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 noinline struct btrfs_fs_devices *find_fsid( 431 const u8 *fsid, const u8 *metadata_fsid) 432 { 433 struct btrfs_fs_devices *fs_devices; 434 435 ASSERT(fsid); 436 437 /* Handle non-split brain cases */ 438 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 439 if (metadata_fsid) { 440 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 441 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 442 BTRFS_FSID_SIZE) == 0) 443 return fs_devices; 444 } else { 445 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 446 return fs_devices; 447 } 448 } 449 return NULL; 450 } 451 452 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 453 struct btrfs_super_block *disk_super) 454 { 455 456 struct btrfs_fs_devices *fs_devices; 457 458 /* 459 * Handle scanned device having completed its fsid change but 460 * belonging to a fs_devices that was created by first scanning 461 * a device which didn't have its fsid/metadata_uuid changed 462 * at all and the CHANGING_FSID_V2 flag set. 463 */ 464 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 465 if (fs_devices->fsid_change && 466 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 467 BTRFS_FSID_SIZE) == 0 && 468 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 469 BTRFS_FSID_SIZE) == 0) { 470 return fs_devices; 471 } 472 } 473 /* 474 * Handle scanned device having completed its fsid change but 475 * belonging to a fs_devices that was created by a device that 476 * has an outdated pair of fsid/metadata_uuid and 477 * CHANGING_FSID_V2 flag set. 478 */ 479 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 480 if (fs_devices->fsid_change && 481 memcmp(fs_devices->metadata_uuid, 482 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 483 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 484 BTRFS_FSID_SIZE) == 0) { 485 return fs_devices; 486 } 487 } 488 489 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 490 } 491 492 493 static int 494 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 495 int flush, struct block_device **bdev, 496 struct btrfs_super_block **disk_super) 497 { 498 int ret; 499 500 *bdev = blkdev_get_by_path(device_path, flags, holder); 501 502 if (IS_ERR(*bdev)) { 503 ret = PTR_ERR(*bdev); 504 goto error; 505 } 506 507 if (flush) 508 sync_blockdev(*bdev); 509 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 510 if (ret) { 511 blkdev_put(*bdev, flags); 512 goto error; 513 } 514 invalidate_bdev(*bdev); 515 *disk_super = btrfs_read_dev_super(*bdev); 516 if (IS_ERR(*disk_super)) { 517 ret = PTR_ERR(*disk_super); 518 blkdev_put(*bdev, flags); 519 goto error; 520 } 521 522 return 0; 523 524 error: 525 *bdev = NULL; 526 return ret; 527 } 528 529 /* 530 * Search and remove all stale devices (which are not mounted). When both 531 * inputs are NULL, it will search and release all stale devices. 532 * 533 * @devt: Optional. When provided will it release all unmounted devices 534 * matching this devt only. 535 * @skip_device: Optional. Will skip this device when searching for the stale 536 * devices. 537 * 538 * Return: 0 for success or if @devt is 0. 539 * -EBUSY if @devt is a mounted device. 540 * -ENOENT if @devt does not match any device in the list. 541 */ 542 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) 543 { 544 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 545 struct btrfs_device *device, *tmp_device; 546 int ret = 0; 547 548 lockdep_assert_held(&uuid_mutex); 549 550 if (devt) 551 ret = -ENOENT; 552 553 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 554 555 mutex_lock(&fs_devices->device_list_mutex); 556 list_for_each_entry_safe(device, tmp_device, 557 &fs_devices->devices, dev_list) { 558 if (skip_device && skip_device == device) 559 continue; 560 if (devt && devt != device->devt) 561 continue; 562 if (fs_devices->opened) { 563 /* for an already deleted device return 0 */ 564 if (devt && ret != 0) 565 ret = -EBUSY; 566 break; 567 } 568 569 /* delete the stale device */ 570 fs_devices->num_devices--; 571 list_del(&device->dev_list); 572 btrfs_free_device(device); 573 574 ret = 0; 575 } 576 mutex_unlock(&fs_devices->device_list_mutex); 577 578 if (fs_devices->num_devices == 0) { 579 btrfs_sysfs_remove_fsid(fs_devices); 580 list_del(&fs_devices->fs_list); 581 free_fs_devices(fs_devices); 582 } 583 } 584 585 return ret; 586 } 587 588 /* 589 * This is only used on mount, and we are protected from competing things 590 * messing with our fs_devices by the uuid_mutex, thus we do not need the 591 * fs_devices->device_list_mutex here. 592 */ 593 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 594 struct btrfs_device *device, fmode_t flags, 595 void *holder) 596 { 597 struct block_device *bdev; 598 struct btrfs_super_block *disk_super; 599 u64 devid; 600 int ret; 601 602 if (device->bdev) 603 return -EINVAL; 604 if (!device->name) 605 return -EINVAL; 606 607 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 608 &bdev, &disk_super); 609 if (ret) 610 return ret; 611 612 devid = btrfs_stack_device_id(&disk_super->dev_item); 613 if (devid != device->devid) 614 goto error_free_page; 615 616 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 617 goto error_free_page; 618 619 device->generation = btrfs_super_generation(disk_super); 620 621 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 622 if (btrfs_super_incompat_flags(disk_super) & 623 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 624 pr_err( 625 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 626 goto error_free_page; 627 } 628 629 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 630 fs_devices->seeding = true; 631 } else { 632 if (bdev_read_only(bdev)) 633 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 634 else 635 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 636 } 637 638 if (!bdev_nonrot(bdev)) 639 fs_devices->rotating = true; 640 641 if (bdev_max_discard_sectors(bdev)) 642 fs_devices->discardable = true; 643 644 device->bdev = bdev; 645 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 646 device->mode = flags; 647 648 fs_devices->open_devices++; 649 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 650 device->devid != BTRFS_DEV_REPLACE_DEVID) { 651 fs_devices->rw_devices++; 652 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 653 } 654 btrfs_release_disk_super(disk_super); 655 656 return 0; 657 658 error_free_page: 659 btrfs_release_disk_super(disk_super); 660 blkdev_put(bdev, flags); 661 662 return -EINVAL; 663 } 664 665 /* 666 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 667 * being created with a disk that has already completed its fsid change. Such 668 * disk can belong to an fs which has its FSID changed or to one which doesn't. 669 * Handle both cases here. 670 */ 671 static struct btrfs_fs_devices *find_fsid_inprogress( 672 struct btrfs_super_block *disk_super) 673 { 674 struct btrfs_fs_devices *fs_devices; 675 676 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 677 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 678 BTRFS_FSID_SIZE) != 0 && 679 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 680 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 681 return fs_devices; 682 } 683 } 684 685 return find_fsid(disk_super->fsid, NULL); 686 } 687 688 689 static struct btrfs_fs_devices *find_fsid_changed( 690 struct btrfs_super_block *disk_super) 691 { 692 struct btrfs_fs_devices *fs_devices; 693 694 /* 695 * Handles the case where scanned device is part of an fs that had 696 * multiple successful changes of FSID but currently device didn't 697 * observe it. Meaning our fsid will be different than theirs. We need 698 * to handle two subcases : 699 * 1 - The fs still continues to have different METADATA/FSID uuids. 700 * 2 - The fs is switched back to its original FSID (METADATA/FSID 701 * are equal). 702 */ 703 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 704 /* Changed UUIDs */ 705 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 706 BTRFS_FSID_SIZE) != 0 && 707 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 708 BTRFS_FSID_SIZE) == 0 && 709 memcmp(fs_devices->fsid, disk_super->fsid, 710 BTRFS_FSID_SIZE) != 0) 711 return fs_devices; 712 713 /* Unchanged UUIDs */ 714 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 715 BTRFS_FSID_SIZE) == 0 && 716 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 717 BTRFS_FSID_SIZE) == 0) 718 return fs_devices; 719 } 720 721 return NULL; 722 } 723 724 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 725 struct btrfs_super_block *disk_super) 726 { 727 struct btrfs_fs_devices *fs_devices; 728 729 /* 730 * Handle the case where the scanned device is part of an fs whose last 731 * metadata UUID change reverted it to the original FSID. At the same 732 * time fs_devices was first created by another constituent device 733 * which didn't fully observe the operation. This results in an 734 * btrfs_fs_devices created with metadata/fsid different AND 735 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 736 * fs_devices equal to the FSID of the disk. 737 */ 738 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 739 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 740 BTRFS_FSID_SIZE) != 0 && 741 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 742 BTRFS_FSID_SIZE) == 0 && 743 fs_devices->fsid_change) 744 return fs_devices; 745 } 746 747 return NULL; 748 } 749 /* 750 * Add new device to list of registered devices 751 * 752 * Returns: 753 * device pointer which was just added or updated when successful 754 * error pointer when failed 755 */ 756 static noinline struct btrfs_device *device_list_add(const char *path, 757 struct btrfs_super_block *disk_super, 758 bool *new_device_added) 759 { 760 struct btrfs_device *device; 761 struct btrfs_fs_devices *fs_devices = NULL; 762 struct rcu_string *name; 763 u64 found_transid = btrfs_super_generation(disk_super); 764 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 765 dev_t path_devt; 766 int error; 767 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 768 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 769 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 770 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 771 772 error = lookup_bdev(path, &path_devt); 773 if (error) { 774 btrfs_err(NULL, "failed to lookup block device for path %s: %d", 775 path, error); 776 return ERR_PTR(error); 777 } 778 779 if (fsid_change_in_progress) { 780 if (!has_metadata_uuid) 781 fs_devices = find_fsid_inprogress(disk_super); 782 else 783 fs_devices = find_fsid_changed(disk_super); 784 } else if (has_metadata_uuid) { 785 fs_devices = find_fsid_with_metadata_uuid(disk_super); 786 } else { 787 fs_devices = find_fsid_reverted_metadata(disk_super); 788 if (!fs_devices) 789 fs_devices = find_fsid(disk_super->fsid, NULL); 790 } 791 792 793 if (!fs_devices) { 794 if (has_metadata_uuid) 795 fs_devices = alloc_fs_devices(disk_super->fsid, 796 disk_super->metadata_uuid); 797 else 798 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 799 800 if (IS_ERR(fs_devices)) 801 return ERR_CAST(fs_devices); 802 803 fs_devices->fsid_change = fsid_change_in_progress; 804 805 mutex_lock(&fs_devices->device_list_mutex); 806 list_add(&fs_devices->fs_list, &fs_uuids); 807 808 device = NULL; 809 } else { 810 struct btrfs_dev_lookup_args args = { 811 .devid = devid, 812 .uuid = disk_super->dev_item.uuid, 813 }; 814 815 mutex_lock(&fs_devices->device_list_mutex); 816 device = btrfs_find_device(fs_devices, &args); 817 818 /* 819 * If this disk has been pulled into an fs devices created by 820 * a device which had the CHANGING_FSID_V2 flag then replace the 821 * metadata_uuid/fsid values of the fs_devices. 822 */ 823 if (fs_devices->fsid_change && 824 found_transid > fs_devices->latest_generation) { 825 memcpy(fs_devices->fsid, disk_super->fsid, 826 BTRFS_FSID_SIZE); 827 828 if (has_metadata_uuid) 829 memcpy(fs_devices->metadata_uuid, 830 disk_super->metadata_uuid, 831 BTRFS_FSID_SIZE); 832 else 833 memcpy(fs_devices->metadata_uuid, 834 disk_super->fsid, BTRFS_FSID_SIZE); 835 836 fs_devices->fsid_change = false; 837 } 838 } 839 840 if (!device) { 841 unsigned int nofs_flag; 842 843 if (fs_devices->opened) { 844 btrfs_err(NULL, 845 "device %s belongs to fsid %pU, and the fs is already mounted", 846 path, fs_devices->fsid); 847 mutex_unlock(&fs_devices->device_list_mutex); 848 return ERR_PTR(-EBUSY); 849 } 850 851 nofs_flag = memalloc_nofs_save(); 852 device = btrfs_alloc_device(NULL, &devid, 853 disk_super->dev_item.uuid, path); 854 memalloc_nofs_restore(nofs_flag); 855 if (IS_ERR(device)) { 856 mutex_unlock(&fs_devices->device_list_mutex); 857 /* we can safely leave the fs_devices entry around */ 858 return device; 859 } 860 861 device->devt = path_devt; 862 863 list_add_rcu(&device->dev_list, &fs_devices->devices); 864 fs_devices->num_devices++; 865 866 device->fs_devices = fs_devices; 867 *new_device_added = true; 868 869 if (disk_super->label[0]) 870 pr_info( 871 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 872 disk_super->label, devid, found_transid, path, 873 current->comm, task_pid_nr(current)); 874 else 875 pr_info( 876 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 877 disk_super->fsid, devid, found_transid, path, 878 current->comm, task_pid_nr(current)); 879 880 } else if (!device->name || strcmp(device->name->str, path)) { 881 /* 882 * When FS is already mounted. 883 * 1. If you are here and if the device->name is NULL that 884 * means this device was missing at time of FS mount. 885 * 2. If you are here and if the device->name is different 886 * from 'path' that means either 887 * a. The same device disappeared and reappeared with 888 * different name. or 889 * b. The missing-disk-which-was-replaced, has 890 * reappeared now. 891 * 892 * We must allow 1 and 2a above. But 2b would be a spurious 893 * and unintentional. 894 * 895 * Further in case of 1 and 2a above, the disk at 'path' 896 * would have missed some transaction when it was away and 897 * in case of 2a the stale bdev has to be updated as well. 898 * 2b must not be allowed at all time. 899 */ 900 901 /* 902 * For now, we do allow update to btrfs_fs_device through the 903 * btrfs dev scan cli after FS has been mounted. We're still 904 * tracking a problem where systems fail mount by subvolume id 905 * when we reject replacement on a mounted FS. 906 */ 907 if (!fs_devices->opened && found_transid < device->generation) { 908 /* 909 * That is if the FS is _not_ mounted and if you 910 * are here, that means there is more than one 911 * disk with same uuid and devid.We keep the one 912 * with larger generation number or the last-in if 913 * generation are equal. 914 */ 915 mutex_unlock(&fs_devices->device_list_mutex); 916 btrfs_err(NULL, 917 "device %s already registered with a higher generation, found %llu expect %llu", 918 path, found_transid, device->generation); 919 return ERR_PTR(-EEXIST); 920 } 921 922 /* 923 * We are going to replace the device path for a given devid, 924 * make sure it's the same device if the device is mounted 925 * 926 * NOTE: the device->fs_info may not be reliable here so pass 927 * in a NULL to message helpers instead. This avoids a possible 928 * use-after-free when the fs_info and fs_info->sb are already 929 * torn down. 930 */ 931 if (device->bdev) { 932 if (device->devt != path_devt) { 933 mutex_unlock(&fs_devices->device_list_mutex); 934 btrfs_warn_in_rcu(NULL, 935 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 936 path, devid, found_transid, 937 current->comm, 938 task_pid_nr(current)); 939 return ERR_PTR(-EEXIST); 940 } 941 btrfs_info_in_rcu(NULL, 942 "devid %llu device path %s changed to %s scanned by %s (%d)", 943 devid, btrfs_dev_name(device), 944 path, current->comm, 945 task_pid_nr(current)); 946 } 947 948 name = rcu_string_strdup(path, GFP_NOFS); 949 if (!name) { 950 mutex_unlock(&fs_devices->device_list_mutex); 951 return ERR_PTR(-ENOMEM); 952 } 953 rcu_string_free(device->name); 954 rcu_assign_pointer(device->name, name); 955 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 956 fs_devices->missing_devices--; 957 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 958 } 959 device->devt = path_devt; 960 } 961 962 /* 963 * Unmount does not free the btrfs_device struct but would zero 964 * generation along with most of the other members. So just update 965 * it back. We need it to pick the disk with largest generation 966 * (as above). 967 */ 968 if (!fs_devices->opened) { 969 device->generation = found_transid; 970 fs_devices->latest_generation = max_t(u64, found_transid, 971 fs_devices->latest_generation); 972 } 973 974 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 975 976 mutex_unlock(&fs_devices->device_list_mutex); 977 return device; 978 } 979 980 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 981 { 982 struct btrfs_fs_devices *fs_devices; 983 struct btrfs_device *device; 984 struct btrfs_device *orig_dev; 985 int ret = 0; 986 987 lockdep_assert_held(&uuid_mutex); 988 989 fs_devices = alloc_fs_devices(orig->fsid, NULL); 990 if (IS_ERR(fs_devices)) 991 return fs_devices; 992 993 fs_devices->total_devices = orig->total_devices; 994 995 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 996 const char *dev_path = NULL; 997 998 /* 999 * This is ok to do without RCU read locked because we hold the 1000 * uuid mutex so nothing we touch in here is going to disappear. 1001 */ 1002 if (orig_dev->name) 1003 dev_path = orig_dev->name->str; 1004 1005 device = btrfs_alloc_device(NULL, &orig_dev->devid, 1006 orig_dev->uuid, dev_path); 1007 if (IS_ERR(device)) { 1008 ret = PTR_ERR(device); 1009 goto error; 1010 } 1011 1012 if (orig_dev->zone_info) { 1013 struct btrfs_zoned_device_info *zone_info; 1014 1015 zone_info = btrfs_clone_dev_zone_info(orig_dev); 1016 if (!zone_info) { 1017 btrfs_free_device(device); 1018 ret = -ENOMEM; 1019 goto error; 1020 } 1021 device->zone_info = zone_info; 1022 } 1023 1024 list_add(&device->dev_list, &fs_devices->devices); 1025 device->fs_devices = fs_devices; 1026 fs_devices->num_devices++; 1027 } 1028 return fs_devices; 1029 error: 1030 free_fs_devices(fs_devices); 1031 return ERR_PTR(ret); 1032 } 1033 1034 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1035 struct btrfs_device **latest_dev) 1036 { 1037 struct btrfs_device *device, *next; 1038 1039 /* This is the initialized path, it is safe to release the devices. */ 1040 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1041 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1042 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1043 &device->dev_state) && 1044 !test_bit(BTRFS_DEV_STATE_MISSING, 1045 &device->dev_state) && 1046 (!*latest_dev || 1047 device->generation > (*latest_dev)->generation)) { 1048 *latest_dev = device; 1049 } 1050 continue; 1051 } 1052 1053 /* 1054 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1055 * in btrfs_init_dev_replace() so just continue. 1056 */ 1057 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1058 continue; 1059 1060 if (device->bdev) { 1061 blkdev_put(device->bdev, device->mode); 1062 device->bdev = NULL; 1063 fs_devices->open_devices--; 1064 } 1065 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1066 list_del_init(&device->dev_alloc_list); 1067 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1068 fs_devices->rw_devices--; 1069 } 1070 list_del_init(&device->dev_list); 1071 fs_devices->num_devices--; 1072 btrfs_free_device(device); 1073 } 1074 1075 } 1076 1077 /* 1078 * After we have read the system tree and know devids belonging to this 1079 * filesystem, remove the device which does not belong there. 1080 */ 1081 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1082 { 1083 struct btrfs_device *latest_dev = NULL; 1084 struct btrfs_fs_devices *seed_dev; 1085 1086 mutex_lock(&uuid_mutex); 1087 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1088 1089 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1090 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1091 1092 fs_devices->latest_dev = latest_dev; 1093 1094 mutex_unlock(&uuid_mutex); 1095 } 1096 1097 static void btrfs_close_bdev(struct btrfs_device *device) 1098 { 1099 if (!device->bdev) 1100 return; 1101 1102 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1103 sync_blockdev(device->bdev); 1104 invalidate_bdev(device->bdev); 1105 } 1106 1107 blkdev_put(device->bdev, device->mode); 1108 } 1109 1110 static void btrfs_close_one_device(struct btrfs_device *device) 1111 { 1112 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1113 1114 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1115 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1116 list_del_init(&device->dev_alloc_list); 1117 fs_devices->rw_devices--; 1118 } 1119 1120 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1121 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1122 1123 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 1124 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 1125 fs_devices->missing_devices--; 1126 } 1127 1128 btrfs_close_bdev(device); 1129 if (device->bdev) { 1130 fs_devices->open_devices--; 1131 device->bdev = NULL; 1132 } 1133 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1134 btrfs_destroy_dev_zone_info(device); 1135 1136 device->fs_info = NULL; 1137 atomic_set(&device->dev_stats_ccnt, 0); 1138 extent_io_tree_release(&device->alloc_state); 1139 1140 /* 1141 * Reset the flush error record. We might have a transient flush error 1142 * in this mount, and if so we aborted the current transaction and set 1143 * the fs to an error state, guaranteeing no super blocks can be further 1144 * committed. However that error might be transient and if we unmount the 1145 * filesystem and mount it again, we should allow the mount to succeed 1146 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1147 * filesystem again we still get flush errors, then we will again abort 1148 * any transaction and set the error state, guaranteeing no commits of 1149 * unsafe super blocks. 1150 */ 1151 device->last_flush_error = 0; 1152 1153 /* Verify the device is back in a pristine state */ 1154 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1155 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1156 WARN_ON(!list_empty(&device->dev_alloc_list)); 1157 WARN_ON(!list_empty(&device->post_commit_list)); 1158 } 1159 1160 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1161 { 1162 struct btrfs_device *device, *tmp; 1163 1164 lockdep_assert_held(&uuid_mutex); 1165 1166 if (--fs_devices->opened > 0) 1167 return; 1168 1169 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1170 btrfs_close_one_device(device); 1171 1172 WARN_ON(fs_devices->open_devices); 1173 WARN_ON(fs_devices->rw_devices); 1174 fs_devices->opened = 0; 1175 fs_devices->seeding = false; 1176 fs_devices->fs_info = NULL; 1177 } 1178 1179 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1180 { 1181 LIST_HEAD(list); 1182 struct btrfs_fs_devices *tmp; 1183 1184 mutex_lock(&uuid_mutex); 1185 close_fs_devices(fs_devices); 1186 if (!fs_devices->opened) { 1187 list_splice_init(&fs_devices->seed_list, &list); 1188 1189 /* 1190 * If the struct btrfs_fs_devices is not assembled with any 1191 * other device, it can be re-initialized during the next mount 1192 * without the needing device-scan step. Therefore, it can be 1193 * fully freed. 1194 */ 1195 if (fs_devices->num_devices == 1) { 1196 list_del(&fs_devices->fs_list); 1197 free_fs_devices(fs_devices); 1198 } 1199 } 1200 1201 1202 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1203 close_fs_devices(fs_devices); 1204 list_del(&fs_devices->seed_list); 1205 free_fs_devices(fs_devices); 1206 } 1207 mutex_unlock(&uuid_mutex); 1208 } 1209 1210 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1211 fmode_t flags, void *holder) 1212 { 1213 struct btrfs_device *device; 1214 struct btrfs_device *latest_dev = NULL; 1215 struct btrfs_device *tmp_device; 1216 1217 flags |= FMODE_EXCL; 1218 1219 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1220 dev_list) { 1221 int ret; 1222 1223 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1224 if (ret == 0 && 1225 (!latest_dev || device->generation > latest_dev->generation)) { 1226 latest_dev = device; 1227 } else if (ret == -ENODATA) { 1228 fs_devices->num_devices--; 1229 list_del(&device->dev_list); 1230 btrfs_free_device(device); 1231 } 1232 } 1233 if (fs_devices->open_devices == 0) 1234 return -EINVAL; 1235 1236 fs_devices->opened = 1; 1237 fs_devices->latest_dev = latest_dev; 1238 fs_devices->total_rw_bytes = 0; 1239 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1240 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1241 1242 return 0; 1243 } 1244 1245 static int devid_cmp(void *priv, const struct list_head *a, 1246 const struct list_head *b) 1247 { 1248 const struct btrfs_device *dev1, *dev2; 1249 1250 dev1 = list_entry(a, struct btrfs_device, dev_list); 1251 dev2 = list_entry(b, struct btrfs_device, dev_list); 1252 1253 if (dev1->devid < dev2->devid) 1254 return -1; 1255 else if (dev1->devid > dev2->devid) 1256 return 1; 1257 return 0; 1258 } 1259 1260 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1261 fmode_t flags, void *holder) 1262 { 1263 int ret; 1264 1265 lockdep_assert_held(&uuid_mutex); 1266 /* 1267 * The device_list_mutex cannot be taken here in case opening the 1268 * underlying device takes further locks like open_mutex. 1269 * 1270 * We also don't need the lock here as this is called during mount and 1271 * exclusion is provided by uuid_mutex 1272 */ 1273 1274 if (fs_devices->opened) { 1275 fs_devices->opened++; 1276 ret = 0; 1277 } else { 1278 list_sort(NULL, &fs_devices->devices, devid_cmp); 1279 ret = open_fs_devices(fs_devices, flags, holder); 1280 } 1281 1282 return ret; 1283 } 1284 1285 void btrfs_release_disk_super(struct btrfs_super_block *super) 1286 { 1287 struct page *page = virt_to_page(super); 1288 1289 put_page(page); 1290 } 1291 1292 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1293 u64 bytenr, u64 bytenr_orig) 1294 { 1295 struct btrfs_super_block *disk_super; 1296 struct page *page; 1297 void *p; 1298 pgoff_t index; 1299 1300 /* make sure our super fits in the device */ 1301 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) 1302 return ERR_PTR(-EINVAL); 1303 1304 /* make sure our super fits in the page */ 1305 if (sizeof(*disk_super) > PAGE_SIZE) 1306 return ERR_PTR(-EINVAL); 1307 1308 /* make sure our super doesn't straddle pages on disk */ 1309 index = bytenr >> PAGE_SHIFT; 1310 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1311 return ERR_PTR(-EINVAL); 1312 1313 /* pull in the page with our super */ 1314 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1315 1316 if (IS_ERR(page)) 1317 return ERR_CAST(page); 1318 1319 p = page_address(page); 1320 1321 /* align our pointer to the offset of the super block */ 1322 disk_super = p + offset_in_page(bytenr); 1323 1324 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1325 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1326 btrfs_release_disk_super(p); 1327 return ERR_PTR(-EINVAL); 1328 } 1329 1330 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1331 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1332 1333 return disk_super; 1334 } 1335 1336 int btrfs_forget_devices(dev_t devt) 1337 { 1338 int ret; 1339 1340 mutex_lock(&uuid_mutex); 1341 ret = btrfs_free_stale_devices(devt, NULL); 1342 mutex_unlock(&uuid_mutex); 1343 1344 return ret; 1345 } 1346 1347 /* 1348 * Look for a btrfs signature on a device. This may be called out of the mount path 1349 * and we are not allowed to call set_blocksize during the scan. The superblock 1350 * is read via pagecache 1351 */ 1352 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1353 void *holder) 1354 { 1355 struct btrfs_super_block *disk_super; 1356 bool new_device_added = false; 1357 struct btrfs_device *device = NULL; 1358 struct block_device *bdev; 1359 u64 bytenr, bytenr_orig; 1360 int ret; 1361 1362 lockdep_assert_held(&uuid_mutex); 1363 1364 /* 1365 * we would like to check all the supers, but that would make 1366 * a btrfs mount succeed after a mkfs from a different FS. 1367 * So, we need to add a special mount option to scan for 1368 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1369 */ 1370 1371 /* 1372 * Avoid using flag |= FMODE_EXCL here, as the systemd-udev may 1373 * initiate the device scan which may race with the user's mount 1374 * or mkfs command, resulting in failure. 1375 * Since the device scan is solely for reading purposes, there is 1376 * no need for FMODE_EXCL. Additionally, the devices are read again 1377 * during the mount process. It is ok to get some inconsistent 1378 * values temporarily, as the device paths of the fsid are the only 1379 * required information for assembling the volume. 