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 | S_VERSION); 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 /* 4085 * Validate target profile against allowed profiles and return true if it's OK. 4086 * Otherwise print the error message and return false. 4087 */ 4088 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4089 const struct btrfs_balance_args *bargs, 4090 u64 allowed, const char *type) 4091 { 4092 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4093 return true; 4094 4095 /* Profile is valid and does not have bits outside of the allowed set */ 4096 if (alloc_profile_is_valid(bargs->target, 1) && 4097 (bargs->target & ~allowed) == 0) 4098 return true; 4099 4100 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4101 type, btrfs_bg_type_to_raid_name(bargs->target)); 4102 return false; 4103 } 4104 4105 /* 4106 * Fill @buf with textual description of balance filter flags @bargs, up to 4107 * @size_buf including the terminating null. The output may be trimmed if it 4108 * does not fit into the provided buffer. 4109 */ 4110 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4111 u32 size_buf) 4112 { 4113 int ret; 4114 u32 size_bp = size_buf; 4115 char *bp = buf; 4116 u64 flags = bargs->flags; 4117 char tmp_buf[128] = {'\0'}; 4118 4119 if (!flags) 4120 return; 4121 4122 #define CHECK_APPEND_NOARG(a) \ 4123 do { \ 4124 ret = snprintf(bp, size_bp, (a)); \ 4125 if (ret < 0 || ret >= size_bp) \ 4126 goto out_overflow; \ 4127 size_bp -= ret; \ 4128 bp += ret; \ 4129 } while (0) 4130 4131 #define CHECK_APPEND_1ARG(a, v1) \ 4132 do { \ 4133 ret = snprintf(bp, size_bp, (a), (v1)); \ 4134 if (ret < 0 || ret >= size_bp) \ 4135 goto out_overflow; \ 4136 size_bp -= ret; \ 4137 bp += ret; \ 4138 } while (0) 4139 4140 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4141 do { \ 4142 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4143 if (ret < 0 || ret >= size_bp) \ 4144 goto out_overflow; \ 4145 size_bp -= ret; \ 4146 bp += ret; \ 4147 } while (0) 4148 4149 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4150 CHECK_APPEND_1ARG("convert=%s,", 4151 btrfs_bg_type_to_raid_name(bargs->target)); 4152 4153 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4154 CHECK_APPEND_NOARG("soft,"); 4155 4156 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4157 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4158 sizeof(tmp_buf)); 4159 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4160 } 4161 4162 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4163 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4164 4165 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4166 CHECK_APPEND_2ARG("usage=%u..%u,", 4167 bargs->usage_min, bargs->usage_max); 4168 4169 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4170 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4171 4172 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4173 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4174 bargs->pstart, bargs->pend); 4175 4176 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4177 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4178 bargs->vstart, bargs->vend); 4179 4180 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4181 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4182 4183 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4184 CHECK_APPEND_2ARG("limit=%u..%u,", 4185 bargs->limit_min, bargs->limit_max); 4186 4187 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4188 CHECK_APPEND_2ARG("stripes=%u..%u,", 4189 bargs->stripes_min, bargs->stripes_max); 4190 4191 #undef CHECK_APPEND_2ARG 4192 #undef CHECK_APPEND_1ARG 4193 #undef CHECK_APPEND_NOARG 4194 4195 out_overflow: 4196 4197 if (size_bp < size_buf) 4198 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4199 else 4200 buf[0] = '\0'; 4201 } 4202 4203 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4204 { 4205 u32 size_buf = 1024; 4206 char tmp_buf[192] = {'\0'}; 4207 char *buf; 4208 char *bp; 4209 u32 size_bp = size_buf; 4210 int ret; 4211 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4212 4213 buf = kzalloc(size_buf, GFP_KERNEL); 4214 if (!buf) 4215 return; 4216 4217 bp = buf; 4218 4219 #define CHECK_APPEND_1ARG(a, v1) \ 4220 do { \ 4221 ret = snprintf(bp, size_bp, (a), (v1)); \ 4222 if (ret < 0 || ret >= size_bp) \ 4223 goto out_overflow; \ 4224 size_bp -= ret; \ 4225 bp += ret; \ 4226 } while (0) 4227 4228 if (bctl->flags & BTRFS_BALANCE_FORCE) 4229 CHECK_APPEND_1ARG("%s", "-f "); 4230 4231 if (bctl->flags & BTRFS_BALANCE_DATA) { 4232 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4233 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4234 } 4235 4236 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4237 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4238 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4239 } 4240 4241 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4242 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4243 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4244 } 4245 4246 #undef CHECK_APPEND_1ARG 4247 4248 out_overflow: 4249 4250 if (size_bp < size_buf) 4251 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4252 btrfs_info(fs_info, "balance: %s %s", 4253 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4254 "resume" : "start", buf); 4255 4256 kfree(buf); 4257 } 4258 4259 /* 4260 * Should be called with balance mutexe held 4261 */ 4262 int btrfs_balance(struct btrfs_fs_info *fs_info, 4263 struct btrfs_balance_control *bctl, 4264 struct btrfs_ioctl_balance_args *bargs) 4265 { 4266 u64 meta_target, data_target; 4267 u64 allowed; 4268 int mixed = 0; 4269 int ret; 4270 u64 num_devices; 4271 unsigned seq; 4272 bool reducing_redundancy; 4273 bool paused = false; 4274 int i; 4275 4276 if (btrfs_fs_closing(fs_info) || 4277 atomic_read(&fs_info->balance_pause_req) || 4278 btrfs_should_cancel_balance(fs_info)) { 4279 ret = -EINVAL; 4280 goto out; 4281 } 4282 4283 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4284 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4285 mixed = 1; 4286 4287 /* 4288 * In case of mixed groups both data and meta should be picked, 4289 * and identical options should be given for both of them. 4290 */ 4291 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4292 if (mixed && (bctl->flags & allowed)) { 4293 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4294 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4295 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4296 btrfs_err(fs_info, 4297 "balance: mixed groups data and metadata options must be the same"); 4298 ret = -EINVAL; 4299 goto out; 4300 } 4301 } 4302 4303 /* 4304 * rw_devices will not change at the moment, device add/delete/replace 4305 * are exclusive 4306 */ 4307 num_devices = fs_info->fs_devices->rw_devices; 4308 4309 /* 4310 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4311 * special bit for it, to make it easier to distinguish. Thus we need 4312 * to set it manually, or balance would refuse the profile. 4313 */ 4314 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4315 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4316 if (num_devices >= btrfs_raid_array[i].devs_min) 4317 allowed |= btrfs_raid_array[i].bg_flag; 4318 4319 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4320 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4321 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4322 ret = -EINVAL; 4323 goto out; 4324 } 4325 4326 /* 4327 * Allow to reduce metadata or system integrity only if force set for 4328 * profiles with redundancy (copies, parity) 4329 */ 4330 allowed = 0; 4331 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4332 if (btrfs_raid_array[i].ncopies >= 2 || 4333 btrfs_raid_array[i].tolerated_failures >= 1) 4334 allowed |= btrfs_raid_array[i].bg_flag; 4335 } 4336 do { 4337 seq = read_seqbegin(&fs_info->profiles_lock); 4338 4339 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4340 (fs_info->avail_system_alloc_bits & allowed) && 4341 !(bctl->sys.target & allowed)) || 4342 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4343 (fs_info->avail_metadata_alloc_bits & allowed) && 4344 !(bctl->meta.target & allowed))) 4345 reducing_redundancy = true; 4346 else 4347 reducing_redundancy = false; 4348 4349 /* if we're not converting, the target field is uninitialized */ 4350 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4351 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4352 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4353 bctl->data.target : fs_info->avail_data_alloc_bits; 4354 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4355 4356 if (reducing_redundancy) { 4357 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4358 btrfs_info(fs_info, 4359 "balance: force reducing metadata redundancy"); 4360 } else { 4361 btrfs_err(fs_info, 4362 "balance: reduces metadata redundancy, use --force if you want this"); 4363 ret = -EINVAL; 4364 goto out; 4365 } 4366 } 4367 4368 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4369 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4370 btrfs_warn(fs_info, 4371 "balance: metadata profile %s has lower redundancy than data profile %s", 4372 btrfs_bg_type_to_raid_name(meta_target), 4373 btrfs_bg_type_to_raid_name(data_target)); 4374 } 4375 4376 ret = insert_balance_item(fs_info, bctl); 4377 if (ret && ret != -EEXIST) 4378 goto out; 4379 4380 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4381 BUG_ON(ret == -EEXIST); 4382 BUG_ON(fs_info->balance_ctl); 4383 spin_lock(&fs_info->balance_lock); 4384 fs_info->balance_ctl = bctl; 4385 spin_unlock(&fs_info->balance_lock); 4386 } else { 4387 BUG_ON(ret != -EEXIST); 4388 spin_lock(&fs_info->balance_lock); 4389 update_balance_args(bctl); 4390 spin_unlock(&fs_info->balance_lock); 4391 } 4392 4393 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4394 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4395 describe_balance_start_or_resume(fs_info); 4396 mutex_unlock(&fs_info->balance_mutex); 4397 4398 ret = __btrfs_balance(fs_info); 4399 4400 mutex_lock(&fs_info->balance_mutex); 4401 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4402 btrfs_info(fs_info, "balance: paused"); 4403 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4404 paused = true; 4405 } 4406 /* 4407 * Balance can be canceled by: 4408 * 4409 * - Regular cancel request 4410 * Then ret == -ECANCELED and balance_cancel_req > 0 4411 * 4412 * - Fatal signal to "btrfs" process 4413 * Either the signal caught by wait_reserve_ticket() and callers 4414 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4415 * got -ECANCELED. 4416 * Either way, in this case balance_cancel_req = 0, and 4417 * ret == -EINTR or ret == -ECANCELED. 4418 * 4419 * So here we only check the return value to catch canceled balance. 4420 */ 4421 else if (ret == -ECANCELED || ret == -EINTR) 4422 btrfs_info(fs_info, "balance: canceled"); 4423 else 4424 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4425 4426 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4427 4428 if (bargs) { 4429 memset(bargs, 0, sizeof(*bargs)); 4430 btrfs_update_ioctl_balance_args(fs_info, bargs); 4431 } 4432 4433 /* We didn't pause, we can clean everything up. */ 4434 if (!paused) { 4435 reset_balance_state(fs_info); 4436 btrfs_exclop_finish(fs_info); 4437 } 4438 4439 wake_up(&fs_info->balance_wait_q); 4440 4441 return ret; 4442 out: 4443 if (bctl->flags & BTRFS_BALANCE_RESUME) 4444 reset_balance_state(fs_info); 4445 else 4446 kfree(bctl); 4447 btrfs_exclop_finish(fs_info); 4448 4449 return ret; 4450 } 4451 4452 static int balance_kthread(void *data) 4453 { 4454 struct btrfs_fs_info *fs_info = data; 4455 int ret = 0; 4456 4457 sb_start_write(fs_info->sb); 4458 mutex_lock(&fs_info->balance_mutex); 4459 if (fs_info->balance_ctl) 4460 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4461 mutex_unlock(&fs_info->balance_mutex); 4462 sb_end_write(fs_info->sb); 4463 4464 return ret; 4465 } 4466 4467 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4468 { 4469 struct task_struct *tsk; 4470 4471 mutex_lock(&fs_info->balance_mutex); 4472 if (!fs_info->balance_ctl) { 4473 mutex_unlock(&fs_info->balance_mutex); 4474 return 0; 4475 } 4476 mutex_unlock(&fs_info->balance_mutex); 4477 4478 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4479 btrfs_info(fs_info, "balance: resume skipped"); 4480 return 0; 4481 } 4482 4483 spin_lock(&fs_info->super_lock); 4484 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4485 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4486 spin_unlock(&fs_info->super_lock); 4487 /* 4488 * A ro->rw remount sequence should continue with the paused balance 4489 * regardless of who pauses it, system or the user as of now, so set 4490 * the resume flag. 4491 */ 4492 spin_lock(&fs_info->balance_lock); 4493 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4494 spin_unlock(&fs_info->balance_lock); 4495 4496 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4497 return PTR_ERR_OR_ZERO(tsk); 4498 } 4499 4500 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4501 { 4502 struct btrfs_balance_control *bctl; 4503 struct btrfs_balance_item *item; 4504 struct btrfs_disk_balance_args disk_bargs; 4505 struct btrfs_path *path; 4506 struct extent_buffer *leaf; 4507 struct btrfs_key key; 4508 int ret; 4509 4510 path = btrfs_alloc_path(); 4511 if (!path) 4512 return -ENOMEM; 4513 4514 key.objectid = BTRFS_BALANCE_OBJECTID; 4515 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4516 key.offset = 0; 4517 4518 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4519 if (ret < 0) 4520 goto out; 4521 if (ret > 0) { /* ret = -ENOENT; */ 4522 ret = 0; 4523 goto out; 4524 } 4525 4526 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4527 if (!bctl) { 4528 ret = -ENOMEM; 4529 goto out; 4530 } 4531 4532 leaf = path->nodes[0]; 4533 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4534 4535 bctl->flags = btrfs_balance_flags(leaf, item); 4536 bctl->flags |= BTRFS_BALANCE_RESUME; 4537 4538 btrfs_balance_data(leaf, item, &disk_bargs); 4539 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4540 btrfs_balance_meta(leaf, item, &disk_bargs); 4541 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4542 btrfs_balance_sys(leaf, item, &disk_bargs); 4543 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4544 4545 /* 4546 * This should never happen, as the paused balance state is recovered 4547 * during mount without any chance of other exclusive ops to collide. 