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 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 374 if (!fs_devs) 375 return ERR_PTR(-ENOMEM); 376 377 mutex_init(&fs_devs->device_list_mutex); 378 379 INIT_LIST_HEAD(&fs_devs->devices); 380 INIT_LIST_HEAD(&fs_devs->alloc_list); 381 INIT_LIST_HEAD(&fs_devs->fs_list); 382 INIT_LIST_HEAD(&fs_devs->seed_list); 383 if (fsid) 384 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 385 386 if (metadata_fsid) 387 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); 388 else if (fsid) 389 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); 390 391 return fs_devs; 392 } 393 394 void btrfs_free_device(struct btrfs_device *device) 395 { 396 WARN_ON(!list_empty(&device->post_commit_list)); 397 rcu_string_free(device->name); 398 extent_io_tree_release(&device->alloc_state); 399 btrfs_destroy_dev_zone_info(device); 400 kfree(device); 401 } 402 403 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 404 { 405 struct btrfs_device *device; 406 407 WARN_ON(fs_devices->opened); 408 while (!list_empty(&fs_devices->devices)) { 409 device = list_entry(fs_devices->devices.next, 410 struct btrfs_device, dev_list); 411 list_del(&device->dev_list); 412 btrfs_free_device(device); 413 } 414 kfree(fs_devices); 415 } 416 417 void __exit btrfs_cleanup_fs_uuids(void) 418 { 419 struct btrfs_fs_devices *fs_devices; 420 421 while (!list_empty(&fs_uuids)) { 422 fs_devices = list_entry(fs_uuids.next, 423 struct btrfs_fs_devices, fs_list); 424 list_del(&fs_devices->fs_list); 425 free_fs_devices(fs_devices); 426 } 427 } 428 429 static noinline struct btrfs_fs_devices *find_fsid( 430 const u8 *fsid, const u8 *metadata_fsid) 431 { 432 struct btrfs_fs_devices *fs_devices; 433 434 ASSERT(fsid); 435 436 /* Handle non-split brain cases */ 437 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 438 if (metadata_fsid) { 439 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 440 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 441 BTRFS_FSID_SIZE) == 0) 442 return fs_devices; 443 } else { 444 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 445 return fs_devices; 446 } 447 } 448 return NULL; 449 } 450 451 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 452 struct btrfs_super_block *disk_super) 453 { 454 455 struct btrfs_fs_devices *fs_devices; 456 457 /* 458 * Handle scanned device having completed its fsid change but 459 * belonging to a fs_devices that was created by first scanning 460 * a device which didn't have its fsid/metadata_uuid changed 461 * at all and the CHANGING_FSID_V2 flag set. 462 */ 463 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 464 if (fs_devices->fsid_change && 465 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 466 BTRFS_FSID_SIZE) == 0 && 467 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 468 BTRFS_FSID_SIZE) == 0) { 469 return fs_devices; 470 } 471 } 472 /* 473 * Handle scanned device having completed its fsid change but 474 * belonging to a fs_devices that was created by a device that 475 * has an outdated pair of fsid/metadata_uuid and 476 * CHANGING_FSID_V2 flag set. 477 */ 478 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 479 if (fs_devices->fsid_change && 480 memcmp(fs_devices->metadata_uuid, 481 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 482 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 483 BTRFS_FSID_SIZE) == 0) { 484 return fs_devices; 485 } 486 } 487 488 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 489 } 490 491 492 static int 493 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 494 int flush, struct block_device **bdev, 495 struct btrfs_super_block **disk_super) 496 { 497 int ret; 498 499 *bdev = blkdev_get_by_path(device_path, flags, holder); 500 501 if (IS_ERR(*bdev)) { 502 ret = PTR_ERR(*bdev); 503 goto error; 504 } 505 506 if (flush) 507 sync_blockdev(*bdev); 508 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 509 if (ret) { 510 blkdev_put(*bdev, flags); 511 goto error; 512 } 513 invalidate_bdev(*bdev); 514 *disk_super = btrfs_read_dev_super(*bdev); 515 if (IS_ERR(*disk_super)) { 516 ret = PTR_ERR(*disk_super); 517 blkdev_put(*bdev, flags); 518 goto error; 519 } 520 521 return 0; 522 523 error: 524 *bdev = NULL; 525 return ret; 526 } 527 528 /* 529 * Search and remove all stale devices (which are not mounted). When both 530 * inputs are NULL, it will search and release all stale devices. 531 * 532 * @devt: Optional. When provided will it release all unmounted devices 533 * matching this devt only. 534 * @skip_device: Optional. Will skip this device when searching for the stale 535 * devices. 536 * 537 * Return: 0 for success or if @devt is 0. 538 * -EBUSY if @devt is a mounted device. 539 * -ENOENT if @devt does not match any device in the list. 540 */ 541 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) 542 { 543 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 544 struct btrfs_device *device, *tmp_device; 545 int ret = 0; 546 547 lockdep_assert_held(&uuid_mutex); 548 549 if (devt) 550 ret = -ENOENT; 551 552 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 553 554 mutex_lock(&fs_devices->device_list_mutex); 555 list_for_each_entry_safe(device, tmp_device, 556 &fs_devices->devices, dev_list) { 557 if (skip_device && skip_device == device) 558 continue; 559 if (devt && devt != device->devt) 560 continue; 561 if (fs_devices->opened) { 562 /* for an already deleted device return 0 */ 563 if (devt && ret != 0) 564 ret = -EBUSY; 565 break; 566 } 567 568 /* delete the stale device */ 569 fs_devices->num_devices--; 570 list_del(&device->dev_list); 571 btrfs_free_device(device); 572 573 ret = 0; 574 } 575 mutex_unlock(&fs_devices->device_list_mutex); 576 577 if (fs_devices->num_devices == 0) { 578 btrfs_sysfs_remove_fsid(fs_devices); 579 list_del(&fs_devices->fs_list); 580 free_fs_devices(fs_devices); 581 } 582 } 583 584 return ret; 585 } 586 587 /* 588 * This is only used on mount, and we are protected from competing things 589 * messing with our fs_devices by the uuid_mutex, thus we do not need the 590 * fs_devices->device_list_mutex here. 591 */ 592 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 593 struct btrfs_device *device, fmode_t flags, 594 void *holder) 595 { 596 struct block_device *bdev; 597 struct btrfs_super_block *disk_super; 598 u64 devid; 599 int ret; 600 601 if (device->bdev) 602 return -EINVAL; 603 if (!device->name) 604 return -EINVAL; 605 606 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 607 &bdev, &disk_super); 608 if (ret) 609 return ret; 610 611 devid = btrfs_stack_device_id(&disk_super->dev_item); 612 if (devid != device->devid) 613 goto error_free_page; 614 615 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 616 goto error_free_page; 617 618 device->generation = btrfs_super_generation(disk_super); 619 620 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 621 if (btrfs_super_incompat_flags(disk_super) & 622 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 623 pr_err( 624 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 625 goto error_free_page; 626 } 627 628 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 629 fs_devices->seeding = true; 630 } else { 631 if (bdev_read_only(bdev)) 632 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 633 else 634 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 635 } 636 637 if (!bdev_nonrot(bdev)) 638 fs_devices->rotating = true; 639 640 if (bdev_max_discard_sectors(bdev)) 641 fs_devices->discardable = true; 642 643 device->bdev = bdev; 644 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 645 device->mode = flags; 646 647 fs_devices->open_devices++; 648 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 649 device->devid != BTRFS_DEV_REPLACE_DEVID) { 650 fs_devices->rw_devices++; 651 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 652 } 653 btrfs_release_disk_super(disk_super); 654 655 return 0; 656 657 error_free_page: 658 btrfs_release_disk_super(disk_super); 659 blkdev_put(bdev, flags); 660 661 return -EINVAL; 662 } 663 664 /* 665 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 666 * being created with a disk that has already completed its fsid change. Such 667 * disk can belong to an fs which has its FSID changed or to one which doesn't. 668 * Handle both cases here. 669 */ 670 static struct btrfs_fs_devices *find_fsid_inprogress( 671 struct btrfs_super_block *disk_super) 672 { 673 struct btrfs_fs_devices *fs_devices; 674 675 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 676 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 677 BTRFS_FSID_SIZE) != 0 && 678 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 679 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 680 return fs_devices; 681 } 682 } 683 684 return find_fsid(disk_super->fsid, NULL); 685 } 686 687 688 static struct btrfs_fs_devices *find_fsid_changed( 689 struct btrfs_super_block *disk_super) 690 { 691 struct btrfs_fs_devices *fs_devices; 692 693 /* 694 * Handles the case where scanned device is part of an fs that had 695 * multiple successful changes of FSID but currently device didn't 696 * observe it. Meaning our fsid will be different than theirs. We need 697 * to handle two subcases : 698 * 1 - The fs still continues to have different METADATA/FSID uuids. 699 * 2 - The fs is switched back to its original FSID (METADATA/FSID 700 * are equal). 701 */ 702 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 703 /* Changed UUIDs */ 704 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 705 BTRFS_FSID_SIZE) != 0 && 706 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 707 BTRFS_FSID_SIZE) == 0 && 708 memcmp(fs_devices->fsid, disk_super->fsid, 709 BTRFS_FSID_SIZE) != 0) 710 return fs_devices; 711 712 /* Unchanged UUIDs */ 713 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 714 BTRFS_FSID_SIZE) == 0 && 715 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 716 BTRFS_FSID_SIZE) == 0) 717 return fs_devices; 718 } 719 720 return NULL; 721 } 722 723 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 724 struct btrfs_super_block *disk_super) 725 { 726 struct btrfs_fs_devices *fs_devices; 727 728 /* 729 * Handle the case where the scanned device is part of an fs whose last 730 * metadata UUID change reverted it to the original FSID. At the same 731 * time fs_devices was first created by another constituent device 732 * which didn't fully observe the operation. This results in an 733 * btrfs_fs_devices created with metadata/fsid different AND 734 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 735 * fs_devices equal to the FSID of the disk. 736 */ 737 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 738 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 739 BTRFS_FSID_SIZE) != 0 && 740 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 741 BTRFS_FSID_SIZE) == 0 && 742 fs_devices->fsid_change) 743 return fs_devices; 744 } 745 746 return NULL; 747 } 748 /* 749 * Add new device to list of registered devices 750 * 751 * Returns: 752 * device pointer which was just added or updated when successful 753 * error pointer when failed 754 */ 755 static noinline struct btrfs_device *device_list_add(const char *path, 756 struct btrfs_super_block *disk_super, 757 bool *new_device_added) 758 { 759 struct btrfs_device *device; 760 struct btrfs_fs_devices *fs_devices = NULL; 761 struct rcu_string *name; 762 u64 found_transid = btrfs_super_generation(disk_super); 763 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 764 dev_t path_devt; 765 int error; 766 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 767 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 768 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 769 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 770 771 error = lookup_bdev(path, &path_devt); 772 if (error) { 773 btrfs_err(NULL, "failed to lookup block device for path %s: %d", 774 path, error); 775 return ERR_PTR(error); 776 } 777 778 if (fsid_change_in_progress) { 779 if (!has_metadata_uuid) 780 fs_devices = find_fsid_inprogress(disk_super); 781 else 782 fs_devices = find_fsid_changed(disk_super); 783 } else if (has_metadata_uuid) { 784 fs_devices = find_fsid_with_metadata_uuid(disk_super); 785 } else { 786 fs_devices = find_fsid_reverted_metadata(disk_super); 787 if (!fs_devices) 788 fs_devices = find_fsid(disk_super->fsid, NULL); 789 } 790 791 792 if (!fs_devices) { 793 if (has_metadata_uuid) 794 fs_devices = alloc_fs_devices(disk_super->fsid, 795 disk_super->metadata_uuid); 796 else 797 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 798 799 if (IS_ERR(fs_devices)) 800 return ERR_CAST(fs_devices); 801 802 fs_devices->fsid_change = fsid_change_in_progress; 803 804 mutex_lock(&fs_devices->device_list_mutex); 805 list_add(&fs_devices->fs_list, &fs_uuids); 806 807 device = NULL; 808 } else { 809 struct btrfs_dev_lookup_args args = { 810 .devid = devid, 811 .uuid = disk_super->dev_item.uuid, 812 }; 813 814 mutex_lock(&fs_devices->device_list_mutex); 815 device = btrfs_find_device(fs_devices, &args); 816 817 /* 818 * If this disk has been pulled into an fs devices created by 819 * a device which had the CHANGING_FSID_V2 flag then replace the 820 * metadata_uuid/fsid values of the fs_devices. 821 */ 822 if (fs_devices->fsid_change && 823 found_transid > fs_devices->latest_generation) { 824 memcpy(fs_devices->fsid, disk_super->fsid, 825 BTRFS_FSID_SIZE); 826 827 if (has_metadata_uuid) 828 memcpy(fs_devices->metadata_uuid, 829 disk_super->metadata_uuid, 830 BTRFS_FSID_SIZE); 831 else 832 memcpy(fs_devices->metadata_uuid, 833 disk_super->fsid, BTRFS_FSID_SIZE); 834 835 fs_devices->fsid_change = false; 836 } 837 } 838 839 if (!device) { 840 unsigned int nofs_flag; 841 842 if (fs_devices->opened) { 843 btrfs_err(NULL, 844 "device %s belongs to fsid %pU, and the fs is already mounted", 845 path, fs_devices->fsid); 846 mutex_unlock(&fs_devices->device_list_mutex); 847 return ERR_PTR(-EBUSY); 848 } 849 850 nofs_flag = memalloc_nofs_save(); 851 device = btrfs_alloc_device(NULL, &devid, 852 disk_super->dev_item.uuid, path); 853 memalloc_nofs_restore(nofs_flag); 854 if (IS_ERR(device)) { 855 mutex_unlock(&fs_devices->device_list_mutex); 856 /* we can safely leave the fs_devices entry around */ 857 return device; 858 } 859 860 device->devt = path_devt; 861 862 list_add_rcu(&device->dev_list, &fs_devices->devices); 863 fs_devices->num_devices++; 864 865 device->fs_devices = fs_devices; 866 *new_device_added = true; 867 868 if (disk_super->label[0]) 869 pr_info( 870 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 871 disk_super->label, devid, found_transid, path, 872 current->comm, task_pid_nr(current)); 873 else 874 pr_info( 875 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 876 disk_super->fsid, devid, found_transid, path, 877 current->comm, task_pid_nr(current)); 878 879 } else if (!device->name || strcmp(device->name->str, path)) { 880 /* 881 * When FS is already mounted. 882 * 1. If you are here and if the device->name is NULL that 883 * means this device was missing at time of FS mount. 884 * 2. If you are here and if the device->name is different 885 * from 'path' that means either 886 * a. The same device disappeared and reappeared with 887 * different name. or 888 * b. The missing-disk-which-was-replaced, has 889 * reappeared now. 890 * 891 * We must allow 1 and 2a above. But 2b would be a spurious 892 * and unintentional. 893 * 894 * Further in case of 1 and 2a above, the disk at 'path' 895 * would have missed some transaction when it was away and 896 * in case of 2a the stale bdev has to be updated as well. 897 * 2b must not be allowed at all time. 898 */ 899 900 /* 901 * For now, we do allow update to btrfs_fs_device through the 902 * btrfs dev scan cli after FS has been mounted. We're still 903 * tracking a problem where systems fail mount by subvolume id 904 * when we reject replacement on a mounted FS. 905 */ 906 if (!fs_devices->opened && found_transid < device->generation) { 907 /* 908 * That is if the FS is _not_ mounted and if you 909 * are here, that means there is more than one 910 * disk with same uuid and devid.We keep the one 911 * with larger generation number or the last-in if 912 * generation are equal. 913 */ 914 mutex_unlock(&fs_devices->device_list_mutex); 915 btrfs_err(NULL, 916 "device %s already registered with a higher generation, found %llu expect %llu", 917 path, found_transid, device->generation); 918 return ERR_PTR(-EEXIST); 919 } 920 921 /* 922 * We are going to replace the device path for a given devid, 923 * make sure it's the same device if the device is mounted 924 * 925 * NOTE: the device->fs_info may not be reliable here so pass 926 * in a NULL to message helpers instead. This avoids a possible 927 * use-after-free when the fs_info and fs_info->sb are already 928 * torn down. 929 */ 930 if (device->bdev) { 931 if (device->devt != path_devt) { 932 mutex_unlock(&fs_devices->device_list_mutex); 933 btrfs_warn_in_rcu(NULL, 934 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 935 path, devid, found_transid, 936 current->comm, 937 task_pid_nr(current)); 938 return ERR_PTR(-EEXIST); 939 } 940 btrfs_info_in_rcu(NULL, 941 "devid %llu device path %s changed to %s scanned by %s (%d)", 942 devid, btrfs_dev_name(device), 943 path, current->comm, 944 task_pid_nr(current)); 945 } 946 947 name = rcu_string_strdup(path, GFP_NOFS); 948 if (!name) { 949 mutex_unlock(&fs_devices->device_list_mutex); 950 return ERR_PTR(-ENOMEM); 951 } 952 rcu_string_free(device->name); 953 rcu_assign_pointer(device->name, name); 954 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 955 fs_devices->missing_devices--; 956 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 957 } 958 device->devt = path_devt; 959 } 960 961 /* 962 * Unmount does not free the btrfs_device struct but would zero 963 * generation along with most of the other members. So just update 964 * it back. We need it to pick the disk with largest generation 965 * (as above). 966 */ 967 if (!fs_devices->opened) { 968 device->generation = found_transid; 969 fs_devices->latest_generation = max_t(u64, found_transid, 970 fs_devices->latest_generation); 971 } 972 973 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 974 975 mutex_unlock(&fs_devices->device_list_mutex); 976 return device; 977 } 978 979 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 980 { 981 struct btrfs_fs_devices *fs_devices; 982 struct btrfs_device *device; 983 struct btrfs_device *orig_dev; 984 int ret = 0; 985 986 lockdep_assert_held(&uuid_mutex); 987 988 fs_devices = alloc_fs_devices(orig->fsid, NULL); 989 if (IS_ERR(fs_devices)) 990 return fs_devices; 991 992 fs_devices->total_devices = orig->total_devices; 993 994 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 995 const char *dev_path = NULL; 996 997 /* 998 * This is ok to do without RCU read locked because we hold the 999 * uuid mutex so nothing we touch in here is going to disappear. 1000 */ 1001 if (orig_dev->name) 1002 dev_path = orig_dev->name->str; 1003 1004 device = btrfs_alloc_device(NULL, &orig_dev->devid, 1005 orig_dev->uuid, dev_path); 1006 if (IS_ERR(device)) { 1007 ret = PTR_ERR(device); 1008 goto error; 1009 } 1010 1011 if (orig_dev->zone_info) { 1012 struct btrfs_zoned_device_info *zone_info; 1013 1014 zone_info = btrfs_clone_dev_zone_info(orig_dev); 1015 if (!zone_info) { 1016 btrfs_free_device(device); 1017 ret = -ENOMEM; 1018 goto error; 1019 } 1020 device->zone_info = zone_info; 1021 } 1022 1023 list_add(&device->dev_list, &fs_devices->devices); 1024 device->fs_devices = fs_devices; 1025 fs_devices->num_devices++; 1026 } 1027 return fs_devices; 1028 error: 1029 free_fs_devices(fs_devices); 1030 return ERR_PTR(ret); 1031 } 1032 1033 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1034 struct btrfs_device **latest_dev) 1035 { 1036 struct btrfs_device *device, *next; 1037 1038 /* This is the initialized path, it is safe to release the devices. */ 1039 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1040 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1041 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1042 &device->dev_state) && 1043 !test_bit(BTRFS_DEV_STATE_MISSING, 1044 &device->dev_state) && 1045 (!*latest_dev || 1046 device->generation > (*latest_dev)->generation)) { 1047 *latest_dev = device; 1048 } 1049 continue; 1050 } 1051 1052 /* 1053 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1054 * in btrfs_init_dev_replace() so just continue. 1055 */ 1056 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1057 continue; 1058 1059 if (device->bdev) { 1060 blkdev_put(device->bdev, device->mode); 1061 device->bdev = NULL; 1062 fs_devices->open_devices--; 1063 } 1064 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1065 list_del_init(&device->dev_alloc_list); 1066 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1067 fs_devices->rw_devices--; 1068 } 1069 list_del_init(&device->dev_list); 1070 fs_devices->num_devices--; 1071 btrfs_free_device(device); 1072 } 1073 1074 } 1075 1076 /* 1077 * After we have read the system tree and know devids belonging to this 1078 * filesystem, remove the device which does not belong there. 1079 */ 1080 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1081 { 1082 struct btrfs_device *latest_dev = NULL; 1083 struct btrfs_fs_devices *seed_dev; 1084 1085 mutex_lock(&uuid_mutex); 1086 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1087 1088 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1089 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1090 1091 fs_devices->latest_dev = latest_dev; 1092 1093 mutex_unlock(&uuid_mutex); 1094 } 1095 1096 static void btrfs_close_bdev(struct btrfs_device *device) 1097 { 1098 if (!device->bdev) 1099 return; 1100 1101 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1102 sync_blockdev(device->bdev); 1103 invalidate_bdev(device->bdev); 1104 } 1105 1106 blkdev_put(device->bdev, device->mode); 1107 } 1108 1109 static void btrfs_close_one_device(struct btrfs_device *device) 1110 { 1111 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1112 1113 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1114 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1115 list_del_init(&device->dev_alloc_list); 1116 fs_devices->rw_devices--; 1117 } 1118 1119 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1120 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1121 1122 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 1123 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 1124 fs_devices->missing_devices--; 1125 } 1126 1127 btrfs_close_bdev(device); 1128 if (device->bdev) { 1129 fs_devices->open_devices--; 1130 device->bdev = NULL; 1131 } 1132 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1133 btrfs_destroy_dev_zone_info(device); 1134 1135 device->fs_info = NULL; 1136 atomic_set(&device->dev_stats_ccnt, 0); 1137 extent_io_tree_release(&device->alloc_state); 1138 1139 /* 1140 * Reset the flush error record. We might have a transient flush error 1141 * in this mount, and if so we aborted the current transaction and set 1142 * the fs to an error state, guaranteeing no super blocks can be further 1143 * committed. However that error might be transient and if we unmount the 1144 * filesystem and mount it again, we should allow the mount to succeed 1145 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1146 * filesystem again we still get flush errors, then we will again abort 1147 * any transaction and set the error state, guaranteeing no commits of 1148 * unsafe super blocks. 1149 */ 1150 device->last_flush_error = 0; 1151 1152 /* Verify the device is back in a pristine state */ 1153 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1154 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1155 ASSERT(list_empty(&device->dev_alloc_list)); 1156 ASSERT(list_empty(&device->post_commit_list)); 1157 } 1158 1159 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1160 { 1161 struct btrfs_device *device, *tmp; 1162 1163 lockdep_assert_held(&uuid_mutex); 1164 1165 if (--fs_devices->opened > 0) 1166 return; 1167 1168 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1169 btrfs_close_one_device(device); 1170 1171 WARN_ON(fs_devices->open_devices); 1172 WARN_ON(fs_devices->rw_devices); 1173 fs_devices->opened = 0; 1174 fs_devices->seeding = false; 1175 fs_devices->fs_info = NULL; 1176 } 1177 1178 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1179 { 1180 LIST_HEAD(list); 1181 struct btrfs_fs_devices *tmp; 1182 1183 mutex_lock(&uuid_mutex); 1184 close_fs_devices(fs_devices); 1185 if (!fs_devices->opened) { 1186 list_splice_init(&fs_devices->seed_list, &list); 1187 1188 /* 1189 * If the struct btrfs_fs_devices is not assembled with any 1190 * other device, it can be re-initialized during the next mount 1191 * without the needing device-scan step. Therefore, it can be 1192 * fully freed. 1193 */ 1194 if (fs_devices->num_devices == 1) { 1195 list_del(&fs_devices->fs_list); 1196 free_fs_devices(fs_devices); 1197 } 1198 } 1199 1200 1201 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1202 close_fs_devices(fs_devices); 1203 list_del(&fs_devices->seed_list); 1204 free_fs_devices(fs_devices); 1205 } 1206 mutex_unlock(&uuid_mutex); 1207 } 1208 1209 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1210 fmode_t flags, void *holder) 1211 { 1212 struct btrfs_device *device; 1213 struct btrfs_device *latest_dev = NULL; 1214 struct btrfs_device *tmp_device; 1215 1216 flags |= FMODE_EXCL; 1217 1218 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1219 dev_list) { 1220 int ret; 1221 1222 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1223 if (ret == 0 && 1224 (!latest_dev || device->generation > latest_dev->generation)) { 1225 latest_dev = device; 1226 } else if (ret == -ENODATA) { 1227 fs_devices->num_devices--; 1228 list_del(&device->dev_list); 1229 btrfs_free_device(device); 1230 } 1231 } 1232 if (fs_devices->open_devices == 0) 1233 return -EINVAL; 1234 1235 fs_devices->opened = 1; 1236 fs_devices->latest_dev = latest_dev; 1237 fs_devices->total_rw_bytes = 0; 1238 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1239 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1240 1241 return 0; 1242 } 1243 1244 static int devid_cmp(void *priv, const struct list_head *a, 1245 const struct list_head *b) 1246 { 1247 const struct btrfs_device *dev1, *dev2; 1248 1249 dev1 = list_entry(a, struct btrfs_device, dev_list); 1250 dev2 = list_entry(b, struct btrfs_device, dev_list); 1251 1252 if (dev1->devid < dev2->devid) 1253 return -1; 1254 else if (dev1->devid > dev2->devid) 1255 return 1; 1256 return 0; 1257 } 1258 1259 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1260 fmode_t flags, void *holder) 1261 { 1262 int ret; 1263 1264 lockdep_assert_held(&uuid_mutex); 1265 /* 1266 * The device_list_mutex cannot be taken here in case opening the 1267 * underlying device takes further locks like open_mutex. 