1380 */ 1381 bdev = blkdev_get_by_path(path, flags, holder); 1382 if (IS_ERR(bdev)) 1383 return ERR_CAST(bdev); 1384 1385 bytenr_orig = btrfs_sb_offset(0); 1386 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1387 if (ret) { 1388 device = ERR_PTR(ret); 1389 goto error_bdev_put; 1390 } 1391 1392 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1393 if (IS_ERR(disk_super)) { 1394 device = ERR_CAST(disk_super); 1395 goto error_bdev_put; 1396 } 1397 1398 device = device_list_add(path, disk_super, &new_device_added); 1399 if (!IS_ERR(device) && new_device_added) 1400 btrfs_free_stale_devices(device->devt, device); 1401 1402 btrfs_release_disk_super(disk_super); 1403 1404 error_bdev_put: 1405 blkdev_put(bdev, flags); 1406 1407 return device; 1408 } 1409 1410 /* 1411 * Try to find a chunk that intersects [start, start + len] range and when one 1412 * such is found, record the end of it in *start 1413 */ 1414 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1415 u64 len) 1416 { 1417 u64 physical_start, physical_end; 1418 1419 lockdep_assert_held(&device->fs_info->chunk_mutex); 1420 1421 if (!find_first_extent_bit(&device->alloc_state, *start, 1422 &physical_start, &physical_end, 1423 CHUNK_ALLOCATED, NULL)) { 1424 1425 if (in_range(physical_start, *start, len) || 1426 in_range(*start, physical_start, 1427 physical_end - physical_start)) { 1428 *start = physical_end + 1; 1429 return true; 1430 } 1431 } 1432 return false; 1433 } 1434 1435 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1436 { 1437 switch (device->fs_devices->chunk_alloc_policy) { 1438 case BTRFS_CHUNK_ALLOC_REGULAR: 1439 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); 1440 case BTRFS_CHUNK_ALLOC_ZONED: 1441 /* 1442 * We don't care about the starting region like regular 1443 * allocator, because we anyway use/reserve the first two zones 1444 * for superblock logging. 1445 */ 1446 return ALIGN(start, device->zone_info->zone_size); 1447 default: 1448 BUG(); 1449 } 1450 } 1451 1452 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1453 u64 *hole_start, u64 *hole_size, 1454 u64 num_bytes) 1455 { 1456 u64 zone_size = device->zone_info->zone_size; 1457 u64 pos; 1458 int ret; 1459 bool changed = false; 1460 1461 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1462 1463 while (*hole_size > 0) { 1464 pos = btrfs_find_allocatable_zones(device, *hole_start, 1465 *hole_start + *hole_size, 1466 num_bytes); 1467 if (pos != *hole_start) { 1468 *hole_size = *hole_start + *hole_size - pos; 1469 *hole_start = pos; 1470 changed = true; 1471 if (*hole_size < num_bytes) 1472 break; 1473 } 1474 1475 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1476 1477 /* Range is ensured to be empty */ 1478 if (!ret) 1479 return changed; 1480 1481 /* Given hole range was invalid (outside of device) */ 1482 if (ret == -ERANGE) { 1483 *hole_start += *hole_size; 1484 *hole_size = 0; 1485 return true; 1486 } 1487 1488 *hole_start += zone_size; 1489 *hole_size -= zone_size; 1490 changed = true; 1491 } 1492 1493 return changed; 1494 } 1495 1496 /* 1497 * Check if specified hole is suitable for allocation. 1498 * 1499 * @device: the device which we have the hole 1500 * @hole_start: starting position of the hole 1501 * @hole_size: the size of the hole 1502 * @num_bytes: the size of the free space that we need 1503 * 1504 * This function may modify @hole_start and @hole_size to reflect the suitable 1505 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1506 */ 1507 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1508 u64 *hole_size, u64 num_bytes) 1509 { 1510 bool changed = false; 1511 u64 hole_end = *hole_start + *hole_size; 1512 1513 for (;;) { 1514 /* 1515 * Check before we set max_hole_start, otherwise we could end up 1516 * sending back this offset anyway. 1517 */ 1518 if (contains_pending_extent(device, hole_start, *hole_size)) { 1519 if (hole_end >= *hole_start) 1520 *hole_size = hole_end - *hole_start; 1521 else 1522 *hole_size = 0; 1523 changed = true; 1524 } 1525 1526 switch (device->fs_devices->chunk_alloc_policy) { 1527 case BTRFS_CHUNK_ALLOC_REGULAR: 1528 /* No extra check */ 1529 break; 1530 case BTRFS_CHUNK_ALLOC_ZONED: 1531 if (dev_extent_hole_check_zoned(device, hole_start, 1532 hole_size, num_bytes)) { 1533 changed = true; 1534 /* 1535 * The changed hole can contain pending extent. 1536 * Loop again to check that. 1537 */ 1538 continue; 1539 } 1540 break; 1541 default: 1542 BUG(); 1543 } 1544 1545 break; 1546 } 1547 1548 return changed; 1549 } 1550 1551 /* 1552 * Find free space in the specified device. 1553 * 1554 * @device: the device which we search the free space in 1555 * @num_bytes: the size of the free space that we need 1556 * @search_start: the position from which to begin the search 1557 * @start: store the start of the free space. 1558 * @len: the size of the free space. that we find, or the size 1559 * of the max free space if we don't find suitable free space 1560 * 1561 * This does a pretty simple search, the expectation is that it is called very 1562 * infrequently and that a given device has a small number of extents. 1563 * 1564 * @start is used to store the start of the free space if we find. But if we 1565 * don't find suitable free space, it will be used to store the start position 1566 * of the max free space. 1567 * 1568 * @len is used to store the size of the free space that we find. 1569 * But if we don't find suitable free space, it is used to store the size of 1570 * the max free space. 1571 * 1572 * NOTE: This function will search *commit* root of device tree, and does extra 1573 * check to ensure dev extents are not double allocated. 1574 * This makes the function safe to allocate dev extents but may not report 1575 * correct usable device space, as device extent freed in current transaction 1576 * is not reported as available. 1577 */ 1578 static int find_free_dev_extent_start(struct btrfs_device *device, 1579 u64 num_bytes, u64 search_start, u64 *start, 1580 u64 *len) 1581 { 1582 struct btrfs_fs_info *fs_info = device->fs_info; 1583 struct btrfs_root *root = fs_info->dev_root; 1584 struct btrfs_key key; 1585 struct btrfs_dev_extent *dev_extent; 1586 struct btrfs_path *path; 1587 u64 hole_size; 1588 u64 max_hole_start; 1589 u64 max_hole_size; 1590 u64 extent_end; 1591 u64 search_end = device->total_bytes; 1592 int ret; 1593 int slot; 1594 struct extent_buffer *l; 1595 1596 search_start = dev_extent_search_start(device, search_start); 1597 1598 WARN_ON(device->zone_info && 1599 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1600 1601 path = btrfs_alloc_path(); 1602 if (!path) 1603 return -ENOMEM; 1604 1605 max_hole_start = search_start; 1606 max_hole_size = 0; 1607 1608 again: 1609 if (search_start >= search_end || 1610 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1611 ret = -ENOSPC; 1612 goto out; 1613 } 1614 1615 path->reada = READA_FORWARD; 1616 path->search_commit_root = 1; 1617 path->skip_locking = 1; 1618 1619 key.objectid = device->devid; 1620 key.offset = search_start; 1621 key.type = BTRFS_DEV_EXTENT_KEY; 1622 1623 ret = btrfs_search_backwards(root, &key, path); 1624 if (ret < 0) 1625 goto out; 1626 1627 while (search_start < search_end) { 1628 l = path->nodes[0]; 1629 slot = path->slots[0]; 1630 if (slot >= btrfs_header_nritems(l)) { 1631 ret = btrfs_next_leaf(root, path); 1632 if (ret == 0) 1633 continue; 1634 if (ret < 0) 1635 goto out; 1636 1637 break; 1638 } 1639 btrfs_item_key_to_cpu(l, &key, slot); 1640 1641 if (key.objectid < device->devid) 1642 goto next; 1643 1644 if (key.objectid > device->devid) 1645 break; 1646 1647 if (key.type != BTRFS_DEV_EXTENT_KEY) 1648 goto next; 1649 1650 if (key.offset > search_end) 1651 break; 1652 1653 if (key.offset > search_start) { 1654 hole_size = key.offset - search_start; 1655 dev_extent_hole_check(device, &search_start, &hole_size, 1656 num_bytes); 1657 1658 if (hole_size > max_hole_size) { 1659 max_hole_start = search_start; 1660 max_hole_size = hole_size; 1661 } 1662 1663 /* 1664 * If this free space is greater than which we need, 1665 * it must be the max free space that we have found 1666 * until now, so max_hole_start must point to the start 1667 * of this free space and the length of this free space 1668 * is stored in max_hole_size. Thus, we return 1669 * max_hole_start and max_hole_size and go back to the 1670 * caller. 1671 */ 1672 if (hole_size >= num_bytes) { 1673 ret = 0; 1674 goto out; 1675 } 1676 } 1677 1678 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1679 extent_end = key.offset + btrfs_dev_extent_length(l, 1680 dev_extent); 1681 if (extent_end > search_start) 1682 search_start = extent_end; 1683 next: 1684 path->slots[0]++; 1685 cond_resched(); 1686 } 1687 1688 /* 1689 * At this point, search_start should be the end of 1690 * allocated dev extents, and when shrinking the device, 1691 * search_end may be smaller than search_start. 1692 */ 1693 if (search_end > search_start) { 1694 hole_size = search_end - search_start; 1695 if (dev_extent_hole_check(device, &search_start, &hole_size, 1696 num_bytes)) { 1697 btrfs_release_path(path); 1698 goto again; 1699 } 1700 1701 if (hole_size > max_hole_size) { 1702 max_hole_start = search_start; 1703 max_hole_size = hole_size; 1704 } 1705 } 1706 1707 /* See above. */ 1708 if (max_hole_size < num_bytes) 1709 ret = -ENOSPC; 1710 else 1711 ret = 0; 1712 1713 ASSERT(max_hole_start + max_hole_size <= search_end); 1714 out: 1715 btrfs_free_path(path); 1716 *start = max_hole_start; 1717 if (len) 1718 *len = max_hole_size; 1719 return ret; 1720 } 1721 1722 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1723 u64 *start, u64 *len) 1724 { 1725 /* FIXME use last free of some kind */ 1726 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1727 } 1728 1729 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1730 struct btrfs_device *device, 1731 u64 start, u64 *dev_extent_len) 1732 { 1733 struct btrfs_fs_info *fs_info = device->fs_info; 1734 struct btrfs_root *root = fs_info->dev_root; 1735 int ret; 1736 struct btrfs_path *path; 1737 struct btrfs_key key; 1738 struct btrfs_key found_key; 1739 struct extent_buffer *leaf = NULL; 1740 struct btrfs_dev_extent *extent = NULL; 1741 1742 path = btrfs_alloc_path(); 1743 if (!path) 1744 return -ENOMEM; 1745 1746 key.objectid = device->devid; 1747 key.offset = start; 1748 key.type = BTRFS_DEV_EXTENT_KEY; 1749 again: 1750 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1751 if (ret > 0) { 1752 ret = btrfs_previous_item(root, path, key.objectid, 1753 BTRFS_DEV_EXTENT_KEY); 1754 if (ret) 1755 goto out; 1756 leaf = path->nodes[0]; 1757 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1758 extent = btrfs_item_ptr(leaf, path->slots[0], 1759 struct btrfs_dev_extent); 1760 BUG_ON(found_key.offset > start || found_key.offset + 1761 btrfs_dev_extent_length(leaf, extent) < start); 1762 key = found_key; 1763 btrfs_release_path(path); 1764 goto again; 1765 } else if (ret == 0) { 1766 leaf = path->nodes[0]; 1767 extent = btrfs_item_ptr(leaf, path->slots[0], 1768 struct btrfs_dev_extent); 1769 } else { 1770 goto out; 1771 } 1772 1773 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1774 1775 ret = btrfs_del_item(trans, root, path); 1776 if (ret == 0) 1777 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1778 out: 1779 btrfs_free_path(path); 1780 return ret; 1781 } 1782 1783 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1784 { 1785 struct extent_map_tree *em_tree; 1786 struct extent_map *em; 1787 struct rb_node *n; 1788 u64 ret = 0; 1789 1790 em_tree = &fs_info->mapping_tree; 1791 read_lock(&em_tree->lock); 1792 n = rb_last(&em_tree->map.rb_root); 1793 if (n) { 1794 em = rb_entry(n, struct extent_map, rb_node); 1795 ret = em->start + em->len; 1796 } 1797 read_unlock(&em_tree->lock); 1798 1799 return ret; 1800 } 1801 1802 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1803 u64 *devid_ret) 1804 { 1805 int ret; 1806 struct btrfs_key key; 1807 struct btrfs_key found_key; 1808 struct btrfs_path *path; 1809 1810 path = btrfs_alloc_path(); 1811 if (!path) 1812 return -ENOMEM; 1813 1814 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1815 key.type = BTRFS_DEV_ITEM_KEY; 1816 key.offset = (u64)-1; 1817 1818 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1819 if (ret < 0) 1820 goto error; 1821 1822 if (ret == 0) { 1823 /* Corruption */ 1824 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1825 ret = -EUCLEAN; 1826 goto error; 1827 } 1828 1829 ret = btrfs_previous_item(fs_info->chunk_root, path, 1830 BTRFS_DEV_ITEMS_OBJECTID, 1831 BTRFS_DEV_ITEM_KEY); 1832 if (ret) { 1833 *devid_ret = 1; 1834 } else { 1835 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1836 path->slots[0]); 1837 *devid_ret = found_key.offset + 1; 1838 } 1839 ret = 0; 1840 error: 1841 btrfs_free_path(path); 1842 return ret; 1843 } 1844 1845 /* 1846 * the device information is stored in the chunk root 1847 * the btrfs_device struct should be fully filled in 1848 */ 1849 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1850 struct btrfs_device *device) 1851 { 1852 int ret; 1853 struct btrfs_path *path; 1854 struct btrfs_dev_item *dev_item; 1855 struct extent_buffer *leaf; 1856 struct btrfs_key key; 1857 unsigned long ptr; 1858 1859 path = btrfs_alloc_path(); 1860 if (!path) 1861 return -ENOMEM; 1862 1863 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1864 key.type = BTRFS_DEV_ITEM_KEY; 1865 key.offset = device->devid; 1866 1867 btrfs_reserve_chunk_metadata(trans, true); 1868 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1869 &key, sizeof(*dev_item)); 1870 btrfs_trans_release_chunk_metadata(trans); 1871 if (ret) 1872 goto out; 1873 1874 leaf = path->nodes[0]; 1875 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1876 1877 btrfs_set_device_id(leaf, dev_item, device->devid); 1878 btrfs_set_device_generation(leaf, dev_item, 0); 1879 btrfs_set_device_type(leaf, dev_item, device->type); 1880 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1881 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1882 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1883 btrfs_set_device_total_bytes(leaf, dev_item, 1884 btrfs_device_get_disk_total_bytes(device)); 1885 btrfs_set_device_bytes_used(leaf, dev_item, 1886 btrfs_device_get_bytes_used(device)); 1887 btrfs_set_device_group(leaf, dev_item, 0); 1888 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1889 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1890 btrfs_set_device_start_offset(leaf, dev_item, 0); 1891 1892 ptr = btrfs_device_uuid(dev_item); 1893 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1894 ptr = btrfs_device_fsid(dev_item); 1895 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1896 ptr, BTRFS_FSID_SIZE); 1897 btrfs_mark_buffer_dirty(leaf); 1898 1899 ret = 0; 1900 out: 1901 btrfs_free_path(path); 1902 return ret; 1903 } 1904 1905 /* 1906 * Function to update ctime/mtime for a given device path. 1907 * Mainly used for ctime/mtime based probe like libblkid. 1908 * 1909 * We don't care about errors here, this is just to be kind to userspace. 1910 */ 1911 static void update_dev_time(const char *device_path) 1912 { 1913 struct path path; 1914 struct timespec64 now; 1915 int ret; 1916 1917 ret = kern_path(device_path, LOOKUP_FOLLOW, &path); 1918 if (ret) 1919 return; 1920 1921 now = current_time(d_inode(path.dentry)); 1922 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); 1923 path_put(&path); 1924 } 1925 1926 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, 1927 struct btrfs_device *device) 1928 { 1929 struct btrfs_root *root = device->fs_info->chunk_root; 1930 int ret; 1931 struct btrfs_path *path; 1932 struct btrfs_key key; 1933 1934 path = btrfs_alloc_path(); 1935 if (!path) 1936 return -ENOMEM; 1937 1938 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1939 key.type = BTRFS_DEV_ITEM_KEY; 1940 key.offset = device->devid; 1941 1942 btrfs_reserve_chunk_metadata(trans, false); 1943 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1944 btrfs_trans_release_chunk_metadata(trans); 1945 if (ret) { 1946 if (ret > 0) 1947 ret = -ENOENT; 1948 goto out; 1949 } 1950 1951 ret = btrfs_del_item(trans, root, path); 1952 out: 1953 btrfs_free_path(path); 1954 return ret; 1955 } 1956 1957 /* 1958 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1959 * filesystem. It's up to the caller to adjust that number regarding eg. device 1960 * replace. 1961 */ 1962 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1963 u64 num_devices) 1964 { 1965 u64 all_avail; 1966 unsigned seq; 1967 int i; 1968 1969 do { 1970 seq = read_seqbegin(&fs_info->profiles_lock); 1971 1972 all_avail = fs_info->avail_data_alloc_bits | 1973 fs_info->avail_system_alloc_bits | 1974 fs_info->avail_metadata_alloc_bits; 1975 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1976 1977 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1978 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1979 continue; 1980 1981 if (num_devices < btrfs_raid_array[i].devs_min) 1982 return btrfs_raid_array[i].mindev_error; 1983 } 1984 1985 return 0; 1986 } 1987 1988 static struct btrfs_device * btrfs_find_next_active_device( 1989 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1990 { 1991 struct btrfs_device *next_device; 1992 1993 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1994 if (next_device != device && 1995 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1996 && next_device->bdev) 1997 return next_device; 1998 } 1999 2000 return NULL; 2001 } 2002 2003 /* 2004 * Helper function to check if the given device is part of s_bdev / latest_dev 2005 * and replace it with the provided or the next active device, in the context 2006 * where this function called, there should be always be another device (or 2007 * this_dev) which is active. 2008 */ 2009 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 2010 struct btrfs_device *next_device) 2011 { 2012 struct btrfs_fs_info *fs_info = device->fs_info; 2013 2014 if (!next_device) 2015 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 2016 device); 2017 ASSERT(next_device); 2018 2019 if (fs_info->sb->s_bdev && 2020 (fs_info->sb->s_bdev == device->bdev)) 2021 fs_info->sb->s_bdev = next_device->bdev; 2022 2023 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2024 fs_info->fs_devices->latest_dev = next_device; 2025 } 2026 2027 /* 2028 * Return btrfs_fs_devices::num_devices excluding the device that's being 2029 * currently replaced. 2030 */ 2031 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2032 { 2033 u64 num_devices = fs_info->fs_devices->num_devices; 2034 2035 down_read(&fs_info->dev_replace.rwsem); 2036 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2037 ASSERT(num_devices > 1); 2038 num_devices--; 2039 } 2040 up_read(&fs_info->dev_replace.rwsem); 2041 2042 return num_devices; 2043 } 2044 2045 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info, 2046 struct block_device *bdev, int copy_num) 2047 { 2048 struct btrfs_super_block *disk_super; 2049 const size_t len = sizeof(disk_super->magic); 2050 const u64 bytenr = btrfs_sb_offset(copy_num); 2051 int ret; 2052 2053 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr); 2054 if (IS_ERR(disk_super)) 2055 return; 2056 2057 memset(&disk_super->magic, 0, len); 2058 folio_mark_dirty(virt_to_folio(disk_super)); 2059 btrfs_release_disk_super(disk_super); 2060 2061 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1); 2062 if (ret) 2063 btrfs_warn(fs_info, "error clearing superblock number %d (%d)", 2064 copy_num, ret); 2065 } 2066 2067 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2068 struct block_device *bdev, 2069 const char *device_path) 2070 { 2071 int copy_num; 2072 2073 if (!bdev) 2074 return; 2075 2076 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2077 if (bdev_is_zoned(bdev)) 2078 btrfs_reset_sb_log_zones(bdev, copy_num); 2079 else 2080 btrfs_scratch_superblock(fs_info, bdev, copy_num); 2081 } 2082 2083 /* Notify udev that device has changed */ 2084 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2085 2086 /* Update ctime/mtime for device path for libblkid */ 2087 update_dev_time(device_path); 2088 } 2089 2090 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2091 struct btrfs_dev_lookup_args *args, 2092 struct block_device **bdev, fmode_t *mode) 2093 { 2094 struct btrfs_trans_handle *trans; 2095 struct btrfs_device *device; 2096 struct btrfs_fs_devices *cur_devices; 2097 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2098 u64 num_devices; 2099 int ret = 0; 2100 2101 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2102 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); 2103 return -EINVAL; 2104 } 2105 2106 /* 2107 * The device list in fs_devices is accessed without locks (neither 2108 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2109 * filesystem and another device rm cannot run. 2110 */ 2111 num_devices = btrfs_num_devices(fs_info); 2112 2113 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2114 if (ret) 2115 return ret; 2116 2117 device = btrfs_find_device(fs_info->fs_devices, args); 2118 if (!device) { 2119 if (args->missing) 2120 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2121 else 2122 ret = -ENOENT; 2123 return ret; 2124 } 2125 2126 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2127 btrfs_warn_in_rcu(fs_info, 2128 "cannot remove device %s (devid %llu) due to active swapfile", 2129 btrfs_dev_name(device), device->devid); 2130 return -ETXTBSY; 2131 } 2132 2133 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 2134 return BTRFS_ERROR_DEV_TGT_REPLACE; 2135 2136 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2137 fs_info->fs_devices->rw_devices == 1) 2138 return BTRFS_ERROR_DEV_ONLY_WRITABLE; 2139 2140 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2141 mutex_lock(&fs_info->chunk_mutex); 2142 list_del_init(&device->dev_alloc_list); 2143 device->fs_devices->rw_devices--; 2144 mutex_unlock(&fs_info->chunk_mutex); 2145 } 2146 2147 ret = btrfs_shrink_device(device, 0); 2148 if (ret) 2149 goto error_undo; 2150 2151 trans = btrfs_start_transaction(fs_info->chunk_root, 0); 2152 if (IS_ERR(trans)) { 2153 ret = PTR_ERR(trans); 2154 goto error_undo; 2155 } 2156 2157 ret = btrfs_rm_dev_item(trans, device); 2158 if (ret) { 2159 /* Any error in dev item removal is critical */ 2160 btrfs_crit(fs_info, 2161 "failed to remove device item for devid %llu: %d", 2162 device->devid, ret); 2163 btrfs_abort_transaction(trans, ret); 2164 btrfs_end_transaction(trans); 2165 return ret; 2166 } 2167 2168 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2169 btrfs_scrub_cancel_dev(device); 2170 2171 /* 2172 * the device list mutex makes sure that we don't change 2173 * the device list while someone else is writing out all 2174 * the device supers. Whoever is writing all supers, should 2175 * lock the device list mutex before getting the number of 2176 * devices in the super block (super_copy). Conversely, 2177 * whoever updates the number of devices in the super block 2178 * (super_copy) should hold the device list mutex. 2179 */ 2180 2181 /* 2182 * In normal cases the cur_devices == fs_devices. But in case 2183 * of deleting a seed device, the cur_devices should point to 2184 * its own fs_devices listed under the fs_devices->seed_list. 2185 */ 2186 cur_devices = device->fs_devices; 2187 mutex_lock(&fs_devices->device_list_mutex); 2188 list_del_rcu(&device->dev_list); 2189 2190 cur_devices->num_devices--; 2191 cur_devices->total_devices--; 2192 /* Update total_devices of the parent fs_devices if it's seed */ 2193 if (cur_devices != fs_devices) 2194 fs_devices->total_devices--; 2195 2196 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2197 cur_devices->missing_devices--; 2198 2199 btrfs_assign_next_active_device(device, NULL); 2200 2201 if (device->bdev) { 2202 cur_devices->open_devices--; 2203 /* remove sysfs entry */ 2204 btrfs_sysfs_remove_device(device); 2205 } 2206 2207 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2208 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2209 mutex_unlock(&fs_devices->device_list_mutex); 2210 2211 /* 2212 * At this point, the device is zero sized and detached from the 2213 * devices list. All that's left is to zero out the old supers and 2214 * free the device. 2215 * 2216 * We cannot call btrfs_close_bdev() here because we're holding the sb 2217 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2218 * block device and it's dependencies. Instead just flush the device 2219 * and let the caller do the final blkdev_put. 2220 */ 2221 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2222 btrfs_scratch_superblocks(fs_info, device->bdev, 2223 device->name->str); 2224 if (device->bdev) { 2225 sync_blockdev(device->bdev); 2226 invalidate_bdev(device->bdev); 2227 } 2228 } 2229 2230 *bdev = device->bdev; 2231 *mode = device->mode; 2232 synchronize_rcu(); 2233 btrfs_free_device(device); 2234 2235 /* 2236 * This can happen if cur_devices is the private seed devices list. We 2237 * cannot call close_fs_devices() here because it expects the uuid_mutex 2238 * to be held, but in fact we don't need that for the private 2239 * seed_devices, we can simply decrement cur_devices->opened and then 2240 * remove it from our list and free the fs_devices. 2241 */ 2242 if (cur_devices->num_devices == 0) { 2243 list_del_init(&cur_devices->seed_list); 2244 ASSERT(cur_devices->opened == 1); 2245 cur_devices->opened--; 2246 free_fs_devices(cur_devices); 2247 } 2248 2249 ret = btrfs_commit_transaction(trans); 2250 2251 return ret; 2252 2253 error_undo: 2254 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2255 mutex_lock(&fs_info->chunk_mutex); 2256 list_add(&device->dev_alloc_list, 2257 &fs_devices->alloc_list); 2258 device->fs_devices->rw_devices++; 2259 mutex_unlock(&fs_info->chunk_mutex); 2260 } 2261 return ret; 2262 } 2263 2264 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2265 { 2266 struct btrfs_fs_devices *fs_devices; 2267 2268 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2269 2270 /* 2271 * in case of fs with no seed, srcdev->fs_devices will point 2272 * to fs_devices of fs_info. However when the dev being replaced is 2273 * a seed dev it will point to the seed's local fs_devices. In short 2274 * srcdev will have its correct fs_devices in both the cases. 2275 */ 2276 fs_devices = srcdev->fs_devices; 2277 2278 list_del_rcu(&srcdev->dev_list); 2279 list_del(&srcdev->dev_alloc_list); 2280 fs_devices->num_devices--; 2281 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2282 fs_devices->missing_devices--; 2283 2284 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2285 fs_devices->rw_devices--; 2286 2287 if (srcdev->bdev) 2288 fs_devices->open_devices--; 2289 } 2290 2291 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2292 { 2293 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2294 2295 mutex_lock(&uuid_mutex); 2296 2297 btrfs_close_bdev(srcdev); 2298 synchronize_rcu(); 2299 btrfs_free_device(srcdev); 2300 2301 /* if this is no devs we rather delete the fs_devices */ 2302 if (!fs_devices->num_devices) { 2303 /* 2304 * On a mounted FS, num_devices can't be zero unless it's a 2305 * seed. In case of a seed device being replaced, the replace 2306 * target added to the sprout FS, so there will be no more 2307 * device left under the seed FS. 2308 */ 2309 ASSERT(fs_devices->seeding); 2310 2311 list_del_init(&fs_devices->seed_list); 2312 close_fs_devices(fs_devices); 2313 free_fs_devices(fs_devices); 2314 } 2315 mutex_unlock(&uuid_mutex); 2316 } 2317 2318 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2319 { 2320 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2321 2322 mutex_lock(&fs_devices->device_list_mutex); 2323 2324 btrfs_sysfs_remove_device(tgtdev); 2325 2326 if (tgtdev->bdev) 2327 fs_devices->open_devices--; 2328 2329 fs_devices->num_devices--; 2330 2331 btrfs_assign_next_active_device(tgtdev, NULL); 2332 2333 list_del_rcu(&tgtdev->dev_list); 2334 2335 mutex_unlock(&fs_devices->device_list_mutex); 2336 2337 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2338 tgtdev->name->str); 2339 2340 btrfs_close_bdev(tgtdev); 2341 synchronize_rcu(); 2342 btrfs_free_device(tgtdev); 2343 } 2344 2345 /* 2346 * Populate args from device at path. 2347 * 2348 * @fs_info: the filesystem 2349 * @args: the args to populate 2350 * @path: the path to the device 2351 * 2352 * This will read the super block of the device at @path and populate @args with 2353 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2354 * lookup a device to operate on, but need to do it before we take any locks. 