1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2007 Oracle. All rights reserved. 4 */ 5 6 #include <linux/sched.h> 7 #include <linux/sched/mm.h> 8 #include <linux/slab.h> 9 #include <linux/ratelimit.h> 10 #include <linux/kthread.h> 11 #include <linux/semaphore.h> 12 #include <linux/uuid.h> 13 #include <linux/list_sort.h> 14 #include <linux/namei.h> 15 #include "misc.h" 16 #include "ctree.h" 17 #include "extent_map.h" 18 #include "disk-io.h" 19 #include "transaction.h" 20 #include "print-tree.h" 21 #include "volumes.h" 22 #include "raid56.h" 23 #include "rcu-string.h" 24 #include "dev-replace.h" 25 #include "sysfs.h" 26 #include "tree-checker.h" 27 #include "space-info.h" 28 #include "block-group.h" 29 #include "discard.h" 30 #include "zoned.h" 31 #include "fs.h" 32 #include "accessors.h" 33 #include "uuid-tree.h" 34 #include "ioctl.h" 35 #include "relocation.h" 36 #include "scrub.h" 37 #include "super.h" 38 39 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ 40 BTRFS_BLOCK_GROUP_RAID10 | \ 41 BTRFS_BLOCK_GROUP_RAID56_MASK) 42 43 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { 44 [BTRFS_RAID_RAID10] = { 45 .sub_stripes = 2, 46 .dev_stripes = 1, 47 .devs_max = 0, /* 0 == as many as possible */ 48 .devs_min = 2, 49 .tolerated_failures = 1, 50 .devs_increment = 2, 51 .ncopies = 2, 52 .nparity = 0, 53 .raid_name = "raid10", 54 .bg_flag = BTRFS_BLOCK_GROUP_RAID10, 55 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, 56 }, 57 [BTRFS_RAID_RAID1] = { 58 .sub_stripes = 1, 59 .dev_stripes = 1, 60 .devs_max = 2, 61 .devs_min = 2, 62 .tolerated_failures = 1, 63 .devs_increment = 2, 64 .ncopies = 2, 65 .nparity = 0, 66 .raid_name = "raid1", 67 .bg_flag = BTRFS_BLOCK_GROUP_RAID1, 68 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, 69 }, 70 [BTRFS_RAID_RAID1C3] = { 71 .sub_stripes = 1, 72 .dev_stripes = 1, 73 .devs_max = 3, 74 .devs_min = 3, 75 .tolerated_failures = 2, 76 .devs_increment = 3, 77 .ncopies = 3, 78 .nparity = 0, 79 .raid_name = "raid1c3", 80 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, 81 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, 82 }, 83 [BTRFS_RAID_RAID1C4] = { 84 .sub_stripes = 1, 85 .dev_stripes = 1, 86 .devs_max = 4, 87 .devs_min = 4, 88 .tolerated_failures = 3, 89 .devs_increment = 4, 90 .ncopies = 4, 91 .nparity = 0, 92 .raid_name = "raid1c4", 93 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, 94 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, 95 }, 96 [BTRFS_RAID_DUP] = { 97 .sub_stripes = 1, 98 .dev_stripes = 2, 99 .devs_max = 1, 100 .devs_min = 1, 101 .tolerated_failures = 0, 102 .devs_increment = 1, 103 .ncopies = 2, 104 .nparity = 0, 105 .raid_name = "dup", 106 .bg_flag = BTRFS_BLOCK_GROUP_DUP, 107 .mindev_error = 0, 108 }, 109 [BTRFS_RAID_RAID0] = { 110 .sub_stripes = 1, 111 .dev_stripes = 1, 112 .devs_max = 0, 113 .devs_min = 1, 114 .tolerated_failures = 0, 115 .devs_increment = 1, 116 .ncopies = 1, 117 .nparity = 0, 118 .raid_name = "raid0", 119 .bg_flag = BTRFS_BLOCK_GROUP_RAID0, 120 .mindev_error = 0, 121 }, 122 [BTRFS_RAID_SINGLE] = { 123 .sub_stripes = 1, 124 .dev_stripes = 1, 125 .devs_max = 1, 126 .devs_min = 1, 127 .tolerated_failures = 0, 128 .devs_increment = 1, 129 .ncopies = 1, 130 .nparity = 0, 131 .raid_name = "single", 132 .bg_flag = 0, 133 .mindev_error = 0, 134 }, 135 [BTRFS_RAID_RAID5] = { 136 .sub_stripes = 1, 137 .dev_stripes = 1, 138 .devs_max = 0, 139 .devs_min = 2, 140 .tolerated_failures = 1, 141 .devs_increment = 1, 142 .ncopies = 1, 143 .nparity = 1, 144 .raid_name = "raid5", 145 .bg_flag = BTRFS_BLOCK_GROUP_RAID5, 146 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, 147 }, 148 [BTRFS_RAID_RAID6] = { 149 .sub_stripes = 1, 150 .dev_stripes = 1, 151 .devs_max = 0, 152 .devs_min = 3, 153 .tolerated_failures = 2, 154 .devs_increment = 1, 155 .ncopies = 1, 156 .nparity = 2, 157 .raid_name = "raid6", 158 .bg_flag = BTRFS_BLOCK_GROUP_RAID6, 159 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, 160 }, 161 }; 162 163 /* 164 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 165 * can be used as index to access btrfs_raid_array[]. 166 */ 167 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) 168 { 169 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK); 170 171 if (!profile) 172 return BTRFS_RAID_SINGLE; 173 174 return BTRFS_BG_FLAG_TO_INDEX(profile); 175 } 176 177 const char *btrfs_bg_type_to_raid_name(u64 flags) 178 { 179 const int index = btrfs_bg_flags_to_raid_index(flags); 180 181 if (index >= BTRFS_NR_RAID_TYPES) 182 return NULL; 183 184 return btrfs_raid_array[index].raid_name; 185 } 186 187 int btrfs_nr_parity_stripes(u64 type) 188 { 189 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type); 190 191 return btrfs_raid_array[index].nparity; 192 } 193 194 /* 195 * Fill @buf with textual description of @bg_flags, no more than @size_buf 196 * bytes including terminating null byte. 197 */ 198 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) 199 { 200 int i; 201 int ret; 202 char *bp = buf; 203 u64 flags = bg_flags; 204 u32 size_bp = size_buf; 205 206 if (!flags) { 207 strcpy(bp, "NONE"); 208 return; 209 } 210 211 #define DESCRIBE_FLAG(flag, desc) \ 212 do { \ 213 if (flags & (flag)) { \ 214 ret = snprintf(bp, size_bp, "%s|", (desc)); \ 215 if (ret < 0 || ret >= size_bp) \ 216 goto out_overflow; \ 217 size_bp -= ret; \ 218 bp += ret; \ 219 flags &= ~(flag); \ 220 } \ 221 } while (0) 222 223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); 224 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); 225 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); 226 227 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); 228 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) 229 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, 230 btrfs_raid_array[i].raid_name); 231 #undef DESCRIBE_FLAG 232 233 if (flags) { 234 ret = snprintf(bp, size_bp, "0x%llx|", flags); 235 size_bp -= ret; 236 } 237 238 if (size_bp < size_buf) 239 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ 240 241 /* 242 * The text is trimmed, it's up to the caller to provide sufficiently 243 * large buffer 244 */ 245 out_overflow:; 246 } 247 248 static int init_first_rw_device(struct btrfs_trans_handle *trans); 249 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); 250 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); 251 252 /* 253 * Device locking 254 * ============== 255 * 256 * There are several mutexes that protect manipulation of devices and low-level 257 * structures like chunks but not block groups, extents or files 258 * 259 * uuid_mutex (global lock) 260 * ------------------------ 261 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from 262 * the SCAN_DEV ioctl registration or from mount either implicitly (the first 263 * device) or requested by the device= mount option 264 * 265 * the mutex can be very coarse and can cover long-running operations 266 * 267 * protects: updates to fs_devices counters like missing devices, rw devices, 268 * seeding, structure cloning, opening/closing devices at mount/umount time 269 * 270 * global::fs_devs - add, remove, updates to the global list 271 * 272 * does not protect: manipulation of the fs_devices::devices list in general 273 * but in mount context it could be used to exclude list modifications by eg. 274 * scan ioctl 275 * 276 * btrfs_device::name - renames (write side), read is RCU 277 * 278 * fs_devices::device_list_mutex (per-fs, with RCU) 279 * ------------------------------------------------ 280 * protects updates to fs_devices::devices, ie. adding and deleting 281 * 282 * simple list traversal with read-only actions can be done with RCU protection 283 * 284 * may be used to exclude some operations from running concurrently without any 285 * modifications to the list (see write_all_supers) 286 * 287 * Is not required at mount and close times, because our device list is 288 * protected by the uuid_mutex at that point. 289 * 290 * balance_mutex 291 * ------------- 292 * protects balance structures (status, state) and context accessed from 293 * several places (internally, ioctl) 294 * 295 * chunk_mutex 296 * ----------- 297 * protects chunks, adding or removing during allocation, trim or when a new 298 * device is added/removed. Additionally it also protects post_commit_list of 299 * individual devices, since they can be added to the transaction's 300 * post_commit_list only with chunk_mutex held. 301 * 302 * cleaner_mutex 303 * ------------- 304 * a big lock that is held by the cleaner thread and prevents running subvolume 305 * cleaning together with relocation or delayed iputs 306 * 307 * 308 * Lock nesting 309 * ============ 310 * 311 * uuid_mutex 312 * device_list_mutex 313 * chunk_mutex 314 * balance_mutex 315 * 316 * 317 * Exclusive operations 318 * ==================== 319 * 320 * Maintains the exclusivity of the following operations that apply to the 321 * whole filesystem and cannot run in parallel. 322 * 323 * - Balance (*) 324 * - Device add 325 * - Device remove 326 * - Device replace (*) 327 * - Resize 328 * 329 * The device operations (as above) can be in one of the following states: 330 * 331 * - Running state 332 * - Paused state 333 * - Completed state 334 * 335 * Only device operations marked with (*) can go into the Paused state for the 336 * following reasons: 337 * 338 * - ioctl (only Balance can be Paused through ioctl) 339 * - filesystem remounted as read-only 340 * - filesystem unmounted and mounted as read-only 341 * - system power-cycle and filesystem mounted as read-only 342 * - filesystem or device errors leading to forced read-only 343 * 344 * The status of exclusive operation is set and cleared atomically. 345 * During the course of Paused state, fs_info::exclusive_operation remains set. 346 * A device operation in Paused or Running state can be canceled or resumed 347 * either by ioctl (Balance only) or when remounted as read-write. 348 * The exclusive status is cleared when the device operation is canceled or 349 * completed. 350 */ 351 352 DEFINE_MUTEX(uuid_mutex); 353 static LIST_HEAD(fs_uuids); 354 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) 355 { 356 return &fs_uuids; 357 } 358 359 /* 360 * alloc_fs_devices - allocate struct btrfs_fs_devices 361 * @fsid: if not NULL, copy the UUID to fs_devices::fsid 362 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid 363 * 364 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). 365 * The returned struct is not linked onto any lists and can be destroyed with 366 * kfree() right away. 367 */ 368 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, 369 const u8 *metadata_fsid) 370 { 371 struct btrfs_fs_devices *fs_devs; 372 373 ASSERT(fsid || !metadata_fsid); 374 375 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 376 if (!fs_devs) 377 return ERR_PTR(-ENOMEM); 378 379 mutex_init(&fs_devs->device_list_mutex); 380 381 INIT_LIST_HEAD(&fs_devs->devices); 382 INIT_LIST_HEAD(&fs_devs->alloc_list); 383 INIT_LIST_HEAD(&fs_devs->fs_list); 384 INIT_LIST_HEAD(&fs_devs->seed_list); 385 386 if (fsid) { 387 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 388 memcpy(fs_devs->metadata_uuid, 389 metadata_fsid ?: fsid, BTRFS_FSID_SIZE); 390 } 391 392 return fs_devs; 393 } 394 395 static void btrfs_free_device(struct btrfs_device *device) 396 { 397 WARN_ON(!list_empty(&device->post_commit_list)); 398 rcu_string_free(device->name); 399 extent_io_tree_release(&device->alloc_state); 400 btrfs_destroy_dev_zone_info(device); 401 kfree(device); 402 } 403 404 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 405 { 406 struct btrfs_device *device; 407 408 WARN_ON(fs_devices->opened); 409 while (!list_empty(&fs_devices->devices)) { 410 device = list_entry(fs_devices->devices.next, 411 struct btrfs_device, dev_list); 412 list_del(&device->dev_list); 413 btrfs_free_device(device); 414 } 415 kfree(fs_devices); 416 } 417 418 void __exit btrfs_cleanup_fs_uuids(void) 419 { 420 struct btrfs_fs_devices *fs_devices; 421 422 while (!list_empty(&fs_uuids)) { 423 fs_devices = list_entry(fs_uuids.next, 424 struct btrfs_fs_devices, fs_list); 425 list_del(&fs_devices->fs_list); 426 free_fs_devices(fs_devices); 427 } 428 } 429 430 static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices, 431 const u8 *fsid, const u8 *metadata_fsid) 432 { 433 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0) 434 return false; 435 436 if (!metadata_fsid) 437 return true; 438 439 if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0) 440 return false; 441 442 return true; 443 } 444 445 static noinline struct btrfs_fs_devices *find_fsid( 446 const u8 *fsid, const u8 *metadata_fsid) 447 { 448 struct btrfs_fs_devices *fs_devices; 449 450 ASSERT(fsid); 451 452 /* Handle non-split brain cases */ 453 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 454 if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid)) 455 return fs_devices; 456 } 457 return NULL; 458 } 459 460 /* 461 * First check if the metadata_uuid is different from the fsid in the given 462 * fs_devices. Then check if the given fsid is the same as the metadata_uuid 463 * in the fs_devices. If it is, return true; otherwise, return false. 464 */ 465 static inline bool check_fsid_changed(const struct btrfs_fs_devices *fs_devices, 466 const u8 *fsid) 467 { 468 return memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 469 BTRFS_FSID_SIZE) != 0 && 470 memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE) == 0; 471 } 472 473 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 474 struct btrfs_super_block *disk_super) 475 { 476 477 struct btrfs_fs_devices *fs_devices; 478 479 /* 480 * Handle scanned device having completed its fsid change but 481 * belonging to a fs_devices that was created by first scanning 482 * a device which didn't have its fsid/metadata_uuid changed 483 * at all and the CHANGING_FSID_V2 flag set. 484 */ 485 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 486 if (!fs_devices->fsid_change) 487 continue; 488 489 if (match_fsid_fs_devices(fs_devices, disk_super->metadata_uuid, 490 fs_devices->fsid)) 491 return fs_devices; 492 } 493 494 /* 495 * Handle scanned device having completed its fsid change but 496 * belonging to a fs_devices that was created by a device that 497 * has an outdated pair of fsid/metadata_uuid and 498 * CHANGING_FSID_V2 flag set. 499 */ 500 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 501 if (!fs_devices->fsid_change) 502 continue; 503 504 if (check_fsid_changed(fs_devices, disk_super->metadata_uuid)) 505 return fs_devices; 506 } 507 508 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 509 } 510 511 512 static int 513 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 514 int flush, struct block_device **bdev, 515 struct btrfs_super_block **disk_super) 516 { 517 int ret; 518 519 *bdev = blkdev_get_by_path(device_path, flags, holder); 520 521 if (IS_ERR(*bdev)) { 522 ret = PTR_ERR(*bdev); 523 goto error; 524 } 525 526 if (flush) 527 sync_blockdev(*bdev); 528 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 529 if (ret) { 530 blkdev_put(*bdev, flags); 531 goto error; 532 } 533 invalidate_bdev(*bdev); 534 *disk_super = btrfs_read_dev_super(*bdev); 535 if (IS_ERR(*disk_super)) { 536 ret = PTR_ERR(*disk_super); 537 blkdev_put(*bdev, flags); 538 goto error; 539 } 540 541 return 0; 542 543 error: 544 *bdev = NULL; 545 return ret; 546 } 547 548 /* 549 * Search and remove all stale devices (which are not mounted). When both 550 * inputs are NULL, it will search and release all stale devices. 551 * 552 * @devt: Optional. When provided will it release all unmounted devices 553 * matching this devt only. 554 * @skip_device: Optional. Will skip this device when searching for the stale 555 * devices. 556 * 557 * Return: 0 for success or if @devt is 0. 558 * -EBUSY if @devt is a mounted device. 559 * -ENOENT if @devt does not match any device in the list. 560 */ 561 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) 562 { 563 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 564 struct btrfs_device *device, *tmp_device; 565 int ret = 0; 566 567 lockdep_assert_held(&uuid_mutex); 568 569 if (devt) 570 ret = -ENOENT; 571 572 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 573 574 mutex_lock(&fs_devices->device_list_mutex); 575 list_for_each_entry_safe(device, tmp_device, 576 &fs_devices->devices, dev_list) { 577 if (skip_device && skip_device == device) 578 continue; 579 if (devt && devt != device->devt) 580 continue; 581 if (fs_devices->opened) { 582 /* for an already deleted device return 0 */ 583 if (devt && ret != 0) 584 ret = -EBUSY; 585 break; 586 } 587 588 /* delete the stale device */ 589 fs_devices->num_devices--; 590 list_del(&device->dev_list); 591 btrfs_free_device(device); 592 593 ret = 0; 594 } 595 mutex_unlock(&fs_devices->device_list_mutex); 596 597 if (fs_devices->num_devices == 0) { 598 btrfs_sysfs_remove_fsid(fs_devices); 599 list_del(&fs_devices->fs_list); 600 free_fs_devices(fs_devices); 601 } 602 } 603 604 return ret; 605 } 606 607 /* 608 * This is only used on mount, and we are protected from competing things 609 * messing with our fs_devices by the uuid_mutex, thus we do not need the 610 * fs_devices->device_list_mutex here. 611 */ 612 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 613 struct btrfs_device *device, fmode_t flags, 614 void *holder) 615 { 616 struct block_device *bdev; 617 struct btrfs_super_block *disk_super; 618 u64 devid; 619 int ret; 620 621 if (device->bdev) 622 return -EINVAL; 623 if (!device->name) 624 return -EINVAL; 625 626 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 627 &bdev, &disk_super); 628 if (ret) 629 return ret; 630 631 devid = btrfs_stack_device_id(&disk_super->dev_item); 632 if (devid != device->devid) 633 goto error_free_page; 634 635 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 636 goto error_free_page; 637 638 device->generation = btrfs_super_generation(disk_super); 639 640 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 641 if (btrfs_super_incompat_flags(disk_super) & 642 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 643 pr_err( 644 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 645 goto error_free_page; 646 } 647 648 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 649 fs_devices->seeding = true; 650 } else { 651 if (bdev_read_only(bdev)) 652 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 653 else 654 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 655 } 656 657 if (!bdev_nonrot(bdev)) 658 fs_devices->rotating = true; 659 660 if (bdev_max_discard_sectors(bdev)) 661 fs_devices->discardable = true; 662 663 device->bdev = bdev; 664 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 665 device->mode = flags; 666 667 fs_devices->open_devices++; 668 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 669 device->devid != BTRFS_DEV_REPLACE_DEVID) { 670 fs_devices->rw_devices++; 671 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 672 } 673 btrfs_release_disk_super(disk_super); 674 675 return 0; 676 677 error_free_page: 678 btrfs_release_disk_super(disk_super); 679 blkdev_put(bdev, flags); 680 681 return -EINVAL; 682 } 683 684 /* 685 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 686 * being created with a disk that has already completed its fsid change. Such 687 * disk can belong to an fs which has its FSID changed or to one which doesn't. 688 * Handle both cases here. 689 */ 690 static struct btrfs_fs_devices *find_fsid_inprogress( 691 struct btrfs_super_block *disk_super) 692 { 693 struct btrfs_fs_devices *fs_devices; 694 695 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 696 if (fs_devices->fsid_change) 697 continue; 698 699 if (check_fsid_changed(fs_devices, disk_super->fsid)) 700 return fs_devices; 701 } 702 703 return find_fsid(disk_super->fsid, NULL); 704 } 705 706 static struct btrfs_fs_devices *find_fsid_changed( 707 struct btrfs_super_block *disk_super) 708 { 709 struct btrfs_fs_devices *fs_devices; 710 711 /* 712 * Handles the case where scanned device is part of an fs that had 713 * multiple successful changes of FSID but currently device didn't 714 * observe it. Meaning our fsid will be different than theirs. We need 715 * to handle two subcases : 716 * 1 - The fs still continues to have different METADATA/FSID uuids. 717 * 2 - The fs is switched back to its original FSID (METADATA/FSID 718 * are equal). 719 */ 720 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 721 /* Changed UUIDs */ 722 if (check_fsid_changed(fs_devices, disk_super->metadata_uuid) && 723 memcmp(fs_devices->fsid, disk_super->fsid, 724 BTRFS_FSID_SIZE) != 0) 725 return fs_devices; 726 727 /* Unchanged UUIDs */ 728 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 729 BTRFS_FSID_SIZE) == 0 && 730 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 731 BTRFS_FSID_SIZE) == 0) 732 return fs_devices; 733 } 734 735 return NULL; 736 } 737 738 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 739 struct btrfs_super_block *disk_super) 740 { 741 struct btrfs_fs_devices *fs_devices; 742 743 /* 744 * Handle the case where the scanned device is part of an fs whose last 745 * metadata UUID change reverted it to the original FSID. At the same 746 * time fs_devices was first created by another constituent device 747 * which didn't fully observe the operation. This results in an 748 * btrfs_fs_devices created with metadata/fsid different AND 749 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 750 * fs_devices equal to the FSID of the disk. 751 */ 752 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 753 if (!fs_devices->fsid_change) 754 continue; 755 756 if (check_fsid_changed(fs_devices, disk_super->fsid)) 757 return fs_devices; 758 } 759 760 return NULL; 761 } 762 /* 763 * Add new device to list of registered devices 764 * 765 * Returns: 766 * device pointer which was just added or updated when successful 767 * error pointer when failed 768 */ 769 static noinline struct btrfs_device *device_list_add(const char *path, 770 struct btrfs_super_block *disk_super, 771 bool *new_device_added) 772 { 773 struct btrfs_device *device; 774 struct btrfs_fs_devices *fs_devices = NULL; 775 struct rcu_string *name; 776 u64 found_transid = btrfs_super_generation(disk_super); 777 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 778 dev_t path_devt; 779 int error; 780 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 781 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 782 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 783 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 784 785 error = lookup_bdev(path, &path_devt); 786 if (error) { 787 btrfs_err(NULL, "failed to lookup block device for path %s: %d", 788 path, error); 789 return ERR_PTR(error); 790 } 791 792 if (fsid_change_in_progress) { 793 if (!has_metadata_uuid) 794 fs_devices = find_fsid_inprogress(disk_super); 795 else 796 fs_devices = find_fsid_changed(disk_super); 797 } else if (has_metadata_uuid) { 798 fs_devices = find_fsid_with_metadata_uuid(disk_super); 799 } else { 800 fs_devices = find_fsid_reverted_metadata(disk_super); 801 if (!fs_devices) 802 fs_devices = find_fsid(disk_super->fsid, NULL); 803 } 804 805 806 if (!fs_devices) { 807 fs_devices = alloc_fs_devices(disk_super->fsid, 808 has_metadata_uuid ? disk_super->metadata_uuid : NULL); 809 if (IS_ERR(fs_devices)) 810 return ERR_CAST(fs_devices); 811 812 fs_devices->fsid_change = fsid_change_in_progress; 813 814 mutex_lock(&fs_devices->device_list_mutex); 815 list_add(&fs_devices->fs_list, &fs_uuids); 816 817 device = NULL; 818 } else { 819 struct btrfs_dev_lookup_args args = { 820 .devid = devid, 821 .uuid = disk_super->dev_item.uuid, 822 }; 823 824 mutex_lock(&fs_devices->device_list_mutex); 825 device = btrfs_find_device(fs_devices, &args); 826 827 /* 828 * If this disk has been pulled into an fs devices created by 829 * a device which had the CHANGING_FSID_V2 flag then replace the 830 * metadata_uuid/fsid values of the fs_devices. 831 */ 832 if (fs_devices->fsid_change && 833 found_transid > fs_devices->latest_generation) { 834 memcpy(fs_devices->fsid, disk_super->fsid, 835 BTRFS_FSID_SIZE); 836 837 if (has_metadata_uuid) 838 memcpy(fs_devices->metadata_uuid, 839 disk_super->metadata_uuid, 840 BTRFS_FSID_SIZE); 841 else 842 memcpy(fs_devices->metadata_uuid, 843 disk_super->fsid, BTRFS_FSID_SIZE); 844 845 fs_devices->fsid_change = false; 846 } 847 } 848 849 if (!device) { 850 unsigned int nofs_flag; 851 852 if (fs_devices->opened) { 853 btrfs_err(NULL, 854 "device %s belongs to fsid %pU, and the fs is already mounted", 855 path, fs_devices->fsid); 856 mutex_unlock(&fs_devices->device_list_mutex); 857 return ERR_PTR(-EBUSY); 858 } 859 860 nofs_flag = memalloc_nofs_save(); 861 device = btrfs_alloc_device(NULL, &devid, 862 disk_super->dev_item.uuid, path); 863 memalloc_nofs_restore(nofs_flag); 864 if (IS_ERR(device)) { 865 mutex_unlock(&fs_devices->device_list_mutex); 866 /* we can safely leave the fs_devices entry around */ 867 return device; 868 } 869 870 device->devt = path_devt; 871 872 list_add_rcu(&device->dev_list, &fs_devices->devices); 873 fs_devices->num_devices++; 874 875 device->fs_devices = fs_devices; 876 *new_device_added = true; 877 878 if (disk_super->label[0]) 879 pr_info( 880 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 881 disk_super->label, devid, found_transid, path, 882 current->comm, task_pid_nr(current)); 883 else 884 pr_info( 885 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 886 disk_super->fsid, devid, found_transid, path, 887 current->comm, task_pid_nr(current)); 888 889 } else if (!device->name || strcmp(device->name->str, path)) { 890 /* 891 * When FS is already mounted. 892 * 1. If you are here and if the device->name is NULL that 893 * means this device was missing at time of FS mount. 894 * 2. If you are here and if the device->name is different 895 * from 'path' that means either 896 * a. The same device disappeared and reappeared with 897 * different name. or 898 * b. The missing-disk-which-was-replaced, has 899 * reappeared now. 900 * 901 * We must allow 1 and 2a above. But 2b would be a spurious 902 * and unintentional. 903 * 904 * Further in case of 1 and 2a above, the disk at 'path' 905 * would have missed some transaction when it was away and 906 * in case of 2a the stale bdev has to be updated as well. 907 * 2b must not be allowed at all time. 908 */ 909 910 /* 911 * For now, we do allow update to btrfs_fs_device through the 912 * btrfs dev scan cli after FS has been mounted. We're still 913 * tracking a problem where systems fail mount by subvolume id 914 * when we reject replacement on a mounted FS. 915 */ 916 if (!fs_devices->opened && found_transid < device->generation) { 917 /* 918 * That is if the FS is _not_ mounted and if you 919 * are here, that means there is more than one 920 * disk with same uuid and devid.We keep the one 921 * with larger generation number or the last-in if 922 * generation are equal. 923 */ 924 mutex_unlock(&fs_devices->device_list_mutex); 925 btrfs_err(NULL, 926 "device %s already registered with a higher generation, found %llu expect %llu", 927 path, found_transid, device->generation); 928 return ERR_PTR(-EEXIST); 929 } 930 931 /* 932 * We are going to replace the device path for a given devid, 933 * make sure it's the same device if the device is mounted 934 * 935 * NOTE: the device->fs_info may not be reliable here so pass 936 * in a NULL to message helpers instead. This avoids a possible 937 * use-after-free when the fs_info and fs_info->sb are already 938 * torn down. 939 */ 940 if (device->bdev) { 941 if (device->devt != path_devt) { 942 mutex_unlock(&fs_devices->device_list_mutex); 943 btrfs_warn_in_rcu(NULL, 944 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 945 path, devid, found_transid, 946 current->comm, 947 task_pid_nr(current)); 948 return ERR_PTR(-EEXIST); 949 } 950 btrfs_info_in_rcu(NULL, 951 "devid %llu device path %s changed to %s scanned by %s (%d)", 952 devid, btrfs_dev_name(device), 953 path, current->comm, 954 task_pid_nr(current)); 955 } 956 957 name = rcu_string_strdup(path, GFP_NOFS); 958 if (!name) { 959 mutex_unlock(&fs_devices->device_list_mutex); 960 return ERR_PTR(-ENOMEM); 961 } 962 rcu_string_free(device->name); 963 rcu_assign_pointer(device->name, name); 964 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 965 fs_devices->missing_devices--; 966 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 967 } 968 device->devt = path_devt; 969 } 970 971 /* 972 * Unmount does not free the btrfs_device struct but would zero 973 * generation along with most of the other members. So just update 974 * it back. We need it to pick the disk with largest generation 975 * (as above). 976 */ 977 if (!fs_devices->opened) { 978 device->generation = found_transid; 979 fs_devices->latest_generation = max_t(u64, found_transid, 980 fs_devices->latest_generation); 981 } 982 983 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 984 985 mutex_unlock(&fs_devices->device_list_mutex); 986 return device; 987 } 988 989 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 990 { 991 struct btrfs_fs_devices *fs_devices; 992 struct btrfs_device *device; 993 struct btrfs_device *orig_dev; 994 int ret = 0; 995 996 lockdep_assert_held(&uuid_mutex); 997 998 fs_devices = alloc_fs_devices(orig->fsid, NULL); 999 if (IS_ERR(fs_devices)) 1000 return fs_devices; 1001 1002 fs_devices->total_devices = orig->total_devices; 1003 1004 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 1005 const char *dev_path = NULL; 1006 1007 /* 1008 * This is ok to do without RCU read locked because we hold the 1009 * uuid mutex so nothing we touch in here is going to disappear. 