1268 * 1269 * We also don't need the lock here as this is called during mount and 1270 * exclusion is provided by uuid_mutex 1271 */ 1272 1273 if (fs_devices->opened) { 1274 fs_devices->opened++; 1275 ret = 0; 1276 } else { 1277 list_sort(NULL, &fs_devices->devices, devid_cmp); 1278 ret = open_fs_devices(fs_devices, flags, holder); 1279 } 1280 1281 return ret; 1282 } 1283 1284 void btrfs_release_disk_super(struct btrfs_super_block *super) 1285 { 1286 struct page *page = virt_to_page(super); 1287 1288 put_page(page); 1289 } 1290 1291 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1292 u64 bytenr, u64 bytenr_orig) 1293 { 1294 struct btrfs_super_block *disk_super; 1295 struct page *page; 1296 void *p; 1297 pgoff_t index; 1298 1299 /* make sure our super fits in the device */ 1300 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) 1301 return ERR_PTR(-EINVAL); 1302 1303 /* make sure our super fits in the page */ 1304 if (sizeof(*disk_super) > PAGE_SIZE) 1305 return ERR_PTR(-EINVAL); 1306 1307 /* make sure our super doesn't straddle pages on disk */ 1308 index = bytenr >> PAGE_SHIFT; 1309 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1310 return ERR_PTR(-EINVAL); 1311 1312 /* pull in the page with our super */ 1313 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1314 1315 if (IS_ERR(page)) 1316 return ERR_CAST(page); 1317 1318 p = page_address(page); 1319 1320 /* align our pointer to the offset of the super block */ 1321 disk_super = p + offset_in_page(bytenr); 1322 1323 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1324 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1325 btrfs_release_disk_super(p); 1326 return ERR_PTR(-EINVAL); 1327 } 1328 1329 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1330 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1331 1332 return disk_super; 1333 } 1334 1335 int btrfs_forget_devices(dev_t devt) 1336 { 1337 int ret; 1338 1339 mutex_lock(&uuid_mutex); 1340 ret = btrfs_free_stale_devices(devt, NULL); 1341 mutex_unlock(&uuid_mutex); 1342 1343 return ret; 1344 } 1345 1346 /* 1347 * Look for a btrfs signature on a device. This may be called out of the mount path 1348 * and we are not allowed to call set_blocksize during the scan. The superblock 1349 * is read via pagecache 1350 */ 1351 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1352 void *holder) 1353 { 1354 struct btrfs_super_block *disk_super; 1355 bool new_device_added = false; 1356 struct btrfs_device *device = NULL; 1357 struct block_device *bdev; 1358 u64 bytenr, bytenr_orig; 1359 int ret; 1360 1361 lockdep_assert_held(&uuid_mutex); 1362 1363 /* 1364 * we would like to check all the supers, but that would make 1365 * a btrfs mount succeed after a mkfs from a different FS. 1366 * So, we need to add a special mount option to scan for 1367 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1368 */ 1369 1370 /* 1371 * Avoid using flag |= FMODE_EXCL here, as the systemd-udev may 1372 * initiate the device scan which may race with the user's mount 1373 * or mkfs command, resulting in failure. 1374 * Since the device scan is solely for reading purposes, there is 1375 * no need for FMODE_EXCL. Additionally, the devices are read again 1376 * during the mount process. It is ok to get some inconsistent 1377 * values temporarily, as the device paths of the fsid are the only 1378 * required information for assembling the volume. 1379 */ 1380 bdev = blkdev_get_by_path(path, flags, holder); 1381 if (IS_ERR(bdev)) 1382 return ERR_CAST(bdev); 1383 1384 bytenr_orig = btrfs_sb_offset(0); 1385 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1386 if (ret) { 1387 device = ERR_PTR(ret); 1388 goto error_bdev_put; 1389 } 1390 1391 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1392 if (IS_ERR(disk_super)) { 1393 device = ERR_CAST(disk_super); 1394 goto error_bdev_put; 1395 } 1396 1397 device = device_list_add(path, disk_super, &new_device_added); 1398 if (!IS_ERR(device) && new_device_added) 1399 btrfs_free_stale_devices(device->devt, device); 1400 1401 btrfs_release_disk_super(disk_super); 1402 1403 error_bdev_put: 1404 blkdev_put(bdev, flags); 1405 1406 return device; 1407 } 1408 1409 /* 1410 * Try to find a chunk that intersects [start, start + len] range and when one 1411 * such is found, record the end of it in *start 1412 */ 1413 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1414 u64 len) 1415 { 1416 u64 physical_start, physical_end; 1417 1418 lockdep_assert_held(&device->fs_info->chunk_mutex); 1419 1420 if (!find_first_extent_bit(&device->alloc_state, *start, 1421 &physical_start, &physical_end, 1422 CHUNK_ALLOCATED, NULL)) { 1423 1424 if (in_range(physical_start, *start, len) || 1425 in_range(*start, physical_start, 1426 physical_end - physical_start)) { 1427 *start = physical_end + 1; 1428 return true; 1429 } 1430 } 1431 return false; 1432 } 1433 1434 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1435 { 1436 switch (device->fs_devices->chunk_alloc_policy) { 1437 case BTRFS_CHUNK_ALLOC_REGULAR: 1438 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); 1439 case BTRFS_CHUNK_ALLOC_ZONED: 1440 /* 1441 * We don't care about the starting region like regular 1442 * allocator, because we anyway use/reserve the first two zones 1443 * for superblock logging. 1444 */ 1445 return ALIGN(start, device->zone_info->zone_size); 1446 default: 1447 BUG(); 1448 } 1449 } 1450 1451 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1452 u64 *hole_start, u64 *hole_size, 1453 u64 num_bytes) 1454 { 1455 u64 zone_size = device->zone_info->zone_size; 1456 u64 pos; 1457 int ret; 1458 bool changed = false; 1459 1460 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1461 1462 while (*hole_size > 0) { 1463 pos = btrfs_find_allocatable_zones(device, *hole_start, 1464 *hole_start + *hole_size, 1465 num_bytes); 1466 if (pos != *hole_start) { 1467 *hole_size = *hole_start + *hole_size - pos; 1468 *hole_start = pos; 1469 changed = true; 1470 if (*hole_size < num_bytes) 1471 break; 1472 } 1473 1474 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1475 1476 /* Range is ensured to be empty */ 1477 if (!ret) 1478 return changed; 1479 1480 /* Given hole range was invalid (outside of device) */ 1481 if (ret == -ERANGE) { 1482 *hole_start += *hole_size; 1483 *hole_size = 0; 1484 return true; 1485 } 1486 1487 *hole_start += zone_size; 1488 *hole_size -= zone_size; 1489 changed = true; 1490 } 1491 1492 return changed; 1493 } 1494 1495 /* 1496 * Check if specified hole is suitable for allocation. 1497 * 1498 * @device: the device which we have the hole 1499 * @hole_start: starting position of the hole 1500 * @hole_size: the size of the hole 1501 * @num_bytes: the size of the free space that we need 1502 * 1503 * This function may modify @hole_start and @hole_size to reflect the suitable 1504 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1505 */ 1506 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1507 u64 *hole_size, u64 num_bytes) 1508 { 1509 bool changed = false; 1510 u64 hole_end = *hole_start + *hole_size; 1511 1512 for (;;) { 1513 /* 1514 * Check before we set max_hole_start, otherwise we could end up 1515 * sending back this offset anyway. 1516 */ 1517 if (contains_pending_extent(device, hole_start, *hole_size)) { 1518 if (hole_end >= *hole_start) 1519 *hole_size = hole_end - *hole_start; 1520 else 1521 *hole_size = 0; 1522 changed = true; 1523 } 1524 1525 switch (device->fs_devices->chunk_alloc_policy) { 1526 case BTRFS_CHUNK_ALLOC_REGULAR: 1527 /* No extra check */ 1528 break; 1529 case BTRFS_CHUNK_ALLOC_ZONED: 1530 if (dev_extent_hole_check_zoned(device, hole_start, 1531 hole_size, num_bytes)) { 1532 changed = true; 1533 /* 1534 * The changed hole can contain pending extent. 1535 * Loop again to check that. 1536 */ 1537 continue; 1538 } 1539 break; 1540 default: 1541 BUG(); 1542 } 1543 1544 break; 1545 } 1546 1547 return changed; 1548 } 1549 1550 /* 1551 * Find free space in the specified device. 1552 * 1553 * @device: the device which we search the free space in 1554 * @num_bytes: the size of the free space that we need 1555 * @search_start: the position from which to begin the search 1556 * @start: store the start of the free space. 1557 * @len: the size of the free space. that we find, or the size 1558 * of the max free space if we don't find suitable free space 1559 * 1560 * This does a pretty simple search, the expectation is that it is called very 1561 * infrequently and that a given device has a small number of extents. 1562 * 1563 * @start is used to store the start of the free space if we find. But if we 1564 * don't find suitable free space, it will be used to store the start position 1565 * of the max free space. 1566 * 1567 * @len is used to store the size of the free space that we find. 1568 * But if we don't find suitable free space, it is used to store the size of 1569 * the max free space. 1570 * 1571 * NOTE: This function will search *commit* root of device tree, and does extra 1572 * check to ensure dev extents are not double allocated. 1573 * This makes the function safe to allocate dev extents but may not report 1574 * correct usable device space, as device extent freed in current transaction 1575 * is not reported as available. 1576 */ 1577 static int find_free_dev_extent_start(struct btrfs_device *device, 1578 u64 num_bytes, u64 search_start, u64 *start, 1579 u64 *len) 1580 { 1581 struct btrfs_fs_info *fs_info = device->fs_info; 1582 struct btrfs_root *root = fs_info->dev_root; 1583 struct btrfs_key key; 1584 struct btrfs_dev_extent *dev_extent; 1585 struct btrfs_path *path; 1586 u64 hole_size; 1587 u64 max_hole_start; 1588 u64 max_hole_size; 1589 u64 extent_end; 1590 u64 search_end = device->total_bytes; 1591 int ret; 1592 int slot; 1593 struct extent_buffer *l; 1594 1595 search_start = dev_extent_search_start(device, search_start); 1596 1597 WARN_ON(device->zone_info && 1598 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1599 1600 path = btrfs_alloc_path(); 1601 if (!path) 1602 return -ENOMEM; 1603 1604 max_hole_start = search_start; 1605 max_hole_size = 0; 1606 1607 again: 1608 if (search_start >= search_end || 1609 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1610 ret = -ENOSPC; 1611 goto out; 1612 } 1613 1614 path->reada = READA_FORWARD; 1615 path->search_commit_root = 1; 1616 path->skip_locking = 1; 1617 1618 key.objectid = device->devid; 1619 key.offset = search_start; 1620 key.type = BTRFS_DEV_EXTENT_KEY; 1621 1622 ret = btrfs_search_backwards(root, &key, path); 1623 if (ret < 0) 1624 goto out; 1625 1626 while (search_start < search_end) { 1627 l = path->nodes[0]; 1628 slot = path->slots[0]; 1629 if (slot >= btrfs_header_nritems(l)) { 1630 ret = btrfs_next_leaf(root, path); 1631 if (ret == 0) 1632 continue; 1633 if (ret < 0) 1634 goto out; 1635 1636 break; 1637 } 1638 btrfs_item_key_to_cpu(l, &key, slot); 1639 1640 if (key.objectid < device->devid) 1641 goto next; 1642 1643 if (key.objectid > device->devid) 1644 break; 1645 1646 if (key.type != BTRFS_DEV_EXTENT_KEY) 1647 goto next; 1648 1649 if (key.offset > search_end) 1650 break; 1651 1652 if (key.offset > search_start) { 1653 hole_size = key.offset - search_start; 1654 dev_extent_hole_check(device, &search_start, &hole_size, 1655 num_bytes); 1656 1657 if (hole_size > max_hole_size) { 1658 max_hole_start = search_start; 1659 max_hole_size = hole_size; 1660 } 1661 1662 /* 1663 * If this free space is greater than which we need, 1664 * it must be the max free space that we have found 1665 * until now, so max_hole_start must point to the start 1666 * of this free space and the length of this free space 1667 * is stored in max_hole_size. Thus, we return 1668 * max_hole_start and max_hole_size and go back to the 1669 * caller. 1670 */ 1671 if (hole_size >= num_bytes) { 1672 ret = 0; 1673 goto out; 1674 } 1675 } 1676 1677 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1678 extent_end = key.offset + btrfs_dev_extent_length(l, 1679 dev_extent); 1680 if (extent_end > search_start) 1681 search_start = extent_end; 1682 next: 1683 path->slots[0]++; 1684 cond_resched(); 1685 } 1686 1687 /* 1688 * At this point, search_start should be the end of 1689 * allocated dev extents, and when shrinking the device, 1690 * search_end may be smaller than search_start. 1691 */ 1692 if (search_end > search_start) { 1693 hole_size = search_end - search_start; 1694 if (dev_extent_hole_check(device, &search_start, &hole_size, 1695 num_bytes)) { 1696 btrfs_release_path(path); 1697 goto again; 1698 } 1699 1700 if (hole_size > max_hole_size) { 1701 max_hole_start = search_start; 1702 max_hole_size = hole_size; 1703 } 1704 } 1705 1706 /* See above. */ 1707 if (max_hole_size < num_bytes) 1708 ret = -ENOSPC; 1709 else 1710 ret = 0; 1711 1712 ASSERT(max_hole_start + max_hole_size <= search_end); 1713 out: 1714 btrfs_free_path(path); 1715 *start = max_hole_start; 1716 if (len) 1717 *len = max_hole_size; 1718 return ret; 1719 } 1720 1721 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1722 u64 *start, u64 *len) 1723 { 1724 /* FIXME use last free of some kind */ 1725 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1726 } 1727 1728 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1729 struct btrfs_device *device, 1730 u64 start, u64 *dev_extent_len) 1731 { 1732 struct btrfs_fs_info *fs_info = device->fs_info; 1733 struct btrfs_root *root = fs_info->dev_root; 1734 int ret; 1735 struct btrfs_path *path; 1736 struct btrfs_key key; 1737 struct btrfs_key found_key; 1738 struct extent_buffer *leaf = NULL; 1739 struct btrfs_dev_extent *extent = NULL; 1740 1741 path = btrfs_alloc_path(); 1742 if (!path) 1743 return -ENOMEM; 1744 1745 key.objectid = device->devid; 1746 key.offset = start; 1747 key.type = BTRFS_DEV_EXTENT_KEY; 1748 again: 1749 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1750 if (ret > 0) { 1751 ret = btrfs_previous_item(root, path, key.objectid, 1752 BTRFS_DEV_EXTENT_KEY); 1753 if (ret) 1754 goto out; 1755 leaf = path->nodes[0]; 1756 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1757 extent = btrfs_item_ptr(leaf, path->slots[0], 1758 struct btrfs_dev_extent); 1759 BUG_ON(found_key.offset > start || found_key.offset + 1760 btrfs_dev_extent_length(leaf, extent) < start); 1761 key = found_key; 1762 btrfs_release_path(path); 1763 goto again; 1764 } else if (ret == 0) { 1765 leaf = path->nodes[0]; 1766 extent = btrfs_item_ptr(leaf, path->slots[0], 1767 struct btrfs_dev_extent); 1768 } else { 1769 goto out; 1770 } 1771 1772 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1773 1774 ret = btrfs_del_item(trans, root, path); 1775 if (ret == 0) 1776 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1777 out: 1778 btrfs_free_path(path); 1779 return ret; 1780 } 1781 1782 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1783 { 1784 struct extent_map_tree *em_tree; 1785 struct extent_map *em; 1786 struct rb_node *n; 1787 u64 ret = 0; 1788 1789 em_tree = &fs_info->mapping_tree; 1790 read_lock(&em_tree->lock); 1791 n = rb_last(&em_tree->map.rb_root); 1792 if (n) { 1793 em = rb_entry(n, struct extent_map, rb_node); 1794 ret = em->start + em->len; 1795 } 1796 read_unlock(&em_tree->lock); 1797 1798 return ret; 1799 } 1800 1801 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1802 u64 *devid_ret) 1803 { 1804 int ret; 1805 struct btrfs_key key; 1806 struct btrfs_key found_key; 1807 struct btrfs_path *path; 1808 1809 path = btrfs_alloc_path(); 1810 if (!path) 1811 return -ENOMEM; 1812 1813 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1814 key.type = BTRFS_DEV_ITEM_KEY; 1815 key.offset = (u64)-1; 1816 1817 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1818 if (ret < 0) 1819 goto error; 1820 1821 if (ret == 0) { 1822 /* Corruption */ 1823 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1824 ret = -EUCLEAN; 1825 goto error; 1826 } 1827 1828 ret = btrfs_previous_item(fs_info->chunk_root, path, 1829 BTRFS_DEV_ITEMS_OBJECTID, 1830 BTRFS_DEV_ITEM_KEY); 1831 if (ret) { 1832 *devid_ret = 1; 1833 } else { 1834 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1835 path->slots[0]); 1836 *devid_ret = found_key.offset + 1; 1837 } 1838 ret = 0; 1839 error: 1840 btrfs_free_path(path); 1841 return ret; 1842 } 1843 1844 /* 1845 * the device information is stored in the chunk root 1846 * the btrfs_device struct should be fully filled in 1847 */ 1848 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1849 struct btrfs_device *device) 1850 { 1851 int ret; 1852 struct btrfs_path *path; 1853 struct btrfs_dev_item *dev_item; 1854 struct extent_buffer *leaf; 1855 struct btrfs_key key; 1856 unsigned long ptr; 1857 1858 path = btrfs_alloc_path(); 1859 if (!path) 1860 return -ENOMEM; 1861 1862 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1863 key.type = BTRFS_DEV_ITEM_KEY; 1864 key.offset = device->devid; 1865 1866 btrfs_reserve_chunk_metadata(trans, true); 1867 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1868 &key, sizeof(*dev_item)); 1869 btrfs_trans_release_chunk_metadata(trans); 1870 if (ret) 1871 goto out; 1872 1873 leaf = path->nodes[0]; 1874 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1875 1876 btrfs_set_device_id(leaf, dev_item, device->devid); 1877 btrfs_set_device_generation(leaf, dev_item, 0); 1878 btrfs_set_device_type(leaf, dev_item, device->type); 1879 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1880 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1881 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1882 btrfs_set_device_total_bytes(leaf, dev_item, 1883 btrfs_device_get_disk_total_bytes(device)); 1884 btrfs_set_device_bytes_used(leaf, dev_item, 1885 btrfs_device_get_bytes_used(device)); 1886 btrfs_set_device_group(leaf, dev_item, 0); 1887 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1888 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1889 btrfs_set_device_start_offset(leaf, dev_item, 0); 1890 1891 ptr = btrfs_device_uuid(dev_item); 1892 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1893 ptr = btrfs_device_fsid(dev_item); 1894 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1895 ptr, BTRFS_FSID_SIZE); 1896 btrfs_mark_buffer_dirty(leaf); 1897 1898 ret = 0; 1899 out: 1900 btrfs_free_path(path); 1901 return ret; 1902 } 1903 1904 /* 1905 * Function to update ctime/mtime for a given device path. 1906 * Mainly used for ctime/mtime based probe like libblkid. 1907 * 1908 * We don't care about errors here, this is just to be kind to userspace. 1909 */ 1910 static void update_dev_time(const char *device_path) 1911 { 1912 struct path path; 1913 struct timespec64 now; 1914 int ret; 1915 1916 ret = kern_path(device_path, LOOKUP_FOLLOW, &path); 1917 if (ret) 1918 return; 1919 1920 now = current_time(d_inode(path.dentry)); 1921 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); 1922 path_put(&path); 1923 } 1924 1925 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, 1926 struct btrfs_device *device) 1927 { 1928 struct btrfs_root *root = device->fs_info->chunk_root; 1929 int ret; 1930 struct btrfs_path *path; 1931 struct btrfs_key key; 1932 1933 path = btrfs_alloc_path(); 1934 if (!path) 1935 return -ENOMEM; 1936 1937 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1938 key.type = BTRFS_DEV_ITEM_KEY; 1939 key.offset = device->devid; 1940 1941 btrfs_reserve_chunk_metadata(trans, false); 1942 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1943 btrfs_trans_release_chunk_metadata(trans); 1944 if (ret) { 1945 if (ret > 0) 1946 ret = -ENOENT; 1947 goto out; 1948 } 1949 1950 ret = btrfs_del_item(trans, root, path); 1951 out: 1952 btrfs_free_path(path); 1953 return ret; 1954 } 1955 1956 /* 1957 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1958 * filesystem. It's up to the caller to adjust that number regarding eg. device 1959 * replace. 1960 */ 1961 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1962 u64 num_devices) 1963 { 1964 u64 all_avail; 1965 unsigned seq; 1966 int i; 1967 1968 do { 1969 seq = read_seqbegin(&fs_info->profiles_lock); 1970 1971 all_avail = fs_info->avail_data_alloc_bits | 1972 fs_info->avail_system_alloc_bits | 1973 fs_info->avail_metadata_alloc_bits; 1974 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1975 1976 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1977 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1978 continue; 1979 1980 if (num_devices < btrfs_raid_array[i].devs_min) 1981 return btrfs_raid_array[i].mindev_error; 1982 } 1983 1984 return 0; 1985 } 1986 1987 static struct btrfs_device * btrfs_find_next_active_device( 1988 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1989 { 1990 struct btrfs_device *next_device; 1991 1992 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1993 if (next_device != device && 1994 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1995 && next_device->bdev) 1996 return next_device; 1997 } 1998 1999 return NULL; 2000 } 2001 2002 /* 2003 * Helper function to check if the given device is part of s_bdev / latest_dev 2004 * and replace it with the provided or the next active device, in the context 2005 * where this function called, there should be always be another device (or 2006 * this_dev) which is active. 2007 */ 2008 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 2009 struct btrfs_device *next_device) 2010 { 2011 struct btrfs_fs_info *fs_info = device->fs_info; 2012 2013 if (!next_device) 2014 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 2015 device); 2016 ASSERT(next_device); 2017 2018 if (fs_info->sb->s_bdev && 2019 (fs_info->sb->s_bdev == device->bdev)) 2020 fs_info->sb->s_bdev = next_device->bdev; 2021 2022 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2023 fs_info->fs_devices->latest_dev = next_device; 2024 } 2025 2026 /* 2027 * Return btrfs_fs_devices::num_devices excluding the device that's being 2028 * currently replaced. 2029 */ 2030 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2031 { 2032 u64 num_devices = fs_info->fs_devices->num_devices; 2033 2034 down_read(&fs_info->dev_replace.rwsem); 2035 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2036 ASSERT(num_devices > 1); 2037 num_devices--; 2038 } 2039 up_read(&fs_info->dev_replace.rwsem); 2040 2041 return num_devices; 2042 } 2043 2044 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info, 2045 struct block_device *bdev, int copy_num) 2046 { 2047 struct btrfs_super_block *disk_super; 2048 const size_t len = sizeof(disk_super->magic); 2049 const u64 bytenr = btrfs_sb_offset(copy_num); 2050 int ret; 2051 2052 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr); 2053 if (IS_ERR(disk_super)) 2054 return; 2055 2056 memset(&disk_super->magic, 0, len); 2057 folio_mark_dirty(virt_to_folio(disk_super)); 2058 btrfs_release_disk_super(disk_super); 2059 2060 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1); 2061 if (ret) 2062 btrfs_warn(fs_info, "error clearing superblock number %d (%d)", 2063 copy_num, ret); 2064 } 2065 2066 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2067 struct block_device *bdev, 2068 const char *device_path) 2069 { 2070 int copy_num; 2071 2072 if (!bdev) 2073 return; 2074 2075 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2076 if (bdev_is_zoned(bdev)) 2077 btrfs_reset_sb_log_zones(bdev, copy_num); 2078 else 2079 btrfs_scratch_superblock(fs_info, bdev, copy_num); 2080 } 2081 2082 /* Notify udev that device has changed */ 2083 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2084 2085 /* Update ctime/mtime for device path for libblkid */ 2086 update_dev_time(device_path); 2087 } 2088 2089 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2090 struct btrfs_dev_lookup_args *args, 2091 struct block_device **bdev, fmode_t *mode) 2092 { 2093 struct btrfs_trans_handle *trans; 2094 struct btrfs_device *device; 2095 struct btrfs_fs_devices *cur_devices; 2096 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2097 u64 num_devices; 2098 int ret = 0; 2099 2100 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2101 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); 2102 return -EINVAL; 2103 } 2104 2105 /* 2106 * The device list in fs_devices is accessed without locks (neither 2107 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2108 * filesystem and another device rm cannot run. 2109 */ 2110 num_devices = btrfs_num_devices(fs_info); 2111 2112 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2113 if (ret) 2114 return ret; 2115 2116 device = btrfs_find_device(fs_info->fs_devices, args); 2117 if (!device) { 2118 if (args->missing) 2119 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2120 else 2121 ret = -ENOENT; 2122 return ret; 2123 } 2124 2125 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2126 btrfs_warn_in_rcu(fs_info, 2127 "cannot remove device %s (devid %llu) due to active swapfile", 2128 btrfs_dev_name(device), device->devid); 2129 return -ETXTBSY; 2130 } 2131 2132 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 2133 return BTRFS_ERROR_DEV_TGT_REPLACE; 2134 2135 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2136 fs_info->fs_devices->rw_devices == 1) 2137 return BTRFS_ERROR_DEV_ONLY_WRITABLE; 2138 2139 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2140 mutex_lock(&fs_info->chunk_mutex); 2141 list_del_init(&device->dev_alloc_list); 2142 device->fs_devices->rw_devices--; 2143 mutex_unlock(&fs_info->chunk_mutex); 2144 } 2145 2146 ret = btrfs_shrink_device(device, 0); 2147 if (ret) 2148 goto error_undo; 2149 2150 trans = btrfs_start_transaction(fs_info->chunk_root, 0); 2151 if (IS_ERR(trans)) { 2152 ret = PTR_ERR(trans); 2153 goto error_undo; 2154 } 2155 2156 ret = btrfs_rm_dev_item(trans, device); 2157 if (ret) { 2158 /* Any error in dev item removal is critical */ 2159 btrfs_crit(fs_info, 2160 "failed to remove device item for devid %llu: %d", 2161 device->devid, ret); 2162 btrfs_abort_transaction(trans, ret); 2163 btrfs_end_transaction(trans); 2164 return ret; 2165 } 2166 2167 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2168 btrfs_scrub_cancel_dev(device); 2169 2170 /* 2171 * the device list mutex makes sure that we don't change 2172 * the device list while someone else is writing out all 2173 * the device supers. Whoever is writing all supers, should 2174 * lock the device list mutex before getting the number of 2175 * devices in the super block (super_copy). Conversely, 2176 * whoever updates the number of devices in the super block 2177 * (super_copy) should hold the device list mutex. 2178 */ 2179 2180 /* 2181 * In normal cases the cur_devices == fs_devices. But in case 2182 * of deleting a seed device, the cur_devices should point to 2183 * its own fs_devices listed under the fs_devices->seed_list. 2184 */ 2185 cur_devices = device->fs_devices; 2186 mutex_lock(&fs_devices->device_list_mutex); 2187 list_del_rcu(&device->dev_list); 2188 2189 cur_devices->num_devices--; 2190 cur_devices->total_devices--; 2191 /* Update total_devices of the parent fs_devices if it's seed */ 2192 if (cur_devices != fs_devices) 2193 fs_devices->total_devices--; 2194 2195 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2196 cur_devices->missing_devices--; 2197 2198 btrfs_assign_next_active_device(device, NULL); 2199 2200 if (device->bdev) { 2201 cur_devices->open_devices--; 2202 /* remove sysfs entry */ 2203 btrfs_sysfs_remove_device(device); 2204 } 2205 2206 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2207 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2208 mutex_unlock(&fs_devices->device_list_mutex); 2209 2210 /* 2211 * At this point, the device is zero sized and detached from the 2212 * devices list. All that's left is to zero out the old supers and 2213 * free the device. 2214 * 2215 * We cannot call btrfs_close_bdev() here because we're holding the sb 2216 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2217 * block device and it's dependencies. Instead just flush the device 2218 * and let the caller do the final blkdev_put. 2219 */ 2220 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2221 btrfs_scratch_superblocks(fs_info, device->bdev, 2222 device->name->str); 2223 if (device->bdev) { 2224 sync_blockdev(device->bdev); 2225 invalidate_bdev(device->bdev); 2226 } 2227 } 2228 2229 *bdev = device->bdev; 2230 *mode = device->mode; 2231 synchronize_rcu(); 2232 btrfs_free_device(device); 2233 2234 /* 2235 * This can happen if cur_devices is the private seed devices list. We 2236 * cannot call close_fs_devices() here because it expects the uuid_mutex 2237 * to be held, but in fact we don't need that for the private 2238 * seed_devices, we can simply decrement cur_devices->opened and then 2239 * remove it from our list and free the fs_devices. 