev-replace.h" 40 41 static u64 btrfs_get_seconds_since_1970(void); 42 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 43 int scrub_ret); 44 static void btrfs_dev_replace_update_device_in_mapping_tree( 45 struct btrfs_fs_info *fs_info, 46 struct btrfs_device *srcdev, 47 struct btrfs_device *tgtdev); 48 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, 49 char *srcdev_name, 50 struct btrfs_device **device); 51 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info); 52 static int btrfs_dev_replace_kthread(void *data); 53 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info); 54 55 56 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) 57 { 58 struct btrfs_key key; 59 struct btrfs_root *dev_root = fs_info->dev_root; 60 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 61 struct extent_buffer *eb; 62 int slot; 63 int ret = 0; 64 struct btrfs_path *path = NULL; 65 int item_size; 66 struct btrfs_dev_replace_item *ptr; 67 u64 src_devid; 68 69 path = btrfs_alloc_path(); 70 if (!path) { 71 ret = -ENOMEM; 72 goto out; 73 } 74 75 key.objectid = 0; 76 key.type = BTRFS_DEV_REPLACE_KEY; 77 key.offset = 0; 78 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); 79 if (ret) { 80 no_valid_dev_replace_entry_found: 81 ret = 0; 82 dev_replace->replace_state = 83 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED; 84 dev_replace->cont_reading_from_srcdev_mode = 85 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; 86 dev_replace->replace_state = 0; 87 dev_replace->time_started = 0; 88 dev_replace->time_stopped = 0; 89 atomic64_set(&dev_replace->num_write_errors, 0); 90 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); 91 dev_replace->cursor_left = 0; 92 dev_replace->committed_cursor_left = 0; 93 dev_replace->cursor_left_last_write_of_item = 0; 94 dev_replace->cursor_right = 0; 95 dev_replace->srcdev = NULL; 96 dev_replace->tgtdev = NULL; 97 dev_replace->is_valid = 0; 98 dev_replace->item_needs_writeback = 0; 99 goto out; 100 } 101 slot = path->slots[0]; 102 eb = path->nodes[0]; 103 item_size = btrfs_item_size_nr(eb, slot); 104 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); 105 106 if (item_size != sizeof(struct btrfs_dev_replace_item)) { 107 pr_warn("btrfs: dev_replace entry found has unexpected size, ignore entry\n"); 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 pr_warn("btrfs: cannot mount because device replace operation is ongoing and\n" "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?\n", 151 (unsigned long long)src_devid); 152 } 153 if (!dev_replace->tgtdev && 154 !btrfs_test_opt(dev_root, DEGRADED)) { 155 ret = -EIO; 156 pr_warn("btrfs: cannot mount because device replace operation is ongoing and\n" "tgtdev (devid %llu) is missing, need to run btrfs dev scan?\n", 157 (unsigned long long)BTRFS_DEV_REPLACE_DEVID); 158 } 159 if (dev_replace->tgtdev) { 160 if (dev_replace->srcdev) { 161 dev_replace->tgtdev->total_bytes = 162 dev_replace->srcdev->total_bytes; 163 dev_replace->tgtdev->disk_total_bytes = 164 dev_replace->srcdev->disk_total_bytes; 165 dev_replace->tgtdev->bytes_used = 166 dev_replace->srcdev->bytes_used; 167 } 168 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1; 169 btrfs_init_dev_replace_tgtdev_for_resume(fs_info, 170 dev_replace->tgtdev); 171 } 172 break; 173 } 174 175 out: 176 if (path) 177 btrfs_free_path(path); 178 return ret; 179 } 180 181 /* 182 * called from commit_transaction. Writes changed device replace state to 183 * disk. 184 */ 185 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans, 186 struct btrfs_fs_info *fs_info) 187 { 188 int ret; 189 struct btrfs_root *dev_root = fs_info->dev_root; 190 struct btrfs_path *path; 191 struct btrfs_key key; 192 struct extent_buffer *eb; 193 struct btrfs_dev_replace_item *ptr; 194 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 195 196 btrfs_dev_replace_lock(dev_replace); 197 if (!dev_replace->is_valid || 198 !dev_replace->item_needs_writeback) { 199 btrfs_dev_replace_unlock(dev_replace); 200 return 0; 201 } 202 btrfs_dev_replace_unlock(dev_replace); 203 204 key.objectid = 0; 205 key.type = BTRFS_DEV_REPLACE_KEY; 206 key.offset = 0; 207 208 path = btrfs_alloc_path(); 209 if (!path) { 210 ret = -ENOMEM; 211 goto out; 212 } 213 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 214 if (ret < 0) { 215 pr_warn("btrfs: error %d while searching for dev_replace item!