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 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 pr_warn("btrfs: dev_replace entry found has unexpected size, ignore entry\n"); 107 goto no_valid_dev_replace_entry_found; 108 } 109 110 src_devid = btrfs_dev_replace_src_devid(eb, ptr); 111 dev_replace->cont_reading_from_srcdev_mode = 112 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); 113 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); 114 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); 115 dev_replace->time_stopped = 116 btrfs_dev_replace_time_stopped(eb, ptr); 117 atomic64_set(&dev_replace->num_write_errors, 118 btrfs_dev_replace_num_write_errors(eb, ptr)); 119 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 120 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); 121 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); 122 dev_replace->committed_cursor_left = dev_replace->cursor_left; 123 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; 124 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); 125 dev_replace->is_valid = 1; 126 127 dev_replace->item_needs_writeback = 0; 128 switch (dev_replace->replace_state) { 129 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 130 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 131 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 132 dev_replace->srcdev = NULL; 133 dev_replace->tgtdev = NULL; 134 break; 135 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 136 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 137 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid, 138 NULL, NULL); 139 dev_replace->tgtdev = btrfs_find_device(fs_info, 140 BTRFS_DEV_REPLACE_DEVID, 141 NULL, NULL); 142 /* 143 * allow 'btrfs dev replace_cancel' if src/tgt device is 144 * missing 145 */ 146 if (!dev_replace->srcdev && 147 !btrfs_test_opt(dev_root, DEGRADED)) { 148 ret = -EIO; 149 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", 150 src_devid); 151 } 152 if (!dev_replace->tgtdev && 153 !btrfs_test_opt(dev_root, DEGRADED)) { 154 ret = -EIO; 155 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", 156 BTRFS_DEV_REPLACE_DEVID); 157 } 158 if (dev_replace->tgtdev) { 159 if (dev_replace->srcdev) { 160 dev_replace->tgtdev->total_bytes = 161 dev_replace->srcdev->total_bytes; 162 dev_replace->tgtdev->disk_total_bytes = 163 dev_replace->srcdev->disk_total_bytes; 164 dev_replace->tgtdev->bytes_used = 165 dev_replace->srcdev->bytes_used; 166 } 167 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1; 168 btrfs_init_dev_replace_tgtdev_for_resume(fs_info, 169 dev_replace->tgtdev); 170 } 171 break; 172 } 173 174 out: 175 if (path) 176 btrfs_free_path(path); 177 return ret; 178 } 179 180 /* 181 * called from commit_transaction. Writes changed device replace state to 182 * disk. 183 */ 184 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans, 185 struct btrfs_fs_info *fs_info) 186 { 187 int ret; 188 struct btrfs_root *dev_root = fs_info->dev_root; 189 struct btrfs_path *path; 190 struct btrfs_key key; 191 struct extent_buffer *eb; 192 struct btrfs_dev_replace_item *ptr; 193 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 194 195 btrfs_dev_replace_lock(dev_replace); 196 if (!dev_replace->is_valid || 197 !dev_replace->item_needs_writeback) { 198 btrfs_dev_replace_unlock(dev_replace); 199 return 0; 200 } 201 btrfs_dev_replace_unlock(dev_replace); 202 203 key.objectid = 0; 204 key.type = BTRFS_DEV_REPLACE_KEY; 205 key.offset = 0; 206 207 path = btrfs_alloc_path(); 208 if (!path) { 209 ret = -ENOMEM; 210 goto out; 211 } 212 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 213 if (ret < 0) { 214 pr_warn("btrfs: error %d while searching for dev_replace item!\n", 215 ret); 216 goto out; 217 } 218 219 if (ret == 0 && 220 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 221 /* 222 * need to delete old one and insert a new one. 223 * Since no attempt is made to recover any old state, if the 224 * dev_replace state is 'running', the data on the target 225 * drive is lost. 226 * It would be possible to recover the state: just make sure 227 * that the beginning of the item is never changed and always 228 * contains all the essential information. Then read this 229 * minimal set of information and use it as a base for the 230 * new state. 