2240 */ 2241 if (cur_devices->num_devices == 0) { 2242 list_del_init(&cur_devices->seed_list); 2243 ASSERT(cur_devices->opened == 1); 2244 cur_devices->opened--; 2245 free_fs_devices(cur_devices); 2246 } 2247 2248 ret = btrfs_commit_transaction(trans); 2249 2250 return ret; 2251 2252 error_undo: 2253 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2254 mutex_lock(&fs_info->chunk_mutex); 2255 list_add(&device->dev_alloc_list, 2256 &fs_devices->alloc_list); 2257 device->fs_devices->rw_devices++; 2258 mutex_unlock(&fs_info->chunk_mutex); 2259 } 2260 return ret; 2261 } 2262 2263 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2264 { 2265 struct btrfs_fs_devices *fs_devices; 2266 2267 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2268 2269 /* 2270 * in case of fs with no seed, srcdev->fs_devices will point 2271 * to fs_devices of fs_info. However when the dev being replaced is 2272 * a seed dev it will point to the seed's local fs_devices. In short 2273 * srcdev will have its correct fs_devices in both the cases. 2274 */ 2275 fs_devices = srcdev->fs_devices; 2276 2277 list_del_rcu(&srcdev->dev_list); 2278 list_del(&srcdev->dev_alloc_list); 2279 fs_devices->num_devices--; 2280 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2281 fs_devices->missing_devices--; 2282 2283 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2284 fs_devices->rw_devices--; 2285 2286 if (srcdev->bdev) 2287 fs_devices->open_devices--; 2288 } 2289 2290 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2291 { 2292 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2293 2294 mutex_lock(&uuid_mutex); 2295 2296 btrfs_close_bdev(srcdev); 2297 synchronize_rcu(); 2298 btrfs_free_device(srcdev); 2299 2300 /* if this is no devs we rather delete the fs_devices */ 2301 if (!fs_devices->num_devices) { 2302 /* 2303 * On a mounted FS, num_devices can't be zero unless it's a 2304 * seed. In case of a seed device being replaced, the replace 2305 * target added to the sprout FS, so there will be no more 2306 * device left under the seed FS. 2307 */ 2308 ASSERT(fs_devices->seeding); 2309 2310 list_del_init(&fs_devices->seed_list); 2311 close_fs_devices(fs_devices); 2312 free_fs_devices(fs_devices); 2313 } 2314 mutex_unlock(&uuid_mutex); 2315 } 2316 2317 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2318 { 2319 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2320 2321 mutex_lock(&fs_devices->device_list_mutex); 2322 2323 btrfs_sysfs_remove_device(tgtdev); 2324 2325 if (tgtdev->bdev) 2326 fs_devices->open_devices--; 2327 2328 fs_devices->num_devices--; 2329 2330 btrfs_assign_next_active_device(tgtdev, NULL); 2331 2332 list_del_rcu(&tgtdev->dev_list); 2333 2334 mutex_unlock(&fs_devices->device_list_mutex); 2335 2336 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2337 tgtdev->name->str); 2338 2339 btrfs_close_bdev(tgtdev); 2340 synchronize_rcu(); 2341 btrfs_free_device(tgtdev); 2342 } 2343 2344 /* 2345 * Populate args from device at path. 2346 * 2347 * @fs_info: the filesystem 2348 * @args: the args to populate 2349 * @path: the path to the device 2350 * 2351 * This will read the super block of the device at @path and populate @args with 2352 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2353 * lookup a device to operate on, but need to do it before we take any locks. 2354 * This properly handles the special case of "missing" that a user may pass in, 2355 * and does some basic sanity checks. The caller must make sure that @path is 2356 * properly NUL terminated before calling in, and must call 2357 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2358 * uuid buffers. 2359 * 2360 * Return: 0 for success, -errno for failure 2361 */ 2362 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2363 struct btrfs_dev_lookup_args *args, 2364 const char *path) 2365 { 2366 struct btrfs_super_block *disk_super; 2367 struct block_device *bdev; 2368 int ret; 2369 2370 if (!path || !path[0]) 2371 return -EINVAL; 2372 if (!strcmp(path, "missing")) { 2373 args->missing = true; 2374 return 0; 2375 } 2376 2377 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2378 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2379 if (!args->uuid || !args->fsid) { 2380 btrfs_put_dev_args_from_path(args); 2381 return -ENOMEM; 2382 } 2383 2384 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2385 &bdev, &disk_super); 2386 if (ret) { 2387 btrfs_put_dev_args_from_path(args); 2388 return ret; 2389 } 2390 2391 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2392 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2393 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2394 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2395 else 2396 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2397 btrfs_release_disk_super(disk_super); 2398 blkdev_put(bdev, FMODE_READ); 2399 return 0; 2400 } 2401 2402 /* 2403 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2404 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2405 * that don't need to be freed. 2406 */ 2407 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2408 { 2409 kfree(args->uuid); 2410 kfree(args->fsid); 2411 args->uuid = NULL; 2412 args->fsid = NULL; 2413 } 2414 2415 struct btrfs_device *btrfs_find_device_by_devspec( 2416 struct btrfs_fs_info *fs_info, u64 devid, 2417 const char *device_path) 2418 { 2419 BTRFS_DEV_LOOKUP_ARGS(args); 2420 struct btrfs_device *device; 2421 int ret; 2422 2423 if (devid) { 2424 args.devid = devid; 2425 device = btrfs_find_device(fs_info->fs_devices, &args); 2426 if (!device) 2427 return ERR_PTR(-ENOENT); 2428 return device; 2429 } 2430 2431 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2432 if (ret) 2433 return ERR_PTR(ret); 2434 device = btrfs_find_device(fs_info->fs_devices, &args); 2435 btrfs_put_dev_args_from_path(&args); 2436 if (!device) 2437 return ERR_PTR(-ENOENT); 2438 return device; 2439 } 2440 2441 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) 2442 { 2443 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2444 struct btrfs_fs_devices *old_devices; 2445 struct btrfs_fs_devices *seed_devices; 2446 2447 lockdep_assert_held(&uuid_mutex); 2448 if (!fs_devices->seeding) 2449 return ERR_PTR(-EINVAL); 2450 2451 /* 2452 * Private copy of the seed devices, anchored at 2453 * fs_info->fs_devices->seed_list 2454 */ 2455 seed_devices = alloc_fs_devices(NULL, NULL); 2456 if (IS_ERR(seed_devices)) 2457 return seed_devices; 2458 2459 /* 2460 * It's necessary to retain a copy of the original seed fs_devices in 2461 * fs_uuids so that filesystems which have been seeded can successfully 2462 * reference the seed device from open_seed_devices. This also supports 2463 * multiple fs seed. 2464 */ 2465 old_devices = clone_fs_devices(fs_devices); 2466 if (IS_ERR(old_devices)) { 2467 kfree(seed_devices); 2468 return old_devices; 2469 } 2470 2471 list_add(&old_devices->fs_list, &fs_uuids); 2472 2473 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2474 seed_devices->opened = 1; 2475 INIT_LIST_HEAD(&seed_devices->devices); 2476 INIT_LIST_HEAD(&seed_devices->alloc_list); 2477 mutex_init(&seed_devices->device_list_mutex); 2478 2479 return seed_devices; 2480 } 2481 2482 /* 2483 * Splice seed devices into the sprout fs_devices. 2484 * Generate a new fsid for the sprouted read-write filesystem. 2485 */ 2486 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, 2487 struct btrfs_fs_devices *seed_devices) 2488 { 2489 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2490 struct btrfs_super_block *disk_super = fs_info->super_copy; 2491 struct btrfs_device *device; 2492 u64 super_flags; 2493 2494 /* 2495 * We are updating the fsid, the thread leading to device_list_add() 2496 * could race, so uuid_mutex is needed. 2497 */ 2498 lockdep_assert_held(&uuid_mutex); 2499 2500 /* 2501 * The threads listed below may traverse dev_list but can do that without 2502 * device_list_mutex: 2503 * - All device ops and balance - as we are in btrfs_exclop_start. 2504 * - Various dev_list readers - are using RCU. 2505 * - btrfs_ioctl_fitrim() - is using RCU. 2506 * 2507 * For-read threads as below are using device_list_mutex: 2508 * - Readonly scrub btrfs_scrub_dev() 2509 * - Readonly scrub btrfs_scrub_progress() 2510 * - btrfs_get_dev_stats() 2511 */ 2512 lockdep_assert_held(&fs_devices->device_list_mutex); 2513 2514 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2515 synchronize_rcu); 2516 list_for_each_entry(device, &seed_devices->devices, dev_list) 2517 device->fs_devices = seed_devices; 2518 2519 fs_devices->seeding = false; 2520 fs_devices->num_devices = 0; 2521 fs_devices->open_devices = 0; 2522 fs_devices->missing_devices = 0; 2523 fs_devices->rotating = false; 2524 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2525 2526 generate_random_uuid(fs_devices->fsid); 2527 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2528 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2529 2530 super_flags = btrfs_super_flags(disk_super) & 2531 ~BTRFS_SUPER_FLAG_SEEDING; 2532 btrfs_set_super_flags(disk_super, super_flags); 2533 } 2534 2535 /* 2536 * Store the expected generation for seed devices in device items. 2537 */ 2538 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2539 { 2540 BTRFS_DEV_LOOKUP_ARGS(args); 2541 struct btrfs_fs_info *fs_info = trans->fs_info; 2542 struct btrfs_root *root = fs_info->chunk_root; 2543 struct btrfs_path *path; 2544 struct extent_buffer *leaf; 2545 struct btrfs_dev_item *dev_item; 2546 struct btrfs_device *device; 2547 struct btrfs_key key; 2548 u8 fs_uuid[BTRFS_FSID_SIZE]; 2549 u8 dev_uuid[BTRFS_UUID_SIZE]; 2550 int ret; 2551 2552 path = btrfs_alloc_path(); 2553 if (!path) 2554 return -ENOMEM; 2555 2556 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2557 key.offset = 0; 2558 key.type = BTRFS_DEV_ITEM_KEY; 2559 2560 while (1) { 2561 btrfs_reserve_chunk_metadata(trans, false); 2562 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2563 btrfs_trans_release_chunk_metadata(trans); 2564 if (ret < 0) 2565 goto error; 2566 2567 leaf = path->nodes[0]; 2568 next_slot: 2569 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2570 ret = btrfs_next_leaf(root, path); 2571 if (ret > 0) 2572 break; 2573 if (ret < 0) 2574 goto error; 2575 leaf = path->nodes[0]; 2576 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2577 btrfs_release_path(path); 2578 continue; 2579 } 2580 2581 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2582 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2583 key.type != BTRFS_DEV_ITEM_KEY) 2584 break; 2585 2586 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2587 struct btrfs_dev_item); 2588 args.devid = btrfs_device_id(leaf, dev_item); 2589 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2590 BTRFS_UUID_SIZE); 2591 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2592 BTRFS_FSID_SIZE); 2593 args.uuid = dev_uuid; 2594 args.fsid = fs_uuid; 2595 device = btrfs_find_device(fs_info->fs_devices, &args); 2596 BUG_ON(!device); /* Logic error */ 2597 2598 if (device->fs_devices->seeding) { 2599 btrfs_set_device_generation(leaf, dev_item, 2600 device->generation); 2601 btrfs_mark_buffer_dirty(leaf); 2602 } 2603 2604 path->slots[0]++; 2605 goto next_slot; 2606 } 2607 ret = 0; 2608 error: 2609 btrfs_free_path(path); 2610 return ret; 2611 } 2612 2613 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2614 { 2615 struct btrfs_root *root = fs_info->dev_root; 2616 struct btrfs_trans_handle *trans; 2617 struct btrfs_device *device; 2618 struct block_device *bdev; 2619 struct super_block *sb = fs_info->sb; 2620 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2621 struct btrfs_fs_devices *seed_devices; 2622 u64 orig_super_total_bytes; 2623 u64 orig_super_num_devices; 2624 int ret = 0; 2625 bool seeding_dev = false; 2626 bool locked = false; 2627 2628 if (sb_rdonly(sb) && !fs_devices->seeding) 2629 return -EROFS; 2630 2631 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2632 fs_info->bdev_holder); 2633 if (IS_ERR(bdev)) 2634 return PTR_ERR(bdev); 2635 2636 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2637 ret = -EINVAL; 2638 goto error; 2639 } 2640 2641 if (fs_devices->seeding) { 2642 seeding_dev = true; 2643 down_write(&sb->s_umount); 2644 mutex_lock(&uuid_mutex); 2645 locked = true; 2646 } 2647 2648 sync_blockdev(bdev); 2649 2650 rcu_read_lock(); 2651 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2652 if (device->bdev == bdev) { 2653 ret = -EEXIST; 2654 rcu_read_unlock(); 2655 goto error; 2656 } 2657 } 2658 rcu_read_unlock(); 2659 2660 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path); 2661 if (IS_ERR(device)) { 2662 /* we can safely leave the fs_devices entry around */ 2663 ret = PTR_ERR(device); 2664 goto error; 2665 } 2666 2667 device->fs_info = fs_info; 2668 device->bdev = bdev; 2669 ret = lookup_bdev(device_path, &device->devt); 2670 if (ret) 2671 goto error_free_device; 2672 2673 ret = btrfs_get_dev_zone_info(device, false); 2674 if (ret) 2675 goto error_free_device; 2676 2677 trans = btrfs_start_transaction(root, 0); 2678 if (IS_ERR(trans)) { 2679 ret = PTR_ERR(trans); 2680 goto error_free_zone; 2681 } 2682 2683 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2684 device->generation = trans->transid; 2685 device->io_width = fs_info->sectorsize; 2686 device->io_align = fs_info->sectorsize; 2687 device->sector_size = fs_info->sectorsize; 2688 device->total_bytes = 2689 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); 2690 device->disk_total_bytes = device->total_bytes; 2691 device->commit_total_bytes = device->total_bytes; 2692 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2693 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2694 device->mode = FMODE_EXCL; 2695 device->dev_stats_valid = 1; 2696 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2697 2698 if (seeding_dev) { 2699 btrfs_clear_sb_rdonly(sb); 2700 2701 /* GFP_KERNEL allocation must not be under device_list_mutex */ 2702 seed_devices = btrfs_init_sprout(fs_info); 2703 if (IS_ERR(seed_devices)) { 2704 ret = PTR_ERR(seed_devices); 2705 btrfs_abort_transaction(trans, ret); 2706 goto error_trans; 2707 } 2708 } 2709 2710 mutex_lock(&fs_devices->device_list_mutex); 2711 if (seeding_dev) { 2712 btrfs_setup_sprout(fs_info, seed_devices); 2713 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2714 device); 2715 } 2716 2717 device->fs_devices = fs_devices; 2718 2719 mutex_lock(&fs_info->chunk_mutex); 2720 list_add_rcu(&device->dev_list, &fs_devices->devices); 2721 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2722 fs_devices->num_devices++; 2723 fs_devices->open_devices++; 2724 fs_devices->rw_devices++; 2725 fs_devices->total_devices++; 2726 fs_devices->total_rw_bytes += device->total_bytes; 2727 2728 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2729 2730 if (!bdev_nonrot(bdev)) 2731 fs_devices->rotating = true; 2732 2733 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2734 btrfs_set_super_total_bytes(fs_info->super_copy, 2735 round_down(orig_super_total_bytes + device->total_bytes, 2736 fs_info->sectorsize)); 2737 2738 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2739 btrfs_set_super_num_devices(fs_info->super_copy, 2740 orig_super_num_devices + 1); 2741 2742 /* 2743 * we've got more storage, clear any full flags on the space 2744 * infos 2745 */ 2746 btrfs_clear_space_info_full(fs_info); 2747 2748 mutex_unlock(&fs_info->chunk_mutex); 2749 2750 /* Add sysfs device entry */ 2751 btrfs_sysfs_add_device(device); 2752 2753 mutex_unlock(&fs_devices->device_list_mutex); 2754 2755 if (seeding_dev) { 2756 mutex_lock(&fs_info->chunk_mutex); 2757 ret = init_first_rw_device(trans); 2758 mutex_unlock(&fs_info->chunk_mutex); 2759 if (ret) { 2760 btrfs_abort_transaction(trans, ret); 2761 goto error_sysfs; 2762 } 2763 } 2764 2765 ret = btrfs_add_dev_item(trans, device); 2766 if (ret) { 2767 btrfs_abort_transaction(trans, ret); 2768 goto error_sysfs; 2769 } 2770 2771 if (seeding_dev) { 2772 ret = btrfs_finish_sprout(trans); 2773 if (ret) { 2774 btrfs_abort_transaction(trans, ret); 2775 goto error_sysfs; 2776 } 2777 2778 /* 2779 * fs_devices now represents the newly sprouted filesystem and 2780 * its fsid has been changed by btrfs_sprout_splice(). 2781 */ 2782 btrfs_sysfs_update_sprout_fsid(fs_devices); 2783 } 2784 2785 ret = btrfs_commit_transaction(trans); 2786 2787 if (seeding_dev) { 2788 mutex_unlock(&uuid_mutex); 2789 up_write(&sb->s_umount); 2790 locked = false; 2791 2792 if (ret) /* transaction commit */ 2793 return ret; 2794 2795 ret = btrfs_relocate_sys_chunks(fs_info); 2796 if (ret < 0) 2797 btrfs_handle_fs_error(fs_info, ret, 2798 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2799 trans = btrfs_attach_transaction(root); 2800 if (IS_ERR(trans)) { 2801 if (PTR_ERR(trans) == -ENOENT) 2802 return 0; 2803 ret = PTR_ERR(trans); 2804 trans = NULL; 2805 goto error_sysfs; 2806 } 2807 ret = btrfs_commit_transaction(trans); 2808 } 2809 2810 /* 2811 * Now that we have written a new super block to this device, check all 2812 * other fs_devices list if device_path alienates any other scanned 2813 * device. 2814 * We can ignore the return value as it typically returns -EINVAL and 2815 * only succeeds if the device was an alien. 2816 */ 2817 btrfs_forget_devices(device->devt); 2818 2819 /* Update ctime/mtime for blkid or udev */ 2820 update_dev_time(device_path); 2821 2822 return ret; 2823 2824 error_sysfs: 2825 btrfs_sysfs_remove_device(device); 2826 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2827 mutex_lock(&fs_info->chunk_mutex); 2828 list_del_rcu(&device->dev_list); 2829 list_del(&device->dev_alloc_list); 2830 fs_info->fs_devices->num_devices--; 2831 fs_info->fs_devices->open_devices--; 2832 fs_info->fs_devices->rw_devices--; 2833 fs_info->fs_devices->total_devices--; 2834 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2835 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2836 btrfs_set_super_total_bytes(fs_info->super_copy, 2837 orig_super_total_bytes); 2838 btrfs_set_super_num_devices(fs_info->super_copy, 2839 orig_super_num_devices); 2840 mutex_unlock(&fs_info->chunk_mutex); 2841 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2842 error_trans: 2843 if (seeding_dev) 2844 btrfs_set_sb_rdonly(sb); 2845 if (trans) 2846 btrfs_end_transaction(trans); 2847 error_free_zone: 2848 btrfs_destroy_dev_zone_info(device); 2849 error_free_device: 2850 btrfs_free_device(device); 2851 error: 2852 blkdev_put(bdev, FMODE_EXCL); 2853 if (locked) { 2854 mutex_unlock(&uuid_mutex); 2855 up_write(&sb->s_umount); 2856 } 2857 return ret; 2858 } 2859 2860 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2861 struct btrfs_device *device) 2862 { 2863 int ret; 2864 struct btrfs_path *path; 2865 struct btrfs_root *root = device->fs_info->chunk_root; 2866 struct btrfs_dev_item *dev_item; 2867 struct extent_buffer *leaf; 2868 struct btrfs_key key; 2869 2870 path = btrfs_alloc_path(); 2871 if (!path) 2872 return -ENOMEM; 2873 2874 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2875 key.type = BTRFS_DEV_ITEM_KEY; 2876 key.offset = device->devid; 2877 2878 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2879 if (ret < 0) 2880 goto out; 2881 2882 if (ret > 0) { 2883 ret = -ENOENT; 2884 goto out; 2885 } 2886 2887 leaf = path->nodes[0]; 2888 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2889 2890 btrfs_set_device_id(leaf, dev_item, device->devid); 2891 btrfs_set_device_type(leaf, dev_item, device->type); 2892 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2893 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2894 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2895 btrfs_set_device_total_bytes(leaf, dev_item, 2896 btrfs_device_get_disk_total_bytes(device)); 2897 btrfs_set_device_bytes_used(leaf, dev_item, 2898 btrfs_device_get_bytes_used(device)); 2899 btrfs_mark_buffer_dirty(leaf); 2900 2901 out: 2902 btrfs_free_path(path); 2903 return ret; 2904 } 2905 2906 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2907 struct btrfs_device *device, u64 new_size) 2908 { 2909 struct btrfs_fs_info *fs_info = device->fs_info; 2910 struct btrfs_super_block *super_copy = fs_info->super_copy; 2911 u64 old_total; 2912 u64 diff; 2913 int ret; 2914 2915 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2916 return -EACCES; 2917 2918 new_size = round_down(new_size, fs_info->sectorsize); 2919 2920 mutex_lock(&fs_info->chunk_mutex); 2921 old_total = btrfs_super_total_bytes(super_copy); 2922 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2923 2924 if (new_size <= device->total_bytes || 2925 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2926 mutex_unlock(&fs_info->chunk_mutex); 2927 return -EINVAL; 2928 } 2929 2930 btrfs_set_super_total_bytes(super_copy, 2931 round_down(old_total + diff, fs_info->sectorsize)); 2932 device->fs_devices->total_rw_bytes += diff; 2933 2934 btrfs_device_set_total_bytes(device, new_size); 2935 btrfs_device_set_disk_total_bytes(device, new_size); 2936 btrfs_clear_space_info_full(device->fs_info); 2937 if (list_empty(&device->post_commit_list)) 2938 list_add_tail(&device->post_commit_list, 2939 &trans->transaction->dev_update_list); 2940 mutex_unlock(&fs_info->chunk_mutex); 2941 2942 btrfs_reserve_chunk_metadata(trans, false); 2943 ret = btrfs_update_device(trans, device); 2944 btrfs_trans_release_chunk_metadata(trans); 2945 2946 return ret; 2947 } 2948 2949 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2950 { 2951 struct btrfs_fs_info *fs_info = trans->fs_info; 2952 struct btrfs_root *root = fs_info->chunk_root; 2953 int ret; 2954 struct btrfs_path *path; 2955 struct btrfs_key key; 2956 2957 path = btrfs_alloc_path(); 2958 if (!path) 2959 return -ENOMEM; 2960 2961 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2962 key.offset = chunk_offset; 2963 key.type = BTRFS_CHUNK_ITEM_KEY; 2964 2965 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2966 if (ret < 0) 2967 goto out; 2968 else if (ret > 0) { /* Logic error or corruption */ 2969 btrfs_handle_fs_error(fs_info, -ENOENT, 2970 "Failed lookup while freeing chunk."); 2971 ret = -ENOENT; 2972 goto out; 2973 } 2974 2975 ret = btrfs_del_item(trans, root, path); 2976 if (ret < 0) 2977 btrfs_handle_fs_error(fs_info, ret, 2978 "Failed to delete chunk item."); 2979 out: 2980 btrfs_free_path(path); 2981 return ret; 2982 } 2983 2984 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2985 { 2986 struct btrfs_super_block *super_copy = fs_info->super_copy; 2987 struct btrfs_disk_key *disk_key; 2988 struct btrfs_chunk *chunk; 2989 u8 *ptr; 2990 int ret = 0; 2991 u32 num_stripes; 2992 u32 array_size; 2993 u32 len = 0; 2994 u32 cur; 2995 struct btrfs_key key; 2996 2997 lockdep_assert_held(&fs_info->chunk_mutex); 2998 array_size = btrfs_super_sys_array_size(super_copy); 2999 3000 ptr = super_copy->sys_chunk_array; 3001 cur = 0; 3002 3003 while (cur < array_size) { 3004 disk_key = (struct btrfs_disk_key *)ptr; 3005 btrfs_disk_key_to_cpu(&key, disk_key); 3006 3007 len = sizeof(*disk_key); 3008 3009 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 3010 chunk = (struct btrfs_chunk *)(ptr + len); 3011 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 3012 len += btrfs_chunk_item_size(num_stripes); 3013 } else { 3014 ret = -EIO; 3015 break; 3016 } 3017 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 3018 key.offset == chunk_offset) { 3019 memmove(ptr, ptr + len, array_size - (cur + len)); 3020 array_size -= len; 3021 btrfs_set_super_sys_array_size(super_copy, array_size); 3022 } else { 3023 ptr += len; 3024 cur += len; 3025 } 3026 } 3027 return ret; 3028 } 3029 3030 /* 3031 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 3032 * @logical: Logical block offset in bytes. 3033 * @length: Length of extent in bytes. 3034 * 3035 * Return: Chunk mapping or ERR_PTR. 3036 */ 3037 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 3038 u64 logical, u64 length) 3039 { 3040 struct extent_map_tree *em_tree; 3041 struct extent_map *em; 3042 3043 em_tree = &fs_info->mapping_tree; 3044 read_lock(&em_tree->lock); 3045 em = lookup_extent_mapping(em_tree, logical, length); 3046 read_unlock(&em_tree->lock); 3047 3048 if (!em) { 3049 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 3050 logical, length); 3051 return ERR_PTR(-EINVAL); 3052 } 3053 3054 if (em->start > logical || em->start + em->len < logical) { 3055 btrfs_crit(fs_info, 3056 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3057 logical, length, em->start, em->start + em->len); 3058 free_extent_map(em); 3059 return ERR_PTR(-EINVAL); 3060 } 3061 3062 /* callers are responsible for dropping em's ref. */ 3063 return em; 3064 } 3065 3066 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3067 struct map_lookup *map, u64 chunk_offset) 3068 { 3069 int i; 3070 3071 /* 3072 * Removing chunk items and updating the device items in the chunks btree 3073 * requires holding the chunk_mutex. 3074 * See the comment at btrfs_chunk_alloc() for the details. 3075 */ 3076 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3077 3078 for (i = 0; i < map->num_stripes; i++) { 3079 int ret; 3080 3081 ret = btrfs_update_device(trans, map->stripes[i].dev); 3082 if (ret) 3083 return ret; 3084 } 3085 3086 return btrfs_free_chunk(trans, chunk_offset); 3087 } 3088 3089 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3090 { 3091 struct btrfs_fs_info *fs_info = trans->fs_info; 3092 struct extent_map *em; 3093 struct map_lookup *map; 3094 u64 dev_extent_len = 0; 3095 int i, ret = 0; 3096 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3097 3098 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3099 if (IS_ERR(em)) { 3100 /* 3101 * This is a logic error, but we don't want to just rely on the 3102 * user having built with ASSERT enabled, so if ASSERT doesn't 3103 * do anything we still error out. 3104 */ 3105 ASSERT(0); 3106 return PTR_ERR(em); 3107 } 3108 map = em->map_lookup; 3109 3110 /* 3111 * First delete the device extent items from the devices btree. 3112 * We take the device_list_mutex to avoid racing with the finishing phase 3113 * of a device replace operation. See the comment below before acquiring 3114 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3115 * because that can result in a deadlock when deleting the device extent 3116 * items from the devices btree - COWing an extent buffer from the btree 3117 * may result in allocating a new metadata chunk, which would attempt to 3118 * lock again fs_info->chunk_mutex. 3119 */ 3120 mutex_lock(&fs_devices->device_list_mutex); 3121 for (i = 0; i < map->num_stripes; i++) { 3122 struct btrfs_device *device = map->stripes[i].dev; 3123 ret = btrfs_free_dev_extent(trans, device, 3124 map->stripes[i].physical, 3125 &dev_extent_len); 3126 if (ret) { 3127 mutex_unlock(&fs_devices->device_list_mutex); 3128 btrfs_abort_transaction(trans, ret); 3129 goto out; 3130 } 3131 3132 if (device->bytes_used > 0) { 3133 mutex_lock(&fs_info->chunk_mutex); 3134 btrfs_device_set_bytes_used(device, 3135 device->bytes_used - dev_extent_len); 3136 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3137 btrfs_clear_space_info_full(fs_info); 3138 mutex_unlock(&fs_info->chunk_mutex); 3139 } 3140 } 3141 mutex_unlock(&fs_devices->device_list_mutex); 3142 3143 /* 3144 * We acquire fs_info->chunk_mutex for 2 reasons: 3145 * 3146 * 1) Just like with the first phase of the chunk allocation, we must 3147 * reserve system space, do all chunk btree updates and deletions, and 3148 * update the system chunk array in the superblock while holding this 3149 * mutex. This is for similar reasons as explained on the comment at 3150 * the top of btrfs_chunk_alloc(); 3151 * 3152 * 2) Prevent races with the final phase of a device replace operation 3153 * that replaces the device object associated with the map's stripes, 3154 * because the device object's id can change at any time during that 3155 * final phase of the device replace operation 3156 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3157 * replaced device and then see it with an ID of 3158 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3159 * the device item, which does not exists on the chunk btree. 3160 * The finishing phase of device replace acquires both the 3161 * device_list_mutex and the chunk_mutex, in that order, so we are 3162 * safe by just acquiring the chunk_mutex. 