\n", 216 ret); 217 goto out; 218 } 219 220 if (ret == 0 && 221 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 222 /* 223 * need to delete old one and insert a new one. 224 * Since no attempt is made to recover any old state, if the 225 * dev_replace state is 'running', the data on the target 226 * drive is lost. 227 * It would be possible to recover the state: just make sure 228 * that the beginning of the item is never changed and always 229 * contains all the essential information. Then read this 230 * minimal set of information and use it as a base for the 231 * new state. 232 */ 233 ret = btrfs_del_item(trans, dev_root, path); 234 if (ret != 0) { 235 pr_warn("btrfs: delete too small dev_replace item failed %d!\n", 236 ret); 237 goto out; 238 } 239 ret = 1; 240 } 241 242 if (ret == 1) { 243 /* need to insert a new item */ 244 btrfs_release_path(path); 245 ret = btrfs_insert_empty_item(trans, dev_root, path, 246 &key, sizeof(*ptr)); 247 if (ret < 0) { 248 pr_warn("btrfs: insert dev_replace item failed %d!\n", 249 ret); 250 goto out; 251 } 252 } 253 254 eb = path->nodes[0]; 255 ptr = btrfs_item_ptr(eb, path->slots[0], 256 struct btrfs_dev_replace_item); 257 258 btrfs_dev_replace_lock(dev_replace); 259 if (dev_replace->srcdev) 260 btrfs_set_dev_replace_src_devid(eb, ptr, 261 dev_replace->srcdev->devid); 262 else 263 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); 264 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, 265 dev_replace->cont_reading_from_srcdev_mode); 266 btrfs_set_dev_replace_replace_state(eb, ptr, 267 dev_replace->replace_state); 268 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); 269 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); 270 btrfs_set_dev_replace_num_write_errors(eb, ptr, 271 atomic64_read(&dev_replace->num_write_errors)); 272 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, 273 atomic64_read(&dev_replace->num_uncorrectable_read_errors)); 274 dev_replace->cursor_left_last_write_of_item = 275 dev_replace->cursor_left; 276 btrfs_set_dev_replace_cursor_left(eb, ptr, 277 dev_replace->cursor_left_last_write_of_item); 278 btrfs_set_dev_replace_cursor_right(eb, ptr, 279 dev_replace->cursor_right); 280 dev_replace->item_needs_writeback = 0; 281 btrfs_dev_replace_unlock(dev_replace); 282 283 btrfs_mark_buffer_dirty(eb); 284 285 out: 286 btrfs_free_path(path); 287 288 return ret; 289 } 290 291 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info) 292 { 293 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 294 295 dev_replace->committed_cursor_left = 296 dev_replace->cursor_left_last_write_of_item; 297 } 298 299 static u64 btrfs_get_seconds_since_1970(void) 300 { 301 struct timespec t = CURRENT_TIME_SEC; 302 303 return t.tv_sec; 304 } 305 306 int btrfs_dev_replace_start(struct btrfs_root *root, 307 struct btrfs_ioctl_dev_replace_args *args) 308 { 309 struct btrfs_trans_handle *trans; 310 struct btrfs_fs_info *fs_info = root->fs_info; 311 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 312 int ret; 313 struct btrfs_device *tgt_device = NULL; 314 struct btrfs_device *src_device = NULL; 315 316 switch (args->start.cont_reading_from_srcdev_mode) { 317 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 318 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 319 break; 320 default: 321 return -EINVAL; 322 } 323 324 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 325 args->start.tgtdev_name[0] == '\0') 326 return -EINVAL; 327 328 mutex_lock(&fs_info->volume_mutex); 329 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name, 330 &tgt_device); 331 if (ret) { 332 pr_err("btrfs: target device %s is invalid!\n", 333 args->start.tgtdev_name); 334 mutex_unlock(&fs_info->volume_mutex); 335 return -EINVAL; 336 } 337 338 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid, 339 args->start.srcdev_name, 340 &src_device); 341 mutex_unlock(&fs_info->volume_mutex); 342 if (ret) { 343 ret = -EINVAL; 344 goto leave_no_lock; 345 } 346 347 if (tgt_device->total_bytes < src_device->total_bytes) { 348 pr_err("btrfs: target device is smaller than source device!