1010 */ 1011 if (orig_dev->name) 1012 dev_path = orig_dev->name->str; 1013 1014 device = btrfs_alloc_device(NULL, &orig_dev->devid, 1015 orig_dev->uuid, dev_path); 1016 if (IS_ERR(device)) { 1017 ret = PTR_ERR(device); 1018 goto error; 1019 } 1020 1021 if (orig_dev->zone_info) { 1022 struct btrfs_zoned_device_info *zone_info; 1023 1024 zone_info = btrfs_clone_dev_zone_info(orig_dev); 1025 if (!zone_info) { 1026 btrfs_free_device(device); 1027 ret = -ENOMEM; 1028 goto error; 1029 } 1030 device->zone_info = zone_info; 1031 } 1032 1033 list_add(&device->dev_list, &fs_devices->devices); 1034 device->fs_devices = fs_devices; 1035 fs_devices->num_devices++; 1036 } 1037 return fs_devices; 1038 error: 1039 free_fs_devices(fs_devices); 1040 return ERR_PTR(ret); 1041 } 1042 1043 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1044 struct btrfs_device **latest_dev) 1045 { 1046 struct btrfs_device *device, *next; 1047 1048 /* This is the initialized path, it is safe to release the devices. */ 1049 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1050 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1051 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1052 &device->dev_state) && 1053 !test_bit(BTRFS_DEV_STATE_MISSING, 1054 &device->dev_state) && 1055 (!*latest_dev || 1056 device->generation > (*latest_dev)->generation)) { 1057 *latest_dev = device; 1058 } 1059 continue; 1060 } 1061 1062 /* 1063 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1064 * in btrfs_init_dev_replace() so just continue. 1065 */ 1066 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1067 continue; 1068 1069 if (device->bdev) { 1070 blkdev_put(device->bdev, device->mode); 1071 device->bdev = NULL; 1072 fs_devices->open_devices--; 1073 } 1074 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1075 list_del_init(&device->dev_alloc_list); 1076 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1077 fs_devices->rw_devices--; 1078 } 1079 list_del_init(&device->dev_list); 1080 fs_devices->num_devices--; 1081 btrfs_free_device(device); 1082 } 1083 1084 } 1085 1086 /* 1087 * After we have read the system tree and know devids belonging to this 1088 * filesystem, remove the device which does not belong there. 1089 */ 1090 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1091 { 1092 struct btrfs_device *latest_dev = NULL; 1093 struct btrfs_fs_devices *seed_dev; 1094 1095 mutex_lock(&uuid_mutex); 1096 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1097 1098 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1099 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1100 1101 fs_devices->latest_dev = latest_dev; 1102 1103 mutex_unlock(&uuid_mutex); 1104 } 1105 1106 static void btrfs_close_bdev(struct btrfs_device *device) 1107 { 1108 if (!device->bdev) 1109 return; 1110 1111 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1112 sync_blockdev(device->bdev); 1113 invalidate_bdev(device->bdev); 1114 } 1115 1116 blkdev_put(device->bdev, device->mode); 1117 } 1118 1119 static void btrfs_close_one_device(struct btrfs_device *device) 1120 { 1121 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1122 1123 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1124 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1125 list_del_init(&device->dev_alloc_list); 1126 fs_devices->rw_devices--; 1127 } 1128 1129 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1130 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1131 1132 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 1133 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 1134 fs_devices->missing_devices--; 1135 } 1136 1137 btrfs_close_bdev(device); 1138 if (device->bdev) { 1139 fs_devices->open_devices--; 1140 device->bdev = NULL; 1141 } 1142 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1143 btrfs_destroy_dev_zone_info(device); 1144 1145 device->fs_info = NULL; 1146 atomic_set(&device->dev_stats_ccnt, 0); 1147 extent_io_tree_release(&device->alloc_state); 1148 1149 /* 1150 * Reset the flush error record. We might have a transient flush error 1151 * in this mount, and if so we aborted the current transaction and set 1152 * the fs to an error state, guaranteeing no super blocks can be further 1153 * committed. However that error might be transient and if we unmount the 1154 * filesystem and mount it again, we should allow the mount to succeed 1155 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1156 * filesystem again we still get flush errors, then we will again abort 1157 * any transaction and set the error state, guaranteeing no commits of 1158 * unsafe super blocks. 1159 */ 1160 device->last_flush_error = 0; 1161 1162 /* Verify the device is back in a pristine state */ 1163 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1164 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1165 WARN_ON(!list_empty(&device->dev_alloc_list)); 1166 WARN_ON(!list_empty(&device->post_commit_list)); 1167 } 1168 1169 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1170 { 1171 struct btrfs_device *device, *tmp; 1172 1173 lockdep_assert_held(&uuid_mutex); 1174 1175 if (--fs_devices->opened > 0) 1176 return; 1177 1178 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1179 btrfs_close_one_device(device); 1180 1181 WARN_ON(fs_devices->open_devices); 1182 WARN_ON(fs_devices->rw_devices); 1183 fs_devices->opened = 0; 1184 fs_devices->seeding = false; 1185 fs_devices->fs_info = NULL; 1186 } 1187 1188 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1189 { 1190 LIST_HEAD(list); 1191 struct btrfs_fs_devices *tmp; 1192 1193 mutex_lock(&uuid_mutex); 1194 close_fs_devices(fs_devices); 1195 if (!fs_devices->opened) { 1196 list_splice_init(&fs_devices->seed_list, &list); 1197 1198 /* 1199 * If the struct btrfs_fs_devices is not assembled with any 1200 * other device, it can be re-initialized during the next mount 1201 * without the needing device-scan step. Therefore, it can be 1202 * fully freed. 1203 */ 1204 if (fs_devices->num_devices == 1) { 1205 list_del(&fs_devices->fs_list); 1206 free_fs_devices(fs_devices); 1207 } 1208 } 1209 1210 1211 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1212 close_fs_devices(fs_devices); 1213 list_del(&fs_devices->seed_list); 1214 free_fs_devices(fs_devices); 1215 } 1216 mutex_unlock(&uuid_mutex); 1217 } 1218 1219 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1220 fmode_t flags, void *holder) 1221 { 1222 struct btrfs_device *device; 1223 struct btrfs_device *latest_dev = NULL; 1224 struct btrfs_device *tmp_device; 1225 1226 flags |= FMODE_EXCL; 1227 1228 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1229 dev_list) { 1230 int ret; 1231 1232 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1233 if (ret == 0 && 1234 (!latest_dev || device->generation > latest_dev->generation)) { 1235 latest_dev = device; 1236 } else if (ret == -ENODATA) { 1237 fs_devices->num_devices--; 1238 list_del(&device->dev_list); 1239 btrfs_free_device(device); 1240 } 1241 } 1242 if (fs_devices->open_devices == 0) 1243 return -EINVAL; 1244 1245 fs_devices->opened = 1; 1246 fs_devices->latest_dev = latest_dev; 1247 fs_devices->total_rw_bytes = 0; 1248 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1249 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1250 1251 return 0; 1252 } 1253 1254 static int devid_cmp(void *priv, const struct list_head *a, 1255 const struct list_head *b) 1256 { 1257 const struct btrfs_device *dev1, *dev2; 1258 1259 dev1 = list_entry(a, struct btrfs_device, dev_list); 1260 dev2 = list_entry(b, struct btrfs_device, dev_list); 1261 1262 if (dev1->devid < dev2->devid) 1263 return -1; 1264 else if (dev1->devid > dev2->devid) 1265 return 1; 1266 return 0; 1267 } 1268 1269 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1270 fmode_t flags, void *holder) 1271 { 1272 int ret; 1273 1274 lockdep_assert_held(&uuid_mutex); 1275 /* 1276 * The device_list_mutex cannot be taken here in case opening the 1277 * underlying device takes further locks like open_mutex. 1278 * 1279 * We also don't need the lock here as this is called during mount and 1280 * exclusion is provided by uuid_mutex 1281 */ 1282 1283 if (fs_devices->opened) { 1284 fs_devices->opened++; 1285 ret = 0; 1286 } else { 1287 list_sort(NULL, &fs_devices->devices, devid_cmp); 1288 ret = open_fs_devices(fs_devices, flags, holder); 1289 } 1290 1291 return ret; 1292 } 1293 1294 void btrfs_release_disk_super(struct btrfs_super_block *super) 1295 { 1296 struct page *page = virt_to_page(super); 1297 1298 put_page(page); 1299 } 1300 1301 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1302 u64 bytenr, u64 bytenr_orig) 1303 { 1304 struct btrfs_super_block *disk_super; 1305 struct page *page; 1306 void *p; 1307 pgoff_t index; 1308 1309 /* make sure our super fits in the device */ 1310 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) 1311 return ERR_PTR(-EINVAL); 1312 1313 /* make sure our super fits in the page */ 1314 if (sizeof(*disk_super) > PAGE_SIZE) 1315 return ERR_PTR(-EINVAL); 1316 1317 /* make sure our super doesn't straddle pages on disk */ 1318 index = bytenr >> PAGE_SHIFT; 1319 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1320 return ERR_PTR(-EINVAL); 1321 1322 /* pull in the page with our super */ 1323 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1324 1325 if (IS_ERR(page)) 1326 return ERR_CAST(page); 1327 1328 p = page_address(page); 1329 1330 /* align our pointer to the offset of the super block */ 1331 disk_super = p + offset_in_page(bytenr); 1332 1333 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1334 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1335 btrfs_release_disk_super(p); 1336 return ERR_PTR(-EINVAL); 1337 } 1338 1339 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1340 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1341 1342 return disk_super; 1343 } 1344 1345 int btrfs_forget_devices(dev_t devt) 1346 { 1347 int ret; 1348 1349 mutex_lock(&uuid_mutex); 1350 ret = btrfs_free_stale_devices(devt, NULL); 1351 mutex_unlock(&uuid_mutex); 1352 1353 return ret; 1354 } 1355 1356 /* 1357 * Look for a btrfs signature on a device. This may be called out of the mount path 1358 * and we are not allowed to call set_blocksize during the scan. The superblock 1359 * is read via pagecache 1360 */ 1361 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1362 void *holder) 1363 { 1364 struct btrfs_super_block *disk_super; 1365 bool new_device_added = false; 1366 struct btrfs_device *device = NULL; 1367 struct block_device *bdev; 1368 u64 bytenr, bytenr_orig; 1369 int ret; 1370 1371 lockdep_assert_held(&uuid_mutex); 1372 1373 /* 1374 * we would like to check all the supers, but that would make 1375 * a btrfs mount succeed after a mkfs from a different FS. 1376 * So, we need to add a special mount option to scan for 1377 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1378 */ 1379 1380 /* 1381 * Avoid using flag |= FMODE_EXCL here, as the systemd-udev may 1382 * initiate the device scan which may race with the user's mount 1383 * or mkfs command, resulting in failure. 1384 * Since the device scan is solely for reading purposes, there is 1385 * no need for FMODE_EXCL. Additionally, the devices are read again 1386 * during the mount process. It is ok to get some inconsistent 1387 * values temporarily, as the device paths of the fsid are the only 1388 * required information for assembling the volume. 1389 */ 1390 bdev = blkdev_get_by_path(path, flags, holder); 1391 if (IS_ERR(bdev)) 1392 return ERR_CAST(bdev); 1393 1394 bytenr_orig = btrfs_sb_offset(0); 1395 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1396 if (ret) { 1397 device = ERR_PTR(ret); 1398 goto error_bdev_put; 1399 } 1400 1401 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1402 if (IS_ERR(disk_super)) { 1403 device = ERR_CAST(disk_super); 1404 goto error_bdev_put; 1405 } 1406 1407 device = device_list_add(path, disk_super, &new_device_added); 1408 if (!IS_ERR(device) && new_device_added) 1409 btrfs_free_stale_devices(device->devt, device); 1410 1411 btrfs_release_disk_super(disk_super); 1412 1413 error_bdev_put: 1414 blkdev_put(bdev, flags); 1415 1416 return device; 1417 } 1418 1419 /* 1420 * Try to find a chunk that intersects [start, start + len] range and when one 1421 * such is found, record the end of it in *start 1422 */ 1423 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1424 u64 len) 1425 { 1426 u64 physical_start, physical_end; 1427 1428 lockdep_assert_held(&device->fs_info->chunk_mutex); 1429 1430 if (!find_first_extent_bit(&device->alloc_state, *start, 1431 &physical_start, &physical_end, 1432 CHUNK_ALLOCATED, NULL)) { 1433 1434 if (in_range(physical_start, *start, len) || 1435 in_range(*start, physical_start, 1436 physical_end - physical_start)) { 1437 *start = physical_end + 1; 1438 return true; 1439 } 1440 } 1441 return false; 1442 } 1443 1444 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1445 { 1446 switch (device->fs_devices->chunk_alloc_policy) { 1447 case BTRFS_CHUNK_ALLOC_REGULAR: 1448 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); 1449 case BTRFS_CHUNK_ALLOC_ZONED: 1450 /* 1451 * We don't care about the starting region like regular 1452 * allocator, because we anyway use/reserve the first two zones 1453 * for superblock logging. 1454 */ 1455 return ALIGN(start, device->zone_info->zone_size); 1456 default: 1457 BUG(); 1458 } 1459 } 1460 1461 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1462 u64 *hole_start, u64 *hole_size, 1463 u64 num_bytes) 1464 { 1465 u64 zone_size = device->zone_info->zone_size; 1466 u64 pos; 1467 int ret; 1468 bool changed = false; 1469 1470 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1471 1472 while (*hole_size > 0) { 1473 pos = btrfs_find_allocatable_zones(device, *hole_start, 1474 *hole_start + *hole_size, 1475 num_bytes); 1476 if (pos != *hole_start) { 1477 *hole_size = *hole_start + *hole_size - pos; 1478 *hole_start = pos; 1479 changed = true; 1480 if (*hole_size < num_bytes) 1481 break; 1482 } 1483 1484 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1485 1486 /* Range is ensured to be empty */ 1487 if (!ret) 1488 return changed; 1489 1490 /* Given hole range was invalid (outside of device) */ 1491 if (ret == -ERANGE) { 1492 *hole_start += *hole_size; 1493 *hole_size = 0; 1494 return true; 1495 } 1496 1497 *hole_start += zone_size; 1498 *hole_size -= zone_size; 1499 changed = true; 1500 } 1501 1502 return changed; 1503 } 1504 1505 /* 1506 * Check if specified hole is suitable for allocation. 1507 * 1508 * @device: the device which we have the hole 1509 * @hole_start: starting position of the hole 1510 * @hole_size: the size of the hole 1511 * @num_bytes: the size of the free space that we need 1512 * 1513 * This function may modify @hole_start and @hole_size to reflect the suitable 1514 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1515 */ 1516 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1517 u64 *hole_size, u64 num_bytes) 1518 { 1519 bool changed = false; 1520 u64 hole_end = *hole_start + *hole_size; 1521 1522 for (;;) { 1523 /* 1524 * Check before we set max_hole_start, otherwise we could end up 1525 * sending back this offset anyway. 1526 */ 1527 if (contains_pending_extent(device, hole_start, *hole_size)) { 1528 if (hole_end >= *hole_start) 1529 *hole_size = hole_end - *hole_start; 1530 else 1531 *hole_size = 0; 1532 changed = true; 1533 } 1534 1535 switch (device->fs_devices->chunk_alloc_policy) { 1536 case BTRFS_CHUNK_ALLOC_REGULAR: 1537 /* No extra check */ 1538 break; 1539 case BTRFS_CHUNK_ALLOC_ZONED: 1540 if (dev_extent_hole_check_zoned(device, hole_start, 1541 hole_size, num_bytes)) { 1542 changed = true; 1543 /* 1544 * The changed hole can contain pending extent. 1545 * Loop again to check that. 1546 */ 1547 continue; 1548 } 1549 break; 1550 default: 1551 BUG(); 1552 } 1553 1554 break; 1555 } 1556 1557 return changed; 1558 } 1559 1560 /* 1561 * Find free space in the specified device. 1562 * 1563 * @device: the device which we search the free space in 1564 * @num_bytes: the size of the free space that we need 1565 * @search_start: the position from which to begin the search 1566 * @start: store the start of the free space. 1567 * @len: the size of the free space. that we find, or the size 1568 * of the max free space if we don't find suitable free space 1569 * 1570 * This does a pretty simple search, the expectation is that it is called very 1571 * infrequently and that a given device has a small number of extents. 1572 * 1573 * @start is used to store the start of the free space if we find. But if we 1574 * don't find suitable free space, it will be used to store the start position 1575 * of the max free space. 1576 * 1577 * @len is used to store the size of the free space that we find. 1578 * But if we don't find suitable free space, it is used to store the size of 1579 * the max free space. 1580 * 1581 * NOTE: This function will search *commit* root of device tree, and does extra 1582 * check to ensure dev extents are not double allocated. 1583 * This makes the function safe to allocate dev extents but may not report 1584 * correct usable device space, as device extent freed in current transaction 1585 * is not reported as available. 1586 */ 1587 static int find_free_dev_extent_start(struct btrfs_device *device, 1588 u64 num_bytes, u64 search_start, u64 *start, 1589 u64 *len) 1590 { 1591 struct btrfs_fs_info *fs_info = device->fs_info; 1592 struct btrfs_root *root = fs_info->dev_root; 1593 struct btrfs_key key; 1594 struct btrfs_dev_extent *dev_extent; 1595 struct btrfs_path *path; 1596 u64 hole_size; 1597 u64 max_hole_start; 1598 u64 max_hole_size; 1599 u64 extent_end; 1600 u64 search_end = device->total_bytes; 1601 int ret; 1602 int slot; 1603 struct extent_buffer *l; 1604 1605 search_start = dev_extent_search_start(device, search_start); 1606 1607 WARN_ON(device->zone_info && 1608 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1609 1610 path = btrfs_alloc_path(); 1611 if (!path) 1612 return -ENOMEM; 1613 1614 max_hole_start = search_start; 1615 max_hole_size = 0; 1616 1617 again: 1618 if (search_start >= search_end || 1619 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1620 ret = -ENOSPC; 1621 goto out; 1622 } 1623 1624 path->reada = READA_FORWARD; 1625 path->search_commit_root = 1; 1626 path->skip_locking = 1; 1627 1628 key.objectid = device->devid; 1629 key.offset = search_start; 1630 key.type = BTRFS_DEV_EXTENT_KEY; 1631 1632 ret = btrfs_search_backwards(root, &key, path); 1633 if (ret < 0) 1634 goto out; 1635 1636 while (search_start < search_end) { 1637 l = path->nodes[0]; 1638 slot = path->slots[0]; 1639 if (slot >= btrfs_header_nritems(l)) { 1640 ret = btrfs_next_leaf(root, path); 1641 if (ret == 0) 1642 continue; 1643 if (ret < 0) 1644 goto out; 1645 1646 break; 1647 } 1648 btrfs_item_key_to_cpu(l, &key, slot); 1649 1650 if (key.objectid < device->devid) 1651 goto next; 1652 1653 if (key.objectid > device->devid) 1654 break; 1655 1656 if (key.type != BTRFS_DEV_EXTENT_KEY) 1657 goto next; 1658 1659 if (key.offset > search_end) 1660 break; 1661 1662 if (key.offset > search_start) { 1663 hole_size = key.offset - search_start; 1664 dev_extent_hole_check(device, &search_start, &hole_size, 1665 num_bytes); 1666 1667 if (hole_size > max_hole_size) { 1668 max_hole_start = search_start; 1669 max_hole_size = hole_size; 1670 } 1671 1672 /* 1673 * If this free space is greater than which we need, 1674 * it must be the max free space that we have found 1675 * until now, so max_hole_start must point to the start 1676 * of this free space and the length of this free space 1677 * is stored in max_hole_size. Thus, we return 1678 * max_hole_start and max_hole_size and go back to the 1679 * caller. 1680 */ 1681 if (hole_size >= num_bytes) { 1682 ret = 0; 1683 goto out; 1684 } 1685 } 1686 1687 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1688 extent_end = key.offset + btrfs_dev_extent_length(l, 1689 dev_extent); 1690 if (extent_end > search_start) 1691 search_start = extent_end; 1692 next: 1693 path->slots[0]++; 1694 cond_resched(); 1695 } 1696 1697 /* 1698 * At this point, search_start should be the end of 1699 * allocated dev extents, and when shrinking the device, 1700 * search_end may be smaller than search_start. 1701 */ 1702 if (search_end > search_start) { 1703 hole_size = search_end - search_start; 1704 if (dev_extent_hole_check(device, &search_start, &hole_size, 1705 num_bytes)) { 1706 btrfs_release_path(path); 1707 goto again; 1708 } 1709 1710 if (hole_size > max_hole_size) { 1711 max_hole_start = search_start; 1712 max_hole_size = hole_size; 1713 } 1714 } 1715 1716 /* See above. */ 1717 if (max_hole_size < num_bytes) 1718 ret = -ENOSPC; 1719 else 1720 ret = 0; 1721 1722 ASSERT(max_hole_start + max_hole_size <= search_end); 1723 out: 1724 btrfs_free_path(path); 1725 *start = max_hole_start; 1726 if (len) 1727 *len = max_hole_size; 1728 return ret; 1729 } 1730 1731 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1732 u64 *start, u64 *len) 1733 { 1734 /* FIXME use last free of some kind */ 1735 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1736 } 1737 1738 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1739 struct btrfs_device *device, 1740 u64 start, u64 *dev_extent_len) 1741 { 1742 struct btrfs_fs_info *fs_info = device->fs_info; 1743 struct btrfs_root *root = fs_info->dev_root; 1744 int ret; 1745 struct btrfs_path *path; 1746 struct btrfs_key key; 1747 struct btrfs_key found_key; 1748 struct extent_buffer *leaf = NULL; 1749 struct btrfs_dev_extent *extent = NULL; 1750 1751 path = btrfs_alloc_path(); 1752 if (!path) 1753 return -ENOMEM; 1754 1755 key.objectid = device->devid; 1756 key.offset = start; 1757 key.type = BTRFS_DEV_EXTENT_KEY; 1758 again: 1759 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1760 if (ret > 0) { 1761 ret = btrfs_previous_item(root, path, key.objectid, 1762 BTRFS_DEV_EXTENT_KEY); 1763 if (ret) 1764 goto out; 1765 leaf = path->nodes[0]; 1766 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1767 extent = btrfs_item_ptr(leaf, path->slots[0], 1768 struct btrfs_dev_extent); 1769 BUG_ON(found_key.offset > start || found_key.offset + 1770 btrfs_dev_extent_length(leaf, extent) < start); 1771 key = found_key; 1772 btrfs_release_path(path); 1773 goto again; 1774 } else if (ret == 0) { 1775 leaf = path->nodes[0]; 1776 extent = btrfs_item_ptr(leaf, path->slots[0], 1777 struct btrfs_dev_extent); 1778 } else { 1779 goto out; 1780 } 1781 1782 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1783 1784 ret = btrfs_del_item(trans, root, path); 1785 if (ret == 0) 1786 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1787 out: 1788 btrfs_free_path(path); 1789 return ret; 1790 } 1791 1792 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1793 { 1794 struct extent_map_tree *em_tree; 1795 struct extent_map *em; 1796 struct rb_node *n; 1797 u64 ret = 0; 1798 1799 em_tree = &fs_info->mapping_tree; 1800 read_lock(&em_tree->lock); 1801 n = rb_last(&em_tree->map.rb_root); 1802 if (n) { 1803 em = rb_entry(n, struct extent_map, rb_node); 1804 ret = em->start + em->len; 1805 } 1806 read_unlock(&em_tree->lock); 1807 1808 return ret; 1809 } 1810 1811 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1812 u64 *devid_ret) 1813 { 1814 int ret; 1815 struct btrfs_key key; 1816 struct btrfs_key found_key; 1817 struct btrfs_path *path; 1818 1819 path = btrfs_alloc_path(); 1820 if (!path) 1821 return -ENOMEM; 1822 1823 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1824 key.type = BTRFS_DEV_ITEM_KEY; 1825 key.offset = (u64)-1; 1826 1827 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1828 if (ret < 0) 1829 goto error; 1830 1831 if (ret == 0) { 1832 /* Corruption */ 1833 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1834 ret = -EUCLEAN; 1835 goto error; 1836 } 1837 1838 ret = btrfs_previous_item(fs_info->chunk_root, path, 1839 BTRFS_DEV_ITEMS_OBJECTID, 1840 BTRFS_DEV_ITEM_KEY); 1841 if (ret) { 1842 *devid_ret = 1; 1843 } else { 1844 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1845 path->slots[0]); 1846 *devid_ret = found_key.offset + 1; 1847 } 1848 ret = 0; 1849 error: 1850 btrfs_free_path(path); 1851 return ret; 1852 } 1853 1854 /* 1855 * the device information is stored in the chunk root 1856 * the btrfs_device struct should be fully filled in 1857 */ 1858 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1859 struct btrfs_device *device) 1860 { 1861 int ret; 1862 struct btrfs_path *path; 1863 struct btrfs_dev_item *dev_item; 1864 struct extent_buffer *leaf; 1865 struct btrfs_key key; 1866 unsigned long ptr; 1867 1868 path = btrfs_alloc_path(); 1869 if (!path) 1870 return -ENOMEM; 1871 1872 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1873 key.type = BTRFS_DEV_ITEM_KEY; 1874 key.offset = device->devid; 1875 1876 btrfs_reserve_chunk_metadata(trans, true); 1877 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1878 &key, sizeof(*dev_item)); 1879 btrfs_trans_release_chunk_metadata(trans); 1880 if (ret) 1881 goto out; 1882 1883 leaf = path->nodes[0]; 1884 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1885 1886 btrfs_set_device_id(leaf, dev_item, device->devid); 1887 btrfs_set_device_generation(leaf, dev_item, 0); 1888 btrfs_set_device_type(leaf, dev_item, device->type); 1889 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1890 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1891 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1892 btrfs_set_device_total_bytes(leaf, dev_item, 1893 btrfs_device_get_disk_total_bytes(device)); 1894 btrfs_set_device_bytes_used(leaf, dev_item, 1895 btrfs_device_get_bytes_used(device)); 1896 btrfs_set_device_group(leaf, dev_item, 0); 1897 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1898 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1899 btrfs_set_device_start_offset(leaf, dev_item, 0); 1900 1901 ptr = btrfs_device_uuid(dev_item); 1902 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1903 ptr = btrfs_device_fsid(dev_item); 1904 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1905 ptr, BTRFS_FSID_SIZE); 1906 btrfs_mark_buffer_dirty(leaf); 1907 1908 ret = 0; 1909 out: 1910 btrfs_free_path(path); 1911 return ret; 1912 } 1913 1914 /* 1915 * Function to update ctime/mtime for a given device path. 1916 * Mainly used for ctime/mtime based probe like libblkid. 1917 * 1918 * We don't care about errors here, this is just to be kind to userspace. 1919 */ 1920 static void update_dev_time(const char *device_path) 1921 { 1922 struct path path; 1923 struct timespec64 now; 1924 int ret; 1925 1926 ret = kern_path(device_path, LOOKUP_FOLLOW, &path); 1927 if (ret) 1928 return; 1929 1930 now = current_time(d_inode(path.dentry)); 1931 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); 1932 path_put(&path); 1933 } 1934 1935 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, 1936 struct btrfs_device *device) 1937 { 1938 struct btrfs_root *root = device->fs_info->chunk_root; 1939 int ret; 1940 struct btrfs_path *path; 1941 struct btrfs_key key; 1942 1943 path = btrfs_alloc_path(); 1944 if (!path) 1945 return -ENOMEM; 1946 1947 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1948 key.type = BTRFS_DEV_ITEM_KEY; 1949 key.offset = device->devid; 1950 1951 btrfs_reserve_chunk_metadata(trans, false); 1952 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1953 btrfs_trans_release_chunk_metadata(trans); 1954 if (ret) { 1955 if (ret > 0) 1956 ret = -ENOENT; 1957 goto out; 1958 } 1959 1960 ret = btrfs_del_item(trans, root, path); 1961 out: 1962 btrfs_free_path(path); 1963 return ret; 1964 } 1965 1966 /* 1967 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1968 * filesystem. It's up to the caller to adjust that number regarding eg. device 1969 * replace. 1970 */ 1971 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1972 u64 num_devices) 1973 { 1974 u64 all_avail; 1975 unsigned seq; 1976 int i; 1977 1978 do { 1979 seq = read_seqbegin(&fs_info->profiles_lock); 1980 1981 all_avail = fs_info->avail_data_alloc_bits | 1982 fs_info->avail_system_alloc_bits | 1983 fs_info->avail_metadata_alloc_bits; 1984 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1985 1986 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1987 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1988 continue; 1989 1990 if (num_devices < btrfs_raid_array[i].devs_min) 1991 return btrfs_raid_array[i].mindev_error; 1992 } 1993 1994 return 0; 1995 } 1996 1997 static struct btrfs_device * btrfs_find_next_active_device( 1998 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1999 { 2000 struct btrfs_device *next_device; 2001 2002 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 2003 if (next_device != device && 2004 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 2005 && next_device->bdev) 2006 return next_device; 2007 } 2008 2009 return NULL; 2010 } 2011 2012 /* 2013 * Helper function to check if the given device is part of s_bdev / latest_dev 2014 * and replace it with the provided or the next active device, in the context 2015 * where this function called, there should be always be another device (or 2016 * this_dev) which is active. 2017 */ 2018 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 2019 struct btrfs_device *next_device) 2020 { 2021 struct btrfs_fs_info *fs_info = device->fs_info; 2022 2023 if (!next_device) 2024 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 2025 device); 2026 ASSERT(next_device); 2027 2028 if (fs_info->sb->s_bdev && 2029 (fs_info->sb->s_bdev == device->bdev)) 2030 fs_info->sb->s_bdev = next_device->bdev; 2031 2032 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2033 fs_info->fs_devices->latest_dev = next_device; 2034 } 2035 2036 /* 2037 * Return btrfs_fs_devices::num_devices excluding the device that's being 2038 * currently replaced. 