2355 * This properly handles the special case of "missing" that a user may pass in, 2356 * and does some basic sanity checks. The caller must make sure that @path is 2357 * properly NUL terminated before calling in, and must call 2358 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2359 * uuid buffers. 2360 * 2361 * Return: 0 for success, -errno for failure 2362 */ 2363 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2364 struct btrfs_dev_lookup_args *args, 2365 const char *path) 2366 { 2367 struct btrfs_super_block *disk_super; 2368 struct block_device *bdev; 2369 int ret; 2370 2371 if (!path || !path[0]) 2372 return -EINVAL; 2373 if (!strcmp(path, "missing")) { 2374 args->missing = true; 2375 return 0; 2376 } 2377 2378 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2379 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2380 if (!args->uuid || !args->fsid) { 2381 btrfs_put_dev_args_from_path(args); 2382 return -ENOMEM; 2383 } 2384 2385 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2386 &bdev, &disk_super); 2387 if (ret) { 2388 btrfs_put_dev_args_from_path(args); 2389 return ret; 2390 } 2391 2392 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2393 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2394 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2395 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2396 else 2397 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2398 btrfs_release_disk_super(disk_super); 2399 blkdev_put(bdev, FMODE_READ); 2400 return 0; 2401 } 2402 2403 /* 2404 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2405 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2406 * that don't need to be freed. 2407 */ 2408 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2409 { 2410 kfree(args->uuid); 2411 kfree(args->fsid); 2412 args->uuid = NULL; 2413 args->fsid = NULL; 2414 } 2415 2416 struct btrfs_device *btrfs_find_device_by_devspec( 2417 struct btrfs_fs_info *fs_info, u64 devid, 2418 const char *device_path) 2419 { 2420 BTRFS_DEV_LOOKUP_ARGS(args); 2421 struct btrfs_device *device; 2422 int ret; 2423 2424 if (devid) { 2425 args.devid = devid; 2426 device = btrfs_find_device(fs_info->fs_devices, &args); 2427 if (!device) 2428 return ERR_PTR(-ENOENT); 2429 return device; 2430 } 2431 2432 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2433 if (ret) 2434 return ERR_PTR(ret); 2435 device = btrfs_find_device(fs_info->fs_devices, &args); 2436 btrfs_put_dev_args_from_path(&args); 2437 if (!device) 2438 return ERR_PTR(-ENOENT); 2439 return device; 2440 } 2441 2442 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) 2443 { 2444 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2445 struct btrfs_fs_devices *old_devices; 2446 struct btrfs_fs_devices *seed_devices; 2447 2448 lockdep_assert_held(&uuid_mutex); 2449 if (!fs_devices->seeding) 2450 return ERR_PTR(-EINVAL); 2451 2452 /* 2453 * Private copy of the seed devices, anchored at 2454 * fs_info->fs_devices->seed_list 2455 */ 2456 seed_devices = alloc_fs_devices(NULL, NULL); 2457 if (IS_ERR(seed_devices)) 2458 return seed_devices; 2459 2460 /* 2461 * It's necessary to retain a copy of the original seed fs_devices in 2462 * fs_uuids so that filesystems which have been seeded can successfully 2463 * reference the seed device from open_seed_devices. This also supports 2464 * multiple fs seed. 2465 */ 2466 old_devices = clone_fs_devices(fs_devices); 2467 if (IS_ERR(old_devices)) { 2468 kfree(seed_devices); 2469 return old_devices; 2470 } 2471 2472 list_add(&old_devices->fs_list, &fs_uuids); 2473 2474 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2475 seed_devices->opened = 1; 2476 INIT_LIST_HEAD(&seed_devices->devices); 2477 INIT_LIST_HEAD(&seed_devices->alloc_list); 2478 mutex_init(&seed_devices->device_list_mutex); 2479 2480 return seed_devices; 2481 } 2482 2483 /* 2484 * Splice seed devices into the sprout fs_devices. 2485 * Generate a new fsid for the sprouted read-write filesystem. 2486 */ 2487 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, 2488 struct btrfs_fs_devices *seed_devices) 2489 { 2490 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2491 struct btrfs_super_block *disk_super = fs_info->super_copy; 2492 struct btrfs_device *device; 2493 u64 super_flags; 2494 2495 /* 2496 * We are updating the fsid, the thread leading to device_list_add() 2497 * could race, so uuid_mutex is needed. 2498 */ 2499 lockdep_assert_held(&uuid_mutex); 2500 2501 /* 2502 * The threads listed below may traverse dev_list but can do that without 2503 * device_list_mutex: 2504 * - All device ops and balance - as we are in btrfs_exclop_start. 2505 * - Various dev_list readers - are using RCU. 2506 * - btrfs_ioctl_fitrim() - is using RCU. 2507 * 2508 * For-read threads as below are using device_list_mutex: 2509 * - Readonly scrub btrfs_scrub_dev() 2510 * - Readonly scrub btrfs_scrub_progress() 2511 * - btrfs_get_dev_stats() 2512 */ 2513 lockdep_assert_held(&fs_devices->device_list_mutex); 2514 2515 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2516 synchronize_rcu); 2517 list_for_each_entry(device, &seed_devices->devices, dev_list) 2518 device->fs_devices = seed_devices; 2519 2520 fs_devices->seeding = false; 2521 fs_devices->num_devices = 0; 2522 fs_devices->open_devices = 0; 2523 fs_devices->missing_devices = 0; 2524 fs_devices->rotating = false; 2525 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2526 2527 generate_random_uuid(fs_devices->fsid); 2528 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2529 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2530 2531 super_flags = btrfs_super_flags(disk_super) & 2532 ~BTRFS_SUPER_FLAG_SEEDING; 2533 btrfs_set_super_flags(disk_super, super_flags); 2534 } 2535 2536 /* 2537 * Store the expected generation for seed devices in device items. 2538 */ 2539 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2540 { 2541 BTRFS_DEV_LOOKUP_ARGS(args); 2542 struct btrfs_fs_info *fs_info = trans->fs_info; 2543 struct btrfs_root *root = fs_info->chunk_root; 2544 struct btrfs_path *path; 2545 struct extent_buffer *leaf; 2546 struct btrfs_dev_item *dev_item; 2547 struct btrfs_device *device; 2548 struct btrfs_key key; 2549 u8 fs_uuid[BTRFS_FSID_SIZE]; 2550 u8 dev_uuid[BTRFS_UUID_SIZE]; 2551 int ret; 2552 2553 path = btrfs_alloc_path(); 2554 if (!path) 2555 return -ENOMEM; 2556 2557 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2558 key.offset = 0; 2559 key.type = BTRFS_DEV_ITEM_KEY; 2560 2561 while (1) { 2562 btrfs_reserve_chunk_metadata(trans, false); 2563 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2564 btrfs_trans_release_chunk_metadata(trans); 2565 if (ret < 0) 2566 goto error; 2567 2568 leaf = path->nodes[0]; 2569 next_slot: 2570 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2571 ret = btrfs_next_leaf(root, path); 2572 if (ret > 0) 2573 break; 2574 if (ret < 0) 2575 goto error; 2576 leaf = path->nodes[0]; 2577 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2578 btrfs_release_path(path); 2579 continue; 2580 } 2581 2582 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2583 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2584 key.type != BTRFS_DEV_ITEM_KEY) 2585 break; 2586 2587 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2588 struct btrfs_dev_item); 2589 args.devid = btrfs_device_id(leaf, dev_item); 2590 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2591 BTRFS_UUID_SIZE); 2592 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2593 BTRFS_FSID_SIZE); 2594 args.uuid = dev_uuid; 2595 args.fsid = fs_uuid; 2596 device = btrfs_find_device(fs_info->fs_devices, &args); 2597 BUG_ON(!device); /* Logic error */ 2598 2599 if (device->fs_devices->seeding) { 2600 btrfs_set_device_generation(leaf, dev_item, 2601 device->generation); 2602 btrfs_mark_buffer_dirty(leaf); 2603 } 2604 2605 path->slots[0]++; 2606 goto next_slot; 2607 } 2608 ret = 0; 2609 error: 2610 btrfs_free_path(path); 2611 return ret; 2612 } 2613 2614 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2615 { 2616 struct btrfs_root *root = fs_info->dev_root; 2617 struct btrfs_trans_handle *trans; 2618 struct btrfs_device *device; 2619 struct block_device *bdev; 2620 struct super_block *sb = fs_info->sb; 2621 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2622 struct btrfs_fs_devices *seed_devices = NULL; 2623 u64 orig_super_total_bytes; 2624 u64 orig_super_num_devices; 2625 int ret = 0; 2626 bool seeding_dev = false; 2627 bool locked = false; 2628 2629 if (sb_rdonly(sb) && !fs_devices->seeding) 2630 return -EROFS; 2631 2632 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2633 fs_info->bdev_holder); 2634 if (IS_ERR(bdev)) 2635 return PTR_ERR(bdev); 2636 2637 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2638 ret = -EINVAL; 2639 goto error; 2640 } 2641 2642 if (fs_devices->seeding) { 2643 seeding_dev = true; 2644 down_write(&sb->s_umount); 2645 mutex_lock(&uuid_mutex); 2646 locked = true; 2647 } 2648 2649 sync_blockdev(bdev); 2650 2651 rcu_read_lock(); 2652 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2653 if (device->bdev == bdev) { 2654 ret = -EEXIST; 2655 rcu_read_unlock(); 2656 goto error; 2657 } 2658 } 2659 rcu_read_unlock(); 2660 2661 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path); 2662 if (IS_ERR(device)) { 2663 /* we can safely leave the fs_devices entry around */ 2664 ret = PTR_ERR(device); 2665 goto error; 2666 } 2667 2668 device->fs_info = fs_info; 2669 device->bdev = bdev; 2670 ret = lookup_bdev(device_path, &device->devt); 2671 if (ret) 2672 goto error_free_device; 2673 2674 ret = btrfs_get_dev_zone_info(device, false); 2675 if (ret) 2676 goto error_free_device; 2677 2678 trans = btrfs_start_transaction(root, 0); 2679 if (IS_ERR(trans)) { 2680 ret = PTR_ERR(trans); 2681 goto error_free_zone; 2682 } 2683 2684 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2685 device->generation = trans->transid; 2686 device->io_width = fs_info->sectorsize; 2687 device->io_align = fs_info->sectorsize; 2688 device->sector_size = fs_info->sectorsize; 2689 device->total_bytes = 2690 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); 2691 device->disk_total_bytes = device->total_bytes; 2692 device->commit_total_bytes = device->total_bytes; 2693 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2694 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2695 device->mode = FMODE_EXCL; 2696 device->dev_stats_valid = 1; 2697 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2698 2699 if (seeding_dev) { 2700 btrfs_clear_sb_rdonly(sb); 2701 2702 /* GFP_KERNEL allocation must not be under device_list_mutex */ 2703 seed_devices = btrfs_init_sprout(fs_info); 2704 if (IS_ERR(seed_devices)) { 2705 ret = PTR_ERR(seed_devices); 2706 btrfs_abort_transaction(trans, ret); 2707 goto error_trans; 2708 } 2709 } 2710 2711 mutex_lock(&fs_devices->device_list_mutex); 2712 if (seeding_dev) { 2713 btrfs_setup_sprout(fs_info, seed_devices); 2714 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2715 device); 2716 } 2717 2718 device->fs_devices = fs_devices; 2719 2720 mutex_lock(&fs_info->chunk_mutex); 2721 list_add_rcu(&device->dev_list, &fs_devices->devices); 2722 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2723 fs_devices->num_devices++; 2724 fs_devices->open_devices++; 2725 fs_devices->rw_devices++; 2726 fs_devices->total_devices++; 2727 fs_devices->total_rw_bytes += device->total_bytes; 2728 2729 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2730 2731 if (!bdev_nonrot(bdev)) 2732 fs_devices->rotating = true; 2733 2734 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2735 btrfs_set_super_total_bytes(fs_info->super_copy, 2736 round_down(orig_super_total_bytes + device->total_bytes, 2737 fs_info->sectorsize)); 2738 2739 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2740 btrfs_set_super_num_devices(fs_info->super_copy, 2741 orig_super_num_devices + 1); 2742 2743 /* 2744 * we've got more storage, clear any full flags on the space 2745 * infos 2746 */ 2747 btrfs_clear_space_info_full(fs_info); 2748 2749 mutex_unlock(&fs_info->chunk_mutex); 2750 2751 /* Add sysfs device entry */ 2752 btrfs_sysfs_add_device(device); 2753 2754 mutex_unlock(&fs_devices->device_list_mutex); 2755 2756 if (seeding_dev) { 2757 mutex_lock(&fs_info->chunk_mutex); 2758 ret = init_first_rw_device(trans); 2759 mutex_unlock(&fs_info->chunk_mutex); 2760 if (ret) { 2761 btrfs_abort_transaction(trans, ret); 2762 goto error_sysfs; 2763 } 2764 } 2765 2766 ret = btrfs_add_dev_item(trans, device); 2767 if (ret) { 2768 btrfs_abort_transaction(trans, ret); 2769 goto error_sysfs; 2770 } 2771 2772 if (seeding_dev) { 2773 ret = btrfs_finish_sprout(trans); 2774 if (ret) { 2775 btrfs_abort_transaction(trans, ret); 2776 goto error_sysfs; 2777 } 2778 2779 /* 2780 * fs_devices now represents the newly sprouted filesystem and 2781 * its fsid has been changed by btrfs_sprout_splice(). 2782 */ 2783 btrfs_sysfs_update_sprout_fsid(fs_devices); 2784 } 2785 2786 ret = btrfs_commit_transaction(trans); 2787 2788 if (seeding_dev) { 2789 mutex_unlock(&uuid_mutex); 2790 up_write(&sb->s_umount); 2791 locked = false; 2792 2793 if (ret) /* transaction commit */ 2794 return ret; 2795 2796 ret = btrfs_relocate_sys_chunks(fs_info); 2797 if (ret < 0) 2798 btrfs_handle_fs_error(fs_info, ret, 2799 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2800 trans = btrfs_attach_transaction(root); 2801 if (IS_ERR(trans)) { 2802 if (PTR_ERR(trans) == -ENOENT) 2803 return 0; 2804 ret = PTR_ERR(trans); 2805 trans = NULL; 2806 goto error_sysfs; 2807 } 2808 ret = btrfs_commit_transaction(trans); 2809 } 2810 2811 /* 2812 * Now that we have written a new super block to this device, check all 2813 * other fs_devices list if device_path alienates any other scanned 2814 * device. 2815 * We can ignore the return value as it typically returns -EINVAL and 2816 * only succeeds if the device was an alien. 2817 */ 2818 btrfs_forget_devices(device->devt); 2819 2820 /* Update ctime/mtime for blkid or udev */ 2821 update_dev_time(device_path); 2822 2823 return ret; 2824 2825 error_sysfs: 2826 btrfs_sysfs_remove_device(device); 2827 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2828 mutex_lock(&fs_info->chunk_mutex); 2829 list_del_rcu(&device->dev_list); 2830 list_del(&device->dev_alloc_list); 2831 fs_info->fs_devices->num_devices--; 2832 fs_info->fs_devices->open_devices--; 2833 fs_info->fs_devices->rw_devices--; 2834 fs_info->fs_devices->total_devices--; 2835 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2836 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2837 btrfs_set_super_total_bytes(fs_info->super_copy, 2838 orig_super_total_bytes); 2839 btrfs_set_super_num_devices(fs_info->super_copy, 2840 orig_super_num_devices); 2841 mutex_unlock(&fs_info->chunk_mutex); 2842 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2843 error_trans: 2844 if (seeding_dev) 2845 btrfs_set_sb_rdonly(sb); 2846 if (trans) 2847 btrfs_end_transaction(trans); 2848 error_free_zone: 2849 btrfs_destroy_dev_zone_info(device); 2850 error_free_device: 2851 btrfs_free_device(device); 2852 error: 2853 blkdev_put(bdev, FMODE_EXCL); 2854 if (locked) { 2855 mutex_unlock(&uuid_mutex); 2856 up_write(&sb->s_umount); 2857 } 2858 return ret; 2859 } 2860 2861 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2862 struct btrfs_device *device) 2863 { 2864 int ret; 2865 struct btrfs_path *path; 2866 struct btrfs_root *root = device->fs_info->chunk_root; 2867 struct btrfs_dev_item *dev_item; 2868 struct extent_buffer *leaf; 2869 struct btrfs_key key; 2870 2871 path = btrfs_alloc_path(); 2872 if (!path) 2873 return -ENOMEM; 2874 2875 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2876 key.type = BTRFS_DEV_ITEM_KEY; 2877 key.offset = device->devid; 2878 2879 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2880 if (ret < 0) 2881 goto out; 2882 2883 if (ret > 0) { 2884 ret = -ENOENT; 2885 goto out; 2886 } 2887 2888 leaf = path->nodes[0]; 2889 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2890 2891 btrfs_set_device_id(leaf, dev_item, device->devid); 2892 btrfs_set_device_type(leaf, dev_item, device->type); 2893 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2894 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2895 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2896 btrfs_set_device_total_bytes(leaf, dev_item, 2897 btrfs_device_get_disk_total_bytes(device)); 2898 btrfs_set_device_bytes_used(leaf, dev_item, 2899 btrfs_device_get_bytes_used(device)); 2900 btrfs_mark_buffer_dirty(leaf); 2901 2902 out: 2903 btrfs_free_path(path); 2904 return ret; 2905 } 2906 2907 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2908 struct btrfs_device *device, u64 new_size) 2909 { 2910 struct btrfs_fs_info *fs_info = device->fs_info; 2911 struct btrfs_super_block *super_copy = fs_info->super_copy; 2912 u64 old_total; 2913 u64 diff; 2914 int ret; 2915 2916 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2917 return -EACCES; 2918 2919 new_size = round_down(new_size, fs_info->sectorsize); 2920 2921 mutex_lock(&fs_info->chunk_mutex); 2922 old_total = btrfs_super_total_bytes(super_copy); 2923 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2924 2925 if (new_size <= device->total_bytes || 2926 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2927 mutex_unlock(&fs_info->chunk_mutex); 2928 return -EINVAL; 2929 } 2930 2931 btrfs_set_super_total_bytes(super_copy, 2932 round_down(old_total + diff, fs_info->sectorsize)); 2933 device->fs_devices->total_rw_bytes += diff; 2934 2935 btrfs_device_set_total_bytes(device, new_size); 2936 btrfs_device_set_disk_total_bytes(device, new_size); 2937 btrfs_clear_space_info_full(device->fs_info); 2938 if (list_empty(&device->post_commit_list)) 2939 list_add_tail(&device->post_commit_list, 2940 &trans->transaction->dev_update_list); 2941 mutex_unlock(&fs_info->chunk_mutex); 2942 2943 btrfs_reserve_chunk_metadata(trans, false); 2944 ret = btrfs_update_device(trans, device); 2945 btrfs_trans_release_chunk_metadata(trans); 2946 2947 return ret; 2948 } 2949 2950 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2951 { 2952 struct btrfs_fs_info *fs_info = trans->fs_info; 2953 struct btrfs_root *root = fs_info->chunk_root; 2954 int ret; 2955 struct btrfs_path *path; 2956 struct btrfs_key key; 2957 2958 path = btrfs_alloc_path(); 2959 if (!path) 2960 return -ENOMEM; 2961 2962 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2963 key.offset = chunk_offset; 2964 key.type = BTRFS_CHUNK_ITEM_KEY; 2965 2966 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2967 if (ret < 0) 2968 goto out; 2969 else if (ret > 0) { /* Logic error or corruption */ 2970 btrfs_handle_fs_error(fs_info, -ENOENT, 2971 "Failed lookup while freeing chunk."); 2972 ret = -ENOENT; 2973 goto out; 2974 } 2975 2976 ret = btrfs_del_item(trans, root, path); 2977 if (ret < 0) 2978 btrfs_handle_fs_error(fs_info, ret, 2979 "Failed to delete chunk item."); 2980 out: 2981 btrfs_free_path(path); 2982 return ret; 2983 } 2984 2985 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2986 { 2987 struct btrfs_super_block *super_copy = fs_info->super_copy; 2988 struct btrfs_disk_key *disk_key; 2989 struct btrfs_chunk *chunk; 2990 u8 *ptr; 2991 int ret = 0; 2992 u32 num_stripes; 2993 u32 array_size; 2994 u32 len = 0; 2995 u32 cur; 2996 struct btrfs_key key; 2997 2998 lockdep_assert_held(&fs_info->chunk_mutex); 2999 array_size = btrfs_super_sys_array_size(super_copy); 3000 3001 ptr = super_copy->sys_chunk_array; 3002 cur = 0; 3003 3004 while (cur < array_size) { 3005 disk_key = (struct btrfs_disk_key *)ptr; 3006 btrfs_disk_key_to_cpu(&key, disk_key); 3007 3008 len = sizeof(*disk_key); 3009 3010 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 3011 chunk = (struct btrfs_chunk *)(ptr + len); 3012 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 3013 len += btrfs_chunk_item_size(num_stripes); 3014 } else { 3015 ret = -EIO; 3016 break; 3017 } 3018 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 3019 key.offset == chunk_offset) { 3020 memmove(ptr, ptr + len, array_size - (cur + len)); 3021 array_size -= len; 3022 btrfs_set_super_sys_array_size(super_copy, array_size); 3023 } else { 3024 ptr += len; 3025 cur += len; 3026 } 3027 } 3028 return ret; 3029 } 3030 3031 /* 3032 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 3033 * @logical: Logical block offset in bytes. 3034 * @length: Length of extent in bytes. 3035 * 3036 * Return: Chunk mapping or ERR_PTR. 3037 */ 3038 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 3039 u64 logical, u64 length) 3040 { 3041 struct extent_map_tree *em_tree; 3042 struct extent_map *em; 3043 3044 em_tree = &fs_info->mapping_tree; 3045 read_lock(&em_tree->lock); 3046 em = lookup_extent_mapping(em_tree, logical, length); 3047 read_unlock(&em_tree->lock); 3048 3049 if (!em) { 3050 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 3051 logical, length); 3052 return ERR_PTR(-EINVAL); 3053 } 3054 3055 if (em->start > logical || em->start + em->len < logical) { 3056 btrfs_crit(fs_info, 3057 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3058 logical, length, em->start, em->start + em->len); 3059 free_extent_map(em); 3060 return ERR_PTR(-EINVAL); 3061 } 3062 3063 /* callers are responsible for dropping em's ref. */ 3064 return em; 3065 } 3066 3067 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3068 struct map_lookup *map, u64 chunk_offset) 3069 { 3070 int i; 3071 3072 /* 3073 * Removing chunk items and updating the device items in the chunks btree 3074 * requires holding the chunk_mutex. 3075 * See the comment at btrfs_chunk_alloc() for the details. 3076 */ 3077 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3078 3079 for (i = 0; i < map->num_stripes; i++) { 3080 int ret; 3081 3082 ret = btrfs_update_device(trans, map->stripes[i].dev); 3083 if (ret) 3084 return ret; 3085 } 3086 3087 return btrfs_free_chunk(trans, chunk_offset); 3088 } 3089 3090 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3091 { 3092 struct btrfs_fs_info *fs_info = trans->fs_info; 3093 struct extent_map *em; 3094 struct map_lookup *map; 3095 u64 dev_extent_len = 0; 3096 int i, ret = 0; 3097 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3098 3099 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3100 if (IS_ERR(em)) { 3101 /* 3102 * This is a logic error, but we don't want to just rely on the 3103 * user having built with ASSERT enabled, so if ASSERT doesn't 3104 * do anything we still error out. 3105 */ 3106 ASSERT(0); 3107 return PTR_ERR(em); 3108 } 3109 map = em->map_lookup; 3110 3111 /* 3112 * First delete the device extent items from the devices btree. 3113 * We take the device_list_mutex to avoid racing with the finishing phase 3114 * of a device replace operation. See the comment below before acquiring 3115 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3116 * because that can result in a deadlock when deleting the device extent 3117 * items from the devices btree - COWing an extent buffer from the btree 3118 * may result in allocating a new metadata chunk, which would attempt to 3119 * lock again fs_info->chunk_mutex. 3120 */ 3121 mutex_lock(&fs_devices->device_list_mutex); 3122 for (i = 0; i < map->num_stripes; i++) { 3123 struct btrfs_device *device = map->stripes[i].dev; 3124 ret = btrfs_free_dev_extent(trans, device, 3125 map->stripes[i].physical, 3126 &dev_extent_len); 3127 if (ret) { 3128 mutex_unlock(&fs_devices->device_list_mutex); 3129 btrfs_abort_transaction(trans, ret); 3130 goto out; 3131 } 3132 3133 if (device->bytes_used > 0) { 3134 mutex_lock(&fs_info->chunk_mutex); 3135 btrfs_device_set_bytes_used(device, 3136 device->bytes_used - dev_extent_len); 3137 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3138 btrfs_clear_space_info_full(fs_info); 3139 mutex_unlock(&fs_info->chunk_mutex); 3140 } 3141 } 3142 mutex_unlock(&fs_devices->device_list_mutex); 3143 3144 /* 3145 * We acquire fs_info->chunk_mutex for 2 reasons: 3146 * 3147 * 1) Just like with the first phase of the chunk allocation, we must 3148 * reserve system space, do all chunk btree updates and deletions, and 3149 * update the system chunk array in the superblock while holding this 3150 * mutex. This is for similar reasons as explained on the comment at 3151 * the top of btrfs_chunk_alloc(); 3152 * 3153 * 2) Prevent races with the final phase of a device replace operation 3154 * that replaces the device object associated with the map's stripes, 3155 * because the device object's id can change at any time during that 3156 * final phase of the device replace operation 3157 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3158 * replaced device and then see it with an ID of 3159 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3160 * the device item, which does not exists on the chunk btree. 3161 * The finishing phase of device replace acquires both the 3162 * device_list_mutex and the chunk_mutex, in that order, so we are 3163 * safe by just acquiring the chunk_mutex. 3164 */ 3165 trans->removing_chunk = true; 3166 mutex_lock(&fs_info->chunk_mutex); 3167 3168 check_system_chunk(trans, map->type); 3169 3170 ret = remove_chunk_item(trans, map, chunk_offset); 3171 /* 3172 * Normally we should not get -ENOSPC since we reserved space before 3173 * through the call to check_system_chunk(). 3174 * 3175 * Despite our system space_info having enough free space, we may not 3176 * be able to allocate extents from its block groups, because all have 3177 * an incompatible profile, which will force us to allocate a new system 3178 * block group with the right profile, or right after we called 3179 * check_system_space() above, a scrub turned the only system block group 3180 * with enough free space into RO mode. 3181 * This is explained with more detail at do_chunk_alloc(). 3182 * 3183 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3184 */ 3185 if (ret == -ENOSPC) { 3186 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3187 struct btrfs_block_group *sys_bg; 3188 3189 sys_bg = btrfs_create_chunk(trans, sys_flags); 3190 if (IS_ERR(sys_bg)) { 3191 ret = PTR_ERR(sys_bg); 3192 btrfs_abort_transaction(trans, ret); 3193 goto out; 3194 } 3195 3196 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3197 if (ret) { 3198 btrfs_abort_transaction(trans, ret); 3199 goto out; 3200 } 3201 3202 ret = remove_chunk_item(trans, map, chunk_offset); 3203 if (ret) { 3204 btrfs_abort_transaction(trans, ret); 3205 goto out; 3206 } 3207 } else if (ret) { 3208 btrfs_abort_transaction(trans, ret); 3209 goto out; 3210 } 3211 3212 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3213 3214 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3215 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3216 if (ret) { 3217 btrfs_abort_transaction(trans, ret); 3218 goto out; 3219 } 3220 } 3221 3222 mutex_unlock(&fs_info->chunk_mutex); 3223 trans->removing_chunk = false; 3224 3225 /* 3226 * We are done with chunk btree updates and deletions, so release the 3227 * system space we previously reserved (with check_system_chunk()). 3228 */ 3229 btrfs_trans_release_chunk_metadata(trans); 3230 3231 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3232 if (ret) { 3233 btrfs_abort_transaction(trans, ret); 3234 goto out; 3235 } 3236 3237 out: 3238 if (trans->removing_chunk) { 3239 mutex_unlock(&fs_info->chunk_mutex); 3240 trans->removing_chunk = false; 3241 } 3242 /* once for us */ 3243 free_extent_map(em); 3244 return ret; 3245 } 3246 3247 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3248 { 3249 struct btrfs_root *root = fs_info->chunk_root; 3250 struct btrfs_trans_handle *trans; 3251 struct btrfs_block_group *block_group; 3252 u64 length; 3253 int ret; 3254 3255 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3256 btrfs_err(fs_info, 3257 "relocate: not supported on extent tree v2 yet"); 3258 return -EINVAL; 3259 } 3260 3261 /* 3262 * Prevent races with automatic removal of unused block groups. 3263 * After we relocate and before we remove the chunk with offset 3264 * chunk_offset, automatic removal of the block group can kick in, 3265 * resulting in a failure when calling btrfs_remove_chunk() below. 3266 * 3267 * Make sure to acquire this mutex before doing a tree search (dev 3268 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3269 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3270 * we release the path used to search the chunk/dev tree and before 3271 * the current task acquires this mutex and calls us. 3272 */ 3273 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3274 3275 /* step one, relocate all the extents inside this chunk */ 3276 btrfs_scrub_pause(fs_info); 3277 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3278 btrfs_scrub_continue(fs_info); 3279 if (ret) { 3280 /* 3281 * If we had a transaction abort, stop all running scrubs. 