\n"); 349 ret = -EINVAL; 350 goto leave_no_lock; 351 } 352 353 btrfs_dev_replace_lock(dev_replace); 354 switch (dev_replace->replace_state) { 355 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 356 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 357 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 358 break; 359 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 360 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 361 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 362 goto leave; 363 } 364 365 dev_replace->cont_reading_from_srcdev_mode = 366 args->start.cont_reading_from_srcdev_mode; 367 WARN_ON(!src_device); 368 dev_replace->srcdev = src_device; 369 WARN_ON(!tgt_device); 370 dev_replace->tgtdev = tgt_device; 371 372 printk_in_rcu(KERN_INFO 373 "btrfs: dev_replace from %s (devid %llu) to %s) started\n", 374 src_device->missing ? "<missing disk>" : 375 rcu_str_deref(src_device->name), 376 src_device->devid, 377 rcu_str_deref(tgt_device->name)); 378 379 tgt_device->total_bytes = src_device->total_bytes; 380 tgt_device->disk_total_bytes = src_device->disk_total_bytes; 381 tgt_device->bytes_used = src_device->bytes_used; 382 383 /* 384 * from now on, the writes to the srcdev are all duplicated to 385 * go to the tgtdev as well (refer to btrfs_map_block()). 386 */ 387 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 388 dev_replace->time_started = btrfs_get_seconds_since_1970(); 389 dev_replace->cursor_left = 0; 390 dev_replace->committed_cursor_left = 0; 391 dev_replace->cursor_left_last_write_of_item = 0; 392 dev_replace->cursor_right = 0; 393 dev_replace->is_valid = 1; 394 dev_replace->item_needs_writeback = 1; 395 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 396 btrfs_dev_replace_unlock(dev_replace); 397 398 btrfs_wait_ordered_extents(root, 0); 399 400 /* force writing the updated state information to disk */ 401 trans = btrfs_start_transaction(root, 0); 402 if (IS_ERR(trans)) { 403 ret = PTR_ERR(trans); 404 btrfs_dev_replace_lock(dev_replace); 405 goto leave; 406 } 407 408 ret = btrfs_commit_transaction(trans, root); 409 WARN_ON(ret); 410 411 /* the disk copy procedure reuses the scrub code */ 412 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 413 src_device->total_bytes, 414 &dev_replace->scrub_progress, 0, 1); 415 416 ret = btrfs_dev_replace_finishing(root->fs_info, ret); 417 WARN_ON(ret); 418 419 return 0; 420 421 leave: 422 dev_replace->srcdev = NULL; 423 dev_replace->tgtdev = NULL; 424 btrfs_dev_replace_unlock(dev_replace); 425 leave_no_lock: 426 if (tgt_device) 427 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 428 return ret; 429 } 430 431 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 432 int scrub_ret) 433 { 434 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 435 struct btrfs_device *tgt_device; 436 struct btrfs_device *src_device; 437 struct btrfs_root *root = fs_info->tree_root; 438 u8 uuid_tmp[BTRFS_UUID_SIZE]; 439 struct btrfs_trans_handle *trans; 440 int ret = 0; 441 442 /* don't allow cancel or unmount to disturb the finishing procedure */ 443 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 444 445 btrfs_dev_replace_lock(dev_replace); 446 /* was the operation canceled, or is it finished? */ 447 if (dev_replace->replace_state != 448 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 449 btrfs_dev_replace_unlock(dev_replace); 450 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 451 return 0; 452 } 453 454 tgt_device = dev_replace->tgtdev; 455 src_device = dev_replace->srcdev; 456 btrfs_dev_replace_unlock(dev_replace); 457 458 /* replace old device with new one in mapping tree */ 459 if (!scrub_ret) 460 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 461 src_device, 462 tgt_device); 463 464 /* 465 * flush all outstanding I/O and inode extent mappings before the 466 * copy operation is declared as being finished 467 */ 468 ret = btrfs_start_delalloc_inodes(root, 0); 469 if (ret) { 470 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 471 return ret; 472 } 473 btrfs_wait_ordered_extents(root, 0); 474 475 trans = btrfs_start_transaction(root, 0); 476 if (IS_ERR(trans)) { 477 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 478 return PTR_ERR(trans); 479 } 480 ret = btrfs_commit_transaction(trans, root); 481 WARN_ON(ret); 482 483 /* keep away write_all_supers() during the finishing procedure */ 484 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 485 btrfs_dev_replace_lock(dev_replace); 486 dev_replace->replace_state = 487 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 488 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 