4548 * 4549 * This gives the exclusive op status to balance and keeps in paused 4550 * state until user intervention (cancel or umount). If the ownership 4551 * cannot be assigned, show a message but do not fail. The balance 4552 * is in a paused state and must have fs_info::balance_ctl properly 4553 * set up. 4554 */ 4555 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4556 btrfs_warn(fs_info, 4557 "balance: cannot set exclusive op status, resume manually"); 4558 4559 btrfs_release_path(path); 4560 4561 mutex_lock(&fs_info->balance_mutex); 4562 BUG_ON(fs_info->balance_ctl); 4563 spin_lock(&fs_info->balance_lock); 4564 fs_info->balance_ctl = bctl; 4565 spin_unlock(&fs_info->balance_lock); 4566 mutex_unlock(&fs_info->balance_mutex); 4567 out: 4568 btrfs_free_path(path); 4569 return ret; 4570 } 4571 4572 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4573 { 4574 int ret = 0; 4575 4576 mutex_lock(&fs_info->balance_mutex); 4577 if (!fs_info->balance_ctl) { 4578 mutex_unlock(&fs_info->balance_mutex); 4579 return -ENOTCONN; 4580 } 4581 4582 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4583 atomic_inc(&fs_info->balance_pause_req); 4584 mutex_unlock(&fs_info->balance_mutex); 4585 4586 wait_event(fs_info->balance_wait_q, 4587 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4588 4589 mutex_lock(&fs_info->balance_mutex); 4590 /* we are good with balance_ctl ripped off from under us */ 4591 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4592 atomic_dec(&fs_info->balance_pause_req); 4593 } else { 4594 ret = -ENOTCONN; 4595 } 4596 4597 mutex_unlock(&fs_info->balance_mutex); 4598 return ret; 4599 } 4600 4601 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4602 { 4603 mutex_lock(&fs_info->balance_mutex); 4604 if (!fs_info->balance_ctl) { 4605 mutex_unlock(&fs_info->balance_mutex); 4606 return -ENOTCONN; 4607 } 4608 4609 /* 4610 * A paused balance with the item stored on disk can be resumed at 4611 * mount time if the mount is read-write. Otherwise it's still paused 4612 * and we must not allow cancelling as it deletes the item. 4613 */ 4614 if (sb_rdonly(fs_info->sb)) { 4615 mutex_unlock(&fs_info->balance_mutex); 4616 return -EROFS; 4617 } 4618 4619 atomic_inc(&fs_info->balance_cancel_req); 4620 /* 4621 * if we are running just wait and return, balance item is 4622 * deleted in btrfs_balance in this case 4623 */ 4624 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4625 mutex_unlock(&fs_info->balance_mutex); 4626 wait_event(fs_info->balance_wait_q, 4627 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4628 mutex_lock(&fs_info->balance_mutex); 4629 } else { 4630 mutex_unlock(&fs_info->balance_mutex); 4631 /* 4632 * Lock released to allow other waiters to continue, we'll 4633 * reexamine the status again. 4634 */ 4635 mutex_lock(&fs_info->balance_mutex); 4636 4637 if (fs_info->balance_ctl) { 4638 reset_balance_state(fs_info); 4639 btrfs_exclop_finish(fs_info); 4640 btrfs_info(fs_info, "balance: canceled"); 4641 } 4642 } 4643 4644 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4645 atomic_dec(&fs_info->balance_cancel_req); 4646 mutex_unlock(&fs_info->balance_mutex); 4647 return 0; 4648 } 4649 4650 int btrfs_uuid_scan_kthread(void *data) 4651 { 4652 struct btrfs_fs_info *fs_info = data; 4653 struct btrfs_root *root = fs_info->tree_root; 4654 struct btrfs_key key; 4655 struct btrfs_path *path = NULL; 4656 int ret = 0; 4657 struct extent_buffer *eb; 4658 int slot; 4659 struct btrfs_root_item root_item; 4660 u32 item_size; 4661 struct btrfs_trans_handle *trans = NULL; 4662 bool closing = false; 4663 4664 path = btrfs_alloc_path(); 4665 if (!path) { 4666 ret = -ENOMEM; 4667 goto out; 4668 } 4669 4670 key.objectid = 0; 4671 key.type = BTRFS_ROOT_ITEM_KEY; 4672 key.offset = 0; 4673 4674 while (1) { 4675 if (btrfs_fs_closing(fs_info)) { 4676 closing = true; 4677 break; 4678 } 4679 ret = btrfs_search_forward(root, &key, path, 4680 BTRFS_OLDEST_GENERATION); 4681 if (ret) { 4682 if (ret > 0) 4683 ret = 0; 4684 break; 4685 } 4686 4687 if (key.type != BTRFS_ROOT_ITEM_KEY || 4688 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4689 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4690 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4691 goto skip; 4692 4693 eb = path->nodes[0]; 4694 slot = path->slots[0]; 4695 item_size = btrfs_item_size(eb, slot); 4696 if (item_size < sizeof(root_item)) 4697 goto skip; 4698 4699 read_extent_buffer(eb, &root_item, 4700 btrfs_item_ptr_offset(eb, slot), 4701 (int)sizeof(root_item)); 4702 if (btrfs_root_refs(&root_item) == 0) 4703 goto skip; 4704 4705 if (!btrfs_is_empty_uuid(root_item.uuid) || 4706 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4707 if (trans) 4708 goto update_tree; 4709 4710 btrfs_release_path(path); 4711 /* 4712 * 1 - subvol uuid item 4713 * 1 - received_subvol uuid item 4714 */ 4715 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4716 if (IS_ERR(trans)) { 4717 ret = PTR_ERR(trans); 4718 break; 4719 } 4720 continue; 4721 } else { 4722 goto skip; 4723 } 4724 update_tree: 4725 btrfs_release_path(path); 4726 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4727 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4728 BTRFS_UUID_KEY_SUBVOL, 4729 key.objectid); 4730 if (ret < 0) { 4731 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4732 ret); 4733 break; 4734 } 4735 } 4736 4737 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4738 ret = btrfs_uuid_tree_add(trans, 4739 root_item.received_uuid, 4740 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4741 key.objectid); 4742 if (ret < 0) { 4743 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4744 ret); 4745 break; 4746 } 4747 } 4748 4749 skip: 4750 btrfs_release_path(path); 4751 if (trans) { 4752 ret = btrfs_end_transaction(trans); 4753 trans = NULL; 4754 if (ret) 4755 break; 4756 } 4757 4758 if (key.offset < (u64)-1) { 4759 key.offset++; 4760 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4761 key.offset = 0; 4762 key.type = BTRFS_ROOT_ITEM_KEY; 4763 } else if (key.objectid < (u64)-1) { 4764 key.offset = 0; 4765 key.type = BTRFS_ROOT_ITEM_KEY; 4766 key.objectid++; 4767 } else { 4768 break; 4769 } 4770 cond_resched(); 4771 } 4772 4773 out: 4774 btrfs_free_path(path); 4775 if (trans && !IS_ERR(trans)) 4776 btrfs_end_transaction(trans); 4777 if (ret) 4778 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4779 else if (!closing) 4780 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4781 up(&fs_info->uuid_tree_rescan_sem); 4782 return 0; 4783 } 4784 4785 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4786 { 4787 struct btrfs_trans_handle *trans; 4788 struct btrfs_root *tree_root = fs_info->tree_root; 4789 struct btrfs_root *uuid_root; 4790 struct task_struct *task; 4791 int ret; 4792 4793 /* 4794 * 1 - root node 4795 * 1 - root item 4796 */ 4797 trans = btrfs_start_transaction(tree_root, 2); 4798 if (IS_ERR(trans)) 4799 return PTR_ERR(trans); 4800 4801 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4802 if (IS_ERR(uuid_root)) { 4803 ret = PTR_ERR(uuid_root); 4804 btrfs_abort_transaction(trans, ret); 4805 btrfs_end_transaction(trans); 4806 return ret; 4807 } 4808 4809 fs_info->uuid_root = uuid_root; 4810 4811 ret = btrfs_commit_transaction(trans); 4812 if (ret) 4813 return ret; 4814 4815 down(&fs_info->uuid_tree_rescan_sem); 4816 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4817 if (IS_ERR(task)) { 4818 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4819 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4820 up(&fs_info->uuid_tree_rescan_sem); 4821 return PTR_ERR(task); 4822 } 4823 4824 return 0; 4825 } 4826 4827 /* 4828 * shrinking a device means finding all of the device extents past 4829 * the new size, and then following the back refs to the chunks. 4830 * The chunk relocation code actually frees the device extent 4831 */ 4832 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4833 { 4834 struct btrfs_fs_info *fs_info = device->fs_info; 4835 struct btrfs_root *root = fs_info->dev_root; 4836 struct btrfs_trans_handle *trans; 4837 struct btrfs_dev_extent *dev_extent = NULL; 4838 struct btrfs_path *path; 4839 u64 length; 4840 u64 chunk_offset; 4841 int ret; 4842 int slot; 4843 int failed = 0; 4844 bool retried = false; 4845 struct extent_buffer *l; 4846 struct btrfs_key key; 4847 struct btrfs_super_block *super_copy = fs_info->super_copy; 4848 u64 old_total = btrfs_super_total_bytes(super_copy); 4849 u64 old_size = btrfs_device_get_total_bytes(device); 4850 u64 diff; 4851 u64 start; 4852 4853 new_size = round_down(new_size, fs_info->sectorsize); 4854 start = new_size; 4855 diff = round_down(old_size - new_size, fs_info->sectorsize); 4856 4857 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4858 return -EINVAL; 4859 4860 path = btrfs_alloc_path(); 4861 if (!path) 4862 return -ENOMEM; 4863 4864 path->reada = READA_BACK; 4865 4866 trans = btrfs_start_transaction(root, 0); 4867 if (IS_ERR(trans)) { 4868 btrfs_free_path(path); 4869 return PTR_ERR(trans); 4870 } 4871 4872 mutex_lock(&fs_info->chunk_mutex); 4873 4874 btrfs_device_set_total_bytes(device, new_size); 4875 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4876 device->fs_devices->total_rw_bytes -= diff; 4877 atomic64_sub(diff, &fs_info->free_chunk_space); 4878 } 4879 4880 /* 4881 * Once the device's size has been set to the new size, ensure all 4882 * in-memory chunks are synced to disk so that the loop below sees them 4883 * and relocates them accordingly. 4884 */ 4885 if (contains_pending_extent(device, &start, diff)) { 4886 mutex_unlock(&fs_info->chunk_mutex); 4887 ret = btrfs_commit_transaction(trans); 4888 if (ret) 4889 goto done; 4890 } else { 4891 mutex_unlock(&fs_info->chunk_mutex); 4892 btrfs_end_transaction(trans); 4893 } 4894 4895 again: 4896 key.objectid = device->devid; 4897 key.offset = (u64)-1; 4898 key.type = BTRFS_DEV_EXTENT_KEY; 4899 4900 do { 4901 mutex_lock(&fs_info->reclaim_bgs_lock); 4902 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4903 if (ret < 0) { 4904 mutex_unlock(&fs_info->reclaim_bgs_lock); 4905 goto done; 4906 } 4907 4908 ret = btrfs_previous_item(root, path, 0, key.type); 4909 if (ret) { 4910 mutex_unlock(&fs_info->reclaim_bgs_lock); 4911 if (ret < 0) 4912 goto done; 4913 ret = 0; 4914 btrfs_release_path(path); 4915 break; 4916 } 4917 4918 l = path->nodes[0]; 4919 slot = path->slots[0]; 4920 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4921 4922 if (key.objectid != device->devid) { 4923 mutex_unlock(&fs_info->reclaim_bgs_lock); 4924 btrfs_release_path(path); 4925 break; 4926 } 4927 4928 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4929 length = btrfs_dev_extent_length(l, dev_extent); 4930 4931 if (key.offset + length <= new_size) { 4932 mutex_unlock(&fs_info->reclaim_bgs_lock); 4933 btrfs_release_path(path); 4934 break; 4935 } 4936 4937 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4938 btrfs_release_path(path); 4939 4940 /* 4941 * We may be relocating the only data chunk we have, 4942 * which could potentially end up with losing data's 4943 * raid profile, so lets allocate an empty one in 4944 * advance. 4945 */ 4946 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4947 if (ret < 0) { 4948 mutex_unlock(&fs_info->reclaim_bgs_lock); 4949 goto done; 4950 } 4951 4952 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4953 mutex_unlock(&fs_info->reclaim_bgs_lock); 4954 if (ret == -ENOSPC) { 4955 failed++; 4956 } else if (ret) { 4957 if (ret == -ETXTBSY) { 4958 btrfs_warn(fs_info, 4959 "could not shrink block group %llu due to active swapfile", 4960 chunk_offset); 4961 } 4962 goto done; 4963 } 4964 } while (key.offset-- > 0); 4965 4966 if (failed && !retried) { 4967 failed = 0; 4968 retried = true; 4969 goto again; 4970 } else if (failed && retried) { 4971 ret = -ENOSPC; 4972 goto done; 4973 } 4974 4975 /* Shrinking succeeded, else we would be at "done". */ 4976 trans = btrfs_start_transaction(root, 0); 4977 if (IS_ERR(trans)) { 4978 ret = PTR_ERR(trans); 4979 goto done; 4980 } 4981 4982 mutex_lock(&fs_info->chunk_mutex); 4983 /* Clear all state bits beyond the shrunk device size */ 4984 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4985 CHUNK_STATE_MASK); 4986 4987 btrfs_device_set_disk_total_bytes(device, new_size); 4988 if (list_empty(&device->post_commit_list)) 4989 list_add_tail(&device->post_commit_list, 4990 &trans->transaction->dev_update_list); 4991 4992 WARN_ON(diff > old_total); 4993 btrfs_set_super_total_bytes(super_copy, 4994 round_down(old_total - diff, fs_info->sectorsize)); 4995 mutex_unlock(&fs_info->chunk_mutex); 4996 4997 btrfs_reserve_chunk_metadata(trans, false); 4998 /* Now btrfs_update_device() will change the on-disk size. */ 4999 ret = btrfs_update_device(trans, device); 5000 btrfs_trans_release_chunk_metadata(trans); 5001 if (ret < 0) { 5002 btrfs_abort_transaction(trans, ret); 5003 btrfs_end_transaction(trans); 5004 } else { 5005 ret = btrfs_commit_transaction(trans); 5006 } 5007 done: 5008 btrfs_free_path(path); 5009 if (ret) { 5010 mutex_lock(&fs_info->chunk_mutex); 5011 btrfs_device_set_total_bytes(device, old_size); 5012 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 5013 device->fs_devices->total_rw_bytes += diff; 5014 atomic64_add(diff, &fs_info->free_chunk_space); 5015 mutex_unlock(&fs_info->chunk_mutex); 5016 } 5017 return ret; 5018 } 5019 5020 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 5021 struct btrfs_key *key, 5022 struct btrfs_chunk *chunk, int item_size) 5023 { 5024 struct btrfs_super_block *super_copy = fs_info->super_copy; 5025 struct btrfs_disk_key disk_key; 5026 u32 array_size; 5027 u8 *ptr; 5028 5029 lockdep_assert_held(&fs_info->chunk_mutex); 5030 5031 array_size = btrfs_super_sys_array_size(super_copy); 5032 if (array_size + item_size + sizeof(disk_key) 5033 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5034 return -EFBIG; 5035 5036 ptr = super_copy->sys_chunk_array + array_size; 5037 btrfs_cpu_key_to_disk(&disk_key, key); 5038 memcpy(ptr, &disk_key, sizeof(disk_key)); 5039 ptr += sizeof(disk_key); 5040 memcpy(ptr, chunk, item_size); 5041 item_size += sizeof(disk_key); 5042 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5043 5044 return 0; 5045 } 5046 5047 /* 5048 * sort the devices in descending order by max_avail, total_avail 5049 */ 5050 static int btrfs_cmp_device_info(const void *a, const void *b) 5051 { 5052 const struct btrfs_device_info *di_a = a; 5053 const struct btrfs_device_info *di_b = b; 5054 5055 if (di_a->max_avail > di_b->max_avail) 5056 return -1; 5057 if (di_a->max_avail < di_b->max_avail) 5058 return 1; 5059 if (di_a->total_avail > di_b->total_avail) 5060 return -1; 5061 if (di_a->total_avail < di_b->total_avail) 5062 return 1; 5063 return 0; 5064 } 5065 5066 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5067 { 5068 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5069 return; 5070 5071 btrfs_set_fs_incompat(info, RAID56); 5072 } 5073 5074 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5075 { 5076 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5077 return; 5078 5079 btrfs_set_fs_incompat(info, RAID1C34); 5080 } 5081 5082 /* 5083 * Structure used internally for btrfs_create_chunk() function. 5084 * Wraps needed parameters. 