2039 */ 2040 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2041 { 2042 u64 num_devices = fs_info->fs_devices->num_devices; 2043 2044 down_read(&fs_info->dev_replace.rwsem); 2045 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2046 ASSERT(num_devices > 1); 2047 num_devices--; 2048 } 2049 up_read(&fs_info->dev_replace.rwsem); 2050 2051 return num_devices; 2052 } 2053 2054 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info, 2055 struct block_device *bdev, int copy_num) 2056 { 2057 struct btrfs_super_block *disk_super; 2058 const size_t len = sizeof(disk_super->magic); 2059 const u64 bytenr = btrfs_sb_offset(copy_num); 2060 int ret; 2061 2062 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr); 2063 if (IS_ERR(disk_super)) 2064 return; 2065 2066 memset(&disk_super->magic, 0, len); 2067 folio_mark_dirty(virt_to_folio(disk_super)); 2068 btrfs_release_disk_super(disk_super); 2069 2070 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1); 2071 if (ret) 2072 btrfs_warn(fs_info, "error clearing superblock number %d (%d)", 2073 copy_num, ret); 2074 } 2075 2076 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2077 struct block_device *bdev, 2078 const char *device_path) 2079 { 2080 int copy_num; 2081 2082 if (!bdev) 2083 return; 2084 2085 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2086 if (bdev_is_zoned(bdev)) 2087 btrfs_reset_sb_log_zones(bdev, copy_num); 2088 else 2089 btrfs_scratch_superblock(fs_info, bdev, copy_num); 2090 } 2091 2092 /* Notify udev that device has changed */ 2093 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2094 2095 /* Update ctime/mtime for device path for libblkid */ 2096 update_dev_time(device_path); 2097 } 2098 2099 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2100 struct btrfs_dev_lookup_args *args, 2101 struct block_device **bdev, fmode_t *mode) 2102 { 2103 struct btrfs_trans_handle *trans; 2104 struct btrfs_device *device; 2105 struct btrfs_fs_devices *cur_devices; 2106 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2107 u64 num_devices; 2108 int ret = 0; 2109 2110 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2111 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); 2112 return -EINVAL; 2113 } 2114 2115 /* 2116 * The device list in fs_devices is accessed without locks (neither 2117 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2118 * filesystem and another device rm cannot run. 2119 */ 2120 num_devices = btrfs_num_devices(fs_info); 2121 2122 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2123 if (ret) 2124 return ret; 2125 2126 device = btrfs_find_device(fs_info->fs_devices, args); 2127 if (!device) { 2128 if (args->missing) 2129 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2130 else 2131 ret = -ENOENT; 2132 return ret; 2133 } 2134 2135 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2136 btrfs_warn_in_rcu(fs_info, 2137 "cannot remove device %s (devid %llu) due to active swapfile", 2138 btrfs_dev_name(device), device->devid); 2139 return -ETXTBSY; 2140 } 2141 2142 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 2143 return BTRFS_ERROR_DEV_TGT_REPLACE; 2144 2145 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2146 fs_info->fs_devices->rw_devices == 1) 2147 return BTRFS_ERROR_DEV_ONLY_WRITABLE; 2148 2149 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2150 mutex_lock(&fs_info->chunk_mutex); 2151 list_del_init(&device->dev_alloc_list); 2152 device->fs_devices->rw_devices--; 2153 mutex_unlock(&fs_info->chunk_mutex); 2154 } 2155 2156 ret = btrfs_shrink_device(device, 0); 2157 if (ret) 2158 goto error_undo; 2159 2160 trans = btrfs_start_transaction(fs_info->chunk_root, 0); 2161 if (IS_ERR(trans)) { 2162 ret = PTR_ERR(trans); 2163 goto error_undo; 2164 } 2165 2166 ret = btrfs_rm_dev_item(trans, device); 2167 if (ret) { 2168 /* Any error in dev item removal is critical */ 2169 btrfs_crit(fs_info, 2170 "failed to remove device item for devid %llu: %d", 2171 device->devid, ret); 2172 btrfs_abort_transaction(trans, ret); 2173 btrfs_end_transaction(trans); 2174 return ret; 2175 } 2176 2177 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2178 btrfs_scrub_cancel_dev(device); 2179 2180 /* 2181 * the device list mutex makes sure that we don't change 2182 * the device list while someone else is writing out all 2183 * the device supers. Whoever is writing all supers, should 2184 * lock the device list mutex before getting the number of 2185 * devices in the super block (super_copy). Conversely, 2186 * whoever updates the number of devices in the super block 2187 * (super_copy) should hold the device list mutex. 2188 */ 2189 2190 /* 2191 * In normal cases the cur_devices == fs_devices. But in case 2192 * of deleting a seed device, the cur_devices should point to 2193 * its own fs_devices listed under the fs_devices->seed_list. 2194 */ 2195 cur_devices = device->fs_devices; 2196 mutex_lock(&fs_devices->device_list_mutex); 2197 list_del_rcu(&device->dev_list); 2198 2199 cur_devices->num_devices--; 2200 cur_devices->total_devices--; 2201 /* Update total_devices of the parent fs_devices if it's seed */ 2202 if (cur_devices != fs_devices) 2203 fs_devices->total_devices--; 2204 2205 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2206 cur_devices->missing_devices--; 2207 2208 btrfs_assign_next_active_device(device, NULL); 2209 2210 if (device->bdev) { 2211 cur_devices->open_devices--; 2212 /* remove sysfs entry */ 2213 btrfs_sysfs_remove_device(device); 2214 } 2215 2216 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2217 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2218 mutex_unlock(&fs_devices->device_list_mutex); 2219 2220 /* 2221 * At this point, the device is zero sized and detached from the 2222 * devices list. All that's left is to zero out the old supers and 2223 * free the device. 2224 * 2225 * We cannot call btrfs_close_bdev() here because we're holding the sb 2226 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2227 * block device and it's dependencies. Instead just flush the device 2228 * and let the caller do the final blkdev_put. 2229 */ 2230 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2231 btrfs_scratch_superblocks(fs_info, device->bdev, 2232 device->name->str); 2233 if (device->bdev) { 2234 sync_blockdev(device->bdev); 2235 invalidate_bdev(device->bdev); 2236 } 2237 } 2238 2239 *bdev = device->bdev; 2240 *mode = device->mode; 2241 synchronize_rcu(); 2242 btrfs_free_device(device); 2243 2244 /* 2245 * This can happen if cur_devices is the private seed devices list. We 2246 * cannot call close_fs_devices() here because it expects the uuid_mutex 2247 * to be held, but in fact we don't need that for the private 2248 * seed_devices, we can simply decrement cur_devices->opened and then 2249 * remove it from our list and free the fs_devices. 2250 */ 2251 if (cur_devices->num_devices == 0) { 2252 list_del_init(&cur_devices->seed_list); 2253 ASSERT(cur_devices->opened == 1); 2254 cur_devices->opened--; 2255 free_fs_devices(cur_devices); 2256 } 2257 2258 ret = btrfs_commit_transaction(trans); 2259 2260 return ret; 2261 2262 error_undo: 2263 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2264 mutex_lock(&fs_info->chunk_mutex); 2265 list_add(&device->dev_alloc_list, 2266 &fs_devices->alloc_list); 2267 device->fs_devices->rw_devices++; 2268 mutex_unlock(&fs_info->chunk_mutex); 2269 } 2270 return ret; 2271 } 2272 2273 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2274 { 2275 struct btrfs_fs_devices *fs_devices; 2276 2277 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2278 2279 /* 2280 * in case of fs with no seed, srcdev->fs_devices will point 2281 * to fs_devices of fs_info. However when the dev being replaced is 2282 * a seed dev it will point to the seed's local fs_devices. In short 2283 * srcdev will have its correct fs_devices in both the cases. 2284 */ 2285 fs_devices = srcdev->fs_devices; 2286 2287 list_del_rcu(&srcdev->dev_list); 2288 list_del(&srcdev->dev_alloc_list); 2289 fs_devices->num_devices--; 2290 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2291 fs_devices->missing_devices--; 2292 2293 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2294 fs_devices->rw_devices--; 2295 2296 if (srcdev->bdev) 2297 fs_devices->open_devices--; 2298 } 2299 2300 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2301 { 2302 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2303 2304 mutex_lock(&uuid_mutex); 2305 2306 btrfs_close_bdev(srcdev); 2307 synchronize_rcu(); 2308 btrfs_free_device(srcdev); 2309 2310 /* if this is no devs we rather delete the fs_devices */ 2311 if (!fs_devices->num_devices) { 2312 /* 2313 * On a mounted FS, num_devices can't be zero unless it's a 2314 * seed. In case of a seed device being replaced, the replace 2315 * target added to the sprout FS, so there will be no more 2316 * device left under the seed FS. 2317 */ 2318 ASSERT(fs_devices->seeding); 2319 2320 list_del_init(&fs_devices->seed_list); 2321 close_fs_devices(fs_devices); 2322 free_fs_devices(fs_devices); 2323 } 2324 mutex_unlock(&uuid_mutex); 2325 } 2326 2327 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2328 { 2329 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2330 2331 mutex_lock(&fs_devices->device_list_mutex); 2332 2333 btrfs_sysfs_remove_device(tgtdev); 2334 2335 if (tgtdev->bdev) 2336 fs_devices->open_devices--; 2337 2338 fs_devices->num_devices--; 2339 2340 btrfs_assign_next_active_device(tgtdev, NULL); 2341 2342 list_del_rcu(&tgtdev->dev_list); 2343 2344 mutex_unlock(&fs_devices->device_list_mutex); 2345 2346 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2347 tgtdev->name->str); 2348 2349 btrfs_close_bdev(tgtdev); 2350 synchronize_rcu(); 2351 btrfs_free_device(tgtdev); 2352 } 2353 2354 /* 2355 * Populate args from device at path. 2356 * 2357 * @fs_info: the filesystem 2358 * @args: the args to populate 2359 * @path: the path to the device 2360 * 2361 * This will read the super block of the device at @path and populate @args with 2362 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2363 * lookup a device to operate on, but need to do it before we take any locks. 2364 * This properly handles the special case of "missing" that a user may pass in, 2365 * and does some basic sanity checks. The caller must make sure that @path is 2366 * properly NUL terminated before calling in, and must call 2367 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2368 * uuid buffers. 2369 * 2370 * Return: 0 for success, -errno for failure 2371 */ 2372 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2373 struct btrfs_dev_lookup_args *args, 2374 const char *path) 2375 { 2376 struct btrfs_super_block *disk_super; 2377 struct block_device *bdev; 2378 int ret; 2379 2380 if (!path || !path[0]) 2381 return -EINVAL; 2382 if (!strcmp(path, "missing")) { 2383 args->missing = true; 2384 return 0; 2385 } 2386 2387 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2388 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2389 if (!args->uuid || !args->fsid) { 2390 btrfs_put_dev_args_from_path(args); 2391 return -ENOMEM; 2392 } 2393 2394 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2395 &bdev, &disk_super); 2396 if (ret) { 2397 btrfs_put_dev_args_from_path(args); 2398 return ret; 2399 } 2400 2401 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2402 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2403 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2404 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2405 else 2406 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2407 btrfs_release_disk_super(disk_super); 2408 blkdev_put(bdev, FMODE_READ); 2409 return 0; 2410 } 2411 2412 /* 2413 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2414 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2415 * that don't need to be freed. 2416 */ 2417 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2418 { 2419 kfree(args->uuid); 2420 kfree(args->fsid); 2421 args->uuid = NULL; 2422 args->fsid = NULL; 2423 } 2424 2425 struct btrfs_device *btrfs_find_device_by_devspec( 2426 struct btrfs_fs_info *fs_info, u64 devid, 2427 const char *device_path) 2428 { 2429 BTRFS_DEV_LOOKUP_ARGS(args); 2430 struct btrfs_device *device; 2431 int ret; 2432 2433 if (devid) { 2434 args.devid = devid; 2435 device = btrfs_find_device(fs_info->fs_devices, &args); 2436 if (!device) 2437 return ERR_PTR(-ENOENT); 2438 return device; 2439 } 2440 2441 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2442 if (ret) 2443 return ERR_PTR(ret); 2444 device = btrfs_find_device(fs_info->fs_devices, &args); 2445 btrfs_put_dev_args_from_path(&args); 2446 if (!device) 2447 return ERR_PTR(-ENOENT); 2448 return device; 2449 } 2450 2451 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) 2452 { 2453 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2454 struct btrfs_fs_devices *old_devices; 2455 struct btrfs_fs_devices *seed_devices; 2456 2457 lockdep_assert_held(&uuid_mutex); 2458 if (!fs_devices->seeding) 2459 return ERR_PTR(-EINVAL); 2460 2461 /* 2462 * Private copy of the seed devices, anchored at 2463 * fs_info->fs_devices->seed_list 2464 */ 2465 seed_devices = alloc_fs_devices(NULL, NULL); 2466 if (IS_ERR(seed_devices)) 2467 return seed_devices; 2468 2469 /* 2470 * It's necessary to retain a copy of the original seed fs_devices in 2471 * fs_uuids so that filesystems which have been seeded can successfully 2472 * reference the seed device from open_seed_devices. This also supports 2473 * multiple fs seed. 2474 */ 2475 old_devices = clone_fs_devices(fs_devices); 2476 if (IS_ERR(old_devices)) { 2477 kfree(seed_devices); 2478 return old_devices; 2479 } 2480 2481 list_add(&old_devices->fs_list, &fs_uuids); 2482 2483 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2484 seed_devices->opened = 1; 2485 INIT_LIST_HEAD(&seed_devices->devices); 2486 INIT_LIST_HEAD(&seed_devices->alloc_list); 2487 mutex_init(&seed_devices->device_list_mutex); 2488 2489 return seed_devices; 2490 } 2491 2492 /* 2493 * Splice seed devices into the sprout fs_devices. 2494 * Generate a new fsid for the sprouted read-write filesystem. 2495 */ 2496 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, 2497 struct btrfs_fs_devices *seed_devices) 2498 { 2499 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2500 struct btrfs_super_block *disk_super = fs_info->super_copy; 2501 struct btrfs_device *device; 2502 u64 super_flags; 2503 2504 /* 2505 * We are updating the fsid, the thread leading to device_list_add() 2506 * could race, so uuid_mutex is needed. 2507 */ 2508 lockdep_assert_held(&uuid_mutex); 2509 2510 /* 2511 * The threads listed below may traverse dev_list but can do that without 2512 * device_list_mutex: 2513 * - All device ops and balance - as we are in btrfs_exclop_start. 2514 * - Various dev_list readers - are using RCU. 2515 * - btrfs_ioctl_fitrim() - is using RCU. 2516 * 2517 * For-read threads as below are using device_list_mutex: 2518 * - Readonly scrub btrfs_scrub_dev() 2519 * - Readonly scrub btrfs_scrub_progress() 2520 * - btrfs_get_dev_stats() 2521 */ 2522 lockdep_assert_held(&fs_devices->device_list_mutex); 2523 2524 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2525 synchronize_rcu); 2526 list_for_each_entry(device, &seed_devices->devices, dev_list) 2527 device->fs_devices = seed_devices; 2528 2529 fs_devices->seeding = false; 2530 fs_devices->num_devices = 0; 2531 fs_devices->open_devices = 0; 2532 fs_devices->missing_devices = 0; 2533 fs_devices->rotating = false; 2534 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2535 2536 generate_random_uuid(fs_devices->fsid); 2537 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2538 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2539 2540 super_flags = btrfs_super_flags(disk_super) & 2541 ~BTRFS_SUPER_FLAG_SEEDING; 2542 btrfs_set_super_flags(disk_super, super_flags); 2543 } 2544 2545 /* 2546 * Store the expected generation for seed devices in device items. 2547 */ 2548 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2549 { 2550 BTRFS_DEV_LOOKUP_ARGS(args); 2551 struct btrfs_fs_info *fs_info = trans->fs_info; 2552 struct btrfs_root *root = fs_info->chunk_root; 2553 struct btrfs_path *path; 2554 struct extent_buffer *leaf; 2555 struct btrfs_dev_item *dev_item; 2556 struct btrfs_device *device; 2557 struct btrfs_key key; 2558 u8 fs_uuid[BTRFS_FSID_SIZE]; 2559 u8 dev_uuid[BTRFS_UUID_SIZE]; 2560 int ret; 2561 2562 path = btrfs_alloc_path(); 2563 if (!path) 2564 return -ENOMEM; 2565 2566 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2567 key.offset = 0; 2568 key.type = BTRFS_DEV_ITEM_KEY; 2569 2570 while (1) { 2571 btrfs_reserve_chunk_metadata(trans, false); 2572 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2573 btrfs_trans_release_chunk_metadata(trans); 2574 if (ret < 0) 2575 goto error; 2576 2577 leaf = path->nodes[0]; 2578 next_slot: 2579 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2580 ret = btrfs_next_leaf(root, path); 2581 if (ret > 0) 2582 break; 2583 if (ret < 0) 2584 goto error; 2585 leaf = path->nodes[0]; 2586 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2587 btrfs_release_path(path); 2588 continue; 2589 } 2590 2591 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2592 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2593 key.type != BTRFS_DEV_ITEM_KEY) 2594 break; 2595 2596 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2597 struct btrfs_dev_item); 2598 args.devid = btrfs_device_id(leaf, dev_item); 2599 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2600 BTRFS_UUID_SIZE); 2601 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2602 BTRFS_FSID_SIZE); 2603 args.uuid = dev_uuid; 2604 args.fsid = fs_uuid; 2605 device = btrfs_find_device(fs_info->fs_devices, &args); 2606 BUG_ON(!device); /* Logic error */ 2607 2608 if (device->fs_devices->seeding) { 2609 btrfs_set_device_generation(leaf, dev_item, 2610 device->generation); 2611 btrfs_mark_buffer_dirty(leaf); 2612 } 2613 2614 path->slots[0]++; 2615 goto next_slot; 2616 } 2617 ret = 0; 2618 error: 2619 btrfs_free_path(path); 2620 return ret; 2621 } 2622 2623 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2624 { 2625 struct btrfs_root *root = fs_info->dev_root; 2626 struct btrfs_trans_handle *trans; 2627 struct btrfs_device *device; 2628 struct block_device *bdev; 2629 struct super_block *sb = fs_info->sb; 2630 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2631 struct btrfs_fs_devices *seed_devices = NULL; 2632 u64 orig_super_total_bytes; 2633 u64 orig_super_num_devices; 2634 int ret = 0; 2635 bool seeding_dev = false; 2636 bool locked = false; 2637 2638 if (sb_rdonly(sb) && !fs_devices->seeding) 2639 return -EROFS; 2640 2641 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2642 fs_info->bdev_holder); 2643 if (IS_ERR(bdev)) 2644 return PTR_ERR(bdev); 2645 2646 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2647 ret = -EINVAL; 2648 goto error; 2649 } 2650 2651 if (fs_devices->seeding) { 2652 seeding_dev = true; 2653 down_write(&sb->s_umount); 2654 mutex_lock(&uuid_mutex); 2655 locked = true; 2656 } 2657 2658 sync_blockdev(bdev); 2659 2660 rcu_read_lock(); 2661 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2662 if (device->bdev == bdev) { 2663 ret = -EEXIST; 2664 rcu_read_unlock(); 2665 goto error; 2666 } 2667 } 2668 rcu_read_unlock(); 2669 2670 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path); 2671 if (IS_ERR(device)) { 2672 /* we can safely leave the fs_devices entry around */ 2673 ret = PTR_ERR(device); 2674 goto error; 2675 } 2676 2677 device->fs_info = fs_info; 2678 device->bdev = bdev; 2679 ret = lookup_bdev(device_path, &device->devt); 2680 if (ret) 2681 goto error_free_device; 2682 2683 ret = btrfs_get_dev_zone_info(device, false); 2684 if (ret) 2685 goto error_free_device; 2686 2687 trans = btrfs_start_transaction(root, 0); 2688 if (IS_ERR(trans)) { 2689 ret = PTR_ERR(trans); 2690 goto error_free_zone; 2691 } 2692 2693 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2694 device->generation = trans->transid; 2695 device->io_width = fs_info->sectorsize; 2696 device->io_align = fs_info->sectorsize; 2697 device->sector_size = fs_info->sectorsize; 2698 device->total_bytes = 2699 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); 2700 device->disk_total_bytes = device->total_bytes; 2701 device->commit_total_bytes = device->total_bytes; 2702 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2703 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2704 device->mode = FMODE_EXCL; 2705 device->dev_stats_valid = 1; 2706 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2707 2708 if (seeding_dev) { 2709 btrfs_clear_sb_rdonly(sb); 2710 2711 /* GFP_KERNEL allocation must not be under device_list_mutex */ 2712 seed_devices = btrfs_init_sprout(fs_info); 2713 if (IS_ERR(seed_devices)) { 2714 ret = PTR_ERR(seed_devices); 2715 btrfs_abort_transaction(trans, ret); 2716 goto error_trans; 2717 } 2718 } 2719 2720 mutex_lock(&fs_devices->device_list_mutex); 2721 if (seeding_dev) { 2722 btrfs_setup_sprout(fs_info, seed_devices); 2723 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2724 device); 2725 } 2726 2727 device->fs_devices = fs_devices; 2728 2729 mutex_lock(&fs_info->chunk_mutex); 2730 list_add_rcu(&device->dev_list, &fs_devices->devices); 2731 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2732 fs_devices->num_devices++; 2733 fs_devices->open_devices++; 2734 fs_devices->rw_devices++; 2735 fs_devices->total_devices++; 2736 fs_devices->total_rw_bytes += device->total_bytes; 2737 2738 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2739 2740 if (!bdev_nonrot(bdev)) 2741 fs_devices->rotating = true; 2742 2743 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2744 btrfs_set_super_total_bytes(fs_info->super_copy, 2745 round_down(orig_super_total_bytes + device->total_bytes, 2746 fs_info->sectorsize)); 2747 2748 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2749 btrfs_set_super_num_devices(fs_info->super_copy, 2750 orig_super_num_devices + 1); 2751 2752 /* 2753 * we've got more storage, clear any full flags on the space 2754 * infos 2755 */ 2756 btrfs_clear_space_info_full(fs_info); 2757 2758 mutex_unlock(&fs_info->chunk_mutex); 2759 2760 /* Add sysfs device entry */ 2761 btrfs_sysfs_add_device(device); 2762 2763 mutex_unlock(&fs_devices->device_list_mutex); 2764 2765 if (seeding_dev) { 2766 mutex_lock(&fs_info->chunk_mutex); 2767 ret = init_first_rw_device(trans); 2768 mutex_unlock(&fs_info->chunk_mutex); 2769 if (ret) { 2770 btrfs_abort_transaction(trans, ret); 2771 goto error_sysfs; 2772 } 2773 } 2774 2775 ret = btrfs_add_dev_item(trans, device); 2776 if (ret) { 2777 btrfs_abort_transaction(trans, ret); 2778 goto error_sysfs; 2779 } 2780 2781 if (seeding_dev) { 2782 ret = btrfs_finish_sprout(trans); 2783 if (ret) { 2784 btrfs_abort_transaction(trans, ret); 2785 goto error_sysfs; 2786 } 2787 2788 /* 2789 * fs_devices now represents the newly sprouted filesystem and 2790 * its fsid has been changed by btrfs_sprout_splice(). 2791 */ 2792 btrfs_sysfs_update_sprout_fsid(fs_devices); 2793 } 2794 2795 ret = btrfs_commit_transaction(trans); 2796 2797 if (seeding_dev) { 2798 mutex_unlock(&uuid_mutex); 2799 up_write(&sb->s_umount); 2800 locked = false; 2801 2802 if (ret) /* transaction commit */ 2803 return ret; 2804 2805 ret = btrfs_relocate_sys_chunks(fs_info); 2806 if (ret < 0) 2807 btrfs_handle_fs_error(fs_info, ret, 2808 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2809 trans = btrfs_attach_transaction(root); 2810 if (IS_ERR(trans)) { 2811 if (PTR_ERR(trans) == -ENOENT) 2812 return 0; 2813 ret = PTR_ERR(trans); 2814 trans = NULL; 2815 goto error_sysfs; 2816 } 2817 ret = btrfs_commit_transaction(trans); 2818 } 2819 2820 /* 2821 * Now that we have written a new super block to this device, check all 2822 * other fs_devices list if device_path alienates any other scanned 2823 * device. 2824 * We can ignore the return value as it typically returns -EINVAL and 2825 * only succeeds if the device was an alien. 2826 */ 2827 btrfs_forget_devices(device->devt); 2828 2829 /* Update ctime/mtime for blkid or udev */ 2830 update_dev_time(device_path); 2831 2832 return ret; 2833 2834 error_sysfs: 2835 btrfs_sysfs_remove_device(device); 2836 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2837 mutex_lock(&fs_info->chunk_mutex); 2838 list_del_rcu(&device->dev_list); 2839 list_del(&device->dev_alloc_list); 2840 fs_info->fs_devices->num_devices--; 2841 fs_info->fs_devices->open_devices--; 2842 fs_info->fs_devices->rw_devices--; 2843 fs_info->fs_devices->total_devices--; 2844 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2845 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2846 btrfs_set_super_total_bytes(fs_info->super_copy, 2847 orig_super_total_bytes); 2848 btrfs_set_super_num_devices(fs_info->super_copy, 2849 orig_super_num_devices); 2850 mutex_unlock(&fs_info->chunk_mutex); 2851 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2852 error_trans: 2853 if (seeding_dev) 2854 btrfs_set_sb_rdonly(sb); 2855 if (trans) 2856 btrfs_end_transaction(trans); 2857 error_free_zone: 2858 btrfs_destroy_dev_zone_info(device); 2859 error_free_device: 2860 btrfs_free_device(device); 2861 error: 2862 blkdev_put(bdev, FMODE_EXCL); 2863 if (locked) { 2864 mutex_unlock(&uuid_mutex); 2865 up_write(&sb->s_umount); 2866 } 2867 return ret; 2868 } 2869 2870 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2871 struct btrfs_device *device) 2872 { 2873 int ret; 2874 struct btrfs_path *path; 2875 struct btrfs_root *root = device->fs_info->chunk_root; 2876 struct btrfs_dev_item *dev_item; 2877 struct extent_buffer *leaf; 2878 struct btrfs_key key; 2879 2880 path = btrfs_alloc_path(); 2881 if (!path) 2882 return -ENOMEM; 2883 2884 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2885 key.type = BTRFS_DEV_ITEM_KEY; 2886 key.offset = device->devid; 2887 2888 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2889 if (ret < 0) 2890 goto out; 2891 2892 if (ret > 0) { 2893 ret = -ENOENT; 2894 goto out; 2895 } 2896 2897 leaf = path->nodes[0]; 2898 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2899 2900 btrfs_set_device_id(leaf, dev_item, device->devid); 2901 btrfs_set_device_type(leaf, dev_item, device->type); 2902 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2903 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2904 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2905 btrfs_set_device_total_bytes(leaf, dev_item, 2906 btrfs_device_get_disk_total_bytes(device)); 2907 btrfs_set_device_bytes_used(leaf, dev_item, 2908 btrfs_device_get_bytes_used(device)); 2909 btrfs_mark_buffer_dirty(leaf); 2910 2911 out: 2912 btrfs_free_path(path); 2913 return ret; 2914 } 2915 2916 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2917 struct btrfs_device *device, u64 new_size) 2918 { 2919 struct btrfs_fs_info *fs_info = device->fs_info; 2920 struct btrfs_super_block *super_copy = fs_info->super_copy; 2921 u64 old_total; 2922 u64 diff; 2923 int ret; 2924 2925 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2926 return -EACCES; 2927 2928 new_size = round_down(new_size, fs_info->sectorsize); 2929 2930 mutex_lock(&fs_info->chunk_mutex); 2931 old_total = btrfs_super_total_bytes(super_copy); 2932 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2933 2934 if (new_size <= device->total_bytes || 2935 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2936 mutex_unlock(&fs_info->chunk_mutex); 2937 return -EINVAL; 2938 } 2939 2940 btrfs_set_super_total_bytes(super_copy, 2941 round_down(old_total + diff, fs_info->sectorsize)); 2942 device->fs_devices->total_rw_bytes += diff; 2943 2944 btrfs_device_set_total_bytes(device, new_size); 2945 btrfs_device_set_disk_total_bytes(device, new_size); 2946 btrfs_clear_space_info_full(device->fs_info); 2947 if (list_empty(&device->post_commit_list)) 2948 list_add_tail(&device->post_commit_list, 2949 &trans->transaction->dev_update_list); 2950 mutex_unlock(&fs_info->chunk_mutex); 2951 2952 btrfs_reserve_chunk_metadata(trans, false); 2953 ret = btrfs_update_device(trans, device); 2954 btrfs_trans_release_chunk_metadata(trans); 2955 2956 return ret; 2957 } 2958 2959 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2960 { 2961 struct btrfs_fs_info *fs_info = trans->fs_info; 2962 struct btrfs_root *root = fs_info->chunk_root; 2963 int ret; 2964 struct btrfs_path *path; 2965 struct btrfs_key key; 2966 2967 path = btrfs_alloc_path(); 2968 if (!path) 2969 return -ENOMEM; 2970 2971 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2972 key.offset = chunk_offset; 2973 key.type = BTRFS_CHUNK_ITEM_KEY; 2974 2975 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2976 if (ret < 0) 2977 goto out; 2978 else if (ret > 0) { /* Logic error or corruption */ 2979 btrfs_handle_fs_error(fs_info, -ENOENT, 2980 "Failed lookup while freeing chunk."); 2981 ret = -ENOENT; 2982 goto out; 2983 } 2984 2985 ret = btrfs_del_item(trans, root, path); 2986 if (ret < 0) 2987 btrfs_handle_fs_error(fs_info, ret, 2988 "Failed to delete chunk item."); 2989 out: 2990 btrfs_free_path(path); 2991 return ret; 2992 } 2993 2994 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2995 { 2996 struct btrfs_super_block *super_copy = fs_info->super_copy; 2997 struct btrfs_disk_key *disk_key; 2998 struct btrfs_chunk *chunk; 2999 u8 *ptr; 3000 int ret = 0; 3001 u32 num_stripes; 3002 u32 array_size; 3003 u32 len = 0; 3004 u32 cur; 3005 struct btrfs_key key; 3006 3007 lockdep_assert_held(&fs_info->chunk_mutex); 3008 array_size = btrfs_super_sys_array_size(super_copy); 3009 3010 ptr = super_copy->sys_chunk_array; 3011 cur = 0; 3012 3013 while (cur < array_size) { 3014 disk_key = (struct btrfs_disk_key *)ptr; 3015 btrfs_disk_key_to_cpu(&key, disk_key); 3016 3017 len = sizeof(*disk_key); 3018 3019 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 3020 chunk = (struct btrfs_chunk *)(ptr + len); 3021 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 3022 len += btrfs_chunk_item_size(num_stripes); 3023 } else { 3024 ret = -EIO; 3025 break; 3026 } 3027 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 3028 key.offset == chunk_offset) { 3029 memmove(ptr, ptr + len, array_size - (cur + len)); 3030 array_size -= len; 3031 btrfs_set_super_sys_array_size(super_copy, array_size); 3032 } else { 3033 ptr += len; 3034 cur += len; 3035 } 3036 } 3037 return ret; 3038 } 3039 3040 /* 3041 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 3042 * @logical: Logical block offset in bytes. 3043 * @length: Length of extent in bytes. 