489 dev_replace->tgtdev = NULL; 490 dev_replace->srcdev = NULL; 491 dev_replace->time_stopped = btrfs_get_seconds_since_1970(); 492 dev_replace->item_needs_writeback = 1; 493 494 if (scrub_ret) { 495 printk_in_rcu(KERN_ERR 496 "btrfs: btrfs_scrub_dev(%s, %llu, %s) failed %d\n", 497 src_device->missing ? "<missing disk>" : 498 rcu_str_deref(src_device->name), 499 src_device->devid, 500 rcu_str_deref(tgt_device->name), scrub_ret); 501 btrfs_dev_replace_unlock(dev_replace); 502 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 503 if (tgt_device) 504 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 505 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 506 507 return 0; 508 } 509 510 printk_in_rcu(KERN_INFO 511 "btrfs: dev_replace from %s (devid %llu) to %s) finished\n", 512 src_device->missing ? "<missing disk>" : 513 rcu_str_deref(src_device->name), 514 src_device->devid, 515 rcu_str_deref(tgt_device->name)); 516 tgt_device->is_tgtdev_for_dev_replace = 0; 517 tgt_device->devid = src_device->devid; 518 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 519 tgt_device->bytes_used = src_device->bytes_used; 520 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 521 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 522 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 523 tgt_device->total_bytes = src_device->total_bytes; 524 tgt_device->disk_total_bytes = src_device->disk_total_bytes; 525 tgt_device->bytes_used = src_device->bytes_used; 526 if (fs_info->sb->s_bdev == src_device->bdev) 527 fs_info->sb->s_bdev = tgt_device->bdev; 528 if (fs_info->fs_devices->latest_bdev == src_device->bdev) 529 fs_info->fs_devices->latest_bdev = tgt_device->bdev; 530 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); 531 532 btrfs_rm_dev_replace_srcdev(fs_info, src_device); 533 if (src_device->bdev) { 534 /* zero out the old super */ 535 btrfs_scratch_superblock(src_device); 536 } 537 /* 538 * this is again a consistent state where no dev_replace procedure 539 * is running, the target device is part of the filesystem, the 540 * source device is not part of the filesystem anymore and its 1st 541 * superblock is scratched out so that it is no longer marked to 542 * belong to this filesystem. 543 */ 544 btrfs_dev_replace_unlock(dev_replace); 545 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 546 547 /* write back the superblocks */ 548 trans = btrfs_start_transaction(root, 0); 549 if (!IS_ERR(trans)) 550 btrfs_commit_transaction(trans, root); 551 552 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 553 554 return 0; 555 } 556 557 static void btrfs_dev_replace_update_device_in_mapping_tree( 558 struct btrfs_fs_info *fs_info, 559 struct btrfs_device *srcdev, 560 struct btrfs_device *tgtdev) 561 { 562 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; 563 struct extent_map *em; 564 struct map_lookup *map; 565 u64 start = 0; 566 int i; 567 568 write_lock(&em_tree->lock); 569 do { 570 em = lookup_extent_mapping(em_tree, start, (u64)-1); 571 if (!em) 572 break; 573 map = (struct map_lookup *)em->bdev; 574 for (i = 0; i < map->num_stripes; i++) 575 if (srcdev == map->stripes[i].dev) 576 map->stripes[i].dev = tgtdev; 577 start = em->start + em->len; 578 free_extent_map(em); 579 } while (start); 580 write_unlock(&em_tree->lock); 581 } 582 583 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, 584 char *srcdev_name, 585 struct btrfs_device **device) 586 { 587 int ret; 588 589 if (srcdevid) { 590 ret = 0; 591 *device = btrfs_find_device(root->fs_info, srcdevid, NULL, 592 NULL); 593 if (!*device) 594 ret = -ENOENT; 595 } else { 596 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name, 597 device); 598 } 599 return ret; 600 } 601 602 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 603 struct btrfs_ioctl_dev_replace_args *args) 604 { 605 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 606 607 btrfs_dev_replace_lock(dev_replace); 608 /* even if !dev_replace_is_valid, the values are good enough for 609 * the replace_status ioctl */ 610 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 611 args->status.replace_state = dev_replace->replace_state; 612 args->status.time_started = dev_replace->time_started; 613 args->status.time_stopped = dev_replace->time_stopped; 614 args->status.num_write_errors = 615 atomic64_read(&dev_replace->num_write_errors); 