5085 */ 5086 struct alloc_chunk_ctl { 5087 u64 start; 5088 u64 type; 5089 /* Total number of stripes to allocate */ 5090 int num_stripes; 5091 /* sub_stripes info for map */ 5092 int sub_stripes; 5093 /* Stripes per device */ 5094 int dev_stripes; 5095 /* Maximum number of devices to use */ 5096 int devs_max; 5097 /* Minimum number of devices to use */ 5098 int devs_min; 5099 /* ndevs has to be a multiple of this */ 5100 int devs_increment; 5101 /* Number of copies */ 5102 int ncopies; 5103 /* Number of stripes worth of bytes to store parity information */ 5104 int nparity; 5105 u64 max_stripe_size; 5106 u64 max_chunk_size; 5107 u64 dev_extent_min; 5108 u64 stripe_size; 5109 u64 chunk_size; 5110 int ndevs; 5111 }; 5112 5113 static void init_alloc_chunk_ctl_policy_regular( 5114 struct btrfs_fs_devices *fs_devices, 5115 struct alloc_chunk_ctl *ctl) 5116 { 5117 struct btrfs_space_info *space_info; 5118 5119 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); 5120 ASSERT(space_info); 5121 5122 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); 5123 ctl->max_stripe_size = ctl->max_chunk_size; 5124 5125 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) 5126 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); 5127 5128 /* We don't want a chunk larger than 10% of writable space */ 5129 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), 5130 ctl->max_chunk_size); 5131 ctl->dev_extent_min = ctl->dev_stripes << BTRFS_STRIPE_LEN_SHIFT; 5132 } 5133 5134 static void init_alloc_chunk_ctl_policy_zoned( 5135 struct btrfs_fs_devices *fs_devices, 5136 struct alloc_chunk_ctl *ctl) 5137 { 5138 u64 zone_size = fs_devices->fs_info->zone_size; 5139 u64 limit; 5140 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5141 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5142 u64 min_chunk_size = min_data_stripes * zone_size; 5143 u64 type = ctl->type; 5144 5145 ctl->max_stripe_size = zone_size; 5146 if (type & BTRFS_BLOCK_GROUP_DATA) { 5147 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5148 zone_size); 5149 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5150 ctl->max_chunk_size = ctl->max_stripe_size; 5151 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5152 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5153 ctl->devs_max = min_t(int, ctl->devs_max, 5154 BTRFS_MAX_DEVS_SYS_CHUNK); 5155 } else { 5156 BUG(); 5157 } 5158 5159 /* We don't want a chunk larger than 10% of writable space */ 5160 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), 5161 zone_size), 5162 min_chunk_size); 5163 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5164 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5165 } 5166 5167 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5168 struct alloc_chunk_ctl *ctl) 5169 { 5170 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5171 5172 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5173 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5174 ctl->devs_max = btrfs_raid_array[index].devs_max; 5175 if (!ctl->devs_max) 5176 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5177 ctl->devs_min = btrfs_raid_array[index].devs_min; 5178 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5179 ctl->ncopies = btrfs_raid_array[index].ncopies; 5180 ctl->nparity = btrfs_raid_array[index].nparity; 5181 ctl->ndevs = 0; 5182 5183 switch (fs_devices->chunk_alloc_policy) { 5184 case BTRFS_CHUNK_ALLOC_REGULAR: 5185 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5186 break; 5187 case BTRFS_CHUNK_ALLOC_ZONED: 5188 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5189 break; 5190 default: 5191 BUG(); 5192 } 5193 } 5194 5195 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5196 struct alloc_chunk_ctl *ctl, 5197 struct btrfs_device_info *devices_info) 5198 { 5199 struct btrfs_fs_info *info = fs_devices->fs_info; 5200 struct btrfs_device *device; 5201 u64 total_avail; 5202 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5203 int ret; 5204 int ndevs = 0; 5205 u64 max_avail; 5206 u64 dev_offset; 5207 5208 /* 5209 * in the first pass through the devices list, we gather information 5210 * about the available holes on each device. 5211 */ 5212 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5213 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5214 WARN(1, KERN_ERR 5215 "BTRFS: read-only device in alloc_list\n"); 5216 continue; 5217 } 5218 5219 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5220 &device->dev_state) || 5221 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5222 continue; 5223 5224 if (device->total_bytes > device->bytes_used) 5225 total_avail = device->total_bytes - device->bytes_used; 5226 else 5227 total_avail = 0; 5228 5229 /* If there is no space on this device, skip it. */ 5230 if (total_avail < ctl->dev_extent_min) 5231 continue; 5232 5233 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5234 &max_avail); 5235 if (ret && ret != -ENOSPC) 5236 return ret; 5237 5238 if (ret == 0) 5239 max_avail = dev_extent_want; 5240 5241 if (max_avail < ctl->dev_extent_min) { 5242 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5243 btrfs_debug(info, 5244 "%s: devid %llu has no free space, have=%llu want=%llu", 5245 __func__, device->devid, max_avail, 5246 ctl->dev_extent_min); 5247 continue; 5248 } 5249 5250 if (ndevs == fs_devices->rw_devices) { 5251 WARN(1, "%s: found more than %llu devices\n", 5252 __func__, fs_devices->rw_devices); 5253 break; 5254 } 5255 devices_info[ndevs].dev_offset = dev_offset; 5256 devices_info[ndevs].max_avail = max_avail; 5257 devices_info[ndevs].total_avail = total_avail; 5258 devices_info[ndevs].dev = device; 5259 ++ndevs; 5260 } 5261 ctl->ndevs = ndevs; 5262 5263 /* 5264 * now sort the devices by hole size / available space 5265 */ 5266 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5267 btrfs_cmp_device_info, NULL); 5268 5269 return 0; 5270 } 5271 5272 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5273 struct btrfs_device_info *devices_info) 5274 { 5275 /* Number of stripes that count for block group size */ 5276 int data_stripes; 5277 5278 /* 5279 * The primary goal is to maximize the number of stripes, so use as 5280 * many devices as possible, even if the stripes are not maximum sized. 5281 * 5282 * The DUP profile stores more than one stripe per device, the 5283 * max_avail is the total size so we have to adjust. 5284 */ 5285 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5286 ctl->dev_stripes); 5287 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5288 5289 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5290 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5291 5292 /* 5293 * Use the number of data stripes to figure out how big this chunk is 5294 * really going to be in terms of logical address space, and compare 5295 * that answer with the max chunk size. If it's higher, we try to 5296 * reduce stripe_size. 5297 */ 5298 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5299 /* 5300 * Reduce stripe_size, round it up to a 16MB boundary again and 5301 * then use it, unless it ends up being even bigger than the 5302 * previous value we had already. 5303 */ 5304 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5305 data_stripes), SZ_16M), 5306 ctl->stripe_size); 5307 } 5308 5309 /* Stripe size should not go beyond 1G. */ 5310 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); 5311 5312 /* Align to BTRFS_STRIPE_LEN */ 5313 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5314 ctl->chunk_size = ctl->stripe_size * data_stripes; 5315 5316 return 0; 5317 } 5318 5319 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5320 struct btrfs_device_info *devices_info) 5321 { 5322 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5323 /* Number of stripes that count for block group size */ 5324 int data_stripes; 5325 5326 /* 5327 * It should hold because: 5328 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5329 */ 5330 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5331 5332 ctl->stripe_size = zone_size; 5333 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5334 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5335 5336 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5337 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5338 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5339 ctl->stripe_size) + ctl->nparity, 5340 ctl->dev_stripes); 5341 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5342 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5343 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5344 } 5345 5346 ctl->chunk_size = ctl->stripe_size * data_stripes; 5347 5348 return 0; 5349 } 5350 5351 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5352 struct alloc_chunk_ctl *ctl, 5353 struct btrfs_device_info *devices_info) 5354 { 5355 struct btrfs_fs_info *info = fs_devices->fs_info; 5356 5357 /* 5358 * Round down to number of usable stripes, devs_increment can be any 5359 * number so we can't use round_down() that requires power of 2, while 5360 * rounddown is safe. 5361 */ 5362 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5363 5364 if (ctl->ndevs < ctl->devs_min) { 5365 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5366 btrfs_debug(info, 5367 "%s: not enough devices with free space: have=%d minimum required=%d", 5368 __func__, ctl->ndevs, ctl->devs_min); 5369 } 5370 return -ENOSPC; 5371 } 5372 5373 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5374 5375 switch (fs_devices->chunk_alloc_policy) { 5376 case BTRFS_CHUNK_ALLOC_REGULAR: 5377 return decide_stripe_size_regular(ctl, devices_info); 5378 case BTRFS_CHUNK_ALLOC_ZONED: 5379 return decide_stripe_size_zoned(ctl, devices_info); 5380 default: 5381 BUG(); 5382 } 5383 } 5384 5385 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5386 struct alloc_chunk_ctl *ctl, 5387 struct btrfs_device_info *devices_info) 5388 { 5389 struct btrfs_fs_info *info = trans->fs_info; 5390 struct map_lookup *map = NULL; 5391 struct extent_map_tree *em_tree; 5392 struct btrfs_block_group *block_group; 5393 struct extent_map *em; 5394 u64 start = ctl->start; 5395 u64 type = ctl->type; 5396 int ret; 5397 int i; 5398 int j; 5399 5400 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5401 if (!map) 5402 return ERR_PTR(-ENOMEM); 5403 map->num_stripes = ctl->num_stripes; 5404 5405 for (i = 0; i < ctl->ndevs; ++i) { 5406 for (j = 0; j < ctl->dev_stripes; ++j) { 5407 int s = i * ctl->dev_stripes + j; 5408 map->stripes[s].dev = devices_info[i].dev; 5409 map->stripes[s].physical = devices_info[i].dev_offset + 5410 j * ctl->stripe_size; 5411 } 5412 } 5413 map->io_align = BTRFS_STRIPE_LEN; 5414 map->io_width = BTRFS_STRIPE_LEN; 5415 map->type = type; 5416 map->sub_stripes = ctl->sub_stripes; 5417 5418 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5419 5420 em = alloc_extent_map(); 5421 if (!em) { 5422 kfree(map); 5423 return ERR_PTR(-ENOMEM); 5424 } 5425 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5426 em->map_lookup = map; 5427 em->start = start; 5428 em->len = ctl->chunk_size; 5429 em->block_start = 0; 5430 em->block_len = em->len; 5431 em->orig_block_len = ctl->stripe_size; 5432 5433 em_tree = &info->mapping_tree; 5434 write_lock(&em_tree->lock); 5435 ret = add_extent_mapping(em_tree, em, 0); 5436 if (ret) { 5437 write_unlock(&em_tree->lock); 5438 free_extent_map(em); 5439 return ERR_PTR(ret); 5440 } 5441 write_unlock(&em_tree->lock); 5442 5443 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size); 5444 if (IS_ERR(block_group)) 5445 goto error_del_extent; 5446 5447 for (i = 0; i < map->num_stripes; i++) { 5448 struct btrfs_device *dev = map->stripes[i].dev; 5449 5450 btrfs_device_set_bytes_used(dev, 5451 dev->bytes_used + ctl->stripe_size); 5452 if (list_empty(&dev->post_commit_list)) 5453 list_add_tail(&dev->post_commit_list, 5454 &trans->transaction->dev_update_list); 5455 } 5456 5457 atomic64_sub(ctl->stripe_size * map->num_stripes, 5458 &info->free_chunk_space); 5459 5460 free_extent_map(em); 5461 check_raid56_incompat_flag(info, type); 5462 check_raid1c34_incompat_flag(info, type); 5463 5464 return block_group; 5465 5466 error_del_extent: 5467 write_lock(&em_tree->lock); 5468 remove_extent_mapping(em_tree, em); 5469 write_unlock(&em_tree->lock); 5470 5471 /* One for our allocation */ 5472 free_extent_map(em); 5473 /* One for the tree reference */ 5474 free_extent_map(em); 5475 5476 return block_group; 5477 } 5478 5479 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5480 u64 type) 5481 { 5482 struct btrfs_fs_info *info = trans->fs_info; 5483 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5484 struct btrfs_device_info *devices_info = NULL; 5485 struct alloc_chunk_ctl ctl; 5486 struct btrfs_block_group *block_group; 5487 int ret; 5488 5489 lockdep_assert_held(&info->chunk_mutex); 5490 5491 if (!alloc_profile_is_valid(type, 0)) { 5492 ASSERT(0); 5493 return ERR_PTR(-EINVAL); 5494 } 5495 5496 if (list_empty(&fs_devices->alloc_list)) { 5497 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5498 btrfs_debug(info, "%s: no writable device", __func__); 5499 return ERR_PTR(-ENOSPC); 5500 } 5501 5502 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5503 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5504 ASSERT(0); 5505 return ERR_PTR(-EINVAL); 5506 } 5507 5508 ctl.start = find_next_chunk(info); 5509 ctl.type = type; 5510 init_alloc_chunk_ctl(fs_devices, &ctl); 5511 5512 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5513 GFP_NOFS); 5514 if (!devices_info) 5515 return ERR_PTR(-ENOMEM); 5516 5517 ret = gather_device_info(fs_devices, &ctl, devices_info); 5518 if (ret < 0) { 5519 block_group = ERR_PTR(ret); 5520 goto out; 5521 } 5522 5523 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5524 if (ret < 0) { 5525 block_group = ERR_PTR(ret); 5526 goto out; 5527 } 5528 5529 block_group = create_chunk(trans, &ctl, devices_info); 5530 5531 out: 5532 kfree(devices_info); 5533 return block_group; 5534 } 5535 5536 /* 5537 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5538 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5539 * chunks. 5540 * 5541 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5542 * phases. 5543 */ 5544 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5545 struct btrfs_block_group *bg) 5546 { 5547 struct btrfs_fs_info *fs_info = trans->fs_info; 5548 struct btrfs_root *chunk_root = fs_info->chunk_root; 5549 struct btrfs_key key; 5550 struct btrfs_chunk *chunk; 5551 struct btrfs_stripe *stripe; 5552 struct extent_map *em; 5553 struct map_lookup *map; 5554 size_t item_size; 5555 int i; 5556 int ret; 5557 5558 /* 5559 * We take the chunk_mutex for 2 reasons: 5560 * 5561 * 1) Updates and insertions in the chunk btree must be done while holding 5562 * the chunk_mutex, as well as updating the system chunk array in the 5563 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5564 * details; 5565 * 5566 * 2) To prevent races with the final phase of a device replace operation 5567 * that replaces the device object associated with the map's stripes, 5568 * because the device object's id can change at any time during that 5569 * final phase of the device replace operation 5570 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5571 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5572 * which would cause a failure when updating the device item, which does 5573 * not exists, or persisting a stripe of the chunk item with such ID. 5574 * Here we can't use the device_list_mutex because our caller already 5575 * has locked the chunk_mutex, and the final phase of device replace 5576 * acquires both mutexes - first the device_list_mutex and then the 5577 * chunk_mutex. Using any of those two mutexes protects us from a 5578 * concurrent device replace. 