3163 */ 3164 trans->removing_chunk = true; 3165 mutex_lock(&fs_info->chunk_mutex); 3166 3167 check_system_chunk(trans, map->type); 3168 3169 ret = remove_chunk_item(trans, map, chunk_offset); 3170 /* 3171 * Normally we should not get -ENOSPC since we reserved space before 3172 * through the call to check_system_chunk(). 3173 * 3174 * Despite our system space_info having enough free space, we may not 3175 * be able to allocate extents from its block groups, because all have 3176 * an incompatible profile, which will force us to allocate a new system 3177 * block group with the right profile, or right after we called 3178 * check_system_space() above, a scrub turned the only system block group 3179 * with enough free space into RO mode. 3180 * This is explained with more detail at do_chunk_alloc(). 3181 * 3182 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3183 */ 3184 if (ret == -ENOSPC) { 3185 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3186 struct btrfs_block_group *sys_bg; 3187 3188 sys_bg = btrfs_create_chunk(trans, sys_flags); 3189 if (IS_ERR(sys_bg)) { 3190 ret = PTR_ERR(sys_bg); 3191 btrfs_abort_transaction(trans, ret); 3192 goto out; 3193 } 3194 3195 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3196 if (ret) { 3197 btrfs_abort_transaction(trans, ret); 3198 goto out; 3199 } 3200 3201 ret = remove_chunk_item(trans, map, chunk_offset); 3202 if (ret) { 3203 btrfs_abort_transaction(trans, ret); 3204 goto out; 3205 } 3206 } else if (ret) { 3207 btrfs_abort_transaction(trans, ret); 3208 goto out; 3209 } 3210 3211 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3212 3213 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3214 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3215 if (ret) { 3216 btrfs_abort_transaction(trans, ret); 3217 goto out; 3218 } 3219 } 3220 3221 mutex_unlock(&fs_info->chunk_mutex); 3222 trans->removing_chunk = false; 3223 3224 /* 3225 * We are done with chunk btree updates and deletions, so release the 3226 * system space we previously reserved (with check_system_chunk()). 3227 */ 3228 btrfs_trans_release_chunk_metadata(trans); 3229 3230 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3231 if (ret) { 3232 btrfs_abort_transaction(trans, ret); 3233 goto out; 3234 } 3235 3236 out: 3237 if (trans->removing_chunk) { 3238 mutex_unlock(&fs_info->chunk_mutex); 3239 trans->removing_chunk = false; 3240 } 3241 /* once for us */ 3242 free_extent_map(em); 3243 return ret; 3244 } 3245 3246 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3247 { 3248 struct btrfs_root *root = fs_info->chunk_root; 3249 struct btrfs_trans_handle *trans; 3250 struct btrfs_block_group *block_group; 3251 u64 length; 3252 int ret; 3253 3254 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3255 btrfs_err(fs_info, 3256 "relocate: not supported on extent tree v2 yet"); 3257 return -EINVAL; 3258 } 3259 3260 /* 3261 * Prevent races with automatic removal of unused block groups. 3262 * After we relocate and before we remove the chunk with offset 3263 * chunk_offset, automatic removal of the block group can kick in, 3264 * resulting in a failure when calling btrfs_remove_chunk() below. 3265 * 3266 * Make sure to acquire this mutex before doing a tree search (dev 3267 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3268 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3269 * we release the path used to search the chunk/dev tree and before 3270 * the current task acquires this mutex and calls us. 3271 */ 3272 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3273 3274 /* step one, relocate all the extents inside this chunk */ 3275 btrfs_scrub_pause(fs_info); 3276 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3277 btrfs_scrub_continue(fs_info); 3278 if (ret) { 3279 /* 3280 * If we had a transaction abort, stop all running scrubs. 3281 * See transaction.c:cleanup_transaction() why we do it here. 3282 */ 3283 if (BTRFS_FS_ERROR(fs_info)) 3284 btrfs_scrub_cancel(fs_info); 3285 return ret; 3286 } 3287 3288 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3289 if (!block_group) 3290 return -ENOENT; 3291 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3292 length = block_group->length; 3293 btrfs_put_block_group(block_group); 3294 3295 /* 3296 * On a zoned file system, discard the whole block group, this will 3297 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3298 * resetting the zone fails, don't treat it as a fatal problem from the 3299 * filesystem's point of view. 3300 */ 3301 if (btrfs_is_zoned(fs_info)) { 3302 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3303 if (ret) 3304 btrfs_info(fs_info, 3305 "failed to reset zone %llu after relocation", 3306 chunk_offset); 3307 } 3308 3309 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3310 chunk_offset); 3311 if (IS_ERR(trans)) { 3312 ret = PTR_ERR(trans); 3313 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3314 return ret; 3315 } 3316 3317 /* 3318 * step two, delete the device extents and the 3319 * chunk tree entries 3320 */ 3321 ret = btrfs_remove_chunk(trans, chunk_offset); 3322 btrfs_end_transaction(trans); 3323 return ret; 3324 } 3325 3326 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3327 { 3328 struct btrfs_root *chunk_root = fs_info->chunk_root; 3329 struct btrfs_path *path; 3330 struct extent_buffer *leaf; 3331 struct btrfs_chunk *chunk; 3332 struct btrfs_key key; 3333 struct btrfs_key found_key; 3334 u64 chunk_type; 3335 bool retried = false; 3336 int failed = 0; 3337 int ret; 3338 3339 path = btrfs_alloc_path(); 3340 if (!path) 3341 return -ENOMEM; 3342 3343 again: 3344 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3345 key.offset = (u64)-1; 3346 key.type = BTRFS_CHUNK_ITEM_KEY; 3347 3348 while (1) { 3349 mutex_lock(&fs_info->reclaim_bgs_lock); 3350 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3351 if (ret < 0) { 3352 mutex_unlock(&fs_info->reclaim_bgs_lock); 3353 goto error; 3354 } 3355 BUG_ON(ret == 0); /* Corruption */ 3356 3357 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3358 key.type); 3359 if (ret) 3360 mutex_unlock(&fs_info->reclaim_bgs_lock); 3361 if (ret < 0) 3362 goto error; 3363 if (ret > 0) 3364 break; 3365 3366 leaf = path->nodes[0]; 3367 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3368 3369 chunk = btrfs_item_ptr(leaf, path->slots[0], 3370 struct btrfs_chunk); 3371 chunk_type = btrfs_chunk_type(leaf, chunk); 3372 btrfs_release_path(path); 3373 3374 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3375 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3376 if (ret == -ENOSPC) 3377 failed++; 3378 else 3379 BUG_ON(ret); 3380 } 3381 mutex_unlock(&fs_info->reclaim_bgs_lock); 3382 3383 if (found_key.offset == 0) 3384 break; 3385 key.offset = found_key.offset - 1; 3386 } 3387 ret = 0; 3388 if (failed && !retried) { 3389 failed = 0; 3390 retried = true; 3391 goto again; 3392 } else if (WARN_ON(failed && retried)) { 3393 ret = -ENOSPC; 3394 } 3395 error: 3396 btrfs_free_path(path); 3397 return ret; 3398 } 3399 3400 /* 3401 * return 1 : allocate a data chunk successfully, 3402 * return <0: errors during allocating a data chunk, 3403 * return 0 : no need to allocate a data chunk. 3404 */ 3405 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3406 u64 chunk_offset) 3407 { 3408 struct btrfs_block_group *cache; 3409 u64 bytes_used; 3410 u64 chunk_type; 3411 3412 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3413 ASSERT(cache); 3414 chunk_type = cache->flags; 3415 btrfs_put_block_group(cache); 3416 3417 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3418 return 0; 3419 3420 spin_lock(&fs_info->data_sinfo->lock); 3421 bytes_used = fs_info->data_sinfo->bytes_used; 3422 spin_unlock(&fs_info->data_sinfo->lock); 3423 3424 if (!bytes_used) { 3425 struct btrfs_trans_handle *trans; 3426 int ret; 3427 3428 trans = btrfs_join_transaction(fs_info->tree_root); 3429 if (IS_ERR(trans)) 3430 return PTR_ERR(trans); 3431 3432 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3433 btrfs_end_transaction(trans); 3434 if (ret < 0) 3435 return ret; 3436 return 1; 3437 } 3438 3439 return 0; 3440 } 3441 3442 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3443 struct btrfs_balance_control *bctl) 3444 { 3445 struct btrfs_root *root = fs_info->tree_root; 3446 struct btrfs_trans_handle *trans; 3447 struct btrfs_balance_item *item; 3448 struct btrfs_disk_balance_args disk_bargs; 3449 struct btrfs_path *path; 3450 struct extent_buffer *leaf; 3451 struct btrfs_key key; 3452 int ret, err; 3453 3454 path = btrfs_alloc_path(); 3455 if (!path) 3456 return -ENOMEM; 3457 3458 trans = btrfs_start_transaction(root, 0); 3459 if (IS_ERR(trans)) { 3460 btrfs_free_path(path); 3461 return PTR_ERR(trans); 3462 } 3463 3464 key.objectid = BTRFS_BALANCE_OBJECTID; 3465 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3466 key.offset = 0; 3467 3468 ret = btrfs_insert_empty_item(trans, root, path, &key, 3469 sizeof(*item)); 3470 if (ret) 3471 goto out; 3472 3473 leaf = path->nodes[0]; 3474 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3475 3476 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3477 3478 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3479 btrfs_set_balance_data(leaf, item, &disk_bargs); 3480 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3481 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3482 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3483 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3484 3485 btrfs_set_balance_flags(leaf, item, bctl->flags); 3486 3487 btrfs_mark_buffer_dirty(leaf); 3488 out: 3489 btrfs_free_path(path); 3490 err = btrfs_commit_transaction(trans); 3491 if (err && !ret) 3492 ret = err; 3493 return ret; 3494 } 3495 3496 static int del_balance_item(struct btrfs_fs_info *fs_info) 3497 { 3498 struct btrfs_root *root = fs_info->tree_root; 3499 struct btrfs_trans_handle *trans; 3500 struct btrfs_path *path; 3501 struct btrfs_key key; 3502 int ret, err; 3503 3504 path = btrfs_alloc_path(); 3505 if (!path) 3506 return -ENOMEM; 3507 3508 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3509 if (IS_ERR(trans)) { 3510 btrfs_free_path(path); 3511 return PTR_ERR(trans); 3512 } 3513 3514 key.objectid = BTRFS_BALANCE_OBJECTID; 3515 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3516 key.offset = 0; 3517 3518 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3519 if (ret < 0) 3520 goto out; 3521 if (ret > 0) { 3522 ret = -ENOENT; 3523 goto out; 3524 } 3525 3526 ret = btrfs_del_item(trans, root, path); 3527 out: 3528 btrfs_free_path(path); 3529 err = btrfs_commit_transaction(trans); 3530 if (err && !ret) 3531 ret = err; 3532 return ret; 3533 } 3534 3535 /* 3536 * This is a heuristic used to reduce the number of chunks balanced on 3537 * resume after balance was interrupted. 3538 */ 3539 static void update_balance_args(struct btrfs_balance_control *bctl) 3540 { 3541 /* 3542 * Turn on soft mode for chunk types that were being converted. 3543 */ 3544 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3545 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3546 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3547 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3548 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3549 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3550 3551 /* 3552 * Turn on usage filter if is not already used. The idea is 3553 * that chunks that we have already balanced should be 3554 * reasonably full. Don't do it for chunks that are being 3555 * converted - that will keep us from relocating unconverted 3556 * (albeit full) chunks. 3557 */ 3558 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3559 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3560 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3561 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3562 bctl->data.usage = 90; 3563 } 3564 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3565 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3566 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3567 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3568 bctl->sys.usage = 90; 3569 } 3570 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3571 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3572 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3573 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3574 bctl->meta.usage = 90; 3575 } 3576 } 3577 3578 /* 3579 * Clear the balance status in fs_info and delete the balance item from disk. 3580 */ 3581 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3582 { 3583 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3584 int ret; 3585 3586 BUG_ON(!fs_info->balance_ctl); 3587 3588 spin_lock(&fs_info->balance_lock); 3589 fs_info->balance_ctl = NULL; 3590 spin_unlock(&fs_info->balance_lock); 3591 3592 kfree(bctl); 3593 ret = del_balance_item(fs_info); 3594 if (ret) 3595 btrfs_handle_fs_error(fs_info, ret, NULL); 3596 } 3597 3598 /* 3599 * Balance filters. Return 1 if chunk should be filtered out 3600 * (should not be balanced). 3601 */ 3602 static int chunk_profiles_filter(u64 chunk_type, 3603 struct btrfs_balance_args *bargs) 3604 { 3605 chunk_type = chunk_to_extended(chunk_type) & 3606 BTRFS_EXTENDED_PROFILE_MASK; 3607 3608 if (bargs->profiles & chunk_type) 3609 return 0; 3610 3611 return 1; 3612 } 3613 3614 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3615 struct btrfs_balance_args *bargs) 3616 { 3617 struct btrfs_block_group *cache; 3618 u64 chunk_used; 3619 u64 user_thresh_min; 3620 u64 user_thresh_max; 3621 int ret = 1; 3622 3623 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3624 chunk_used = cache->used; 3625 3626 if (bargs->usage_min == 0) 3627 user_thresh_min = 0; 3628 else 3629 user_thresh_min = mult_perc(cache->length, bargs->usage_min); 3630 3631 if (bargs->usage_max == 0) 3632 user_thresh_max = 1; 3633 else if (bargs->usage_max > 100) 3634 user_thresh_max = cache->length; 3635 else 3636 user_thresh_max = mult_perc(cache->length, bargs->usage_max); 3637 3638 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3639 ret = 0; 3640 3641 btrfs_put_block_group(cache); 3642 return ret; 3643 } 3644 3645 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3646 u64 chunk_offset, struct btrfs_balance_args *bargs) 3647 { 3648 struct btrfs_block_group *cache; 3649 u64 chunk_used, user_thresh; 3650 int ret = 1; 3651 3652 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3653 chunk_used = cache->used; 3654 3655 if (bargs->usage_min == 0) 3656 user_thresh = 1; 3657 else if (bargs->usage > 100) 3658 user_thresh = cache->length; 3659 else 3660 user_thresh = mult_perc(cache->length, bargs->usage); 3661 3662 if (chunk_used < user_thresh) 3663 ret = 0; 3664 3665 btrfs_put_block_group(cache); 3666 return ret; 3667 } 3668 3669 static int chunk_devid_filter(struct extent_buffer *leaf, 3670 struct btrfs_chunk *chunk, 3671 struct btrfs_balance_args *bargs) 3672 { 3673 struct btrfs_stripe *stripe; 3674 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3675 int i; 3676 3677 for (i = 0; i < num_stripes; i++) { 3678 stripe = btrfs_stripe_nr(chunk, i); 3679 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3680 return 0; 3681 } 3682 3683 return 1; 3684 } 3685 3686 static u64 calc_data_stripes(u64 type, int num_stripes) 3687 { 3688 const int index = btrfs_bg_flags_to_raid_index(type); 3689 const int ncopies = btrfs_raid_array[index].ncopies; 3690 const int nparity = btrfs_raid_array[index].nparity; 3691 3692 return (num_stripes - nparity) / ncopies; 3693 } 3694 3695 /* [pstart, pend) */ 3696 static int chunk_drange_filter(struct extent_buffer *leaf, 3697 struct btrfs_chunk *chunk, 3698 struct btrfs_balance_args *bargs) 3699 { 3700 struct btrfs_stripe *stripe; 3701 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3702 u64 stripe_offset; 3703 u64 stripe_length; 3704 u64 type; 3705 int factor; 3706 int i; 3707 3708 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3709 return 0; 3710 3711 type = btrfs_chunk_type(leaf, chunk); 3712 factor = calc_data_stripes(type, num_stripes); 3713 3714 for (i = 0; i < num_stripes; i++) { 3715 stripe = btrfs_stripe_nr(chunk, i); 3716 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3717 continue; 3718 3719 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3720 stripe_length = btrfs_chunk_length(leaf, chunk); 3721 stripe_length = div_u64(stripe_length, factor); 3722 3723 if (stripe_offset < bargs->pend && 3724 stripe_offset + stripe_length > bargs->pstart) 3725 return 0; 3726 } 3727 3728 return 1; 3729 } 3730 3731 /* [vstart, vend) */ 3732 static int chunk_vrange_filter(struct extent_buffer *leaf, 3733 struct btrfs_chunk *chunk, 3734 u64 chunk_offset, 3735 struct btrfs_balance_args *bargs) 3736 { 3737 if (chunk_offset < bargs->vend && 3738 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3739 /* at least part of the chunk is inside this vrange */ 3740 return 0; 3741 3742 return 1; 3743 } 3744 3745 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3746 struct btrfs_chunk *chunk, 3747 struct btrfs_balance_args *bargs) 3748 { 3749 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3750 3751 if (bargs->stripes_min <= num_stripes 3752 && num_stripes <= bargs->stripes_max) 3753 return 0; 3754 3755 return 1; 3756 } 3757 3758 static int chunk_soft_convert_filter(u64 chunk_type, 3759 struct btrfs_balance_args *bargs) 3760 { 3761 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3762 return 0; 3763 3764 chunk_type = chunk_to_extended(chunk_type) & 3765 BTRFS_EXTENDED_PROFILE_MASK; 3766 3767 if (bargs->target == chunk_type) 3768 return 1; 3769 3770 return 0; 3771 } 3772 3773 static int should_balance_chunk(struct extent_buffer *leaf, 3774 struct btrfs_chunk *chunk, u64 chunk_offset) 3775 { 3776 struct btrfs_fs_info *fs_info = leaf->fs_info; 3777 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3778 struct btrfs_balance_args *bargs = NULL; 3779 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3780 3781 /* type filter */ 3782 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3783 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3784 return 0; 3785 } 3786 3787 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3788 bargs = &bctl->data; 3789 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3790 bargs = &bctl->sys; 3791 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3792 bargs = &bctl->meta; 3793 3794 /* profiles filter */ 3795 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3796 chunk_profiles_filter(chunk_type, bargs)) { 3797 return 0; 3798 } 3799 3800 /* usage filter */ 3801 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3802 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3803 return 0; 3804 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3805 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3806 return 0; 3807 } 3808 3809 /* devid filter */ 3810 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3811 chunk_devid_filter(leaf, chunk, bargs)) { 3812 return 0; 3813 } 3814 3815 /* drange filter, makes sense only with devid filter */ 3816 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3817 chunk_drange_filter(leaf, chunk, bargs)) { 3818 return 0; 3819 } 3820 3821 /* vrange filter */ 3822 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3823 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3824 return 0; 3825 } 3826 3827 /* stripes filter */ 3828 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3829 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3830 return 0; 3831 } 3832 3833 /* soft profile changing mode */ 3834 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3835 chunk_soft_convert_filter(chunk_type, bargs)) { 3836 return 0; 3837 } 3838 3839 /* 3840 * limited by count, must be the last filter 3841 */ 3842 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3843 if (bargs->limit == 0) 3844 return 0; 3845 else 3846 bargs->limit--; 3847 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3848 /* 3849 * Same logic as the 'limit' filter; the minimum cannot be 3850 * determined here because we do not have the global information 3851 * about the count of all chunks that satisfy the filters. 3852 */ 3853 if (bargs->limit_max == 0) 3854 return 0; 3855 else 3856 bargs->limit_max--; 3857 } 3858 3859 return 1; 3860 } 3861 3862 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3863 { 3864 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3865 struct btrfs_root *chunk_root = fs_info->chunk_root; 3866 u64 chunk_type; 3867 struct btrfs_chunk *chunk; 3868 struct btrfs_path *path = NULL; 3869 struct btrfs_key key; 3870 struct btrfs_key found_key; 3871 struct extent_buffer *leaf; 3872 int slot; 3873 int ret; 3874 int enospc_errors = 0; 3875 bool counting = true; 3876 /* The single value limit and min/max limits use the same bytes in the */ 3877 u64 limit_data = bctl->data.limit; 3878 u64 limit_meta = bctl->meta.limit; 3879 u64 limit_sys = bctl->sys.limit; 3880 u32 count_data = 0; 3881 u32 count_meta = 0; 3882 u32 count_sys = 0; 3883 int chunk_reserved = 0; 3884 3885 path = btrfs_alloc_path(); 3886 if (!path) { 3887 ret = -ENOMEM; 3888 goto error; 3889 } 3890 3891 /* zero out stat counters */ 3892 spin_lock(&fs_info->balance_lock); 3893 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3894 spin_unlock(&fs_info->balance_lock); 3895 again: 3896 if (!counting) { 3897 /* 3898 * The single value limit and min/max limits use the same bytes 3899 * in the 3900 */ 3901 bctl->data.limit = limit_data; 3902 bctl->meta.limit = limit_meta; 3903 bctl->sys.limit = limit_sys; 3904 } 3905 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3906 key.offset = (u64)-1; 3907 key.type = BTRFS_CHUNK_ITEM_KEY; 3908 3909 while (1) { 3910 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3911 atomic_read(&fs_info->balance_cancel_req)) { 3912 ret = -ECANCELED; 3913 goto error; 3914 } 3915 3916 mutex_lock(&fs_info->reclaim_bgs_lock); 3917 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3918 if (ret < 0) { 3919 mutex_unlock(&fs_info->reclaim_bgs_lock); 3920 goto error; 3921 } 3922 3923 /* 3924 * this shouldn't happen, it means the last relocate 3925 * failed 3926 */ 3927 if (ret == 0) 3928 BUG(); /* FIXME break ? */ 3929 3930 ret = btrfs_previous_item(chunk_root, path, 0, 3931 BTRFS_CHUNK_ITEM_KEY); 3932 if (ret) { 3933 mutex_unlock(&fs_info->reclaim_bgs_lock); 3934 ret = 0; 3935 break; 3936 } 3937 3938 leaf = path->nodes[0]; 3939 slot = path->slots[0]; 3940 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3941 3942 if (found_key.objectid != key.objectid) { 3943 mutex_unlock(&fs_info->reclaim_bgs_lock); 3944 break; 3945 } 3946 3947 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3948 chunk_type = btrfs_chunk_type(leaf, chunk); 3949 3950 if (!counting) { 3951 spin_lock(&fs_info->balance_lock); 3952 bctl->stat.considered++; 3953 spin_unlock(&fs_info->balance_lock); 3954 } 3955 3956 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3957 3958 btrfs_release_path(path); 3959 if (!ret) { 3960 mutex_unlock(&fs_info->reclaim_bgs_lock); 3961 goto loop; 3962 } 3963 3964 if (counting) { 3965 mutex_unlock(&fs_info->reclaim_bgs_lock); 3966 spin_lock(&fs_info->balance_lock); 3967 bctl->stat.expected++; 3968 spin_unlock(&fs_info->balance_lock); 3969 3970 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3971 count_data++; 3972 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3973 count_sys++; 3974 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3975 count_meta++; 3976 3977 goto loop; 3978 } 3979 3980 /* 3981 * Apply limit_min filter, no need to check if the LIMITS 3982 * filter is used, limit_min is 0 by default 3983 */ 3984 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3985 count_data < bctl->data.limit_min) 3986 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3987 count_meta < bctl->meta.limit_min) 3988 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3989 count_sys < bctl->sys.limit_min)) { 3990 mutex_unlock(&fs_info->reclaim_bgs_lock); 3991 goto loop; 3992 } 3993 3994 if (!chunk_reserved) { 3995 /* 3996 * We may be relocating the only data chunk we have, 3997 * which could potentially end up with losing data's 3998 * raid profile, so lets allocate an empty one in 3999 * advance. 4000 */ 4001 ret = btrfs_may_alloc_data_chunk(fs_info, 4002 found_key.offset); 4003 if (ret < 0) { 4004 mutex_unlock(&fs_info->reclaim_bgs_lock); 4005 goto error; 4006 } else if (ret == 1) { 4007 chunk_reserved = 1; 4008 } 4009 } 4010 4011 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 4012 mutex_unlock(&fs_info->reclaim_bgs_lock); 4013 if (ret == -ENOSPC) { 4014 enospc_errors++; 4015 } else if (ret == -ETXTBSY) { 4016 btrfs_info(fs_info, 4017 "skipping relocation of block group %llu due to active swapfile", 4018 found_key.offset); 4019 ret = 0; 4020 } else if (ret) { 4021 goto error; 4022 } else { 4023 spin_lock(&fs_info->balance_lock); 4024 bctl->stat.completed++; 4025 spin_unlock(&fs_info->balance_lock); 4026 } 4027 loop: 4028 if (found_key.offset == 0) 4029 break; 4030 key.offset = found_key.offset - 1; 4031 } 4032 4033 if (counting) { 4034 btrfs_release_path(path); 4035 counting = false; 4036 goto again; 4037 } 4038 error: 4039 btrfs_free_path(path); 4040 if (enospc_errors) { 4041 btrfs_info(fs_info, "%d enospc errors during balance", 4042 enospc_errors); 4043 if (!ret) 4044 ret = -ENOSPC; 4045 } 4046 4047 return ret; 4048 } 4049 4050 /* 4051 * See if a given profile is valid and reduced. 4052 * 4053 * @flags: profile to validate 4054 * @extended: if true @flags is treated as an extended profile 4055 */ 4056 static int alloc_profile_is_valid(u64 flags, int extended) 4057 { 4058 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 4059 BTRFS_BLOCK_GROUP_PROFILE_MASK); 4060 4061 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 4062 4063 /* 1) check that all other bits are zeroed */ 4064 if (flags & ~mask) 4065 return 0; 4066 4067 /* 2) see if profile is reduced */ 4068 if (flags == 0) 4069 return !extended; /* "0" is valid for usual profiles */ 4070 4071 return has_single_bit_set(flags); 4072 } 4073 4074 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4075 { 4076 /* cancel requested || normal exit path */ 4077 return atomic_read(&fs_info->balance_cancel_req) || 4078 (atomic_read(&fs_info->balance_pause_req) == 0 && 4079 atomic_read(&fs_info->balance_cancel_req) == 0); 4080 } 4081 4082 /* 4083 * Validate target profile against allowed profiles and return true if it's OK. 4084 * Otherwise print the error message and return false. 4085 */ 4086 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4087 const struct btrfs_balance_args *bargs, 4088 u64 allowed, const char *type) 4089 { 4090 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4091 return true; 4092 4093 /* Profile is valid and does not have bits outside of the allowed set */ 4094 if (alloc_profile_is_valid(bargs->target, 1) && 4095 (bargs->target & ~allowed) == 0) 4096 return true; 4097 4098 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4099 type, btrfs_bg_type_to_raid_name(bargs->target)); 4100 return false; 4101 } 4102 4103 /* 4104 * Fill @buf with textual description of balance filter flags @bargs, up to 4105 * @size_buf including the terminating null. The output may be trimmed if it 4106 * does not fit into the provided buffer. 