616 args->status.num_uncorrectable_read_errors = 617 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 618 switch (dev_replace->replace_state) { 619 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 620 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 621 args->status.progress_1000 = 0; 622 break; 623 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 624 args->status.progress_1000 = 1000; 625 break; 626 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 627 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 628 args->status.progress_1000 = div64_u64(dev_replace->cursor_left, 629 div64_u64(dev_replace->srcdev->total_bytes, 1000)); 630 break; 631 } 632 btrfs_dev_replace_unlock(dev_replace); 633 } 634 635 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info, 636 struct btrfs_ioctl_dev_replace_args *args) 637 { 638 args->result = __btrfs_dev_replace_cancel(fs_info); 639 return 0; 640 } 641 642 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 643 { 644 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 645 struct btrfs_device *tgt_device = NULL; 646 struct btrfs_trans_handle *trans; 647 struct btrfs_root *root = fs_info->tree_root; 648 u64 result; 649 int ret; 650 651 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 652 btrfs_dev_replace_lock(dev_replace); 653 switch (dev_replace->replace_state) { 654 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 655 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 656 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 657 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 658 btrfs_dev_replace_unlock(dev_replace); 659 goto leave; 660 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 661 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 662 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 663 tgt_device = dev_replace->tgtdev; 664 dev_replace->tgtdev = NULL; 665 dev_replace->srcdev = NULL; 666 break; 667 } 668 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 669 dev_replace->time_stopped = btrfs_get_seconds_since_1970(); 670 dev_replace->item_needs_writeback = 1; 671 btrfs_dev_replace_unlock(dev_replace); 672 btrfs_scrub_cancel(fs_info); 673 674 trans = btrfs_start_transaction(root, 0); 675 if (IS_ERR(trans)) { 676 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 677 return PTR_ERR(trans); 678 } 679 ret = btrfs_commit_transaction(trans, root); 680 WARN_ON(ret); 681 if (tgt_device) 682 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 683 684 leave: 685 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 686 return result; 687 } 688 689 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 690 { 691 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 692 693 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 694 btrfs_dev_replace_lock(dev_replace); 695 switch (dev_replace->replace_state) { 696 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 697 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 698 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 699 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 700 break; 701 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 702 dev_replace->replace_state = 703 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 704 dev_replace->time_stopped = btrfs_get_seconds_since_1970(); 705 dev_replace->item_needs_writeback = 1; 706 pr_info("btrfs: suspending dev_replace for unmount\n"); 707 break; 708 } 709 710 btrfs_dev_replace_unlock(dev_replace); 711 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 712 } 713 714 /* resume dev_replace procedure that was interrupted by unmount */ 715 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 716 { 717 struct task_struct *task; 718 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 719 720 btrfs_dev_replace_lock(dev_replace); 721 switch (dev_replace->replace_state) { 722 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 723 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 724 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 725 btrfs_dev_replace_unlock(dev_replace); 726 return 0; 727 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 728 break; 729 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 730 dev_replace->replace_state = 731 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 732 break; 733 } 734 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 