5579 */ 5580 lockdep_assert_held(&fs_info->chunk_mutex); 5581 5582 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5583 if (IS_ERR(em)) { 5584 ret = PTR_ERR(em); 5585 btrfs_abort_transaction(trans, ret); 5586 return ret; 5587 } 5588 5589 map = em->map_lookup; 5590 item_size = btrfs_chunk_item_size(map->num_stripes); 5591 5592 chunk = kzalloc(item_size, GFP_NOFS); 5593 if (!chunk) { 5594 ret = -ENOMEM; 5595 btrfs_abort_transaction(trans, ret); 5596 goto out; 5597 } 5598 5599 for (i = 0; i < map->num_stripes; i++) { 5600 struct btrfs_device *device = map->stripes[i].dev; 5601 5602 ret = btrfs_update_device(trans, device); 5603 if (ret) 5604 goto out; 5605 } 5606 5607 stripe = &chunk->stripe; 5608 for (i = 0; i < map->num_stripes; i++) { 5609 struct btrfs_device *device = map->stripes[i].dev; 5610 const u64 dev_offset = map->stripes[i].physical; 5611 5612 btrfs_set_stack_stripe_devid(stripe, device->devid); 5613 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5614 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5615 stripe++; 5616 } 5617 5618 btrfs_set_stack_chunk_length(chunk, bg->length); 5619 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5620 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN); 5621 btrfs_set_stack_chunk_type(chunk, map->type); 5622 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5623 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN); 5624 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN); 5625 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5626 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5627 5628 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5629 key.type = BTRFS_CHUNK_ITEM_KEY; 5630 key.offset = bg->start; 5631 5632 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5633 if (ret) 5634 goto out; 5635 5636 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); 5637 5638 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5639 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5640 if (ret) 5641 goto out; 5642 } 5643 5644 out: 5645 kfree(chunk); 5646 free_extent_map(em); 5647 return ret; 5648 } 5649 5650 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5651 { 5652 struct btrfs_fs_info *fs_info = trans->fs_info; 5653 u64 alloc_profile; 5654 struct btrfs_block_group *meta_bg; 5655 struct btrfs_block_group *sys_bg; 5656 5657 /* 5658 * When adding a new device for sprouting, the seed device is read-only 5659 * so we must first allocate a metadata and a system chunk. But before 5660 * adding the block group items to the extent, device and chunk btrees, 5661 * we must first: 5662 * 5663 * 1) Create both chunks without doing any changes to the btrees, as 5664 * otherwise we would get -ENOSPC since the block groups from the 5665 * seed device are read-only; 5666 * 5667 * 2) Add the device item for the new sprout device - finishing the setup 5668 * of a new block group requires updating the device item in the chunk 5669 * btree, so it must exist when we attempt to do it. The previous step 5670 * ensures this does not fail with -ENOSPC. 5671 * 5672 * After that we can add the block group items to their btrees: 5673 * update existing device item in the chunk btree, add a new block group 5674 * item to the extent btree, add a new chunk item to the chunk btree and 5675 * finally add the new device extent items to the devices btree. 5676 */ 5677 5678 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5679 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5680 if (IS_ERR(meta_bg)) 5681 return PTR_ERR(meta_bg); 5682 5683 alloc_profile = btrfs_system_alloc_profile(fs_info); 5684 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5685 if (IS_ERR(sys_bg)) 5686 return PTR_ERR(sys_bg); 5687 5688 return 0; 5689 } 5690 5691 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5692 { 5693 const int index = btrfs_bg_flags_to_raid_index(map->type); 5694 5695 return btrfs_raid_array[index].tolerated_failures; 5696 } 5697 5698 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5699 { 5700 struct extent_map *em; 5701 struct map_lookup *map; 5702 int miss_ndevs = 0; 5703 int i; 5704 bool ret = true; 5705 5706 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5707 if (IS_ERR(em)) 5708 return false; 5709 5710 map = em->map_lookup; 5711 for (i = 0; i < map->num_stripes; i++) { 5712 if (test_bit(BTRFS_DEV_STATE_MISSING, 5713 &map->stripes[i].dev->dev_state)) { 5714 miss_ndevs++; 5715 continue; 5716 } 5717 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5718 &map->stripes[i].dev->dev_state)) { 5719 ret = false; 5720 goto end; 5721 } 5722 } 5723 5724 /* 5725 * If the number of missing devices is larger than max errors, we can 5726 * not write the data into that chunk successfully. 5727 */ 5728 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5729 ret = false; 5730 end: 5731 free_extent_map(em); 5732 return ret; 5733 } 5734 5735 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5736 { 5737 struct extent_map *em; 5738 5739 while (1) { 5740 write_lock(&tree->lock); 5741 em = lookup_extent_mapping(tree, 0, (u64)-1); 5742 if (em) 5743 remove_extent_mapping(tree, em); 5744 write_unlock(&tree->lock); 5745 if (!em) 5746 break; 5747 /* once for us */ 5748 free_extent_map(em); 5749 /* once for the tree */ 5750 free_extent_map(em); 5751 } 5752 } 5753 5754 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5755 { 5756 struct extent_map *em; 5757 struct map_lookup *map; 5758 enum btrfs_raid_types index; 5759 int ret = 1; 5760 5761 em = btrfs_get_chunk_map(fs_info, logical, len); 5762 if (IS_ERR(em)) 5763 /* 5764 * We could return errors for these cases, but that could get 5765 * ugly and we'd probably do the same thing which is just not do 5766 * anything else and exit, so return 1 so the callers don't try 5767 * to use other copies. 5768 */ 5769 return 1; 5770 5771 map = em->map_lookup; 5772 index = btrfs_bg_flags_to_raid_index(map->type); 5773 5774 /* Non-RAID56, use their ncopies from btrfs_raid_array. */ 5775 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5776 ret = btrfs_raid_array[index].ncopies; 5777 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5778 ret = 2; 5779 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5780 /* 5781 * There could be two corrupted data stripes, we need 5782 * to loop retry in order to rebuild the correct data. 5783 * 5784 * Fail a stripe at a time on every retry except the 5785 * stripe under reconstruction. 5786 */ 5787 ret = map->num_stripes; 5788 free_extent_map(em); 5789 return ret; 5790 } 5791 5792 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5793 u64 logical) 5794 { 5795 struct extent_map *em; 5796 struct map_lookup *map; 5797 unsigned long len = fs_info->sectorsize; 5798 5799 if (!btrfs_fs_incompat(fs_info, RAID56)) 5800 return len; 5801 5802 em = btrfs_get_chunk_map(fs_info, logical, len); 5803 5804 if (!WARN_ON(IS_ERR(em))) { 5805 map = em->map_lookup; 5806 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5807 len = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT; 5808 free_extent_map(em); 5809 } 5810 return len; 5811 } 5812 5813 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5814 { 5815 struct extent_map *em; 5816 struct map_lookup *map; 5817 int ret = 0; 5818 5819 if (!btrfs_fs_incompat(fs_info, RAID56)) 5820 return 0; 5821 5822 em = btrfs_get_chunk_map(fs_info, logical, len); 5823 5824 if(!WARN_ON(IS_ERR(em))) { 5825 map = em->map_lookup; 5826 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5827 ret = 1; 5828 free_extent_map(em); 5829 } 5830 return ret; 5831 } 5832 5833 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5834 struct map_lookup *map, int first, 5835 int dev_replace_is_ongoing) 5836 { 5837 int i; 5838 int num_stripes; 5839 int preferred_mirror; 5840 int tolerance; 5841 struct btrfs_device *srcdev; 5842 5843 ASSERT((map->type & 5844 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5845 5846 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5847 num_stripes = map->sub_stripes; 5848 else 5849 num_stripes = map->num_stripes; 5850 5851 switch (fs_info->fs_devices->read_policy) { 5852 default: 5853 /* Shouldn't happen, just warn and use pid instead of failing */ 5854 btrfs_warn_rl(fs_info, 5855 "unknown read_policy type %u, reset to pid", 5856 fs_info->fs_devices->read_policy); 5857 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5858 fallthrough; 5859 case BTRFS_READ_POLICY_PID: 5860 preferred_mirror = first + (current->pid % num_stripes); 5861 break; 5862 } 5863 5864 if (dev_replace_is_ongoing && 5865 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5866 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5867 srcdev = fs_info->dev_replace.srcdev; 5868 else 5869 srcdev = NULL; 5870 5871 /* 5872 * try to avoid the drive that is the source drive for a 5873 * dev-replace procedure, only choose it if no other non-missing 5874 * mirror is available 5875 */ 5876 for (tolerance = 0; tolerance < 2; tolerance++) { 5877 if (map->stripes[preferred_mirror].dev->bdev && 5878 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5879 return preferred_mirror; 5880 for (i = first; i < first + num_stripes; i++) { 5881 if (map->stripes[i].dev->bdev && 5882 (tolerance || map->stripes[i].dev != srcdev)) 5883 return i; 5884 } 5885 } 5886 5887 /* we couldn't find one that doesn't fail. Just return something 5888 * and the io error handling code will clean up eventually 5889 */ 5890 return preferred_mirror; 5891 } 5892 5893 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5894 u16 total_stripes) 5895 { 5896 struct btrfs_io_context *bioc; 5897 5898 bioc = kzalloc( 5899 /* The size of btrfs_io_context */ 5900 sizeof(struct btrfs_io_context) + 5901 /* Plus the variable array for the stripes */ 5902 sizeof(struct btrfs_io_stripe) * (total_stripes), 5903 GFP_NOFS); 5904 5905 if (!bioc) 5906 return NULL; 5907 5908 refcount_set(&bioc->refs, 1); 5909 5910 bioc->fs_info = fs_info; 5911 bioc->replace_stripe_src = -1; 5912 bioc->full_stripe_logical = (u64)-1; 5913 5914 return bioc; 5915 } 5916 5917 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5918 { 5919 WARN_ON(!refcount_read(&bioc->refs)); 5920 refcount_inc(&bioc->refs); 5921 } 5922 5923 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5924 { 5925 if (!bioc) 5926 return; 5927 if (refcount_dec_and_test(&bioc->refs)) 5928 kfree(bioc); 5929 } 5930 5931 /* 5932 * Please note that, discard won't be sent to target device of device 5933 * replace. 5934 */ 5935 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, 5936 u64 logical, u64 *length_ret, 5937 u32 *num_stripes) 5938 { 5939 struct extent_map *em; 5940 struct map_lookup *map; 5941 struct btrfs_discard_stripe *stripes; 5942 u64 length = *length_ret; 5943 u64 offset; 5944 u32 stripe_nr; 5945 u32 stripe_nr_end; 5946 u32 stripe_cnt; 5947 u64 stripe_end_offset; 5948 u64 stripe_offset; 5949 u32 stripe_index; 5950 u32 factor = 0; 5951 u32 sub_stripes = 0; 5952 u32 stripes_per_dev = 0; 5953 u32 remaining_stripes = 0; 5954 u32 last_stripe = 0; 5955 int ret; 5956 int i; 5957 5958 em = btrfs_get_chunk_map(fs_info, logical, length); 5959 if (IS_ERR(em)) 5960 return ERR_CAST(em); 5961 5962 map = em->map_lookup; 5963 5964 /* we don't discard raid56 yet */ 5965 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5966 ret = -EOPNOTSUPP; 5967 goto out_free_map; 5968 } 5969 5970 offset = logical - em->start; 5971 length = min_t(u64, em->start + em->len - logical, length); 5972 *length_ret = length; 5973 5974 /* 5975 * stripe_nr counts the total number of stripes we have to stride 5976 * to get to this block 5977 */ 5978 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 5979 5980 /* stripe_offset is the offset of this block in its stripe */ 5981 stripe_offset = offset - (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 5982 5983 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >> 5984 BTRFS_STRIPE_LEN_SHIFT; 5985 stripe_cnt = stripe_nr_end - stripe_nr; 5986 stripe_end_offset = (stripe_nr_end << BTRFS_STRIPE_LEN_SHIFT) - 5987 (offset + length); 5988 /* 5989 * after this, stripe_nr is the number of stripes on this 5990 * device we have to walk to find the data, and stripe_index is 5991 * the number of our device in the stripe array 5992 */ 5993 *num_stripes = 1; 5994 stripe_index = 0; 5995 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5996 BTRFS_BLOCK_GROUP_RAID10)) { 5997 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 5998 sub_stripes = 1; 5999 else 6000 sub_stripes = map->sub_stripes; 6001 6002 factor = map->num_stripes / sub_stripes; 6003 *num_stripes = min_t(u64, map->num_stripes, 6004 sub_stripes * stripe_cnt); 6005 stripe_index = stripe_nr % factor; 6006 stripe_nr /= factor; 6007 stripe_index *= sub_stripes; 6008 6009 remaining_stripes = stripe_cnt % factor; 6010 stripes_per_dev = stripe_cnt / factor; 6011 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes; 6012 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 6013 BTRFS_BLOCK_GROUP_DUP)) { 6014 *num_stripes = map->num_stripes; 6015 } else { 6016 stripe_index = stripe_nr % map->num_stripes; 6017 stripe_nr /= map->num_stripes; 6018 } 6019 6020 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); 6021 if (!stripes) { 6022 ret = -ENOMEM; 6023 goto out_free_map; 6024 } 6025 6026 for (i = 0; i < *num_stripes; i++) { 6027 stripes[i].physical = 6028 map->stripes[stripe_index].physical + 6029 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6030 stripes[i].dev = map->stripes[stripe_index].dev; 6031 6032 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6033 BTRFS_BLOCK_GROUP_RAID10)) { 6034 stripes[i].length = stripes_per_dev << BTRFS_STRIPE_LEN_SHIFT; 6035 6036 if (i / sub_stripes < remaining_stripes) 6037 stripes[i].length += BTRFS_STRIPE_LEN; 6038 6039 /* 6040 * Special for the first stripe and 6041 * the last stripe: 6042 * 6043 * |-------|...|-------| 6044 * |----------| 6045 * off end_off 6046 */ 6047 if (i < sub_stripes) 6048 stripes[i].length -= stripe_offset; 6049 6050 if (stripe_index >= last_stripe && 6051 stripe_index <= (last_stripe + 6052 sub_stripes - 1)) 6053 stripes[i].length -= stripe_end_offset; 6054 6055 if (i == sub_stripes - 1) 6056 stripe_offset = 0; 6057 } else { 6058 stripes[i].length = length; 6059 } 6060 6061 stripe_index++; 6062 if (stripe_index == map->num_stripes) { 6063 stripe_index = 0; 6064 stripe_nr++; 6065 } 6066 } 6067 6068 free_extent_map(em); 6069 return stripes; 6070 out_free_map: 6071 free_extent_map(em); 6072 return ERR_PTR(ret); 6073 } 6074 6075 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6076 { 6077 struct btrfs_block_group *cache; 6078 bool ret; 6079 6080 /* Non zoned filesystem does not use "to_copy" flag */ 6081 if (!btrfs_is_zoned(fs_info)) 6082 return false; 6083 6084 cache = btrfs_lookup_block_group(fs_info, logical); 6085 6086 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 6087 6088 btrfs_put_block_group(cache); 6089 return ret; 6090 } 6091 6092 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6093 struct btrfs_io_context *bioc, 6094 struct btrfs_dev_replace *dev_replace, 6095 u64 logical, 6096 int *num_stripes_ret, int *max_errors_ret) 6097 { 6098 u64 srcdev_devid = dev_replace->srcdev->devid; 6099 /* 6100 * At this stage, num_stripes is still the real number of stripes, 6101 * excluding the duplicated stripes. 