3282 * See transaction.c:cleanup_transaction() why we do it here. 3283 */ 3284 if (BTRFS_FS_ERROR(fs_info)) 3285 btrfs_scrub_cancel(fs_info); 3286 return ret; 3287 } 3288 3289 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3290 if (!block_group) 3291 return -ENOENT; 3292 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3293 length = block_group->length; 3294 btrfs_put_block_group(block_group); 3295 3296 /* 3297 * On a zoned file system, discard the whole block group, this will 3298 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3299 * resetting the zone fails, don't treat it as a fatal problem from the 3300 * filesystem's point of view. 3301 */ 3302 if (btrfs_is_zoned(fs_info)) { 3303 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3304 if (ret) 3305 btrfs_info(fs_info, 3306 "failed to reset zone %llu after relocation", 3307 chunk_offset); 3308 } 3309 3310 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3311 chunk_offset); 3312 if (IS_ERR(trans)) { 3313 ret = PTR_ERR(trans); 3314 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3315 return ret; 3316 } 3317 3318 /* 3319 * step two, delete the device extents and the 3320 * chunk tree entries 3321 */ 3322 ret = btrfs_remove_chunk(trans, chunk_offset); 3323 btrfs_end_transaction(trans); 3324 return ret; 3325 } 3326 3327 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3328 { 3329 struct btrfs_root *chunk_root = fs_info->chunk_root; 3330 struct btrfs_path *path; 3331 struct extent_buffer *leaf; 3332 struct btrfs_chunk *chunk; 3333 struct btrfs_key key; 3334 struct btrfs_key found_key; 3335 u64 chunk_type; 3336 bool retried = false; 3337 int failed = 0; 3338 int ret; 3339 3340 path = btrfs_alloc_path(); 3341 if (!path) 3342 return -ENOMEM; 3343 3344 again: 3345 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3346 key.offset = (u64)-1; 3347 key.type = BTRFS_CHUNK_ITEM_KEY; 3348 3349 while (1) { 3350 mutex_lock(&fs_info->reclaim_bgs_lock); 3351 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3352 if (ret < 0) { 3353 mutex_unlock(&fs_info->reclaim_bgs_lock); 3354 goto error; 3355 } 3356 BUG_ON(ret == 0); /* Corruption */ 3357 3358 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3359 key.type); 3360 if (ret) 3361 mutex_unlock(&fs_info->reclaim_bgs_lock); 3362 if (ret < 0) 3363 goto error; 3364 if (ret > 0) 3365 break; 3366 3367 leaf = path->nodes[0]; 3368 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3369 3370 chunk = btrfs_item_ptr(leaf, path->slots[0], 3371 struct btrfs_chunk); 3372 chunk_type = btrfs_chunk_type(leaf, chunk); 3373 btrfs_release_path(path); 3374 3375 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3376 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3377 if (ret == -ENOSPC) 3378 failed++; 3379 else 3380 BUG_ON(ret); 3381 } 3382 mutex_unlock(&fs_info->reclaim_bgs_lock); 3383 3384 if (found_key.offset == 0) 3385 break; 3386 key.offset = found_key.offset - 1; 3387 } 3388 ret = 0; 3389 if (failed && !retried) { 3390 failed = 0; 3391 retried = true; 3392 goto again; 3393 } else if (WARN_ON(failed && retried)) { 3394 ret = -ENOSPC; 3395 } 3396 error: 3397 btrfs_free_path(path); 3398 return ret; 3399 } 3400 3401 /* 3402 * return 1 : allocate a data chunk successfully, 3403 * return <0: errors during allocating a data chunk, 3404 * return 0 : no need to allocate a data chunk. 3405 */ 3406 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3407 u64 chunk_offset) 3408 { 3409 struct btrfs_block_group *cache; 3410 u64 bytes_used; 3411 u64 chunk_type; 3412 3413 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3414 ASSERT(cache); 3415 chunk_type = cache->flags; 3416 btrfs_put_block_group(cache); 3417 3418 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3419 return 0; 3420 3421 spin_lock(&fs_info->data_sinfo->lock); 3422 bytes_used = fs_info->data_sinfo->bytes_used; 3423 spin_unlock(&fs_info->data_sinfo->lock); 3424 3425 if (!bytes_used) { 3426 struct btrfs_trans_handle *trans; 3427 int ret; 3428 3429 trans = btrfs_join_transaction(fs_info->tree_root); 3430 if (IS_ERR(trans)) 3431 return PTR_ERR(trans); 3432 3433 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3434 btrfs_end_transaction(trans); 3435 if (ret < 0) 3436 return ret; 3437 return 1; 3438 } 3439 3440 return 0; 3441 } 3442 3443 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3444 struct btrfs_balance_control *bctl) 3445 { 3446 struct btrfs_root *root = fs_info->tree_root; 3447 struct btrfs_trans_handle *trans; 3448 struct btrfs_balance_item *item; 3449 struct btrfs_disk_balance_args disk_bargs; 3450 struct btrfs_path *path; 3451 struct extent_buffer *leaf; 3452 struct btrfs_key key; 3453 int ret, err; 3454 3455 path = btrfs_alloc_path(); 3456 if (!path) 3457 return -ENOMEM; 3458 3459 trans = btrfs_start_transaction(root, 0); 3460 if (IS_ERR(trans)) { 3461 btrfs_free_path(path); 3462 return PTR_ERR(trans); 3463 } 3464 3465 key.objectid = BTRFS_BALANCE_OBJECTID; 3466 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3467 key.offset = 0; 3468 3469 ret = btrfs_insert_empty_item(trans, root, path, &key, 3470 sizeof(*item)); 3471 if (ret) 3472 goto out; 3473 3474 leaf = path->nodes[0]; 3475 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3476 3477 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3478 3479 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3480 btrfs_set_balance_data(leaf, item, &disk_bargs); 3481 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3482 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3483 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3484 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3485 3486 btrfs_set_balance_flags(leaf, item, bctl->flags); 3487 3488 btrfs_mark_buffer_dirty(leaf); 3489 out: 3490 btrfs_free_path(path); 3491 err = btrfs_commit_transaction(trans); 3492 if (err && !ret) 3493 ret = err; 3494 return ret; 3495 } 3496 3497 static int del_balance_item(struct btrfs_fs_info *fs_info) 3498 { 3499 struct btrfs_root *root = fs_info->tree_root; 3500 struct btrfs_trans_handle *trans; 3501 struct btrfs_path *path; 3502 struct btrfs_key key; 3503 int ret, err; 3504 3505 path = btrfs_alloc_path(); 3506 if (!path) 3507 return -ENOMEM; 3508 3509 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3510 if (IS_ERR(trans)) { 3511 btrfs_free_path(path); 3512 return PTR_ERR(trans); 3513 } 3514 3515 key.objectid = BTRFS_BALANCE_OBJECTID; 3516 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3517 key.offset = 0; 3518 3519 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3520 if (ret < 0) 3521 goto out; 3522 if (ret > 0) { 3523 ret = -ENOENT; 3524 goto out; 3525 } 3526 3527 ret = btrfs_del_item(trans, root, path); 3528 out: 3529 btrfs_free_path(path); 3530 err = btrfs_commit_transaction(trans); 3531 if (err && !ret) 3532 ret = err; 3533 return ret; 3534 } 3535 3536 /* 3537 * This is a heuristic used to reduce the number of chunks balanced on 3538 * resume after balance was interrupted. 3539 */ 3540 static void update_balance_args(struct btrfs_balance_control *bctl) 3541 { 3542 /* 3543 * Turn on soft mode for chunk types that were being converted. 3544 */ 3545 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3546 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3547 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3548 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3549 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3550 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3551 3552 /* 3553 * Turn on usage filter if is not already used. The idea is 3554 * that chunks that we have already balanced should be 3555 * reasonably full. Don't do it for chunks that are being 3556 * converted - that will keep us from relocating unconverted 3557 * (albeit full) chunks. 3558 */ 3559 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3560 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3561 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3562 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3563 bctl->data.usage = 90; 3564 } 3565 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3566 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3567 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3568 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3569 bctl->sys.usage = 90; 3570 } 3571 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3572 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3573 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3574 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3575 bctl->meta.usage = 90; 3576 } 3577 } 3578 3579 /* 3580 * Clear the balance status in fs_info and delete the balance item from disk. 3581 */ 3582 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3583 { 3584 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3585 int ret; 3586 3587 BUG_ON(!fs_info->balance_ctl); 3588 3589 spin_lock(&fs_info->balance_lock); 3590 fs_info->balance_ctl = NULL; 3591 spin_unlock(&fs_info->balance_lock); 3592 3593 kfree(bctl); 3594 ret = del_balance_item(fs_info); 3595 if (ret) 3596 btrfs_handle_fs_error(fs_info, ret, NULL); 3597 } 3598 3599 /* 3600 * Balance filters. Return 1 if chunk should be filtered out 3601 * (should not be balanced). 3602 */ 3603 static int chunk_profiles_filter(u64 chunk_type, 3604 struct btrfs_balance_args *bargs) 3605 { 3606 chunk_type = chunk_to_extended(chunk_type) & 3607 BTRFS_EXTENDED_PROFILE_MASK; 3608 3609 if (bargs->profiles & chunk_type) 3610 return 0; 3611 3612 return 1; 3613 } 3614 3615 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3616 struct btrfs_balance_args *bargs) 3617 { 3618 struct btrfs_block_group *cache; 3619 u64 chunk_used; 3620 u64 user_thresh_min; 3621 u64 user_thresh_max; 3622 int ret = 1; 3623 3624 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3625 chunk_used = cache->used; 3626 3627 if (bargs->usage_min == 0) 3628 user_thresh_min = 0; 3629 else 3630 user_thresh_min = mult_perc(cache->length, bargs->usage_min); 3631 3632 if (bargs->usage_max == 0) 3633 user_thresh_max = 1; 3634 else if (bargs->usage_max > 100) 3635 user_thresh_max = cache->length; 3636 else 3637 user_thresh_max = mult_perc(cache->length, bargs->usage_max); 3638 3639 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3640 ret = 0; 3641 3642 btrfs_put_block_group(cache); 3643 return ret; 3644 } 3645 3646 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3647 u64 chunk_offset, struct btrfs_balance_args *bargs) 3648 { 3649 struct btrfs_block_group *cache; 3650 u64 chunk_used, user_thresh; 3651 int ret = 1; 3652 3653 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3654 chunk_used = cache->used; 3655 3656 if (bargs->usage_min == 0) 3657 user_thresh = 1; 3658 else if (bargs->usage > 100) 3659 user_thresh = cache->length; 3660 else 3661 user_thresh = mult_perc(cache->length, bargs->usage); 3662 3663 if (chunk_used < user_thresh) 3664 ret = 0; 3665 3666 btrfs_put_block_group(cache); 3667 return ret; 3668 } 3669 3670 static int chunk_devid_filter(struct extent_buffer *leaf, 3671 struct btrfs_chunk *chunk, 3672 struct btrfs_balance_args *bargs) 3673 { 3674 struct btrfs_stripe *stripe; 3675 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3676 int i; 3677 3678 for (i = 0; i < num_stripes; i++) { 3679 stripe = btrfs_stripe_nr(chunk, i); 3680 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3681 return 0; 3682 } 3683 3684 return 1; 3685 } 3686 3687 static u64 calc_data_stripes(u64 type, int num_stripes) 3688 { 3689 const int index = btrfs_bg_flags_to_raid_index(type); 3690 const int ncopies = btrfs_raid_array[index].ncopies; 3691 const int nparity = btrfs_raid_array[index].nparity; 3692 3693 return (num_stripes - nparity) / ncopies; 3694 } 3695 3696 /* [pstart, pend) */ 3697 static int chunk_drange_filter(struct extent_buffer *leaf, 3698 struct btrfs_chunk *chunk, 3699 struct btrfs_balance_args *bargs) 3700 { 3701 struct btrfs_stripe *stripe; 3702 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3703 u64 stripe_offset; 3704 u64 stripe_length; 3705 u64 type; 3706 int factor; 3707 int i; 3708 3709 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3710 return 0; 3711 3712 type = btrfs_chunk_type(leaf, chunk); 3713 factor = calc_data_stripes(type, num_stripes); 3714 3715 for (i = 0; i < num_stripes; i++) { 3716 stripe = btrfs_stripe_nr(chunk, i); 3717 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3718 continue; 3719 3720 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3721 stripe_length = btrfs_chunk_length(leaf, chunk); 3722 stripe_length = div_u64(stripe_length, factor); 3723 3724 if (stripe_offset < bargs->pend && 3725 stripe_offset + stripe_length > bargs->pstart) 3726 return 0; 3727 } 3728 3729 return 1; 3730 } 3731 3732 /* [vstart, vend) */ 3733 static int chunk_vrange_filter(struct extent_buffer *leaf, 3734 struct btrfs_chunk *chunk, 3735 u64 chunk_offset, 3736 struct btrfs_balance_args *bargs) 3737 { 3738 if (chunk_offset < bargs->vend && 3739 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3740 /* at least part of the chunk is inside this vrange */ 3741 return 0; 3742 3743 return 1; 3744 } 3745 3746 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3747 struct btrfs_chunk *chunk, 3748 struct btrfs_balance_args *bargs) 3749 { 3750 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3751 3752 if (bargs->stripes_min <= num_stripes 3753 && num_stripes <= bargs->stripes_max) 3754 return 0; 3755 3756 return 1; 3757 } 3758 3759 static int chunk_soft_convert_filter(u64 chunk_type, 3760 struct btrfs_balance_args *bargs) 3761 { 3762 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3763 return 0; 3764 3765 chunk_type = chunk_to_extended(chunk_type) & 3766 BTRFS_EXTENDED_PROFILE_MASK; 3767 3768 if (bargs->target == chunk_type) 3769 return 1; 3770 3771 return 0; 3772 } 3773 3774 static int should_balance_chunk(struct extent_buffer *leaf, 3775 struct btrfs_chunk *chunk, u64 chunk_offset) 3776 { 3777 struct btrfs_fs_info *fs_info = leaf->fs_info; 3778 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3779 struct btrfs_balance_args *bargs = NULL; 3780 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3781 3782 /* type filter */ 3783 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3784 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3785 return 0; 3786 } 3787 3788 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3789 bargs = &bctl->data; 3790 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3791 bargs = &bctl->sys; 3792 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3793 bargs = &bctl->meta; 3794 3795 /* profiles filter */ 3796 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3797 chunk_profiles_filter(chunk_type, bargs)) { 3798 return 0; 3799 } 3800 3801 /* usage filter */ 3802 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3803 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3804 return 0; 3805 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3806 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3807 return 0; 3808 } 3809 3810 /* devid filter */ 3811 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3812 chunk_devid_filter(leaf, chunk, bargs)) { 3813 return 0; 3814 } 3815 3816 /* drange filter, makes sense only with devid filter */ 3817 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3818 chunk_drange_filter(leaf, chunk, bargs)) { 3819 return 0; 3820 } 3821 3822 /* vrange filter */ 3823 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3824 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3825 return 0; 3826 } 3827 3828 /* stripes filter */ 3829 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3830 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3831 return 0; 3832 } 3833 3834 /* soft profile changing mode */ 3835 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3836 chunk_soft_convert_filter(chunk_type, bargs)) { 3837 return 0; 3838 } 3839 3840 /* 3841 * limited by count, must be the last filter 3842 */ 3843 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3844 if (bargs->limit == 0) 3845 return 0; 3846 else 3847 bargs->limit--; 3848 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3849 /* 3850 * Same logic as the 'limit' filter; the minimum cannot be 3851 * determined here because we do not have the global information 3852 * about the count of all chunks that satisfy the filters. 3853 */ 3854 if (bargs->limit_max == 0) 3855 return 0; 3856 else 3857 bargs->limit_max--; 3858 } 3859 3860 return 1; 3861 } 3862 3863 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3864 { 3865 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3866 struct btrfs_root *chunk_root = fs_info->chunk_root; 3867 u64 chunk_type; 3868 struct btrfs_chunk *chunk; 3869 struct btrfs_path *path = NULL; 3870 struct btrfs_key key; 3871 struct btrfs_key found_key; 3872 struct extent_buffer *leaf; 3873 int slot; 3874 int ret; 3875 int enospc_errors = 0; 3876 bool counting = true; 3877 /* The single value limit and min/max limits use the same bytes in the */ 3878 u64 limit_data = bctl->data.limit; 3879 u64 limit_meta = bctl->meta.limit; 3880 u64 limit_sys = bctl->sys.limit; 3881 u32 count_data = 0; 3882 u32 count_meta = 0; 3883 u32 count_sys = 0; 3884 int chunk_reserved = 0; 3885 3886 path = btrfs_alloc_path(); 3887 if (!path) { 3888 ret = -ENOMEM; 3889 goto error; 3890 } 3891 3892 /* zero out stat counters */ 3893 spin_lock(&fs_info->balance_lock); 3894 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3895 spin_unlock(&fs_info->balance_lock); 3896 again: 3897 if (!counting) { 3898 /* 3899 * The single value limit and min/max limits use the same bytes 3900 * in the 3901 */ 3902 bctl->data.limit = limit_data; 3903 bctl->meta.limit = limit_meta; 3904 bctl->sys.limit = limit_sys; 3905 } 3906 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3907 key.offset = (u64)-1; 3908 key.type = BTRFS_CHUNK_ITEM_KEY; 3909 3910 while (1) { 3911 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3912 atomic_read(&fs_info->balance_cancel_req)) { 3913 ret = -ECANCELED; 3914 goto error; 3915 } 3916 3917 mutex_lock(&fs_info->reclaim_bgs_lock); 3918 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3919 if (ret < 0) { 3920 mutex_unlock(&fs_info->reclaim_bgs_lock); 3921 goto error; 3922 } 3923 3924 /* 3925 * this shouldn't happen, it means the last relocate 3926 * failed 3927 */ 3928 if (ret == 0) 3929 BUG(); /* FIXME break ? */ 3930 3931 ret = btrfs_previous_item(chunk_root, path, 0, 3932 BTRFS_CHUNK_ITEM_KEY); 3933 if (ret) { 3934 mutex_unlock(&fs_info->reclaim_bgs_lock); 3935 ret = 0; 3936 break; 3937 } 3938 3939 leaf = path->nodes[0]; 3940 slot = path->slots[0]; 3941 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3942 3943 if (found_key.objectid != key.objectid) { 3944 mutex_unlock(&fs_info->reclaim_bgs_lock); 3945 break; 3946 } 3947 3948 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3949 chunk_type = btrfs_chunk_type(leaf, chunk); 3950 3951 if (!counting) { 3952 spin_lock(&fs_info->balance_lock); 3953 bctl->stat.considered++; 3954 spin_unlock(&fs_info->balance_lock); 3955 } 3956 3957 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3958 3959 btrfs_release_path(path); 3960 if (!ret) { 3961 mutex_unlock(&fs_info->reclaim_bgs_lock); 3962 goto loop; 3963 } 3964 3965 if (counting) { 3966 mutex_unlock(&fs_info->reclaim_bgs_lock); 3967 spin_lock(&fs_info->balance_lock); 3968 bctl->stat.expected++; 3969 spin_unlock(&fs_info->balance_lock); 3970 3971 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3972 count_data++; 3973 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3974 count_sys++; 3975 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3976 count_meta++; 3977 3978 goto loop; 3979 } 3980 3981 /* 3982 * Apply limit_min filter, no need to check if the LIMITS 3983 * filter is used, limit_min is 0 by default 3984 */ 3985 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3986 count_data < bctl->data.limit_min) 3987 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3988 count_meta < bctl->meta.limit_min) 3989 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3990 count_sys < bctl->sys.limit_min)) { 3991 mutex_unlock(&fs_info->reclaim_bgs_lock); 3992 goto loop; 3993 } 3994 3995 if (!chunk_reserved) { 3996 /* 3997 * We may be relocating the only data chunk we have, 3998 * which could potentially end up with losing data's 3999 * raid profile, so lets allocate an empty one in 4000 * advance. 4001 */ 4002 ret = btrfs_may_alloc_data_chunk(fs_info, 4003 found_key.offset); 4004 if (ret < 0) { 4005 mutex_unlock(&fs_info->reclaim_bgs_lock); 4006 goto error; 4007 } else if (ret == 1) { 4008 chunk_reserved = 1; 4009 } 4010 } 4011 4012 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 4013 mutex_unlock(&fs_info->reclaim_bgs_lock); 4014 if (ret == -ENOSPC) { 4015 enospc_errors++; 4016 } else if (ret == -ETXTBSY) { 4017 btrfs_info(fs_info, 4018 "skipping relocation of block group %llu due to active swapfile", 4019 found_key.offset); 4020 ret = 0; 4021 } else if (ret) { 4022 goto error; 4023 } else { 4024 spin_lock(&fs_info->balance_lock); 4025 bctl->stat.completed++; 4026 spin_unlock(&fs_info->balance_lock); 4027 } 4028 loop: 4029 if (found_key.offset == 0) 4030 break; 4031 key.offset = found_key.offset - 1; 4032 } 4033 4034 if (counting) { 4035 btrfs_release_path(path); 4036 counting = false; 4037 goto again; 4038 } 4039 error: 4040 btrfs_free_path(path); 4041 if (enospc_errors) { 4042 btrfs_info(fs_info, "%d enospc errors during balance", 4043 enospc_errors); 4044 if (!ret) 4045 ret = -ENOSPC; 4046 } 4047 4048 return ret; 4049 } 4050 4051 /* 4052 * See if a given profile is valid and reduced. 4053 * 4054 * @flags: profile to validate 4055 * @extended: if true @flags is treated as an extended profile 4056 */ 4057 static int alloc_profile_is_valid(u64 flags, int extended) 4058 { 4059 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 4060 BTRFS_BLOCK_GROUP_PROFILE_MASK); 4061 4062 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 4063 4064 /* 1) check that all other bits are zeroed */ 4065 if (flags & ~mask) 4066 return 0; 4067 4068 /* 2) see if profile is reduced */ 4069 if (flags == 0) 4070 return !extended; /* "0" is valid for usual profiles */ 4071 4072 return has_single_bit_set(flags); 4073 } 4074 4075 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4076 { 4077 /* cancel requested || normal exit path */ 4078 return atomic_read(&fs_info->balance_cancel_req) || 4079 (atomic_read(&fs_info->balance_pause_req) == 0 && 4080 atomic_read(&fs_info->balance_cancel_req) == 0); 4081 } 4082 4083 /* 4084 * Validate target profile against allowed profiles and return true if it's OK. 4085 * Otherwise print the error message and return false. 4086 */ 4087 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4088 const struct btrfs_balance_args *bargs, 4089 u64 allowed, const char *type) 4090 { 4091 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4092 return true; 4093 4094 /* Profile is valid and does not have bits outside of the allowed set */ 4095 if (alloc_profile_is_valid(bargs->target, 1) && 4096 (bargs->target & ~allowed) == 0) 4097 return true; 4098 4099 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4100 type, btrfs_bg_type_to_raid_name(bargs->target)); 4101 return false; 4102 } 4103 4104 /* 4105 * Fill @buf with textual description of balance filter flags @bargs, up to 4106 * @size_buf including the terminating null. The output may be trimmed if it 4107 * does not fit into the provided buffer. 4108 */ 4109 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4110 u32 size_buf) 4111 { 4112 int ret; 4113 u32 size_bp = size_buf; 4114 char *bp = buf; 4115 u64 flags = bargs->flags; 4116 char tmp_buf[128] = {'\0'}; 4117 4118 if (!flags) 4119 return; 4120 4121 #define CHECK_APPEND_NOARG(a) \ 4122 do { \ 4123 ret = snprintf(bp, size_bp, (a)); \ 4124 if (ret < 0 || ret >= size_bp) \ 4125 goto out_overflow; \ 4126 size_bp -= ret; \ 4127 bp += ret; \ 4128 } while (0) 4129 4130 #define CHECK_APPEND_1ARG(a, v1) \ 4131 do { \ 4132 ret = snprintf(bp, size_bp, (a), (v1)); \ 4133 if (ret < 0 || ret >= size_bp) \ 4134 goto out_overflow; \ 4135 size_bp -= ret; \ 4136 bp += ret; \ 4137 } while (0) 4138 4139 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4140 do { \ 4141 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4142 if (ret < 0 || ret >= size_bp) \ 4143 goto out_overflow; \ 4144 size_bp -= ret; \ 4145 bp += ret; \ 4146 } while (0) 4147 4148 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4149 CHECK_APPEND_1ARG("convert=%s,", 4150 btrfs_bg_type_to_raid_name(bargs->target)); 4151 4152 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4153 CHECK_APPEND_NOARG("soft,"); 4154 4155 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4156 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4157 sizeof(tmp_buf)); 4158 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4159 } 4160 4161 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4162 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4163 4164 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4165 CHECK_APPEND_2ARG("usage=%u..%u,", 4166 bargs->usage_min, bargs->usage_max); 4167 4168 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4169 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4170 4171 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4172 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4173 bargs->pstart, bargs->pend); 4174 4175 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4176 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4177 bargs->vstart, bargs->vend); 4178 4179 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4180 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4181 4182 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4183 CHECK_APPEND_2ARG("limit=%u..%u,", 4184 bargs->limit_min, bargs->limit_max); 4185 4186 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4187 CHECK_APPEND_2ARG("stripes=%u..%u,", 4188 bargs->stripes_min, bargs->stripes_max); 4189 4190 #undef CHECK_APPEND_2ARG 4191 #undef CHECK_APPEND_1ARG 4192 #undef CHECK_APPEND_NOARG 4193 4194 out_overflow: 4195 4196 if (size_bp < size_buf) 4197 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4198 else 4199 buf[0] = '\0'; 4200 } 4201 4202 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4203 { 4204 u32 size_buf = 1024; 4205 char tmp_buf[192] = {'\0'}; 4206 char *buf; 4207 char *bp; 4208 u32 size_bp = size_buf; 4209 int ret; 4210 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4211 4212 buf = kzalloc(size_buf, GFP_KERNEL); 4213 if (!buf) 4214 return; 4215 4216 bp = buf; 4217 4218 #define CHECK_APPEND_1ARG(a, v1) \ 4219 do { \ 4220 ret = snprintf(bp, size_bp, (a), (v1)); \ 4221 if (ret < 0 || ret >= size_bp) \ 4222 goto out_overflow; \ 4223 size_bp -= ret; \ 4224 bp += ret; \ 4225 } while (0) 4226 4227 if (bctl->flags & BTRFS_BALANCE_FORCE) 4228 CHECK_APPEND_1ARG("%s", "-f "); 4229 4230 if (bctl->flags & BTRFS_BALANCE_DATA) { 4231 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4232 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4233 } 4234 4235 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4236 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4237 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4238 } 4239 4240 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4241 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4242 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4243 } 4244 4245 #undef CHECK_APPEND_1ARG 4246 4247 out_overflow: 4248 4249 if (size_bp < size_buf) 4250 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4251 btrfs_info(fs_info, "balance: %s %s", 4252 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4253 "resume" : "start", buf); 4254 4255 kfree(buf); 4256 } 4257 4258 /* 4259 * Should be called with balance mutexe held 4260 */ 4261 int btrfs_balance(struct btrfs_fs_info *fs_info, 4262 struct btrfs_balance_control *bctl, 4263 struct btrfs_ioctl_balance_args *bargs) 4264 { 4265 u64 meta_target, data_target; 4266 u64 allowed; 4267 int mixed = 0; 4268 int ret; 4269 u64 num_devices; 4270 unsigned seq; 4271 bool reducing_redundancy; 4272 int i; 4273 4274 if (btrfs_fs_closing(fs_info) || 4275 atomic_read(&fs_info->balance_pause_req) || 4276 btrfs_should_cancel_balance(fs_info)) { 4277 ret = -EINVAL; 4278 goto out; 4279 } 4280 4281 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4282 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4283 mixed = 1; 4284 4285 /* 4286 * In case of mixed groups both data and meta should be picked, 4287 * and identical options should be given for both of them. 4288 */ 4289 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4290 if (mixed && (bctl->flags & allowed)) { 4291 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4292 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4293 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4294 btrfs_err(fs_info, 4295 "balance: mixed groups data and metadata options must be the same"); 4296 ret = -EINVAL; 4297 goto out; 4298 } 4299 } 4300 4301 /* 4302 * rw_devices will not change at the moment, device add/delete/replace 4303 * are exclusive 4304 */ 4305 num_devices = fs_info->fs_devices->rw_devices; 4306 4307 /* 4308 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4309 * special bit for it, to make it easier to distinguish. Thus we need 4310 * to set it manually, or balance would refuse the profile. 