3044 * 3045 * Return: Chunk mapping or ERR_PTR. 3046 */ 3047 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 3048 u64 logical, u64 length) 3049 { 3050 struct extent_map_tree *em_tree; 3051 struct extent_map *em; 3052 3053 em_tree = &fs_info->mapping_tree; 3054 read_lock(&em_tree->lock); 3055 em = lookup_extent_mapping(em_tree, logical, length); 3056 read_unlock(&em_tree->lock); 3057 3058 if (!em) { 3059 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 3060 logical, length); 3061 return ERR_PTR(-EINVAL); 3062 } 3063 3064 if (em->start > logical || em->start + em->len < logical) { 3065 btrfs_crit(fs_info, 3066 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3067 logical, length, em->start, em->start + em->len); 3068 free_extent_map(em); 3069 return ERR_PTR(-EINVAL); 3070 } 3071 3072 /* callers are responsible for dropping em's ref. */ 3073 return em; 3074 } 3075 3076 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3077 struct map_lookup *map, u64 chunk_offset) 3078 { 3079 int i; 3080 3081 /* 3082 * Removing chunk items and updating the device items in the chunks btree 3083 * requires holding the chunk_mutex. 3084 * See the comment at btrfs_chunk_alloc() for the details. 3085 */ 3086 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3087 3088 for (i = 0; i < map->num_stripes; i++) { 3089 int ret; 3090 3091 ret = btrfs_update_device(trans, map->stripes[i].dev); 3092 if (ret) 3093 return ret; 3094 } 3095 3096 return btrfs_free_chunk(trans, chunk_offset); 3097 } 3098 3099 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3100 { 3101 struct btrfs_fs_info *fs_info = trans->fs_info; 3102 struct extent_map *em; 3103 struct map_lookup *map; 3104 u64 dev_extent_len = 0; 3105 int i, ret = 0; 3106 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3107 3108 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3109 if (IS_ERR(em)) { 3110 /* 3111 * This is a logic error, but we don't want to just rely on the 3112 * user having built with ASSERT enabled, so if ASSERT doesn't 3113 * do anything we still error out. 3114 */ 3115 ASSERT(0); 3116 return PTR_ERR(em); 3117 } 3118 map = em->map_lookup; 3119 3120 /* 3121 * First delete the device extent items from the devices btree. 3122 * We take the device_list_mutex to avoid racing with the finishing phase 3123 * of a device replace operation. See the comment below before acquiring 3124 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3125 * because that can result in a deadlock when deleting the device extent 3126 * items from the devices btree - COWing an extent buffer from the btree 3127 * may result in allocating a new metadata chunk, which would attempt to 3128 * lock again fs_info->chunk_mutex. 3129 */ 3130 mutex_lock(&fs_devices->device_list_mutex); 3131 for (i = 0; i < map->num_stripes; i++) { 3132 struct btrfs_device *device = map->stripes[i].dev; 3133 ret = btrfs_free_dev_extent(trans, device, 3134 map->stripes[i].physical, 3135 &dev_extent_len); 3136 if (ret) { 3137 mutex_unlock(&fs_devices->device_list_mutex); 3138 btrfs_abort_transaction(trans, ret); 3139 goto out; 3140 } 3141 3142 if (device->bytes_used > 0) { 3143 mutex_lock(&fs_info->chunk_mutex); 3144 btrfs_device_set_bytes_used(device, 3145 device->bytes_used - dev_extent_len); 3146 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3147 btrfs_clear_space_info_full(fs_info); 3148 mutex_unlock(&fs_info->chunk_mutex); 3149 } 3150 } 3151 mutex_unlock(&fs_devices->device_list_mutex); 3152 3153 /* 3154 * We acquire fs_info->chunk_mutex for 2 reasons: 3155 * 3156 * 1) Just like with the first phase of the chunk allocation, we must 3157 * reserve system space, do all chunk btree updates and deletions, and 3158 * update the system chunk array in the superblock while holding this 3159 * mutex. This is for similar reasons as explained on the comment at 3160 * the top of btrfs_chunk_alloc(); 3161 * 3162 * 2) Prevent races with the final phase of a device replace operation 3163 * that replaces the device object associated with the map's stripes, 3164 * because the device object's id can change at any time during that 3165 * final phase of the device replace operation 3166 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3167 * replaced device and then see it with an ID of 3168 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3169 * the device item, which does not exists on the chunk btree. 3170 * The finishing phase of device replace acquires both the 3171 * device_list_mutex and the chunk_mutex, in that order, so we are 3172 * safe by just acquiring the chunk_mutex. 3173 */ 3174 trans->removing_chunk = true; 3175 mutex_lock(&fs_info->chunk_mutex); 3176 3177 check_system_chunk(trans, map->type); 3178 3179 ret = remove_chunk_item(trans, map, chunk_offset); 3180 /* 3181 * Normally we should not get -ENOSPC since we reserved space before 3182 * through the call to check_system_chunk(). 3183 * 3184 * Despite our system space_info having enough free space, we may not 3185 * be able to allocate extents from its block groups, because all have 3186 * an incompatible profile, which will force us to allocate a new system 3187 * block group with the right profile, or right after we called 3188 * check_system_space() above, a scrub turned the only system block group 3189 * with enough free space into RO mode. 3190 * This is explained with more detail at do_chunk_alloc(). 3191 * 3192 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3193 */ 3194 if (ret == -ENOSPC) { 3195 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3196 struct btrfs_block_group *sys_bg; 3197 3198 sys_bg = btrfs_create_chunk(trans, sys_flags); 3199 if (IS_ERR(sys_bg)) { 3200 ret = PTR_ERR(sys_bg); 3201 btrfs_abort_transaction(trans, ret); 3202 goto out; 3203 } 3204 3205 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3206 if (ret) { 3207 btrfs_abort_transaction(trans, ret); 3208 goto out; 3209 } 3210 3211 ret = remove_chunk_item(trans, map, chunk_offset); 3212 if (ret) { 3213 btrfs_abort_transaction(trans, ret); 3214 goto out; 3215 } 3216 } else if (ret) { 3217 btrfs_abort_transaction(trans, ret); 3218 goto out; 3219 } 3220 3221 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3222 3223 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3224 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3225 if (ret) { 3226 btrfs_abort_transaction(trans, ret); 3227 goto out; 3228 } 3229 } 3230 3231 mutex_unlock(&fs_info->chunk_mutex); 3232 trans->removing_chunk = false; 3233 3234 /* 3235 * We are done with chunk btree updates and deletions, so release the 3236 * system space we previously reserved (with check_system_chunk()). 3237 */ 3238 btrfs_trans_release_chunk_metadata(trans); 3239 3240 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3241 if (ret) { 3242 btrfs_abort_transaction(trans, ret); 3243 goto out; 3244 } 3245 3246 out: 3247 if (trans->removing_chunk) { 3248 mutex_unlock(&fs_info->chunk_mutex); 3249 trans->removing_chunk = false; 3250 } 3251 /* once for us */ 3252 free_extent_map(em); 3253 return ret; 3254 } 3255 3256 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3257 { 3258 struct btrfs_root *root = fs_info->chunk_root; 3259 struct btrfs_trans_handle *trans; 3260 struct btrfs_block_group *block_group; 3261 u64 length; 3262 int ret; 3263 3264 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3265 btrfs_err(fs_info, 3266 "relocate: not supported on extent tree v2 yet"); 3267 return -EINVAL; 3268 } 3269 3270 /* 3271 * Prevent races with automatic removal of unused block groups. 3272 * After we relocate and before we remove the chunk with offset 3273 * chunk_offset, automatic removal of the block group can kick in, 3274 * resulting in a failure when calling btrfs_remove_chunk() below. 3275 * 3276 * Make sure to acquire this mutex before doing a tree search (dev 3277 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3278 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3279 * we release the path used to search the chunk/dev tree and before 3280 * the current task acquires this mutex and calls us. 3281 */ 3282 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3283 3284 /* step one, relocate all the extents inside this chunk */ 3285 btrfs_scrub_pause(fs_info); 3286 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3287 btrfs_scrub_continue(fs_info); 3288 if (ret) { 3289 /* 3290 * If we had a transaction abort, stop all running scrubs. 3291 * See transaction.c:cleanup_transaction() why we do it here. 3292 */ 3293 if (BTRFS_FS_ERROR(fs_info)) 3294 btrfs_scrub_cancel(fs_info); 3295 return ret; 3296 } 3297 3298 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3299 if (!block_group) 3300 return -ENOENT; 3301 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3302 length = block_group->length; 3303 btrfs_put_block_group(block_group); 3304 3305 /* 3306 * On a zoned file system, discard the whole block group, this will 3307 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3308 * resetting the zone fails, don't treat it as a fatal problem from the 3309 * filesystem's point of view. 3310 */ 3311 if (btrfs_is_zoned(fs_info)) { 3312 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3313 if (ret) 3314 btrfs_info(fs_info, 3315 "failed to reset zone %llu after relocation", 3316 chunk_offset); 3317 } 3318 3319 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3320 chunk_offset); 3321 if (IS_ERR(trans)) { 3322 ret = PTR_ERR(trans); 3323 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3324 return ret; 3325 } 3326 3327 /* 3328 * step two, delete the device extents and the 3329 * chunk tree entries 3330 */ 3331 ret = btrfs_remove_chunk(trans, chunk_offset); 3332 btrfs_end_transaction(trans); 3333 return ret; 3334 } 3335 3336 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3337 { 3338 struct btrfs_root *chunk_root = fs_info->chunk_root; 3339 struct btrfs_path *path; 3340 struct extent_buffer *leaf; 3341 struct btrfs_chunk *chunk; 3342 struct btrfs_key key; 3343 struct btrfs_key found_key; 3344 u64 chunk_type; 3345 bool retried = false; 3346 int failed = 0; 3347 int ret; 3348 3349 path = btrfs_alloc_path(); 3350 if (!path) 3351 return -ENOMEM; 3352 3353 again: 3354 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3355 key.offset = (u64)-1; 3356 key.type = BTRFS_CHUNK_ITEM_KEY; 3357 3358 while (1) { 3359 mutex_lock(&fs_info->reclaim_bgs_lock); 3360 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3361 if (ret < 0) { 3362 mutex_unlock(&fs_info->reclaim_bgs_lock); 3363 goto error; 3364 } 3365 BUG_ON(ret == 0); /* Corruption */ 3366 3367 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3368 key.type); 3369 if (ret) 3370 mutex_unlock(&fs_info->reclaim_bgs_lock); 3371 if (ret < 0) 3372 goto error; 3373 if (ret > 0) 3374 break; 3375 3376 leaf = path->nodes[0]; 3377 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3378 3379 chunk = btrfs_item_ptr(leaf, path->slots[0], 3380 struct btrfs_chunk); 3381 chunk_type = btrfs_chunk_type(leaf, chunk); 3382 btrfs_release_path(path); 3383 3384 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3385 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3386 if (ret == -ENOSPC) 3387 failed++; 3388 else 3389 BUG_ON(ret); 3390 } 3391 mutex_unlock(&fs_info->reclaim_bgs_lock); 3392 3393 if (found_key.offset == 0) 3394 break; 3395 key.offset = found_key.offset - 1; 3396 } 3397 ret = 0; 3398 if (failed && !retried) { 3399 failed = 0; 3400 retried = true; 3401 goto again; 3402 } else if (WARN_ON(failed && retried)) { 3403 ret = -ENOSPC; 3404 } 3405 error: 3406 btrfs_free_path(path); 3407 return ret; 3408 } 3409 3410 /* 3411 * return 1 : allocate a data chunk successfully, 3412 * return <0: errors during allocating a data chunk, 3413 * return 0 : no need to allocate a data chunk. 3414 */ 3415 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3416 u64 chunk_offset) 3417 { 3418 struct btrfs_block_group *cache; 3419 u64 bytes_used; 3420 u64 chunk_type; 3421 3422 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3423 ASSERT(cache); 3424 chunk_type = cache->flags; 3425 btrfs_put_block_group(cache); 3426 3427 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3428 return 0; 3429 3430 spin_lock(&fs_info->data_sinfo->lock); 3431 bytes_used = fs_info->data_sinfo->bytes_used; 3432 spin_unlock(&fs_info->data_sinfo->lock); 3433 3434 if (!bytes_used) { 3435 struct btrfs_trans_handle *trans; 3436 int ret; 3437 3438 trans = btrfs_join_transaction(fs_info->tree_root); 3439 if (IS_ERR(trans)) 3440 return PTR_ERR(trans); 3441 3442 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3443 btrfs_end_transaction(trans); 3444 if (ret < 0) 3445 return ret; 3446 return 1; 3447 } 3448 3449 return 0; 3450 } 3451 3452 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3453 struct btrfs_balance_control *bctl) 3454 { 3455 struct btrfs_root *root = fs_info->tree_root; 3456 struct btrfs_trans_handle *trans; 3457 struct btrfs_balance_item *item; 3458 struct btrfs_disk_balance_args disk_bargs; 3459 struct btrfs_path *path; 3460 struct extent_buffer *leaf; 3461 struct btrfs_key key; 3462 int ret, err; 3463 3464 path = btrfs_alloc_path(); 3465 if (!path) 3466 return -ENOMEM; 3467 3468 trans = btrfs_start_transaction(root, 0); 3469 if (IS_ERR(trans)) { 3470 btrfs_free_path(path); 3471 return PTR_ERR(trans); 3472 } 3473 3474 key.objectid = BTRFS_BALANCE_OBJECTID; 3475 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3476 key.offset = 0; 3477 3478 ret = btrfs_insert_empty_item(trans, root, path, &key, 3479 sizeof(*item)); 3480 if (ret) 3481 goto out; 3482 3483 leaf = path->nodes[0]; 3484 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3485 3486 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3487 3488 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3489 btrfs_set_balance_data(leaf, item, &disk_bargs); 3490 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3491 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3492 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3493 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3494 3495 btrfs_set_balance_flags(leaf, item, bctl->flags); 3496 3497 btrfs_mark_buffer_dirty(leaf); 3498 out: 3499 btrfs_free_path(path); 3500 err = btrfs_commit_transaction(trans); 3501 if (err && !ret) 3502 ret = err; 3503 return ret; 3504 } 3505 3506 static int del_balance_item(struct btrfs_fs_info *fs_info) 3507 { 3508 struct btrfs_root *root = fs_info->tree_root; 3509 struct btrfs_trans_handle *trans; 3510 struct btrfs_path *path; 3511 struct btrfs_key key; 3512 int ret, err; 3513 3514 path = btrfs_alloc_path(); 3515 if (!path) 3516 return -ENOMEM; 3517 3518 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3519 if (IS_ERR(trans)) { 3520 btrfs_free_path(path); 3521 return PTR_ERR(trans); 3522 } 3523 3524 key.objectid = BTRFS_BALANCE_OBJECTID; 3525 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3526 key.offset = 0; 3527 3528 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3529 if (ret < 0) 3530 goto out; 3531 if (ret > 0) { 3532 ret = -ENOENT; 3533 goto out; 3534 } 3535 3536 ret = btrfs_del_item(trans, root, path); 3537 out: 3538 btrfs_free_path(path); 3539 err = btrfs_commit_transaction(trans); 3540 if (err && !ret) 3541 ret = err; 3542 return ret; 3543 } 3544 3545 /* 3546 * This is a heuristic used to reduce the number of chunks balanced on 3547 * resume after balance was interrupted. 3548 */ 3549 static void update_balance_args(struct btrfs_balance_control *bctl) 3550 { 3551 /* 3552 * Turn on soft mode for chunk types that were being converted. 3553 */ 3554 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3555 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3556 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3557 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3558 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3559 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3560 3561 /* 3562 * Turn on usage filter if is not already used. The idea is 3563 * that chunks that we have already balanced should be 3564 * reasonably full. Don't do it for chunks that are being 3565 * converted - that will keep us from relocating unconverted 3566 * (albeit full) chunks. 3567 */ 3568 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3569 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3570 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3571 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3572 bctl->data.usage = 90; 3573 } 3574 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3575 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3576 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3577 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3578 bctl->sys.usage = 90; 3579 } 3580 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3581 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3582 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3583 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3584 bctl->meta.usage = 90; 3585 } 3586 } 3587 3588 /* 3589 * Clear the balance status in fs_info and delete the balance item from disk. 3590 */ 3591 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3592 { 3593 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3594 int ret; 3595 3596 BUG_ON(!fs_info->balance_ctl); 3597 3598 spin_lock(&fs_info->balance_lock); 3599 fs_info->balance_ctl = NULL; 3600 spin_unlock(&fs_info->balance_lock); 3601 3602 kfree(bctl); 3603 ret = del_balance_item(fs_info); 3604 if (ret) 3605 btrfs_handle_fs_error(fs_info, ret, NULL); 3606 } 3607 3608 /* 3609 * Balance filters. Return 1 if chunk should be filtered out 3610 * (should not be balanced). 3611 */ 3612 static int chunk_profiles_filter(u64 chunk_type, 3613 struct btrfs_balance_args *bargs) 3614 { 3615 chunk_type = chunk_to_extended(chunk_type) & 3616 BTRFS_EXTENDED_PROFILE_MASK; 3617 3618 if (bargs->profiles & chunk_type) 3619 return 0; 3620 3621 return 1; 3622 } 3623 3624 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3625 struct btrfs_balance_args *bargs) 3626 { 3627 struct btrfs_block_group *cache; 3628 u64 chunk_used; 3629 u64 user_thresh_min; 3630 u64 user_thresh_max; 3631 int ret = 1; 3632 3633 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3634 chunk_used = cache->used; 3635 3636 if (bargs->usage_min == 0) 3637 user_thresh_min = 0; 3638 else 3639 user_thresh_min = mult_perc(cache->length, bargs->usage_min); 3640 3641 if (bargs->usage_max == 0) 3642 user_thresh_max = 1; 3643 else if (bargs->usage_max > 100) 3644 user_thresh_max = cache->length; 3645 else 3646 user_thresh_max = mult_perc(cache->length, bargs->usage_max); 3647 3648 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3649 ret = 0; 3650 3651 btrfs_put_block_group(cache); 3652 return ret; 3653 } 3654 3655 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3656 u64 chunk_offset, struct btrfs_balance_args *bargs) 3657 { 3658 struct btrfs_block_group *cache; 3659 u64 chunk_used, user_thresh; 3660 int ret = 1; 3661 3662 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3663 chunk_used = cache->used; 3664 3665 if (bargs->usage_min == 0) 3666 user_thresh = 1; 3667 else if (bargs->usage > 100) 3668 user_thresh = cache->length; 3669 else 3670 user_thresh = mult_perc(cache->length, bargs->usage); 3671 3672 if (chunk_used < user_thresh) 3673 ret = 0; 3674 3675 btrfs_put_block_group(cache); 3676 return ret; 3677 } 3678 3679 static int chunk_devid_filter(struct extent_buffer *leaf, 3680 struct btrfs_chunk *chunk, 3681 struct btrfs_balance_args *bargs) 3682 { 3683 struct btrfs_stripe *stripe; 3684 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3685 int i; 3686 3687 for (i = 0; i < num_stripes; i++) { 3688 stripe = btrfs_stripe_nr(chunk, i); 3689 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3690 return 0; 3691 } 3692 3693 return 1; 3694 } 3695 3696 static u64 calc_data_stripes(u64 type, int num_stripes) 3697 { 3698 const int index = btrfs_bg_flags_to_raid_index(type); 3699 const int ncopies = btrfs_raid_array[index].ncopies; 3700 const int nparity = btrfs_raid_array[index].nparity; 3701 3702 return (num_stripes - nparity) / ncopies; 3703 } 3704 3705 /* [pstart, pend) */ 3706 static int chunk_drange_filter(struct extent_buffer *leaf, 3707 struct btrfs_chunk *chunk, 3708 struct btrfs_balance_args *bargs) 3709 { 3710 struct btrfs_stripe *stripe; 3711 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3712 u64 stripe_offset; 3713 u64 stripe_length; 3714 u64 type; 3715 int factor; 3716 int i; 3717 3718 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3719 return 0; 3720 3721 type = btrfs_chunk_type(leaf, chunk); 3722 factor = calc_data_stripes(type, num_stripes); 3723 3724 for (i = 0; i < num_stripes; i++) { 3725 stripe = btrfs_stripe_nr(chunk, i); 3726 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3727 continue; 3728 3729 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3730 stripe_length = btrfs_chunk_length(leaf, chunk); 3731 stripe_length = div_u64(stripe_length, factor); 3732 3733 if (stripe_offset < bargs->pend && 3734 stripe_offset + stripe_length > bargs->pstart) 3735 return 0; 3736 } 3737 3738 return 1; 3739 } 3740 3741 /* [vstart, vend) */ 3742 static int chunk_vrange_filter(struct extent_buffer *leaf, 3743 struct btrfs_chunk *chunk, 3744 u64 chunk_offset, 3745 struct btrfs_balance_args *bargs) 3746 { 3747 if (chunk_offset < bargs->vend && 3748 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3749 /* at least part of the chunk is inside this vrange */ 3750 return 0; 3751 3752 return 1; 3753 } 3754 3755 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3756 struct btrfs_chunk *chunk, 3757 struct btrfs_balance_args *bargs) 3758 { 3759 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3760 3761 if (bargs->stripes_min <= num_stripes 3762 && num_stripes <= bargs->stripes_max) 3763 return 0; 3764 3765 return 1; 3766 } 3767 3768 static int chunk_soft_convert_filter(u64 chunk_type, 3769 struct btrfs_balance_args *bargs) 3770 { 3771 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3772 return 0; 3773 3774 chunk_type = chunk_to_extended(chunk_type) & 3775 BTRFS_EXTENDED_PROFILE_MASK; 3776 3777 if (bargs->target == chunk_type) 3778 return 1; 3779 3780 return 0; 3781 } 3782 3783 static int should_balance_chunk(struct extent_buffer *leaf, 3784 struct btrfs_chunk *chunk, u64 chunk_offset) 3785 { 3786 struct btrfs_fs_info *fs_info = leaf->fs_info; 3787 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3788 struct btrfs_balance_args *bargs = NULL; 3789 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3790 3791 /* type filter */ 3792 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3793 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3794 return 0; 3795 } 3796 3797 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3798 bargs = &bctl->data; 3799 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3800 bargs = &bctl->sys; 3801 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3802 bargs = &bctl->meta; 3803 3804 /* profiles filter */ 3805 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3806 chunk_profiles_filter(chunk_type, bargs)) { 3807 return 0; 3808 } 3809 3810 /* usage filter */ 3811 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3812 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3813 return 0; 3814 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3815 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3816 return 0; 3817 } 3818 3819 /* devid filter */ 3820 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3821 chunk_devid_filter(leaf, chunk, bargs)) { 3822 return 0; 3823 } 3824 3825 /* drange filter, makes sense only with devid filter */ 3826 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3827 chunk_drange_filter(leaf, chunk, bargs)) { 3828 return 0; 3829 } 3830 3831 /* vrange filter */ 3832 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3833 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3834 return 0; 3835 } 3836 3837 /* stripes filter */ 3838 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3839 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3840 return 0; 3841 } 3842 3843 /* soft profile changing mode */ 3844 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3845 chunk_soft_convert_filter(chunk_type, bargs)) { 3846 return 0; 3847 } 3848 3849 /* 3850 * limited by count, must be the last filter 3851 */ 3852 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3853 if (bargs->limit == 0) 3854 return 0; 3855 else 3856 bargs->limit--; 3857 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3858 /* 3859 * Same logic as the 'limit' filter; the minimum cannot be 3860 * determined here because we do not have the global information 3861 * about the count of all chunks that satisfy the filters. 3862 */ 3863 if (bargs->limit_max == 0) 3864 return 0; 3865 else 3866 bargs->limit_max--; 3867 } 3868 3869 return 1; 3870 } 3871 3872 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3873 { 3874 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3875 struct btrfs_root *chunk_root = fs_info->chunk_root; 3876 u64 chunk_type; 3877 struct btrfs_chunk *chunk; 3878 struct btrfs_path *path = NULL; 3879 struct btrfs_key key; 3880 struct btrfs_key found_key; 3881 struct extent_buffer *leaf; 3882 int slot; 3883 int ret; 3884 int enospc_errors = 0; 3885 bool counting = true; 3886 /* The single value limit and min/max limits use the same bytes in the */ 3887 u64 limit_data = bctl->data.limit; 3888 u64 limit_meta = bctl->meta.limit; 3889 u64 limit_sys = bctl->sys.limit; 3890 u32 count_data = 0; 3891 u32 count_meta = 0; 3892 u32 count_sys = 0; 3893 int chunk_reserved = 0; 3894 3895 path = btrfs_alloc_path(); 3896 if (!path) { 3897 ret = -ENOMEM; 3898 goto error; 3899 } 3900 3901 /* zero out stat counters */ 3902 spin_lock(&fs_info->balance_lock); 3903 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3904 spin_unlock(&fs_info->balance_lock); 3905 again: 3906 if (!counting) { 3907 /* 3908 * The single value limit and min/max limits use the same bytes 3909 * in the 3910 */ 3911 bctl->data.limit = limit_data; 3912 bctl->meta.limit = limit_meta; 3913 bctl->sys.limit = limit_sys; 3914 } 3915 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3916 key.offset = (u64)-1; 3917 key.type = BTRFS_CHUNK_ITEM_KEY; 3918 3919 while (1) { 3920 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3921 atomic_read(&fs_info->balance_cancel_req)) { 3922 ret = -ECANCELED; 3923 goto error; 3924 } 3925 3926 mutex_lock(&fs_info->reclaim_bgs_lock); 3927 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3928 if (ret < 0) { 3929 mutex_unlock(&fs_info->reclaim_bgs_lock); 3930 goto error; 3931 } 3932 3933 /* 3934 * this shouldn't happen, it means the last relocate 3935 * failed 3936 */ 3937 if (ret == 0) 3938 BUG(); /* FIXME break ? */ 3939 3940 ret = btrfs_previous_item(chunk_root, path, 0, 3941 BTRFS_CHUNK_ITEM_KEY); 3942 if (ret) { 3943 mutex_unlock(&fs_info->reclaim_bgs_lock); 3944 ret = 0; 3945 break; 3946 } 3947 3948 leaf = path->nodes[0]; 3949 slot = path->slots[0]; 3950 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3951 3952 if (found_key.objectid != key.objectid) { 3953 mutex_unlock(&fs_info->reclaim_bgs_lock); 3954 break; 3955 } 3956 3957 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3958 chunk_type = btrfs_chunk_type(leaf, chunk); 3959 3960 if (!counting) { 3961 spin_lock(&fs_info->balance_lock); 3962 bctl->stat.considered++; 3963 spin_unlock(&fs_info->balance_lock); 3964 } 3965 3966 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3967 3968 btrfs_release_path(path); 3969 if (!ret) { 3970 mutex_unlock(&fs_info->reclaim_bgs_lock); 3971 goto loop; 3972 } 3973 3974 if (counting) { 3975 mutex_unlock(&fs_info->reclaim_bgs_lock); 3976 spin_lock(&fs_info->balance_lock); 3977 bctl->stat.expected++; 3978 spin_unlock(&fs_info->balance_lock); 3979 3980 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3981 count_data++; 3982 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3983 count_sys++; 3984 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3985 count_meta++; 3986 3987 goto loop; 3988 } 3989 3990 /* 3991 * Apply limit_min filter, no need to check if the LIMITS 3992 * filter is used, limit_min is 0 by default 3993 */ 3994 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3995 count_data < bctl->data.limit_min) 3996 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3997 count_meta < bctl->meta.limit_min) 3998 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3999 count_sys < bctl->sys.limit_min)) { 4000 mutex_unlock(&fs_info->reclaim_bgs_lock); 4001 goto loop; 4002 } 4003 4004 if (!chunk_reserved) { 4005 /* 4006 * We may be relocating the only data chunk we have, 4007 * which could potentially end up with losing data's 4008 * raid profile, so lets allocate an empty one in 4009 * advance. 4010 */ 4011 ret = btrfs_may_alloc_data_chunk(fs_info, 4012 found_key.offset); 4013 if (ret < 0) { 4014 mutex_unlock(&fs_info->reclaim_bgs_lock); 4015 goto error; 4016 } else if (ret == 1) { 4017 chunk_reserved = 1; 4018 } 4019 } 4020 4021 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 4022 mutex_unlock(&fs_info->reclaim_bgs_lock); 4023 if (ret == -ENOSPC) { 4024 enospc_errors++; 4025 } else if (ret == -ETXTBSY) { 4026 btrfs_info(fs_info, 4027 "skipping relocation of block group %llu due to active swapfile", 4028 found_key.offset); 4029 ret = 0; 4030 } else if (ret) { 4031 goto error; 4032 } else { 4033 spin_lock(&fs_info->balance_lock); 4034 bctl->stat.completed++; 4035 spin_unlock(&fs_info->balance_lock); 4036 } 4037 loop: 4038 if (found_key.offset == 0) 4039 break; 4040 key.offset = found_key.offset - 1; 4041 } 4042 4043 if (counting) { 4044 btrfs_release_path(path); 4045 counting = false; 4046 goto again; 4047 } 4048 error: 4049 btrfs_free_path(path); 4050 if (enospc_errors) { 4051 btrfs_info(fs_info, "%d enospc errors during balance", 4052 enospc_errors); 4053 if (!ret) 4054 ret = -ENOSPC; 4055 } 4056 4057 return ret; 4058 } 4059 4060 /* 4061 * See if a given profile is valid and reduced. 4062 * 4063 * @flags: profile to validate 4064 * @extended: if true @flags is treated as an extended profile 4065 */ 4066 static int alloc_profile_is_valid(u64 flags, int extended) 4067 { 4068 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 4069 BTRFS_BLOCK_GROUP_PROFILE_MASK); 4070 4071 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 4072 4073 /* 1) check that all other bits are zeroed */ 4074 if (flags & ~mask) 4075 return 0; 4076 4077 /* 2) see if profile is reduced */ 4078 if (flags == 0) 4079 return !extended; /* "0" is valid for usual profiles */ 4080 4081 return has_single_bit_set(flags); 4082 } 4083 4084 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4085 { 4086 /* cancel requested || normal exit path */ 4087 return atomic_read(&fs_info->balance_cancel_req) || 4088 (atomic_read(&fs_info->balance_pause_req) == 0 && 4089 atomic_read(&fs_info->balance_cancel_req) == 0); 4090 } 4091 4092 /* 4093 * Validate target profile against allowed profiles and return true if it's OK. 4094 * Otherwise print the error message and return false. 