4107 */ 4108 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4109 u32 size_buf) 4110 { 4111 int ret; 4112 u32 size_bp = size_buf; 4113 char *bp = buf; 4114 u64 flags = bargs->flags; 4115 char tmp_buf[128] = {'\0'}; 4116 4117 if (!flags) 4118 return; 4119 4120 #define CHECK_APPEND_NOARG(a) \ 4121 do { \ 4122 ret = snprintf(bp, size_bp, (a)); \ 4123 if (ret < 0 || ret >= size_bp) \ 4124 goto out_overflow; \ 4125 size_bp -= ret; \ 4126 bp += ret; \ 4127 } while (0) 4128 4129 #define CHECK_APPEND_1ARG(a, v1) \ 4130 do { \ 4131 ret = snprintf(bp, size_bp, (a), (v1)); \ 4132 if (ret < 0 || ret >= size_bp) \ 4133 goto out_overflow; \ 4134 size_bp -= ret; \ 4135 bp += ret; \ 4136 } while (0) 4137 4138 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4139 do { \ 4140 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4141 if (ret < 0 || ret >= size_bp) \ 4142 goto out_overflow; \ 4143 size_bp -= ret; \ 4144 bp += ret; \ 4145 } while (0) 4146 4147 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4148 CHECK_APPEND_1ARG("convert=%s,", 4149 btrfs_bg_type_to_raid_name(bargs->target)); 4150 4151 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4152 CHECK_APPEND_NOARG("soft,"); 4153 4154 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4155 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4156 sizeof(tmp_buf)); 4157 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4158 } 4159 4160 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4161 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4162 4163 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4164 CHECK_APPEND_2ARG("usage=%u..%u,", 4165 bargs->usage_min, bargs->usage_max); 4166 4167 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4168 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4169 4170 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4171 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4172 bargs->pstart, bargs->pend); 4173 4174 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4175 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4176 bargs->vstart, bargs->vend); 4177 4178 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4179 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4180 4181 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4182 CHECK_APPEND_2ARG("limit=%u..%u,", 4183 bargs->limit_min, bargs->limit_max); 4184 4185 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4186 CHECK_APPEND_2ARG("stripes=%u..%u,", 4187 bargs->stripes_min, bargs->stripes_max); 4188 4189 #undef CHECK_APPEND_2ARG 4190 #undef CHECK_APPEND_1ARG 4191 #undef CHECK_APPEND_NOARG 4192 4193 out_overflow: 4194 4195 if (size_bp < size_buf) 4196 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4197 else 4198 buf[0] = '\0'; 4199 } 4200 4201 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4202 { 4203 u32 size_buf = 1024; 4204 char tmp_buf[192] = {'\0'}; 4205 char *buf; 4206 char *bp; 4207 u32 size_bp = size_buf; 4208 int ret; 4209 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4210 4211 buf = kzalloc(size_buf, GFP_KERNEL); 4212 if (!buf) 4213 return; 4214 4215 bp = buf; 4216 4217 #define CHECK_APPEND_1ARG(a, v1) \ 4218 do { \ 4219 ret = snprintf(bp, size_bp, (a), (v1)); \ 4220 if (ret < 0 || ret >= size_bp) \ 4221 goto out_overflow; \ 4222 size_bp -= ret; \ 4223 bp += ret; \ 4224 } while (0) 4225 4226 if (bctl->flags & BTRFS_BALANCE_FORCE) 4227 CHECK_APPEND_1ARG("%s", "-f "); 4228 4229 if (bctl->flags & BTRFS_BALANCE_DATA) { 4230 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4231 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4232 } 4233 4234 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4235 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4236 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4237 } 4238 4239 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4240 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4241 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4242 } 4243 4244 #undef CHECK_APPEND_1ARG 4245 4246 out_overflow: 4247 4248 if (size_bp < size_buf) 4249 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4250 btrfs_info(fs_info, "balance: %s %s", 4251 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4252 "resume" : "start", buf); 4253 4254 kfree(buf); 4255 } 4256 4257 /* 4258 * Should be called with balance mutexe held 4259 */ 4260 int btrfs_balance(struct btrfs_fs_info *fs_info, 4261 struct btrfs_balance_control *bctl, 4262 struct btrfs_ioctl_balance_args *bargs) 4263 { 4264 u64 meta_target, data_target; 4265 u64 allowed; 4266 int mixed = 0; 4267 int ret; 4268 u64 num_devices; 4269 unsigned seq; 4270 bool reducing_redundancy; 4271 int i; 4272 4273 if (btrfs_fs_closing(fs_info) || 4274 atomic_read(&fs_info->balance_pause_req) || 4275 btrfs_should_cancel_balance(fs_info)) { 4276 ret = -EINVAL; 4277 goto out; 4278 } 4279 4280 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4281 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4282 mixed = 1; 4283 4284 /* 4285 * In case of mixed groups both data and meta should be picked, 4286 * and identical options should be given for both of them. 4287 */ 4288 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4289 if (mixed && (bctl->flags & allowed)) { 4290 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4291 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4292 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4293 btrfs_err(fs_info, 4294 "balance: mixed groups data and metadata options must be the same"); 4295 ret = -EINVAL; 4296 goto out; 4297 } 4298 } 4299 4300 /* 4301 * rw_devices will not change at the moment, device add/delete/replace 4302 * are exclusive 4303 */ 4304 num_devices = fs_info->fs_devices->rw_devices; 4305 4306 /* 4307 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4308 * special bit for it, to make it easier to distinguish. Thus we need 4309 * to set it manually, or balance would refuse the profile. 4310 */ 4311 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4312 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4313 if (num_devices >= btrfs_raid_array[i].devs_min) 4314 allowed |= btrfs_raid_array[i].bg_flag; 4315 4316 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4317 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4318 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4319 ret = -EINVAL; 4320 goto out; 4321 } 4322 4323 /* 4324 * Allow to reduce metadata or system integrity only if force set for 4325 * profiles with redundancy (copies, parity) 4326 */ 4327 allowed = 0; 4328 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4329 if (btrfs_raid_array[i].ncopies >= 2 || 4330 btrfs_raid_array[i].tolerated_failures >= 1) 4331 allowed |= btrfs_raid_array[i].bg_flag; 4332 } 4333 do { 4334 seq = read_seqbegin(&fs_info->profiles_lock); 4335 4336 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4337 (fs_info->avail_system_alloc_bits & allowed) && 4338 !(bctl->sys.target & allowed)) || 4339 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4340 (fs_info->avail_metadata_alloc_bits & allowed) && 4341 !(bctl->meta.target & allowed))) 4342 reducing_redundancy = true; 4343 else 4344 reducing_redundancy = false; 4345 4346 /* if we're not converting, the target field is uninitialized */ 4347 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4348 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4349 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4350 bctl->data.target : fs_info->avail_data_alloc_bits; 4351 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4352 4353 if (reducing_redundancy) { 4354 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4355 btrfs_info(fs_info, 4356 "balance: force reducing metadata redundancy"); 4357 } else { 4358 btrfs_err(fs_info, 4359 "balance: reduces metadata redundancy, use --force if you want this"); 4360 ret = -EINVAL; 4361 goto out; 4362 } 4363 } 4364 4365 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4366 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4367 btrfs_warn(fs_info, 4368 "balance: metadata profile %s has lower redundancy than data profile %s", 4369 btrfs_bg_type_to_raid_name(meta_target), 4370 btrfs_bg_type_to_raid_name(data_target)); 4371 } 4372 4373 ret = insert_balance_item(fs_info, bctl); 4374 if (ret && ret != -EEXIST) 4375 goto out; 4376 4377 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4378 BUG_ON(ret == -EEXIST); 4379 BUG_ON(fs_info->balance_ctl); 4380 spin_lock(&fs_info->balance_lock); 4381 fs_info->balance_ctl = bctl; 4382 spin_unlock(&fs_info->balance_lock); 4383 } else { 4384 BUG_ON(ret != -EEXIST); 4385 spin_lock(&fs_info->balance_lock); 4386 update_balance_args(bctl); 4387 spin_unlock(&fs_info->balance_lock); 4388 } 4389 4390 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4391 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4392 describe_balance_start_or_resume(fs_info); 4393 mutex_unlock(&fs_info->balance_mutex); 4394 4395 ret = __btrfs_balance(fs_info); 4396 4397 mutex_lock(&fs_info->balance_mutex); 4398 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4399 btrfs_info(fs_info, "balance: paused"); 4400 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4401 } 4402 /* 4403 * Balance can be canceled by: 4404 * 4405 * - Regular cancel request 4406 * Then ret == -ECANCELED and balance_cancel_req > 0 4407 * 4408 * - Fatal signal to "btrfs" process 4409 * Either the signal caught by wait_reserve_ticket() and callers 4410 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4411 * got -ECANCELED. 4412 * Either way, in this case balance_cancel_req = 0, and 4413 * ret == -EINTR or ret == -ECANCELED. 4414 * 4415 * So here we only check the return value to catch canceled balance. 4416 */ 4417 else if (ret == -ECANCELED || ret == -EINTR) 4418 btrfs_info(fs_info, "balance: canceled"); 4419 else 4420 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4421 4422 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4423 4424 if (bargs) { 4425 memset(bargs, 0, sizeof(*bargs)); 4426 btrfs_update_ioctl_balance_args(fs_info, bargs); 4427 } 4428 4429 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4430 balance_need_close(fs_info)) { 4431 reset_balance_state(fs_info); 4432 btrfs_exclop_finish(fs_info); 4433 } 4434 4435 wake_up(&fs_info->balance_wait_q); 4436 4437 return ret; 4438 out: 4439 if (bctl->flags & BTRFS_BALANCE_RESUME) 4440 reset_balance_state(fs_info); 4441 else 4442 kfree(bctl); 4443 btrfs_exclop_finish(fs_info); 4444 4445 return ret; 4446 } 4447 4448 static int balance_kthread(void *data) 4449 { 4450 struct btrfs_fs_info *fs_info = data; 4451 int ret = 0; 4452 4453 sb_start_write(fs_info->sb); 4454 mutex_lock(&fs_info->balance_mutex); 4455 if (fs_info->balance_ctl) 4456 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4457 mutex_unlock(&fs_info->balance_mutex); 4458 sb_end_write(fs_info->sb); 4459 4460 return ret; 4461 } 4462 4463 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4464 { 4465 struct task_struct *tsk; 4466 4467 mutex_lock(&fs_info->balance_mutex); 4468 if (!fs_info->balance_ctl) { 4469 mutex_unlock(&fs_info->balance_mutex); 4470 return 0; 4471 } 4472 mutex_unlock(&fs_info->balance_mutex); 4473 4474 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4475 btrfs_info(fs_info, "balance: resume skipped"); 4476 return 0; 4477 } 4478 4479 spin_lock(&fs_info->super_lock); 4480 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4481 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4482 spin_unlock(&fs_info->super_lock); 4483 /* 4484 * A ro->rw remount sequence should continue with the paused balance 4485 * regardless of who pauses it, system or the user as of now, so set 4486 * the resume flag. 4487 */ 4488 spin_lock(&fs_info->balance_lock); 4489 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4490 spin_unlock(&fs_info->balance_lock); 4491 4492 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4493 return PTR_ERR_OR_ZERO(tsk); 4494 } 4495 4496 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4497 { 4498 struct btrfs_balance_control *bctl; 4499 struct btrfs_balance_item *item; 4500 struct btrfs_disk_balance_args disk_bargs; 4501 struct btrfs_path *path; 4502 struct extent_buffer *leaf; 4503 struct btrfs_key key; 4504 int ret; 4505 4506 path = btrfs_alloc_path(); 4507 if (!path) 4508 return -ENOMEM; 4509 4510 key.objectid = BTRFS_BALANCE_OBJECTID; 4511 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4512 key.offset = 0; 4513 4514 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4515 if (ret < 0) 4516 goto out; 4517 if (ret > 0) { /* ret = -ENOENT; */ 4518 ret = 0; 4519 goto out; 4520 } 4521 4522 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4523 if (!bctl) { 4524 ret = -ENOMEM; 4525 goto out; 4526 } 4527 4528 leaf = path->nodes[0]; 4529 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4530 4531 bctl->flags = btrfs_balance_flags(leaf, item); 4532 bctl->flags |= BTRFS_BALANCE_RESUME; 4533 4534 btrfs_balance_data(leaf, item, &disk_bargs); 4535 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4536 btrfs_balance_meta(leaf, item, &disk_bargs); 4537 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4538 btrfs_balance_sys(leaf, item, &disk_bargs); 4539 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4540 4541 /* 4542 * This should never happen, as the paused balance state is recovered 4543 * during mount without any chance of other exclusive ops to collide. 4544 * 4545 * This gives the exclusive op status to balance and keeps in paused 4546 * state until user intervention (cancel or umount). If the ownership 4547 * cannot be assigned, show a message but do not fail. The balance 4548 * is in a paused state and must have fs_info::balance_ctl properly 4549 * set up. 4550 */ 4551 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4552 btrfs_warn(fs_info, 4553 "balance: cannot set exclusive op status, resume manually"); 4554 4555 btrfs_release_path(path); 4556 4557 mutex_lock(&fs_info->balance_mutex); 4558 BUG_ON(fs_info->balance_ctl); 4559 spin_lock(&fs_info->balance_lock); 4560 fs_info->balance_ctl = bctl; 4561 spin_unlock(&fs_info->balance_lock); 4562 mutex_unlock(&fs_info->balance_mutex); 4563 out: 4564 btrfs_free_path(path); 4565 return ret; 4566 } 4567 4568 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4569 { 4570 int ret = 0; 4571 4572 mutex_lock(&fs_info->balance_mutex); 4573 if (!fs_info->balance_ctl) { 4574 mutex_unlock(&fs_info->balance_mutex); 4575 return -ENOTCONN; 4576 } 4577 4578 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4579 atomic_inc(&fs_info->balance_pause_req); 4580 mutex_unlock(&fs_info->balance_mutex); 4581 4582 wait_event(fs_info->balance_wait_q, 4583 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4584 4585 mutex_lock(&fs_info->balance_mutex); 4586 /* we are good with balance_ctl ripped off from under us */ 4587 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4588 atomic_dec(&fs_info->balance_pause_req); 4589 } else { 4590 ret = -ENOTCONN; 4591 } 4592 4593 mutex_unlock(&fs_info->balance_mutex); 4594 return ret; 4595 } 4596 4597 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4598 { 4599 mutex_lock(&fs_info->balance_mutex); 4600 if (!fs_info->balance_ctl) { 4601 mutex_unlock(&fs_info->balance_mutex); 4602 return -ENOTCONN; 4603 } 4604 4605 /* 4606 * A paused balance with the item stored on disk can be resumed at 4607 * mount time if the mount is read-write. Otherwise it's still paused 4608 * and we must not allow cancelling as it deletes the item. 4609 */ 4610 if (sb_rdonly(fs_info->sb)) { 4611 mutex_unlock(&fs_info->balance_mutex); 4612 return -EROFS; 4613 } 4614 4615 atomic_inc(&fs_info->balance_cancel_req); 4616 /* 4617 * if we are running just wait and return, balance item is 4618 * deleted in btrfs_balance in this case 4619 */ 4620 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4621 mutex_unlock(&fs_info->balance_mutex); 4622 wait_event(fs_info->balance_wait_q, 4623 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4624 mutex_lock(&fs_info->balance_mutex); 4625 } else { 4626 mutex_unlock(&fs_info->balance_mutex); 4627 /* 4628 * Lock released to allow other waiters to continue, we'll 4629 * reexamine the status again. 4630 */ 4631 mutex_lock(&fs_info->balance_mutex); 4632 4633 if (fs_info->balance_ctl) { 4634 reset_balance_state(fs_info); 4635 btrfs_exclop_finish(fs_info); 4636 btrfs_info(fs_info, "balance: canceled"); 4637 } 4638 } 4639 4640 BUG_ON(fs_info->balance_ctl || 4641 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4642 atomic_dec(&fs_info->balance_cancel_req); 4643 mutex_unlock(&fs_info->balance_mutex); 4644 return 0; 4645 } 4646 4647 int btrfs_uuid_scan_kthread(void *data) 4648 { 4649 struct btrfs_fs_info *fs_info = data; 4650 struct btrfs_root *root = fs_info->tree_root; 4651 struct btrfs_key key; 4652 struct btrfs_path *path = NULL; 4653 int ret = 0; 4654 struct extent_buffer *eb; 4655 int slot; 4656 struct btrfs_root_item root_item; 4657 u32 item_size; 4658 struct btrfs_trans_handle *trans = NULL; 4659 bool closing = false; 4660 4661 path = btrfs_alloc_path(); 4662 if (!path) { 4663 ret = -ENOMEM; 4664 goto out; 4665 } 4666 4667 key.objectid = 0; 4668 key.type = BTRFS_ROOT_ITEM_KEY; 4669 key.offset = 0; 4670 4671 while (1) { 4672 if (btrfs_fs_closing(fs_info)) { 4673 closing = true; 4674 break; 4675 } 4676 ret = btrfs_search_forward(root, &key, path, 4677 BTRFS_OLDEST_GENERATION); 4678 if (ret) { 4679 if (ret > 0) 4680 ret = 0; 4681 break; 4682 } 4683 4684 if (key.type != BTRFS_ROOT_ITEM_KEY || 4685 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4686 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4687 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4688 goto skip; 4689 4690 eb = path->nodes[0]; 4691 slot = path->slots[0]; 4692 item_size = btrfs_item_size(eb, slot); 4693 if (item_size < sizeof(root_item)) 4694 goto skip; 4695 4696 read_extent_buffer(eb, &root_item, 4697 btrfs_item_ptr_offset(eb, slot), 4698 (int)sizeof(root_item)); 4699 if (btrfs_root_refs(&root_item) == 0) 4700 goto skip; 4701 4702 if (!btrfs_is_empty_uuid(root_item.uuid) || 4703 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4704 if (trans) 4705 goto update_tree; 4706 4707 btrfs_release_path(path); 4708 /* 4709 * 1 - subvol uuid item 4710 * 1 - received_subvol uuid item 4711 */ 4712 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4713 if (IS_ERR(trans)) { 4714 ret = PTR_ERR(trans); 4715 break; 4716 } 4717 continue; 4718 } else { 4719 goto skip; 4720 } 4721 update_tree: 4722 btrfs_release_path(path); 4723 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4724 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4725 BTRFS_UUID_KEY_SUBVOL, 4726 key.objectid); 4727 if (ret < 0) { 4728 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4729 ret); 4730 break; 4731 } 4732 } 4733 4734 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4735 ret = btrfs_uuid_tree_add(trans, 4736 root_item.received_uuid, 4737 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4738 key.objectid); 4739 if (ret < 0) { 4740 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4741 ret); 4742 break; 4743 } 4744 } 4745 4746 skip: 4747 btrfs_release_path(path); 4748 if (trans) { 4749 ret = btrfs_end_transaction(trans); 4750 trans = NULL; 4751 if (ret) 4752 break; 4753 } 4754 4755 if (key.offset < (u64)-1) { 4756 key.offset++; 4757 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4758 key.offset = 0; 4759 key.type = BTRFS_ROOT_ITEM_KEY; 4760 } else if (key.objectid < (u64)-1) { 4761 key.offset = 0; 4762 key.type = BTRFS_ROOT_ITEM_KEY; 4763 key.objectid++; 4764 } else { 4765 break; 4766 } 4767 cond_resched(); 4768 } 4769 4770 out: 4771 btrfs_free_path(path); 4772 if (trans && !IS_ERR(trans)) 4773 btrfs_end_transaction(trans); 4774 if (ret) 4775 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4776 else if (!closing) 4777 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4778 up(&fs_info->uuid_tree_rescan_sem); 4779 return 0; 4780 } 4781 4782 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4783 { 4784 struct btrfs_trans_handle *trans; 4785 struct btrfs_root *tree_root = fs_info->tree_root; 4786 struct btrfs_root *uuid_root; 4787 struct task_struct *task; 4788 int ret; 4789 4790 /* 4791 * 1 - root node 4792 * 1 - root item 4793 */ 4794 trans = btrfs_start_transaction(tree_root, 2); 4795 if (IS_ERR(trans)) 4796 return PTR_ERR(trans); 4797 4798 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4799 if (IS_ERR(uuid_root)) { 4800 ret = PTR_ERR(uuid_root); 4801 btrfs_abort_transaction(trans, ret); 4802 btrfs_end_transaction(trans); 4803 return ret; 4804 } 4805 4806 fs_info->uuid_root = uuid_root; 4807 4808 ret = btrfs_commit_transaction(trans); 4809 if (ret) 4810 return ret; 4811 4812 down(&fs_info->uuid_tree_rescan_sem); 4813 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4814 if (IS_ERR(task)) { 4815 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4816 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4817 up(&fs_info->uuid_tree_rescan_sem); 4818 return PTR_ERR(task); 4819 } 4820 4821 return 0; 4822 } 4823 4824 /* 4825 * shrinking a device means finding all of the device extents past 4826 * the new size, and then following the back refs to the chunks. 4827 * The chunk relocation code actually frees the device extent 4828 */ 4829 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4830 { 4831 struct btrfs_fs_info *fs_info = device->fs_info; 4832 struct btrfs_root *root = fs_info->dev_root; 4833 struct btrfs_trans_handle *trans; 4834 struct btrfs_dev_extent *dev_extent = NULL; 4835 struct btrfs_path *path; 4836 u64 length; 4837 u64 chunk_offset; 4838 int ret; 4839 int slot; 4840 int failed = 0; 4841 bool retried = false; 4842 struct extent_buffer *l; 4843 struct btrfs_key key; 4844 struct btrfs_super_block *super_copy = fs_info->super_copy; 4845 u64 old_total = btrfs_super_total_bytes(super_copy); 4846 u64 old_size = btrfs_device_get_total_bytes(device); 4847 u64 diff; 4848 u64 start; 4849 4850 new_size = round_down(new_size, fs_info->sectorsize); 4851 start = new_size; 4852 diff = round_down(old_size - new_size, fs_info->sectorsize); 4853 4854 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4855 return -EINVAL; 4856 4857 path = btrfs_alloc_path(); 4858 if (!path) 4859 return -ENOMEM; 4860 4861 path->reada = READA_BACK; 4862 4863 trans = btrfs_start_transaction(root, 0); 4864 if (IS_ERR(trans)) { 4865 btrfs_free_path(path); 4866 return PTR_ERR(trans); 4867 } 4868 4869 mutex_lock(&fs_info->chunk_mutex); 4870 4871 btrfs_device_set_total_bytes(device, new_size); 4872 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4873 device->fs_devices->total_rw_bytes -= diff; 4874 atomic64_sub(diff, &fs_info->free_chunk_space); 4875 } 4876 4877 /* 4878 * Once the device's size has been set to the new size, ensure all 4879 * in-memory chunks are synced to disk so that the loop below sees them 4880 * and relocates them accordingly. 4881 */ 4882 if (contains_pending_extent(device, &start, diff)) { 4883 mutex_unlock(&fs_info->chunk_mutex); 4884 ret = btrfs_commit_transaction(trans); 4885 if (ret) 4886 goto done; 4887 } else { 4888 mutex_unlock(&fs_info->chunk_mutex); 4889 btrfs_end_transaction(trans); 4890 } 4891 4892 again: 4893 key.objectid = device->devid; 4894 key.offset = (u64)-1; 4895 key.type = BTRFS_DEV_EXTENT_KEY; 4896 4897 do { 4898 mutex_lock(&fs_info->reclaim_bgs_lock); 4899 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4900 if (ret < 0) { 4901 mutex_unlock(&fs_info->reclaim_bgs_lock); 4902 goto done; 4903 } 4904 4905 ret = btrfs_previous_item(root, path, 0, key.type); 4906 if (ret) { 4907 mutex_unlock(&fs_info->reclaim_bgs_lock); 4908 if (ret < 0) 4909 goto done; 4910 ret = 0; 4911 btrfs_release_path(path); 4912 break; 4913 } 4914 4915 l = path->nodes[0]; 4916 slot = path->slots[0]; 4917 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4918 4919 if (key.objectid != device->devid) { 4920 mutex_unlock(&fs_info->reclaim_bgs_lock); 4921 btrfs_release_path(path); 4922 break; 4923 } 4924 4925 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4926 length = btrfs_dev_extent_length(l, dev_extent); 4927 4928 if (key.offset + length <= new_size) { 4929 mutex_unlock(&fs_info->reclaim_bgs_lock); 4930 btrfs_release_path(path); 4931 break; 4932 } 4933 4934 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4935 btrfs_release_path(path); 4936 4937 /* 4938 * We may be relocating the only data chunk we have, 4939 * which could potentially end up with losing data's 4940 * raid profile, so lets allocate an empty one in 4941 * advance. 4942 */ 4943 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4944 if (ret < 0) { 4945 mutex_unlock(&fs_info->reclaim_bgs_lock); 4946 goto done; 4947 } 4948 4949 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4950 mutex_unlock(&fs_info->reclaim_bgs_lock); 4951 if (ret == -ENOSPC) { 4952 failed++; 4953 } else if (ret) { 4954 if (ret == -ETXTBSY) { 4955 btrfs_warn(fs_info, 4956 "could not shrink block group %llu due to active swapfile", 4957 chunk_offset); 4958 } 4959 goto done; 4960 } 4961 } while (key.offset-- > 0); 4962 4963 if (failed && !retried) { 4964 failed = 0; 4965 retried = true; 4966 goto again; 4967 } else if (failed && retried) { 4968 ret = -ENOSPC; 4969 goto done; 4970 } 4971 4972 /* Shrinking succeeded, else we would be at "done". */ 4973 trans = btrfs_start_transaction(root, 0); 4974 if (IS_ERR(trans)) { 4975 ret = PTR_ERR(trans); 4976 goto done; 4977 } 4978 4979 mutex_lock(&fs_info->chunk_mutex); 4980 /* Clear all state bits beyond the shrunk device size */ 4981 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4982 CHUNK_STATE_MASK); 4983 4984 btrfs_device_set_disk_total_bytes(device, new_size); 4985 if (list_empty(&device->post_commit_list)) 4986 list_add_tail(&device->post_commit_list, 4987 &trans->transaction->dev_update_list); 4988 4989 WARN_ON(diff > old_total); 4990 btrfs_set_super_total_bytes(super_copy, 4991 round_down(old_total - diff, fs_info->sectorsize)); 4992 mutex_unlock(&fs_info->chunk_mutex); 4993 4994 btrfs_reserve_chunk_metadata(trans, false); 4995 /* Now btrfs_update_device() will change the on-disk size. */ 4996 ret = btrfs_update_device(trans, device); 4997 btrfs_trans_release_chunk_metadata(trans); 4998 if (ret < 0) { 4999 btrfs_abort_transaction(trans, ret); 5000 btrfs_end_transaction(trans); 5001 } else { 5002 ret = btrfs_commit_transaction(trans); 5003 } 5004 done: 5005 btrfs_free_path(path); 5006 if (ret) { 5007 mutex_lock(&fs_info->chunk_mutex); 5008 btrfs_device_set_total_bytes(device, old_size); 5009 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 5010 device->fs_devices->total_rw_bytes += diff; 5011 atomic64_add(diff, &fs_info->free_chunk_space); 5012 mutex_unlock(&fs_info->chunk_mutex); 5013 } 5014 return ret; 5015 } 5016 5017 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 5018 struct btrfs_key *key, 5019 struct btrfs_chunk *chunk, int item_size) 5020 { 5021 struct btrfs_super_block *super_copy = fs_info->super_copy; 5022 struct btrfs_disk_key disk_key; 5023 u32 array_size; 5024 u8 *ptr; 5025 5026 lockdep_assert_held(&fs_info->chunk_mutex); 5027 5028 array_size = btrfs_super_sys_array_size(super_copy); 5029 if (array_size + item_size + sizeof(disk_key) 5030 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5031 return -EFBIG; 5032 5033 ptr = super_copy->sys_chunk_array + array_size; 5034 btrfs_cpu_key_to_disk(&disk_key, key); 5035 memcpy(ptr, &disk_key, sizeof(disk_key)); 5036 ptr += sizeof(disk_key); 5037 memcpy(ptr, chunk, item_size); 5038 item_size += sizeof(disk_key); 5039 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5040 5041 return 0; 5042 } 5043 5044 /* 5045 * sort the devices in descending order by max_avail, total_avail 5046 */ 5047 static int btrfs_cmp_device_info(const void *a, const void *b) 5048 { 5049 const struct btrfs_device_info *di_a = a; 5050 const struct btrfs_device_info *di_b = b; 5051 5052 if (di_a->max_avail > di_b->max_avail) 5053 return -1; 5054 if (di_a->max_avail < di_b->max_avail) 5055 return 1; 5056 if (di_a->total_avail > di_b->total_avail) 5057 return -1; 5058 if (di_a->total_avail < di_b->total_avail) 5059 return 1; 5060 return 0; 5061 } 5062 5063 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5064 { 5065 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5066 return; 5067 5068 btrfs_set_fs_incompat(info, RAID56); 5069 } 5070 5071 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5072 { 5073 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5074 return; 5075 5076 btrfs_set_fs_incompat(info, RAID1C34); 5077 } 5078 5079 /* 5080 * Structure used internally for btrfs_create_chunk() function. 