4311 */ 4312 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4313 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4314 if (num_devices >= btrfs_raid_array[i].devs_min) 4315 allowed |= btrfs_raid_array[i].bg_flag; 4316 4317 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4318 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4319 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4320 ret = -EINVAL; 4321 goto out; 4322 } 4323 4324 /* 4325 * Allow to reduce metadata or system integrity only if force set for 4326 * profiles with redundancy (copies, parity) 4327 */ 4328 allowed = 0; 4329 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4330 if (btrfs_raid_array[i].ncopies >= 2 || 4331 btrfs_raid_array[i].tolerated_failures >= 1) 4332 allowed |= btrfs_raid_array[i].bg_flag; 4333 } 4334 do { 4335 seq = read_seqbegin(&fs_info->profiles_lock); 4336 4337 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4338 (fs_info->avail_system_alloc_bits & allowed) && 4339 !(bctl->sys.target & allowed)) || 4340 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4341 (fs_info->avail_metadata_alloc_bits & allowed) && 4342 !(bctl->meta.target & allowed))) 4343 reducing_redundancy = true; 4344 else 4345 reducing_redundancy = false; 4346 4347 /* if we're not converting, the target field is uninitialized */ 4348 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4349 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4350 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4351 bctl->data.target : fs_info->avail_data_alloc_bits; 4352 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4353 4354 if (reducing_redundancy) { 4355 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4356 btrfs_info(fs_info, 4357 "balance: force reducing metadata redundancy"); 4358 } else { 4359 btrfs_err(fs_info, 4360 "balance: reduces metadata redundancy, use --force if you want this"); 4361 ret = -EINVAL; 4362 goto out; 4363 } 4364 } 4365 4366 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4367 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4368 btrfs_warn(fs_info, 4369 "balance: metadata profile %s has lower redundancy than data profile %s", 4370 btrfs_bg_type_to_raid_name(meta_target), 4371 btrfs_bg_type_to_raid_name(data_target)); 4372 } 4373 4374 ret = insert_balance_item(fs_info, bctl); 4375 if (ret && ret != -EEXIST) 4376 goto out; 4377 4378 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4379 BUG_ON(ret == -EEXIST); 4380 BUG_ON(fs_info->balance_ctl); 4381 spin_lock(&fs_info->balance_lock); 4382 fs_info->balance_ctl = bctl; 4383 spin_unlock(&fs_info->balance_lock); 4384 } else { 4385 BUG_ON(ret != -EEXIST); 4386 spin_lock(&fs_info->balance_lock); 4387 update_balance_args(bctl); 4388 spin_unlock(&fs_info->balance_lock); 4389 } 4390 4391 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4392 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4393 describe_balance_start_or_resume(fs_info); 4394 mutex_unlock(&fs_info->balance_mutex); 4395 4396 ret = __btrfs_balance(fs_info); 4397 4398 mutex_lock(&fs_info->balance_mutex); 4399 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4400 btrfs_info(fs_info, "balance: paused"); 4401 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4402 } 4403 /* 4404 * Balance can be canceled by: 4405 * 4406 * - Regular cancel request 4407 * Then ret == -ECANCELED and balance_cancel_req > 0 4408 * 4409 * - Fatal signal to "btrfs" process 4410 * Either the signal caught by wait_reserve_ticket() and callers 4411 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4412 * got -ECANCELED. 4413 * Either way, in this case balance_cancel_req = 0, and 4414 * ret == -EINTR or ret == -ECANCELED. 4415 * 4416 * So here we only check the return value to catch canceled balance. 4417 */ 4418 else if (ret == -ECANCELED || ret == -EINTR) 4419 btrfs_info(fs_info, "balance: canceled"); 4420 else 4421 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4422 4423 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4424 4425 if (bargs) { 4426 memset(bargs, 0, sizeof(*bargs)); 4427 btrfs_update_ioctl_balance_args(fs_info, bargs); 4428 } 4429 4430 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4431 balance_need_close(fs_info)) { 4432 reset_balance_state(fs_info); 4433 btrfs_exclop_finish(fs_info); 4434 } 4435 4436 wake_up(&fs_info->balance_wait_q); 4437 4438 return ret; 4439 out: 4440 if (bctl->flags & BTRFS_BALANCE_RESUME) 4441 reset_balance_state(fs_info); 4442 else 4443 kfree(bctl); 4444 btrfs_exclop_finish(fs_info); 4445 4446 return ret; 4447 } 4448 4449 static int balance_kthread(void *data) 4450 { 4451 struct btrfs_fs_info *fs_info = data; 4452 int ret = 0; 4453 4454 sb_start_write(fs_info->sb); 4455 mutex_lock(&fs_info->balance_mutex); 4456 if (fs_info->balance_ctl) 4457 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4458 mutex_unlock(&fs_info->balance_mutex); 4459 sb_end_write(fs_info->sb); 4460 4461 return ret; 4462 } 4463 4464 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4465 { 4466 struct task_struct *tsk; 4467 4468 mutex_lock(&fs_info->balance_mutex); 4469 if (!fs_info->balance_ctl) { 4470 mutex_unlock(&fs_info->balance_mutex); 4471 return 0; 4472 } 4473 mutex_unlock(&fs_info->balance_mutex); 4474 4475 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4476 btrfs_info(fs_info, "balance: resume skipped"); 4477 return 0; 4478 } 4479 4480 spin_lock(&fs_info->super_lock); 4481 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4482 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4483 spin_unlock(&fs_info->super_lock); 4484 /* 4485 * A ro->rw remount sequence should continue with the paused balance 4486 * regardless of who pauses it, system or the user as of now, so set 4487 * the resume flag. 4488 */ 4489 spin_lock(&fs_info->balance_lock); 4490 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4491 spin_unlock(&fs_info->balance_lock); 4492 4493 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4494 return PTR_ERR_OR_ZERO(tsk); 4495 } 4496 4497 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4498 { 4499 struct btrfs_balance_control *bctl; 4500 struct btrfs_balance_item *item; 4501 struct btrfs_disk_balance_args disk_bargs; 4502 struct btrfs_path *path; 4503 struct extent_buffer *leaf; 4504 struct btrfs_key key; 4505 int ret; 4506 4507 path = btrfs_alloc_path(); 4508 if (!path) 4509 return -ENOMEM; 4510 4511 key.objectid = BTRFS_BALANCE_OBJECTID; 4512 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4513 key.offset = 0; 4514 4515 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4516 if (ret < 0) 4517 goto out; 4518 if (ret > 0) { /* ret = -ENOENT; */ 4519 ret = 0; 4520 goto out; 4521 } 4522 4523 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4524 if (!bctl) { 4525 ret = -ENOMEM; 4526 goto out; 4527 } 4528 4529 leaf = path->nodes[0]; 4530 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4531 4532 bctl->flags = btrfs_balance_flags(leaf, item); 4533 bctl->flags |= BTRFS_BALANCE_RESUME; 4534 4535 btrfs_balance_data(leaf, item, &disk_bargs); 4536 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4537 btrfs_balance_meta(leaf, item, &disk_bargs); 4538 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4539 btrfs_balance_sys(leaf, item, &disk_bargs); 4540 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4541 4542 /* 4543 * This should never happen, as the paused balance state is recovered 4544 * during mount without any chance of other exclusive ops to collide. 4545 * 4546 * This gives the exclusive op status to balance and keeps in paused 4547 * state until user intervention (cancel or umount). If the ownership 4548 * cannot be assigned, show a message but do not fail. The balance 4549 * is in a paused state and must have fs_info::balance_ctl properly 4550 * set up. 4551 */ 4552 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4553 btrfs_warn(fs_info, 4554 "balance: cannot set exclusive op status, resume manually"); 4555 4556 btrfs_release_path(path); 4557 4558 mutex_lock(&fs_info->balance_mutex); 4559 BUG_ON(fs_info->balance_ctl); 4560 spin_lock(&fs_info->balance_lock); 4561 fs_info->balance_ctl = bctl; 4562 spin_unlock(&fs_info->balance_lock); 4563 mutex_unlock(&fs_info->balance_mutex); 4564 out: 4565 btrfs_free_path(path); 4566 return ret; 4567 } 4568 4569 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4570 { 4571 int ret = 0; 4572 4573 mutex_lock(&fs_info->balance_mutex); 4574 if (!fs_info->balance_ctl) { 4575 mutex_unlock(&fs_info->balance_mutex); 4576 return -ENOTCONN; 4577 } 4578 4579 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4580 atomic_inc(&fs_info->balance_pause_req); 4581 mutex_unlock(&fs_info->balance_mutex); 4582 4583 wait_event(fs_info->balance_wait_q, 4584 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4585 4586 mutex_lock(&fs_info->balance_mutex); 4587 /* we are good with balance_ctl ripped off from under us */ 4588 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4589 atomic_dec(&fs_info->balance_pause_req); 4590 } else { 4591 ret = -ENOTCONN; 4592 } 4593 4594 mutex_unlock(&fs_info->balance_mutex); 4595 return ret; 4596 } 4597 4598 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4599 { 4600 mutex_lock(&fs_info->balance_mutex); 4601 if (!fs_info->balance_ctl) { 4602 mutex_unlock(&fs_info->balance_mutex); 4603 return -ENOTCONN; 4604 } 4605 4606 /* 4607 * A paused balance with the item stored on disk can be resumed at 4608 * mount time if the mount is read-write. Otherwise it's still paused 4609 * and we must not allow cancelling as it deletes the item. 4610 */ 4611 if (sb_rdonly(fs_info->sb)) { 4612 mutex_unlock(&fs_info->balance_mutex); 4613 return -EROFS; 4614 } 4615 4616 atomic_inc(&fs_info->balance_cancel_req); 4617 /* 4618 * if we are running just wait and return, balance item is 4619 * deleted in btrfs_balance in this case 4620 */ 4621 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4622 mutex_unlock(&fs_info->balance_mutex); 4623 wait_event(fs_info->balance_wait_q, 4624 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4625 mutex_lock(&fs_info->balance_mutex); 4626 } else { 4627 mutex_unlock(&fs_info->balance_mutex); 4628 /* 4629 * Lock released to allow other waiters to continue, we'll 4630 * reexamine the status again. 4631 */ 4632 mutex_lock(&fs_info->balance_mutex); 4633 4634 if (fs_info->balance_ctl) { 4635 reset_balance_state(fs_info); 4636 btrfs_exclop_finish(fs_info); 4637 btrfs_info(fs_info, "balance: canceled"); 4638 } 4639 } 4640 4641 BUG_ON(fs_info->balance_ctl || 4642 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4643 atomic_dec(&fs_info->balance_cancel_req); 4644 mutex_unlock(&fs_info->balance_mutex); 4645 return 0; 4646 } 4647 4648 int btrfs_uuid_scan_kthread(void *data) 4649 { 4650 struct btrfs_fs_info *fs_info = data; 4651 struct btrfs_root *root = fs_info->tree_root; 4652 struct btrfs_key key; 4653 struct btrfs_path *path = NULL; 4654 int ret = 0; 4655 struct extent_buffer *eb; 4656 int slot; 4657 struct btrfs_root_item root_item; 4658 u32 item_size; 4659 struct btrfs_trans_handle *trans = NULL; 4660 bool closing = false; 4661 4662 path = btrfs_alloc_path(); 4663 if (!path) { 4664 ret = -ENOMEM; 4665 goto out; 4666 } 4667 4668 key.objectid = 0; 4669 key.type = BTRFS_ROOT_ITEM_KEY; 4670 key.offset = 0; 4671 4672 while (1) { 4673 if (btrfs_fs_closing(fs_info)) { 4674 closing = true; 4675 break; 4676 } 4677 ret = btrfs_search_forward(root, &key, path, 4678 BTRFS_OLDEST_GENERATION); 4679 if (ret) { 4680 if (ret > 0) 4681 ret = 0; 4682 break; 4683 } 4684 4685 if (key.type != BTRFS_ROOT_ITEM_KEY || 4686 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4687 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4688 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4689 goto skip; 4690 4691 eb = path->nodes[0]; 4692 slot = path->slots[0]; 4693 item_size = btrfs_item_size(eb, slot); 4694 if (item_size < sizeof(root_item)) 4695 goto skip; 4696 4697 read_extent_buffer(eb, &root_item, 4698 btrfs_item_ptr_offset(eb, slot), 4699 (int)sizeof(root_item)); 4700 if (btrfs_root_refs(&root_item) == 0) 4701 goto skip; 4702 4703 if (!btrfs_is_empty_uuid(root_item.uuid) || 4704 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4705 if (trans) 4706 goto update_tree; 4707 4708 btrfs_release_path(path); 4709 /* 4710 * 1 - subvol uuid item 4711 * 1 - received_subvol uuid item 4712 */ 4713 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4714 if (IS_ERR(trans)) { 4715 ret = PTR_ERR(trans); 4716 break; 4717 } 4718 continue; 4719 } else { 4720 goto skip; 4721 } 4722 update_tree: 4723 btrfs_release_path(path); 4724 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4725 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4726 BTRFS_UUID_KEY_SUBVOL, 4727 key.objectid); 4728 if (ret < 0) { 4729 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4730 ret); 4731 break; 4732 } 4733 } 4734 4735 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4736 ret = btrfs_uuid_tree_add(trans, 4737 root_item.received_uuid, 4738 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4739 key.objectid); 4740 if (ret < 0) { 4741 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4742 ret); 4743 break; 4744 } 4745 } 4746 4747 skip: 4748 btrfs_release_path(path); 4749 if (trans) { 4750 ret = btrfs_end_transaction(trans); 4751 trans = NULL; 4752 if (ret) 4753 break; 4754 } 4755 4756 if (key.offset < (u64)-1) { 4757 key.offset++; 4758 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4759 key.offset = 0; 4760 key.type = BTRFS_ROOT_ITEM_KEY; 4761 } else if (key.objectid < (u64)-1) { 4762 key.offset = 0; 4763 key.type = BTRFS_ROOT_ITEM_KEY; 4764 key.objectid++; 4765 } else { 4766 break; 4767 } 4768 cond_resched(); 4769 } 4770 4771 out: 4772 btrfs_free_path(path); 4773 if (trans && !IS_ERR(trans)) 4774 btrfs_end_transaction(trans); 4775 if (ret) 4776 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4777 else if (!closing) 4778 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4779 up(&fs_info->uuid_tree_rescan_sem); 4780 return 0; 4781 } 4782 4783 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4784 { 4785 struct btrfs_trans_handle *trans; 4786 struct btrfs_root *tree_root = fs_info->tree_root; 4787 struct btrfs_root *uuid_root; 4788 struct task_struct *task; 4789 int ret; 4790 4791 /* 4792 * 1 - root node 4793 * 1 - root item 4794 */ 4795 trans = btrfs_start_transaction(tree_root, 2); 4796 if (IS_ERR(trans)) 4797 return PTR_ERR(trans); 4798 4799 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4800 if (IS_ERR(uuid_root)) { 4801 ret = PTR_ERR(uuid_root); 4802 btrfs_abort_transaction(trans, ret); 4803 btrfs_end_transaction(trans); 4804 return ret; 4805 } 4806 4807 fs_info->uuid_root = uuid_root; 4808 4809 ret = btrfs_commit_transaction(trans); 4810 if (ret) 4811 return ret; 4812 4813 down(&fs_info->uuid_tree_rescan_sem); 4814 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4815 if (IS_ERR(task)) { 4816 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4817 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4818 up(&fs_info->uuid_tree_rescan_sem); 4819 return PTR_ERR(task); 4820 } 4821 4822 return 0; 4823 } 4824 4825 /* 4826 * shrinking a device means finding all of the device extents past 4827 * the new size, and then following the back refs to the chunks. 4828 * The chunk relocation code actually frees the device extent 4829 */ 4830 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4831 { 4832 struct btrfs_fs_info *fs_info = device->fs_info; 4833 struct btrfs_root *root = fs_info->dev_root; 4834 struct btrfs_trans_handle *trans; 4835 struct btrfs_dev_extent *dev_extent = NULL; 4836 struct btrfs_path *path; 4837 u64 length; 4838 u64 chunk_offset; 4839 int ret; 4840 int slot; 4841 int failed = 0; 4842 bool retried = false; 4843 struct extent_buffer *l; 4844 struct btrfs_key key; 4845 struct btrfs_super_block *super_copy = fs_info->super_copy; 4846 u64 old_total = btrfs_super_total_bytes(super_copy); 4847 u64 old_size = btrfs_device_get_total_bytes(device); 4848 u64 diff; 4849 u64 start; 4850 4851 new_size = round_down(new_size, fs_info->sectorsize); 4852 start = new_size; 4853 diff = round_down(old_size - new_size, fs_info->sectorsize); 4854 4855 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4856 return -EINVAL; 4857 4858 path = btrfs_alloc_path(); 4859 if (!path) 4860 return -ENOMEM; 4861 4862 path->reada = READA_BACK; 4863 4864 trans = btrfs_start_transaction(root, 0); 4865 if (IS_ERR(trans)) { 4866 btrfs_free_path(path); 4867 return PTR_ERR(trans); 4868 } 4869 4870 mutex_lock(&fs_info->chunk_mutex); 4871 4872 btrfs_device_set_total_bytes(device, new_size); 4873 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4874 device->fs_devices->total_rw_bytes -= diff; 4875 atomic64_sub(diff, &fs_info->free_chunk_space); 4876 } 4877 4878 /* 4879 * Once the device's size has been set to the new size, ensure all 4880 * in-memory chunks are synced to disk so that the loop below sees them 4881 * and relocates them accordingly. 4882 */ 4883 if (contains_pending_extent(device, &start, diff)) { 4884 mutex_unlock(&fs_info->chunk_mutex); 4885 ret = btrfs_commit_transaction(trans); 4886 if (ret) 4887 goto done; 4888 } else { 4889 mutex_unlock(&fs_info->chunk_mutex); 4890 btrfs_end_transaction(trans); 4891 } 4892 4893 again: 4894 key.objectid = device->devid; 4895 key.offset = (u64)-1; 4896 key.type = BTRFS_DEV_EXTENT_KEY; 4897 4898 do { 4899 mutex_lock(&fs_info->reclaim_bgs_lock); 4900 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4901 if (ret < 0) { 4902 mutex_unlock(&fs_info->reclaim_bgs_lock); 4903 goto done; 4904 } 4905 4906 ret = btrfs_previous_item(root, path, 0, key.type); 4907 if (ret) { 4908 mutex_unlock(&fs_info->reclaim_bgs_lock); 4909 if (ret < 0) 4910 goto done; 4911 ret = 0; 4912 btrfs_release_path(path); 4913 break; 4914 } 4915 4916 l = path->nodes[0]; 4917 slot = path->slots[0]; 4918 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4919 4920 if (key.objectid != device->devid) { 4921 mutex_unlock(&fs_info->reclaim_bgs_lock); 4922 btrfs_release_path(path); 4923 break; 4924 } 4925 4926 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4927 length = btrfs_dev_extent_length(l, dev_extent); 4928 4929 if (key.offset + length <= new_size) { 4930 mutex_unlock(&fs_info->reclaim_bgs_lock); 4931 btrfs_release_path(path); 4932 break; 4933 } 4934 4935 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4936 btrfs_release_path(path); 4937 4938 /* 4939 * We may be relocating the only data chunk we have, 4940 * which could potentially end up with losing data's 4941 * raid profile, so lets allocate an empty one in 4942 * advance. 4943 */ 4944 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4945 if (ret < 0) { 4946 mutex_unlock(&fs_info->reclaim_bgs_lock); 4947 goto done; 4948 } 4949 4950 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4951 mutex_unlock(&fs_info->reclaim_bgs_lock); 4952 if (ret == -ENOSPC) { 4953 failed++; 4954 } else if (ret) { 4955 if (ret == -ETXTBSY) { 4956 btrfs_warn(fs_info, 4957 "could not shrink block group %llu due to active swapfile", 4958 chunk_offset); 4959 } 4960 goto done; 4961 } 4962 } while (key.offset-- > 0); 4963 4964 if (failed && !retried) { 4965 failed = 0; 4966 retried = true; 4967 goto again; 4968 } else if (failed && retried) { 4969 ret = -ENOSPC; 4970 goto done; 4971 } 4972 4973 /* Shrinking succeeded, else we would be at "done". */ 4974 trans = btrfs_start_transaction(root, 0); 4975 if (IS_ERR(trans)) { 4976 ret = PTR_ERR(trans); 4977 goto done; 4978 } 4979 4980 mutex_lock(&fs_info->chunk_mutex); 4981 /* Clear all state bits beyond the shrunk device size */ 4982 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4983 CHUNK_STATE_MASK); 4984 4985 btrfs_device_set_disk_total_bytes(device, new_size); 4986 if (list_empty(&device->post_commit_list)) 4987 list_add_tail(&device->post_commit_list, 4988 &trans->transaction->dev_update_list); 4989 4990 WARN_ON(diff > old_total); 4991 btrfs_set_super_total_bytes(super_copy, 4992 round_down(old_total - diff, fs_info->sectorsize)); 4993 mutex_unlock(&fs_info->chunk_mutex); 4994 4995 btrfs_reserve_chunk_metadata(trans, false); 4996 /* Now btrfs_update_device() will change the on-disk size. */ 4997 ret = btrfs_update_device(trans, device); 4998 btrfs_trans_release_chunk_metadata(trans); 4999 if (ret < 0) { 5000 btrfs_abort_transaction(trans, ret); 5001 btrfs_end_transaction(trans); 5002 } else { 5003 ret = btrfs_commit_transaction(trans); 5004 } 5005 done: 5006 btrfs_free_path(path); 5007 if (ret) { 5008 mutex_lock(&fs_info->chunk_mutex); 5009 btrfs_device_set_total_bytes(device, old_size); 5010 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 5011 device->fs_devices->total_rw_bytes += diff; 5012 atomic64_add(diff, &fs_info->free_chunk_space); 5013 mutex_unlock(&fs_info->chunk_mutex); 5014 } 5015 return ret; 5016 } 5017 5018 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 5019 struct btrfs_key *key, 5020 struct btrfs_chunk *chunk, int item_size) 5021 { 5022 struct btrfs_super_block *super_copy = fs_info->super_copy; 5023 struct btrfs_disk_key disk_key; 5024 u32 array_size; 5025 u8 *ptr; 5026 5027 lockdep_assert_held(&fs_info->chunk_mutex); 5028 5029 array_size = btrfs_super_sys_array_size(super_copy); 5030 if (array_size + item_size + sizeof(disk_key) 5031 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5032 return -EFBIG; 5033 5034 ptr = super_copy->sys_chunk_array + array_size; 5035 btrfs_cpu_key_to_disk(&disk_key, key); 5036 memcpy(ptr, &disk_key, sizeof(disk_key)); 5037 ptr += sizeof(disk_key); 5038 memcpy(ptr, chunk, item_size); 5039 item_size += sizeof(disk_key); 5040 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5041 5042 return 0; 5043 } 5044 5045 /* 5046 * sort the devices in descending order by max_avail, total_avail 5047 */ 5048 static int btrfs_cmp_device_info(const void *a, const void *b) 5049 { 5050 const struct btrfs_device_info *di_a = a; 5051 const struct btrfs_device_info *di_b = b; 5052 5053 if (di_a->max_avail > di_b->max_avail) 5054 return -1; 5055 if (di_a->max_avail < di_b->max_avail) 5056 return 1; 5057 if (di_a->total_avail > di_b->total_avail) 5058 return -1; 5059 if (di_a->total_avail < di_b->total_avail) 5060 return 1; 5061 return 0; 5062 } 5063 5064 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5065 { 5066 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5067 return; 5068 5069 btrfs_set_fs_incompat(info, RAID56); 5070 } 5071 5072 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5073 { 5074 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5075 return; 5076 5077 btrfs_set_fs_incompat(info, RAID1C34); 5078 } 5079 5080 /* 5081 * Structure used internally for btrfs_create_chunk() function. 5082 * Wraps needed parameters. 5083 */ 5084 struct alloc_chunk_ctl { 5085 u64 start; 5086 u64 type; 5087 /* Total number of stripes to allocate */ 5088 int num_stripes; 5089 /* sub_stripes info for map */ 5090 int sub_stripes; 5091 /* Stripes per device */ 5092 int dev_stripes; 5093 /* Maximum number of devices to use */ 5094 int devs_max; 5095 /* Minimum number of devices to use */ 5096 int devs_min; 5097 /* ndevs has to be a multiple of this */ 5098 int devs_increment; 5099 /* Number of copies */ 5100 int ncopies; 5101 /* Number of stripes worth of bytes to store parity information */ 5102 int nparity; 5103 u64 max_stripe_size; 5104 u64 max_chunk_size; 5105 u64 dev_extent_min; 5106 u64 stripe_size; 5107 u64 chunk_size; 5108 int ndevs; 5109 }; 5110 5111 static void init_alloc_chunk_ctl_policy_regular( 5112 struct btrfs_fs_devices *fs_devices, 5113 struct alloc_chunk_ctl *ctl) 5114 { 5115 struct btrfs_space_info *space_info; 5116 5117 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); 5118 ASSERT(space_info); 5119 5120 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); 5121 ctl->max_stripe_size = ctl->max_chunk_size; 5122 5123 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) 5124 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); 5125 5126 /* We don't want a chunk larger than 10% of writable space */ 5127 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), 5128 ctl->max_chunk_size); 5129 ctl->dev_extent_min = ctl->dev_stripes << BTRFS_STRIPE_LEN_SHIFT; 5130 } 5131 5132 static void init_alloc_chunk_ctl_policy_zoned( 5133 struct btrfs_fs_devices *fs_devices, 5134 struct alloc_chunk_ctl *ctl) 5135 { 5136 u64 zone_size = fs_devices->fs_info->zone_size; 5137 u64 limit; 5138 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5139 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5140 u64 min_chunk_size = min_data_stripes * zone_size; 5141 u64 type = ctl->type; 5142 5143 ctl->max_stripe_size = zone_size; 5144 if (type & BTRFS_BLOCK_GROUP_DATA) { 5145 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5146 zone_size); 5147 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5148 ctl->max_chunk_size = ctl->max_stripe_size; 5149 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5150 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5151 ctl->devs_max = min_t(int, ctl->devs_max, 5152 BTRFS_MAX_DEVS_SYS_CHUNK); 5153 } else { 5154 BUG(); 5155 } 5156 5157 /* We don't want a chunk larger than 10% of writable space */ 5158 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), 5159 zone_size), 5160 min_chunk_size); 5161 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5162 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5163 } 5164 5165 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5166 struct alloc_chunk_ctl *ctl) 5167 { 5168 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5169 5170 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5171 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5172 ctl->devs_max = btrfs_raid_array[index].devs_max; 5173 if (!ctl->devs_max) 5174 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5175 ctl->devs_min = btrfs_raid_array[index].devs_min; 5176 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5177 ctl->ncopies = btrfs_raid_array[index].ncopies; 5178 ctl->nparity = btrfs_raid_array[index].nparity; 5179 ctl->ndevs = 0; 5180 5181 switch (fs_devices->chunk_alloc_policy) { 5182 case BTRFS_CHUNK_ALLOC_REGULAR: 5183 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5184 break; 5185 case BTRFS_CHUNK_ALLOC_ZONED: 5186 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5187 break; 5188 default: 5189 BUG(); 5190 } 5191 } 5192 5193 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5194 struct alloc_chunk_ctl *ctl, 5195 struct btrfs_device_info *devices_info) 5196 { 5197 struct btrfs_fs_info *info = fs_devices->fs_info; 5198 struct btrfs_device *device; 5199 u64 total_avail; 5200 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5201 int ret; 5202 int ndevs = 0; 5203 u64 max_avail; 5204 u64 dev_offset; 5205 5206 /* 5207 * in the first pass through the devices list, we gather information 5208 * about the available holes on each device. 