4095 */ 4096 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4097 const struct btrfs_balance_args *bargs, 4098 u64 allowed, const char *type) 4099 { 4100 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4101 return true; 4102 4103 /* Profile is valid and does not have bits outside of the allowed set */ 4104 if (alloc_profile_is_valid(bargs->target, 1) && 4105 (bargs->target & ~allowed) == 0) 4106 return true; 4107 4108 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4109 type, btrfs_bg_type_to_raid_name(bargs->target)); 4110 return false; 4111 } 4112 4113 /* 4114 * Fill @buf with textual description of balance filter flags @bargs, up to 4115 * @size_buf including the terminating null. The output may be trimmed if it 4116 * does not fit into the provided buffer. 4117 */ 4118 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4119 u32 size_buf) 4120 { 4121 int ret; 4122 u32 size_bp = size_buf; 4123 char *bp = buf; 4124 u64 flags = bargs->flags; 4125 char tmp_buf[128] = {'\0'}; 4126 4127 if (!flags) 4128 return; 4129 4130 #define CHECK_APPEND_NOARG(a) \ 4131 do { \ 4132 ret = snprintf(bp, size_bp, (a)); \ 4133 if (ret < 0 || ret >= size_bp) \ 4134 goto out_overflow; \ 4135 size_bp -= ret; \ 4136 bp += ret; \ 4137 } while (0) 4138 4139 #define CHECK_APPEND_1ARG(a, v1) \ 4140 do { \ 4141 ret = snprintf(bp, size_bp, (a), (v1)); \ 4142 if (ret < 0 || ret >= size_bp) \ 4143 goto out_overflow; \ 4144 size_bp -= ret; \ 4145 bp += ret; \ 4146 } while (0) 4147 4148 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4149 do { \ 4150 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4151 if (ret < 0 || ret >= size_bp) \ 4152 goto out_overflow; \ 4153 size_bp -= ret; \ 4154 bp += ret; \ 4155 } while (0) 4156 4157 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4158 CHECK_APPEND_1ARG("convert=%s,", 4159 btrfs_bg_type_to_raid_name(bargs->target)); 4160 4161 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4162 CHECK_APPEND_NOARG("soft,"); 4163 4164 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4165 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4166 sizeof(tmp_buf)); 4167 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4168 } 4169 4170 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4171 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4172 4173 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4174 CHECK_APPEND_2ARG("usage=%u..%u,", 4175 bargs->usage_min, bargs->usage_max); 4176 4177 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4178 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4179 4180 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4181 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4182 bargs->pstart, bargs->pend); 4183 4184 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4185 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4186 bargs->vstart, bargs->vend); 4187 4188 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4189 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4190 4191 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4192 CHECK_APPEND_2ARG("limit=%u..%u,", 4193 bargs->limit_min, bargs->limit_max); 4194 4195 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4196 CHECK_APPEND_2ARG("stripes=%u..%u,", 4197 bargs->stripes_min, bargs->stripes_max); 4198 4199 #undef CHECK_APPEND_2ARG 4200 #undef CHECK_APPEND_1ARG 4201 #undef CHECK_APPEND_NOARG 4202 4203 out_overflow: 4204 4205 if (size_bp < size_buf) 4206 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4207 else 4208 buf[0] = '\0'; 4209 } 4210 4211 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4212 { 4213 u32 size_buf = 1024; 4214 char tmp_buf[192] = {'\0'}; 4215 char *buf; 4216 char *bp; 4217 u32 size_bp = size_buf; 4218 int ret; 4219 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4220 4221 buf = kzalloc(size_buf, GFP_KERNEL); 4222 if (!buf) 4223 return; 4224 4225 bp = buf; 4226 4227 #define CHECK_APPEND_1ARG(a, v1) \ 4228 do { \ 4229 ret = snprintf(bp, size_bp, (a), (v1)); \ 4230 if (ret < 0 || ret >= size_bp) \ 4231 goto out_overflow; \ 4232 size_bp -= ret; \ 4233 bp += ret; \ 4234 } while (0) 4235 4236 if (bctl->flags & BTRFS_BALANCE_FORCE) 4237 CHECK_APPEND_1ARG("%s", "-f "); 4238 4239 if (bctl->flags & BTRFS_BALANCE_DATA) { 4240 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4241 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4242 } 4243 4244 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4245 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4246 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4247 } 4248 4249 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4250 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4251 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4252 } 4253 4254 #undef CHECK_APPEND_1ARG 4255 4256 out_overflow: 4257 4258 if (size_bp < size_buf) 4259 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4260 btrfs_info(fs_info, "balance: %s %s", 4261 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4262 "resume" : "start", buf); 4263 4264 kfree(buf); 4265 } 4266 4267 /* 4268 * Should be called with balance mutexe held 4269 */ 4270 int btrfs_balance(struct btrfs_fs_info *fs_info, 4271 struct btrfs_balance_control *bctl, 4272 struct btrfs_ioctl_balance_args *bargs) 4273 { 4274 u64 meta_target, data_target; 4275 u64 allowed; 4276 int mixed = 0; 4277 int ret; 4278 u64 num_devices; 4279 unsigned seq; 4280 bool reducing_redundancy; 4281 int i; 4282 4283 if (btrfs_fs_closing(fs_info) || 4284 atomic_read(&fs_info->balance_pause_req) || 4285 btrfs_should_cancel_balance(fs_info)) { 4286 ret = -EINVAL; 4287 goto out; 4288 } 4289 4290 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4291 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4292 mixed = 1; 4293 4294 /* 4295 * In case of mixed groups both data and meta should be picked, 4296 * and identical options should be given for both of them. 4297 */ 4298 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4299 if (mixed && (bctl->flags & allowed)) { 4300 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4301 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4302 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4303 btrfs_err(fs_info, 4304 "balance: mixed groups data and metadata options must be the same"); 4305 ret = -EINVAL; 4306 goto out; 4307 } 4308 } 4309 4310 /* 4311 * rw_devices will not change at the moment, device add/delete/replace 4312 * are exclusive 4313 */ 4314 num_devices = fs_info->fs_devices->rw_devices; 4315 4316 /* 4317 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4318 * special bit for it, to make it easier to distinguish. Thus we need 4319 * to set it manually, or balance would refuse the profile. 4320 */ 4321 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4322 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4323 if (num_devices >= btrfs_raid_array[i].devs_min) 4324 allowed |= btrfs_raid_array[i].bg_flag; 4325 4326 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4327 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4328 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4329 ret = -EINVAL; 4330 goto out; 4331 } 4332 4333 /* 4334 * Allow to reduce metadata or system integrity only if force set for 4335 * profiles with redundancy (copies, parity) 4336 */ 4337 allowed = 0; 4338 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4339 if (btrfs_raid_array[i].ncopies >= 2 || 4340 btrfs_raid_array[i].tolerated_failures >= 1) 4341 allowed |= btrfs_raid_array[i].bg_flag; 4342 } 4343 do { 4344 seq = read_seqbegin(&fs_info->profiles_lock); 4345 4346 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4347 (fs_info->avail_system_alloc_bits & allowed) && 4348 !(bctl->sys.target & allowed)) || 4349 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4350 (fs_info->avail_metadata_alloc_bits & allowed) && 4351 !(bctl->meta.target & allowed))) 4352 reducing_redundancy = true; 4353 else 4354 reducing_redundancy = false; 4355 4356 /* if we're not converting, the target field is uninitialized */ 4357 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4358 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4359 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4360 bctl->data.target : fs_info->avail_data_alloc_bits; 4361 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4362 4363 if (reducing_redundancy) { 4364 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4365 btrfs_info(fs_info, 4366 "balance: force reducing metadata redundancy"); 4367 } else { 4368 btrfs_err(fs_info, 4369 "balance: reduces metadata redundancy, use --force if you want this"); 4370 ret = -EINVAL; 4371 goto out; 4372 } 4373 } 4374 4375 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4376 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4377 btrfs_warn(fs_info, 4378 "balance: metadata profile %s has lower redundancy than data profile %s", 4379 btrfs_bg_type_to_raid_name(meta_target), 4380 btrfs_bg_type_to_raid_name(data_target)); 4381 } 4382 4383 ret = insert_balance_item(fs_info, bctl); 4384 if (ret && ret != -EEXIST) 4385 goto out; 4386 4387 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4388 BUG_ON(ret == -EEXIST); 4389 BUG_ON(fs_info->balance_ctl); 4390 spin_lock(&fs_info->balance_lock); 4391 fs_info->balance_ctl = bctl; 4392 spin_unlock(&fs_info->balance_lock); 4393 } else { 4394 BUG_ON(ret != -EEXIST); 4395 spin_lock(&fs_info->balance_lock); 4396 update_balance_args(bctl); 4397 spin_unlock(&fs_info->balance_lock); 4398 } 4399 4400 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4401 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4402 describe_balance_start_or_resume(fs_info); 4403 mutex_unlock(&fs_info->balance_mutex); 4404 4405 ret = __btrfs_balance(fs_info); 4406 4407 mutex_lock(&fs_info->balance_mutex); 4408 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4409 btrfs_info(fs_info, "balance: paused"); 4410 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4411 } 4412 /* 4413 * Balance can be canceled by: 4414 * 4415 * - Regular cancel request 4416 * Then ret == -ECANCELED and balance_cancel_req > 0 4417 * 4418 * - Fatal signal to "btrfs" process 4419 * Either the signal caught by wait_reserve_ticket() and callers 4420 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4421 * got -ECANCELED. 4422 * Either way, in this case balance_cancel_req = 0, and 4423 * ret == -EINTR or ret == -ECANCELED. 4424 * 4425 * So here we only check the return value to catch canceled balance. 4426 */ 4427 else if (ret == -ECANCELED || ret == -EINTR) 4428 btrfs_info(fs_info, "balance: canceled"); 4429 else 4430 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4431 4432 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4433 4434 if (bargs) { 4435 memset(bargs, 0, sizeof(*bargs)); 4436 btrfs_update_ioctl_balance_args(fs_info, bargs); 4437 } 4438 4439 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4440 balance_need_close(fs_info)) { 4441 reset_balance_state(fs_info); 4442 btrfs_exclop_finish(fs_info); 4443 } 4444 4445 wake_up(&fs_info->balance_wait_q); 4446 4447 return ret; 4448 out: 4449 if (bctl->flags & BTRFS_BALANCE_RESUME) 4450 reset_balance_state(fs_info); 4451 else 4452 kfree(bctl); 4453 btrfs_exclop_finish(fs_info); 4454 4455 return ret; 4456 } 4457 4458 static int balance_kthread(void *data) 4459 { 4460 struct btrfs_fs_info *fs_info = data; 4461 int ret = 0; 4462 4463 sb_start_write(fs_info->sb); 4464 mutex_lock(&fs_info->balance_mutex); 4465 if (fs_info->balance_ctl) 4466 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4467 mutex_unlock(&fs_info->balance_mutex); 4468 sb_end_write(fs_info->sb); 4469 4470 return ret; 4471 } 4472 4473 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4474 { 4475 struct task_struct *tsk; 4476 4477 mutex_lock(&fs_info->balance_mutex); 4478 if (!fs_info->balance_ctl) { 4479 mutex_unlock(&fs_info->balance_mutex); 4480 return 0; 4481 } 4482 mutex_unlock(&fs_info->balance_mutex); 4483 4484 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4485 btrfs_info(fs_info, "balance: resume skipped"); 4486 return 0; 4487 } 4488 4489 spin_lock(&fs_info->super_lock); 4490 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4491 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4492 spin_unlock(&fs_info->super_lock); 4493 /* 4494 * A ro->rw remount sequence should continue with the paused balance 4495 * regardless of who pauses it, system or the user as of now, so set 4496 * the resume flag. 4497 */ 4498 spin_lock(&fs_info->balance_lock); 4499 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4500 spin_unlock(&fs_info->balance_lock); 4501 4502 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4503 return PTR_ERR_OR_ZERO(tsk); 4504 } 4505 4506 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4507 { 4508 struct btrfs_balance_control *bctl; 4509 struct btrfs_balance_item *item; 4510 struct btrfs_disk_balance_args disk_bargs; 4511 struct btrfs_path *path; 4512 struct extent_buffer *leaf; 4513 struct btrfs_key key; 4514 int ret; 4515 4516 path = btrfs_alloc_path(); 4517 if (!path) 4518 return -ENOMEM; 4519 4520 key.objectid = BTRFS_BALANCE_OBJECTID; 4521 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4522 key.offset = 0; 4523 4524 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4525 if (ret < 0) 4526 goto out; 4527 if (ret > 0) { /* ret = -ENOENT; */ 4528 ret = 0; 4529 goto out; 4530 } 4531 4532 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4533 if (!bctl) { 4534 ret = -ENOMEM; 4535 goto out; 4536 } 4537 4538 leaf = path->nodes[0]; 4539 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4540 4541 bctl->flags = btrfs_balance_flags(leaf, item); 4542 bctl->flags |= BTRFS_BALANCE_RESUME; 4543 4544 btrfs_balance_data(leaf, item, &disk_bargs); 4545 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4546 btrfs_balance_meta(leaf, item, &disk_bargs); 4547 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4548 btrfs_balance_sys(leaf, item, &disk_bargs); 4549 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4550 4551 /* 4552 * This should never happen, as the paused balance state is recovered 4553 * during mount without any chance of other exclusive ops to collide. 4554 * 4555 * This gives the exclusive op status to balance and keeps in paused 4556 * state until user intervention (cancel or umount). If the ownership 4557 * cannot be assigned, show a message but do not fail. The balance 4558 * is in a paused state and must have fs_info::balance_ctl properly 4559 * set up. 4560 */ 4561 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4562 btrfs_warn(fs_info, 4563 "balance: cannot set exclusive op status, resume manually"); 4564 4565 btrfs_release_path(path); 4566 4567 mutex_lock(&fs_info->balance_mutex); 4568 BUG_ON(fs_info->balance_ctl); 4569 spin_lock(&fs_info->balance_lock); 4570 fs_info->balance_ctl = bctl; 4571 spin_unlock(&fs_info->balance_lock); 4572 mutex_unlock(&fs_info->balance_mutex); 4573 out: 4574 btrfs_free_path(path); 4575 return ret; 4576 } 4577 4578 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4579 { 4580 int ret = 0; 4581 4582 mutex_lock(&fs_info->balance_mutex); 4583 if (!fs_info->balance_ctl) { 4584 mutex_unlock(&fs_info->balance_mutex); 4585 return -ENOTCONN; 4586 } 4587 4588 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4589 atomic_inc(&fs_info->balance_pause_req); 4590 mutex_unlock(&fs_info->balance_mutex); 4591 4592 wait_event(fs_info->balance_wait_q, 4593 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4594 4595 mutex_lock(&fs_info->balance_mutex); 4596 /* we are good with balance_ctl ripped off from under us */ 4597 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4598 atomic_dec(&fs_info->balance_pause_req); 4599 } else { 4600 ret = -ENOTCONN; 4601 } 4602 4603 mutex_unlock(&fs_info->balance_mutex); 4604 return ret; 4605 } 4606 4607 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4608 { 4609 mutex_lock(&fs_info->balance_mutex); 4610 if (!fs_info->balance_ctl) { 4611 mutex_unlock(&fs_info->balance_mutex); 4612 return -ENOTCONN; 4613 } 4614 4615 /* 4616 * A paused balance with the item stored on disk can be resumed at 4617 * mount time if the mount is read-write. Otherwise it's still paused 4618 * and we must not allow cancelling as it deletes the item. 4619 */ 4620 if (sb_rdonly(fs_info->sb)) { 4621 mutex_unlock(&fs_info->balance_mutex); 4622 return -EROFS; 4623 } 4624 4625 atomic_inc(&fs_info->balance_cancel_req); 4626 /* 4627 * if we are running just wait and return, balance item is 4628 * deleted in btrfs_balance in this case 4629 */ 4630 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4631 mutex_unlock(&fs_info->balance_mutex); 4632 wait_event(fs_info->balance_wait_q, 4633 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4634 mutex_lock(&fs_info->balance_mutex); 4635 } else { 4636 mutex_unlock(&fs_info->balance_mutex); 4637 /* 4638 * Lock released to allow other waiters to continue, we'll 4639 * reexamine the status again. 4640 */ 4641 mutex_lock(&fs_info->balance_mutex); 4642 4643 if (fs_info->balance_ctl) { 4644 reset_balance_state(fs_info); 4645 btrfs_exclop_finish(fs_info); 4646 btrfs_info(fs_info, "balance: canceled"); 4647 } 4648 } 4649 4650 BUG_ON(fs_info->balance_ctl || 4651 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4652 atomic_dec(&fs_info->balance_cancel_req); 4653 mutex_unlock(&fs_info->balance_mutex); 4654 return 0; 4655 } 4656 4657 int btrfs_uuid_scan_kthread(void *data) 4658 { 4659 struct btrfs_fs_info *fs_info = data; 4660 struct btrfs_root *root = fs_info->tree_root; 4661 struct btrfs_key key; 4662 struct btrfs_path *path = NULL; 4663 int ret = 0; 4664 struct extent_buffer *eb; 4665 int slot; 4666 struct btrfs_root_item root_item; 4667 u32 item_size; 4668 struct btrfs_trans_handle *trans = NULL; 4669 bool closing = false; 4670 4671 path = btrfs_alloc_path(); 4672 if (!path) { 4673 ret = -ENOMEM; 4674 goto out; 4675 } 4676 4677 key.objectid = 0; 4678 key.type = BTRFS_ROOT_ITEM_KEY; 4679 key.offset = 0; 4680 4681 while (1) { 4682 if (btrfs_fs_closing(fs_info)) { 4683 closing = true; 4684 break; 4685 } 4686 ret = btrfs_search_forward(root, &key, path, 4687 BTRFS_OLDEST_GENERATION); 4688 if (ret) { 4689 if (ret > 0) 4690 ret = 0; 4691 break; 4692 } 4693 4694 if (key.type != BTRFS_ROOT_ITEM_KEY || 4695 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4696 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4697 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4698 goto skip; 4699 4700 eb = path->nodes[0]; 4701 slot = path->slots[0]; 4702 item_size = btrfs_item_size(eb, slot); 4703 if (item_size < sizeof(root_item)) 4704 goto skip; 4705 4706 read_extent_buffer(eb, &root_item, 4707 btrfs_item_ptr_offset(eb, slot), 4708 (int)sizeof(root_item)); 4709 if (btrfs_root_refs(&root_item) == 0) 4710 goto skip; 4711 4712 if (!btrfs_is_empty_uuid(root_item.uuid) || 4713 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4714 if (trans) 4715 goto update_tree; 4716 4717 btrfs_release_path(path); 4718 /* 4719 * 1 - subvol uuid item 4720 * 1 - received_subvol uuid item 4721 */ 4722 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4723 if (IS_ERR(trans)) { 4724 ret = PTR_ERR(trans); 4725 break; 4726 } 4727 continue; 4728 } else { 4729 goto skip; 4730 } 4731 update_tree: 4732 btrfs_release_path(path); 4733 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4734 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4735 BTRFS_UUID_KEY_SUBVOL, 4736 key.objectid); 4737 if (ret < 0) { 4738 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4739 ret); 4740 break; 4741 } 4742 } 4743 4744 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4745 ret = btrfs_uuid_tree_add(trans, 4746 root_item.received_uuid, 4747 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4748 key.objectid); 4749 if (ret < 0) { 4750 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4751 ret); 4752 break; 4753 } 4754 } 4755 4756 skip: 4757 btrfs_release_path(path); 4758 if (trans) { 4759 ret = btrfs_end_transaction(trans); 4760 trans = NULL; 4761 if (ret) 4762 break; 4763 } 4764 4765 if (key.offset < (u64)-1) { 4766 key.offset++; 4767 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4768 key.offset = 0; 4769 key.type = BTRFS_ROOT_ITEM_KEY; 4770 } else if (key.objectid < (u64)-1) { 4771 key.offset = 0; 4772 key.type = BTRFS_ROOT_ITEM_KEY; 4773 key.objectid++; 4774 } else { 4775 break; 4776 } 4777 cond_resched(); 4778 } 4779 4780 out: 4781 btrfs_free_path(path); 4782 if (trans && !IS_ERR(trans)) 4783 btrfs_end_transaction(trans); 4784 if (ret) 4785 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4786 else if (!closing) 4787 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4788 up(&fs_info->uuid_tree_rescan_sem); 4789 return 0; 4790 } 4791 4792 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4793 { 4794 struct btrfs_trans_handle *trans; 4795 struct btrfs_root *tree_root = fs_info->tree_root; 4796 struct btrfs_root *uuid_root; 4797 struct task_struct *task; 4798 int ret; 4799 4800 /* 4801 * 1 - root node 4802 * 1 - root item 4803 */ 4804 trans = btrfs_start_transaction(tree_root, 2); 4805 if (IS_ERR(trans)) 4806 return PTR_ERR(trans); 4807 4808 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4809 if (IS_ERR(uuid_root)) { 4810 ret = PTR_ERR(uuid_root); 4811 btrfs_abort_transaction(trans, ret); 4812 btrfs_end_transaction(trans); 4813 return ret; 4814 } 4815 4816 fs_info->uuid_root = uuid_root; 4817 4818 ret = btrfs_commit_transaction(trans); 4819 if (ret) 4820 return ret; 4821 4822 down(&fs_info->uuid_tree_rescan_sem); 4823 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4824 if (IS_ERR(task)) { 4825 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4826 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4827 up(&fs_info->uuid_tree_rescan_sem); 4828 return PTR_ERR(task); 4829 } 4830 4831 return 0; 4832 } 4833 4834 /* 4835 * shrinking a device means finding all of the device extents past 4836 * the new size, and then following the back refs to the chunks. 4837 * The chunk relocation code actually frees the device extent 4838 */ 4839 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4840 { 4841 struct btrfs_fs_info *fs_info = device->fs_info; 4842 struct btrfs_root *root = fs_info->dev_root; 4843 struct btrfs_trans_handle *trans; 4844 struct btrfs_dev_extent *dev_extent = NULL; 4845 struct btrfs_path *path; 4846 u64 length; 4847 u64 chunk_offset; 4848 int ret; 4849 int slot; 4850 int failed = 0; 4851 bool retried = false; 4852 struct extent_buffer *l; 4853 struct btrfs_key key; 4854 struct btrfs_super_block *super_copy = fs_info->super_copy; 4855 u64 old_total = btrfs_super_total_bytes(super_copy); 4856 u64 old_size = btrfs_device_get_total_bytes(device); 4857 u64 diff; 4858 u64 start; 4859 4860 new_size = round_down(new_size, fs_info->sectorsize); 4861 start = new_size; 4862 diff = round_down(old_size - new_size, fs_info->sectorsize); 4863 4864 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4865 return -EINVAL; 4866 4867 path = btrfs_alloc_path(); 4868 if (!path) 4869 return -ENOMEM; 4870 4871 path->reada = READA_BACK; 4872 4873 trans = btrfs_start_transaction(root, 0); 4874 if (IS_ERR(trans)) { 4875 btrfs_free_path(path); 4876 return PTR_ERR(trans); 4877 } 4878 4879 mutex_lock(&fs_info->chunk_mutex); 4880 4881 btrfs_device_set_total_bytes(device, new_size); 4882 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4883 device->fs_devices->total_rw_bytes -= diff; 4884 atomic64_sub(diff, &fs_info->free_chunk_space); 4885 } 4886 4887 /* 4888 * Once the device's size has been set to the new size, ensure all 4889 * in-memory chunks are synced to disk so that the loop below sees them 4890 * and relocates them accordingly. 4891 */ 4892 if (contains_pending_extent(device, &start, diff)) { 4893 mutex_unlock(&fs_info->chunk_mutex); 4894 ret = btrfs_commit_transaction(trans); 4895 if (ret) 4896 goto done; 4897 } else { 4898 mutex_unlock(&fs_info->chunk_mutex); 4899 btrfs_end_transaction(trans); 4900 } 4901 4902 again: 4903 key.objectid = device->devid; 4904 key.offset = (u64)-1; 4905 key.type = BTRFS_DEV_EXTENT_KEY; 4906 4907 do { 4908 mutex_lock(&fs_info->reclaim_bgs_lock); 4909 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4910 if (ret < 0) { 4911 mutex_unlock(&fs_info->reclaim_bgs_lock); 4912 goto done; 4913 } 4914 4915 ret = btrfs_previous_item(root, path, 0, key.type); 4916 if (ret) { 4917 mutex_unlock(&fs_info->reclaim_bgs_lock); 4918 if (ret < 0) 4919 goto done; 4920 ret = 0; 4921 btrfs_release_path(path); 4922 break; 4923 } 4924 4925 l = path->nodes[0]; 4926 slot = path->slots[0]; 4927 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4928 4929 if (key.objectid != device->devid) { 4930 mutex_unlock(&fs_info->reclaim_bgs_lock); 4931 btrfs_release_path(path); 4932 break; 4933 } 4934 4935 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4936 length = btrfs_dev_extent_length(l, dev_extent); 4937 4938 if (key.offset + length <= new_size) { 4939 mutex_unlock(&fs_info->reclaim_bgs_lock); 4940 btrfs_release_path(path); 4941 break; 4942 } 4943 4944 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4945 btrfs_release_path(path); 4946 4947 /* 4948 * We may be relocating the only data chunk we have, 4949 * which could potentially end up with losing data's 4950 * raid profile, so lets allocate an empty one in 4951 * advance. 4952 */ 4953 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4954 if (ret < 0) { 4955 mutex_unlock(&fs_info->reclaim_bgs_lock); 4956 goto done; 4957 } 4958 4959 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4960 mutex_unlock(&fs_info->reclaim_bgs_lock); 4961 if (ret == -ENOSPC) { 4962 failed++; 4963 } else if (ret) { 4964 if (ret == -ETXTBSY) { 4965 btrfs_warn(fs_info, 4966 "could not shrink block group %llu due to active swapfile", 4967 chunk_offset); 4968 } 4969 goto done; 4970 } 4971 } while (key.offset-- > 0); 4972 4973 if (failed && !retried) { 4974 failed = 0; 4975 retried = true; 4976 goto again; 4977 } else if (failed && retried) { 4978 ret = -ENOSPC; 4979 goto done; 4980 } 4981 4982 /* Shrinking succeeded, else we would be at "done". */ 4983 trans = btrfs_start_transaction(root, 0); 4984 if (IS_ERR(trans)) { 4985 ret = PTR_ERR(trans); 4986 goto done; 4987 } 4988 4989 mutex_lock(&fs_info->chunk_mutex); 4990 /* Clear all state bits beyond the shrunk device size */ 4991 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4992 CHUNK_STATE_MASK); 4993 4994 btrfs_device_set_disk_total_bytes(device, new_size); 4995 if (list_empty(&device->post_commit_list)) 4996 list_add_tail(&device->post_commit_list, 4997 &trans->transaction->dev_update_list); 4998 4999 WARN_ON(diff > old_total); 5000 btrfs_set_super_total_bytes(super_copy, 5001 round_down(old_total - diff, fs_info->sectorsize)); 5002 mutex_unlock(&fs_info->chunk_mutex); 5003 5004 btrfs_reserve_chunk_metadata(trans, false); 5005 /* Now btrfs_update_device() will change the on-disk size. */ 5006 ret = btrfs_update_device(trans, device); 5007 btrfs_trans_release_chunk_metadata(trans); 5008 if (ret < 0) { 5009 btrfs_abort_transaction(trans, ret); 5010 btrfs_end_transaction(trans); 5011 } else { 5012 ret = btrfs_commit_transaction(trans); 5013 } 5014 done: 5015 btrfs_free_path(path); 5016 if (ret) { 5017 mutex_lock(&fs_info->chunk_mutex); 5018 btrfs_device_set_total_bytes(device, old_size); 5019 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 5020 device->fs_devices->total_rw_bytes += diff; 5021 atomic64_add(diff, &fs_info->free_chunk_space); 5022 mutex_unlock(&fs_info->chunk_mutex); 5023 } 5024 return ret; 5025 } 5026 5027 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 5028 struct btrfs_key *key, 5029 struct btrfs_chunk *chunk, int item_size) 5030 { 5031 struct btrfs_super_block *super_copy = fs_info->super_copy; 5032 struct btrfs_disk_key disk_key; 5033 u32 array_size; 5034 u8 *ptr; 5035 5036 lockdep_assert_held(&fs_info->chunk_mutex); 5037 5038 array_size = btrfs_super_sys_array_size(super_copy); 5039 if (array_size + item_size + sizeof(disk_key) 5040 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5041 return -EFBIG; 5042 5043 ptr = super_copy->sys_chunk_array + array_size; 5044 btrfs_cpu_key_to_disk(&disk_key, key); 5045 memcpy(ptr, &disk_key, sizeof(disk_key)); 5046 ptr += sizeof(disk_key); 5047 memcpy(ptr, chunk, item_size); 5048 item_size += sizeof(disk_key); 5049 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5050 5051 return 0; 5052 } 5053 5054 /* 5055 * sort the devices in descending order by max_avail, total_avail 5056 */ 5057 static int btrfs_cmp_device_info(const void *a, const void *b) 5058 { 5059 const struct btrfs_device_info *di_a = a; 5060 const struct btrfs_device_info *di_b = b; 5061 5062 if (di_a->max_avail > di_b->max_avail) 5063 return -1; 5064 if (di_a->max_avail < di_b->max_avail) 5065 return 1; 5066 if (di_a->total_avail > di_b->total_avail) 5067 return -1; 5068 if (di_a->total_avail < di_b->total_avail) 5069 return 1; 5070 return 0; 5071 } 5072 5073 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5074 { 5075 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5076 return; 5077 5078 btrfs_set_fs_incompat(info, RAID56); 5079 } 5080 5081 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5082 { 5083 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5084 return; 5085 5086 btrfs_set_fs_incompat(info, RAID1C34); 5087 } 5088 5089 /* 5090 * Structure used internally for btrfs_create_chunk() function. 5091 * Wraps needed parameters. 