6102 */ 6103 int num_stripes = *num_stripes_ret; 6104 int nr_extra_stripes = 0; 6105 int max_errors = *max_errors_ret; 6106 int i; 6107 6108 /* 6109 * A block group which has "to_copy" set will eventually be copied by 6110 * the dev-replace process. We can avoid cloning IO here. 6111 */ 6112 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6113 return; 6114 6115 /* 6116 * Duplicate the write operations while the dev-replace procedure is 6117 * running. Since the copying of the old disk to the new disk takes 6118 * place at run time while the filesystem is mounted writable, the 6119 * regular write operations to the old disk have to be duplicated to go 6120 * to the new disk as well. 6121 * 6122 * Note that device->missing is handled by the caller, and that the 6123 * write to the old disk is already set up in the stripes array. 6124 */ 6125 for (i = 0; i < num_stripes; i++) { 6126 struct btrfs_io_stripe *old = &bioc->stripes[i]; 6127 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes]; 6128 6129 if (old->dev->devid != srcdev_devid) 6130 continue; 6131 6132 new->physical = old->physical; 6133 new->dev = dev_replace->tgtdev; 6134 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) 6135 bioc->replace_stripe_src = i; 6136 nr_extra_stripes++; 6137 } 6138 6139 /* We can only have at most 2 extra nr_stripes (for DUP). */ 6140 ASSERT(nr_extra_stripes <= 2); 6141 /* 6142 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for 6143 * replace. 6144 * If we have 2 extra stripes, only choose the one with smaller physical. 6145 */ 6146 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) { 6147 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes]; 6148 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1]; 6149 6150 /* Only DUP can have two extra stripes. */ 6151 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP); 6152 6153 /* 6154 * Swap the last stripe stripes and reduce @nr_extra_stripes. 6155 * The extra stripe would still be there, but won't be accessed. 6156 */ 6157 if (first->physical > second->physical) { 6158 swap(second->physical, first->physical); 6159 swap(second->dev, first->dev); 6160 nr_extra_stripes--; 6161 } 6162 } 6163 6164 *num_stripes_ret = num_stripes + nr_extra_stripes; 6165 *max_errors_ret = max_errors + nr_extra_stripes; 6166 bioc->replace_nr_stripes = nr_extra_stripes; 6167 } 6168 6169 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op, 6170 u64 offset, u32 *stripe_nr, u64 *stripe_offset, 6171 u64 *full_stripe_start) 6172 { 6173 /* 6174 * Stripe_nr is the stripe where this block falls. stripe_offset is 6175 * the offset of this block in its stripe. 6176 */ 6177 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK; 6178 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 6179 ASSERT(*stripe_offset < U32_MAX); 6180 6181 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6182 unsigned long full_stripe_len = nr_data_stripes(map) << 6183 BTRFS_STRIPE_LEN_SHIFT; 6184 6185 /* 6186 * For full stripe start, we use previously calculated 6187 * @stripe_nr. Align it to nr_data_stripes, then multiply with 6188 * STRIPE_LEN. 6189 * 6190 * By this we can avoid u64 division completely. And we have 6191 * to go rounddown(), not round_down(), as nr_data_stripes is 6192 * not ensured to be power of 2. 6193 */ 6194 *full_stripe_start = 6195 rounddown(*stripe_nr, nr_data_stripes(map)) << 6196 BTRFS_STRIPE_LEN_SHIFT; 6197 6198 /* 6199 * For writes to RAID56, allow to write a full stripe set, but 6200 * no straddling of stripe sets. 6201 */ 6202 if (op == BTRFS_MAP_WRITE) 6203 return full_stripe_len - (offset - *full_stripe_start); 6204 } 6205 6206 /* 6207 * For other RAID types and for RAID56 reads, allow a single stripe (on 6208 * a single disk). 6209 */ 6210 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) 6211 return BTRFS_STRIPE_LEN - *stripe_offset; 6212 return U64_MAX; 6213 } 6214 6215 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, 6216 u32 stripe_index, u64 stripe_offset, u32 stripe_nr) 6217 { 6218 dst->dev = map->stripes[stripe_index].dev; 6219 dst->physical = map->stripes[stripe_index].physical + 6220 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6221 } 6222 6223 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6224 u64 logical, u64 *length, 6225 struct btrfs_io_context **bioc_ret, 6226 struct btrfs_io_stripe *smap, int *mirror_num_ret, 6227 int need_raid_map) 6228 { 6229 struct extent_map *em; 6230 struct map_lookup *map; 6231 u64 map_offset; 6232 u64 stripe_offset; 6233 u32 stripe_nr; 6234 u32 stripe_index; 6235 int data_stripes; 6236 int i; 6237 int ret = 0; 6238 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); 6239 int num_stripes; 6240 int num_copies; 6241 int max_errors = 0; 6242 struct btrfs_io_context *bioc = NULL; 6243 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6244 int dev_replace_is_ongoing = 0; 6245 u16 num_alloc_stripes; 6246 u64 raid56_full_stripe_start = (u64)-1; 6247 u64 max_len; 6248 6249 ASSERT(bioc_ret); 6250 6251 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize); 6252 if (mirror_num > num_copies) 6253 return -EINVAL; 6254 6255 em = btrfs_get_chunk_map(fs_info, logical, *length); 6256 if (IS_ERR(em)) 6257 return PTR_ERR(em); 6258 6259 map = em->map_lookup; 6260 data_stripes = nr_data_stripes(map); 6261 6262 map_offset = logical - em->start; 6263 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr, 6264 &stripe_offset, &raid56_full_stripe_start); 6265 *length = min_t(u64, em->len - map_offset, max_len); 6266 6267 down_read(&dev_replace->rwsem); 6268 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6269 /* 6270 * Hold the semaphore for read during the whole operation, write is 6271 * requested at commit time but must wait. 6272 */ 6273 if (!dev_replace_is_ongoing) 6274 up_read(&dev_replace->rwsem); 6275 6276 num_stripes = 1; 6277 stripe_index = 0; 6278 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6279 stripe_index = stripe_nr % map->num_stripes; 6280 stripe_nr /= map->num_stripes; 6281 if (op == BTRFS_MAP_READ) 6282 mirror_num = 1; 6283 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6284 if (op != BTRFS_MAP_READ) { 6285 num_stripes = map->num_stripes; 6286 } else if (mirror_num) { 6287 stripe_index = mirror_num - 1; 6288 } else { 6289 stripe_index = find_live_mirror(fs_info, map, 0, 6290 dev_replace_is_ongoing); 6291 mirror_num = stripe_index + 1; 6292 } 6293 6294 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6295 if (op != BTRFS_MAP_READ) { 6296 num_stripes = map->num_stripes; 6297 } else if (mirror_num) { 6298 stripe_index = mirror_num - 1; 6299 } else { 6300 mirror_num = 1; 6301 } 6302 6303 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6304 u32 factor = map->num_stripes / map->sub_stripes; 6305 6306 stripe_index = (stripe_nr % factor) * map->sub_stripes; 6307 stripe_nr /= factor; 6308 6309 if (op != BTRFS_MAP_READ) 6310 num_stripes = map->sub_stripes; 6311 else if (mirror_num) 6312 stripe_index += mirror_num - 1; 6313 else { 6314 int old_stripe_index = stripe_index; 6315 stripe_index = find_live_mirror(fs_info, map, 6316 stripe_index, 6317 dev_replace_is_ongoing); 6318 mirror_num = stripe_index - old_stripe_index + 1; 6319 } 6320 6321 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6322 if (need_raid_map && (op != BTRFS_MAP_READ || mirror_num > 1)) { 6323 /* 6324 * Push stripe_nr back to the start of the full stripe 6325 * For those cases needing a full stripe, @stripe_nr 6326 * is the full stripe number. 6327 * 6328 * Originally we go raid56_full_stripe_start / full_stripe_len, 6329 * but that can be expensive. Here we just divide 6330 * @stripe_nr with @data_stripes. 6331 */ 6332 stripe_nr /= data_stripes; 6333 6334 /* RAID[56] write or recovery. Return all stripes */ 6335 num_stripes = map->num_stripes; 6336 max_errors = btrfs_chunk_max_errors(map); 6337 6338 /* Return the length to the full stripe end */ 6339 *length = min(logical + *length, 6340 raid56_full_stripe_start + em->start + 6341 (data_stripes << BTRFS_STRIPE_LEN_SHIFT)) - logical; 6342 stripe_index = 0; 6343 stripe_offset = 0; 6344 } else { 6345 /* 6346 * Mirror #0 or #1 means the original data block. 6347 * Mirror #2 is RAID5 parity block. 6348 * Mirror #3 is RAID6 Q block. 6349 */ 6350 stripe_index = stripe_nr % data_stripes; 6351 stripe_nr /= data_stripes; 6352 if (mirror_num > 1) 6353 stripe_index = data_stripes + mirror_num - 2; 6354 6355 /* We distribute the parity blocks across stripes */ 6356 stripe_index = (stripe_nr + stripe_index) % map->num_stripes; 6357 if (op == BTRFS_MAP_READ && mirror_num <= 1) 6358 mirror_num = 1; 6359 } 6360 } else { 6361 /* 6362 * After this, stripe_nr is the number of stripes on this 6363 * device we have to walk to find the data, and stripe_index is 6364 * the number of our device in the stripe array 6365 */ 6366 stripe_index = stripe_nr % map->num_stripes; 6367 stripe_nr /= map->num_stripes; 6368 mirror_num = stripe_index + 1; 6369 } 6370 if (stripe_index >= map->num_stripes) { 6371 btrfs_crit(fs_info, 6372 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6373 stripe_index, map->num_stripes); 6374 ret = -EINVAL; 6375 goto out; 6376 } 6377 6378 num_alloc_stripes = num_stripes; 6379 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6380 op != BTRFS_MAP_READ) 6381 /* 6382 * For replace case, we need to add extra stripes for extra 6383 * duplicated stripes. 6384 * 6385 * For both WRITE and GET_READ_MIRRORS, we may have at most 6386 * 2 more stripes (DUP types, otherwise 1). 6387 */ 6388 num_alloc_stripes += 2; 6389 6390 /* 6391 * If this I/O maps to a single device, try to return the device and 6392 * physical block information on the stack instead of allocating an 6393 * I/O context structure. 6394 */ 6395 if (smap && num_alloc_stripes == 1 && 6396 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && 6397 (op == BTRFS_MAP_READ || !dev_replace_is_ongoing || 6398 !dev_replace->tgtdev)) { 6399 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr); 6400 if (mirror_num_ret) 6401 *mirror_num_ret = mirror_num; 6402 *bioc_ret = NULL; 6403 ret = 0; 6404 goto out; 6405 } 6406 6407 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes); 6408 if (!bioc) { 6409 ret = -ENOMEM; 6410 goto out; 6411 } 6412 bioc->map_type = map->type; 6413 6414 /* 6415 * For RAID56 full map, we need to make sure the stripes[] follows the 6416 * rule that data stripes are all ordered, then followed with P and Q 6417 * (if we have). 6418 * 6419 * It's still mostly the same as other profiles, just with extra rotation. 6420 */ 6421 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6422 (op != BTRFS_MAP_READ || mirror_num > 1)) { 6423 /* 6424 * For RAID56 @stripe_nr is already the number of full stripes 6425 * before us, which is also the rotation value (needs to modulo 6426 * with num_stripes). 6427 * 6428 * In this case, we just add @stripe_nr with @i, then do the 6429 * modulo, to reduce one modulo call. 6430 */ 6431 bioc->full_stripe_logical = em->start + 6432 ((stripe_nr * data_stripes) << BTRFS_STRIPE_LEN_SHIFT); 6433 for (i = 0; i < num_stripes; i++) 6434 set_io_stripe(&bioc->stripes[i], map, 6435 (i + stripe_nr) % num_stripes, 6436 stripe_offset, stripe_nr); 6437 } else { 6438 /* 6439 * For all other non-RAID56 profiles, just copy the target 6440 * stripe into the bioc. 6441 */ 6442 for (i = 0; i < num_stripes; i++) { 6443 set_io_stripe(&bioc->stripes[i], map, stripe_index, 6444 stripe_offset, stripe_nr); 6445 stripe_index++; 6446 } 6447 } 6448 6449 if (op != BTRFS_MAP_READ) 6450 max_errors = btrfs_chunk_max_errors(map); 6451 6452 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6453 op != BTRFS_MAP_READ) { 6454 handle_ops_on_dev_replace(op, bioc, dev_replace, logical, 6455 &num_stripes, &max_errors); 6456 } 6457 6458 *bioc_ret = bioc; 6459 bioc->num_stripes = num_stripes; 6460 bioc->max_errors = max_errors; 6461 bioc->mirror_num = mirror_num; 6462 6463 out: 6464 if (dev_replace_is_ongoing) { 6465 lockdep_assert_held(&dev_replace->rwsem); 6466 /* Unlock and let waiting writers proceed */ 6467 up_read(&dev_replace->rwsem); 6468 } 6469 free_extent_map(em); 6470 return ret; 6471 } 6472 6473 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6474 const struct btrfs_fs_devices *fs_devices) 6475 { 6476 if (args->fsid == NULL) 6477 return true; 6478 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6479 return true; 6480 return false; 6481 } 6482 6483 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6484 const struct btrfs_device *device) 6485 { 6486 if (args->missing) { 6487 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6488 !device->bdev) 6489 return true; 6490 return false; 6491 } 6492 6493 if (device->devid != args->devid) 6494 return false; 6495 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6496 return false; 6497 return true; 6498 } 6499 6500 /* 6501 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6502 * return NULL. 6503 * 6504 * If devid and uuid are both specified, the match must be exact, otherwise 6505 * only devid is used. 6506 */ 6507 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6508 const struct btrfs_dev_lookup_args *args) 6509 { 6510 struct btrfs_device *device; 6511 struct btrfs_fs_devices *seed_devs; 6512 6513 if (dev_args_match_fs_devices(args, fs_devices)) { 6514 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6515 if (dev_args_match_device(args, device)) 6516 return device; 6517 } 6518 } 6519 6520 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6521 if (!dev_args_match_fs_devices(args, seed_devs)) 6522 continue; 6523 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6524 if (dev_args_match_device(args, device)) 6525 return device; 6526 } 6527 } 6528 6529 return NULL; 6530 } 6531 6532 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6533 u64 devid, u8 *dev_uuid) 6534 { 6535 struct btrfs_device *device; 6536 unsigned int nofs_flag; 6537 6538 /* 6539 * We call this under the chunk_mutex, so we want to use NOFS for this 6540 * allocation, however we don't want to change btrfs_alloc_device() to 6541 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6542 * places. 