5081 * Wraps needed parameters. 5082 */ 5083 struct alloc_chunk_ctl { 5084 u64 start; 5085 u64 type; 5086 /* Total number of stripes to allocate */ 5087 int num_stripes; 5088 /* sub_stripes info for map */ 5089 int sub_stripes; 5090 /* Stripes per device */ 5091 int dev_stripes; 5092 /* Maximum number of devices to use */ 5093 int devs_max; 5094 /* Minimum number of devices to use */ 5095 int devs_min; 5096 /* ndevs has to be a multiple of this */ 5097 int devs_increment; 5098 /* Number of copies */ 5099 int ncopies; 5100 /* Number of stripes worth of bytes to store parity information */ 5101 int nparity; 5102 u64 max_stripe_size; 5103 u64 max_chunk_size; 5104 u64 dev_extent_min; 5105 u64 stripe_size; 5106 u64 chunk_size; 5107 int ndevs; 5108 }; 5109 5110 static void init_alloc_chunk_ctl_policy_regular( 5111 struct btrfs_fs_devices *fs_devices, 5112 struct alloc_chunk_ctl *ctl) 5113 { 5114 struct btrfs_space_info *space_info; 5115 5116 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); 5117 ASSERT(space_info); 5118 5119 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); 5120 ctl->max_stripe_size = ctl->max_chunk_size; 5121 5122 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) 5123 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); 5124 5125 /* We don't want a chunk larger than 10% of writable space */ 5126 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), 5127 ctl->max_chunk_size); 5128 ctl->dev_extent_min = ctl->dev_stripes << BTRFS_STRIPE_LEN_SHIFT; 5129 } 5130 5131 static void init_alloc_chunk_ctl_policy_zoned( 5132 struct btrfs_fs_devices *fs_devices, 5133 struct alloc_chunk_ctl *ctl) 5134 { 5135 u64 zone_size = fs_devices->fs_info->zone_size; 5136 u64 limit; 5137 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5138 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5139 u64 min_chunk_size = min_data_stripes * zone_size; 5140 u64 type = ctl->type; 5141 5142 ctl->max_stripe_size = zone_size; 5143 if (type & BTRFS_BLOCK_GROUP_DATA) { 5144 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5145 zone_size); 5146 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5147 ctl->max_chunk_size = ctl->max_stripe_size; 5148 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5149 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5150 ctl->devs_max = min_t(int, ctl->devs_max, 5151 BTRFS_MAX_DEVS_SYS_CHUNK); 5152 } else { 5153 BUG(); 5154 } 5155 5156 /* We don't want a chunk larger than 10% of writable space */ 5157 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), 5158 zone_size), 5159 min_chunk_size); 5160 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5161 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5162 } 5163 5164 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5165 struct alloc_chunk_ctl *ctl) 5166 { 5167 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5168 5169 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5170 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5171 ctl->devs_max = btrfs_raid_array[index].devs_max; 5172 if (!ctl->devs_max) 5173 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5174 ctl->devs_min = btrfs_raid_array[index].devs_min; 5175 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5176 ctl->ncopies = btrfs_raid_array[index].ncopies; 5177 ctl->nparity = btrfs_raid_array[index].nparity; 5178 ctl->ndevs = 0; 5179 5180 switch (fs_devices->chunk_alloc_policy) { 5181 case BTRFS_CHUNK_ALLOC_REGULAR: 5182 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5183 break; 5184 case BTRFS_CHUNK_ALLOC_ZONED: 5185 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5186 break; 5187 default: 5188 BUG(); 5189 } 5190 } 5191 5192 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5193 struct alloc_chunk_ctl *ctl, 5194 struct btrfs_device_info *devices_info) 5195 { 5196 struct btrfs_fs_info *info = fs_devices->fs_info; 5197 struct btrfs_device *device; 5198 u64 total_avail; 5199 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5200 int ret; 5201 int ndevs = 0; 5202 u64 max_avail; 5203 u64 dev_offset; 5204 5205 /* 5206 * in the first pass through the devices list, we gather information 5207 * about the available holes on each device. 5208 */ 5209 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5210 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5211 WARN(1, KERN_ERR 5212 "BTRFS: read-only device in alloc_list\n"); 5213 continue; 5214 } 5215 5216 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5217 &device->dev_state) || 5218 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5219 continue; 5220 5221 if (device->total_bytes > device->bytes_used) 5222 total_avail = device->total_bytes - device->bytes_used; 5223 else 5224 total_avail = 0; 5225 5226 /* If there is no space on this device, skip it. */ 5227 if (total_avail < ctl->dev_extent_min) 5228 continue; 5229 5230 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5231 &max_avail); 5232 if (ret && ret != -ENOSPC) 5233 return ret; 5234 5235 if (ret == 0) 5236 max_avail = dev_extent_want; 5237 5238 if (max_avail < ctl->dev_extent_min) { 5239 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5240 btrfs_debug(info, 5241 "%s: devid %llu has no free space, have=%llu want=%llu", 5242 __func__, device->devid, max_avail, 5243 ctl->dev_extent_min); 5244 continue; 5245 } 5246 5247 if (ndevs == fs_devices->rw_devices) { 5248 WARN(1, "%s: found more than %llu devices\n", 5249 __func__, fs_devices->rw_devices); 5250 break; 5251 } 5252 devices_info[ndevs].dev_offset = dev_offset; 5253 devices_info[ndevs].max_avail = max_avail; 5254 devices_info[ndevs].total_avail = total_avail; 5255 devices_info[ndevs].dev = device; 5256 ++ndevs; 5257 } 5258 ctl->ndevs = ndevs; 5259 5260 /* 5261 * now sort the devices by hole size / available space 5262 */ 5263 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5264 btrfs_cmp_device_info, NULL); 5265 5266 return 0; 5267 } 5268 5269 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5270 struct btrfs_device_info *devices_info) 5271 { 5272 /* Number of stripes that count for block group size */ 5273 int data_stripes; 5274 5275 /* 5276 * The primary goal is to maximize the number of stripes, so use as 5277 * many devices as possible, even if the stripes are not maximum sized. 5278 * 5279 * The DUP profile stores more than one stripe per device, the 5280 * max_avail is the total size so we have to adjust. 5281 */ 5282 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5283 ctl->dev_stripes); 5284 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5285 5286 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5287 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5288 5289 /* 5290 * Use the number of data stripes to figure out how big this chunk is 5291 * really going to be in terms of logical address space, and compare 5292 * that answer with the max chunk size. If it's higher, we try to 5293 * reduce stripe_size. 5294 */ 5295 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5296 /* 5297 * Reduce stripe_size, round it up to a 16MB boundary again and 5298 * then use it, unless it ends up being even bigger than the 5299 * previous value we had already. 5300 */ 5301 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5302 data_stripes), SZ_16M), 5303 ctl->stripe_size); 5304 } 5305 5306 /* Stripe size should not go beyond 1G. */ 5307 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); 5308 5309 /* Align to BTRFS_STRIPE_LEN */ 5310 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5311 ctl->chunk_size = ctl->stripe_size * data_stripes; 5312 5313 return 0; 5314 } 5315 5316 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5317 struct btrfs_device_info *devices_info) 5318 { 5319 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5320 /* Number of stripes that count for block group size */ 5321 int data_stripes; 5322 5323 /* 5324 * It should hold because: 5325 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5326 */ 5327 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5328 5329 ctl->stripe_size = zone_size; 5330 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5331 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5332 5333 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5334 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5335 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5336 ctl->stripe_size) + ctl->nparity, 5337 ctl->dev_stripes); 5338 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5339 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5340 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5341 } 5342 5343 ctl->chunk_size = ctl->stripe_size * data_stripes; 5344 5345 return 0; 5346 } 5347 5348 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5349 struct alloc_chunk_ctl *ctl, 5350 struct btrfs_device_info *devices_info) 5351 { 5352 struct btrfs_fs_info *info = fs_devices->fs_info; 5353 5354 /* 5355 * Round down to number of usable stripes, devs_increment can be any 5356 * number so we can't use round_down() that requires power of 2, while 5357 * rounddown is safe. 5358 */ 5359 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5360 5361 if (ctl->ndevs < ctl->devs_min) { 5362 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5363 btrfs_debug(info, 5364 "%s: not enough devices with free space: have=%d minimum required=%d", 5365 __func__, ctl->ndevs, ctl->devs_min); 5366 } 5367 return -ENOSPC; 5368 } 5369 5370 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5371 5372 switch (fs_devices->chunk_alloc_policy) { 5373 case BTRFS_CHUNK_ALLOC_REGULAR: 5374 return decide_stripe_size_regular(ctl, devices_info); 5375 case BTRFS_CHUNK_ALLOC_ZONED: 5376 return decide_stripe_size_zoned(ctl, devices_info); 5377 default: 5378 BUG(); 5379 } 5380 } 5381 5382 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5383 struct alloc_chunk_ctl *ctl, 5384 struct btrfs_device_info *devices_info) 5385 { 5386 struct btrfs_fs_info *info = trans->fs_info; 5387 struct map_lookup *map = NULL; 5388 struct extent_map_tree *em_tree; 5389 struct btrfs_block_group *block_group; 5390 struct extent_map *em; 5391 u64 start = ctl->start; 5392 u64 type = ctl->type; 5393 int ret; 5394 int i; 5395 int j; 5396 5397 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5398 if (!map) 5399 return ERR_PTR(-ENOMEM); 5400 map->num_stripes = ctl->num_stripes; 5401 5402 for (i = 0; i < ctl->ndevs; ++i) { 5403 for (j = 0; j < ctl->dev_stripes; ++j) { 5404 int s = i * ctl->dev_stripes + j; 5405 map->stripes[s].dev = devices_info[i].dev; 5406 map->stripes[s].physical = devices_info[i].dev_offset + 5407 j * ctl->stripe_size; 5408 } 5409 } 5410 map->io_align = BTRFS_STRIPE_LEN; 5411 map->io_width = BTRFS_STRIPE_LEN; 5412 map->type = type; 5413 map->sub_stripes = ctl->sub_stripes; 5414 5415 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5416 5417 em = alloc_extent_map(); 5418 if (!em) { 5419 kfree(map); 5420 return ERR_PTR(-ENOMEM); 5421 } 5422 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5423 em->map_lookup = map; 5424 em->start = start; 5425 em->len = ctl->chunk_size; 5426 em->block_start = 0; 5427 em->block_len = em->len; 5428 em->orig_block_len = ctl->stripe_size; 5429 5430 em_tree = &info->mapping_tree; 5431 write_lock(&em_tree->lock); 5432 ret = add_extent_mapping(em_tree, em, 0); 5433 if (ret) { 5434 write_unlock(&em_tree->lock); 5435 free_extent_map(em); 5436 return ERR_PTR(ret); 5437 } 5438 write_unlock(&em_tree->lock); 5439 5440 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5441 if (IS_ERR(block_group)) 5442 goto error_del_extent; 5443 5444 for (i = 0; i < map->num_stripes; i++) { 5445 struct btrfs_device *dev = map->stripes[i].dev; 5446 5447 btrfs_device_set_bytes_used(dev, 5448 dev->bytes_used + ctl->stripe_size); 5449 if (list_empty(&dev->post_commit_list)) 5450 list_add_tail(&dev->post_commit_list, 5451 &trans->transaction->dev_update_list); 5452 } 5453 5454 atomic64_sub(ctl->stripe_size * map->num_stripes, 5455 &info->free_chunk_space); 5456 5457 free_extent_map(em); 5458 check_raid56_incompat_flag(info, type); 5459 check_raid1c34_incompat_flag(info, type); 5460 5461 return block_group; 5462 5463 error_del_extent: 5464 write_lock(&em_tree->lock); 5465 remove_extent_mapping(em_tree, em); 5466 write_unlock(&em_tree->lock); 5467 5468 /* One for our allocation */ 5469 free_extent_map(em); 5470 /* One for the tree reference */ 5471 free_extent_map(em); 5472 5473 return block_group; 5474 } 5475 5476 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5477 u64 type) 5478 { 5479 struct btrfs_fs_info *info = trans->fs_info; 5480 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5481 struct btrfs_device_info *devices_info = NULL; 5482 struct alloc_chunk_ctl ctl; 5483 struct btrfs_block_group *block_group; 5484 int ret; 5485 5486 lockdep_assert_held(&info->chunk_mutex); 5487 5488 if (!alloc_profile_is_valid(type, 0)) { 5489 ASSERT(0); 5490 return ERR_PTR(-EINVAL); 5491 } 5492 5493 if (list_empty(&fs_devices->alloc_list)) { 5494 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5495 btrfs_debug(info, "%s: no writable device", __func__); 5496 return ERR_PTR(-ENOSPC); 5497 } 5498 5499 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5500 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5501 ASSERT(0); 5502 return ERR_PTR(-EINVAL); 5503 } 5504 5505 ctl.start = find_next_chunk(info); 5506 ctl.type = type; 5507 init_alloc_chunk_ctl(fs_devices, &ctl); 5508 5509 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5510 GFP_NOFS); 5511 if (!devices_info) 5512 return ERR_PTR(-ENOMEM); 5513 5514 ret = gather_device_info(fs_devices, &ctl, devices_info); 5515 if (ret < 0) { 5516 block_group = ERR_PTR(ret); 5517 goto out; 5518 } 5519 5520 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5521 if (ret < 0) { 5522 block_group = ERR_PTR(ret); 5523 goto out; 5524 } 5525 5526 block_group = create_chunk(trans, &ctl, devices_info); 5527 5528 out: 5529 kfree(devices_info); 5530 return block_group; 5531 } 5532 5533 /* 5534 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5535 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5536 * chunks. 5537 * 5538 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5539 * phases. 5540 */ 5541 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5542 struct btrfs_block_group *bg) 5543 { 5544 struct btrfs_fs_info *fs_info = trans->fs_info; 5545 struct btrfs_root *chunk_root = fs_info->chunk_root; 5546 struct btrfs_key key; 5547 struct btrfs_chunk *chunk; 5548 struct btrfs_stripe *stripe; 5549 struct extent_map *em; 5550 struct map_lookup *map; 5551 size_t item_size; 5552 int i; 5553 int ret; 5554 5555 /* 5556 * We take the chunk_mutex for 2 reasons: 5557 * 5558 * 1) Updates and insertions in the chunk btree must be done while holding 5559 * the chunk_mutex, as well as updating the system chunk array in the 5560 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5561 * details; 5562 * 5563 * 2) To prevent races with the final phase of a device replace operation 5564 * that replaces the device object associated with the map's stripes, 5565 * because the device object's id can change at any time during that 5566 * final phase of the device replace operation 5567 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5568 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5569 * which would cause a failure when updating the device item, which does 5570 * not exists, or persisting a stripe of the chunk item with such ID. 5571 * Here we can't use the device_list_mutex because our caller already 5572 * has locked the chunk_mutex, and the final phase of device replace 5573 * acquires both mutexes - first the device_list_mutex and then the 5574 * chunk_mutex. Using any of those two mutexes protects us from a 5575 * concurrent device replace. 5576 */ 5577 lockdep_assert_held(&fs_info->chunk_mutex); 5578 5579 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5580 if (IS_ERR(em)) { 5581 ret = PTR_ERR(em); 5582 btrfs_abort_transaction(trans, ret); 5583 return ret; 5584 } 5585 5586 map = em->map_lookup; 5587 item_size = btrfs_chunk_item_size(map->num_stripes); 5588 5589 chunk = kzalloc(item_size, GFP_NOFS); 5590 if (!chunk) { 5591 ret = -ENOMEM; 5592 btrfs_abort_transaction(trans, ret); 5593 goto out; 5594 } 5595 5596 for (i = 0; i < map->num_stripes; i++) { 5597 struct btrfs_device *device = map->stripes[i].dev; 5598 5599 ret = btrfs_update_device(trans, device); 5600 if (ret) 5601 goto out; 5602 } 5603 5604 stripe = &chunk->stripe; 5605 for (i = 0; i < map->num_stripes; i++) { 5606 struct btrfs_device *device = map->stripes[i].dev; 5607 const u64 dev_offset = map->stripes[i].physical; 5608 5609 btrfs_set_stack_stripe_devid(stripe, device->devid); 5610 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5611 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5612 stripe++; 5613 } 5614 5615 btrfs_set_stack_chunk_length(chunk, bg->length); 5616 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5617 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN); 5618 btrfs_set_stack_chunk_type(chunk, map->type); 5619 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5620 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN); 5621 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN); 5622 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5623 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5624 5625 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5626 key.type = BTRFS_CHUNK_ITEM_KEY; 5627 key.offset = bg->start; 5628 5629 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5630 if (ret) 5631 goto out; 5632 5633 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); 5634 5635 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5636 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5637 if (ret) 5638 goto out; 5639 } 5640 5641 out: 5642 kfree(chunk); 5643 free_extent_map(em); 5644 return ret; 5645 } 5646 5647 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5648 { 5649 struct btrfs_fs_info *fs_info = trans->fs_info; 5650 u64 alloc_profile; 5651 struct btrfs_block_group *meta_bg; 5652 struct btrfs_block_group *sys_bg; 5653 5654 /* 5655 * When adding a new device for sprouting, the seed device is read-only 5656 * so we must first allocate a metadata and a system chunk. But before 5657 * adding the block group items to the extent, device and chunk btrees, 5658 * we must first: 5659 * 5660 * 1) Create both chunks without doing any changes to the btrees, as 5661 * otherwise we would get -ENOSPC since the block groups from the 5662 * seed device are read-only; 5663 * 5664 * 2) Add the device item for the new sprout device - finishing the setup 5665 * of a new block group requires updating the device item in the chunk 5666 * btree, so it must exist when we attempt to do it. The previous step 5667 * ensures this does not fail with -ENOSPC. 5668 * 5669 * After that we can add the block group items to their btrees: 5670 * update existing device item in the chunk btree, add a new block group 5671 * item to the extent btree, add a new chunk item to the chunk btree and 5672 * finally add the new device extent items to the devices btree. 5673 */ 5674 5675 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5676 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5677 if (IS_ERR(meta_bg)) 5678 return PTR_ERR(meta_bg); 5679 5680 alloc_profile = btrfs_system_alloc_profile(fs_info); 5681 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5682 if (IS_ERR(sys_bg)) 5683 return PTR_ERR(sys_bg); 5684 5685 return 0; 5686 } 5687 5688 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5689 { 5690 const int index = btrfs_bg_flags_to_raid_index(map->type); 5691 5692 return btrfs_raid_array[index].tolerated_failures; 5693 } 5694 5695 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5696 { 5697 struct extent_map *em; 5698 struct map_lookup *map; 5699 int miss_ndevs = 0; 5700 int i; 5701 bool ret = true; 5702 5703 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5704 if (IS_ERR(em)) 5705 return false; 5706 5707 map = em->map_lookup; 5708 for (i = 0; i < map->num_stripes; i++) { 5709 if (test_bit(BTRFS_DEV_STATE_MISSING, 5710 &map->stripes[i].dev->dev_state)) { 5711 miss_ndevs++; 5712 continue; 5713 } 5714 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5715 &map->stripes[i].dev->dev_state)) { 5716 ret = false; 5717 goto end; 5718 } 5719 } 5720 5721 /* 5722 * If the number of missing devices is larger than max errors, we can 5723 * not write the data into that chunk successfully. 5724 */ 5725 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5726 ret = false; 5727 end: 5728 free_extent_map(em); 5729 return ret; 5730 } 5731 5732 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5733 { 5734 struct extent_map *em; 5735 5736 while (1) { 5737 write_lock(&tree->lock); 5738 em = lookup_extent_mapping(tree, 0, (u64)-1); 5739 if (em) 5740 remove_extent_mapping(tree, em); 5741 write_unlock(&tree->lock); 5742 if (!em) 5743 break; 5744 /* once for us */ 5745 free_extent_map(em); 5746 /* once for the tree */ 5747 free_extent_map(em); 5748 } 5749 } 5750 5751 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5752 { 5753 struct extent_map *em; 5754 struct map_lookup *map; 5755 enum btrfs_raid_types index; 5756 int ret = 1; 5757 5758 em = btrfs_get_chunk_map(fs_info, logical, len); 5759 if (IS_ERR(em)) 5760 /* 5761 * We could return errors for these cases, but that could get 5762 * ugly and we'd probably do the same thing which is just not do 5763 * anything else and exit, so return 1 so the callers don't try 5764 * to use other copies. 5765 */ 5766 return 1; 5767 5768 map = em->map_lookup; 5769 index = btrfs_bg_flags_to_raid_index(map->type); 5770 5771 /* Non-RAID56, use their ncopies from btrfs_raid_array. */ 5772 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5773 ret = btrfs_raid_array[index].ncopies; 5774 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5775 ret = 2; 5776 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5777 /* 5778 * There could be two corrupted data stripes, we need 5779 * to loop retry in order to rebuild the correct data. 5780 * 5781 * Fail a stripe at a time on every retry except the 5782 * stripe under reconstruction. 5783 */ 5784 ret = map->num_stripes; 5785 free_extent_map(em); 5786 5787 down_read(&fs_info->dev_replace.rwsem); 5788 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5789 fs_info->dev_replace.tgtdev) 5790 ret++; 5791 up_read(&fs_info->dev_replace.rwsem); 5792 5793 return ret; 5794 } 5795 5796 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5797 u64 logical) 5798 { 5799 struct extent_map *em; 5800 struct map_lookup *map; 5801 unsigned long len = fs_info->sectorsize; 5802 5803 if (!btrfs_fs_incompat(fs_info, RAID56)) 5804 return len; 5805 5806 em = btrfs_get_chunk_map(fs_info, logical, len); 5807 5808 if (!WARN_ON(IS_ERR(em))) { 5809 map = em->map_lookup; 5810 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5811 len = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT; 5812 free_extent_map(em); 5813 } 5814 return len; 5815 } 5816 5817 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5818 { 5819 struct extent_map *em; 5820 struct map_lookup *map; 5821 int ret = 0; 5822 5823 if (!btrfs_fs_incompat(fs_info, RAID56)) 5824 return 0; 5825 5826 em = btrfs_get_chunk_map(fs_info, logical, len); 5827 5828 if(!WARN_ON(IS_ERR(em))) { 5829 map = em->map_lookup; 5830 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5831 ret = 1; 5832 free_extent_map(em); 5833 } 5834 return ret; 5835 } 5836 5837 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5838 struct map_lookup *map, int first, 5839 int dev_replace_is_ongoing) 5840 { 5841 int i; 5842 int num_stripes; 5843 int preferred_mirror; 5844 int tolerance; 5845 struct btrfs_device *srcdev; 5846 5847 ASSERT((map->type & 5848 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5849 5850 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5851 num_stripes = map->sub_stripes; 5852 else 5853 num_stripes = map->num_stripes; 5854 5855 switch (fs_info->fs_devices->read_policy) { 5856 default: 5857 /* Shouldn't happen, just warn and use pid instead of failing */ 5858 btrfs_warn_rl(fs_info, 5859 "unknown read_policy type %u, reset to pid", 5860 fs_info->fs_devices->read_policy); 5861 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5862 fallthrough; 5863 case BTRFS_READ_POLICY_PID: 5864 preferred_mirror = first + (current->pid % num_stripes); 5865 break; 5866 } 5867 5868 if (dev_replace_is_ongoing && 5869 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5870 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5871 srcdev = fs_info->dev_replace.srcdev; 5872 else 5873 srcdev = NULL; 5874 5875 /* 5876 * try to avoid the drive that is the source drive for a 5877 * dev-replace procedure, only choose it if no other non-missing 5878 * mirror is available 5879 */ 5880 for (tolerance = 0; tolerance < 2; tolerance++) { 5881 if (map->stripes[preferred_mirror].dev->bdev && 5882 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5883 return preferred_mirror; 5884 for (i = first; i < first + num_stripes; i++) { 5885 if (map->stripes[i].dev->bdev && 5886 (tolerance || map->stripes[i].dev != srcdev)) 5887 return i; 5888 } 5889 } 5890 5891 /* we couldn't find one that doesn't fail. Just return something 5892 * and the io error handling code will clean up eventually 5893 */ 5894 return preferred_mirror; 5895 } 5896 5897 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5898 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5899 { 5900 int i; 5901 int again = 1; 5902 5903 while (again) { 5904 again = 0; 5905 for (i = 0; i < num_stripes - 1; i++) { 5906 /* Swap if parity is on a smaller index */ 5907 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5908 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5909 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5910 again = 1; 5911 } 5912 } 5913 } 5914 } 5915 5916 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5917 int total_stripes, 5918 int real_stripes) 5919 { 5920 struct btrfs_io_context *bioc = kzalloc( 5921 /* The size of btrfs_io_context */ 5922 sizeof(struct btrfs_io_context) + 5923 /* Plus the variable array for the stripes */ 5924 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5925 /* Plus the variable array for the tgt dev */ 5926 sizeof(int) * (real_stripes) + 5927 /* 5928 * Plus the raid_map, which includes both the tgt dev 5929 * and the stripes. 5930 */ 5931 sizeof(u64) * (total_stripes), 5932 GFP_NOFS); 5933 5934 if (!bioc) 5935 return NULL; 5936 5937 refcount_set(&bioc->refs, 1); 5938 5939 bioc->fs_info = fs_info; 5940 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5941 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5942 5943 return bioc; 5944 } 5945 5946 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5947 { 5948 WARN_ON(!refcount_read(&bioc->refs)); 5949 refcount_inc(&bioc->refs); 5950 } 5951 5952 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5953 { 5954 if (!bioc) 5955 return; 5956 if (refcount_dec_and_test(&bioc->refs)) 5957 kfree(bioc); 5958 } 5959 5960 /* 5961 * Please note that, discard won't be sent to target device of device 5962 * replace. 