5092 */ 5093 struct alloc_chunk_ctl { 5094 u64 start; 5095 u64 type; 5096 /* Total number of stripes to allocate */ 5097 int num_stripes; 5098 /* sub_stripes info for map */ 5099 int sub_stripes; 5100 /* Stripes per device */ 5101 int dev_stripes; 5102 /* Maximum number of devices to use */ 5103 int devs_max; 5104 /* Minimum number of devices to use */ 5105 int devs_min; 5106 /* ndevs has to be a multiple of this */ 5107 int devs_increment; 5108 /* Number of copies */ 5109 int ncopies; 5110 /* Number of stripes worth of bytes to store parity information */ 5111 int nparity; 5112 u64 max_stripe_size; 5113 u64 max_chunk_size; 5114 u64 dev_extent_min; 5115 u64 stripe_size; 5116 u64 chunk_size; 5117 int ndevs; 5118 }; 5119 5120 static void init_alloc_chunk_ctl_policy_regular( 5121 struct btrfs_fs_devices *fs_devices, 5122 struct alloc_chunk_ctl *ctl) 5123 { 5124 struct btrfs_space_info *space_info; 5125 5126 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); 5127 ASSERT(space_info); 5128 5129 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); 5130 ctl->max_stripe_size = ctl->max_chunk_size; 5131 5132 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) 5133 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); 5134 5135 /* We don't want a chunk larger than 10% of writable space */ 5136 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), 5137 ctl->max_chunk_size); 5138 ctl->dev_extent_min = ctl->dev_stripes << BTRFS_STRIPE_LEN_SHIFT; 5139 } 5140 5141 static void init_alloc_chunk_ctl_policy_zoned( 5142 struct btrfs_fs_devices *fs_devices, 5143 struct alloc_chunk_ctl *ctl) 5144 { 5145 u64 zone_size = fs_devices->fs_info->zone_size; 5146 u64 limit; 5147 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5148 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5149 u64 min_chunk_size = min_data_stripes * zone_size; 5150 u64 type = ctl->type; 5151 5152 ctl->max_stripe_size = zone_size; 5153 if (type & BTRFS_BLOCK_GROUP_DATA) { 5154 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5155 zone_size); 5156 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5157 ctl->max_chunk_size = ctl->max_stripe_size; 5158 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5159 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5160 ctl->devs_max = min_t(int, ctl->devs_max, 5161 BTRFS_MAX_DEVS_SYS_CHUNK); 5162 } else { 5163 BUG(); 5164 } 5165 5166 /* We don't want a chunk larger than 10% of writable space */ 5167 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), 5168 zone_size), 5169 min_chunk_size); 5170 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5171 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5172 } 5173 5174 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5175 struct alloc_chunk_ctl *ctl) 5176 { 5177 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5178 5179 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5180 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5181 ctl->devs_max = btrfs_raid_array[index].devs_max; 5182 if (!ctl->devs_max) 5183 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5184 ctl->devs_min = btrfs_raid_array[index].devs_min; 5185 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5186 ctl->ncopies = btrfs_raid_array[index].ncopies; 5187 ctl->nparity = btrfs_raid_array[index].nparity; 5188 ctl->ndevs = 0; 5189 5190 switch (fs_devices->chunk_alloc_policy) { 5191 case BTRFS_CHUNK_ALLOC_REGULAR: 5192 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5193 break; 5194 case BTRFS_CHUNK_ALLOC_ZONED: 5195 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5196 break; 5197 default: 5198 BUG(); 5199 } 5200 } 5201 5202 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5203 struct alloc_chunk_ctl *ctl, 5204 struct btrfs_device_info *devices_info) 5205 { 5206 struct btrfs_fs_info *info = fs_devices->fs_info; 5207 struct btrfs_device *device; 5208 u64 total_avail; 5209 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5210 int ret; 5211 int ndevs = 0; 5212 u64 max_avail; 5213 u64 dev_offset; 5214 5215 /* 5216 * in the first pass through the devices list, we gather information 5217 * about the available holes on each device. 5218 */ 5219 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5220 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5221 WARN(1, KERN_ERR 5222 "BTRFS: read-only device in alloc_list\n"); 5223 continue; 5224 } 5225 5226 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5227 &device->dev_state) || 5228 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5229 continue; 5230 5231 if (device->total_bytes > device->bytes_used) 5232 total_avail = device->total_bytes - device->bytes_used; 5233 else 5234 total_avail = 0; 5235 5236 /* If there is no space on this device, skip it. */ 5237 if (total_avail < ctl->dev_extent_min) 5238 continue; 5239 5240 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5241 &max_avail); 5242 if (ret && ret != -ENOSPC) 5243 return ret; 5244 5245 if (ret == 0) 5246 max_avail = dev_extent_want; 5247 5248 if (max_avail < ctl->dev_extent_min) { 5249 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5250 btrfs_debug(info, 5251 "%s: devid %llu has no free space, have=%llu want=%llu", 5252 __func__, device->devid, max_avail, 5253 ctl->dev_extent_min); 5254 continue; 5255 } 5256 5257 if (ndevs == fs_devices->rw_devices) { 5258 WARN(1, "%s: found more than %llu devices\n", 5259 __func__, fs_devices->rw_devices); 5260 break; 5261 } 5262 devices_info[ndevs].dev_offset = dev_offset; 5263 devices_info[ndevs].max_avail = max_avail; 5264 devices_info[ndevs].total_avail = total_avail; 5265 devices_info[ndevs].dev = device; 5266 ++ndevs; 5267 } 5268 ctl->ndevs = ndevs; 5269 5270 /* 5271 * now sort the devices by hole size / available space 5272 */ 5273 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5274 btrfs_cmp_device_info, NULL); 5275 5276 return 0; 5277 } 5278 5279 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5280 struct btrfs_device_info *devices_info) 5281 { 5282 /* Number of stripes that count for block group size */ 5283 int data_stripes; 5284 5285 /* 5286 * The primary goal is to maximize the number of stripes, so use as 5287 * many devices as possible, even if the stripes are not maximum sized. 5288 * 5289 * The DUP profile stores more than one stripe per device, the 5290 * max_avail is the total size so we have to adjust. 5291 */ 5292 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5293 ctl->dev_stripes); 5294 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5295 5296 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5297 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5298 5299 /* 5300 * Use the number of data stripes to figure out how big this chunk is 5301 * really going to be in terms of logical address space, and compare 5302 * that answer with the max chunk size. If it's higher, we try to 5303 * reduce stripe_size. 5304 */ 5305 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5306 /* 5307 * Reduce stripe_size, round it up to a 16MB boundary again and 5308 * then use it, unless it ends up being even bigger than the 5309 * previous value we had already. 5310 */ 5311 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5312 data_stripes), SZ_16M), 5313 ctl->stripe_size); 5314 } 5315 5316 /* Stripe size should not go beyond 1G. */ 5317 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); 5318 5319 /* Align to BTRFS_STRIPE_LEN */ 5320 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5321 ctl->chunk_size = ctl->stripe_size * data_stripes; 5322 5323 return 0; 5324 } 5325 5326 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5327 struct btrfs_device_info *devices_info) 5328 { 5329 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5330 /* Number of stripes that count for block group size */ 5331 int data_stripes; 5332 5333 /* 5334 * It should hold because: 5335 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5336 */ 5337 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5338 5339 ctl->stripe_size = zone_size; 5340 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5341 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5342 5343 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5344 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5345 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5346 ctl->stripe_size) + ctl->nparity, 5347 ctl->dev_stripes); 5348 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5349 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5350 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5351 } 5352 5353 ctl->chunk_size = ctl->stripe_size * data_stripes; 5354 5355 return 0; 5356 } 5357 5358 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5359 struct alloc_chunk_ctl *ctl, 5360 struct btrfs_device_info *devices_info) 5361 { 5362 struct btrfs_fs_info *info = fs_devices->fs_info; 5363 5364 /* 5365 * Round down to number of usable stripes, devs_increment can be any 5366 * number so we can't use round_down() that requires power of 2, while 5367 * rounddown is safe. 5368 */ 5369 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5370 5371 if (ctl->ndevs < ctl->devs_min) { 5372 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5373 btrfs_debug(info, 5374 "%s: not enough devices with free space: have=%d minimum required=%d", 5375 __func__, ctl->ndevs, ctl->devs_min); 5376 } 5377 return -ENOSPC; 5378 } 5379 5380 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5381 5382 switch (fs_devices->chunk_alloc_policy) { 5383 case BTRFS_CHUNK_ALLOC_REGULAR: 5384 return decide_stripe_size_regular(ctl, devices_info); 5385 case BTRFS_CHUNK_ALLOC_ZONED: 5386 return decide_stripe_size_zoned(ctl, devices_info); 5387 default: 5388 BUG(); 5389 } 5390 } 5391 5392 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5393 struct alloc_chunk_ctl *ctl, 5394 struct btrfs_device_info *devices_info) 5395 { 5396 struct btrfs_fs_info *info = trans->fs_info; 5397 struct map_lookup *map = NULL; 5398 struct extent_map_tree *em_tree; 5399 struct btrfs_block_group *block_group; 5400 struct extent_map *em; 5401 u64 start = ctl->start; 5402 u64 type = ctl->type; 5403 int ret; 5404 int i; 5405 int j; 5406 5407 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5408 if (!map) 5409 return ERR_PTR(-ENOMEM); 5410 map->num_stripes = ctl->num_stripes; 5411 5412 for (i = 0; i < ctl->ndevs; ++i) { 5413 for (j = 0; j < ctl->dev_stripes; ++j) { 5414 int s = i * ctl->dev_stripes + j; 5415 map->stripes[s].dev = devices_info[i].dev; 5416 map->stripes[s].physical = devices_info[i].dev_offset + 5417 j * ctl->stripe_size; 5418 } 5419 } 5420 map->io_align = BTRFS_STRIPE_LEN; 5421 map->io_width = BTRFS_STRIPE_LEN; 5422 map->type = type; 5423 map->sub_stripes = ctl->sub_stripes; 5424 5425 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5426 5427 em = alloc_extent_map(); 5428 if (!em) { 5429 kfree(map); 5430 return ERR_PTR(-ENOMEM); 5431 } 5432 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5433 em->map_lookup = map; 5434 em->start = start; 5435 em->len = ctl->chunk_size; 5436 em->block_start = 0; 5437 em->block_len = em->len; 5438 em->orig_block_len = ctl->stripe_size; 5439 5440 em_tree = &info->mapping_tree; 5441 write_lock(&em_tree->lock); 5442 ret = add_extent_mapping(em_tree, em, 0); 5443 if (ret) { 5444 write_unlock(&em_tree->lock); 5445 free_extent_map(em); 5446 return ERR_PTR(ret); 5447 } 5448 write_unlock(&em_tree->lock); 5449 5450 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size); 5451 if (IS_ERR(block_group)) 5452 goto error_del_extent; 5453 5454 for (i = 0; i < map->num_stripes; i++) { 5455 struct btrfs_device *dev = map->stripes[i].dev; 5456 5457 btrfs_device_set_bytes_used(dev, 5458 dev->bytes_used + ctl->stripe_size); 5459 if (list_empty(&dev->post_commit_list)) 5460 list_add_tail(&dev->post_commit_list, 5461 &trans->transaction->dev_update_list); 5462 } 5463 5464 atomic64_sub(ctl->stripe_size * map->num_stripes, 5465 &info->free_chunk_space); 5466 5467 free_extent_map(em); 5468 check_raid56_incompat_flag(info, type); 5469 check_raid1c34_incompat_flag(info, type); 5470 5471 return block_group; 5472 5473 error_del_extent: 5474 write_lock(&em_tree->lock); 5475 remove_extent_mapping(em_tree, em); 5476 write_unlock(&em_tree->lock); 5477 5478 /* One for our allocation */ 5479 free_extent_map(em); 5480 /* One for the tree reference */ 5481 free_extent_map(em); 5482 5483 return block_group; 5484 } 5485 5486 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5487 u64 type) 5488 { 5489 struct btrfs_fs_info *info = trans->fs_info; 5490 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5491 struct btrfs_device_info *devices_info = NULL; 5492 struct alloc_chunk_ctl ctl; 5493 struct btrfs_block_group *block_group; 5494 int ret; 5495 5496 lockdep_assert_held(&info->chunk_mutex); 5497 5498 if (!alloc_profile_is_valid(type, 0)) { 5499 ASSERT(0); 5500 return ERR_PTR(-EINVAL); 5501 } 5502 5503 if (list_empty(&fs_devices->alloc_list)) { 5504 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5505 btrfs_debug(info, "%s: no writable device", __func__); 5506 return ERR_PTR(-ENOSPC); 5507 } 5508 5509 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5510 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5511 ASSERT(0); 5512 return ERR_PTR(-EINVAL); 5513 } 5514 5515 ctl.start = find_next_chunk(info); 5516 ctl.type = type; 5517 init_alloc_chunk_ctl(fs_devices, &ctl); 5518 5519 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5520 GFP_NOFS); 5521 if (!devices_info) 5522 return ERR_PTR(-ENOMEM); 5523 5524 ret = gather_device_info(fs_devices, &ctl, devices_info); 5525 if (ret < 0) { 5526 block_group = ERR_PTR(ret); 5527 goto out; 5528 } 5529 5530 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5531 if (ret < 0) { 5532 block_group = ERR_PTR(ret); 5533 goto out; 5534 } 5535 5536 block_group = create_chunk(trans, &ctl, devices_info); 5537 5538 out: 5539 kfree(devices_info); 5540 return block_group; 5541 } 5542 5543 /* 5544 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5545 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5546 * chunks. 5547 * 5548 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5549 * phases. 5550 */ 5551 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5552 struct btrfs_block_group *bg) 5553 { 5554 struct btrfs_fs_info *fs_info = trans->fs_info; 5555 struct btrfs_root *chunk_root = fs_info->chunk_root; 5556 struct btrfs_key key; 5557 struct btrfs_chunk *chunk; 5558 struct btrfs_stripe *stripe; 5559 struct extent_map *em; 5560 struct map_lookup *map; 5561 size_t item_size; 5562 int i; 5563 int ret; 5564 5565 /* 5566 * We take the chunk_mutex for 2 reasons: 5567 * 5568 * 1) Updates and insertions in the chunk btree must be done while holding 5569 * the chunk_mutex, as well as updating the system chunk array in the 5570 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5571 * details; 5572 * 5573 * 2) To prevent races with the final phase of a device replace operation 5574 * that replaces the device object associated with the map's stripes, 5575 * because the device object's id can change at any time during that 5576 * final phase of the device replace operation 5577 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5578 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5579 * which would cause a failure when updating the device item, which does 5580 * not exists, or persisting a stripe of the chunk item with such ID. 5581 * Here we can't use the device_list_mutex because our caller already 5582 * has locked the chunk_mutex, and the final phase of device replace 5583 * acquires both mutexes - first the device_list_mutex and then the 5584 * chunk_mutex. Using any of those two mutexes protects us from a 5585 * concurrent device replace. 5586 */ 5587 lockdep_assert_held(&fs_info->chunk_mutex); 5588 5589 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5590 if (IS_ERR(em)) { 5591 ret = PTR_ERR(em); 5592 btrfs_abort_transaction(trans, ret); 5593 return ret; 5594 } 5595 5596 map = em->map_lookup; 5597 item_size = btrfs_chunk_item_size(map->num_stripes); 5598 5599 chunk = kzalloc(item_size, GFP_NOFS); 5600 if (!chunk) { 5601 ret = -ENOMEM; 5602 btrfs_abort_transaction(trans, ret); 5603 goto out; 5604 } 5605 5606 for (i = 0; i < map->num_stripes; i++) { 5607 struct btrfs_device *device = map->stripes[i].dev; 5608 5609 ret = btrfs_update_device(trans, device); 5610 if (ret) 5611 goto out; 5612 } 5613 5614 stripe = &chunk->stripe; 5615 for (i = 0; i < map->num_stripes; i++) { 5616 struct btrfs_device *device = map->stripes[i].dev; 5617 const u64 dev_offset = map->stripes[i].physical; 5618 5619 btrfs_set_stack_stripe_devid(stripe, device->devid); 5620 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5621 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5622 stripe++; 5623 } 5624 5625 btrfs_set_stack_chunk_length(chunk, bg->length); 5626 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5627 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN); 5628 btrfs_set_stack_chunk_type(chunk, map->type); 5629 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5630 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN); 5631 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN); 5632 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5633 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5634 5635 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5636 key.type = BTRFS_CHUNK_ITEM_KEY; 5637 key.offset = bg->start; 5638 5639 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5640 if (ret) 5641 goto out; 5642 5643 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); 5644 5645 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5646 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5647 if (ret) 5648 goto out; 5649 } 5650 5651 out: 5652 kfree(chunk); 5653 free_extent_map(em); 5654 return ret; 5655 } 5656 5657 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5658 { 5659 struct btrfs_fs_info *fs_info = trans->fs_info; 5660 u64 alloc_profile; 5661 struct btrfs_block_group *meta_bg; 5662 struct btrfs_block_group *sys_bg; 5663 5664 /* 5665 * When adding a new device for sprouting, the seed device is read-only 5666 * so we must first allocate a metadata and a system chunk. But before 5667 * adding the block group items to the extent, device and chunk btrees, 5668 * we must first: 5669 * 5670 * 1) Create both chunks without doing any changes to the btrees, as 5671 * otherwise we would get -ENOSPC since the block groups from the 5672 * seed device are read-only; 5673 * 5674 * 2) Add the device item for the new sprout device - finishing the setup 5675 * of a new block group requires updating the device item in the chunk 5676 * btree, so it must exist when we attempt to do it. The previous step 5677 * ensures this does not fail with -ENOSPC. 5678 * 5679 * After that we can add the block group items to their btrees: 5680 * update existing device item in the chunk btree, add a new block group 5681 * item to the extent btree, add a new chunk item to the chunk btree and 5682 * finally add the new device extent items to the devices btree. 5683 */ 5684 5685 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5686 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5687 if (IS_ERR(meta_bg)) 5688 return PTR_ERR(meta_bg); 5689 5690 alloc_profile = btrfs_system_alloc_profile(fs_info); 5691 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5692 if (IS_ERR(sys_bg)) 5693 return PTR_ERR(sys_bg); 5694 5695 return 0; 5696 } 5697 5698 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5699 { 5700 const int index = btrfs_bg_flags_to_raid_index(map->type); 5701 5702 return btrfs_raid_array[index].tolerated_failures; 5703 } 5704 5705 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5706 { 5707 struct extent_map *em; 5708 struct map_lookup *map; 5709 int miss_ndevs = 0; 5710 int i; 5711 bool ret = true; 5712 5713 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5714 if (IS_ERR(em)) 5715 return false; 5716 5717 map = em->map_lookup; 5718 for (i = 0; i < map->num_stripes; i++) { 5719 if (test_bit(BTRFS_DEV_STATE_MISSING, 5720 &map->stripes[i].dev->dev_state)) { 5721 miss_ndevs++; 5722 continue; 5723 } 5724 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5725 &map->stripes[i].dev->dev_state)) { 5726 ret = false; 5727 goto end; 5728 } 5729 } 5730 5731 /* 5732 * If the number of missing devices is larger than max errors, we can 5733 * not write the data into that chunk successfully. 5734 */ 5735 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5736 ret = false; 5737 end: 5738 free_extent_map(em); 5739 return ret; 5740 } 5741 5742 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5743 { 5744 struct extent_map *em; 5745 5746 while (1) { 5747 write_lock(&tree->lock); 5748 em = lookup_extent_mapping(tree, 0, (u64)-1); 5749 if (em) 5750 remove_extent_mapping(tree, em); 5751 write_unlock(&tree->lock); 5752 if (!em) 5753 break; 5754 /* once for us */ 5755 free_extent_map(em); 5756 /* once for the tree */ 5757 free_extent_map(em); 5758 } 5759 } 5760 5761 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5762 { 5763 struct extent_map *em; 5764 struct map_lookup *map; 5765 enum btrfs_raid_types index; 5766 int ret = 1; 5767 5768 em = btrfs_get_chunk_map(fs_info, logical, len); 5769 if (IS_ERR(em)) 5770 /* 5771 * We could return errors for these cases, but that could get 5772 * ugly and we'd probably do the same thing which is just not do 5773 * anything else and exit, so return 1 so the callers don't try 5774 * to use other copies. 5775 */ 5776 return 1; 5777 5778 map = em->map_lookup; 5779 index = btrfs_bg_flags_to_raid_index(map->type); 5780 5781 /* Non-RAID56, use their ncopies from btrfs_raid_array. */ 5782 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5783 ret = btrfs_raid_array[index].ncopies; 5784 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5785 ret = 2; 5786 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5787 /* 5788 * There could be two corrupted data stripes, we need 5789 * to loop retry in order to rebuild the correct data. 5790 * 5791 * Fail a stripe at a time on every retry except the 5792 * stripe under reconstruction. 5793 */ 5794 ret = map->num_stripes; 5795 free_extent_map(em); 5796 return ret; 5797 } 5798 5799 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5800 u64 logical) 5801 { 5802 struct extent_map *em; 5803 struct map_lookup *map; 5804 unsigned long len = fs_info->sectorsize; 5805 5806 if (!btrfs_fs_incompat(fs_info, RAID56)) 5807 return len; 5808 5809 em = btrfs_get_chunk_map(fs_info, logical, len); 5810 5811 if (!WARN_ON(IS_ERR(em))) { 5812 map = em->map_lookup; 5813 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5814 len = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT; 5815 free_extent_map(em); 5816 } 5817 return len; 5818 } 5819 5820 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5821 { 5822 struct extent_map *em; 5823 struct map_lookup *map; 5824 int ret = 0; 5825 5826 if (!btrfs_fs_incompat(fs_info, RAID56)) 5827 return 0; 5828 5829 em = btrfs_get_chunk_map(fs_info, logical, len); 5830 5831 if(!WARN_ON(IS_ERR(em))) { 5832 map = em->map_lookup; 5833 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5834 ret = 1; 5835 free_extent_map(em); 5836 } 5837 return ret; 5838 } 5839 5840 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5841 struct map_lookup *map, int first, 5842 int dev_replace_is_ongoing) 5843 { 5844 int i; 5845 int num_stripes; 5846 int preferred_mirror; 5847 int tolerance; 5848 struct btrfs_device *srcdev; 5849 5850 ASSERT((map->type & 5851 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5852 5853 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5854 num_stripes = map->sub_stripes; 5855 else 5856 num_stripes = map->num_stripes; 5857 5858 switch (fs_info->fs_devices->read_policy) { 5859 default: 5860 /* Shouldn't happen, just warn and use pid instead of failing */ 5861 btrfs_warn_rl(fs_info, 5862 "unknown read_policy type %u, reset to pid", 5863 fs_info->fs_devices->read_policy); 5864 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5865 fallthrough; 5866 case BTRFS_READ_POLICY_PID: 5867 preferred_mirror = first + (current->pid % num_stripes); 5868 break; 5869 } 5870 5871 if (dev_replace_is_ongoing && 5872 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5873 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5874 srcdev = fs_info->dev_replace.srcdev; 5875 else 5876 srcdev = NULL; 5877 5878 /* 5879 * try to avoid the drive that is the source drive for a 5880 * dev-replace procedure, only choose it if no other non-missing 5881 * mirror is available 5882 */ 5883 for (tolerance = 0; tolerance < 2; tolerance++) { 5884 if (map->stripes[preferred_mirror].dev->bdev && 5885 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5886 return preferred_mirror; 5887 for (i = first; i < first + num_stripes; i++) { 5888 if (map->stripes[i].dev->bdev && 5889 (tolerance || map->stripes[i].dev != srcdev)) 5890 return i; 5891 } 5892 } 5893 5894 /* we couldn't find one that doesn't fail. Just return something 5895 * and the io error handling code will clean up eventually 5896 */ 5897 return preferred_mirror; 5898 } 5899 5900 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5901 u16 total_stripes) 5902 { 5903 struct btrfs_io_context *bioc; 5904 5905 bioc = kzalloc( 5906 /* The size of btrfs_io_context */ 5907 sizeof(struct btrfs_io_context) + 5908 /* Plus the variable array for the stripes */ 5909 sizeof(struct btrfs_io_stripe) * (total_stripes), 5910 GFP_NOFS); 5911 5912 if (!bioc) 5913 return NULL; 5914 5915 refcount_set(&bioc->refs, 1); 5916 5917 bioc->fs_info = fs_info; 5918 bioc->replace_stripe_src = -1; 5919 bioc->full_stripe_logical = (u64)-1; 5920 5921 return bioc; 5922 } 5923 5924 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5925 { 5926 WARN_ON(!refcount_read(&bioc->refs)); 5927 refcount_inc(&bioc->refs); 5928 } 5929 5930 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5931 { 5932 if (!bioc) 5933 return; 5934 if (refcount_dec_and_test(&bioc->refs)) 5935 kfree(bioc); 5936 } 5937 5938 /* 5939 * Please note that, discard won't be sent to target device of device 5940 * replace. 5941 */ 5942 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, 5943 u64 logical, u64 *length_ret, 5944 u32 *num_stripes) 5945 { 5946 struct extent_map *em; 5947 struct map_lookup *map; 5948 struct btrfs_discard_stripe *stripes; 5949 u64 length = *length_ret; 5950 u64 offset; 5951 u32 stripe_nr; 5952 u32 stripe_nr_end; 5953 u32 stripe_cnt; 5954 u64 stripe_end_offset; 5955 u64 stripe_offset; 5956 u32 stripe_index; 5957 u32 factor = 0; 5958 u32 sub_stripes = 0; 5959 u32 stripes_per_dev = 0; 5960 u32 remaining_stripes = 0; 5961 u32 last_stripe = 0; 5962 int ret; 5963 int i; 5964 5965 em = btrfs_get_chunk_map(fs_info, logical, length); 5966 if (IS_ERR(em)) 5967 return ERR_CAST(em); 5968 5969 map = em->map_lookup; 5970 5971 /* we don't discard raid56 yet */ 5972 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5973 ret = -EOPNOTSUPP; 5974 goto out_free_map; 5975 } 5976 5977 offset = logical - em->start; 5978 length = min_t(u64, em->start + em->len - logical, length); 5979 *length_ret = length; 5980 5981 /* 5982 * stripe_nr counts the total number of stripes we have to stride 5983 * to get to this block 5984 */ 5985 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 5986 5987 /* stripe_offset is the offset of this block in its stripe */ 5988 stripe_offset = offset - (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 5989 5990 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >> 5991 BTRFS_STRIPE_LEN_SHIFT; 5992 stripe_cnt = stripe_nr_end - stripe_nr; 5993 stripe_end_offset = (stripe_nr_end << BTRFS_STRIPE_LEN_SHIFT) - 5994 (offset + length); 5995 /* 5996 * after this, stripe_nr is the number of stripes on this 5997 * device we have to walk to find the data, and stripe_index is 5998 * the number of our device in the stripe array 5999 */ 6000 *num_stripes = 1; 6001 stripe_index = 0; 6002 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6003 BTRFS_BLOCK_GROUP_RAID10)) { 6004 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 6005 sub_stripes = 1; 6006 else 6007 sub_stripes = map->sub_stripes; 6008 6009 factor = map->num_stripes / sub_stripes; 6010 *num_stripes = min_t(u64, map->num_stripes, 6011 sub_stripes * stripe_cnt); 6012 stripe_index = stripe_nr % factor; 6013 stripe_nr /= factor; 6014 stripe_index *= sub_stripes; 6015 6016 remaining_stripes = stripe_cnt % factor; 6017 stripes_per_dev = stripe_cnt / factor; 6018 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes; 6019 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 6020 BTRFS_BLOCK_GROUP_DUP)) { 6021 *num_stripes = map->num_stripes; 6022 } else { 6023 stripe_index = stripe_nr % map->num_stripes; 6024 stripe_nr /= map->num_stripes; 6025 } 6026 6027 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); 6028 if (!stripes) { 6029 ret = -ENOMEM; 6030 goto out_free_map; 6031 } 6032 6033 for (i = 0; i < *num_stripes; i++) { 6034 stripes[i].physical = 6035 map->stripes[stripe_index].physical + 6036 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6037 stripes[i].dev = map->stripes[stripe_index].dev; 6038 6039 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6040 BTRFS_BLOCK_GROUP_RAID10)) { 6041 stripes[i].length = stripes_per_dev << BTRFS_STRIPE_LEN_SHIFT; 6042 6043 if (i / sub_stripes < remaining_stripes) 6044 stripes[i].length += BTRFS_STRIPE_LEN; 6045 6046 /* 6047 * Special for the first stripe and 6048 * the last stripe: 6049 * 6050 * |-------|...|-------| 6051 * |----------| 6052 * off end_off 6053 */ 6054 if (i < sub_stripes) 6055 stripes[i].length -= stripe_offset; 6056 6057 if (stripe_index >= last_stripe && 6058 stripe_index <= (last_stripe + 6059 sub_stripes - 1)) 6060 stripes[i].length -= stripe_end_offset; 6061 6062 if (i == sub_stripes - 1) 6063 stripe_offset = 0; 6064 } else { 6065 stripes[i].length = length; 6066 } 6067 6068 stripe_index++; 6069 if (stripe_index == map->num_stripes) { 6070 stripe_index = 0; 6071 stripe_nr++; 6072 } 6073 } 6074 6075 free_extent_map(em); 6076 return stripes; 6077 out_free_map: 6078 free_extent_map(em); 6079 return ERR_PTR(ret); 6080 } 6081 6082 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6083 { 6084 struct btrfs_block_group *cache; 6085 bool ret; 6086 6087 /* Non zoned filesystem does not use "to_copy" flag */ 6088 if (!btrfs_is_zoned(fs_info)) 6089 return false; 6090 6091 cache = btrfs_lookup_block_group(fs_info, logical); 6092 6093 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 6094 6095 btrfs_put_block_group(cache); 6096 return ret; 6097 } 6098 6099 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6100 struct btrfs_io_context *bioc, 6101 struct btrfs_dev_replace *dev_replace, 6102 u64 logical, 6103 int *num_stripes_ret, int *max_errors_ret) 6104 { 6105 u64 srcdev_devid = dev_replace->srcdev->devid; 6106 /* 6107 * At this stage, num_stripes is still the real number of stripes, 6108 * excluding the duplicated stripes. 