6543 */ 6544 6545 nofs_flag = memalloc_nofs_save(); 6546 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); 6547 memalloc_nofs_restore(nofs_flag); 6548 if (IS_ERR(device)) 6549 return device; 6550 6551 list_add(&device->dev_list, &fs_devices->devices); 6552 device->fs_devices = fs_devices; 6553 fs_devices->num_devices++; 6554 6555 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6556 fs_devices->missing_devices++; 6557 6558 return device; 6559 } 6560 6561 /* 6562 * Allocate new device struct, set up devid and UUID. 6563 * 6564 * @fs_info: used only for generating a new devid, can be NULL if 6565 * devid is provided (i.e. @devid != NULL). 6566 * @devid: a pointer to devid for this device. If NULL a new devid 6567 * is generated. 6568 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6569 * is generated. 6570 * @path: a pointer to device path if available, NULL otherwise. 6571 * 6572 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6573 * on error. Returned struct is not linked onto any lists and must be 6574 * destroyed with btrfs_free_device. 6575 */ 6576 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6577 const u64 *devid, const u8 *uuid, 6578 const char *path) 6579 { 6580 struct btrfs_device *dev; 6581 u64 tmp; 6582 6583 if (WARN_ON(!devid && !fs_info)) 6584 return ERR_PTR(-EINVAL); 6585 6586 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6587 if (!dev) 6588 return ERR_PTR(-ENOMEM); 6589 6590 INIT_LIST_HEAD(&dev->dev_list); 6591 INIT_LIST_HEAD(&dev->dev_alloc_list); 6592 INIT_LIST_HEAD(&dev->post_commit_list); 6593 6594 atomic_set(&dev->dev_stats_ccnt, 0); 6595 btrfs_device_data_ordered_init(dev); 6596 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); 6597 6598 if (devid) 6599 tmp = *devid; 6600 else { 6601 int ret; 6602 6603 ret = find_next_devid(fs_info, &tmp); 6604 if (ret) { 6605 btrfs_free_device(dev); 6606 return ERR_PTR(ret); 6607 } 6608 } 6609 dev->devid = tmp; 6610 6611 if (uuid) 6612 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6613 else 6614 generate_random_uuid(dev->uuid); 6615 6616 if (path) { 6617 struct rcu_string *name; 6618 6619 name = rcu_string_strdup(path, GFP_KERNEL); 6620 if (!name) { 6621 btrfs_free_device(dev); 6622 return ERR_PTR(-ENOMEM); 6623 } 6624 rcu_assign_pointer(dev->name, name); 6625 } 6626 6627 return dev; 6628 } 6629 6630 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6631 u64 devid, u8 *uuid, bool error) 6632 { 6633 if (error) 6634 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6635 devid, uuid); 6636 else 6637 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6638 devid, uuid); 6639 } 6640 6641 u64 btrfs_calc_stripe_length(const struct extent_map *em) 6642 { 6643 const struct map_lookup *map = em->map_lookup; 6644 const int data_stripes = calc_data_stripes(map->type, map->num_stripes); 6645 6646 return div_u64(em->len, data_stripes); 6647 } 6648 6649 #if BITS_PER_LONG == 32 6650 /* 6651 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6652 * can't be accessed on 32bit systems. 6653 * 6654 * This function do mount time check to reject the fs if it already has 6655 * metadata chunk beyond that limit. 6656 */ 6657 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6658 u64 logical, u64 length, u64 type) 6659 { 6660 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6661 return 0; 6662 6663 if (logical + length < MAX_LFS_FILESIZE) 6664 return 0; 6665 6666 btrfs_err_32bit_limit(fs_info); 6667 return -EOVERFLOW; 6668 } 6669 6670 /* 6671 * This is to give early warning for any metadata chunk reaching 6672 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6673 * Although we can still access the metadata, it's not going to be possible 6674 * once the limit is reached. 6675 */ 6676 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6677 u64 logical, u64 length, u64 type) 6678 { 6679 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6680 return; 6681 6682 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 6683 return; 6684 6685 btrfs_warn_32bit_limit(fs_info); 6686 } 6687 #endif 6688 6689 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 6690 u64 devid, u8 *uuid) 6691 { 6692 struct btrfs_device *dev; 6693 6694 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6695 btrfs_report_missing_device(fs_info, devid, uuid, true); 6696 return ERR_PTR(-ENOENT); 6697 } 6698 6699 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 6700 if (IS_ERR(dev)) { 6701 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 6702 devid, PTR_ERR(dev)); 6703 return dev; 6704 } 6705 btrfs_report_missing_device(fs_info, devid, uuid, false); 6706 6707 return dev; 6708 } 6709 6710 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 6711 struct btrfs_chunk *chunk) 6712 { 6713 BTRFS_DEV_LOOKUP_ARGS(args); 6714 struct btrfs_fs_info *fs_info = leaf->fs_info; 6715 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 6716 struct map_lookup *map; 6717 struct extent_map *em; 6718 u64 logical; 6719 u64 length; 6720 u64 devid; 6721 u64 type; 6722 u8 uuid[BTRFS_UUID_SIZE]; 6723 int index; 6724 int num_stripes; 6725 int ret; 6726 int i; 6727 6728 logical = key->offset; 6729 length = btrfs_chunk_length(leaf, chunk); 6730 type = btrfs_chunk_type(leaf, chunk); 6731 index = btrfs_bg_flags_to_raid_index(type); 6732 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 6733 6734 #if BITS_PER_LONG == 32 6735 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 6736 if (ret < 0) 6737 return ret; 6738 warn_32bit_meta_chunk(fs_info, logical, length, type); 6739 #endif 6740 6741 /* 6742 * Only need to verify chunk item if we're reading from sys chunk array, 6743 * as chunk item in tree block is already verified by tree-checker. 6744 */ 6745 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 6746 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 6747 if (ret) 6748 return ret; 6749 } 6750 6751 read_lock(&map_tree->lock); 6752 em = lookup_extent_mapping(map_tree, logical, 1); 6753 read_unlock(&map_tree->lock); 6754 6755 /* already mapped? */ 6756 if (em && em->start <= logical && em->start + em->len > logical) { 6757 free_extent_map(em); 6758 return 0; 6759 } else if (em) { 6760 free_extent_map(em); 6761 } 6762 6763 em = alloc_extent_map(); 6764 if (!em) 6765 return -ENOMEM; 6766 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 6767 if (!map) { 6768 free_extent_map(em); 6769 return -ENOMEM; 6770 } 6771 6772 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 6773 em->map_lookup = map; 6774 em->start = logical; 6775 em->len = length; 6776 em->orig_start = 0; 6777 em->block_start = 0; 6778 em->block_len = em->len; 6779 6780 map->num_stripes = num_stripes; 6781 map->io_width = btrfs_chunk_io_width(leaf, chunk); 6782 map->io_align = btrfs_chunk_io_align(leaf, chunk); 6783 map->type = type; 6784 /* 6785 * We can't use the sub_stripes value, as for profiles other than 6786 * RAID10, they may have 0 as sub_stripes for filesystems created by 6787 * older mkfs (<v5.4). 6788 * In that case, it can cause divide-by-zero errors later. 6789 * Since currently sub_stripes is fixed for each profile, let's 6790 * use the trusted value instead. 6791 */ 6792 map->sub_stripes = btrfs_raid_array[index].sub_stripes; 6793 map->verified_stripes = 0; 6794 em->orig_block_len = btrfs_calc_stripe_length(em); 6795 for (i = 0; i < num_stripes; i++) { 6796 map->stripes[i].physical = 6797 btrfs_stripe_offset_nr(leaf, chunk, i); 6798 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 6799 args.devid = devid; 6800 read_extent_buffer(leaf, uuid, (unsigned long) 6801 btrfs_stripe_dev_uuid_nr(chunk, i), 6802 BTRFS_UUID_SIZE); 6803 args.uuid = uuid; 6804 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 6805 if (!map->stripes[i].dev) { 6806 map->stripes[i].dev = handle_missing_device(fs_info, 6807 devid, uuid); 6808 if (IS_ERR(map->stripes[i].dev)) { 6809 ret = PTR_ERR(map->stripes[i].dev); 6810 free_extent_map(em); 6811 return ret; 6812 } 6813 } 6814 6815 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 6816 &(map->stripes[i].dev->dev_state)); 6817 } 6818 6819 write_lock(&map_tree->lock); 6820 ret = add_extent_mapping(map_tree, em, 0); 6821 write_unlock(&map_tree->lock); 6822 if (ret < 0) { 6823 btrfs_err(fs_info, 6824 "failed to add chunk map, start=%llu len=%llu: %d", 6825 em->start, em->len, ret); 6826 } 6827 free_extent_map(em); 6828 6829 return ret; 6830 } 6831 6832 static void fill_device_from_item(struct extent_buffer *leaf, 6833 struct btrfs_dev_item *dev_item, 6834 struct btrfs_device *device) 6835 { 6836 unsigned long ptr; 6837 6838 device->devid = btrfs_device_id(leaf, dev_item); 6839 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 6840 device->total_bytes = device->disk_total_bytes; 6841 device->commit_total_bytes = device->disk_total_bytes; 6842 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 6843 device->commit_bytes_used = device->bytes_used; 6844 device->type = btrfs_device_type(leaf, dev_item); 6845 device->io_align = btrfs_device_io_align(leaf, dev_item); 6846 device->io_width = btrfs_device_io_width(leaf, dev_item); 6847 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 6848 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 6849 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 6850 6851 ptr = btrfs_device_uuid(dev_item); 6852 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 6853 } 6854 6855 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 6856 u8 *fsid) 6857 { 6858 struct btrfs_fs_devices *fs_devices; 6859 int ret; 6860 6861 lockdep_assert_held(&uuid_mutex); 6862 ASSERT(fsid); 6863 6864 /* This will match only for multi-device seed fs */ 6865 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 6866 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 6867 return fs_devices; 6868 6869 6870 fs_devices = find_fsid(fsid, NULL); 6871 if (!fs_devices) { 6872 if (!btrfs_test_opt(fs_info, DEGRADED)) 6873 return ERR_PTR(-ENOENT); 6874 6875 fs_devices = alloc_fs_devices(fsid, NULL); 6876 if (IS_ERR(fs_devices)) 6877 return fs_devices; 6878 6879 fs_devices->seeding = true; 6880 fs_devices->opened = 1; 6881 return fs_devices; 6882 } 6883 6884 /* 6885 * Upon first call for a seed fs fsid, just create a private copy of the 6886 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 6887 */ 6888 fs_devices = clone_fs_devices(fs_devices); 6889 if (IS_ERR(fs_devices)) 6890 return fs_devices; 6891 6892 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 6893 if (ret) { 6894 free_fs_devices(fs_devices); 6895 return ERR_PTR(ret); 6896 } 6897 6898 if (!fs_devices->seeding) { 6899 close_fs_devices(fs_devices); 6900 free_fs_devices(fs_devices); 6901 return ERR_PTR(-EINVAL); 6902 } 6903 6904 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 6905 6906 return fs_devices; 6907 } 6908 6909 static int read_one_dev(struct extent_buffer *leaf, 6910 struct btrfs_dev_item *dev_item) 6911 { 6912 BTRFS_DEV_LOOKUP_ARGS(args); 6913 struct btrfs_fs_info *fs_info = leaf->fs_info; 6914 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 6915 struct btrfs_device *device; 6916 u64 devid; 6917 int ret; 6918 u8 fs_uuid[BTRFS_FSID_SIZE]; 6919 u8 dev_uuid[BTRFS_UUID_SIZE]; 6920 6921 devid = btrfs_device_id(leaf, dev_item); 6922 args.devid = devid; 6923 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 6924 BTRFS_UUID_SIZE); 6925 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 6926 BTRFS_FSID_SIZE); 6927 args.uuid = dev_uuid; 6928 args.fsid = fs_uuid; 6929 6930 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 6931 fs_devices = open_seed_devices(fs_info, fs_uuid); 6932 if (IS_ERR(fs_devices)) 6933 return PTR_ERR(fs_devices); 6934 } 6935 6936 device = btrfs_find_device(fs_info->fs_devices, &args); 6937 if (!device) { 6938 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6939 btrfs_report_missing_device(fs_info, devid, 6940 dev_uuid, true); 6941 return -ENOENT; 6942 } 6943 6944 device = add_missing_dev(fs_devices, devid, dev_uuid); 6945 if (IS_ERR(device)) { 6946 btrfs_err(fs_info, 6947 "failed to add missing dev %llu: %ld", 6948 devid, PTR_ERR(device)); 6949 return PTR_ERR(device); 6950 } 6951 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 6952 } else { 6953 if (!device->bdev) { 6954 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6955 btrfs_report_missing_device(fs_info, 6956 devid, dev_uuid, true); 6957 return -ENOENT; 6958 } 6959 btrfs_report_missing_device(fs_info, devid, 6960 dev_uuid, false); 6961 } 6962 6963 if (!device->bdev && 6964 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 6965 /* 6966 * this happens when a device that was properly setup 6967 * in the device info lists suddenly goes bad. 6968 * device->bdev is NULL, and so we have to set 6969 * device->missing to one here 6970 */ 6971 device->fs_devices->missing_devices++; 6972 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6973 } 6974 6975 /* Move the device to its own fs_devices */ 6976 if (device->fs_devices != fs_devices) { 6977 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 6978 &device->dev_state)); 6979 6980 list_move(&device->dev_list, &fs_devices->devices); 6981 device->fs_devices->num_devices--; 6982 fs_devices->num_devices++; 6983 6984 device->fs_devices->missing_devices--; 6985 fs_devices->missing_devices++; 6986 6987 device->fs_devices = fs_devices; 6988 } 6989 } 6990 6991 if (device->fs_devices != fs_info->fs_devices) { 6992 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 6993 if (device->generation != 6994 btrfs_device_generation(leaf, dev_item)) 6995 return -EINVAL; 6996 } 6997 6998 fill_device_from_item(leaf, dev_item, device); 6999 if (device->bdev) { 7000 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7001 7002 if (device->total_bytes > max_total_bytes) { 7003 btrfs_err(fs_info, 7004 "device total_bytes should be at most %llu but found %llu", 7005 max_total_bytes, device->total_bytes); 7006 return -EINVAL; 7007 } 7008 } 7009 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7010 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7011 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7012 device->fs_devices->total_rw_bytes += device->total_bytes; 7013 atomic64_add(device->total_bytes - device->bytes_used, 7014 &fs_info->free_chunk_space); 7015 } 7016 ret = 0; 7017 return ret; 7018 } 7019 7020 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7021 { 7022 struct btrfs_super_block *super_copy = fs_info->super_copy; 7023 struct extent_buffer *sb; 7024 struct btrfs_disk_key *disk_key; 7025 struct btrfs_chunk *chunk; 7026 u8 *array_ptr; 7027 unsigned long sb_array_offset; 7028 int ret = 0; 7029 u32 num_stripes; 7030 u32 array_size; 7031 u32 len = 0; 7032 u32 cur_offset; 7033 u64 type; 7034 struct btrfs_key key; 7035 7036 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7037 7038 /* 7039 * We allocated a dummy extent, just to use extent buffer accessors. 7040 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but 7041 * that's fine, we will not go beyond system chunk array anyway. 