5209 */ 5210 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5211 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5212 WARN(1, KERN_ERR 5213 "BTRFS: read-only device in alloc_list\n"); 5214 continue; 5215 } 5216 5217 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5218 &device->dev_state) || 5219 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5220 continue; 5221 5222 if (device->total_bytes > device->bytes_used) 5223 total_avail = device->total_bytes - device->bytes_used; 5224 else 5225 total_avail = 0; 5226 5227 /* If there is no space on this device, skip it. */ 5228 if (total_avail < ctl->dev_extent_min) 5229 continue; 5230 5231 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5232 &max_avail); 5233 if (ret && ret != -ENOSPC) 5234 return ret; 5235 5236 if (ret == 0) 5237 max_avail = dev_extent_want; 5238 5239 if (max_avail < ctl->dev_extent_min) { 5240 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5241 btrfs_debug(info, 5242 "%s: devid %llu has no free space, have=%llu want=%llu", 5243 __func__, device->devid, max_avail, 5244 ctl->dev_extent_min); 5245 continue; 5246 } 5247 5248 if (ndevs == fs_devices->rw_devices) { 5249 WARN(1, "%s: found more than %llu devices\n", 5250 __func__, fs_devices->rw_devices); 5251 break; 5252 } 5253 devices_info[ndevs].dev_offset = dev_offset; 5254 devices_info[ndevs].max_avail = max_avail; 5255 devices_info[ndevs].total_avail = total_avail; 5256 devices_info[ndevs].dev = device; 5257 ++ndevs; 5258 } 5259 ctl->ndevs = ndevs; 5260 5261 /* 5262 * now sort the devices by hole size / available space 5263 */ 5264 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5265 btrfs_cmp_device_info, NULL); 5266 5267 return 0; 5268 } 5269 5270 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5271 struct btrfs_device_info *devices_info) 5272 { 5273 /* Number of stripes that count for block group size */ 5274 int data_stripes; 5275 5276 /* 5277 * The primary goal is to maximize the number of stripes, so use as 5278 * many devices as possible, even if the stripes are not maximum sized. 5279 * 5280 * The DUP profile stores more than one stripe per device, the 5281 * max_avail is the total size so we have to adjust. 5282 */ 5283 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5284 ctl->dev_stripes); 5285 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5286 5287 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5288 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5289 5290 /* 5291 * Use the number of data stripes to figure out how big this chunk is 5292 * really going to be in terms of logical address space, and compare 5293 * that answer with the max chunk size. If it's higher, we try to 5294 * reduce stripe_size. 5295 */ 5296 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5297 /* 5298 * Reduce stripe_size, round it up to a 16MB boundary again and 5299 * then use it, unless it ends up being even bigger than the 5300 * previous value we had already. 5301 */ 5302 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5303 data_stripes), SZ_16M), 5304 ctl->stripe_size); 5305 } 5306 5307 /* Stripe size should not go beyond 1G. */ 5308 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); 5309 5310 /* Align to BTRFS_STRIPE_LEN */ 5311 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5312 ctl->chunk_size = ctl->stripe_size * data_stripes; 5313 5314 return 0; 5315 } 5316 5317 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5318 struct btrfs_device_info *devices_info) 5319 { 5320 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5321 /* Number of stripes that count for block group size */ 5322 int data_stripes; 5323 5324 /* 5325 * It should hold because: 5326 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5327 */ 5328 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5329 5330 ctl->stripe_size = zone_size; 5331 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5332 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5333 5334 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5335 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5336 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5337 ctl->stripe_size) + ctl->nparity, 5338 ctl->dev_stripes); 5339 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5340 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5341 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5342 } 5343 5344 ctl->chunk_size = ctl->stripe_size * data_stripes; 5345 5346 return 0; 5347 } 5348 5349 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5350 struct alloc_chunk_ctl *ctl, 5351 struct btrfs_device_info *devices_info) 5352 { 5353 struct btrfs_fs_info *info = fs_devices->fs_info; 5354 5355 /* 5356 * Round down to number of usable stripes, devs_increment can be any 5357 * number so we can't use round_down() that requires power of 2, while 5358 * rounddown is safe. 5359 */ 5360 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5361 5362 if (ctl->ndevs < ctl->devs_min) { 5363 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5364 btrfs_debug(info, 5365 "%s: not enough devices with free space: have=%d minimum required=%d", 5366 __func__, ctl->ndevs, ctl->devs_min); 5367 } 5368 return -ENOSPC; 5369 } 5370 5371 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5372 5373 switch (fs_devices->chunk_alloc_policy) { 5374 case BTRFS_CHUNK_ALLOC_REGULAR: 5375 return decide_stripe_size_regular(ctl, devices_info); 5376 case BTRFS_CHUNK_ALLOC_ZONED: 5377 return decide_stripe_size_zoned(ctl, devices_info); 5378 default: 5379 BUG(); 5380 } 5381 } 5382 5383 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5384 struct alloc_chunk_ctl *ctl, 5385 struct btrfs_device_info *devices_info) 5386 { 5387 struct btrfs_fs_info *info = trans->fs_info; 5388 struct map_lookup *map = NULL; 5389 struct extent_map_tree *em_tree; 5390 struct btrfs_block_group *block_group; 5391 struct extent_map *em; 5392 u64 start = ctl->start; 5393 u64 type = ctl->type; 5394 int ret; 5395 int i; 5396 int j; 5397 5398 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5399 if (!map) 5400 return ERR_PTR(-ENOMEM); 5401 map->num_stripes = ctl->num_stripes; 5402 5403 for (i = 0; i < ctl->ndevs; ++i) { 5404 for (j = 0; j < ctl->dev_stripes; ++j) { 5405 int s = i * ctl->dev_stripes + j; 5406 map->stripes[s].dev = devices_info[i].dev; 5407 map->stripes[s].physical = devices_info[i].dev_offset + 5408 j * ctl->stripe_size; 5409 } 5410 } 5411 map->io_align = BTRFS_STRIPE_LEN; 5412 map->io_width = BTRFS_STRIPE_LEN; 5413 map->type = type; 5414 map->sub_stripes = ctl->sub_stripes; 5415 5416 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5417 5418 em = alloc_extent_map(); 5419 if (!em) { 5420 kfree(map); 5421 return ERR_PTR(-ENOMEM); 5422 } 5423 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5424 em->map_lookup = map; 5425 em->start = start; 5426 em->len = ctl->chunk_size; 5427 em->block_start = 0; 5428 em->block_len = em->len; 5429 em->orig_block_len = ctl->stripe_size; 5430 5431 em_tree = &info->mapping_tree; 5432 write_lock(&em_tree->lock); 5433 ret = add_extent_mapping(em_tree, em, 0); 5434 if (ret) { 5435 write_unlock(&em_tree->lock); 5436 free_extent_map(em); 5437 return ERR_PTR(ret); 5438 } 5439 write_unlock(&em_tree->lock); 5440 5441 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size); 5442 if (IS_ERR(block_group)) 5443 goto error_del_extent; 5444 5445 for (i = 0; i < map->num_stripes; i++) { 5446 struct btrfs_device *dev = map->stripes[i].dev; 5447 5448 btrfs_device_set_bytes_used(dev, 5449 dev->bytes_used + ctl->stripe_size); 5450 if (list_empty(&dev->post_commit_list)) 5451 list_add_tail(&dev->post_commit_list, 5452 &trans->transaction->dev_update_list); 5453 } 5454 5455 atomic64_sub(ctl->stripe_size * map->num_stripes, 5456 &info->free_chunk_space); 5457 5458 free_extent_map(em); 5459 check_raid56_incompat_flag(info, type); 5460 check_raid1c34_incompat_flag(info, type); 5461 5462 return block_group; 5463 5464 error_del_extent: 5465 write_lock(&em_tree->lock); 5466 remove_extent_mapping(em_tree, em); 5467 write_unlock(&em_tree->lock); 5468 5469 /* One for our allocation */ 5470 free_extent_map(em); 5471 /* One for the tree reference */ 5472 free_extent_map(em); 5473 5474 return block_group; 5475 } 5476 5477 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5478 u64 type) 5479 { 5480 struct btrfs_fs_info *info = trans->fs_info; 5481 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5482 struct btrfs_device_info *devices_info = NULL; 5483 struct alloc_chunk_ctl ctl; 5484 struct btrfs_block_group *block_group; 5485 int ret; 5486 5487 lockdep_assert_held(&info->chunk_mutex); 5488 5489 if (!alloc_profile_is_valid(type, 0)) { 5490 ASSERT(0); 5491 return ERR_PTR(-EINVAL); 5492 } 5493 5494 if (list_empty(&fs_devices->alloc_list)) { 5495 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5496 btrfs_debug(info, "%s: no writable device", __func__); 5497 return ERR_PTR(-ENOSPC); 5498 } 5499 5500 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5501 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5502 ASSERT(0); 5503 return ERR_PTR(-EINVAL); 5504 } 5505 5506 ctl.start = find_next_chunk(info); 5507 ctl.type = type; 5508 init_alloc_chunk_ctl(fs_devices, &ctl); 5509 5510 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5511 GFP_NOFS); 5512 if (!devices_info) 5513 return ERR_PTR(-ENOMEM); 5514 5515 ret = gather_device_info(fs_devices, &ctl, devices_info); 5516 if (ret < 0) { 5517 block_group = ERR_PTR(ret); 5518 goto out; 5519 } 5520 5521 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5522 if (ret < 0) { 5523 block_group = ERR_PTR(ret); 5524 goto out; 5525 } 5526 5527 block_group = create_chunk(trans, &ctl, devices_info); 5528 5529 out: 5530 kfree(devices_info); 5531 return block_group; 5532 } 5533 5534 /* 5535 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5536 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5537 * chunks. 5538 * 5539 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5540 * phases. 5541 */ 5542 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5543 struct btrfs_block_group *bg) 5544 { 5545 struct btrfs_fs_info *fs_info = trans->fs_info; 5546 struct btrfs_root *chunk_root = fs_info->chunk_root; 5547 struct btrfs_key key; 5548 struct btrfs_chunk *chunk; 5549 struct btrfs_stripe *stripe; 5550 struct extent_map *em; 5551 struct map_lookup *map; 5552 size_t item_size; 5553 int i; 5554 int ret; 5555 5556 /* 5557 * We take the chunk_mutex for 2 reasons: 5558 * 5559 * 1) Updates and insertions in the chunk btree must be done while holding 5560 * the chunk_mutex, as well as updating the system chunk array in the 5561 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5562 * details; 5563 * 5564 * 2) To prevent races with the final phase of a device replace operation 5565 * that replaces the device object associated with the map's stripes, 5566 * because the device object's id can change at any time during that 5567 * final phase of the device replace operation 5568 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5569 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5570 * which would cause a failure when updating the device item, which does 5571 * not exists, or persisting a stripe of the chunk item with such ID. 5572 * Here we can't use the device_list_mutex because our caller already 5573 * has locked the chunk_mutex, and the final phase of device replace 5574 * acquires both mutexes - first the device_list_mutex and then the 5575 * chunk_mutex. Using any of those two mutexes protects us from a 5576 * concurrent device replace. 5577 */ 5578 lockdep_assert_held(&fs_info->chunk_mutex); 5579 5580 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5581 if (IS_ERR(em)) { 5582 ret = PTR_ERR(em); 5583 btrfs_abort_transaction(trans, ret); 5584 return ret; 5585 } 5586 5587 map = em->map_lookup; 5588 item_size = btrfs_chunk_item_size(map->num_stripes); 5589 5590 chunk = kzalloc(item_size, GFP_NOFS); 5591 if (!chunk) { 5592 ret = -ENOMEM; 5593 btrfs_abort_transaction(trans, ret); 5594 goto out; 5595 } 5596 5597 for (i = 0; i < map->num_stripes; i++) { 5598 struct btrfs_device *device = map->stripes[i].dev; 5599 5600 ret = btrfs_update_device(trans, device); 5601 if (ret) 5602 goto out; 5603 } 5604 5605 stripe = &chunk->stripe; 5606 for (i = 0; i < map->num_stripes; i++) { 5607 struct btrfs_device *device = map->stripes[i].dev; 5608 const u64 dev_offset = map->stripes[i].physical; 5609 5610 btrfs_set_stack_stripe_devid(stripe, device->devid); 5611 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5612 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5613 stripe++; 5614 } 5615 5616 btrfs_set_stack_chunk_length(chunk, bg->length); 5617 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5618 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN); 5619 btrfs_set_stack_chunk_type(chunk, map->type); 5620 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5621 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN); 5622 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN); 5623 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5624 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5625 5626 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5627 key.type = BTRFS_CHUNK_ITEM_KEY; 5628 key.offset = bg->start; 5629 5630 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5631 if (ret) 5632 goto out; 5633 5634 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); 5635 5636 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5637 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5638 if (ret) 5639 goto out; 5640 } 5641 5642 out: 5643 kfree(chunk); 5644 free_extent_map(em); 5645 return ret; 5646 } 5647 5648 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5649 { 5650 struct btrfs_fs_info *fs_info = trans->fs_info; 5651 u64 alloc_profile; 5652 struct btrfs_block_group *meta_bg; 5653 struct btrfs_block_group *sys_bg; 5654 5655 /* 5656 * When adding a new device for sprouting, the seed device is read-only 5657 * so we must first allocate a metadata and a system chunk. But before 5658 * adding the block group items to the extent, device and chunk btrees, 5659 * we must first: 5660 * 5661 * 1) Create both chunks without doing any changes to the btrees, as 5662 * otherwise we would get -ENOSPC since the block groups from the 5663 * seed device are read-only; 5664 * 5665 * 2) Add the device item for the new sprout device - finishing the setup 5666 * of a new block group requires updating the device item in the chunk 5667 * btree, so it must exist when we attempt to do it. The previous step 5668 * ensures this does not fail with -ENOSPC. 5669 * 5670 * After that we can add the block group items to their btrees: 5671 * update existing device item in the chunk btree, add a new block group 5672 * item to the extent btree, add a new chunk item to the chunk btree and 5673 * finally add the new device extent items to the devices btree. 5674 */ 5675 5676 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5677 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5678 if (IS_ERR(meta_bg)) 5679 return PTR_ERR(meta_bg); 5680 5681 alloc_profile = btrfs_system_alloc_profile(fs_info); 5682 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5683 if (IS_ERR(sys_bg)) 5684 return PTR_ERR(sys_bg); 5685 5686 return 0; 5687 } 5688 5689 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5690 { 5691 const int index = btrfs_bg_flags_to_raid_index(map->type); 5692 5693 return btrfs_raid_array[index].tolerated_failures; 5694 } 5695 5696 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5697 { 5698 struct extent_map *em; 5699 struct map_lookup *map; 5700 int miss_ndevs = 0; 5701 int i; 5702 bool ret = true; 5703 5704 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5705 if (IS_ERR(em)) 5706 return false; 5707 5708 map = em->map_lookup; 5709 for (i = 0; i < map->num_stripes; i++) { 5710 if (test_bit(BTRFS_DEV_STATE_MISSING, 5711 &map->stripes[i].dev->dev_state)) { 5712 miss_ndevs++; 5713 continue; 5714 } 5715 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5716 &map->stripes[i].dev->dev_state)) { 5717 ret = false; 5718 goto end; 5719 } 5720 } 5721 5722 /* 5723 * If the number of missing devices is larger than max errors, we can 5724 * not write the data into that chunk successfully. 5725 */ 5726 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5727 ret = false; 5728 end: 5729 free_extent_map(em); 5730 return ret; 5731 } 5732 5733 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5734 { 5735 struct extent_map *em; 5736 5737 while (1) { 5738 write_lock(&tree->lock); 5739 em = lookup_extent_mapping(tree, 0, (u64)-1); 5740 if (em) 5741 remove_extent_mapping(tree, em); 5742 write_unlock(&tree->lock); 5743 if (!em) 5744 break; 5745 /* once for us */ 5746 free_extent_map(em); 5747 /* once for the tree */ 5748 free_extent_map(em); 5749 } 5750 } 5751 5752 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5753 { 5754 struct extent_map *em; 5755 struct map_lookup *map; 5756 enum btrfs_raid_types index; 5757 int ret = 1; 5758 5759 em = btrfs_get_chunk_map(fs_info, logical, len); 5760 if (IS_ERR(em)) 5761 /* 5762 * We could return errors for these cases, but that could get 5763 * ugly and we'd probably do the same thing which is just not do 5764 * anything else and exit, so return 1 so the callers don't try 5765 * to use other copies. 5766 */ 5767 return 1; 5768 5769 map = em->map_lookup; 5770 index = btrfs_bg_flags_to_raid_index(map->type); 5771 5772 /* Non-RAID56, use their ncopies from btrfs_raid_array. */ 5773 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5774 ret = btrfs_raid_array[index].ncopies; 5775 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5776 ret = 2; 5777 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5778 /* 5779 * There could be two corrupted data stripes, we need 5780 * to loop retry in order to rebuild the correct data. 5781 * 5782 * Fail a stripe at a time on every retry except the 5783 * stripe under reconstruction. 5784 */ 5785 ret = map->num_stripes; 5786 free_extent_map(em); 5787 return ret; 5788 } 5789 5790 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5791 u64 logical) 5792 { 5793 struct extent_map *em; 5794 struct map_lookup *map; 5795 unsigned long len = fs_info->sectorsize; 5796 5797 if (!btrfs_fs_incompat(fs_info, RAID56)) 5798 return len; 5799 5800 em = btrfs_get_chunk_map(fs_info, logical, len); 5801 5802 if (!WARN_ON(IS_ERR(em))) { 5803 map = em->map_lookup; 5804 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5805 len = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT; 5806 free_extent_map(em); 5807 } 5808 return len; 5809 } 5810 5811 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5812 { 5813 struct extent_map *em; 5814 struct map_lookup *map; 5815 int ret = 0; 5816 5817 if (!btrfs_fs_incompat(fs_info, RAID56)) 5818 return 0; 5819 5820 em = btrfs_get_chunk_map(fs_info, logical, len); 5821 5822 if(!WARN_ON(IS_ERR(em))) { 5823 map = em->map_lookup; 5824 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5825 ret = 1; 5826 free_extent_map(em); 5827 } 5828 return ret; 5829 } 5830 5831 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5832 struct map_lookup *map, int first, 5833 int dev_replace_is_ongoing) 5834 { 5835 int i; 5836 int num_stripes; 5837 int preferred_mirror; 5838 int tolerance; 5839 struct btrfs_device *srcdev; 5840 5841 ASSERT((map->type & 5842 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5843 5844 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5845 num_stripes = map->sub_stripes; 5846 else 5847 num_stripes = map->num_stripes; 5848 5849 switch (fs_info->fs_devices->read_policy) { 5850 default: 5851 /* Shouldn't happen, just warn and use pid instead of failing */ 5852 btrfs_warn_rl(fs_info, 5853 "unknown read_policy type %u, reset to pid", 5854 fs_info->fs_devices->read_policy); 5855 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5856 fallthrough; 5857 case BTRFS_READ_POLICY_PID: 5858 preferred_mirror = first + (current->pid % num_stripes); 5859 break; 5860 } 5861 5862 if (dev_replace_is_ongoing && 5863 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5864 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5865 srcdev = fs_info->dev_replace.srcdev; 5866 else 5867 srcdev = NULL; 5868 5869 /* 5870 * try to avoid the drive that is the source drive for a 5871 * dev-replace procedure, only choose it if no other non-missing 5872 * mirror is available 5873 */ 5874 for (tolerance = 0; tolerance < 2; tolerance++) { 5875 if (map->stripes[preferred_mirror].dev->bdev && 5876 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5877 return preferred_mirror; 5878 for (i = first; i < first + num_stripes; i++) { 5879 if (map->stripes[i].dev->bdev && 5880 (tolerance || map->stripes[i].dev != srcdev)) 5881 return i; 5882 } 5883 } 5884 5885 /* we couldn't find one that doesn't fail. Just return something 5886 * and the io error handling code will clean up eventually 5887 */ 5888 return preferred_mirror; 5889 } 5890 5891 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5892 u16 total_stripes) 5893 { 5894 struct btrfs_io_context *bioc; 5895 5896 bioc = kzalloc( 5897 /* The size of btrfs_io_context */ 5898 sizeof(struct btrfs_io_context) + 5899 /* Plus the variable array for the stripes */ 5900 sizeof(struct btrfs_io_stripe) * (total_stripes), 5901 GFP_NOFS); 5902 5903 if (!bioc) 5904 return NULL; 5905 5906 refcount_set(&bioc->refs, 1); 5907 5908 bioc->fs_info = fs_info; 5909 bioc->replace_stripe_src = -1; 5910 bioc->full_stripe_logical = (u64)-1; 5911 5912 return bioc; 5913 } 5914 5915 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5916 { 5917 WARN_ON(!refcount_read(&bioc->refs)); 5918 refcount_inc(&bioc->refs); 5919 } 5920 5921 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5922 { 5923 if (!bioc) 5924 return; 5925 if (refcount_dec_and_test(&bioc->refs)) 5926 kfree(bioc); 5927 } 5928 5929 /* 5930 * Please note that, discard won't be sent to target device of device 5931 * replace. 5932 */ 5933 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, 5934 u64 logical, u64 *length_ret, 5935 u32 *num_stripes) 5936 { 5937 struct extent_map *em; 5938 struct map_lookup *map; 5939 struct btrfs_discard_stripe *stripes; 5940 u64 length = *length_ret; 5941 u64 offset; 5942 u32 stripe_nr; 5943 u32 stripe_nr_end; 5944 u32 stripe_cnt; 5945 u64 stripe_end_offset; 5946 u64 stripe_offset; 5947 u32 stripe_index; 5948 u32 factor = 0; 5949 u32 sub_stripes = 0; 5950 u32 stripes_per_dev = 0; 5951 u32 remaining_stripes = 0; 5952 u32 last_stripe = 0; 5953 int ret; 5954 int i; 5955 5956 em = btrfs_get_chunk_map(fs_info, logical, length); 5957 if (IS_ERR(em)) 5958 return ERR_CAST(em); 5959 5960 map = em->map_lookup; 5961 5962 /* we don't discard raid56 yet */ 5963 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5964 ret = -EOPNOTSUPP; 5965 goto out_free_map; 5966 } 5967 5968 offset = logical - em->start; 5969 length = min_t(u64, em->start + em->len - logical, length); 5970 *length_ret = length; 5971 5972 /* 5973 * stripe_nr counts the total number of stripes we have to stride 5974 * to get to this block 5975 */ 5976 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 5977 5978 /* stripe_offset is the offset of this block in its stripe */ 5979 stripe_offset = offset - (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 5980 5981 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >> 5982 BTRFS_STRIPE_LEN_SHIFT; 5983 stripe_cnt = stripe_nr_end - stripe_nr; 5984 stripe_end_offset = (stripe_nr_end << BTRFS_STRIPE_LEN_SHIFT) - 5985 (offset + length); 5986 /* 5987 * after this, stripe_nr is the number of stripes on this 5988 * device we have to walk to find the data, and stripe_index is 5989 * the number of our device in the stripe array 5990 */ 5991 *num_stripes = 1; 5992 stripe_index = 0; 5993 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5994 BTRFS_BLOCK_GROUP_RAID10)) { 5995 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 5996 sub_stripes = 1; 5997 else 5998 sub_stripes = map->sub_stripes; 5999 6000 factor = map->num_stripes / sub_stripes; 6001 *num_stripes = min_t(u64, map->num_stripes, 6002 sub_stripes * stripe_cnt); 6003 stripe_index = stripe_nr % factor; 6004 stripe_nr /= factor; 6005 stripe_index *= sub_stripes; 6006 6007 remaining_stripes = stripe_cnt % factor; 6008 stripes_per_dev = stripe_cnt / factor; 6009 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes; 6010 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 6011 BTRFS_BLOCK_GROUP_DUP)) { 6012 *num_stripes = map->num_stripes; 6013 } else { 6014 stripe_index = stripe_nr % map->num_stripes; 6015 stripe_nr /= map->num_stripes; 6016 } 6017 6018 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); 6019 if (!stripes) { 6020 ret = -ENOMEM; 6021 goto out_free_map; 6022 } 6023 6024 for (i = 0; i < *num_stripes; i++) { 6025 stripes[i].physical = 6026 map->stripes[stripe_index].physical + 6027 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6028 stripes[i].dev = map->stripes[stripe_index].dev; 6029 6030 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6031 BTRFS_BLOCK_GROUP_RAID10)) { 6032 stripes[i].length = stripes_per_dev << BTRFS_STRIPE_LEN_SHIFT; 6033 6034 if (i / sub_stripes < remaining_stripes) 6035 stripes[i].length += BTRFS_STRIPE_LEN; 6036 6037 /* 6038 * Special for the first stripe and 6039 * the last stripe: 6040 * 6041 * |-------|...|-------| 6042 * |----------| 6043 * off end_off 6044 */ 6045 if (i < sub_stripes) 6046 stripes[i].length -= stripe_offset; 6047 6048 if (stripe_index >= last_stripe && 6049 stripe_index <= (last_stripe + 6050 sub_stripes - 1)) 6051 stripes[i].length -= stripe_end_offset; 6052 6053 if (i == sub_stripes - 1) 6054 stripe_offset = 0; 6055 } else { 6056 stripes[i].length = length; 6057 } 6058 6059 stripe_index++; 6060 if (stripe_index == map->num_stripes) { 6061 stripe_index = 0; 6062 stripe_nr++; 6063 } 6064 } 6065 6066 free_extent_map(em); 6067 return stripes; 6068 out_free_map: 6069 free_extent_map(em); 6070 return ERR_PTR(ret); 6071 } 6072 6073 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6074 { 6075 struct btrfs_block_group *cache; 6076 bool ret; 6077 6078 /* Non zoned filesystem does not use "to_copy" flag */ 6079 if (!btrfs_is_zoned(fs_info)) 6080 return false; 6081 6082 cache = btrfs_lookup_block_group(fs_info, logical); 6083 6084 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 6085 6086 btrfs_put_block_group(cache); 6087 return ret; 6088 } 6089 6090 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6091 struct btrfs_io_context *bioc, 6092 struct btrfs_dev_replace *dev_replace, 6093 u64 logical, 6094 int *num_stripes_ret, int *max_errors_ret) 6095 { 6096 u64 srcdev_devid = dev_replace->srcdev->devid; 6097 /* 6098 * At this stage, num_stripes is still the real number of stripes, 6099 * excluding the duplicated stripes. 6100 */ 6101 int num_stripes = *num_stripes_ret; 6102 int nr_extra_stripes = 0; 6103 int max_errors = *max_errors_ret; 6104 int i; 6105 6106 /* 6107 * A block group which has "to_copy" set will eventually be copied by 6108 * the dev-replace process. We can avoid cloning IO here. 6109 */ 6110 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6111 return; 6112 6113 /* 6114 * Duplicate the write operations while the dev-replace procedure is 6115 * running. Since the copying of the old disk to the new disk takes 6116 * place at run time while the filesystem is mounted writable, the 6117 * regular write operations to the old disk have to be duplicated to go 6118 * to the new disk as well. 6119 * 6120 * Note that device->missing is handled by the caller, and that the 6121 * write to the old disk is already set up in the stripes array. 6122 */ 6123 for (i = 0; i < num_stripes; i++) { 6124 struct btrfs_io_stripe *old = &bioc->stripes[i]; 6125 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes]; 6126 6127 if (old->dev->devid != srcdev_devid) 6128 continue; 6129 6130 new->physical = old->physical; 6131 new->dev = dev_replace->tgtdev; 6132 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) 6133 bioc->replace_stripe_src = i; 6134 nr_extra_stripes++; 6135 } 6136 6137 /* We can only have at most 2 extra nr_stripes (for DUP). */ 6138 ASSERT(nr_extra_stripes <= 2); 6139 /* 6140 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for 6141 * replace. 6142 * If we have 2 extra stripes, only choose the one with smaller physical. 6143 */ 6144 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) { 6145 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes]; 6146 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1]; 6147 6148 /* Only DUP can have two extra stripes. */ 6149 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP); 6150 6151 /* 6152 * Swap the last stripe stripes and reduce @nr_extra_stripes. 6153 * The extra stripe would still be there, but won't be accessed. 6154 */ 6155 if (first->physical > second->physical) { 6156 swap(second->physical, first->physical); 6157 swap(second->dev, first->dev); 6158 nr_extra_stripes--; 6159 } 6160 } 6161 6162 *num_stripes_ret = num_stripes + nr_extra_stripes; 6163 *max_errors_ret = max_errors + nr_extra_stripes; 6164 bioc->replace_nr_stripes = nr_extra_stripes; 6165 } 6166 6167 static bool need_full_stripe(enum btrfs_map_op op) 6168 { 6169 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6170 } 6171 6172 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op, 6173 u64 offset, u32 *stripe_nr, u64 *stripe_offset, 6174 u64 *full_stripe_start) 6175 { 6176 ASSERT(op != BTRFS_MAP_DISCARD); 6177 6178 /* 6179 * Stripe_nr is the stripe where this block falls. stripe_offset is 6180 * the offset of this block in its stripe. 