231 */ 232 ret = btrfs_del_item(trans, dev_root, path); 233 if (ret != 0) { 234 pr_warn("btrfs: delete too small dev_replace item failed %d!\n", 235 ret); 236 goto out; 237 } 238 ret = 1; 239 } 240 241 if (ret == 1) { 242 /* need to insert a new item */ 243 btrfs_release_path(path); 244 ret = btrfs_insert_empty_item(trans, dev_root, path, 245 &key, sizeof(*ptr)); 246 if (ret < 0) { 247 pr_warn("btrfs: insert dev_replace item failed %d!\n", 248 ret); 249 goto out; 250 } 251 } 252 253 eb = path->nodes[0]; 254 ptr = btrfs_item_ptr(eb, path->slots[0], 255 struct btrfs_dev_replace_item); 256 257 btrfs_dev_replace_lock(dev_replace); 258 if (dev_replace->srcdev) 259 btrfs_set_dev_replace_src_devid(eb, ptr, 260 dev_replace->srcdev->devid); 261 else 262 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); 263 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, 264 dev_replace->cont_reading_from_srcdev_mode); 265 btrfs_set_dev_replace_replace_state(eb, ptr, 266 dev_replace->replace_state); 267 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); 268 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); 269 btrfs_set_dev_replace_num_write_errors(eb, ptr, 270 atomic64_read(&dev_replace->num_write_errors)); 271 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, 272 atomic64_read(&dev_replace->num_uncorrectable_read_errors)); 273 dev_replace->cursor_left_last_write_of_item = 274 dev_replace->cursor_left; 275 btrfs_set_dev_replace_cursor_left(eb, ptr, 276 dev_replace->cursor_left_last_write_of_item); 277 btrfs_set_dev_replace_cursor_right(eb, ptr, 278 dev_replace->cursor_right); 279 dev_replace->item_needs_writeback = 0; 280 btrfs_dev_replace_unlock(dev_replace); 281 282 btrfs_mark_buffer_dirty(eb); 283 284 out: 285 btrfs_free_path(path); 286 287 return ret; 288 } 289 290 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info) 291 { 292 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 293 294 dev_replace->committed_cursor_left = 295 dev_replace->cursor_left_last_write_of_item; 296 } 297 298 int btrfs_dev_replace_start(struct btrfs_root *root, 299 struct btrfs_ioctl_dev_replace_args *args) 300 { 301 struct btrfs_trans_handle *trans; 302 struct btrfs_fs_info *fs_info = root->fs_info; 303 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 304 int ret; 305 struct btrfs_device *tgt_device = NULL; 306 struct btrfs_device *src_device = NULL; 307 308 if (btrfs_fs_incompat(fs_info, RAID56)) { 309 pr_warn("btrfs: dev_replace cannot yet handle RAID5/RAID6\n"); 310 return -EINVAL; 311 } 312 313 switch (args->start.cont_reading_from_srcdev_mode) { 314 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: 315 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: 316 break; 317 default: 318 return -EINVAL; 319 } 320 321 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || 322 args->start.tgtdev_name[0] == '\0') 323 return -EINVAL; 324 325 mutex_lock(&fs_info->volume_mutex); 326 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name, 327 &tgt_device); 328 if (ret) { 329 pr_err("btrfs: target device %s is invalid!\n", 330 args->start.tgtdev_name); 331 mutex_unlock(&fs_info->volume_mutex); 332 return -EINVAL; 333 } 334 335 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid, 336 args->start.srcdev_name, 337 &src_device); 338 mutex_unlock(&fs_info->volume_mutex); 339 if (ret) { 340 ret = -EINVAL; 341 goto leave_no_lock; 342 } 343 344 if (tgt_device->total_bytes < src_device->total_bytes) { 345 pr_err("btrfs: target device is smaller than source device!\n"); 346 ret = -EINVAL; 347 goto leave_no_lock; 348 } 349 350 btrfs_dev_replace_lock(dev_replace); 351 switch (dev_replace->replace_state) { 352 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 353 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 354 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 355 break; 356 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 357 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 358 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; 359 goto leave; 360 } 361 362 dev_replace->cont_reading_from_srcdev_mode = 363 args->start.cont_reading_from_srcdev_mode; 364 WARN_ON(!src_device); 365 dev_replace->srcdev = src_device; 366 WARN_ON(!tgt_device); 367 dev_replace->tgtdev = tgt_device; 368 369 printk_in_rcu(KERN_INFO 370 "btrfs: dev_replace from %s (devid %llu) to %s) started\n", 371 src_device->missing ? "<missing disk>" : 372 rcu_str_deref(src_device->name), 373 src_device->devid, 374 rcu_str_deref(tgt_device->name)); 375 376 tgt_device->total_bytes = src_device->total_bytes; 377 tgt_device->disk_total_bytes = src_device->disk_total_bytes; 378 tgt_device->bytes_used = src_device->bytes_used; 379 380 /* 381 * from now on, the writes to the srcdev are all duplicated to 382 * go to the tgtdev as well (refer to btrfs_map_block()). 