7042 */ 7043 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); 7044 if (!sb) 7045 return -ENOMEM; 7046 set_extent_buffer_uptodate(sb); 7047 7048 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7049 array_size = btrfs_super_sys_array_size(super_copy); 7050 7051 array_ptr = super_copy->sys_chunk_array; 7052 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7053 cur_offset = 0; 7054 7055 while (cur_offset < array_size) { 7056 disk_key = (struct btrfs_disk_key *)array_ptr; 7057 len = sizeof(*disk_key); 7058 if (cur_offset + len > array_size) 7059 goto out_short_read; 7060 7061 btrfs_disk_key_to_cpu(&key, disk_key); 7062 7063 array_ptr += len; 7064 sb_array_offset += len; 7065 cur_offset += len; 7066 7067 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7068 btrfs_err(fs_info, 7069 "unexpected item type %u in sys_array at offset %u", 7070 (u32)key.type, cur_offset); 7071 ret = -EIO; 7072 break; 7073 } 7074 7075 chunk = (struct btrfs_chunk *)sb_array_offset; 7076 /* 7077 * At least one btrfs_chunk with one stripe must be present, 7078 * exact stripe count check comes afterwards 7079 */ 7080 len = btrfs_chunk_item_size(1); 7081 if (cur_offset + len > array_size) 7082 goto out_short_read; 7083 7084 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7085 if (!num_stripes) { 7086 btrfs_err(fs_info, 7087 "invalid number of stripes %u in sys_array at offset %u", 7088 num_stripes, cur_offset); 7089 ret = -EIO; 7090 break; 7091 } 7092 7093 type = btrfs_chunk_type(sb, chunk); 7094 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7095 btrfs_err(fs_info, 7096 "invalid chunk type %llu in sys_array at offset %u", 7097 type, cur_offset); 7098 ret = -EIO; 7099 break; 7100 } 7101 7102 len = btrfs_chunk_item_size(num_stripes); 7103 if (cur_offset + len > array_size) 7104 goto out_short_read; 7105 7106 ret = read_one_chunk(&key, sb, chunk); 7107 if (ret) 7108 break; 7109 7110 array_ptr += len; 7111 sb_array_offset += len; 7112 cur_offset += len; 7113 } 7114 clear_extent_buffer_uptodate(sb); 7115 free_extent_buffer_stale(sb); 7116 return ret; 7117 7118 out_short_read: 7119 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7120 len, cur_offset); 7121 clear_extent_buffer_uptodate(sb); 7122 free_extent_buffer_stale(sb); 7123 return -EIO; 7124 } 7125 7126 /* 7127 * Check if all chunks in the fs are OK for read-write degraded mount 7128 * 7129 * If the @failing_dev is specified, it's accounted as missing. 7130 * 7131 * Return true if all chunks meet the minimal RW mount requirements. 7132 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7133 */ 7134 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7135 struct btrfs_device *failing_dev) 7136 { 7137 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7138 struct extent_map *em; 7139 u64 next_start = 0; 7140 bool ret = true; 7141 7142 read_lock(&map_tree->lock); 7143 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7144 read_unlock(&map_tree->lock); 7145 /* No chunk at all? Return false anyway */ 7146 if (!em) { 7147 ret = false; 7148 goto out; 7149 } 7150 while (em) { 7151 struct map_lookup *map; 7152 int missing = 0; 7153 int max_tolerated; 7154 int i; 7155 7156 map = em->map_lookup; 7157 max_tolerated = 7158 btrfs_get_num_tolerated_disk_barrier_failures( 7159 map->type); 7160 for (i = 0; i < map->num_stripes; i++) { 7161 struct btrfs_device *dev = map->stripes[i].dev; 7162 7163 if (!dev || !dev->bdev || 7164 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7165 dev->last_flush_error) 7166 missing++; 7167 else if (failing_dev && failing_dev == dev) 7168 missing++; 7169 } 7170 if (missing > max_tolerated) { 7171 if (!failing_dev) 7172 btrfs_warn(fs_info, 7173 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7174 em->start, missing, max_tolerated); 7175 free_extent_map(em); 7176 ret = false; 7177 goto out; 7178 } 7179 next_start = extent_map_end(em); 7180 free_extent_map(em); 7181 7182 read_lock(&map_tree->lock); 7183 em = lookup_extent_mapping(map_tree, next_start, 7184 (u64)(-1) - next_start); 7185 read_unlock(&map_tree->lock); 7186 } 7187 out: 7188 return ret; 7189 } 7190 7191 static void readahead_tree_node_children(struct extent_buffer *node) 7192 { 7193 int i; 7194 const int nr_items = btrfs_header_nritems(node); 7195 7196 for (i = 0; i < nr_items; i++) 7197 btrfs_readahead_node_child(node, i); 7198 } 7199 7200 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7201 { 7202 struct btrfs_root *root = fs_info->chunk_root; 7203 struct btrfs_path *path; 7204 struct extent_buffer *leaf; 7205 struct btrfs_key key; 7206 struct btrfs_key found_key; 7207 int ret; 7208 int slot; 7209 int iter_ret = 0; 7210 u64 total_dev = 0; 7211 u64 last_ra_node = 0; 7212 7213 path = btrfs_alloc_path(); 7214 if (!path) 7215 return -ENOMEM; 7216 7217 /* 7218 * uuid_mutex is needed only if we are mounting a sprout FS 7219 * otherwise we don't need it. 7220 */ 7221 mutex_lock(&uuid_mutex); 7222 7223 /* 7224 * It is possible for mount and umount to race in such a way that 7225 * we execute this code path, but open_fs_devices failed to clear 7226 * total_rw_bytes. We certainly want it cleared before reading the 7227 * device items, so clear it here. 7228 */ 7229 fs_info->fs_devices->total_rw_bytes = 0; 7230 7231 /* 7232 * Lockdep complains about possible circular locking dependency between 7233 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7234 * used for freeze procection of a fs (struct super_block.s_writers), 7235 * which we take when starting a transaction, and extent buffers of the 7236 * chunk tree if we call read_one_dev() while holding a lock on an 7237 * extent buffer of the chunk tree. Since we are mounting the filesystem 7238 * and at this point there can't be any concurrent task modifying the 7239 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7240 */ 7241 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7242 path->skip_locking = 1; 7243 7244 /* 7245 * Read all device items, and then all the chunk items. All 7246 * device items are found before any chunk item (their object id 7247 * is smaller than the lowest possible object id for a chunk 7248 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7249 */ 7250 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7251 key.offset = 0; 7252 key.type = 0; 7253 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 7254 struct extent_buffer *node = path->nodes[1]; 7255 7256 leaf = path->nodes[0]; 7257 slot = path->slots[0]; 7258 7259 if (node) { 7260 if (last_ra_node != node->start) { 7261 readahead_tree_node_children(node); 7262 last_ra_node = node->start; 7263 } 7264 } 7265 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7266 struct btrfs_dev_item *dev_item; 7267 dev_item = btrfs_item_ptr(leaf, slot, 7268 struct btrfs_dev_item); 7269 ret = read_one_dev(leaf, dev_item); 7270 if (ret) 7271 goto error; 7272 total_dev++; 7273 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7274 struct btrfs_chunk *chunk; 7275 7276 /* 7277 * We are only called at mount time, so no need to take 7278 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7279 * we always lock first fs_info->chunk_mutex before 7280 * acquiring any locks on the chunk tree. This is a 7281 * requirement for chunk allocation, see the comment on 7282 * top of btrfs_chunk_alloc() for details. 7283 */ 7284 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7285 ret = read_one_chunk(&found_key, leaf, chunk); 7286 if (ret) 7287 goto error; 7288 } 7289 } 7290 /* Catch error found during iteration */ 7291 if (iter_ret < 0) { 7292 ret = iter_ret; 7293 goto error; 7294 } 7295 7296 /* 7297 * After loading chunk tree, we've got all device information, 7298 * do another round of validation checks. 7299 */ 7300 if (total_dev != fs_info->fs_devices->total_devices) { 7301 btrfs_warn(fs_info, 7302 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", 7303 btrfs_super_num_devices(fs_info->super_copy), 7304 total_dev); 7305 fs_info->fs_devices->total_devices = total_dev; 7306 btrfs_set_super_num_devices(fs_info->super_copy, total_dev); 7307 } 7308 if (btrfs_super_total_bytes(fs_info->super_copy) < 7309 fs_info->fs_devices->total_rw_bytes) { 7310 btrfs_err(fs_info, 7311 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7312 btrfs_super_total_bytes(fs_info->super_copy), 7313 fs_info->fs_devices->total_rw_bytes); 7314 ret = -EINVAL; 7315 goto error; 7316 } 7317 ret = 0; 7318 error: 7319 mutex_unlock(&uuid_mutex); 7320 7321 btrfs_free_path(path); 7322 return ret; 7323 } 7324 7325 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7326 { 7327 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7328 struct btrfs_device *device; 7329 int ret = 0; 7330 7331 fs_devices->fs_info = fs_info; 7332 7333 mutex_lock(&fs_devices->device_list_mutex); 7334 list_for_each_entry(device, &fs_devices->devices, dev_list) 7335 device->fs_info = fs_info; 7336 7337 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7338 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7339 device->fs_info = fs_info; 7340 ret = btrfs_get_dev_zone_info(device, false); 7341 if (ret) 7342 break; 7343 } 7344 7345 seed_devs->fs_info = fs_info; 7346 } 7347 mutex_unlock(&fs_devices->device_list_mutex); 7348 7349 return ret; 7350 } 7351 7352 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7353 const struct btrfs_dev_stats_item *ptr, 7354 int index) 7355 { 7356 u64 val; 7357 7358 read_extent_buffer(eb, &val, 7359 offsetof(struct btrfs_dev_stats_item, values) + 7360 ((unsigned long)ptr) + (index * sizeof(u64)), 7361 sizeof(val)); 7362 return val; 7363 } 7364 7365 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7366 struct btrfs_dev_stats_item *ptr, 7367 int index, u64 val) 7368 { 7369 write_extent_buffer(eb, &val, 7370 offsetof(struct btrfs_dev_stats_item, values) + 7371 ((unsigned long)ptr) + (index * sizeof(u64)), 7372 sizeof(val)); 7373 } 7374 7375 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7376 struct btrfs_path *path) 7377 { 7378 struct btrfs_dev_stats_item *ptr; 7379 struct extent_buffer *eb; 7380 struct btrfs_key key; 7381 int item_size; 7382 int i, ret, slot; 7383 7384 if (!device->fs_info->dev_root) 7385 return 0; 7386 7387 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7388 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7389 key.offset = device->devid; 7390 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7391 if (ret) { 7392 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7393 btrfs_dev_stat_set(device, i, 0); 7394 device->dev_stats_valid = 1; 7395 btrfs_release_path(path); 7396 return ret < 0 ? ret : 0; 7397 } 7398 slot = path->slots[0]; 7399 eb = path->nodes[0]; 7400 item_size = btrfs_item_size(eb, slot); 7401 7402 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7403 7404 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7405 if (item_size >= (1 + i) * sizeof(__le64)) 7406 btrfs_dev_stat_set(device, i, 7407 btrfs_dev_stats_value(eb, ptr, i)); 7408 else 7409 btrfs_dev_stat_set(device, i, 0); 7410 } 7411 7412 device->dev_stats_valid = 1; 7413 btrfs_dev_stat_print_on_load(device); 7414 btrfs_release_path(path); 7415 7416 return 0; 7417 } 7418 7419 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7420 { 7421 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7422 struct btrfs_device *device; 7423 struct btrfs_path *path = NULL; 7424 int ret = 0; 7425 7426 path = btrfs_alloc_path(); 7427 if (!path) 7428 return -ENOMEM; 7429 7430 mutex_lock(&fs_devices->device_list_mutex); 7431 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7432 ret = btrfs_device_init_dev_stats(device, path); 7433 if (ret) 7434 goto out; 7435 } 7436 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7437 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7438 ret = btrfs_device_init_dev_stats(device, path); 7439 if (ret) 7440 goto out; 7441 } 7442 } 7443 out: 7444 mutex_unlock(&fs_devices->device_list_mutex); 7445 7446 btrfs_free_path(path); 7447 return ret; 7448 } 7449 7450 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7451 struct btrfs_device *device) 7452 { 7453 struct btrfs_fs_info *fs_info = trans->fs_info; 7454 struct btrfs_root *dev_root = fs_info->dev_root; 7455 struct btrfs_path *path; 7456 struct btrfs_key key; 7457 struct extent_buffer *eb; 7458 struct btrfs_dev_stats_item *ptr; 7459 int ret; 7460 int i; 7461 7462 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7463 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7464 key.offset = device->devid; 7465 7466 path = btrfs_alloc_path(); 7467 if (!path) 7468 return -ENOMEM; 7469 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7470 if (ret < 0) { 7471 btrfs_warn_in_rcu(fs_info, 7472 "error %d while searching for dev_stats item for device %s", 7473 ret, btrfs_dev_name(device)); 7474 goto out; 7475 } 7476 7477 if (ret == 0 && 7478 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7479 /* need to delete old one and insert a new one */ 7480 ret = btrfs_del_item(trans, dev_root, path); 7481 if (ret != 0) { 7482 btrfs_warn_in_rcu(fs_info, 7483 "delete too small dev_stats item for device %s failed %d", 7484 btrfs_dev_name(device), ret); 7485 goto out; 7486 } 7487 ret = 1; 7488 } 7489 7490 if (ret == 1) { 7491 /* need to insert a new item */ 7492 btrfs_release_path(path); 7493 ret = btrfs_insert_empty_item(trans, dev_root, path, 7494 &key, sizeof(*ptr)); 7495 if (ret < 0) { 7496 btrfs_warn_in_rcu(fs_info, 7497 "insert dev_stats item for device %s failed %d", 7498 btrfs_dev_name(device), ret); 7499 goto out; 7500 } 7501 } 7502 7503 eb = path->nodes[0]; 7504 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7505 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7506 btrfs_set_dev_stats_value(eb, ptr, i, 7507 btrfs_dev_stat_read(device, i)); 7508 btrfs_mark_buffer_dirty(eb); 7509 7510 out: 7511 btrfs_free_path(path); 7512 return ret; 7513 } 7514 7515 /* 7516 * called from commit_transaction. Writes all changed device stats to disk. 7517 */ 7518 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7519 { 7520 struct btrfs_fs_info *fs_info = trans->fs_info; 7521 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7522 struct btrfs_device *device; 7523 int stats_cnt; 7524 int ret = 0; 7525 7526 mutex_lock(&fs_devices->device_list_mutex); 7527 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7528 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7529 if (!device->dev_stats_valid || stats_cnt == 0) 7530 continue; 7531 7532 7533 /* 7534 * There is a LOAD-LOAD control dependency between the value of 7535 * dev_stats_ccnt and updating the on-disk values which requires 7536 * reading the in-memory counters. Such control dependencies 7537 * require explicit read memory barriers. 