5963 */ 5964 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, 5965 u64 logical, u64 *length_ret, 5966 u32 *num_stripes) 5967 { 5968 struct extent_map *em; 5969 struct map_lookup *map; 5970 struct btrfs_discard_stripe *stripes; 5971 u64 length = *length_ret; 5972 u64 offset; 5973 u32 stripe_nr; 5974 u32 stripe_nr_end; 5975 u32 stripe_cnt; 5976 u64 stripe_end_offset; 5977 u64 stripe_offset; 5978 u32 stripe_index; 5979 u32 factor = 0; 5980 u32 sub_stripes = 0; 5981 u32 stripes_per_dev = 0; 5982 u32 remaining_stripes = 0; 5983 u32 last_stripe = 0; 5984 int ret; 5985 int i; 5986 5987 em = btrfs_get_chunk_map(fs_info, logical, length); 5988 if (IS_ERR(em)) 5989 return ERR_CAST(em); 5990 5991 map = em->map_lookup; 5992 5993 /* we don't discard raid56 yet */ 5994 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5995 ret = -EOPNOTSUPP; 5996 goto out_free_map; 5997 } 5998 5999 offset = logical - em->start; 6000 length = min_t(u64, em->start + em->len - logical, length); 6001 *length_ret = length; 6002 6003 /* 6004 * stripe_nr counts the total number of stripes we have to stride 6005 * to get to this block 6006 */ 6007 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 6008 6009 /* stripe_offset is the offset of this block in its stripe */ 6010 stripe_offset = offset - (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6011 6012 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >> 6013 BTRFS_STRIPE_LEN_SHIFT; 6014 stripe_cnt = stripe_nr_end - stripe_nr; 6015 stripe_end_offset = (stripe_nr_end << BTRFS_STRIPE_LEN_SHIFT) - 6016 (offset + length); 6017 /* 6018 * after this, stripe_nr is the number of stripes on this 6019 * device we have to walk to find the data, and stripe_index is 6020 * the number of our device in the stripe array 6021 */ 6022 *num_stripes = 1; 6023 stripe_index = 0; 6024 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6025 BTRFS_BLOCK_GROUP_RAID10)) { 6026 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 6027 sub_stripes = 1; 6028 else 6029 sub_stripes = map->sub_stripes; 6030 6031 factor = map->num_stripes / sub_stripes; 6032 *num_stripes = min_t(u64, map->num_stripes, 6033 sub_stripes * stripe_cnt); 6034 stripe_index = stripe_nr % factor; 6035 stripe_nr /= factor; 6036 stripe_index *= sub_stripes; 6037 6038 remaining_stripes = stripe_cnt % factor; 6039 stripes_per_dev = stripe_cnt / factor; 6040 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes; 6041 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 6042 BTRFS_BLOCK_GROUP_DUP)) { 6043 *num_stripes = map->num_stripes; 6044 } else { 6045 stripe_index = stripe_nr % map->num_stripes; 6046 stripe_nr /= map->num_stripes; 6047 } 6048 6049 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); 6050 if (!stripes) { 6051 ret = -ENOMEM; 6052 goto out_free_map; 6053 } 6054 6055 for (i = 0; i < *num_stripes; i++) { 6056 stripes[i].physical = 6057 map->stripes[stripe_index].physical + 6058 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6059 stripes[i].dev = map->stripes[stripe_index].dev; 6060 6061 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6062 BTRFS_BLOCK_GROUP_RAID10)) { 6063 stripes[i].length = stripes_per_dev << BTRFS_STRIPE_LEN_SHIFT; 6064 6065 if (i / sub_stripes < remaining_stripes) 6066 stripes[i].length += BTRFS_STRIPE_LEN; 6067 6068 /* 6069 * Special for the first stripe and 6070 * the last stripe: 6071 * 6072 * |-------|...|-------| 6073 * |----------| 6074 * off end_off 6075 */ 6076 if (i < sub_stripes) 6077 stripes[i].length -= stripe_offset; 6078 6079 if (stripe_index >= last_stripe && 6080 stripe_index <= (last_stripe + 6081 sub_stripes - 1)) 6082 stripes[i].length -= stripe_end_offset; 6083 6084 if (i == sub_stripes - 1) 6085 stripe_offset = 0; 6086 } else { 6087 stripes[i].length = length; 6088 } 6089 6090 stripe_index++; 6091 if (stripe_index == map->num_stripes) { 6092 stripe_index = 0; 6093 stripe_nr++; 6094 } 6095 } 6096 6097 free_extent_map(em); 6098 return stripes; 6099 out_free_map: 6100 free_extent_map(em); 6101 return ERR_PTR(ret); 6102 } 6103 6104 /* 6105 * In dev-replace case, for repair case (that's the only case where the mirror 6106 * is selected explicitly when calling btrfs_map_block), blocks left of the 6107 * left cursor can also be read from the target drive. 6108 * 6109 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6110 * array of stripes. 6111 * For READ, it also needs to be supported using the same mirror number. 6112 * 6113 * If the requested block is not left of the left cursor, EIO is returned. This 6114 * can happen because btrfs_num_copies() returns one more in the dev-replace 6115 * case. 6116 */ 6117 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6118 u64 logical, u64 length, 6119 u64 srcdev_devid, int *mirror_num, 6120 u64 *physical) 6121 { 6122 struct btrfs_io_context *bioc = NULL; 6123 int num_stripes; 6124 int index_srcdev = 0; 6125 int found = 0; 6126 u64 physical_of_found = 0; 6127 int i; 6128 int ret = 0; 6129 6130 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6131 logical, &length, &bioc, NULL, NULL, 0); 6132 if (ret) { 6133 ASSERT(bioc == NULL); 6134 return ret; 6135 } 6136 6137 num_stripes = bioc->num_stripes; 6138 if (*mirror_num > num_stripes) { 6139 /* 6140 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6141 * that means that the requested area is not left of the left 6142 * cursor 6143 */ 6144 btrfs_put_bioc(bioc); 6145 return -EIO; 6146 } 6147 6148 /* 6149 * process the rest of the function using the mirror_num of the source 6150 * drive. Therefore look it up first. At the end, patch the device 6151 * pointer to the one of the target drive. 6152 */ 6153 for (i = 0; i < num_stripes; i++) { 6154 if (bioc->stripes[i].dev->devid != srcdev_devid) 6155 continue; 6156 6157 /* 6158 * In case of DUP, in order to keep it simple, only add the 6159 * mirror with the lowest physical address 6160 */ 6161 if (found && 6162 physical_of_found <= bioc->stripes[i].physical) 6163 continue; 6164 6165 index_srcdev = i; 6166 found = 1; 6167 physical_of_found = bioc->stripes[i].physical; 6168 } 6169 6170 btrfs_put_bioc(bioc); 6171 6172 ASSERT(found); 6173 if (!found) 6174 return -EIO; 6175 6176 *mirror_num = index_srcdev + 1; 6177 *physical = physical_of_found; 6178 return ret; 6179 } 6180 6181 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6182 { 6183 struct btrfs_block_group *cache; 6184 bool ret; 6185 6186 /* Non zoned filesystem does not use "to_copy" flag */ 6187 if (!btrfs_is_zoned(fs_info)) 6188 return false; 6189 6190 cache = btrfs_lookup_block_group(fs_info, logical); 6191 6192 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 6193 6194 btrfs_put_block_group(cache); 6195 return ret; 6196 } 6197 6198 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6199 struct btrfs_io_context **bioc_ret, 6200 struct btrfs_dev_replace *dev_replace, 6201 u64 logical, 6202 int *num_stripes_ret, int *max_errors_ret) 6203 { 6204 struct btrfs_io_context *bioc = *bioc_ret; 6205 u64 srcdev_devid = dev_replace->srcdev->devid; 6206 int tgtdev_indexes = 0; 6207 int num_stripes = *num_stripes_ret; 6208 int max_errors = *max_errors_ret; 6209 int i; 6210 6211 if (op == BTRFS_MAP_WRITE) { 6212 int index_where_to_add; 6213 6214 /* 6215 * A block group which have "to_copy" set will eventually 6216 * copied by dev-replace process. We can avoid cloning IO here. 6217 */ 6218 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6219 return; 6220 6221 /* 6222 * duplicate the write operations while the dev replace 6223 * procedure is running. Since the copying of the old disk to 6224 * the new disk takes place at run time while the filesystem is 6225 * mounted writable, the regular write operations to the old 6226 * disk have to be duplicated to go to the new disk as well. 6227 * 6228 * Note that device->missing is handled by the caller, and that 6229 * the write to the old disk is already set up in the stripes 6230 * array. 6231 */ 6232 index_where_to_add = num_stripes; 6233 for (i = 0; i < num_stripes; i++) { 6234 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6235 /* write to new disk, too */ 6236 struct btrfs_io_stripe *new = 6237 bioc->stripes + index_where_to_add; 6238 struct btrfs_io_stripe *old = 6239 bioc->stripes + i; 6240 6241 new->physical = old->physical; 6242 new->dev = dev_replace->tgtdev; 6243 bioc->tgtdev_map[i] = index_where_to_add; 6244 index_where_to_add++; 6245 max_errors++; 6246 tgtdev_indexes++; 6247 } 6248 } 6249 num_stripes = index_where_to_add; 6250 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6251 int index_srcdev = 0; 6252 int found = 0; 6253 u64 physical_of_found = 0; 6254 6255 /* 6256 * During the dev-replace procedure, the target drive can also 6257 * be used to read data in case it is needed to repair a corrupt 6258 * block elsewhere. This is possible if the requested area is 6259 * left of the left cursor. In this area, the target drive is a 6260 * full copy of the source drive. 6261 */ 6262 for (i = 0; i < num_stripes; i++) { 6263 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6264 /* 6265 * In case of DUP, in order to keep it simple, 6266 * only add the mirror with the lowest physical 6267 * address 6268 */ 6269 if (found && 6270 physical_of_found <= bioc->stripes[i].physical) 6271 continue; 6272 index_srcdev = i; 6273 found = 1; 6274 physical_of_found = bioc->stripes[i].physical; 6275 } 6276 } 6277 if (found) { 6278 struct btrfs_io_stripe *tgtdev_stripe = 6279 bioc->stripes + num_stripes; 6280 6281 tgtdev_stripe->physical = physical_of_found; 6282 tgtdev_stripe->dev = dev_replace->tgtdev; 6283 bioc->tgtdev_map[index_srcdev] = num_stripes; 6284 6285 tgtdev_indexes++; 6286 num_stripes++; 6287 } 6288 } 6289 6290 *num_stripes_ret = num_stripes; 6291 *max_errors_ret = max_errors; 6292 bioc->num_tgtdevs = tgtdev_indexes; 6293 *bioc_ret = bioc; 6294 } 6295 6296 static bool need_full_stripe(enum btrfs_map_op op) 6297 { 6298 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6299 } 6300 6301 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op, 6302 u64 offset, u32 *stripe_nr, u64 *stripe_offset, 6303 u64 *full_stripe_start) 6304 { 6305 ASSERT(op != BTRFS_MAP_DISCARD); 6306 6307 /* 6308 * Stripe_nr is the stripe where this block falls. stripe_offset is 6309 * the offset of this block in its stripe. 6310 */ 6311 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK; 6312 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; 6313 ASSERT(*stripe_offset < U32_MAX); 6314 6315 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6316 unsigned long full_stripe_len = nr_data_stripes(map) << 6317 BTRFS_STRIPE_LEN_SHIFT; 6318 6319 /* 6320 * For full stripe start, we use previously calculated 6321 * @stripe_nr. Align it to nr_data_stripes, then multiply with 6322 * STRIPE_LEN. 6323 * 6324 * By this we can avoid u64 division completely. And we have 6325 * to go rounddown(), not round_down(), as nr_data_stripes is 6326 * not ensured to be power of 2. 6327 */ 6328 *full_stripe_start = 6329 rounddown(*stripe_nr, nr_data_stripes(map)) << 6330 BTRFS_STRIPE_LEN_SHIFT; 6331 6332 /* 6333 * For writes to RAID56, allow to write a full stripe set, but 6334 * no straddling of stripe sets. 6335 */ 6336 if (op == BTRFS_MAP_WRITE) 6337 return full_stripe_len - (offset - *full_stripe_start); 6338 } 6339 6340 /* 6341 * For other RAID types and for RAID56 reads, allow a single stripe (on 6342 * a single disk). 6343 */ 6344 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) 6345 return BTRFS_STRIPE_LEN - *stripe_offset; 6346 return U64_MAX; 6347 } 6348 6349 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, 6350 u32 stripe_index, u64 stripe_offset, u32 stripe_nr) 6351 { 6352 dst->dev = map->stripes[stripe_index].dev; 6353 dst->physical = map->stripes[stripe_index].physical + 6354 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); 6355 } 6356 6357 int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6358 u64 logical, u64 *length, 6359 struct btrfs_io_context **bioc_ret, 6360 struct btrfs_io_stripe *smap, int *mirror_num_ret, 6361 int need_raid_map) 6362 { 6363 struct extent_map *em; 6364 struct map_lookup *map; 6365 u64 map_offset; 6366 u64 stripe_offset; 6367 u32 stripe_nr; 6368 u32 stripe_index; 6369 int data_stripes; 6370 int i; 6371 int ret = 0; 6372 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); 6373 int num_stripes; 6374 int max_errors = 0; 6375 int tgtdev_indexes = 0; 6376 struct btrfs_io_context *bioc = NULL; 6377 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6378 int dev_replace_is_ongoing = 0; 6379 int num_alloc_stripes; 6380 int patch_the_first_stripe_for_dev_replace = 0; 6381 u64 physical_to_patch_in_first_stripe = 0; 6382 u64 raid56_full_stripe_start = (u64)-1; 6383 u64 max_len; 6384 6385 ASSERT(bioc_ret); 6386 ASSERT(op != BTRFS_MAP_DISCARD); 6387 6388 em = btrfs_get_chunk_map(fs_info, logical, *length); 6389 if (IS_ERR(em)) 6390 return PTR_ERR(em); 6391 6392 map = em->map_lookup; 6393 data_stripes = nr_data_stripes(map); 6394 6395 map_offset = logical - em->start; 6396 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr, 6397 &stripe_offset, &raid56_full_stripe_start); 6398 *length = min_t(u64, em->len - map_offset, max_len); 6399 6400 down_read(&dev_replace->rwsem); 6401 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6402 /* 6403 * Hold the semaphore for read during the whole operation, write is 6404 * requested at commit time but must wait. 6405 */ 6406 if (!dev_replace_is_ongoing) 6407 up_read(&dev_replace->rwsem); 6408 6409 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6410 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6411 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6412 dev_replace->srcdev->devid, 6413 &mirror_num, 6414 &physical_to_patch_in_first_stripe); 6415 if (ret) 6416 goto out; 6417 else 6418 patch_the_first_stripe_for_dev_replace = 1; 6419 } else if (mirror_num > map->num_stripes) { 6420 mirror_num = 0; 6421 } 6422 6423 num_stripes = 1; 6424 stripe_index = 0; 6425 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6426 stripe_index = stripe_nr % map->num_stripes; 6427 stripe_nr /= map->num_stripes; 6428 if (!need_full_stripe(op)) 6429 mirror_num = 1; 6430 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6431 if (need_full_stripe(op)) 6432 num_stripes = map->num_stripes; 6433 else if (mirror_num) 6434 stripe_index = mirror_num - 1; 6435 else { 6436 stripe_index = find_live_mirror(fs_info, map, 0, 6437 dev_replace_is_ongoing); 6438 mirror_num = stripe_index + 1; 6439 } 6440 6441 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6442 if (need_full_stripe(op)) { 6443 num_stripes = map->num_stripes; 6444 } else if (mirror_num) { 6445 stripe_index = mirror_num - 1; 6446 } else { 6447 mirror_num = 1; 6448 } 6449 6450 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6451 u32 factor = map->num_stripes / map->sub_stripes; 6452 6453 stripe_index = (stripe_nr % factor) * map->sub_stripes; 6454 stripe_nr /= factor; 6455 6456 if (need_full_stripe(op)) 6457 num_stripes = map->sub_stripes; 6458 else if (mirror_num) 6459 stripe_index += mirror_num - 1; 6460 else { 6461 int old_stripe_index = stripe_index; 6462 stripe_index = find_live_mirror(fs_info, map, 6463 stripe_index, 6464 dev_replace_is_ongoing); 6465 mirror_num = stripe_index - old_stripe_index + 1; 6466 } 6467 6468 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6469 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6470 /* 6471 * Push stripe_nr back to the start of the full stripe 6472 * For those cases needing a full stripe, @stripe_nr 6473 * is the full stripe number. 6474 * 6475 * Originally we go raid56_full_stripe_start / full_stripe_len, 6476 * but that can be expensive. Here we just divide 6477 * @stripe_nr with @data_stripes. 6478 */ 6479 stripe_nr /= data_stripes; 6480 6481 /* RAID[56] write or recovery. Return all stripes */ 6482 num_stripes = map->num_stripes; 6483 max_errors = btrfs_chunk_max_errors(map); 6484 6485 /* Return the length to the full stripe end */ 6486 *length = min(logical + *length, 6487 raid56_full_stripe_start + em->start + 6488 (data_stripes << BTRFS_STRIPE_LEN_SHIFT)) - logical; 6489 stripe_index = 0; 6490 stripe_offset = 0; 6491 } else { 6492 /* 6493 * Mirror #0 or #1 means the original data block. 6494 * Mirror #2 is RAID5 parity block. 6495 * Mirror #3 is RAID6 Q block. 6496 */ 6497 stripe_index = stripe_nr % data_stripes; 6498 stripe_nr /= data_stripes; 6499 if (mirror_num > 1) 6500 stripe_index = data_stripes + mirror_num - 2; 6501 6502 /* We distribute the parity blocks across stripes */ 6503 stripe_index = (stripe_nr + stripe_index) % map->num_stripes; 6504 if (!need_full_stripe(op) && mirror_num <= 1) 6505 mirror_num = 1; 6506 } 6507 } else { 6508 /* 6509 * After this, stripe_nr is the number of stripes on this 6510 * device we have to walk to find the data, and stripe_index is 6511 * the number of our device in the stripe array 6512 */ 6513 stripe_index = stripe_nr % map->num_stripes; 6514 stripe_nr /= map->num_stripes; 6515 mirror_num = stripe_index + 1; 6516 } 6517 if (stripe_index >= map->num_stripes) { 6518 btrfs_crit(fs_info, 6519 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6520 stripe_index, map->num_stripes); 6521 ret = -EINVAL; 6522 goto out; 6523 } 6524 6525 num_alloc_stripes = num_stripes; 6526 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6527 if (op == BTRFS_MAP_WRITE) 6528 num_alloc_stripes <<= 1; 6529 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6530 num_alloc_stripes++; 6531 tgtdev_indexes = num_stripes; 6532 } 6533 6534 /* 6535 * If this I/O maps to a single device, try to return the device and 6536 * physical block information on the stack instead of allocating an 6537 * I/O context structure. 6538 */ 6539 if (smap && num_alloc_stripes == 1 && 6540 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && 6541 (!need_full_stripe(op) || !dev_replace_is_ongoing || 6542 !dev_replace->tgtdev)) { 6543 if (patch_the_first_stripe_for_dev_replace) { 6544 smap->dev = dev_replace->tgtdev; 6545 smap->physical = physical_to_patch_in_first_stripe; 6546 *mirror_num_ret = map->num_stripes + 1; 6547 } else { 6548 set_io_stripe(smap, map, stripe_index, stripe_offset, 6549 stripe_nr); 6550 *mirror_num_ret = mirror_num; 6551 } 6552 *bioc_ret = NULL; 6553 ret = 0; 6554 goto out; 6555 } 6556 6557 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6558 if (!bioc) { 6559 ret = -ENOMEM; 6560 goto out; 6561 } 6562 6563 for (i = 0; i < num_stripes; i++) { 6564 set_io_stripe(&bioc->stripes[i], map, stripe_index, stripe_offset, 6565 stripe_nr); 6566 stripe_index++; 6567 } 6568 6569 /* Build raid_map */ 6570 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6571 (need_full_stripe(op) || mirror_num > 1)) { 6572 u64 tmp; 6573 unsigned rot; 6574 6575 /* Work out the disk rotation on this stripe-set */ 6576 rot = stripe_nr % num_stripes; 6577 6578 /* Fill in the logical address of each stripe */ 6579 tmp = stripe_nr * data_stripes; 6580 for (i = 0; i < data_stripes; i++) 6581 bioc->raid_map[(i + rot) % num_stripes] = 6582 em->start + ((tmp + i) << BTRFS_STRIPE_LEN_SHIFT); 6583 6584 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6585 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6586 bioc->raid_map[(i + rot + 1) % num_stripes] = 6587 RAID6_Q_STRIPE; 6588 6589 sort_parity_stripes(bioc, num_stripes); 6590 } 6591 6592 if (need_full_stripe(op)) 6593 max_errors = btrfs_chunk_max_errors(map); 6594 6595 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6596 need_full_stripe(op)) { 6597 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6598 &num_stripes, &max_errors); 6599 } 6600 6601 *bioc_ret = bioc; 6602 bioc->map_type = map->type; 6603 bioc->num_stripes = num_stripes; 6604 bioc->max_errors = max_errors; 6605 bioc->mirror_num = mirror_num; 6606 6607 /* 6608 * this is the case that REQ_READ && dev_replace_is_ongoing && 6609 * mirror_num == num_stripes + 1 && dev_replace target drive is 6610 * available as a mirror 6611 */ 6612 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6613 WARN_ON(num_stripes > 1); 6614 bioc->stripes[0].dev = dev_replace->tgtdev; 6615 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6616 bioc->mirror_num = map->num_stripes + 1; 6617 } 6618 out: 6619 if (dev_replace_is_ongoing) { 6620 lockdep_assert_held(&dev_replace->rwsem); 6621 /* Unlock and let waiting writers proceed */ 6622 up_read(&dev_replace->rwsem); 6623 } 6624 free_extent_map(em); 6625 return ret; 6626 } 6627 6628 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6629 u64 logical, u64 *length, 6630 struct btrfs_io_context **bioc_ret, int mirror_num) 6631 { 6632 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6633 NULL, &mirror_num, 0); 6634 } 6635 6636 /* For Scrub/replace */ 6637 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6638 u64 logical, u64 *length, 6639 struct btrfs_io_context **bioc_ret) 6640 { 6641 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6642 NULL, NULL, 1); 6643 } 6644 6645 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6646 const struct btrfs_fs_devices *fs_devices) 6647 { 6648 if (args->fsid == NULL) 6649 return true; 6650 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6651 return true; 6652 return false; 6653 } 6654 6655 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6656 const struct btrfs_device *device) 6657 { 6658 if (args->missing) { 6659 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6660 !device->bdev) 6661 return true; 6662 return false; 6663 } 6664 6665 if (device->devid != args->devid) 6666 return false; 6667 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6668 return false; 6669 return true; 6670 } 6671 6672 /* 6673 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6674 * return NULL. 6675 * 6676 * If devid and uuid are both specified, the match must be exact, otherwise 6677 * only devid is used. 6678 */ 6679 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6680 const struct btrfs_dev_lookup_args *args) 6681 { 6682 struct btrfs_device *device; 6683 struct btrfs_fs_devices *seed_devs; 6684 6685 if (dev_args_match_fs_devices(args, fs_devices)) { 6686 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6687 if (dev_args_match_device(args, device)) 6688 return device; 6689 } 6690 } 6691 6692 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6693 if (!dev_args_match_fs_devices(args, seed_devs)) 6694 continue; 6695 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6696 if (dev_args_match_device(args, device)) 6697 return device; 6698 } 6699 } 6700 6701 return NULL; 6702 } 6703 6704 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6705 u64 devid, u8 *dev_uuid) 6706 { 6707 struct btrfs_device *device; 6708 unsigned int nofs_flag; 6709 6710 /* 6711 * We call this under the chunk_mutex, so we want to use NOFS for this 6712 * allocation, however we don't want to change btrfs_alloc_device() to 6713 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6714 * places. 6715 */ 6716 6717 nofs_flag = memalloc_nofs_save(); 6718 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); 6719 memalloc_nofs_restore(nofs_flag); 6720 if (IS_ERR(device)) 6721 return device; 6722 6723 list_add(&device->dev_list, &fs_devices->devices); 6724 device->fs_devices = fs_devices; 6725 fs_devices->num_devices++; 6726 6727 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6728 fs_devices->missing_devices++; 6729 6730 return device; 6731 } 6732 6733 /* 6734 * Allocate new device struct, set up devid and UUID. 6735 * 6736 * @fs_info: used only for generating a new devid, can be NULL if 6737 * devid is provided (i.e. @devid != NULL). 6738 * @devid: a pointer to devid for this device. If NULL a new devid 6739 * is generated. 6740 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6741 * is generated. 6742 * @path: a pointer to device path if available, NULL otherwise. 6743 * 6744 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6745 * on error. Returned struct is not linked onto any lists and must be 6746 * destroyed with btrfs_free_device. 6747 */ 6748 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6749 const u64 *devid, const u8 *uuid, 6750 const char *path) 6751 { 6752 struct btrfs_device *dev; 6753 u64 tmp; 6754 6755 if (WARN_ON(!devid && !fs_info)) 6756 return ERR_PTR(-EINVAL); 6757 6758 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6759 if (!dev) 6760 return ERR_PTR(-ENOMEM); 6761 6762 INIT_LIST_HEAD(&dev->dev_list); 6763 INIT_LIST_HEAD(&dev->dev_alloc_list); 6764 INIT_LIST_HEAD(&dev->post_commit_list); 6765 6766 atomic_set(&dev->dev_stats_ccnt, 0); 6767 btrfs_device_data_ordered_init(dev); 6768 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); 6769 6770 if (devid) 6771 tmp = *devid; 6772 else { 6773 int ret; 6774 6775 ret = find_next_devid(fs_info, &tmp); 6776 if (ret) { 6777 btrfs_free_device(dev); 6778 return ERR_PTR(ret); 6779 } 6780 } 6781 dev->devid = tmp; 6782 6783 if (uuid) 6784 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6785 else 6786 generate_random_uuid(dev->uuid); 6787 6788 if (path) { 6789 struct rcu_string *name; 6790 6791 name = rcu_string_strdup(path, GFP_KERNEL); 6792 if (!name) { 6793 btrfs_free_device(dev); 6794 return ERR_PTR(-ENOMEM); 6795 } 6796 rcu_assign_pointer(dev->name, name); 6797 } 6798 6799 return dev; 6800 } 6801 6802 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6803 u64 devid, u8 *uuid, bool error) 6804 { 6805 if (error) 6806 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6807 devid, uuid); 6808 else 6809 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6810 devid, uuid); 6811 } 6812 6813 u64 btrfs_calc_stripe_length(const struct extent_map *em) 6814 { 6815 const struct map_lookup *map = em->map_lookup; 6816 const int data_stripes = calc_data_stripes(map->type, map->num_stripes); 6817 6818 return div_u64(em->len, data_stripes); 6819 } 6820 6821 #if BITS_PER_LONG == 32 6822 /* 6823 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6824 * can't be accessed on 32bit systems. 6825 * 6826 * This function do mount time check to reject the fs if it already has 6827 * metadata chunk beyond that limit. 6828 */ 6829 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6830 u64 logical, u64 length, u64 type) 6831 { 6832 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6833 return 0; 6834 6835 if (logical + length < MAX_LFS_FILESIZE) 6836 return 0; 6837 6838 btrfs_err_32bit_limit(fs_info); 6839 return -EOVERFLOW; 6840 } 6841 6842 /* 6843 * This is to give early warning for any metadata chunk reaching 6844 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6845 * Although we can still access the metadata, it's not going to be possible 6846 * once the limit is reached. 6847 */ 6848 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6849 u64 logical, u64 length, u64 type) 6850 { 6851 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6852 return; 6853 6854 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 6855 return; 6856 6857 btrfs_warn_32bit_limit(fs_info); 6858 } 6859 #endif 6860 6861 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 6862 u64 devid, u8 *uuid) 6863 { 6864 struct btrfs_device *dev; 6865 6866 if (!btrfs_test_opt(fs_info, DEGRADED)) { 6867 btrfs_report_missing_device(fs_info, devid, uuid, true); 6868 return ERR_PTR(-ENOENT); 6869 } 6870 6871 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 6872 if (IS_ERR(dev)) { 6873 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 6874 devid, PTR_ERR(dev)); 6875 return dev; 6876 } 6877 btrfs_report_missing_device(fs_info, devid, uuid, false); 6878 6879 return dev; 6880 } 6881 6882 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 6883 struct btrfs_chunk *chunk) 6884 { 6885 BTRFS_DEV_LOOKUP_ARGS(args); 6886 struct btrfs_fs_info *fs_info = leaf->fs_info; 6887 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 6888 struct map_lookup *map; 6889 struct extent_map *em; 6890 u64 logical; 6891 u64 length; 6892 u64 devid; 6893 u64 type; 6894 u8 uuid[BTRFS_UUID_SIZE]; 6895 int index; 6896 int num_stripes; 6897 int ret; 6898 int i; 6899 6900 logical = key->offset; 6901 length = btrfs_chunk_length(leaf, chunk); 6902 type = btrfs_chunk_type(leaf, chunk); 6903 index = btrfs_bg_flags_to_raid_index(type); 6904 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 6905 6906 #if BITS_PER_LONG == 32 6907 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 6908 if (ret < 0) 6909 return ret; 6910 warn_32bit_meta_chunk(fs_info, logical, length, type); 6911 #endif 6912 6913 /* 6914 * Only need to verify chunk item if we're reading from sys chunk array, 6915 * as chunk item in tree block is already verified by tree-checker. 