6181 */ 6182 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK; 6183 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 6184 ASSERT(*stripe_offset < U32_MAX); 6185 6186 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6187 unsigned long full_stripe_len = nr_data_stripes(map) << 6188 BTRFS_STRIPE_LEN_SHIFT; 6189 6190 /* 6191 * For full stripe start, we use previously calculated 6192 * @stripe_nr. Align it to nr_data_stripes, then multiply with 6193 * STRIPE_LEN. 6194 * 6195 * By this we can avoid u64 division completely. And we have 6196 * to go rounddown(), not round_down(), as nr_data_stripes is 6197 * not ensured to be power of 2. 6198 */ 6199 *full_stripe_start = 6200 rounddown(*stripe_nr, nr_data_stripes(map)) << 6201 BTRFS_STRIPE_LEN_SHIFT; 6202 6203 /* 6204 * For writes to RAID56, allow to write a full stripe set, but 6205 * no straddling of stripe sets. 6206 */ 6207 if (op == BTRFS_MAP_WRITE) 6208 return full_stripe_len - (offset - *full_stripe_start); 6209 } 6210 6211 /* 6212 * For other RAID types and for RAID56 reads, allow a single stripe (on 6213 * a single disk). 6214 */ 6215 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) 6216 return BTRFS_STRIPE_LEN - *stripe_offset; 6217 return U64_MAX; 6218 } 6219 6220 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, 6221 u32 stripe_index, u64 stripe_offset, u32 stripe_nr) 6222 { 6223 dst->dev = map->stripes[stripe_index].dev; 6224 dst->physical = map->stripes[stripe_index].physical + 6225 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6226 } 6227 6228 int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6229 u64 logical, u64 *length, 6230 struct btrfs_io_context **bioc_ret, 6231 struct btrfs_io_stripe *smap, int *mirror_num_ret, 6232 int need_raid_map) 6233 { 6234 struct extent_map *em; 6235 struct map_lookup *map; 6236 u64 map_offset; 6237 u64 stripe_offset; 6238 u32 stripe_nr; 6239 u32 stripe_index; 6240 int data_stripes; 6241 int i; 6242 int ret = 0; 6243 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); 6244 int num_stripes; 6245 int num_copies; 6246 int max_errors = 0; 6247 struct btrfs_io_context *bioc = NULL; 6248 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6249 int dev_replace_is_ongoing = 0; 6250 u16 num_alloc_stripes; 6251 u64 raid56_full_stripe_start = (u64)-1; 6252 u64 max_len; 6253 6254 ASSERT(bioc_ret); 6255 ASSERT(op != BTRFS_MAP_DISCARD); 6256 6257 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize); 6258 if (mirror_num > num_copies) 6259 return -EINVAL; 6260 6261 em = btrfs_get_chunk_map(fs_info, logical, *length); 6262 if (IS_ERR(em)) 6263 return PTR_ERR(em); 6264 6265 map = em->map_lookup; 6266 data_stripes = nr_data_stripes(map); 6267 6268 map_offset = logical - em->start; 6269 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr, 6270 &stripe_offset, &raid56_full_stripe_start); 6271 *length = min_t(u64, em->len - map_offset, max_len); 6272 6273 down_read(&dev_replace->rwsem); 6274 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6275 /* 6276 * Hold the semaphore for read during the whole operation, write is 6277 * requested at commit time but must wait. 6278 */ 6279 if (!dev_replace_is_ongoing) 6280 up_read(&dev_replace->rwsem); 6281 6282 num_stripes = 1; 6283 stripe_index = 0; 6284 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6285 stripe_index = stripe_nr % map->num_stripes; 6286 stripe_nr /= map->num_stripes; 6287 if (!need_full_stripe(op)) 6288 mirror_num = 1; 6289 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6290 if (need_full_stripe(op)) 6291 num_stripes = map->num_stripes; 6292 else if (mirror_num) 6293 stripe_index = mirror_num - 1; 6294 else { 6295 stripe_index = find_live_mirror(fs_info, map, 0, 6296 dev_replace_is_ongoing); 6297 mirror_num = stripe_index + 1; 6298 } 6299 6300 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6301 if (need_full_stripe(op)) { 6302 num_stripes = map->num_stripes; 6303 } else if (mirror_num) { 6304 stripe_index = mirror_num - 1; 6305 } else { 6306 mirror_num = 1; 6307 } 6308 6309 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6310 u32 factor = map->num_stripes / map->sub_stripes; 6311 6312 stripe_index = (stripe_nr % factor) * map->sub_stripes; 6313 stripe_nr /= factor; 6314 6315 if (need_full_stripe(op)) 6316 num_stripes = map->sub_stripes; 6317 else if (mirror_num) 6318 stripe_index += mirror_num - 1; 6319 else { 6320 int old_stripe_index = stripe_index; 6321 stripe_index = find_live_mirror(fs_info, map, 6322 stripe_index, 6323 dev_replace_is_ongoing); 6324 mirror_num = stripe_index - old_stripe_index + 1; 6325 } 6326 6327 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6328 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6329 /* 6330 * Push stripe_nr back to the start of the full stripe 6331 * For those cases needing a full stripe, @stripe_nr 6332 * is the full stripe number. 6333 * 6334 * Originally we go raid56_full_stripe_start / full_stripe_len, 6335 * but that can be expensive. Here we just divide 6336 * @stripe_nr with @data_stripes. 6337 */ 6338 stripe_nr /= data_stripes; 6339 6340 /* RAID[56] write or recovery. Return all stripes */ 6341 num_stripes = map->num_stripes; 6342 max_errors = btrfs_chunk_max_errors(map); 6343 6344 /* Return the length to the full stripe end */ 6345 *length = min(logical + *length, 6346 raid56_full_stripe_start + em->start + 6347 (data_stripes << BTRFS_STRIPE_LEN_SHIFT)) - logical; 6348 stripe_index = 0; 6349 stripe_offset = 0; 6350 } else { 6351 /* 6352 * Mirror #0 or #1 means the original data block. 6353 * Mirror #2 is RAID5 parity block. 6354 * Mirror #3 is RAID6 Q block. 6355 */ 6356 stripe_index = stripe_nr % data_stripes; 6357 stripe_nr /= data_stripes; 6358 if (mirror_num > 1) 6359 stripe_index = data_stripes + mirror_num - 2; 6360 6361 /* We distribute the parity blocks across stripes */ 6362 stripe_index = (stripe_nr + stripe_index) % map->num_stripes; 6363 if (!need_full_stripe(op) && mirror_num <= 1) 6364 mirror_num = 1; 6365 } 6366 } else { 6367 /* 6368 * After this, stripe_nr is the number of stripes on this 6369 * device we have to walk to find the data, and stripe_index is 6370 * the number of our device in the stripe array 6371 */ 6372 stripe_index = stripe_nr % map->num_stripes; 6373 stripe_nr /= map->num_stripes; 6374 mirror_num = stripe_index + 1; 6375 } 6376 if (stripe_index >= map->num_stripes) { 6377 btrfs_crit(fs_info, 6378 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6379 stripe_index, map->num_stripes); 6380 ret = -EINVAL; 6381 goto out; 6382 } 6383 6384 num_alloc_stripes = num_stripes; 6385 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6386 op != BTRFS_MAP_READ) 6387 /* 6388 * For replace case, we need to add extra stripes for extra 6389 * duplicated stripes. 6390 * 6391 * For both WRITE and GET_READ_MIRRORS, we may have at most 6392 * 2 more stripes (DUP types, otherwise 1). 6393 */ 6394 num_alloc_stripes += 2; 6395 6396 /* 6397 * If this I/O maps to a single device, try to return the device and 6398 * physical block information on the stack instead of allocating an 6399 * I/O context structure. 6400 */ 6401 if (smap && num_alloc_stripes == 1 && 6402 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && 6403 (!need_full_stripe(op) || !dev_replace_is_ongoing || 6404 !dev_replace->tgtdev)) { 6405 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr); 6406 *mirror_num_ret = mirror_num; 6407 *bioc_ret = NULL; 6408 ret = 0; 6409 goto out; 6410 } 6411 6412 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes); 6413 if (!bioc) { 6414 ret = -ENOMEM; 6415 goto out; 6416 } 6417 bioc->map_type = map->type; 6418 6419 /* 6420 * For RAID56 full map, we need to make sure the stripes[] follows the 6421 * rule that data stripes are all ordered, then followed with P and Q 6422 * (if we have). 6423 * 6424 * It's still mostly the same as other profiles, just with extra rotation. 6425 */ 6426 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6427 (need_full_stripe(op) || mirror_num > 1)) { 6428 /* 6429 * For RAID56 @stripe_nr is already the number of full stripes 6430 * before us, which is also the rotation value (needs to modulo 6431 * with num_stripes). 6432 * 6433 * In this case, we just add @stripe_nr with @i, then do the 6434 * modulo, to reduce one modulo call. 6435 */ 6436 bioc->full_stripe_logical = em->start + 6437 ((stripe_nr * data_stripes) << BTRFS_STRIPE_LEN_SHIFT); 6438 for (i = 0; i < num_stripes; i++) 6439 set_io_stripe(&bioc->stripes[i], map, 6440 (i + stripe_nr) % num_stripes, 6441 stripe_offset, stripe_nr); 6442 } else { 6443 /* 6444 * For all other non-RAID56 profiles, just copy the target 6445 * stripe into the bioc. 6446 */ 6447 for (i = 0; i < num_stripes; i++) { 6448 set_io_stripe(&bioc->stripes[i], map, stripe_index, 6449 stripe_offset, stripe_nr); 6450 stripe_index++; 6451 } 6452 } 6453 6454 if (need_full_stripe(op)) 6455 max_errors = btrfs_chunk_max_errors(map); 6456 6457 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6458 need_full_stripe(op)) { 6459 handle_ops_on_dev_replace(op, bioc, dev_replace, logical, 6460 &num_stripes, &max_errors); 6461 } 6462 6463 *bioc_ret = bioc; 6464 bioc->num_stripes = num_stripes; 6465 bioc->max_errors = max_errors; 6466 bioc->mirror_num = mirror_num; 6467 6468 out: 6469 if (dev_replace_is_ongoing) { 6470 lockdep_assert_held(&dev_replace->rwsem); 6471 /* Unlock and let waiting writers proceed */ 6472 up_read(&dev_replace->rwsem); 6473 } 6474 free_extent_map(em); 6475 return ret; 6476 } 6477 6478 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6479 u64 logical, u64 *length, 6480 struct btrfs_io_context **bioc_ret, int mirror_num) 6481 { 6482 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6483 NULL, &mirror_num, 0); 6484 } 6485 6486 /* For Scrub/replace */ 6487 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6488 u64 logical, u64 *length, 6489 struct btrfs_io_context **bioc_ret) 6490 { 6491 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6492 NULL, NULL, 1); 6493 } 6494 6495 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6496 const struct btrfs_fs_devices *fs_devices) 6497 { 6498 if (args->fsid == NULL) 6499 return true; 6500 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6501 return true; 6502 return false; 6503 } 6504 6505 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6506 const struct btrfs_device *device) 6507 { 6508 if (args->missing) { 6509 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6510 !device->bdev) 6511 return true; 6512 return false; 6513 } 6514 6515 if (device->devid != args->devid) 6516 return false; 6517 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6518 return false; 6519 return true; 6520 } 6521 6522 /* 6523 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6524 * return NULL. 6525 * 6526 * If devid and uuid are both specified, the match must be exact, otherwise 6527 * only devid is used. 6528 */ 6529 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6530 const struct btrfs_dev_lookup_args *args) 6531 { 6532 struct btrfs_device *device; 6533 struct btrfs_fs_devices *seed_devs; 6534 6535 if (dev_args_match_fs_devices(args, fs_devices)) { 6536 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6537 if (dev_args_match_device(args, device)) 6538 return device; 6539 } 6540 } 6541 6542 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6543 if (!dev_args_match_fs_devices(args, seed_devs)) 6544 continue; 6545 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6546 if (dev_args_match_device(args, device)) 6547 return device; 6548 } 6549 } 6550 6551 return NULL; 6552 } 6553 6554 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6555 u64 devid, u8 *dev_uuid) 6556 { 6557 struct btrfs_device *device; 6558 unsigned int nofs_flag; 6559 6560 /* 6561 * We call this under the chunk_mutex, so we want to use NOFS for this 6562 * allocation, however we don't want to change btrfs_alloc_device() to 6563 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6564 * places. 6565 */ 6566 6567 nofs_flag = memalloc_nofs_save(); 6568 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); 6569 memalloc_nofs_restore(nofs_flag); 6570 if (IS_ERR(device)) 6571 return device; 6572 6573 list_add(&device->dev_list, &fs_devices->devices); 6574 device->fs_devices = fs_devices; 6575 fs_devices->num_devices++; 6576 6577 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6578 fs_devices->missing_devices++; 6579 6580 return device; 6581 } 6582 6583 /* 6584 * Allocate new device struct, set up devid and UUID. 6585 * 6586 * @fs_info: used only for generating a new devid, can be NULL if 6587 * devid is provided (i.e. @devid != NULL). 6588 * @devid: a pointer to devid for this device. If NULL a new devid 6589 * is generated. 6590 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6591 * is generated. 6592 * @path: a pointer to device path if available, NULL otherwise. 6593 * 6594 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6595 * on error. Returned struct is not linked onto any lists and must be 6596 * destroyed with btrfs_free_device. 6597 */ 6598 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6599 const u64 *devid, const u8 *uuid, 6600 const char *path) 6601 { 6602 struct btrfs_device *dev; 6603 u64 tmp; 6604 6605 if (WARN_ON(!devid && !fs_info)) 6606 return ERR_PTR(-EINVAL); 6607 6608 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6609 if (!dev) 6610 return ERR_PTR(-ENOMEM); 6611 6612 INIT_LIST_HEAD(&dev->dev_list); 6613 INIT_LIST_HEAD(&dev->dev_alloc_list); 6614 INIT_LIST_HEAD(&dev->post_commit_list); 6615 6616 atomic_set(&dev->dev_stats_ccnt, 0); 6617 btrfs_device_data_ordered_init(dev); 6618 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); 6619 6620 if (devid) 6621 tmp = *devid; 6622 else { 6623 int ret; 6624 6625 ret = find_next_devid(fs_info, &tmp); 6626 if (ret) { 6627 btrfs_free_device(dev); 6628 return ERR_PTR(ret); 6629 } 6630 } 6631 dev->devid = tmp; 6632 6633 if (uuid) 6634 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6635 else 6636 generate_random_uuid(dev->uuid); 6637 6638 if (path) { 6639 struct rcu_string *name; 6640 6641 name = rcu_string_strdup(path, GFP_KERNEL); 6642 if (!name) { 6643 btrfs_free_device(dev); 6644 return ERR_PTR(-ENOMEM); 6645 } 6646 rcu_assign_pointer(dev->name, name); 6647 } 6648 6649 return dev; 6650 } 6651 6652 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6653 u64 devid, u8 *uuid, bool error) 6654 { 6655 if (error) 6656 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6657 devid, uuid); 6658 else 6659 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6660 devid, uuid); 6661 } 6662 6663 u64 btrfs_calc_stripe_length(const struct extent_map *em) 6664 { 6665 const struct map_lookup *map = em->map_lookup; 6666 const int data_stripes = calc_data_stripes(map->type, map->num_stripes); 6667 6668 return div_u64(em->len, data_stripes); 6669 } 6670 6671 #if BITS_PER_LONG == 32 6672 /* 6673 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6674 * can't be accessed on 32bit systems. 6675 * 6676 * This function do mount time check to reject the fs if it already has 6677 * metadata chunk beyond that limit. 6678 */ 6679 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6680 u64 logical, u64 length, u64 type) 6681 { 6682 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6683 return 0; 6684 6685 if (logical + length < MAX_LFS_FILESIZE) 6686 return 0; 6687 6688 btrfs_err_32bit_limit(fs_info); 6689 return -EOVERFLOW; 6690 } 6691 6692 /* 6693 * This is to give early warning for any metadata chunk reaching 6694 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6695 * Although we can still access the metadata, it's not going to be possible 6696 * once the limit is reached. 6697 */ 6698 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6699 u64 logical, u64 length, u64 type) 6700 { 6701 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6702 return; 6703 6704 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 6705 return; 6706 6707 btrfs_warn_32bit_limit(fs_info); 6708 } 6709 #endif 6710 6711 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 6712 u64 devid, u8 *uuid) 6713 { 6714 struct btrfs_device *dev; 6715 6716 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6717 btrfs_report_missing_device(fs_info, devid, uuid, true); 6718 return ERR_PTR(-ENOENT); 6719 } 6720 6721 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 6722 if (IS_ERR(dev)) { 6723 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 6724 devid, PTR_ERR(dev)); 6725 return dev; 6726 } 6727 btrfs_report_missing_device(fs_info, devid, uuid, false); 6728 6729 return dev; 6730 } 6731 6732 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 6733 struct btrfs_chunk *chunk) 6734 { 6735 BTRFS_DEV_LOOKUP_ARGS(args); 6736 struct btrfs_fs_info *fs_info = leaf->fs_info; 6737 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 6738 struct map_lookup *map; 6739 struct extent_map *em; 6740 u64 logical; 6741 u64 length; 6742 u64 devid; 6743 u64 type; 6744 u8 uuid[BTRFS_UUID_SIZE]; 6745 int index; 6746 int num_stripes; 6747 int ret; 6748 int i; 6749 6750 logical = key->offset; 6751 length = btrfs_chunk_length(leaf, chunk); 6752 type = btrfs_chunk_type(leaf, chunk); 6753 index = btrfs_bg_flags_to_raid_index(type); 6754 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 6755 6756 #if BITS_PER_LONG == 32 6757 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 6758 if (ret < 0) 6759 return ret; 6760 warn_32bit_meta_chunk(fs_info, logical, length, type); 6761 #endif 6762 6763 /* 6764 * Only need to verify chunk item if we're reading from sys chunk array, 6765 * as chunk item in tree block is already verified by tree-checker. 6766 */ 6767 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 6768 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 6769 if (ret) 6770 return ret; 6771 } 6772 6773 read_lock(&map_tree->lock); 6774 em = lookup_extent_mapping(map_tree, logical, 1); 6775 read_unlock(&map_tree->lock); 6776 6777 /* already mapped? */ 6778 if (em && em->start <= logical && em->start + em->len > logical) { 6779 free_extent_map(em); 6780 return 0; 6781 } else if (em) { 6782 free_extent_map(em); 6783 } 6784 6785 em = alloc_extent_map(); 6786 if (!em) 6787 return -ENOMEM; 6788 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 6789 if (!map) { 6790 free_extent_map(em); 6791 return -ENOMEM; 6792 } 6793 6794 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 6795 em->map_lookup = map; 6796 em->start = logical; 6797 em->len = length; 6798 em->orig_start = 0; 6799 em->block_start = 0; 6800 em->block_len = em->len; 6801 6802 map->num_stripes = num_stripes; 6803 map->io_width = btrfs_chunk_io_width(leaf, chunk); 6804 map->io_align = btrfs_chunk_io_align(leaf, chunk); 6805 map->type = type; 6806 /* 6807 * We can't use the sub_stripes value, as for profiles other than 6808 * RAID10, they may have 0 as sub_stripes for filesystems created by 6809 * older mkfs (<v5.4). 6810 * In that case, it can cause divide-by-zero errors later. 6811 * Since currently sub_stripes is fixed for each profile, let's 6812 * use the trusted value instead. 6813 */ 6814 map->sub_stripes = btrfs_raid_array[index].sub_stripes; 6815 map->verified_stripes = 0; 6816 em->orig_block_len = btrfs_calc_stripe_length(em); 6817 for (i = 0; i < num_stripes; i++) { 6818 map->stripes[i].physical = 6819 btrfs_stripe_offset_nr(leaf, chunk, i); 6820 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 6821 args.devid = devid; 6822 read_extent_buffer(leaf, uuid, (unsigned long) 6823 btrfs_stripe_dev_uuid_nr(chunk, i), 6824 BTRFS_UUID_SIZE); 6825 args.uuid = uuid; 6826 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 6827 if (!map->stripes[i].dev) { 6828 map->stripes[i].dev = handle_missing_device(fs_info, 6829 devid, uuid); 6830 if (IS_ERR(map->stripes[i].dev)) { 6831 ret = PTR_ERR(map->stripes[i].dev); 6832 free_extent_map(em); 6833 return ret; 6834 } 6835 } 6836 6837 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 6838 &(map->stripes[i].dev->dev_state)); 6839 } 6840 6841 write_lock(&map_tree->lock); 6842 ret = add_extent_mapping(map_tree, em, 0); 6843 write_unlock(&map_tree->lock); 6844 if (ret < 0) { 6845 btrfs_err(fs_info, 6846 "failed to add chunk map, start=%llu len=%llu: %d", 6847 em->start, em->len, ret); 6848 } 6849 free_extent_map(em); 6850 6851 return ret; 6852 } 6853 6854 static void fill_device_from_item(struct extent_buffer *leaf, 6855 struct btrfs_dev_item *dev_item, 6856 struct btrfs_device *device) 6857 { 6858 unsigned long ptr; 6859 6860 device->devid = btrfs_device_id(leaf, dev_item); 6861 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 6862 device->total_bytes = device->disk_total_bytes; 6863 device->commit_total_bytes = device->disk_total_bytes; 6864 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 6865 device->commit_bytes_used = device->bytes_used; 6866 device->type = btrfs_device_type(leaf, dev_item); 6867 device->io_align = btrfs_device_io_align(leaf, dev_item); 6868 device->io_width = btrfs_device_io_width(leaf, dev_item); 6869 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 6870 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 6871 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 6872 6873 ptr = btrfs_device_uuid(dev_item); 6874 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 6875 } 6876 6877 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 6878 u8 *fsid) 6879 { 6880 struct btrfs_fs_devices *fs_devices; 6881 int ret; 6882 6883 lockdep_assert_held(&uuid_mutex); 6884 ASSERT(fsid); 6885 6886 /* This will match only for multi-device seed fs */ 6887 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 6888 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 6889 return fs_devices; 6890 6891 6892 fs_devices = find_fsid(fsid, NULL); 6893 if (!fs_devices) { 6894 if (!btrfs_test_opt(fs_info, DEGRADED)) 6895 return ERR_PTR(-ENOENT); 6896 6897 fs_devices = alloc_fs_devices(fsid, NULL); 6898 if (IS_ERR(fs_devices)) 6899 return fs_devices; 6900 6901 fs_devices->seeding = true; 6902 fs_devices->opened = 1; 6903 return fs_devices; 6904 } 6905 6906 /* 6907 * Upon first call for a seed fs fsid, just create a private copy of the 6908 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 6909 */ 6910 fs_devices = clone_fs_devices(fs_devices); 6911 if (IS_ERR(fs_devices)) 6912 return fs_devices; 6913 6914 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 6915 if (ret) { 6916 free_fs_devices(fs_devices); 6917 return ERR_PTR(ret); 6918 } 6919 6920 if (!fs_devices->seeding) { 6921 close_fs_devices(fs_devices); 6922 free_fs_devices(fs_devices); 6923 return ERR_PTR(-EINVAL); 6924 } 6925 6926 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 6927 6928 return fs_devices; 6929 } 6930 6931 static int read_one_dev(struct extent_buffer *leaf, 6932 struct btrfs_dev_item *dev_item) 6933 { 6934 BTRFS_DEV_LOOKUP_ARGS(args); 6935 struct btrfs_fs_info *fs_info = leaf->fs_info; 6936 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 6937 struct btrfs_device *device; 6938 u64 devid; 6939 int ret; 6940 u8 fs_uuid[BTRFS_FSID_SIZE]; 6941 u8 dev_uuid[BTRFS_UUID_SIZE]; 6942 6943 devid = btrfs_device_id(leaf, dev_item); 6944 args.devid = devid; 6945 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 6946 BTRFS_UUID_SIZE); 6947 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 6948 BTRFS_FSID_SIZE); 6949 args.uuid = dev_uuid; 6950 args.fsid = fs_uuid; 6951 6952 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 6953 fs_devices = open_seed_devices(fs_info, fs_uuid); 6954 if (IS_ERR(fs_devices)) 6955 return PTR_ERR(fs_devices); 6956 } 6957 6958 device = btrfs_find_device(fs_info->fs_devices, &args); 6959 if (!device) { 6960 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6961 btrfs_report_missing_device(fs_info, devid, 6962 dev_uuid, true); 6963 return -ENOENT; 6964 } 6965 6966 device = add_missing_dev(fs_devices, devid, dev_uuid); 6967 if (IS_ERR(device)) { 6968 btrfs_err(fs_info, 6969 "failed to add missing dev %llu: %ld", 6970 devid, PTR_ERR(device)); 6971 return PTR_ERR(device); 6972 } 6973 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 6974 } else { 6975 if (!device->bdev) { 6976 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6977 btrfs_report_missing_device(fs_info, 6978 devid, dev_uuid, true); 6979 return -ENOENT; 6980 } 6981 btrfs_report_missing_device(fs_info, devid, 6982 dev_uuid, false); 6983 } 6984 6985 if (!device->bdev && 6986 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 6987 /* 6988 * this happens when a device that was properly setup 6989 * in the device info lists suddenly goes bad. 6990 * device->bdev is NULL, and so we have to set 6991 * device->missing to one here 6992 */ 6993 device->fs_devices->missing_devices++; 6994 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6995 } 6996 6997 /* Move the device to its own fs_devices */ 6998 if (device->fs_devices != fs_devices) { 6999 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7000 &device->dev_state)); 7001 7002 list_move(&device->dev_list, &fs_devices->devices); 7003 device->fs_devices->num_devices--; 7004 fs_devices->num_devices++; 7005 7006 device->fs_devices->missing_devices--; 7007 fs_devices->missing_devices++; 7008 7009 device->fs_devices = fs_devices; 7010 } 7011 } 7012 7013 if (device->fs_devices != fs_info->fs_devices) { 7014 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7015 if (device->generation != 7016 btrfs_device_generation(leaf, dev_item)) 7017 return -EINVAL; 7018 } 7019 7020 fill_device_from_item(leaf, dev_item, device); 7021 if (device->bdev) { 7022 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7023 7024 if (device->total_bytes > max_total_bytes) { 7025 btrfs_err(fs_info, 7026 "device total_bytes should be at most %llu but found %llu", 7027 max_total_bytes, device->total_bytes); 7028 return -EINVAL; 7029 } 7030 } 7031 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7032 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7033 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7034 device->fs_devices->total_rw_bytes += device->total_bytes; 7035 atomic64_add(device->total_bytes - device->bytes_used, 7036 &fs_info->free_chunk_space); 7037 } 7038 ret = 0; 7039 return ret; 7040 } 7041 7042 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7043 { 7044 struct btrfs_super_block *super_copy = fs_info->super_copy; 7045 struct extent_buffer *sb; 7046 struct btrfs_disk_key *disk_key; 7047 struct btrfs_chunk *chunk; 7048 u8 *array_ptr; 7049 unsigned long sb_array_offset; 7050 int ret = 0; 7051 u32 num_stripes; 7052 u32 array_size; 7053 u32 len = 0; 7054 u32 cur_offset; 7055 u64 type; 7056 struct btrfs_key key; 7057 7058 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7059 7060 /* 7061 * We allocated a dummy extent, just to use extent buffer accessors. 7062 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but 7063 * that's fine, we will not go beyond system chunk array anyway. 7064 */ 7065 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); 7066 if (!sb) 7067 return -ENOMEM; 7068 set_extent_buffer_uptodate(sb); 7069 7070 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7071 array_size = btrfs_super_sys_array_size(super_copy); 7072 7073 array_ptr = super_copy->sys_chunk_array; 7074 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7075 cur_offset = 0; 7076 7077 while (cur_offset < array_size) { 7078 disk_key = (struct btrfs_disk_key *)array_ptr; 7079 len = sizeof(*disk_key); 7080 if (cur_offset + len > array_size) 7081 goto out_short_read; 7082 7083 btrfs_disk_key_to_cpu(&key, disk_key); 7084 7085 array_ptr += len; 7086 sb_array_offset += len; 7087 cur_offset += len; 7088 7089 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7090 btrfs_err(fs_info, 7091 "unexpected item type %u in sys_array at offset %u", 7092 (u32)key.type, cur_offset); 7093 ret = -EIO; 7094 break; 7095 } 7096 7097 chunk = (struct btrfs_chunk *)sb_array_offset; 7098 /* 7099 * At least one btrfs_chunk with one stripe must be present, 7100 * exact stripe count check comes afterwards 7101 */ 7102 len = btrfs_chunk_item_size(1); 7103 if (cur_offset + len > array_size) 7104 goto out_short_read; 7105 7106 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7107 if (!num_stripes) { 7108 btrfs_err(fs_info, 7109 "invalid number of stripes %u in sys_array at offset %u", 7110 num_stripes, cur_offset); 7111 ret = -EIO; 7112 break; 7113 } 7114 7115 type = btrfs_chunk_type(sb, chunk); 7116 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7117 btrfs_err(fs_info, 7118 "invalid chunk type %llu in sys_array at offset %u", 7119 type, cur_offset); 7120 ret = -EIO; 7121 break; 7122 } 7123 7124 len = btrfs_chunk_item_size(num_stripes); 7125 if (cur_offset + len > array_size) 7126 goto out_short_read; 7127 7128 ret = read_one_chunk(&key, sb, chunk); 7129 if (ret) 7130 break; 7131 7132 array_ptr += len; 7133 sb_array_offset += len; 7134 cur_offset += len; 7135 } 7136 clear_extent_buffer_uptodate(sb); 7137 free_extent_buffer_stale(sb); 7138 return ret; 7139 7140 out_short_read: 7141 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7142 len, cur_offset); 7143 clear_extent_buffer_uptodate(sb); 7144 free_extent_buffer_stale(sb); 7145 return -EIO; 7146 } 7147 7148 /* 7149 * Check if all chunks in the fs are OK for read-write degraded mount 7150 * 7151 * If the @failing_dev is specified, it's accounted as missing. 