6109 */ 6110 int num_stripes = *num_stripes_ret; 6111 int nr_extra_stripes = 0; 6112 int max_errors = *max_errors_ret; 6113 int i; 6114 6115 /* 6116 * A block group which has "to_copy" set will eventually be copied by 6117 * the dev-replace process. We can avoid cloning IO here. 6118 */ 6119 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6120 return; 6121 6122 /* 6123 * Duplicate the write operations while the dev-replace procedure is 6124 * running. Since the copying of the old disk to the new disk takes 6125 * place at run time while the filesystem is mounted writable, the 6126 * regular write operations to the old disk have to be duplicated to go 6127 * to the new disk as well. 6128 * 6129 * Note that device->missing is handled by the caller, and that the 6130 * write to the old disk is already set up in the stripes array. 6131 */ 6132 for (i = 0; i < num_stripes; i++) { 6133 struct btrfs_io_stripe *old = &bioc->stripes[i]; 6134 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes]; 6135 6136 if (old->dev->devid != srcdev_devid) 6137 continue; 6138 6139 new->physical = old->physical; 6140 new->dev = dev_replace->tgtdev; 6141 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) 6142 bioc->replace_stripe_src = i; 6143 nr_extra_stripes++; 6144 } 6145 6146 /* We can only have at most 2 extra nr_stripes (for DUP). */ 6147 ASSERT(nr_extra_stripes <= 2); 6148 /* 6149 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for 6150 * replace. 6151 * If we have 2 extra stripes, only choose the one with smaller physical. 6152 */ 6153 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) { 6154 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes]; 6155 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1]; 6156 6157 /* Only DUP can have two extra stripes. */ 6158 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP); 6159 6160 /* 6161 * Swap the last stripe stripes and reduce @nr_extra_stripes. 6162 * The extra stripe would still be there, but won't be accessed. 6163 */ 6164 if (first->physical > second->physical) { 6165 swap(second->physical, first->physical); 6166 swap(second->dev, first->dev); 6167 nr_extra_stripes--; 6168 } 6169 } 6170 6171 *num_stripes_ret = num_stripes + nr_extra_stripes; 6172 *max_errors_ret = max_errors + nr_extra_stripes; 6173 bioc->replace_nr_stripes = nr_extra_stripes; 6174 } 6175 6176 static bool need_full_stripe(enum btrfs_map_op op) 6177 { 6178 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6179 } 6180 6181 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op, 6182 u64 offset, u32 *stripe_nr, u64 *stripe_offset, 6183 u64 *full_stripe_start) 6184 { 6185 /* 6186 * Stripe_nr is the stripe where this block falls. stripe_offset is 6187 * the offset of this block in its stripe. 6188 */ 6189 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK; 6190 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 6191 ASSERT(*stripe_offset < U32_MAX); 6192 6193 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6194 unsigned long full_stripe_len = nr_data_stripes(map) << 6195 BTRFS_STRIPE_LEN_SHIFT; 6196 6197 /* 6198 * For full stripe start, we use previously calculated 6199 * @stripe_nr. Align it to nr_data_stripes, then multiply with 6200 * STRIPE_LEN. 6201 * 6202 * By this we can avoid u64 division completely. And we have 6203 * to go rounddown(), not round_down(), as nr_data_stripes is 6204 * not ensured to be power of 2. 6205 */ 6206 *full_stripe_start = 6207 rounddown(*stripe_nr, nr_data_stripes(map)) << 6208 BTRFS_STRIPE_LEN_SHIFT; 6209 6210 /* 6211 * For writes to RAID56, allow to write a full stripe set, but 6212 * no straddling of stripe sets. 6213 */ 6214 if (op == BTRFS_MAP_WRITE) 6215 return full_stripe_len - (offset - *full_stripe_start); 6216 } 6217 6218 /* 6219 * For other RAID types and for RAID56 reads, allow a single stripe (on 6220 * a single disk). 6221 */ 6222 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) 6223 return BTRFS_STRIPE_LEN - *stripe_offset; 6224 return U64_MAX; 6225 } 6226 6227 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, 6228 u32 stripe_index, u64 stripe_offset, u32 stripe_nr) 6229 { 6230 dst->dev = map->stripes[stripe_index].dev; 6231 dst->physical = map->stripes[stripe_index].physical + 6232 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6233 } 6234 6235 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6236 u64 logical, u64 *length, 6237 struct btrfs_io_context **bioc_ret, 6238 struct btrfs_io_stripe *smap, int *mirror_num_ret, 6239 int need_raid_map) 6240 { 6241 struct extent_map *em; 6242 struct map_lookup *map; 6243 u64 map_offset; 6244 u64 stripe_offset; 6245 u32 stripe_nr; 6246 u32 stripe_index; 6247 int data_stripes; 6248 int i; 6249 int ret = 0; 6250 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); 6251 int num_stripes; 6252 int num_copies; 6253 int max_errors = 0; 6254 struct btrfs_io_context *bioc = NULL; 6255 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6256 int dev_replace_is_ongoing = 0; 6257 u16 num_alloc_stripes; 6258 u64 raid56_full_stripe_start = (u64)-1; 6259 u64 max_len; 6260 6261 ASSERT(bioc_ret); 6262 6263 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize); 6264 if (mirror_num > num_copies) 6265 return -EINVAL; 6266 6267 em = btrfs_get_chunk_map(fs_info, logical, *length); 6268 if (IS_ERR(em)) 6269 return PTR_ERR(em); 6270 6271 map = em->map_lookup; 6272 data_stripes = nr_data_stripes(map); 6273 6274 map_offset = logical - em->start; 6275 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr, 6276 &stripe_offset, &raid56_full_stripe_start); 6277 *length = min_t(u64, em->len - map_offset, max_len); 6278 6279 down_read(&dev_replace->rwsem); 6280 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6281 /* 6282 * Hold the semaphore for read during the whole operation, write is 6283 * requested at commit time but must wait. 6284 */ 6285 if (!dev_replace_is_ongoing) 6286 up_read(&dev_replace->rwsem); 6287 6288 num_stripes = 1; 6289 stripe_index = 0; 6290 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6291 stripe_index = stripe_nr % map->num_stripes; 6292 stripe_nr /= map->num_stripes; 6293 if (!need_full_stripe(op)) 6294 mirror_num = 1; 6295 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6296 if (need_full_stripe(op)) 6297 num_stripes = map->num_stripes; 6298 else if (mirror_num) 6299 stripe_index = mirror_num - 1; 6300 else { 6301 stripe_index = find_live_mirror(fs_info, map, 0, 6302 dev_replace_is_ongoing); 6303 mirror_num = stripe_index + 1; 6304 } 6305 6306 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6307 if (need_full_stripe(op)) { 6308 num_stripes = map->num_stripes; 6309 } else if (mirror_num) { 6310 stripe_index = mirror_num - 1; 6311 } else { 6312 mirror_num = 1; 6313 } 6314 6315 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6316 u32 factor = map->num_stripes / map->sub_stripes; 6317 6318 stripe_index = (stripe_nr % factor) * map->sub_stripes; 6319 stripe_nr /= factor; 6320 6321 if (need_full_stripe(op)) 6322 num_stripes = map->sub_stripes; 6323 else if (mirror_num) 6324 stripe_index += mirror_num - 1; 6325 else { 6326 int old_stripe_index = stripe_index; 6327 stripe_index = find_live_mirror(fs_info, map, 6328 stripe_index, 6329 dev_replace_is_ongoing); 6330 mirror_num = stripe_index - old_stripe_index + 1; 6331 } 6332 6333 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6334 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6335 /* 6336 * Push stripe_nr back to the start of the full stripe 6337 * For those cases needing a full stripe, @stripe_nr 6338 * is the full stripe number. 6339 * 6340 * Originally we go raid56_full_stripe_start / full_stripe_len, 6341 * but that can be expensive. Here we just divide 6342 * @stripe_nr with @data_stripes. 6343 */ 6344 stripe_nr /= data_stripes; 6345 6346 /* RAID[56] write or recovery. Return all stripes */ 6347 num_stripes = map->num_stripes; 6348 max_errors = btrfs_chunk_max_errors(map); 6349 6350 /* Return the length to the full stripe end */ 6351 *length = min(logical + *length, 6352 raid56_full_stripe_start + em->start + 6353 (data_stripes << BTRFS_STRIPE_LEN_SHIFT)) - logical; 6354 stripe_index = 0; 6355 stripe_offset = 0; 6356 } else { 6357 /* 6358 * Mirror #0 or #1 means the original data block. 6359 * Mirror #2 is RAID5 parity block. 6360 * Mirror #3 is RAID6 Q block. 6361 */ 6362 stripe_index = stripe_nr % data_stripes; 6363 stripe_nr /= data_stripes; 6364 if (mirror_num > 1) 6365 stripe_index = data_stripes + mirror_num - 2; 6366 6367 /* We distribute the parity blocks across stripes */ 6368 stripe_index = (stripe_nr + stripe_index) % map->num_stripes; 6369 if (!need_full_stripe(op) && mirror_num <= 1) 6370 mirror_num = 1; 6371 } 6372 } else { 6373 /* 6374 * After this, stripe_nr is the number of stripes on this 6375 * device we have to walk to find the data, and stripe_index is 6376 * the number of our device in the stripe array 6377 */ 6378 stripe_index = stripe_nr % map->num_stripes; 6379 stripe_nr /= map->num_stripes; 6380 mirror_num = stripe_index + 1; 6381 } 6382 if (stripe_index >= map->num_stripes) { 6383 btrfs_crit(fs_info, 6384 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6385 stripe_index, map->num_stripes); 6386 ret = -EINVAL; 6387 goto out; 6388 } 6389 6390 num_alloc_stripes = num_stripes; 6391 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6392 op != BTRFS_MAP_READ) 6393 /* 6394 * For replace case, we need to add extra stripes for extra 6395 * duplicated stripes. 6396 * 6397 * For both WRITE and GET_READ_MIRRORS, we may have at most 6398 * 2 more stripes (DUP types, otherwise 1). 6399 */ 6400 num_alloc_stripes += 2; 6401 6402 /* 6403 * If this I/O maps to a single device, try to return the device and 6404 * physical block information on the stack instead of allocating an 6405 * I/O context structure. 6406 */ 6407 if (smap && num_alloc_stripes == 1 && 6408 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && 6409 (!need_full_stripe(op) || !dev_replace_is_ongoing || 6410 !dev_replace->tgtdev)) { 6411 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr); 6412 *mirror_num_ret = mirror_num; 6413 *bioc_ret = NULL; 6414 ret = 0; 6415 goto out; 6416 } 6417 6418 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes); 6419 if (!bioc) { 6420 ret = -ENOMEM; 6421 goto out; 6422 } 6423 bioc->map_type = map->type; 6424 6425 /* 6426 * For RAID56 full map, we need to make sure the stripes[] follows the 6427 * rule that data stripes are all ordered, then followed with P and Q 6428 * (if we have). 6429 * 6430 * It's still mostly the same as other profiles, just with extra rotation. 6431 */ 6432 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6433 (need_full_stripe(op) || mirror_num > 1)) { 6434 /* 6435 * For RAID56 @stripe_nr is already the number of full stripes 6436 * before us, which is also the rotation value (needs to modulo 6437 * with num_stripes). 6438 * 6439 * In this case, we just add @stripe_nr with @i, then do the 6440 * modulo, to reduce one modulo call. 6441 */ 6442 bioc->full_stripe_logical = em->start + 6443 ((stripe_nr * data_stripes) << BTRFS_STRIPE_LEN_SHIFT); 6444 for (i = 0; i < num_stripes; i++) 6445 set_io_stripe(&bioc->stripes[i], map, 6446 (i + stripe_nr) % num_stripes, 6447 stripe_offset, stripe_nr); 6448 } else { 6449 /* 6450 * For all other non-RAID56 profiles, just copy the target 6451 * stripe into the bioc. 6452 */ 6453 for (i = 0; i < num_stripes; i++) { 6454 set_io_stripe(&bioc->stripes[i], map, stripe_index, 6455 stripe_offset, stripe_nr); 6456 stripe_index++; 6457 } 6458 } 6459 6460 if (need_full_stripe(op)) 6461 max_errors = btrfs_chunk_max_errors(map); 6462 6463 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6464 need_full_stripe(op)) { 6465 handle_ops_on_dev_replace(op, bioc, dev_replace, logical, 6466 &num_stripes, &max_errors); 6467 } 6468 6469 *bioc_ret = bioc; 6470 bioc->num_stripes = num_stripes; 6471 bioc->max_errors = max_errors; 6472 bioc->mirror_num = mirror_num; 6473 6474 out: 6475 if (dev_replace_is_ongoing) { 6476 lockdep_assert_held(&dev_replace->rwsem); 6477 /* Unlock and let waiting writers proceed */ 6478 up_read(&dev_replace->rwsem); 6479 } 6480 free_extent_map(em); 6481 return ret; 6482 } 6483 6484 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6485 const struct btrfs_fs_devices *fs_devices) 6486 { 6487 if (args->fsid == NULL) 6488 return true; 6489 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6490 return true; 6491 return false; 6492 } 6493 6494 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6495 const struct btrfs_device *device) 6496 { 6497 if (args->missing) { 6498 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6499 !device->bdev) 6500 return true; 6501 return false; 6502 } 6503 6504 if (device->devid != args->devid) 6505 return false; 6506 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6507 return false; 6508 return true; 6509 } 6510 6511 /* 6512 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6513 * return NULL. 6514 * 6515 * If devid and uuid are both specified, the match must be exact, otherwise 6516 * only devid is used. 6517 */ 6518 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6519 const struct btrfs_dev_lookup_args *args) 6520 { 6521 struct btrfs_device *device; 6522 struct btrfs_fs_devices *seed_devs; 6523 6524 if (dev_args_match_fs_devices(args, fs_devices)) { 6525 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6526 if (dev_args_match_device(args, device)) 6527 return device; 6528 } 6529 } 6530 6531 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6532 if (!dev_args_match_fs_devices(args, seed_devs)) 6533 continue; 6534 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6535 if (dev_args_match_device(args, device)) 6536 return device; 6537 } 6538 } 6539 6540 return NULL; 6541 } 6542 6543 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6544 u64 devid, u8 *dev_uuid) 6545 { 6546 struct btrfs_device *device; 6547 unsigned int nofs_flag; 6548 6549 /* 6550 * We call this under the chunk_mutex, so we want to use NOFS for this 6551 * allocation, however we don't want to change btrfs_alloc_device() to 6552 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6553 * places. 6554 */ 6555 6556 nofs_flag = memalloc_nofs_save(); 6557 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); 6558 memalloc_nofs_restore(nofs_flag); 6559 if (IS_ERR(device)) 6560 return device; 6561 6562 list_add(&device->dev_list, &fs_devices->devices); 6563 device->fs_devices = fs_devices; 6564 fs_devices->num_devices++; 6565 6566 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6567 fs_devices->missing_devices++; 6568 6569 return device; 6570 } 6571 6572 /* 6573 * Allocate new device struct, set up devid and UUID. 6574 * 6575 * @fs_info: used only for generating a new devid, can be NULL if 6576 * devid is provided (i.e. @devid != NULL). 6577 * @devid: a pointer to devid for this device. If NULL a new devid 6578 * is generated. 6579 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6580 * is generated. 6581 * @path: a pointer to device path if available, NULL otherwise. 6582 * 6583 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6584 * on error. Returned struct is not linked onto any lists and must be 6585 * destroyed with btrfs_free_device. 6586 */ 6587 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6588 const u64 *devid, const u8 *uuid, 6589 const char *path) 6590 { 6591 struct btrfs_device *dev; 6592 u64 tmp; 6593 6594 if (WARN_ON(!devid && !fs_info)) 6595 return ERR_PTR(-EINVAL); 6596 6597 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6598 if (!dev) 6599 return ERR_PTR(-ENOMEM); 6600 6601 INIT_LIST_HEAD(&dev->dev_list); 6602 INIT_LIST_HEAD(&dev->dev_alloc_list); 6603 INIT_LIST_HEAD(&dev->post_commit_list); 6604 6605 atomic_set(&dev->dev_stats_ccnt, 0); 6606 btrfs_device_data_ordered_init(dev); 6607 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); 6608 6609 if (devid) 6610 tmp = *devid; 6611 else { 6612 int ret; 6613 6614 ret = find_next_devid(fs_info, &tmp); 6615 if (ret) { 6616 btrfs_free_device(dev); 6617 return ERR_PTR(ret); 6618 } 6619 } 6620 dev->devid = tmp; 6621 6622 if (uuid) 6623 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6624 else 6625 generate_random_uuid(dev->uuid); 6626 6627 if (path) { 6628 struct rcu_string *name; 6629 6630 name = rcu_string_strdup(path, GFP_KERNEL); 6631 if (!name) { 6632 btrfs_free_device(dev); 6633 return ERR_PTR(-ENOMEM); 6634 } 6635 rcu_assign_pointer(dev->name, name); 6636 } 6637 6638 return dev; 6639 } 6640 6641 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6642 u64 devid, u8 *uuid, bool error) 6643 { 6644 if (error) 6645 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6646 devid, uuid); 6647 else 6648 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6649 devid, uuid); 6650 } 6651 6652 u64 btrfs_calc_stripe_length(const struct extent_map *em) 6653 { 6654 const struct map_lookup *map = em->map_lookup; 6655 const int data_stripes = calc_data_stripes(map->type, map->num_stripes); 6656 6657 return div_u64(em->len, data_stripes); 6658 } 6659 6660 #if BITS_PER_LONG == 32 6661 /* 6662 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6663 * can't be accessed on 32bit systems. 6664 * 6665 * This function do mount time check to reject the fs if it already has 6666 * metadata chunk beyond that limit. 6667 */ 6668 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6669 u64 logical, u64 length, u64 type) 6670 { 6671 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6672 return 0; 6673 6674 if (logical + length < MAX_LFS_FILESIZE) 6675 return 0; 6676 6677 btrfs_err_32bit_limit(fs_info); 6678 return -EOVERFLOW; 6679 } 6680 6681 /* 6682 * This is to give early warning for any metadata chunk reaching 6683 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6684 * Although we can still access the metadata, it's not going to be possible 6685 * once the limit is reached. 6686 */ 6687 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6688 u64 logical, u64 length, u64 type) 6689 { 6690 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6691 return; 6692 6693 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 6694 return; 6695 6696 btrfs_warn_32bit_limit(fs_info); 6697 } 6698 #endif 6699 6700 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 6701 u64 devid, u8 *uuid) 6702 { 6703 struct btrfs_device *dev; 6704 6705 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6706 btrfs_report_missing_device(fs_info, devid, uuid, true); 6707 return ERR_PTR(-ENOENT); 6708 } 6709 6710 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 6711 if (IS_ERR(dev)) { 6712 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 6713 devid, PTR_ERR(dev)); 6714 return dev; 6715 } 6716 btrfs_report_missing_device(fs_info, devid, uuid, false); 6717 6718 return dev; 6719 } 6720 6721 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 6722 struct btrfs_chunk *chunk) 6723 { 6724 BTRFS_DEV_LOOKUP_ARGS(args); 6725 struct btrfs_fs_info *fs_info = leaf->fs_info; 6726 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 6727 struct map_lookup *map; 6728 struct extent_map *em; 6729 u64 logical; 6730 u64 length; 6731 u64 devid; 6732 u64 type; 6733 u8 uuid[BTRFS_UUID_SIZE]; 6734 int index; 6735 int num_stripes; 6736 int ret; 6737 int i; 6738 6739 logical = key->offset; 6740 length = btrfs_chunk_length(leaf, chunk); 6741 type = btrfs_chunk_type(leaf, chunk); 6742 index = btrfs_bg_flags_to_raid_index(type); 6743 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 6744 6745 #if BITS_PER_LONG == 32 6746 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 6747 if (ret < 0) 6748 return ret; 6749 warn_32bit_meta_chunk(fs_info, logical, length, type); 6750 #endif 6751 6752 /* 6753 * Only need to verify chunk item if we're reading from sys chunk array, 6754 * as chunk item in tree block is already verified by tree-checker. 6755 */ 6756 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 6757 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 6758 if (ret) 6759 return ret; 6760 } 6761 6762 read_lock(&map_tree->lock); 6763 em = lookup_extent_mapping(map_tree, logical, 1); 6764 read_unlock(&map_tree->lock); 6765 6766 /* already mapped? */ 6767 if (em && em->start <= logical && em->start + em->len > logical) { 6768 free_extent_map(em); 6769 return 0; 6770 } else if (em) { 6771 free_extent_map(em); 6772 } 6773 6774 em = alloc_extent_map(); 6775 if (!em) 6776 return -ENOMEM; 6777 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 6778 if (!map) { 6779 free_extent_map(em); 6780 return -ENOMEM; 6781 } 6782 6783 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 6784 em->map_lookup = map; 6785 em->start = logical; 6786 em->len = length; 6787 em->orig_start = 0; 6788 em->block_start = 0; 6789 em->block_len = em->len; 6790 6791 map->num_stripes = num_stripes; 6792 map->io_width = btrfs_chunk_io_width(leaf, chunk); 6793 map->io_align = btrfs_chunk_io_align(leaf, chunk); 6794 map->type = type; 6795 /* 6796 * We can't use the sub_stripes value, as for profiles other than 6797 * RAID10, they may have 0 as sub_stripes for filesystems created by 6798 * older mkfs (<v5.4). 6799 * In that case, it can cause divide-by-zero errors later. 6800 * Since currently sub_stripes is fixed for each profile, let's 6801 * use the trusted value instead. 6802 */ 6803 map->sub_stripes = btrfs_raid_array[index].sub_stripes; 6804 map->verified_stripes = 0; 6805 em->orig_block_len = btrfs_calc_stripe_length(em); 6806 for (i = 0; i < num_stripes; i++) { 6807 map->stripes[i].physical = 6808 btrfs_stripe_offset_nr(leaf, chunk, i); 6809 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 6810 args.devid = devid; 6811 read_extent_buffer(leaf, uuid, (unsigned long) 6812 btrfs_stripe_dev_uuid_nr(chunk, i), 6813 BTRFS_UUID_SIZE); 6814 args.uuid = uuid; 6815 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 6816 if (!map->stripes[i].dev) { 6817 map->stripes[i].dev = handle_missing_device(fs_info, 6818 devid, uuid); 6819 if (IS_ERR(map->stripes[i].dev)) { 6820 ret = PTR_ERR(map->stripes[i].dev); 6821 free_extent_map(em); 6822 return ret; 6823 } 6824 } 6825 6826 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 6827 &(map->stripes[i].dev->dev_state)); 6828 } 6829 6830 write_lock(&map_tree->lock); 6831 ret = add_extent_mapping(map_tree, em, 0); 6832 write_unlock(&map_tree->lock); 6833 if (ret < 0) { 6834 btrfs_err(fs_info, 6835 "failed to add chunk map, start=%llu len=%llu: %d", 6836 em->start, em->len, ret); 6837 } 6838 free_extent_map(em); 6839 6840 return ret; 6841 } 6842 6843 static void fill_device_from_item(struct extent_buffer *leaf, 6844 struct btrfs_dev_item *dev_item, 6845 struct btrfs_device *device) 6846 { 6847 unsigned long ptr; 6848 6849 device->devid = btrfs_device_id(leaf, dev_item); 6850 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 6851 device->total_bytes = device->disk_total_bytes; 6852 device->commit_total_bytes = device->disk_total_bytes; 6853 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 6854 device->commit_bytes_used = device->bytes_used; 6855 device->type = btrfs_device_type(leaf, dev_item); 6856 device->io_align = btrfs_device_io_align(leaf, dev_item); 6857 device->io_width = btrfs_device_io_width(leaf, dev_item); 6858 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 6859 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 6860 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 6861 6862 ptr = btrfs_device_uuid(dev_item); 6863 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 6864 } 6865 6866 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 6867 u8 *fsid) 6868 { 6869 struct btrfs_fs_devices *fs_devices; 6870 int ret; 6871 6872 lockdep_assert_held(&uuid_mutex); 6873 ASSERT(fsid); 6874 6875 /* This will match only for multi-device seed fs */ 6876 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 6877 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 6878 return fs_devices; 6879 6880 6881 fs_devices = find_fsid(fsid, NULL); 6882 if (!fs_devices) { 6883 if (!btrfs_test_opt(fs_info, DEGRADED)) 6884 return ERR_PTR(-ENOENT); 6885 6886 fs_devices = alloc_fs_devices(fsid, NULL); 6887 if (IS_ERR(fs_devices)) 6888 return fs_devices; 6889 6890 fs_devices->seeding = true; 6891 fs_devices->opened = 1; 6892 return fs_devices; 6893 } 6894 6895 /* 6896 * Upon first call for a seed fs fsid, just create a private copy of the 6897 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 6898 */ 6899 fs_devices = clone_fs_devices(fs_devices); 6900 if (IS_ERR(fs_devices)) 6901 return fs_devices; 6902 6903 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 6904 if (ret) { 6905 free_fs_devices(fs_devices); 6906 return ERR_PTR(ret); 6907 } 6908 6909 if (!fs_devices->seeding) { 6910 close_fs_devices(fs_devices); 6911 free_fs_devices(fs_devices); 6912 return ERR_PTR(-EINVAL); 6913 } 6914 6915 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 6916 6917 return fs_devices; 6918 } 6919 6920 static int read_one_dev(struct extent_buffer *leaf, 6921 struct btrfs_dev_item *dev_item) 6922 { 6923 BTRFS_DEV_LOOKUP_ARGS(args); 6924 struct btrfs_fs_info *fs_info = leaf->fs_info; 6925 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 6926 struct btrfs_device *device; 6927 u64 devid; 6928 int ret; 6929 u8 fs_uuid[BTRFS_FSID_SIZE]; 6930 u8 dev_uuid[BTRFS_UUID_SIZE]; 6931 6932 devid = btrfs_device_id(leaf, dev_item); 6933 args.devid = devid; 6934 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 6935 BTRFS_UUID_SIZE); 6936 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 6937 BTRFS_FSID_SIZE); 6938 args.uuid = dev_uuid; 6939 args.fsid = fs_uuid; 6940 6941 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 6942 fs_devices = open_seed_devices(fs_info, fs_uuid); 6943 if (IS_ERR(fs_devices)) 6944 return PTR_ERR(fs_devices); 6945 } 6946 6947 device = btrfs_find_device(fs_info->fs_devices, &args); 6948 if (!device) { 6949 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6950 btrfs_report_missing_device(fs_info, devid, 6951 dev_uuid, true); 6952 return -ENOENT; 6953 } 6954 6955 device = add_missing_dev(fs_devices, devid, dev_uuid); 6956 if (IS_ERR(device)) { 6957 btrfs_err(fs_info, 6958 "failed to add missing dev %llu: %ld", 6959 devid, PTR_ERR(device)); 6960 return PTR_ERR(device); 6961 } 6962 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 6963 } else { 6964 if (!device->bdev) { 6965 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6966 btrfs_report_missing_device(fs_info, 6967 devid, dev_uuid, true); 6968 return -ENOENT; 6969 } 6970 btrfs_report_missing_device(fs_info, devid, 6971 dev_uuid, false); 6972 } 6973 6974 if (!device->bdev && 6975 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 6976 /* 6977 * this happens when a device that was properly setup 6978 * in the device info lists suddenly goes bad. 6979 * device->bdev is NULL, and so we have to set 6980 * device->missing to one here 6981 */ 6982 device->fs_devices->missing_devices++; 6983 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6984 } 6985 6986 /* Move the device to its own fs_devices */ 6987 if (device->fs_devices != fs_devices) { 6988 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 6989 &device->dev_state)); 6990 6991 list_move(&device->dev_list, &fs_devices->devices); 6992 device->fs_devices->num_devices--; 6993 fs_devices->num_devices++; 6994 6995 device->fs_devices->missing_devices--; 6996 fs_devices->missing_devices++; 6997 6998 device->fs_devices = fs_devices; 6999 } 7000 } 7001 7002 if (device->fs_devices != fs_info->fs_devices) { 7003 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7004 if (device->generation != 7005 btrfs_device_generation(leaf, dev_item)) 7006 return -EINVAL; 7007 } 7008 7009 fill_device_from_item(leaf, dev_item, device); 7010 if (device->bdev) { 7011 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7012 7013 if (device->total_bytes > max_total_bytes) { 7014 btrfs_err(fs_info, 7015 "device total_bytes should be at most %llu but found %llu", 7016 max_total_bytes, device->total_bytes); 7017 return -EINVAL; 7018 } 7019 } 7020 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7021 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7022 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7023 device->fs_devices->total_rw_bytes += device->total_bytes; 7024 atomic64_add(device->total_bytes - device->bytes_used, 7025 &fs_info->free_chunk_space); 7026 } 7027 ret = 0; 7028 return ret; 7029 } 7030 7031 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7032 { 7033 struct btrfs_super_block *super_copy = fs_info->super_copy; 7034 struct extent_buffer *sb; 7035 struct btrfs_disk_key *disk_key; 7036 struct btrfs_chunk *chunk; 7037 u8 *array_ptr; 7038 unsigned long sb_array_offset; 7039 int ret = 0; 7040 u32 num_stripes; 7041 u32 array_size; 7042 u32 len = 0; 7043 u32 cur_offset; 7044 u64 type; 7045 struct btrfs_key key; 7046 7047 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7048 7049 /* 7050 * We allocated a dummy extent, just to use extent buffer accessors. 7051 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but 7052 * that's fine, we will not go beyond system chunk array anyway. 