383 */ 384 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 385 dev_replace->time_started = get_seconds(); 386 dev_replace->cursor_left = 0; 387 dev_replace->committed_cursor_left = 0; 388 dev_replace->cursor_left_last_write_of_item = 0; 389 dev_replace->cursor_right = 0; 390 dev_replace->is_valid = 1; 391 dev_replace->item_needs_writeback = 1; 392 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 393 btrfs_dev_replace_unlock(dev_replace); 394 395 btrfs_wait_all_ordered_extents(root->fs_info); 396 397 /* force writing the updated state information to disk */ 398 trans = btrfs_start_transaction(root, 0); 399 if (IS_ERR(trans)) { 400 ret = PTR_ERR(trans); 401 btrfs_dev_replace_lock(dev_replace); 402 goto leave; 403 } 404 405 ret = btrfs_commit_transaction(trans, root); 406 WARN_ON(ret); 407 408 /* the disk copy procedure reuses the scrub code */ 409 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, 410 src_device->total_bytes, 411 &dev_replace->scrub_progress, 0, 1); 412 413 ret = btrfs_dev_replace_finishing(root->fs_info, ret); 414 WARN_ON(ret); 415 416 return 0; 417 418 leave: 419 dev_replace->srcdev = NULL; 420 dev_replace->tgtdev = NULL; 421 btrfs_dev_replace_unlock(dev_replace); 422 leave_no_lock: 423 if (tgt_device) 424 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 425 return ret; 426 } 427 428 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, 429 int scrub_ret) 430 { 431 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 432 struct btrfs_device *tgt_device; 433 struct btrfs_device *src_device; 434 struct btrfs_root *root = fs_info->tree_root; 435 u8 uuid_tmp[BTRFS_UUID_SIZE]; 436 struct btrfs_trans_handle *trans; 437 int ret = 0; 438 439 /* don't allow cancel or unmount to disturb the finishing procedure */ 440 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 441 442 btrfs_dev_replace_lock(dev_replace); 443 /* was the operation canceled, or is it finished? */ 444 if (dev_replace->replace_state != 445 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { 446 btrfs_dev_replace_unlock(dev_replace); 447 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 448 return 0; 449 } 450 451 tgt_device = dev_replace->tgtdev; 452 src_device = dev_replace->srcdev; 453 btrfs_dev_replace_unlock(dev_replace); 454 455 /* replace old device with new one in mapping tree */ 456 if (!scrub_ret) 457 btrfs_dev_replace_update_device_in_mapping_tree(fs_info, 458 src_device, 459 tgt_device); 460 461 /* 462 * flush all outstanding I/O and inode extent mappings before the 463 * copy operation is declared as being finished 464 */ 465 ret = btrfs_start_all_delalloc_inodes(root->fs_info, 0); 466 if (ret) { 467 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 468 return ret; 469 } 470 btrfs_wait_all_ordered_extents(root->fs_info); 471 472 trans = btrfs_start_transaction(root, 0); 473 if (IS_ERR(trans)) { 474 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 475 return PTR_ERR(trans); 476 } 477 ret = btrfs_commit_transaction(trans, root); 478 WARN_ON(ret); 479 480 /* keep away write_all_supers() during the finishing procedure */ 481 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 482 btrfs_dev_replace_lock(dev_replace); 483 dev_replace->replace_state = 484 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED 485 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; 486 dev_replace->tgtdev = NULL; 487 dev_replace->srcdev = NULL; 488 dev_replace->time_stopped = get_seconds(); 489 dev_replace->item_needs_writeback = 1; 490 491 if (scrub_ret) { 492 printk_in_rcu(KERN_ERR 493 "btrfs: btrfs_scrub_dev(%s, %llu, %s) failed %d\n", 494 