7152 * 7153 * Return true if all chunks meet the minimal RW mount requirements. 7154 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7155 */ 7156 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7157 struct btrfs_device *failing_dev) 7158 { 7159 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7160 struct extent_map *em; 7161 u64 next_start = 0; 7162 bool ret = true; 7163 7164 read_lock(&map_tree->lock); 7165 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7166 read_unlock(&map_tree->lock); 7167 /* No chunk at all? Return false anyway */ 7168 if (!em) { 7169 ret = false; 7170 goto out; 7171 } 7172 while (em) { 7173 struct map_lookup *map; 7174 int missing = 0; 7175 int max_tolerated; 7176 int i; 7177 7178 map = em->map_lookup; 7179 max_tolerated = 7180 btrfs_get_num_tolerated_disk_barrier_failures( 7181 map->type); 7182 for (i = 0; i < map->num_stripes; i++) { 7183 struct btrfs_device *dev = map->stripes[i].dev; 7184 7185 if (!dev || !dev->bdev || 7186 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7187 dev->last_flush_error) 7188 missing++; 7189 else if (failing_dev && failing_dev == dev) 7190 missing++; 7191 } 7192 if (missing > max_tolerated) { 7193 if (!failing_dev) 7194 btrfs_warn(fs_info, 7195 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7196 em->start, missing, max_tolerated); 7197 free_extent_map(em); 7198 ret = false; 7199 goto out; 7200 } 7201 next_start = extent_map_end(em); 7202 free_extent_map(em); 7203 7204 read_lock(&map_tree->lock); 7205 em = lookup_extent_mapping(map_tree, next_start, 7206 (u64)(-1) - next_start); 7207 read_unlock(&map_tree->lock); 7208 } 7209 out: 7210 return ret; 7211 } 7212 7213 static void readahead_tree_node_children(struct extent_buffer *node) 7214 { 7215 int i; 7216 const int nr_items = btrfs_header_nritems(node); 7217 7218 for (i = 0; i < nr_items; i++) 7219 btrfs_readahead_node_child(node, i); 7220 } 7221 7222 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7223 { 7224 struct btrfs_root *root = fs_info->chunk_root; 7225 struct btrfs_path *path; 7226 struct extent_buffer *leaf; 7227 struct btrfs_key key; 7228 struct btrfs_key found_key; 7229 int ret; 7230 int slot; 7231 int iter_ret = 0; 7232 u64 total_dev = 0; 7233 u64 last_ra_node = 0; 7234 7235 path = btrfs_alloc_path(); 7236 if (!path) 7237 return -ENOMEM; 7238 7239 /* 7240 * uuid_mutex is needed only if we are mounting a sprout FS 7241 * otherwise we don't need it. 7242 */ 7243 mutex_lock(&uuid_mutex); 7244 7245 /* 7246 * It is possible for mount and umount to race in such a way that 7247 * we execute this code path, but open_fs_devices failed to clear 7248 * total_rw_bytes. We certainly want it cleared before reading the 7249 * device items, so clear it here. 7250 */ 7251 fs_info->fs_devices->total_rw_bytes = 0; 7252 7253 /* 7254 * Lockdep complains about possible circular locking dependency between 7255 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7256 * used for freeze procection of a fs (struct super_block.s_writers), 7257 * which we take when starting a transaction, and extent buffers of the 7258 * chunk tree if we call read_one_dev() while holding a lock on an 7259 * extent buffer of the chunk tree. Since we are mounting the filesystem 7260 * and at this point there can't be any concurrent task modifying the 7261 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7262 */ 7263 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7264 path->skip_locking = 1; 7265 7266 /* 7267 * Read all device items, and then all the chunk items. All 7268 * device items are found before any chunk item (their object id 7269 * is smaller than the lowest possible object id for a chunk 7270 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7271 */ 7272 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7273 key.offset = 0; 7274 key.type = 0; 7275 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 7276 struct extent_buffer *node = path->nodes[1]; 7277 7278 leaf = path->nodes[0]; 7279 slot = path->slots[0]; 7280 7281 if (node) { 7282 if (last_ra_node != node->start) { 7283 readahead_tree_node_children(node); 7284 last_ra_node = node->start; 7285 } 7286 } 7287 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7288 struct btrfs_dev_item *dev_item; 7289 dev_item = btrfs_item_ptr(leaf, slot, 7290 struct btrfs_dev_item); 7291 ret = read_one_dev(leaf, dev_item); 7292 if (ret) 7293 goto error; 7294 total_dev++; 7295 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7296 struct btrfs_chunk *chunk; 7297 7298 /* 7299 * We are only called at mount time, so no need to take 7300 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7301 * we always lock first fs_info->chunk_mutex before 7302 * acquiring any locks on the chunk tree. This is a 7303 * requirement for chunk allocation, see the comment on 7304 * top of btrfs_chunk_alloc() for details. 7305 */ 7306 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7307 ret = read_one_chunk(&found_key, leaf, chunk); 7308 if (ret) 7309 goto error; 7310 } 7311 } 7312 /* Catch error found during iteration */ 7313 if (iter_ret < 0) { 7314 ret = iter_ret; 7315 goto error; 7316 } 7317 7318 /* 7319 * After loading chunk tree, we've got all device information, 7320 * do another round of validation checks. 7321 */ 7322 if (total_dev != fs_info->fs_devices->total_devices) { 7323 btrfs_warn(fs_info, 7324 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", 7325 btrfs_super_num_devices(fs_info->super_copy), 7326 total_dev); 7327 fs_info->fs_devices->total_devices = total_dev; 7328 btrfs_set_super_num_devices(fs_info->super_copy, total_dev); 7329 } 7330 if (btrfs_super_total_bytes(fs_info->super_copy) < 7331 fs_info->fs_devices->total_rw_bytes) { 7332 btrfs_err(fs_info, 7333 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7334 btrfs_super_total_bytes(fs_info->super_copy), 7335 fs_info->fs_devices->total_rw_bytes); 7336 ret = -EINVAL; 7337 goto error; 7338 } 7339 ret = 0; 7340 error: 7341 mutex_unlock(&uuid_mutex); 7342 7343 btrfs_free_path(path); 7344 return ret; 7345 } 7346 7347 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7348 { 7349 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7350 struct btrfs_device *device; 7351 int ret = 0; 7352 7353 fs_devices->fs_info = fs_info; 7354 7355 mutex_lock(&fs_devices->device_list_mutex); 7356 list_for_each_entry(device, &fs_devices->devices, dev_list) 7357 device->fs_info = fs_info; 7358 7359 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7360 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7361 device->fs_info = fs_info; 7362 ret = btrfs_get_dev_zone_info(device, false); 7363 if (ret) 7364 break; 7365 } 7366 7367 seed_devs->fs_info = fs_info; 7368 } 7369 mutex_unlock(&fs_devices->device_list_mutex); 7370 7371 return ret; 7372 } 7373 7374 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7375 const struct btrfs_dev_stats_item *ptr, 7376 int index) 7377 { 7378 u64 val; 7379 7380 read_extent_buffer(eb, &val, 7381 offsetof(struct btrfs_dev_stats_item, values) + 7382 ((unsigned long)ptr) + (index * sizeof(u64)), 7383 sizeof(val)); 7384 return val; 7385 } 7386 7387 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7388 struct btrfs_dev_stats_item *ptr, 7389 int index, u64 val) 7390 { 7391 write_extent_buffer(eb, &val, 7392 offsetof(struct btrfs_dev_stats_item, values) + 7393 ((unsigned long)ptr) + (index * sizeof(u64)), 7394 sizeof(val)); 7395 } 7396 7397 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7398 struct btrfs_path *path) 7399 { 7400 struct btrfs_dev_stats_item *ptr; 7401 struct extent_buffer *eb; 7402 struct btrfs_key key; 7403 int item_size; 7404 int i, ret, slot; 7405 7406 if (!device->fs_info->dev_root) 7407 return 0; 7408 7409 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7410 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7411 key.offset = device->devid; 7412 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7413 if (ret) { 7414 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7415 btrfs_dev_stat_set(device, i, 0); 7416 device->dev_stats_valid = 1; 7417 btrfs_release_path(path); 7418 return ret < 0 ? ret : 0; 7419 } 7420 slot = path->slots[0]; 7421 eb = path->nodes[0]; 7422 item_size = btrfs_item_size(eb, slot); 7423 7424 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7425 7426 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7427 if (item_size >= (1 + i) * sizeof(__le64)) 7428 btrfs_dev_stat_set(device, i, 7429 btrfs_dev_stats_value(eb, ptr, i)); 7430 else 7431 btrfs_dev_stat_set(device, i, 0); 7432 } 7433 7434 device->dev_stats_valid = 1; 7435 btrfs_dev_stat_print_on_load(device); 7436 btrfs_release_path(path); 7437 7438 return 0; 7439 } 7440 7441 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7442 { 7443 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7444 struct btrfs_device *device; 7445 struct btrfs_path *path = NULL; 7446 int ret = 0; 7447 7448 path = btrfs_alloc_path(); 7449 if (!path) 7450 return -ENOMEM; 7451 7452 mutex_lock(&fs_devices->device_list_mutex); 7453 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7454 ret = btrfs_device_init_dev_stats(device, path); 7455 if (ret) 7456 goto out; 7457 } 7458 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7459 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7460 ret = btrfs_device_init_dev_stats(device, path); 7461 if (ret) 7462 goto out; 7463 } 7464 } 7465 out: 7466 mutex_unlock(&fs_devices->device_list_mutex); 7467 7468 btrfs_free_path(path); 7469 return ret; 7470 } 7471 7472 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7473 struct btrfs_device *device) 7474 { 7475 struct btrfs_fs_info *fs_info = trans->fs_info; 7476 struct btrfs_root *dev_root = fs_info->dev_root; 7477 struct btrfs_path *path; 7478 struct btrfs_key key; 7479 struct extent_buffer *eb; 7480 struct btrfs_dev_stats_item *ptr; 7481 int ret; 7482 int i; 7483 7484 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7485 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7486 key.offset = device->devid; 7487 7488 path = btrfs_alloc_path(); 7489 if (!path) 7490 return -ENOMEM; 7491 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7492 if (ret < 0) { 7493 btrfs_warn_in_rcu(fs_info, 7494 "error %d while searching for dev_stats item for device %s", 7495 ret, btrfs_dev_name(device)); 7496 goto out; 7497 } 7498 7499 if (ret == 0 && 7500 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7501 /* need to delete old one and insert a new one */ 7502 ret = btrfs_del_item(trans, dev_root, path); 7503 if (ret != 0) { 7504 btrfs_warn_in_rcu(fs_info, 7505 "delete too small dev_stats item for device %s failed %d", 7506 btrfs_dev_name(device), ret); 7507 goto out; 7508 } 7509 ret = 1; 7510 } 7511 7512 if (ret == 1) { 7513 /* need to insert a new item */ 7514 btrfs_release_path(path); 7515 ret = btrfs_insert_empty_item(trans, dev_root, path, 7516 &key, sizeof(*ptr)); 7517 if (ret < 0) { 7518 btrfs_warn_in_rcu(fs_info, 7519 "insert dev_stats item for device %s failed %d", 7520 btrfs_dev_name(device), ret); 7521 goto out; 7522 } 7523 } 7524 7525 eb = path->nodes[0]; 7526 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7527 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7528 btrfs_set_dev_stats_value(eb, ptr, i, 7529 btrfs_dev_stat_read(device, i)); 7530 btrfs_mark_buffer_dirty(eb); 7531 7532 out: 7533 btrfs_free_path(path); 7534 return ret; 7535 } 7536 7537 /* 7538 * called from commit_transaction. Writes all changed device stats to disk. 7539 */ 7540 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7541 { 7542 struct btrfs_fs_info *fs_info = trans->fs_info; 7543 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7544 struct btrfs_device *device; 7545 int stats_cnt; 7546 int ret = 0; 7547 7548 mutex_lock(&fs_devices->device_list_mutex); 7549 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7550 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7551 if (!device->dev_stats_valid || stats_cnt == 0) 7552 continue; 7553 7554 7555 /* 7556 * There is a LOAD-LOAD control dependency between the value of 7557 * dev_stats_ccnt and updating the on-disk values which requires 7558 * reading the in-memory counters. Such control dependencies 7559 * require explicit read memory barriers. 7560 * 7561 * This memory barriers pairs with smp_mb__before_atomic in 7562 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7563 * barrier implied by atomic_xchg in 7564 * btrfs_dev_stats_read_and_reset 7565 */ 7566 smp_rmb(); 7567 7568 ret = update_dev_stat_item(trans, device); 7569 if (!ret) 7570 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7571 } 7572 mutex_unlock(&fs_devices->device_list_mutex); 7573 7574 return ret; 7575 } 7576 7577 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7578 { 7579 btrfs_dev_stat_inc(dev, index); 7580 7581 if (!dev->dev_stats_valid) 7582 return; 7583 btrfs_err_rl_in_rcu(dev->fs_info, 7584 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7585 btrfs_dev_name(dev), 7586 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7587 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7588 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7589 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7590 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7591 } 7592 7593 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7594 { 7595 int i; 7596 7597 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7598 if (btrfs_dev_stat_read(dev, i) != 0) 7599 break; 7600 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7601 return; /* all values == 0, suppress message */ 7602 7603 btrfs_info_in_rcu(dev->fs_info, 7604 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7605 btrfs_dev_name(dev), 7606 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7607 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7608 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7609 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7610 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7611 } 7612 7613 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7614 struct btrfs_ioctl_get_dev_stats *stats) 7615 { 7616 BTRFS_DEV_LOOKUP_ARGS(args); 7617 struct btrfs_device *dev; 7618 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7619 int i; 7620 7621 mutex_lock(&fs_devices->device_list_mutex); 7622 args.devid = stats->devid; 7623 dev = btrfs_find_device(fs_info->fs_devices, &args); 7624 mutex_unlock(&fs_devices->device_list_mutex); 7625 7626 if (!dev) { 7627 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7628 return -ENODEV; 7629 } else if (!dev->dev_stats_valid) { 7630 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7631 return -ENODEV; 7632 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7633 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7634 if (stats->nr_items > i) 7635 stats->values[i] = 7636 btrfs_dev_stat_read_and_reset(dev, i); 7637 else 7638 btrfs_dev_stat_set(dev, i, 0); 7639 } 7640 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7641 current->comm, task_pid_nr(current)); 7642 } else { 7643 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7644 if (stats->nr_items > i) 7645 stats->values[i] = btrfs_dev_stat_read(dev, i); 7646 } 7647 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7648 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7649 return 0; 7650 } 7651 7652 /* 7653 * Update the size and bytes used for each device where it changed. This is 7654 * delayed since we would otherwise get errors while writing out the 7655 * superblocks. 7656 * 7657 * Must be invoked during transaction commit. 7658 */ 7659 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7660 { 7661 struct btrfs_device *curr, *next; 7662 7663 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7664 7665 if (list_empty(&trans->dev_update_list)) 7666 return; 7667 7668 /* 7669 * We don't need the device_list_mutex here. This list is owned by the 7670 * transaction and the transaction must complete before the device is 7671 * released. 7672 */ 7673 mutex_lock(&trans->fs_info->chunk_mutex); 7674 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7675 post_commit_list) { 7676 list_del_init(&curr->post_commit_list); 7677 curr->commit_total_bytes = curr->disk_total_bytes; 7678 curr->commit_bytes_used = curr->bytes_used; 7679 } 7680 mutex_unlock(&trans->fs_info->chunk_mutex); 7681 } 7682 7683 /* 7684 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7685 */ 7686 int btrfs_bg_type_to_factor(u64 flags) 7687 { 7688 const int index = btrfs_bg_flags_to_raid_index(flags); 7689 7690 return btrfs_raid_array[index].ncopies; 7691 } 7692 7693 7694 7695 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7696 u64 chunk_offset, u64 devid, 7697 u64 physical_offset, u64 physical_len) 7698 { 7699 struct btrfs_dev_lookup_args args = { .devid = devid }; 7700 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7701 struct extent_map *em; 7702 struct map_lookup *map; 7703 struct btrfs_device *dev; 7704 u64 stripe_len; 7705 bool found = false; 7706 int ret = 0; 7707 int i; 7708 7709 read_lock(&em_tree->lock); 7710 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7711 read_unlock(&em_tree->lock); 7712 7713 if (!em) { 7714 btrfs_err(fs_info, 7715 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 7716 physical_offset, devid); 7717 ret = -EUCLEAN; 7718 goto out; 7719 } 7720 7721 map = em->map_lookup; 7722 stripe_len = btrfs_calc_stripe_length(em); 7723 if (physical_len != stripe_len) { 7724 btrfs_err(fs_info, 7725 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 7726 physical_offset, devid, em->start, physical_len, 7727 stripe_len); 7728 ret = -EUCLEAN; 7729 goto out; 7730 } 7731 7732 /* 7733 * Very old mkfs.btrfs (before v4.1) will not respect the reserved 7734 * space. Although kernel can handle it without problem, better to warn 7735 * the users. 7736 */ 7737 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) 7738 btrfs_warn(fs_info, 7739 "devid %llu physical %llu len %llu inside the reserved space", 7740 devid, physical_offset, physical_len); 7741 7742 for (i = 0; i < map->num_stripes; i++) { 7743 if (map->stripes[i].dev->devid == devid && 7744 map->stripes[i].physical == physical_offset) { 7745 found = true; 7746 if (map->verified_stripes >= map->num_stripes) { 7747 btrfs_err(fs_info, 7748 "too many dev extents for chunk %llu found", 7749 em->start); 7750 ret = -EUCLEAN; 7751 goto out; 7752 } 7753 map->verified_stripes++; 7754 break; 7755 } 7756 } 7757 if (!found) { 7758 btrfs_err(fs_info, 7759 "dev extent physical offset %llu devid %llu has no corresponding chunk", 7760 physical_offset, devid); 7761 ret = -EUCLEAN; 7762 } 7763 7764 /* Make sure no dev extent is beyond device boundary */ 7765 dev = btrfs_find_device(fs_info->fs_devices, &args); 7766 if (!dev) { 7767 btrfs_err(fs_info, "failed to find devid %llu", devid); 7768 ret = -EUCLEAN; 7769 goto out; 7770 } 7771 7772 if (physical_offset + physical_len > dev->disk_total_bytes) { 7773 btrfs_err(fs_info, 7774 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 7775 devid, physical_offset, physical_len, 7776 dev->disk_total_bytes); 7777 ret = -EUCLEAN; 7778 goto out; 7779 } 7780 7781 if (dev->zone_info) { 7782 u64 zone_size = dev->zone_info->zone_size; 7783 7784 if (!IS_ALIGNED(physical_offset, zone_size) || 7785 !IS_ALIGNED(physical_len, zone_size)) { 7786 btrfs_err(fs_info, 7787 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 7788 devid, physical_offset, physical_len); 7789 ret = -EUCLEAN; 7790 goto out; 7791 } 7792 } 7793 7794 out: 7795 free_extent_map(em); 7796 return ret; 7797 } 7798 7799 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 7800 { 7801 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7802 struct extent_map *em; 7803 struct rb_node *node; 7804 int ret = 0; 7805 7806 read_lock(&em_tree->lock); 7807 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 7808 em = rb_entry(node, struct extent_map, rb_node); 7809 if (em->map_lookup->num_stripes != 7810 em->map_lookup->verified_stripes) { 7811 btrfs_err(fs_info, 7812 "chunk %llu has missing dev extent, have %d expect %d", 7813 em->start, em->map_lookup->verified_stripes, 7814 em->map_lookup->num_stripes); 7815 ret = -EUCLEAN; 7816 goto out; 7817 } 7818 } 7819 out: 7820 read_unlock(&em_tree->lock); 7821 return ret; 7822 } 7823 7824 /* 7825 * Ensure that all dev extents are mapped to correct chunk, otherwise 7826 * later chunk allocation/free would cause unexpected behavior. 7827 * 7828 * NOTE: This will iterate through the whole device tree, which should be of 7829 * the same size level as the chunk tree. This slightly increases mount time. 7830 */ 7831 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 7832 { 7833 struct btrfs_path *path; 7834 struct btrfs_root *root = fs_info->dev_root; 7835 struct btrfs_key key; 7836 u64 prev_devid = 0; 7837 u64 prev_dev_ext_end = 0; 7838 int ret = 0; 7839 7840 /* 7841 * We don't have a dev_root because we mounted with ignorebadroots and 7842 * failed to load the root, so we want to skip the verification in this 7843 * case for sure. 7844 * 7845 * However if the dev root is fine, but the tree itself is corrupted 7846 * we'd still fail to mount. This verification is only to make sure 7847 * writes can happen safely, so instead just bypass this check 7848 * completely in the case of IGNOREBADROOTS. 7849 */ 7850 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 7851 return 0; 7852 7853 key.objectid = 1; 7854 key.type = BTRFS_DEV_EXTENT_KEY; 7855 key.offset = 0; 7856 7857 path = btrfs_alloc_path(); 7858 if (!path) 7859 return -ENOMEM; 7860 7861 path->reada = READA_FORWARD; 7862 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7863 if (ret < 0) 7864 goto out; 7865 7866 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 7867 ret = btrfs_next_leaf(root, path); 7868 if (ret < 0) 7869 goto out; 7870 /* No dev extents at all? Not good */ 7871 if (ret > 0) { 7872 ret = -EUCLEAN; 7873 goto out; 7874 } 7875 } 7876 while (1) { 7877 struct extent_buffer *leaf = path->nodes[0]; 7878 struct btrfs_dev_extent *dext; 7879 int slot = path->slots[0]; 7880 u64 chunk_offset; 7881 u64 physical_offset; 7882 u64 physical_len; 7883 u64 devid; 7884 7885 btrfs_item_key_to_cpu(leaf, &key, slot); 7886 if (key.type != BTRFS_DEV_EXTENT_KEY) 7887 break; 7888 devid = key.objectid; 7889 physical_offset = key.offset; 7890 7891 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 7892 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 7893 physical_len = btrfs_dev_extent_length(leaf, dext); 7894 7895 /* Check if this dev extent overlaps with the previous one */ 7896 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 7897 btrfs_err(fs_info, 7898 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 7899 devid, physical_offset, prev_dev_ext_end); 7900 ret = -EUCLEAN; 7901 goto out; 7902 } 7903 7904 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 7905 physical_offset, physical_len); 7906 if (ret < 0) 7907 goto out; 7908 prev_devid = devid; 7909 prev_dev_ext_end = physical_offset + physical_len; 7910 7911 ret = btrfs_next_item(root, path); 7912 if (ret < 0) 7913 goto out; 7914 if (ret > 0) { 7915 ret = 0; 7916 break; 7917 } 7918 } 7919 7920 /* Ensure all chunks have corresponding dev extents */ 7921 ret = verify_chunk_dev_extent_mapping(fs_info); 7922 out: 7923 btrfs_free_path(path); 7924 return ret; 7925 } 7926 7927 /* 7928 * Check whether the given block group or device is pinned by any inode being 7929 * used as a swapfile. 7930 */ 7931 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 7932 { 7933 struct btrfs_swapfile_pin *sp; 7934 struct rb_node *node; 7935 7936 spin_lock(&fs_info->swapfile_pins_lock); 7937 node = fs_info->swapfile_pins.rb_node; 7938 while (node) { 7939 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 7940 if (ptr < sp->ptr) 7941 node = node->rb_left; 7942 else if (ptr > sp->ptr) 7943 node = node->rb_right; 7944 else 7945 break; 7946 } 7947 spin_unlock(&fs_info->swapfile_pins_lock); 7948 return node != NULL; 7949 } 7950 7951 static int relocating_repair_kthread(void *data) 7952 { 7953 struct btrfs_block_group *cache = data; 7954 struct btrfs_fs_info *fs_info = cache->fs_info; 7955 u64 target; 7956 int ret = 0; 7957 7958 target = cache->start; 7959 btrfs_put_block_group(cache); 7960 7961 sb_start_write(fs_info->sb); 7962 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 7963 btrfs_info(fs_info, 7964 "zoned: skip relocating block group %llu to repair: EBUSY", 7965 target); 7966 sb_end_write(fs_info->sb); 7967 return -EBUSY; 7968 } 7969 7970 mutex_lock(&fs_info->reclaim_bgs_lock); 7971 7972 /* Ensure block group still exists */ 7973 cache = btrfs_lookup_block_group(fs_info, target); 7974 if (!cache) 7975 goto out; 7976 7977 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) 7978 goto out; 7979 7980 ret = btrfs_may_alloc_data_chunk(fs_info, target); 7981 if (ret < 0) 7982 goto out; 7983 7984 btrfs_info(fs_info, 7985 "zoned: relocating block group %llu to repair IO failure", 7986 target); 7987 ret = btrfs_relocate_chunk(fs_info, target); 7988 7989 out: 7990 if (cache) 7991 btrfs_put_block_group(cache); 7992 mutex_unlock(&fs_info->reclaim_bgs_lock); 7993 btrfs_exclop_finish(fs_info); 7994 sb_end_write(fs_info->sb); 7995 7996 return ret; 7997 } 7998 7999 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8000 { 8001 struct btrfs_block_group *cache; 8002 8003 if (!btrfs_is_zoned(fs_info)) 8004 return false; 8005 8006 /* Do not attempt to repair in degraded state */ 8007 if (btrfs_test_opt(fs_info, DEGRADED)) 8008 return true; 8009 8010 cache = btrfs_lookup_block_group(fs_info, logical); 8011 if (!cache) 8012 return true; 8013 8014 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { 8015 btrfs_put_block_group(cache); 8016 return true; 8017 } 8018 8019 kthread_run(relocating_repair_kthread, cache, 8020 "btrfs-relocating-repair"); 8021 8022 return true; 8023 } 8024 8025 static void map_raid56_repair_block(struct btrfs_io_context *bioc, 8026 struct btrfs_io_stripe *smap, 8027 u64 logical) 8028 { 8029 int data_stripes = nr_bioc_data_stripes(bioc); 8030 int i; 8031 8032 for (i = 0; i < data_stripes; i++) { 8033 u64 stripe_start = bioc->full_stripe_logical + 8034 (i << BTRFS_STRIPE_LEN_SHIFT); 8035 8036 if (logical >= stripe_start && 8037 logical < stripe_start + BTRFS_STRIPE_LEN) 8038 break; 8039 } 8040 ASSERT(i < data_stripes); 8041 smap->dev = bioc->stripes[i].dev; 8042 smap->physical = bioc->stripes[i].physical + 8043 ((logical - bioc->full_stripe_logical) & 8044 BTRFS_STRIPE_LEN_MASK); 8045 } 8046 8047 /* 8048 * Map a repair write into a single device. 8049 * 8050 * A repair write is triggered by read time repair or scrub, which would only 8051 * update the contents of a single device. 8052 * Not update any other mirrors nor go through RMW path. 8053 * 8054 * Callers should ensure: 8055 * 8056 * - Call btrfs_bio_counter_inc_blocked() first 8057 * - The range does not cross stripe boundary 8058 * - Has a valid @mirror_num passed in. 8059 */ 8060 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info, 8061 struct btrfs_io_stripe *smap, u64 logical, 8062 u32 length, int mirror_num) 8063 { 8064 struct btrfs_io_context *bioc = NULL; 8065 u64 map_length = length; 8066 int mirror_ret = mirror_num; 8067 int ret; 8068 8069 ASSERT(mirror_num > 0); 8070 8071 ret = __btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length, 8072 &bioc, smap, &mirror_ret, true); 8073 if (ret < 0) 8074 return ret; 8075 8076 /* The map range should not cross stripe boundary. */ 8077 ASSERT(map_length >= length); 8078 8079 /* Already mapped to single stripe. */ 8080 if (!bioc) 8081 goto out; 8082 8083 /* Map the RAID56 multi-stripe writes to a single one. */ 8084 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 8085 map_raid56_repair_block(bioc, smap, logical); 8086 goto out; 8087 } 8088 8089 ASSERT(mirror_num <= bioc->num_stripes); 8090 smap->dev = bioc->stripes[mirror_num - 1].dev; 8091 smap->physical = bioc->stripes[mirror_num - 1].physical; 8092 out: 8093 btrfs_put_bioc(bioc); 8094 ASSERT(smap->dev); 8095 return 0; 8096 } 8097