7538 * 7539 * This memory barriers pairs with smp_mb__before_atomic in 7540 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7541 * barrier implied by atomic_xchg in 7542 * btrfs_dev_stats_read_and_reset 7543 */ 7544 smp_rmb(); 7545 7546 ret = update_dev_stat_item(trans, device); 7547 if (!ret) 7548 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7549 } 7550 mutex_unlock(&fs_devices->device_list_mutex); 7551 7552 return ret; 7553 } 7554 7555 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7556 { 7557 btrfs_dev_stat_inc(dev, index); 7558 7559 if (!dev->dev_stats_valid) 7560 return; 7561 btrfs_err_rl_in_rcu(dev->fs_info, 7562 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7563 btrfs_dev_name(dev), 7564 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7565 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7566 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7567 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7568 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7569 } 7570 7571 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7572 { 7573 int i; 7574 7575 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7576 if (btrfs_dev_stat_read(dev, i) != 0) 7577 break; 7578 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7579 return; /* all values == 0, suppress message */ 7580 7581 btrfs_info_in_rcu(dev->fs_info, 7582 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7583 btrfs_dev_name(dev), 7584 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7585 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7586 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7587 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7588 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7589 } 7590 7591 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7592 struct btrfs_ioctl_get_dev_stats *stats) 7593 { 7594 BTRFS_DEV_LOOKUP_ARGS(args); 7595 struct btrfs_device *dev; 7596 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7597 int i; 7598 7599 mutex_lock(&fs_devices->device_list_mutex); 7600 args.devid = stats->devid; 7601 dev = btrfs_find_device(fs_info->fs_devices, &args); 7602 mutex_unlock(&fs_devices->device_list_mutex); 7603 7604 if (!dev) { 7605 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7606 return -ENODEV; 7607 } else if (!dev->dev_stats_valid) { 7608 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7609 return -ENODEV; 7610 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7611 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7612 if (stats->nr_items > i) 7613 stats->values[i] = 7614 btrfs_dev_stat_read_and_reset(dev, i); 7615 else 7616 btrfs_dev_stat_set(dev, i, 0); 7617 } 7618 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7619 current->comm, task_pid_nr(current)); 7620 } else { 7621 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7622 if (stats->nr_items > i) 7623 stats->values[i] = btrfs_dev_stat_read(dev, i); 7624 } 7625 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7626 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7627 return 0; 7628 } 7629 7630 /* 7631 * Update the size and bytes used for each device where it changed. This is 7632 * delayed since we would otherwise get errors while writing out the 7633 * superblocks. 7634 * 7635 * Must be invoked during transaction commit. 7636 */ 7637 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7638 { 7639 struct btrfs_device *curr, *next; 7640 7641 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7642 7643 if (list_empty(&trans->dev_update_list)) 7644 return; 7645 7646 /* 7647 * We don't need the device_list_mutex here. This list is owned by the 7648 * transaction and the transaction must complete before the device is 7649 * released. 7650 */ 7651 mutex_lock(&trans->fs_info->chunk_mutex); 7652 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7653 post_commit_list) { 7654 list_del_init(&curr->post_commit_list); 7655 curr->commit_total_bytes = curr->disk_total_bytes; 7656 curr->commit_bytes_used = curr->bytes_used; 7657 } 7658 mutex_unlock(&trans->fs_info->chunk_mutex); 7659 } 7660 7661 /* 7662 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7663 */ 7664 int btrfs_bg_type_to_factor(u64 flags) 7665 { 7666 const int index = btrfs_bg_flags_to_raid_index(flags); 7667 7668 return btrfs_raid_array[index].ncopies; 7669 } 7670 7671 7672 7673 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7674 u64 chunk_offset, u64 devid, 7675 u64 physical_offset, u64 physical_len) 7676 { 7677 struct btrfs_dev_lookup_args args = { .devid = devid }; 7678 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7679 struct extent_map *em; 7680 struct map_lookup *map; 7681 struct btrfs_device *dev; 7682 u64 stripe_len; 7683 bool found = false; 7684 int ret = 0; 7685 int i; 7686 7687 read_lock(&em_tree->lock); 7688 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7689 read_unlock(&em_tree->lock); 7690 7691 if (!em) { 7692 btrfs_err(fs_info, 7693 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 7694 physical_offset, devid); 7695 ret = -EUCLEAN; 7696 goto out; 7697 } 7698 7699 map = em->map_lookup; 7700 stripe_len = btrfs_calc_stripe_length(em); 7701 if (physical_len != stripe_len) { 7702 btrfs_err(fs_info, 7703 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 7704 physical_offset, devid, em->start, physical_len, 7705 stripe_len); 7706 ret = -EUCLEAN; 7707 goto out; 7708 } 7709 7710 /* 7711 * Very old mkfs.btrfs (before v4.1) will not respect the reserved 7712 * space. Although kernel can handle it without problem, better to warn 7713 * the users. 7714 */ 7715 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) 7716 btrfs_warn(fs_info, 7717 "devid %llu physical %llu len %llu inside the reserved space", 7718 devid, physical_offset, physical_len); 7719 7720 for (i = 0; i < map->num_stripes; i++) { 7721 if (map->stripes[i].dev->devid == devid && 7722 map->stripes[i].physical == physical_offset) { 7723 found = true; 7724 if (map->verified_stripes >= map->num_stripes) { 7725 btrfs_err(fs_info, 7726 "too many dev extents for chunk %llu found", 7727 em->start); 7728 ret = -EUCLEAN; 7729 goto out; 7730 } 7731 map->verified_stripes++; 7732 break; 7733 } 7734 } 7735 if (!found) { 7736 btrfs_err(fs_info, 7737 "dev extent physical offset %llu devid %llu has no corresponding chunk", 7738 physical_offset, devid); 7739 ret = -EUCLEAN; 7740 } 7741 7742 /* Make sure no dev extent is beyond device boundary */ 7743 dev = btrfs_find_device(fs_info->fs_devices, &args); 7744 if (!dev) { 7745 btrfs_err(fs_info, "failed to find devid %llu", devid); 7746 ret = -EUCLEAN; 7747 goto out; 7748 } 7749 7750 if (physical_offset + physical_len > dev->disk_total_bytes) { 7751 btrfs_err(fs_info, 7752 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 7753 devid, physical_offset, physical_len, 7754 dev->disk_total_bytes); 7755 ret = -EUCLEAN; 7756 goto out; 7757 } 7758 7759 if (dev->zone_info) { 7760 u64 zone_size = dev->zone_info->zone_size; 7761 7762 if (!IS_ALIGNED(physical_offset, zone_size) || 7763 !IS_ALIGNED(physical_len, zone_size)) { 7764 btrfs_err(fs_info, 7765 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 7766 devid, physical_offset, physical_len); 7767 ret = -EUCLEAN; 7768 goto out; 7769 } 7770 } 7771 7772 out: 7773 free_extent_map(em); 7774 return ret; 7775 } 7776 7777 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 7778 { 7779 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7780 struct extent_map *em; 7781 struct rb_node *node; 7782 int ret = 0; 7783 7784 read_lock(&em_tree->lock); 7785 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 7786 em = rb_entry(node, struct extent_map, rb_node); 7787 if (em->map_lookup->num_stripes != 7788 em->map_lookup->verified_stripes) { 7789 btrfs_err(fs_info, 7790 "chunk %llu has missing dev extent, have %d expect %d", 7791 em->start, em->map_lookup->verified_stripes, 7792 em->map_lookup->num_stripes); 7793 ret = -EUCLEAN; 7794 goto out; 7795 } 7796 } 7797 out: 7798 read_unlock(&em_tree->lock); 7799 return ret; 7800 } 7801 7802 /* 7803 * Ensure that all dev extents are mapped to correct chunk, otherwise 7804 * later chunk allocation/free would cause unexpected behavior. 7805 * 7806 * NOTE: This will iterate through the whole device tree, which should be of 7807 * the same size level as the chunk tree. This slightly increases mount time. 7808 */ 7809 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 7810 { 7811 struct btrfs_path *path; 7812 struct btrfs_root *root = fs_info->dev_root; 7813 struct btrfs_key key; 7814 u64 prev_devid = 0; 7815 u64 prev_dev_ext_end = 0; 7816 int ret = 0; 7817 7818 /* 7819 * We don't have a dev_root because we mounted with ignorebadroots and 7820 * failed to load the root, so we want to skip the verification in this 7821 * case for sure. 7822 * 7823 * However if the dev root is fine, but the tree itself is corrupted 7824 * we'd still fail to mount. This verification is only to make sure 7825 * writes can happen safely, so instead just bypass this check 7826 * completely in the case of IGNOREBADROOTS. 7827 */ 7828 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 7829 return 0; 7830 7831 key.objectid = 1; 7832 key.type = BTRFS_DEV_EXTENT_KEY; 7833 key.offset = 0; 7834 7835 path = btrfs_alloc_path(); 7836 if (!path) 7837 return -ENOMEM; 7838 7839 path->reada = READA_FORWARD; 7840 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7841 if (ret < 0) 7842 goto out; 7843 7844 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 7845 ret = btrfs_next_leaf(root, path); 7846 if (ret < 0) 7847 goto out; 7848 /* No dev extents at all? Not good */ 7849 if (ret > 0) { 7850 ret = -EUCLEAN; 7851 goto out; 7852 } 7853 } 7854 while (1) { 7855 struct extent_buffer *leaf = path->nodes[0]; 7856 struct btrfs_dev_extent *dext; 7857 int slot = path->slots[0]; 7858 u64 chunk_offset; 7859 u64 physical_offset; 7860 u64 physical_len; 7861 u64 devid; 7862 7863 btrfs_item_key_to_cpu(leaf, &key, slot); 7864 if (key.type != BTRFS_DEV_EXTENT_KEY) 7865 break; 7866 devid = key.objectid; 7867 physical_offset = key.offset; 7868 7869 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 7870 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 7871 physical_len = btrfs_dev_extent_length(leaf, dext); 7872 7873 /* Check if this dev extent overlaps with the previous one */ 7874 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 7875 btrfs_err(fs_info, 7876 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 7877 devid, physical_offset, prev_dev_ext_end); 7878 ret = -EUCLEAN; 7879 goto out; 7880 } 7881 7882 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 7883 physical_offset, physical_len); 7884 if (ret < 0) 7885 goto out; 7886 prev_devid = devid; 7887 prev_dev_ext_end = physical_offset + physical_len; 7888 7889 ret = btrfs_next_item(root, path); 7890 if (ret < 0) 7891 goto out; 7892 if (ret > 0) { 7893 ret = 0; 7894 break; 7895 } 7896 } 7897 7898 /* Ensure all chunks have corresponding dev extents */ 7899 ret = verify_chunk_dev_extent_mapping(fs_info); 7900 out: 7901 btrfs_free_path(path); 7902 return ret; 7903 } 7904 7905 /* 7906 * Check whether the given block group or device is pinned by any inode being 7907 * used as a swapfile. 7908 */ 7909 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 7910 { 7911 struct btrfs_swapfile_pin *sp; 7912 struct rb_node *node; 7913 7914 spin_lock(&fs_info->swapfile_pins_lock); 7915 node = fs_info->swapfile_pins.rb_node; 7916 while (node) { 7917 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 7918 if (ptr < sp->ptr) 7919 node = node->rb_left; 7920 else if (ptr > sp->ptr) 7921 node = node->rb_right; 7922 else 7923 break; 7924 } 7925 spin_unlock(&fs_info->swapfile_pins_lock); 7926 return node != NULL; 7927 } 7928 7929 static int relocating_repair_kthread(void *data) 7930 { 7931 struct btrfs_block_group *cache = data; 7932 struct btrfs_fs_info *fs_info = cache->fs_info; 7933 u64 target; 7934 int ret = 0; 7935 7936 target = cache->start; 7937 btrfs_put_block_group(cache); 7938 7939 sb_start_write(fs_info->sb); 7940 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 7941 btrfs_info(fs_info, 7942 "zoned: skip relocating block group %llu to repair: EBUSY", 7943 target); 7944 sb_end_write(fs_info->sb); 7945 return -EBUSY; 7946 } 7947 7948 mutex_lock(&fs_info->reclaim_bgs_lock); 7949 7950 /* Ensure block group still exists */ 7951 cache = btrfs_lookup_block_group(fs_info, target); 7952 if (!cache) 7953 goto out; 7954 7955 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) 7956 goto out; 7957 7958 ret = btrfs_may_alloc_data_chunk(fs_info, target); 7959 if (ret < 0) 7960 goto out; 7961 7962 btrfs_info(fs_info, 7963 "zoned: relocating block group %llu to repair IO failure", 7964 target); 7965 ret = btrfs_relocate_chunk(fs_info, target); 7966 7967 out: 7968 if (cache) 7969 btrfs_put_block_group(cache); 7970 mutex_unlock(&fs_info->reclaim_bgs_lock); 7971 btrfs_exclop_finish(fs_info); 7972 sb_end_write(fs_info->sb); 7973 7974 return ret; 7975 } 7976 7977 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 7978 { 7979 struct btrfs_block_group *cache; 7980 7981 if (!btrfs_is_zoned(fs_info)) 7982 return false; 7983 7984 /* Do not attempt to repair in degraded state */ 7985 if (btrfs_test_opt(fs_info, DEGRADED)) 7986 return true; 7987 7988 cache = btrfs_lookup_block_group(fs_info, logical); 7989 if (!cache) 7990 return true; 7991 7992 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { 7993 btrfs_put_block_group(cache); 7994 return true; 7995 } 7996 7997 kthread_run(relocating_repair_kthread, cache, 7998 "btrfs-relocating-repair"); 7999 8000 return true; 8001 } 8002 8003 static void map_raid56_repair_block(struct btrfs_io_context *bioc, 8004 struct btrfs_io_stripe *smap, 8005 u64 logical) 8006 { 8007 int data_stripes = nr_bioc_data_stripes(bioc); 8008 int i; 8009 8010 for (i = 0; i < data_stripes; i++) { 8011 u64 stripe_start = bioc->full_stripe_logical + 8012 (i << BTRFS_STRIPE_LEN_SHIFT); 8013 8014 if (logical >= stripe_start && 8015 logical < stripe_start + BTRFS_STRIPE_LEN) 8016 break; 8017 } 8018 ASSERT(i < data_stripes); 8019 smap->dev = bioc->stripes[i].dev; 8020 smap->physical = bioc->stripes[i].physical + 8021 ((logical - bioc->full_stripe_logical) & 8022 BTRFS_STRIPE_LEN_MASK); 8023 } 8024 8025 /* 8026 * Map a repair write into a single device. 8027 * 8028 * A repair write is triggered by read time repair or scrub, which would only 8029 * update the contents of a single device. 8030 * Not update any other mirrors nor go through RMW path. 8031 * 8032 * Callers should ensure: 8033 * 8034 * - Call btrfs_bio_counter_inc_blocked() first 8035 * - The range does not cross stripe boundary 8036 * - Has a valid @mirror_num passed in. 8037 */ 8038 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info, 8039 struct btrfs_io_stripe *smap, u64 logical, 8040 u32 length, int mirror_num) 8041 { 8042 struct btrfs_io_context *bioc = NULL; 8043 u64 map_length = length; 8044 int mirror_ret = mirror_num; 8045 int ret; 8046 8047 ASSERT(mirror_num > 0); 8048 8049 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length, 8050 &bioc, smap, &mirror_ret, true); 8051 if (ret < 0) 8052 return ret; 8053 8054 /* The map range should not cross stripe boundary. */ 8055 ASSERT(map_length >= length); 8056 8057 /* Already mapped to single stripe. */ 8058 if (!bioc) 8059 goto out; 8060 8061 /* Map the RAID56 multi-stripe writes to a single one. */ 8062 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 8063 map_raid56_repair_block(bioc, smap, logical); 8064 goto out; 8065 } 8066 8067 ASSERT(mirror_num <= bioc->num_stripes); 8068 smap->dev = bioc->stripes[mirror_num - 1].dev; 8069 smap->physical = bioc->stripes[mirror_num - 1].physical; 8070 out: 8071 btrfs_put_bioc(bioc); 8072 ASSERT(smap->dev); 8073 return 0; 8074 } 8075