src_device->missing ? "<missing disk>" : 495 rcu_str_deref(src_device->name), 496 src_device->devid, 497 rcu_str_deref(tgt_device->name), scrub_ret); 498 btrfs_dev_replace_unlock(dev_replace); 499 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 500 if (tgt_device) 501 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 502 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 503 504 return 0; 505 } 506 507 printk_in_rcu(KERN_INFO 508 "btrfs: dev_replace from %s (devid %llu) to %s) finished\n", 509 src_device->missing ? "<missing disk>" : 510 rcu_str_deref(src_device->name), 511 src_device->devid, 512 rcu_str_deref(tgt_device->name)); 513 tgt_device->is_tgtdev_for_dev_replace = 0; 514 tgt_device->devid = src_device->devid; 515 src_device->devid = BTRFS_DEV_REPLACE_DEVID; 516 tgt_device->bytes_used = src_device->bytes_used; 517 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); 518 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); 519 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); 520 tgt_device->total_bytes = src_device->total_bytes; 521 tgt_device->disk_total_bytes = src_device->disk_total_bytes; 522 tgt_device->bytes_used = src_device->bytes_used; 523 if (fs_info->sb->s_bdev == src_device->bdev) 524 fs_info->sb->s_bdev = tgt_device->bdev; 525 if (fs_info->fs_devices->latest_bdev == src_device->bdev) 526 fs_info->fs_devices->latest_bdev = tgt_device->bdev; 527 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); 528 529 btrfs_rm_dev_replace_srcdev(fs_info, src_device); 530 531 /* 532 * this is again a consistent state where no dev_replace procedure 533 * is running, the target device is part of the filesystem, the 534 * source device is not part of the filesystem anymore and its 1st 535 * superblock is scratched out so that it is no longer marked to 536 * belong to this filesystem. 537 */ 538 btrfs_dev_replace_unlock(dev_replace); 539 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 540 541 /* write back the superblocks */ 542 trans = btrfs_start_transaction(root, 0); 543 if (!IS_ERR(trans)) 544 btrfs_commit_transaction(trans, root); 545 546 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 547 548 return 0; 549 } 550 551 static void btrfs_dev_replace_update_device_in_mapping_tree( 552 struct btrfs_fs_info *fs_info, 553 struct btrfs_device *srcdev, 554 struct btrfs_device *tgtdev) 555 { 556 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; 557 struct extent_map *em; 558 struct map_lookup *map; 559 u64 start = 0; 560 int i; 561 562 write_lock(&em_tree->lock); 563 do { 564 em = lookup_extent_mapping(em_tree, start, (u64)-1); 565 if (!em) 566 break; 567 map = (struct map_lookup *)em->bdev; 568 for (i = 0; i < map->num_stripes; i++) 569 if (srcdev == map->stripes[i].dev) 570 map->stripes[i].dev = tgtdev; 571 start = em->start + em->len; 572 free_extent_map(em); 573 } while (start); 574 write_unlock(&em_tree->lock); 575 } 576 577 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, 578 char *srcdev_name, 579 struct btrfs_device **device) 580 { 581 int ret; 582 583 if (srcdevid) { 584 ret = 0; 585 *device = btrfs_find_device(root->fs_info, srcdevid, NULL, 586 NULL); 587 if (!*device) 588 ret = -ENOENT; 589 } else { 590 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name, 591 device); 592 } 593 return ret; 594 } 595 596 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, 597 struct btrfs_ioctl_dev_replace_args *args) 598 { 599 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 600 601 btrfs_dev_replace_lock(dev_replace); 602 /* even if !dev_replace_is_valid, the values are good enough for 603 * the replace_status ioctl */ 604 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 605 args->status.replace_state = dev_replace->replace_state; 606 args->status.time_started = dev_replace->time_started; 607 args->status.time_stopped = dev_replace->time_stopped; 608 args->status.num_write_errors = 609 atomic64_read(&dev_replace->num_write_errors); 610 args->status.num_uncorrectable_read_errors = 611 atomic64_read(&dev_replace->num_uncorrectable_read_errors); 612 switch (dev_replace->replace_state) { 613 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 614 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 615 