7053 */ 7054 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); 7055 if (!sb) 7056 return -ENOMEM; 7057 set_extent_buffer_uptodate(sb); 7058 7059 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7060 array_size = btrfs_super_sys_array_size(super_copy); 7061 7062 array_ptr = super_copy->sys_chunk_array; 7063 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7064 cur_offset = 0; 7065 7066 while (cur_offset < array_size) { 7067 disk_key = (struct btrfs_disk_key *)array_ptr; 7068 len = sizeof(*disk_key); 7069 if (cur_offset + len > array_size) 7070 goto out_short_read; 7071 7072 btrfs_disk_key_to_cpu(&key, disk_key); 7073 7074 array_ptr += len; 7075 sb_array_offset += len; 7076 cur_offset += len; 7077 7078 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7079 btrfs_err(fs_info, 7080 "unexpected item type %u in sys_array at offset %u", 7081 (u32)key.type, cur_offset); 7082 ret = -EIO; 7083 break; 7084 } 7085 7086 chunk = (struct btrfs_chunk *)sb_array_offset; 7087 /* 7088 * At least one btrfs_chunk with one stripe must be present, 7089 * exact stripe count check comes afterwards 7090 */ 7091 len = btrfs_chunk_item_size(1); 7092 if (cur_offset + len > array_size) 7093 goto out_short_read; 7094 7095 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7096 if (!num_stripes) { 7097 btrfs_err(fs_info, 7098 "invalid number of stripes %u in sys_array at offset %u", 7099 num_stripes, cur_offset); 7100 ret = -EIO; 7101 break; 7102 } 7103 7104 type = btrfs_chunk_type(sb, chunk); 7105 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7106 btrfs_err(fs_info, 7107 "invalid chunk type %llu in sys_array at offset %u", 7108 type, cur_offset); 7109 ret = -EIO; 7110 break; 7111 } 7112 7113 len = btrfs_chunk_item_size(num_stripes); 7114 if (cur_offset + len > array_size) 7115 goto out_short_read; 7116 7117 ret = read_one_chunk(&key, sb, chunk); 7118 if (ret) 7119 break; 7120 7121 array_ptr += len; 7122 sb_array_offset += len; 7123 cur_offset += len; 7124 } 7125 clear_extent_buffer_uptodate(sb); 7126 free_extent_buffer_stale(sb); 7127 return ret; 7128 7129 out_short_read: 7130 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7131 len, cur_offset); 7132 clear_extent_buffer_uptodate(sb); 7133 free_extent_buffer_stale(sb); 7134 return -EIO; 7135 } 7136 7137 /* 7138 * Check if all chunks in the fs are OK for read-write degraded mount 7139 * 7140 * If the @failing_dev is specified, it's accounted as missing. 7141 * 7142 * Return true if all chunks meet the minimal RW mount requirements. 7143 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7144 */ 7145 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7146 struct btrfs_device *failing_dev) 7147 { 7148 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7149 struct extent_map *em; 7150 u64 next_start = 0; 7151 bool ret = true; 7152 7153 read_lock(&map_tree->lock); 7154 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7155 read_unlock(&map_tree->lock); 7156 /* No chunk at all? Return false anyway */ 7157 if (!em) { 7158 ret = false; 7159 goto out; 7160 } 7161 while (em) { 7162 struct map_lookup *map; 7163 int missing = 0; 7164 int max_tolerated; 7165 int i; 7166 7167 map = em->map_lookup; 7168 max_tolerated = 7169 btrfs_get_num_tolerated_disk_barrier_failures( 7170 map->type); 7171 for (i = 0; i < map->num_stripes; i++) { 7172 struct btrfs_device *dev = map->stripes[i].dev; 7173 7174 if (!dev || !dev->bdev || 7175 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7176 dev->last_flush_error) 7177 missing++; 7178 else if (failing_dev && failing_dev == dev) 7179 missing++; 7180 } 7181 if (missing > max_tolerated) { 7182 if (!failing_dev) 7183 btrfs_warn(fs_info, 7184 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7185 em->start, missing, max_tolerated); 7186 free_extent_map(em); 7187 ret = false; 7188 goto out; 7189 } 7190 next_start = extent_map_end(em); 7191 free_extent_map(em); 7192 7193 read_lock(&map_tree->lock); 7194 em = lookup_extent_mapping(map_tree, next_start, 7195 (u64)(-1) - next_start); 7196 read_unlock(&map_tree->lock); 7197 } 7198 out: 7199 return ret; 7200 } 7201 7202 static void readahead_tree_node_children(struct extent_buffer *node) 7203 { 7204 int i; 7205 const int nr_items = btrfs_header_nritems(node); 7206 7207 for (i = 0; i < nr_items; i++) 7208 btrfs_readahead_node_child(node, i); 7209 } 7210 7211 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7212 { 7213 struct btrfs_root *root = fs_info->chunk_root; 7214 struct btrfs_path *path; 7215 struct extent_buffer *leaf; 7216 struct btrfs_key key; 7217 struct btrfs_key found_key; 7218 int ret; 7219 int slot; 7220 int iter_ret = 0; 7221 u64 total_dev = 0; 7222 u64 last_ra_node = 0; 7223 7224 path = btrfs_alloc_path(); 7225 if (!path) 7226 return -ENOMEM; 7227 7228 /* 7229 * uuid_mutex is needed only if we are mounting a sprout FS 7230 * otherwise we don't need it. 7231 */ 7232 mutex_lock(&uuid_mutex); 7233 7234 /* 7235 * It is possible for mount and umount to race in such a way that 7236 * we execute this code path, but open_fs_devices failed to clear 7237 * total_rw_bytes. We certainly want it cleared before reading the 7238 * device items, so clear it here. 7239 */ 7240 fs_info->fs_devices->total_rw_bytes = 0; 7241 7242 /* 7243 * Lockdep complains about possible circular locking dependency between 7244 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7245 * used for freeze procection of a fs (struct super_block.s_writers), 7246 * which we take when starting a transaction, and extent buffers of the 7247 * chunk tree if we call read_one_dev() while holding a lock on an 7248 * extent buffer of the chunk tree. Since we are mounting the filesystem 7249 * and at this point there can't be any concurrent task modifying the 7250 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7251 */ 7252 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7253 path->skip_locking = 1; 7254 7255 /* 7256 * Read all device items, and then all the chunk items. All 7257 * device items are found before any chunk item (their object id 7258 * is smaller than the lowest possible object id for a chunk 7259 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7260 */ 7261 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7262 key.offset = 0; 7263 key.type = 0; 7264 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 7265 struct extent_buffer *node = path->nodes[1]; 7266 7267 leaf = path->nodes[0]; 7268 slot = path->slots[0]; 7269 7270 if (node) { 7271 if (last_ra_node != node->start) { 7272 readahead_tree_node_children(node); 7273 last_ra_node = node->start; 7274 } 7275 } 7276 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7277 struct btrfs_dev_item *dev_item; 7278 dev_item = btrfs_item_ptr(leaf, slot, 7279 struct btrfs_dev_item); 7280 ret = read_one_dev(leaf, dev_item); 7281 if (ret) 7282 goto error; 7283 total_dev++; 7284 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7285 struct btrfs_chunk *chunk; 7286 7287 /* 7288 * We are only called at mount time, so no need to take 7289 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7290 * we always lock first fs_info->chunk_mutex before 7291 * acquiring any locks on the chunk tree. This is a 7292 * requirement for chunk allocation, see the comment on 7293 * top of btrfs_chunk_alloc() for details. 7294 */ 7295 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7296 ret = read_one_chunk(&found_key, leaf, chunk); 7297 if (ret) 7298 goto error; 7299 } 7300 } 7301 /* Catch error found during iteration */ 7302 if (iter_ret < 0) { 7303 ret = iter_ret; 7304 goto error; 7305 } 7306 7307 /* 7308 * After loading chunk tree, we've got all device information, 7309 * do another round of validation checks. 7310 */ 7311 if (total_dev != fs_info->fs_devices->total_devices) { 7312 btrfs_warn(fs_info, 7313 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", 7314 btrfs_super_num_devices(fs_info->super_copy), 7315 total_dev); 7316 fs_info->fs_devices->total_devices = total_dev; 7317 btrfs_set_super_num_devices(fs_info->super_copy, total_dev); 7318 } 7319 if (btrfs_super_total_bytes(fs_info->super_copy) < 7320 fs_info->fs_devices->total_rw_bytes) { 7321 btrfs_err(fs_info, 7322 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7323 btrfs_super_total_bytes(fs_info->super_copy), 7324 fs_info->fs_devices->total_rw_bytes); 7325 ret = -EINVAL; 7326 goto error; 7327 } 7328 ret = 0; 7329 error: 7330 mutex_unlock(&uuid_mutex); 7331 7332 btrfs_free_path(path); 7333 return ret; 7334 } 7335 7336 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7337 { 7338 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7339 struct btrfs_device *device; 7340 int ret = 0; 7341 7342 fs_devices->fs_info = fs_info; 7343 7344 mutex_lock(&fs_devices->device_list_mutex); 7345 list_for_each_entry(device, &fs_devices->devices, dev_list) 7346 device->fs_info = fs_info; 7347 7348 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7349 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7350 device->fs_info = fs_info; 7351 ret = btrfs_get_dev_zone_info(device, false); 7352 if (ret) 7353 break; 7354 } 7355 7356 seed_devs->fs_info = fs_info; 7357 } 7358 mutex_unlock(&fs_devices->device_list_mutex); 7359 7360 return ret; 7361 } 7362 7363 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7364 const struct btrfs_dev_stats_item *ptr, 7365 int index) 7366 { 7367 u64 val; 7368 7369 read_extent_buffer(eb, &val, 7370 offsetof(struct btrfs_dev_stats_item, values) + 7371 ((unsigned long)ptr) + (index * sizeof(u64)), 7372 sizeof(val)); 7373 return val; 7374 } 7375 7376 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7377 struct btrfs_dev_stats_item *ptr, 7378 int index, u64 val) 7379 { 7380 write_extent_buffer(eb, &val, 7381 offsetof(struct btrfs_dev_stats_item, values) + 7382 ((unsigned long)ptr) + (index * sizeof(u64)), 7383 sizeof(val)); 7384 } 7385 7386 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7387 struct btrfs_path *path) 7388 { 7389 struct btrfs_dev_stats_item *ptr; 7390 struct extent_buffer *eb; 7391 struct btrfs_key key; 7392 int item_size; 7393 int i, ret, slot; 7394 7395 if (!device->fs_info->dev_root) 7396 return 0; 7397 7398 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7399 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7400 key.offset = device->devid; 7401 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7402 if (ret) { 7403 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7404 btrfs_dev_stat_set(device, i, 0); 7405 device->dev_stats_valid = 1; 7406 btrfs_release_path(path); 7407 return ret < 0 ? ret : 0; 7408 } 7409 slot = path->slots[0]; 7410 eb = path->nodes[0]; 7411 item_size = btrfs_item_size(eb, slot); 7412 7413 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7414 7415 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7416 if (item_size >= (1 + i) * sizeof(__le64)) 7417 btrfs_dev_stat_set(device, i, 7418 btrfs_dev_stats_value(eb, ptr, i)); 7419 else 7420 btrfs_dev_stat_set(device, i, 0); 7421 } 7422 7423 device->dev_stats_valid = 1; 7424 btrfs_dev_stat_print_on_load(device); 7425 btrfs_release_path(path); 7426 7427 return 0; 7428 } 7429 7430 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7431 { 7432 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7433 struct btrfs_device *device; 7434 struct btrfs_path *path = NULL; 7435 int ret = 0; 7436 7437 path = btrfs_alloc_path(); 7438 if (!path) 7439 return -ENOMEM; 7440 7441 mutex_lock(&fs_devices->device_list_mutex); 7442 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7443 ret = btrfs_device_init_dev_stats(device, path); 7444 if (ret) 7445 goto out; 7446 } 7447 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7448 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7449 ret = btrfs_device_init_dev_stats(device, path); 7450 if (ret) 7451 goto out; 7452 } 7453 } 7454 out: 7455 mutex_unlock(&fs_devices->device_list_mutex); 7456 7457 btrfs_free_path(path); 7458 return ret; 7459 } 7460 7461 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7462 struct btrfs_device *device) 7463 { 7464 struct btrfs_fs_info *fs_info = trans->fs_info; 7465 struct btrfs_root *dev_root = fs_info->dev_root; 7466 struct btrfs_path *path; 7467 struct btrfs_key key; 7468 struct extent_buffer *eb; 7469 struct btrfs_dev_stats_item *ptr; 7470 int ret; 7471 int i; 7472 7473 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7474 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7475 key.offset = device->devid; 7476 7477 path = btrfs_alloc_path(); 7478 if (!path) 7479 return -ENOMEM; 7480 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7481 if (ret < 0) { 7482 btrfs_warn_in_rcu(fs_info, 7483 "error %d while searching for dev_stats item for device %s", 7484 ret, btrfs_dev_name(device)); 7485 goto out; 7486 } 7487 7488 if (ret == 0 && 7489 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7490 /* need to delete old one and insert a new one */ 7491 ret = btrfs_del_item(trans, dev_root, path); 7492 if (ret != 0) { 7493 btrfs_warn_in_rcu(fs_info, 7494 "delete too small dev_stats item for device %s failed %d", 7495 btrfs_dev_name(device), ret); 7496 goto out; 7497 } 7498 ret = 1; 7499 } 7500 7501 if (ret == 1) { 7502 /* need to insert a new item */ 7503 btrfs_release_path(path); 7504 ret = btrfs_insert_empty_item(trans, dev_root, path, 7505 &key, sizeof(*ptr)); 7506 if (ret < 0) { 7507 btrfs_warn_in_rcu(fs_info, 7508 "insert dev_stats item for device %s failed %d", 7509 btrfs_dev_name(device), ret); 7510 goto out; 7511 } 7512 } 7513 7514 eb = path->nodes[0]; 7515 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7516 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7517 btrfs_set_dev_stats_value(eb, ptr, i, 7518 btrfs_dev_stat_read(device, i)); 7519 btrfs_mark_buffer_dirty(eb); 7520 7521 out: 7522 btrfs_free_path(path); 7523 return ret; 7524 } 7525 7526 /* 7527 * called from commit_transaction. Writes all changed device stats to disk. 7528 */ 7529 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7530 { 7531 struct btrfs_fs_info *fs_info = trans->fs_info; 7532 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7533 struct btrfs_device *device; 7534 int stats_cnt; 7535 int ret = 0; 7536 7537 mutex_lock(&fs_devices->device_list_mutex); 7538 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7539 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7540 if (!device->dev_stats_valid || stats_cnt == 0) 7541 continue; 7542 7543 7544 /* 7545 * There is a LOAD-LOAD control dependency between the value of 7546 * dev_stats_ccnt and updating the on-disk values which requires 7547 * reading the in-memory counters. Such control dependencies 7548 * require explicit read memory barriers. 7549 * 7550 * This memory barriers pairs with smp_mb__before_atomic in 7551 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7552 * barrier implied by atomic_xchg in 7553 * btrfs_dev_stats_read_and_reset 7554 */ 7555 smp_rmb(); 7556 7557 ret = update_dev_stat_item(trans, device); 7558 if (!ret) 7559 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7560 } 7561 mutex_unlock(&fs_devices->device_list_mutex); 7562 7563 return ret; 7564 } 7565 7566 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7567 { 7568 btrfs_dev_stat_inc(dev, index); 7569 7570 if (!dev->dev_stats_valid) 7571 return; 7572 btrfs_err_rl_in_rcu(dev->fs_info, 7573 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7574 btrfs_dev_name(dev), 7575 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7576 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7577 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7578 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7579 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7580 } 7581 7582 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7583 { 7584 int i; 7585 7586 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7587 if (btrfs_dev_stat_read(dev, i) != 0) 7588 break; 7589 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7590 return; /* all values == 0, suppress message */ 7591 7592 btrfs_info_in_rcu(dev->fs_info, 7593 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7594 btrfs_dev_name(dev), 7595 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7596 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7597 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7598 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7599 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7600 } 7601 7602 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7603 struct btrfs_ioctl_get_dev_stats *stats) 7604 { 7605 BTRFS_DEV_LOOKUP_ARGS(args); 7606 struct btrfs_device *dev; 7607 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7608 int i; 7609 7610 mutex_lock(&fs_devices->device_list_mutex); 7611 args.devid = stats->devid; 7612 dev = btrfs_find_device(fs_info->fs_devices, &args); 7613 mutex_unlock(&fs_devices->device_list_mutex); 7614 7615 if (!dev) { 7616 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7617 return -ENODEV; 7618 } else if (!dev->dev_stats_valid) { 7619 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7620 return -ENODEV; 7621 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7622 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7623 if (stats->nr_items > i) 7624 stats->values[i] = 7625 btrfs_dev_stat_read_and_reset(dev, i); 7626 else 7627 btrfs_dev_stat_set(dev, i, 0); 7628 } 7629 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7630 current->comm, task_pid_nr(current)); 7631 } else { 7632 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7633 if (stats->nr_items > i) 7634 stats->values[i] = btrfs_dev_stat_read(dev, i); 7635 } 7636 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7637 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7638 return 0; 7639 } 7640 7641 /* 7642 * Update the size and bytes used for each device where it changed. This is 7643 * delayed since we would otherwise get errors while writing out the 7644 * superblocks. 7645 * 7646 * Must be invoked during transaction commit. 7647 */ 7648 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7649 { 7650 struct btrfs_device *curr, *next; 7651 7652 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7653 7654 if (list_empty(&trans->dev_update_list)) 7655 return; 7656 7657 /* 7658 * We don't need the device_list_mutex here. This list is owned by the 7659 * transaction and the transaction must complete before the device is 7660 * released. 7661 */ 7662 mutex_lock(&trans->fs_info->chunk_mutex); 7663 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7664 post_commit_list) { 7665 list_del_init(&curr->post_commit_list); 7666 curr->commit_total_bytes = curr->disk_total_bytes; 7667 curr->commit_bytes_used = curr->bytes_used; 7668 } 7669 mutex_unlock(&trans->fs_info->chunk_mutex); 7670 } 7671 7672 /* 7673 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7674 */ 7675 int btrfs_bg_type_to_factor(u64 flags) 7676 { 7677 const int index = btrfs_bg_flags_to_raid_index(flags); 7678 7679 return btrfs_raid_array[index].ncopies; 7680 } 7681 7682 7683 7684 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7685 u64 chunk_offset, u64 devid, 7686 u64 physical_offset, u64 physical_len) 7687 { 7688 struct btrfs_dev_lookup_args args = { .devid = devid }; 7689 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7690 struct extent_map *em; 7691 struct map_lookup *map; 7692 struct btrfs_device *dev; 7693 u64 stripe_len; 7694 bool found = false; 7695 int ret = 0; 7696 int i; 7697 7698 read_lock(&em_tree->lock); 7699 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7700 read_unlock(&em_tree->lock); 7701 7702 if (!em) { 7703 btrfs_err(fs_info, 7704 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 7705 physical_offset, devid); 7706 ret = -EUCLEAN; 7707 goto out; 7708 } 7709 7710 map = em->map_lookup; 7711 stripe_len = btrfs_calc_stripe_length(em); 7712 if (physical_len != stripe_len) { 7713 btrfs_err(fs_info, 7714 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 7715 physical_offset, devid, em->start, physical_len, 7716 stripe_len); 7717 ret = -EUCLEAN; 7718 goto out; 7719 } 7720 7721 /* 7722 * Very old mkfs.btrfs (before v4.1) will not respect the reserved 7723 * space. Although kernel can handle it without problem, better to warn 7724 * the users. 7725 */ 7726 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) 7727 btrfs_warn(fs_info, 7728 "devid %llu physical %llu len %llu inside the reserved space", 7729 devid, physical_offset, physical_len); 7730 7731 for (i = 0; i < map->num_stripes; i++) { 7732 if (map->stripes[i].dev->devid == devid && 7733 map->stripes[i].physical == physical_offset) { 7734 found = true; 7735 if (map->verified_stripes >= map->num_stripes) { 7736 btrfs_err(fs_info, 7737 "too many dev extents for chunk %llu found", 7738 em->start); 7739 ret = -EUCLEAN; 7740 goto out; 7741 } 7742 map->verified_stripes++; 7743 break; 7744 } 7745 } 7746 if (!found) { 7747 btrfs_err(fs_info, 7748 "dev extent physical offset %llu devid %llu has no corresponding chunk", 7749 physical_offset, devid); 7750 ret = -EUCLEAN; 7751 } 7752 7753 /* Make sure no dev extent is beyond device boundary */ 7754 dev = btrfs_find_device(fs_info->fs_devices, &args); 7755 if (!dev) { 7756 btrfs_err(fs_info, "failed to find devid %llu", devid); 7757 ret = -EUCLEAN; 7758 goto out; 7759 } 7760 7761 if (physical_offset + physical_len > dev->disk_total_bytes) { 7762 btrfs_err(fs_info, 7763 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 7764 devid, physical_offset, physical_len, 7765 dev->disk_total_bytes); 7766 ret = -EUCLEAN; 7767 goto out; 7768 } 7769 7770 if (dev->zone_info) { 7771 u64 zone_size = dev->zone_info->zone_size; 7772 7773 if (!IS_ALIGNED(physical_offset, zone_size) || 7774 !IS_ALIGNED(physical_len, zone_size)) { 7775 btrfs_err(fs_info, 7776 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 7777 devid, physical_offset, physical_len); 7778 ret = -EUCLEAN; 7779 goto out; 7780 } 7781 } 7782 7783 out: 7784 free_extent_map(em); 7785 return ret; 7786 } 7787 7788 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 7789 { 7790 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7791 struct extent_map *em; 7792 struct rb_node *node; 7793 int ret = 0; 7794 7795 read_lock(&em_tree->lock); 7796 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 7797 em = rb_entry(node, struct extent_map, rb_node); 7798 if (em->map_lookup->num_stripes != 7799 em->map_lookup->verified_stripes) { 7800 btrfs_err(fs_info, 7801 "chunk %llu has missing dev extent, have %d expect %d", 7802 em->start, em->map_lookup->verified_stripes, 7803 em->map_lookup->num_stripes); 7804 ret = -EUCLEAN; 7805 goto out; 7806 } 7807 } 7808 out: 7809 read_unlock(&em_tree->lock); 7810 return ret; 7811 } 7812 7813 /* 7814 * Ensure that all dev extents are mapped to correct chunk, otherwise 7815 * later chunk allocation/free would cause unexpected behavior. 7816 * 7817 * NOTE: This will iterate through the whole device tree, which should be of 7818 * the same size level as the chunk tree. This slightly increases mount time. 7819 */ 7820 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 7821 { 7822 struct btrfs_path *path; 7823 struct btrfs_root *root = fs_info->dev_root; 7824 struct btrfs_key key; 7825 u64 prev_devid = 0; 7826 u64 prev_dev_ext_end = 0; 7827 int ret = 0; 7828 7829 /* 7830 * We don't have a dev_root because we mounted with ignorebadroots and 7831 * failed to load the root, so we want to skip the verification in this 7832 * case for sure. 7833 * 7834 * However if the dev root is fine, but the tree itself is corrupted 7835 * we'd still fail to mount. This verification is only to make sure 7836 * writes can happen safely, so instead just bypass this check 7837 * completely in the case of IGNOREBADROOTS. 7838 */ 7839 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 7840 return 0; 7841 7842 key.objectid = 1; 7843 key.type = BTRFS_DEV_EXTENT_KEY; 7844 key.offset = 0; 7845 7846 path = btrfs_alloc_path(); 7847 if (!path) 7848 return -ENOMEM; 7849 7850 path->reada = READA_FORWARD; 7851 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7852 if (ret < 0) 7853 goto out; 7854 7855 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 7856 ret = btrfs_next_leaf(root, path); 7857 if (ret < 0) 7858 goto out; 7859 /* No dev extents at all? Not good */ 7860 if (ret > 0) { 7861 ret = -EUCLEAN; 7862 goto out; 7863 } 7864 } 7865 while (1) { 7866 struct extent_buffer *leaf = path->nodes[0]; 7867 struct btrfs_dev_extent *dext; 7868 int slot = path->slots[0]; 7869 u64 chunk_offset; 7870 u64 physical_offset; 7871 u64 physical_len; 7872 u64 devid; 7873 7874 btrfs_item_key_to_cpu(leaf, &key, slot); 7875 if (key.type != BTRFS_DEV_EXTENT_KEY) 7876 break; 7877 devid = key.objectid; 7878 physical_offset = key.offset; 7879 7880 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 7881 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 7882 physical_len = btrfs_dev_extent_length(leaf, dext); 7883 7884 /* Check if this dev extent overlaps with the previous one */ 7885 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 7886 btrfs_err(fs_info, 7887 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 7888 devid, physical_offset, prev_dev_ext_end); 7889 ret = -EUCLEAN; 7890 goto out; 7891 } 7892 7893 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 7894 physical_offset, physical_len); 7895 if (ret < 0) 7896 goto out; 7897 prev_devid = devid; 7898 prev_dev_ext_end = physical_offset + physical_len; 7899 7900 ret = btrfs_next_item(root, path); 7901 if (ret < 0) 7902 goto out; 7903 if (ret > 0) { 7904 ret = 0; 7905 break; 7906 } 7907 } 7908 7909 /* Ensure all chunks have corresponding dev extents */ 7910 ret = verify_chunk_dev_extent_mapping(fs_info); 7911 out: 7912 btrfs_free_path(path); 7913 return ret; 7914 } 7915 7916 /* 7917 * Check whether the given block group or device is pinned by any inode being 7918 * used as a swapfile. 7919 */ 7920 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 7921 { 7922 struct btrfs_swapfile_pin *sp; 7923 struct rb_node *node; 7924 7925 spin_lock(&fs_info->swapfile_pins_lock); 7926 node = fs_info->swapfile_pins.rb_node; 7927 while (node) { 7928 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 7929 if (ptr < sp->ptr) 7930 node = node->rb_left; 7931 else if (ptr > sp->ptr) 7932 node = node->rb_right; 7933 else 7934 break; 7935 } 7936 spin_unlock(&fs_info->swapfile_pins_lock); 7937 return node != NULL; 7938 } 7939 7940 static int relocating_repair_kthread(void *data) 7941 { 7942 struct btrfs_block_group *cache = data; 7943 struct btrfs_fs_info *fs_info = cache->fs_info; 7944 u64 target; 7945 int ret = 0; 7946 7947 target = cache->start; 7948 btrfs_put_block_group(cache); 7949 7950 sb_start_write(fs_info->sb); 7951 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 7952 btrfs_info(fs_info, 7953 "zoned: skip relocating block group %llu to repair: EBUSY", 7954 target); 7955 sb_end_write(fs_info->sb); 7956 return -EBUSY; 7957 } 7958 7959 mutex_lock(&fs_info->reclaim_bgs_lock); 7960 7961 /* Ensure block group still exists */ 7962 cache = btrfs_lookup_block_group(fs_info, target); 7963 if (!cache) 7964 goto out; 7965 7966 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) 7967 goto out; 7968 7969 ret = btrfs_may_alloc_data_chunk(fs_info, target); 7970 if (ret < 0) 7971 goto out; 7972 7973 btrfs_info(fs_info, 7974 "zoned: relocating block group %llu to repair IO failure", 7975 target); 7976 ret = btrfs_relocate_chunk(fs_info, target); 7977 7978 out: 7979 if (cache) 7980 btrfs_put_block_group(cache); 7981 mutex_unlock(&fs_info->reclaim_bgs_lock); 7982 btrfs_exclop_finish(fs_info); 7983 sb_end_write(fs_info->sb); 7984 7985 return ret; 7986 } 7987 7988 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 7989 { 7990 struct btrfs_block_group *cache; 7991 7992 if (!btrfs_is_zoned(fs_info)) 7993 return false; 7994 7995 /* Do not attempt to repair in degraded state */ 7996 if (btrfs_test_opt(fs_info, DEGRADED)) 7997 return true; 7998 7999 cache = btrfs_lookup_block_group(fs_info, logical); 8000 if (!cache) 8001 return true; 8002 8003 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { 8004 btrfs_put_block_group(cache); 8005 return true; 8006 } 8007 8008 kthread_run(relocating_repair_kthread, cache, 8009 "btrfs-relocating-repair"); 8010 8011 return true; 8012 } 8013 8014 static void map_raid56_repair_block(struct btrfs_io_context *bioc, 8015 struct btrfs_io_stripe *smap, 8016 u64 logical) 8017 { 8018 int data_stripes = nr_bioc_data_stripes(bioc); 8019 int i; 8020 8021 for (i = 0; i < data_stripes; i++) { 8022 u64 stripe_start = bioc->full_stripe_logical + 8023 (i << BTRFS_STRIPE_LEN_SHIFT); 8024 8025 if (logical >= stripe_start && 8026 logical < stripe_start + BTRFS_STRIPE_LEN) 8027 break; 8028 } 8029 ASSERT(i < data_stripes); 8030 smap->dev = bioc->stripes[i].dev; 8031 smap->physical = bioc->stripes[i].physical + 8032 ((logical - bioc->full_stripe_logical) & 8033 BTRFS_STRIPE_LEN_MASK); 8034 } 8035 8036 /* 8037 * Map a repair write into a single device. 8038 * 8039 * A repair write is triggered by read time repair or scrub, which would only 8040 * update the contents of a single device. 8041 * Not update any other mirrors nor go through RMW path. 8042 * 8043 * Callers should ensure: 8044 * 8045 * - Call btrfs_bio_counter_inc_blocked() first 8046 * - The range does not cross stripe boundary 8047 * - Has a valid @mirror_num passed in. 8048 */ 8049 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info, 8050 struct btrfs_io_stripe *smap, u64 logical, 8051 u32 length, int mirror_num) 8052 { 8053 struct btrfs_io_context *bioc = NULL; 8054 u64 map_length = length; 8055 int mirror_ret = mirror_num; 8056 int ret; 8057 8058 ASSERT(mirror_num > 0); 8059 8060 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length, 8061 &bioc, smap, &mirror_ret, true); 8062 if (ret < 0) 8063 return ret; 8064 8065 /* The map range should not cross stripe boundary. */ 8066 ASSERT(map_length >= length); 8067 8068 /* Already mapped to single stripe. */ 8069 if (!bioc) 8070 goto out; 8071 8072 /* Map the RAID56 multi-stripe writes to a single one. */ 8073 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 8074 map_raid56_repair_block(bioc, smap, logical); 8075 goto out; 8076 } 8077 8078 ASSERT(mirror_num <= bioc->num_stripes); 8079 smap->dev = bioc->stripes[mirror_num - 1].dev; 8080 smap->physical = bioc->stripes[mirror_num - 1].physical; 8081 out: 8082 btrfs_put_bioc(bioc); 8083 ASSERT(smap->dev); 8084 return 0; 8085 } 8086