735 pr_info("btrfs: cannot continue dev_replace, tgtdev is missing\n" 736 "btrfs: you may cancel the operation after 'mount -o degraded'\n"); 737 btrfs_dev_replace_unlock(dev_replace); 738 return 0; 739 } 740 btrfs_dev_replace_unlock(dev_replace); 741 742 WARN_ON(atomic_xchg( 743 &fs_info->mutually_exclusive_operation_running, 1)); 744 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 745 return PTR_RET(task); 746 } 747 748 static int btrfs_dev_replace_kthread(void *data) 749 { 750 struct btrfs_fs_info *fs_info = data; 751 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 752 struct btrfs_ioctl_dev_replace_args *status_args; 753 u64 progress; 754 755 status_args = kzalloc(sizeof(*status_args), GFP_NOFS); 756 if (status_args) { 757 btrfs_dev_replace_status(fs_info, status_args); 758 progress = status_args->status.progress_1000; 759 kfree(status_args); 760 do_div(progress, 10); 761 printk_in_rcu(KERN_INFO 762 "btrfs: continuing dev_replace from %s (devid %llu) to %s @%u%%\n", 763 dev_replace->srcdev->missing ? "<missing disk>" : 764 rcu_str_deref(dev_replace->srcdev->name), 765 dev_replace->srcdev->devid, 766 dev_replace->tgtdev ? 767 rcu_str_deref(dev_replace->tgtdev->name) : 768 "<missing target disk>", 769 (unsigned int)progress); 770 } 771 btrfs_dev_replace_continue_on_mount(fs_info); 772 atomic_set(&fs_info->mutually_exclusive_operation_running, 0); 773 774 return 0; 775 } 776 777 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info) 778 { 779 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 780 int ret; 781 782 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 783 dev_replace->committed_cursor_left, 784 dev_replace->srcdev->total_bytes, 785 &dev_replace->scrub_progress, 0, 1); 786 ret = btrfs_dev_replace_finishing(fs_info, ret); 787 WARN_ON(ret); 788 return 0; 789 } 790 791 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 792 { 793 if (!dev_replace->is_valid) 794 return 0; 795 796 switch (dev_replace->replace_state) { 797 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 798 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 799 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 800 return 0; 801 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 802 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 803 /* 804 * return true even if tgtdev is missing (this is 805 * something that can happen if the dev_replace 806 * procedure is suspended by an umount and then 807 * the tgtdev is missing (or "btrfs dev scan") was 808 * not called and the the filesystem is remounted 809 * in degraded state. This does not stop the 810 * dev_replace procedure. It needs to be canceled 811 * manually if the cancelation is wanted. 812 */ 813 break; 814 } 815 return 1; 816 } 817 818 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace) 819 { 820 /* the beginning is just an optimization for the typical case */ 821 if (atomic_read(&dev_replace->nesting_level) == 0) { 822 acquire_lock: 823 /* this is not a nested case where the same thread 824 * is trying to acqurire the same lock twice */ 825 mutex_lock(&dev_replace->lock); 826 mutex_lock(&dev_replace->lock_management_lock); 827 dev_replace->lock_owner = current->pid; 828 atomic_inc(&dev_replace->nesting_level); 829 mutex_unlock(&dev_replace->lock_management_lock); 830 return; 831 } 832 833 mutex_lock(&dev_replace->lock_management_lock); 834 if (atomic_read(&dev_replace->nesting_level) > 0 && 835 dev_replace->lock_owner == current->pid) { 836 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 837 atomic_inc(&dev_replace->nesting_level); 838 mutex_unlock(&dev_replace->lock_management_lock); 839 return; 840 } 841 842 mutex_unlock(&dev_replace->lock_management_lock); 843 goto acquire_lock; 844 } 845 846 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace) 847 { 848 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 849 mutex_lock(&dev_replace->lock_management_lock); 850 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1); 851 WARN_ON(dev_replace->lock_owner != current->pid); 852 atomic_dec(&dev_replace->nesting_level); 853 if (atomic_read(&dev_replace->nesting_level) == 0) { 854 dev_replace->lock_owner = 0; 855 mutex_unlock(&dev_replace->lock_management_lock); 856 mutex_unlock(&dev_replace->lock); 857 } else { 858 mutex_unlock(&dev_replace->lock_management_lock); 859 } 860 } 861