args->status.progress_1000 = 0; 616 break; 617 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 618 args->status.progress_1000 = 1000; 619 break; 620 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 621 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 622 args->status.progress_1000 = div64_u64(dev_replace->cursor_left, 623 div64_u64(dev_replace->srcdev->total_bytes, 1000)); 624 break; 625 } 626 btrfs_dev_replace_unlock(dev_replace); 627 } 628 629 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info, 630 struct btrfs_ioctl_dev_replace_args *args) 631 { 632 args->result = __btrfs_dev_replace_cancel(fs_info); 633 return 0; 634 } 635 636 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) 637 { 638 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 639 struct btrfs_device *tgt_device = NULL; 640 struct btrfs_trans_handle *trans; 641 struct btrfs_root *root = fs_info->tree_root; 642 u64 result; 643 int ret; 644 645 if (fs_info->sb->s_flags & MS_RDONLY) 646 return -EROFS; 647 648 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 649 btrfs_dev_replace_lock(dev_replace); 650 switch (dev_replace->replace_state) { 651 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 652 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 653 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 654 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; 655 btrfs_dev_replace_unlock(dev_replace); 656 goto leave; 657 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 658 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 659 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; 660 tgt_device = dev_replace->tgtdev; 661 dev_replace->tgtdev = NULL; 662 dev_replace->srcdev = NULL; 663 break; 664 } 665 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; 666 dev_replace->time_stopped = get_seconds(); 667 dev_replace->item_needs_writeback = 1; 668 btrfs_dev_replace_unlock(dev_replace); 669 btrfs_scrub_cancel(fs_info); 670 671 trans = btrfs_start_transaction(root, 0); 672 if (IS_ERR(trans)) { 673 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 674 return PTR_ERR(trans); 675 } 676 ret = btrfs_commit_transaction(trans, root); 677 WARN_ON(ret); 678 if (tgt_device) 679 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); 680 681 leave: 682 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 683 return result; 684 } 685 686 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) 687 { 688 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 689 690 mutex_lock(&dev_replace->lock_finishing_cancel_unmount); 691 btrfs_dev_replace_lock(dev_replace); 692 switch (dev_replace->replace_state) { 693 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 694 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 695 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 696 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 697 break; 698 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 699 dev_replace->replace_state = 700 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; 701 dev_replace->time_stopped = get_seconds(); 702 dev_replace->item_needs_writeback = 1; 703 pr_info("btrfs: suspending dev_replace for unmount\n"); 704 break; 705 } 706 707 btrfs_dev_replace_unlock(dev_replace); 708 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); 709 } 710 711 /* resume dev_replace procedure that was interrupted by unmount */ 712 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) 713 { 714 struct task_struct *task; 715 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 716 717 btrfs_dev_replace_lock(dev_replace); 718 switch (dev_replace->replace_state) { 719 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 720 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 721 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 722 btrfs_dev_replace_unlock(dev_replace); 723 return 0; 724 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 725 break; 726 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 727 dev_replace->replace_state = 728 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; 729 break; 730 } 731 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { 732 pr_info("btrfs: cannot continue dev_replace, tgtdev is missing\n" 733 "btrfs: you may cancel the operation after 'mount -o degraded'\n"); 734 btrfs_dev_replace_unlock(dev_replace); 735 return 0; 736 } 737 btrfs_dev_replace_unlock(dev_replace); 738 739 WARN_ON(atomic_xchg( 740 &fs_info->mutually_exclusive_operation_running, 1)); 741 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); 742 return PTR_ERR_OR_ZERO(task); 743 } 744 745 static int btrfs_dev_replace_kthread(void *data) 746 { 747 struct btrfs_fs_info *fs_info = data; 748 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 749 struct btrfs_ioctl_dev_replace_args *status_args; 750 u64 progress; 751 752 status_args = kzalloc(sizeof(*status_args), GFP_NOFS); 753 if (status_args) { 754 btrfs_dev_replace_status(fs_info, status_args); 755 progress = status_args->status.progress_1000; 756 kfree(status_args); 757 do_div(progress, 10); 758 printk_in_rcu(KERN_INFO 759 "btrfs: continuing dev_replace from %s (devid %llu) to %s @%u%%\n", 760 dev_replace->srcdev->missing ? "<missing disk>" : 761 rcu_str_deref(dev_replace->srcdev->name), 762 dev_replace->srcdev->devid, 763 dev_replace->tgtdev ? 764 rcu_str_deref(dev_replace->tgtdev->name) : 765 "<missing target disk>", 766 (unsigned int)progress); 767 } 768 btrfs_dev_replace_continue_on_mount(fs_info); 769 atomic_set(&fs_info->mutually_exclusive_operation_running, 0); 770 771 return 0; 772 } 773 774 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info) 775 { 776 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 777 int ret; 778 779 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, 780 dev_replace->committed_cursor_left, 781 dev_replace->srcdev->total_bytes, 782 &dev_replace->scrub_progress, 0, 1); 783 ret = btrfs_dev_replace_finishing(fs_info, ret); 784 WARN_ON(ret); 785 return 0; 786 } 787 788 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) 789 { 790 if (!dev_replace->is_valid) 791 return 0; 792 793 switch (dev_replace->replace_state) { 794 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: 795 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: 796 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: 797 return 0; 798 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: 799 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: 800 /* 801 * return true even if tgtdev is missing (this is 802 * something that can happen if the dev_replace 803 * procedure is suspended by an umount and then 804 * the tgtdev is missing (or "btrfs dev scan") was 805 * not called and the the filesystem is remounted 806 * in degraded state. This does not stop the 807 * dev_replace procedure. It needs to be canceled 808 * manually if the cancelation is wanted. 809 */ 810 break; 811 } 812 return 1; 813 } 814 815 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace) 816 { 817 /* the beginning is just an optimization for the typical case */ 818 if (atomic_read(&dev_replace->nesting_level) == 0) { 819 acquire_lock: 820 /* this is not a nested case where the same thread 821 * is trying to acqurire the same lock twice */ 822 mutex_lock(&dev_replace->lock); 823 mutex_lock(&dev_replace->lock_management_lock); 824 dev_replace->lock_owner = current->pid; 825 atomic_inc(&dev_replace->nesting_level); 826 mutex_unlock(&dev_replace->lock_management_lock); 827 return; 828 } 829 830 mutex_lock(&dev_replace->lock_management_lock); 831 if (atomic_read(&dev_replace->nesting_level) > 0 && 832 dev_replace->lock_owner == current->pid) { 833 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 834 atomic_inc(&dev_replace->nesting_level); 835 mutex_unlock(&dev_replace->lock_management_lock); 836 return; 837 } 838 839 mutex_unlock(&dev_replace->lock_management_lock); 840 goto acquire_lock; 841 } 842 843 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace) 844 { 845 WARN_ON(!mutex_is_locked(&dev_replace->lock)); 846 mutex_lock(&dev_replace->lock_management_lock); 847 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1); 848 WARN_ON(dev_replace->lock_owner != current->pid); 849 atomic_dec(&dev_replace->nesting_level); 850 if (atomic_read(&dev_replace->nesting_level) == 0) { 851 dev_replace->lock_owner = 0; 852 mutex_unlock(&dev_replace->lock_management_lock); 853 mutex_unlock(&dev_replace->lock); 854 } else { 855 mutex_unlock(&dev_replace->lock_management_lock); 856 } 857 } 858