6916 */ 6917 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 6918 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 6919 if (ret) 6920 return ret; 6921 } 6922 6923 read_lock(&map_tree->lock); 6924 em = lookup_extent_mapping(map_tree, logical, 1); 6925 read_unlock(&map_tree->lock); 6926 6927 /* already mapped? */ 6928 if (em && em->start <= logical && em->start + em->len > logical) { 6929 free_extent_map(em); 6930 return 0; 6931 } else if (em) { 6932 free_extent_map(em); 6933 } 6934 6935 em = alloc_extent_map(); 6936 if (!em) 6937 return -ENOMEM; 6938 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 6939 if (!map) { 6940 free_extent_map(em); 6941 return -ENOMEM; 6942 } 6943 6944 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 6945 em->map_lookup = map; 6946 em->start = logical; 6947 em->len = length; 6948 em->orig_start = 0; 6949 em->block_start = 0; 6950 em->block_len = em->len; 6951 6952 map->num_stripes = num_stripes; 6953 map->io_width = btrfs_chunk_io_width(leaf, chunk); 6954 map->io_align = btrfs_chunk_io_align(leaf, chunk); 6955 map->type = type; 6956 /* 6957 * We can't use the sub_stripes value, as for profiles other than 6958 * RAID10, they may have 0 as sub_stripes for filesystems created by 6959 * older mkfs (<v5.4). 6960 * In that case, it can cause divide-by-zero errors later. 6961 * Since currently sub_stripes is fixed for each profile, let's 6962 * use the trusted value instead. 6963 */ 6964 map->sub_stripes = btrfs_raid_array[index].sub_stripes; 6965 map->verified_stripes = 0; 6966 em->orig_block_len = btrfs_calc_stripe_length(em); 6967 for (i = 0; i < num_stripes; i++) { 6968 map->stripes[i].physical = 6969 btrfs_stripe_offset_nr(leaf, chunk, i); 6970 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 6971 args.devid = devid; 6972 read_extent_buffer(leaf, uuid, (unsigned long) 6973 btrfs_stripe_dev_uuid_nr(chunk, i), 6974 BTRFS_UUID_SIZE); 6975 args.uuid = uuid; 6976 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 6977 if (!map->stripes[i].dev) { 6978 map->stripes[i].dev = handle_missing_device(fs_info, 6979 devid, uuid); 6980 if (IS_ERR(map->stripes[i].dev)) { 6981 ret = PTR_ERR(map->stripes[i].dev); 6982 free_extent_map(em); 6983 return ret; 6984 } 6985 } 6986 6987 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 6988 &(map->stripes[i].dev->dev_state)); 6989 } 6990 6991 write_lock(&map_tree->lock); 6992 ret = add_extent_mapping(map_tree, em, 0); 6993 write_unlock(&map_tree->lock); 6994 if (ret < 0) { 6995 btrfs_err(fs_info, 6996 "failed to add chunk map, start=%llu len=%llu: %d", 6997 em->start, em->len, ret); 6998 } 6999 free_extent_map(em); 7000 7001 return ret; 7002 } 7003 7004 static void fill_device_from_item(struct extent_buffer *leaf, 7005 struct btrfs_dev_item *dev_item, 7006 struct btrfs_device *device) 7007 { 7008 unsigned long ptr; 7009 7010 device->devid = btrfs_device_id(leaf, dev_item); 7011 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7012 device->total_bytes = device->disk_total_bytes; 7013 device->commit_total_bytes = device->disk_total_bytes; 7014 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7015 device->commit_bytes_used = device->bytes_used; 7016 device->type = btrfs_device_type(leaf, dev_item); 7017 device->io_align = btrfs_device_io_align(leaf, dev_item); 7018 device->io_width = btrfs_device_io_width(leaf, dev_item); 7019 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7020 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7021 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7022 7023 ptr = btrfs_device_uuid(dev_item); 7024 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7025 } 7026 7027 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7028 u8 *fsid) 7029 { 7030 struct btrfs_fs_devices *fs_devices; 7031 int ret; 7032 7033 lockdep_assert_held(&uuid_mutex); 7034 ASSERT(fsid); 7035 7036 /* This will match only for multi-device seed fs */ 7037 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7038 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7039 return fs_devices; 7040 7041 7042 fs_devices = find_fsid(fsid, NULL); 7043 if (!fs_devices) { 7044 if (!btrfs_test_opt(fs_info, DEGRADED)) 7045 return ERR_PTR(-ENOENT); 7046 7047 fs_devices = alloc_fs_devices(fsid, NULL); 7048 if (IS_ERR(fs_devices)) 7049 return fs_devices; 7050 7051 fs_devices->seeding = true; 7052 fs_devices->opened = 1; 7053 return fs_devices; 7054 } 7055 7056 /* 7057 * Upon first call for a seed fs fsid, just create a private copy of the 7058 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7059 */ 7060 fs_devices = clone_fs_devices(fs_devices); 7061 if (IS_ERR(fs_devices)) 7062 return fs_devices; 7063 7064 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7065 if (ret) { 7066 free_fs_devices(fs_devices); 7067 return ERR_PTR(ret); 7068 } 7069 7070 if (!fs_devices->seeding) { 7071 close_fs_devices(fs_devices); 7072 free_fs_devices(fs_devices); 7073 return ERR_PTR(-EINVAL); 7074 } 7075 7076 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7077 7078 return fs_devices; 7079 } 7080 7081 static int read_one_dev(struct extent_buffer *leaf, 7082 struct btrfs_dev_item *dev_item) 7083 { 7084 BTRFS_DEV_LOOKUP_ARGS(args); 7085 struct btrfs_fs_info *fs_info = leaf->fs_info; 7086 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7087 struct btrfs_device *device; 7088 u64 devid; 7089 int ret; 7090 u8 fs_uuid[BTRFS_FSID_SIZE]; 7091 u8 dev_uuid[BTRFS_UUID_SIZE]; 7092 7093 devid = btrfs_device_id(leaf, dev_item); 7094 args.devid = devid; 7095 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7096 BTRFS_UUID_SIZE); 7097 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7098 BTRFS_FSID_SIZE); 7099 args.uuid = dev_uuid; 7100 args.fsid = fs_uuid; 7101 7102 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7103 fs_devices = open_seed_devices(fs_info, fs_uuid); 7104 if (IS_ERR(fs_devices)) 7105 return PTR_ERR(fs_devices); 7106 } 7107 7108 device = btrfs_find_device(fs_info->fs_devices, &args); 7109 if (!device) { 7110 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7111 btrfs_report_missing_device(fs_info, devid, 7112 dev_uuid, true); 7113 return -ENOENT; 7114 } 7115 7116 device = add_missing_dev(fs_devices, devid, dev_uuid); 7117 if (IS_ERR(device)) { 7118 btrfs_err(fs_info, 7119 "failed to add missing dev %llu: %ld", 7120 devid, PTR_ERR(device)); 7121 return PTR_ERR(device); 7122 } 7123 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7124 } else { 7125 if (!device->bdev) { 7126 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7127 btrfs_report_missing_device(fs_info, 7128 devid, dev_uuid, true); 7129 return -ENOENT; 7130 } 7131 btrfs_report_missing_device(fs_info, devid, 7132 dev_uuid, false); 7133 } 7134 7135 if (!device->bdev && 7136 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7137 /* 7138 * this happens when a device that was properly setup 7139 * in the device info lists suddenly goes bad. 7140 * device->bdev is NULL, and so we have to set 7141 * device->missing to one here 7142 */ 7143 device->fs_devices->missing_devices++; 7144 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7145 } 7146 7147 /* Move the device to its own fs_devices */ 7148 if (device->fs_devices != fs_devices) { 7149 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7150 &device->dev_state)); 7151 7152 list_move(&device->dev_list, &fs_devices->devices); 7153 device->fs_devices->num_devices--; 7154 fs_devices->num_devices++; 7155 7156 device->fs_devices->missing_devices--; 7157 fs_devices->missing_devices++; 7158 7159 device->fs_devices = fs_devices; 7160 } 7161 } 7162 7163 if (device->fs_devices != fs_info->fs_devices) { 7164 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7165 if (device->generation != 7166 btrfs_device_generation(leaf, dev_item)) 7167 return -EINVAL; 7168 } 7169 7170 fill_device_from_item(leaf, dev_item, device); 7171 if (device->bdev) { 7172 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7173 7174 if (device->total_bytes > max_total_bytes) { 7175 btrfs_err(fs_info, 7176 "device total_bytes should be at most %llu but found %llu", 7177 max_total_bytes, device->total_bytes); 7178 return -EINVAL; 7179 } 7180 } 7181 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7182 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7183 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7184 device->fs_devices->total_rw_bytes += device->total_bytes; 7185 atomic64_add(device->total_bytes - device->bytes_used, 7186 &fs_info->free_chunk_space); 7187 } 7188 ret = 0; 7189 return ret; 7190 } 7191 7192 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7193 { 7194 struct btrfs_super_block *super_copy = fs_info->super_copy; 7195 struct extent_buffer *sb; 7196 struct btrfs_disk_key *disk_key; 7197 struct btrfs_chunk *chunk; 7198 u8 *array_ptr; 7199 unsigned long sb_array_offset; 7200 int ret = 0; 7201 u32 num_stripes; 7202 u32 array_size; 7203 u32 len = 0; 7204 u32 cur_offset; 7205 u64 type; 7206 struct btrfs_key key; 7207 7208 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7209 7210 /* 7211 * We allocated a dummy extent, just to use extent buffer accessors. 7212 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but 7213 * that's fine, we will not go beyond system chunk array anyway. 7214 */ 7215 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); 7216 if (!sb) 7217 return -ENOMEM; 7218 set_extent_buffer_uptodate(sb); 7219 7220 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7221 array_size = btrfs_super_sys_array_size(super_copy); 7222 7223 array_ptr = super_copy->sys_chunk_array; 7224 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7225 cur_offset = 0; 7226 7227 while (cur_offset < array_size) { 7228 disk_key = (struct btrfs_disk_key *)array_ptr; 7229 len = sizeof(*disk_key); 7230 if (cur_offset + len > array_size) 7231 goto out_short_read; 7232 7233 btrfs_disk_key_to_cpu(&key, disk_key); 7234 7235 array_ptr += len; 7236 sb_array_offset += len; 7237 cur_offset += len; 7238 7239 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7240 btrfs_err(fs_info, 7241 "unexpected item type %u in sys_array at offset %u", 7242 (u32)key.type, cur_offset); 7243 ret = -EIO; 7244 break; 7245 } 7246 7247 chunk = (struct btrfs_chunk *)sb_array_offset; 7248 /* 7249 * At least one btrfs_chunk with one stripe must be present, 7250 * exact stripe count check comes afterwards 7251 */ 7252 len = btrfs_chunk_item_size(1); 7253 if (cur_offset + len > array_size) 7254 goto out_short_read; 7255 7256 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7257 if (!num_stripes) { 7258 btrfs_err(fs_info, 7259 "invalid number of stripes %u in sys_array at offset %u", 7260 num_stripes, cur_offset); 7261 ret = -EIO; 7262 break; 7263 } 7264 7265 type = btrfs_chunk_type(sb, chunk); 7266 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7267 btrfs_err(fs_info, 7268 "invalid chunk type %llu in sys_array at offset %u", 7269 type, cur_offset); 7270 ret = -EIO; 7271 break; 7272 } 7273 7274 len = btrfs_chunk_item_size(num_stripes); 7275 if (cur_offset + len > array_size) 7276 goto out_short_read; 7277 7278 ret = read_one_chunk(&key, sb, chunk); 7279 if (ret) 7280 break; 7281 7282 array_ptr += len; 7283 sb_array_offset += len; 7284 cur_offset += len; 7285 } 7286 clear_extent_buffer_uptodate(sb); 7287 free_extent_buffer_stale(sb); 7288 return ret; 7289 7290 out_short_read: 7291 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7292 len, cur_offset); 7293 clear_extent_buffer_uptodate(sb); 7294 free_extent_buffer_stale(sb); 7295 return -EIO; 7296 } 7297 7298 /* 7299 * Check if all chunks in the fs are OK for read-write degraded mount 7300 * 7301 * If the @failing_dev is specified, it's accounted as missing. 7302 * 7303 * Return true if all chunks meet the minimal RW mount requirements. 7304 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7305 */ 7306 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7307 struct btrfs_device *failing_dev) 7308 { 7309 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7310 struct extent_map *em; 7311 u64 next_start = 0; 7312 bool ret = true; 7313 7314 read_lock(&map_tree->lock); 7315 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7316 read_unlock(&map_tree->lock); 7317 /* No chunk at all? Return false anyway */ 7318 if (!em) { 7319 ret = false; 7320 goto out; 7321 } 7322 while (em) { 7323 struct map_lookup *map; 7324 int missing = 0; 7325 int max_tolerated; 7326 int i; 7327 7328 map = em->map_lookup; 7329 max_tolerated = 7330 btrfs_get_num_tolerated_disk_barrier_failures( 7331 map->type); 7332 for (i = 0; i < map->num_stripes; i++) { 7333 struct btrfs_device *dev = map->stripes[i].dev; 7334 7335 if (!dev || !dev->bdev || 7336 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7337 dev->last_flush_error) 7338 missing++; 7339 else if (failing_dev && failing_dev == dev) 7340 missing++; 7341 } 7342 if (missing > max_tolerated) { 7343 if (!failing_dev) 7344 btrfs_warn(fs_info, 7345 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7346 em->start, missing, max_tolerated); 7347 free_extent_map(em); 7348 ret = false; 7349 goto out; 7350 } 7351 next_start = extent_map_end(em); 7352 free_extent_map(em); 7353 7354 read_lock(&map_tree->lock); 7355 em = lookup_extent_mapping(map_tree, next_start, 7356 (u64)(-1) - next_start); 7357 read_unlock(&map_tree->lock); 7358 } 7359 out: 7360 return ret; 7361 } 7362 7363 static void readahead_tree_node_children(struct extent_buffer *node) 7364 { 7365 int i; 7366 const int nr_items = btrfs_header_nritems(node); 7367 7368 for (i = 0; i < nr_items; i++) 7369 btrfs_readahead_node_child(node, i); 7370 } 7371 7372 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7373 { 7374 struct btrfs_root *root = fs_info->chunk_root; 7375 struct btrfs_path *path; 7376 struct extent_buffer *leaf; 7377 struct btrfs_key key; 7378 struct btrfs_key found_key; 7379 int ret; 7380 int slot; 7381 int iter_ret = 0; 7382 u64 total_dev = 0; 7383 u64 last_ra_node = 0; 7384 7385 path = btrfs_alloc_path(); 7386 if (!path) 7387 return -ENOMEM; 7388 7389 /* 7390 * uuid_mutex is needed only if we are mounting a sprout FS 7391 * otherwise we don't need it. 7392 */ 7393 mutex_lock(&uuid_mutex); 7394 7395 /* 7396 * It is possible for mount and umount to race in such a way that 7397 * we execute this code path, but open_fs_devices failed to clear 7398 * total_rw_bytes. We certainly want it cleared before reading the 7399 * device items, so clear it here. 7400 */ 7401 fs_info->fs_devices->total_rw_bytes = 0; 7402 7403 /* 7404 * Lockdep complains about possible circular locking dependency between 7405 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7406 * used for freeze procection of a fs (struct super_block.s_writers), 7407 * which we take when starting a transaction, and extent buffers of the 7408 * chunk tree if we call read_one_dev() while holding a lock on an 7409 * extent buffer of the chunk tree. Since we are mounting the filesystem 7410 * and at this point there can't be any concurrent task modifying the 7411 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7412 */ 7413 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7414 path->skip_locking = 1; 7415 7416 /* 7417 * Read all device items, and then all the chunk items. All 7418 * device items are found before any chunk item (their object id 7419 * is smaller than the lowest possible object id for a chunk 7420 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7421 */ 7422 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7423 key.offset = 0; 7424 key.type = 0; 7425 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 7426 struct extent_buffer *node = path->nodes[1]; 7427 7428 leaf = path->nodes[0]; 7429 slot = path->slots[0]; 7430 7431 if (node) { 7432 if (last_ra_node != node->start) { 7433 readahead_tree_node_children(node); 7434 last_ra_node = node->start; 7435 } 7436 } 7437 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7438 struct btrfs_dev_item *dev_item; 7439 dev_item = btrfs_item_ptr(leaf, slot, 7440 struct btrfs_dev_item); 7441 ret = read_one_dev(leaf, dev_item); 7442 if (ret) 7443 goto error; 7444 total_dev++; 7445 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7446 struct btrfs_chunk *chunk; 7447 7448 /* 7449 * We are only called at mount time, so no need to take 7450 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7451 * we always lock first fs_info->chunk_mutex before 7452 * acquiring any locks on the chunk tree. This is a 7453 * requirement for chunk allocation, see the comment on 7454 * top of btrfs_chunk_alloc() for details. 7455 */ 7456 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7457 ret = read_one_chunk(&found_key, leaf, chunk); 7458 if (ret) 7459 goto error; 7460 } 7461 } 7462 /* Catch error found during iteration */ 7463 if (iter_ret < 0) { 7464 ret = iter_ret; 7465 goto error; 7466 } 7467 7468 /* 7469 * After loading chunk tree, we've got all device information, 7470 * do another round of validation checks. 7471 */ 7472 if (total_dev != fs_info->fs_devices->total_devices) { 7473 btrfs_warn(fs_info, 7474 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", 7475 btrfs_super_num_devices(fs_info->super_copy), 7476 total_dev); 7477 fs_info->fs_devices->total_devices = total_dev; 7478 btrfs_set_super_num_devices(fs_info->super_copy, total_dev); 7479 } 7480 if (btrfs_super_total_bytes(fs_info->super_copy) < 7481 fs_info->fs_devices->total_rw_bytes) { 7482 btrfs_err(fs_info, 7483 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7484 btrfs_super_total_bytes(fs_info->super_copy), 7485 fs_info->fs_devices->total_rw_bytes); 7486 ret = -EINVAL; 7487 goto error; 7488 } 7489 ret = 0; 7490 error: 7491 mutex_unlock(&uuid_mutex); 7492 7493 btrfs_free_path(path); 7494 return ret; 7495 } 7496 7497 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7498 { 7499 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7500 struct btrfs_device *device; 7501 int ret = 0; 7502 7503 fs_devices->fs_info = fs_info; 7504 7505 mutex_lock(&fs_devices->device_list_mutex); 7506 list_for_each_entry(device, &fs_devices->devices, dev_list) 7507 device->fs_info = fs_info; 7508 7509 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7510 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7511 device->fs_info = fs_info; 7512 ret = btrfs_get_dev_zone_info(device, false); 7513 if (ret) 7514 break; 7515 } 7516 7517 seed_devs->fs_info = fs_info; 7518 } 7519 mutex_unlock(&fs_devices->device_list_mutex); 7520 7521 return ret; 7522 } 7523 7524 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7525 const struct btrfs_dev_stats_item *ptr, 7526 int index) 7527 { 7528 u64 val; 7529 7530 read_extent_buffer(eb, &val, 7531 offsetof(struct btrfs_dev_stats_item, values) + 7532 ((unsigned long)ptr) + (index * sizeof(u64)), 7533 sizeof(val)); 7534 return val; 7535 } 7536 7537 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7538 struct btrfs_dev_stats_item *ptr, 7539 int index, u64 val) 7540 { 7541 write_extent_buffer(eb, &val, 7542 offsetof(struct btrfs_dev_stats_item, values) + 7543 ((unsigned long)ptr) + (index * sizeof(u64)), 7544 sizeof(val)); 7545 } 7546 7547 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7548 struct btrfs_path *path) 7549 { 7550 struct btrfs_dev_stats_item *ptr; 7551 struct extent_buffer *eb; 7552 struct btrfs_key key; 7553 int item_size; 7554 int i, ret, slot; 7555 7556 if (!device->fs_info->dev_root) 7557 return 0; 7558 7559 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7560 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7561 key.offset = device->devid; 7562 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7563 if (ret) { 7564 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7565 btrfs_dev_stat_set(device, i, 0); 7566 device->dev_stats_valid = 1; 7567 btrfs_release_path(path); 7568 return ret < 0 ? ret : 0; 7569 } 7570 slot = path->slots[0]; 7571 eb = path->nodes[0]; 7572 item_size = btrfs_item_size(eb, slot); 7573 7574 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7575 7576 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7577 if (item_size >= (1 + i) * sizeof(__le64)) 7578 btrfs_dev_stat_set(device, i, 7579 btrfs_dev_stats_value(eb, ptr, i)); 7580 else 7581 btrfs_dev_stat_set(device, i, 0); 7582 } 7583 7584 device->dev_stats_valid = 1; 7585 btrfs_dev_stat_print_on_load(device); 7586 btrfs_release_path(path); 7587 7588 return 0; 7589 } 7590 7591 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7592 { 7593 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7594 struct btrfs_device *device; 7595 struct btrfs_path *path = NULL; 7596 int ret = 0; 7597 7598 path = btrfs_alloc_path(); 7599 if (!path) 7600 return -ENOMEM; 7601 7602 mutex_lock(&fs_devices->device_list_mutex); 7603 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7604 ret = btrfs_device_init_dev_stats(device, path); 7605 if (ret) 7606 goto out; 7607 } 7608 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7609 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7610 ret = btrfs_device_init_dev_stats(device, path); 7611 if (ret) 7612 goto out; 7613 } 7614 } 7615 out: 7616 mutex_unlock(&fs_devices->device_list_mutex); 7617 7618 btrfs_free_path(path); 7619 return ret; 7620 } 7621 7622 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7623 struct btrfs_device *device) 7624 { 7625 struct btrfs_fs_info *fs_info = trans->fs_info; 7626 struct btrfs_root *dev_root = fs_info->dev_root; 7627 struct btrfs_path *path; 7628 struct btrfs_key key; 7629 struct extent_buffer *eb; 7630 struct btrfs_dev_stats_item *ptr; 7631 int ret; 7632 int i; 7633 7634 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7635 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7636 key.offset = device->devid; 7637 7638 path = btrfs_alloc_path(); 7639 if (!path) 7640 return -ENOMEM; 7641 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7642 if (ret < 0) { 7643 btrfs_warn_in_rcu(fs_info, 7644 "error %d while searching for dev_stats item for device %s", 7645 ret, btrfs_dev_name(device)); 7646 goto out; 7647 } 7648 7649 if (ret == 0 && 7650 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7651 /* need to delete old one and insert a new one */ 7652 ret = btrfs_del_item(trans, dev_root, path); 7653 if (ret != 0) { 7654 btrfs_warn_in_rcu(fs_info, 7655 "delete too small dev_stats item for device %s failed %d", 7656 btrfs_dev_name(device), ret); 7657 goto out; 7658 } 7659 ret = 1; 7660 } 7661 7662 if (ret == 1) { 7663 /* need to insert a new item */ 7664 btrfs_release_path(path); 7665 ret = btrfs_insert_empty_item(trans, dev_root, path, 7666 &key, sizeof(*ptr)); 7667 if (ret < 0) { 7668 btrfs_warn_in_rcu(fs_info, 7669 "insert dev_stats item for device %s failed %d", 7670 btrfs_dev_name(device), ret); 7671 goto out; 7672 } 7673 } 7674 7675 eb = path->nodes[0]; 7676 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7677 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7678 btrfs_set_dev_stats_value(eb, ptr, i, 7679 btrfs_dev_stat_read(device, i)); 7680 btrfs_mark_buffer_dirty(eb); 7681 7682 out: 7683 btrfs_free_path(path); 7684 return ret; 7685 } 7686 7687 /* 7688 * called from commit_transaction. Writes all changed device stats to disk. 7689 */ 7690 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7691 { 7692 struct btrfs_fs_info *fs_info = trans->fs_info; 7693 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7694 struct btrfs_device *device; 7695 int stats_cnt; 7696 int ret = 0; 7697 7698 mutex_lock(&fs_devices->device_list_mutex); 7699 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7700 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7701 if (!device->dev_stats_valid || stats_cnt == 0) 7702 continue; 7703 7704 7705 /* 7706 * There is a LOAD-LOAD control dependency between the value of 7707 * dev_stats_ccnt and updating the on-disk values which requires 7708 * reading the in-memory counters. Such control dependencies 7709 * require explicit read memory barriers. 7710 * 7711 * This memory barriers pairs with smp_mb__before_atomic in 7712 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7713 * barrier implied by atomic_xchg in 7714 * btrfs_dev_stats_read_and_reset 7715 */ 7716 smp_rmb(); 7717 7718 ret = update_dev_stat_item(trans, device); 7719 if (!ret) 7720 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7721 } 7722 mutex_unlock(&fs_devices->device_list_mutex); 7723 7724 return ret; 7725 } 7726 7727 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7728 { 7729 btrfs_dev_stat_inc(dev, index); 7730 7731 if (!dev->dev_stats_valid) 7732 return; 7733 btrfs_err_rl_in_rcu(dev->fs_info, 7734 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7735 btrfs_dev_name(dev), 7736 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7737 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7738 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7739 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7740 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7741 } 7742 7743 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7744 { 7745 int i; 7746 7747 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7748 if (btrfs_dev_stat_read(dev, i) != 0) 7749 break; 7750 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7751 return; /* all values == 0, suppress message */ 7752 7753 btrfs_info_in_rcu(dev->fs_info, 7754 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7755 btrfs_dev_name(dev), 7756 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7757 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7758 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7759 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7760 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7761 } 7762 7763 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7764 struct btrfs_ioctl_get_dev_stats *stats) 7765 { 7766 BTRFS_DEV_LOOKUP_ARGS(args); 7767 struct btrfs_device *dev; 7768 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7769 int i; 7770 7771 mutex_lock(&fs_devices->device_list_mutex); 7772 args.devid = stats->devid; 7773 dev = btrfs_find_device(fs_info->fs_devices, &args); 7774 mutex_unlock(&fs_devices->device_list_mutex); 7775 7776 if (!dev) { 7777 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7778 return -ENODEV; 7779 } else if (!dev->dev_stats_valid) { 7780 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7781 return -ENODEV; 7782 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7783 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7784 if (stats->nr_items > i) 7785 stats->values[i] = 7786 btrfs_dev_stat_read_and_reset(dev, i); 7787 else 7788 btrfs_dev_stat_set(dev, i, 0); 7789 } 7790 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7791 current->comm, task_pid_nr(current)); 7792 } else { 7793 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7794 if (stats->nr_items > i) 7795 stats->values[i] = btrfs_dev_stat_read(dev, i); 7796 } 7797 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7798 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7799 return 0; 7800 } 7801 7802 /* 7803 * Update the size and bytes used for each device where it changed. This is 7804 * delayed since we would otherwise get errors while writing out the 7805 * superblocks. 7806 * 7807 * Must be invoked during transaction commit. 7808 */ 7809 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7810 { 7811 struct btrfs_device *curr, *next; 7812 7813 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7814 7815 if (list_empty(&trans->dev_update_list)) 7816 return; 7817 7818 /* 7819 * We don't need the device_list_mutex here. This list is owned by the 7820 * transaction and the transaction must complete before the device is 7821 * released. 7822 */ 7823 mutex_lock(&trans->fs_info->chunk_mutex); 7824 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7825 post_commit_list) { 7826 list_del_init(&curr->post_commit_list); 7827 curr->commit_total_bytes = curr->disk_total_bytes; 7828 curr->commit_bytes_used = curr->bytes_used; 7829 } 7830 mutex_unlock(&trans->fs_info->chunk_mutex); 7831 } 7832 7833 /* 7834 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7835 */ 7836 int btrfs_bg_type_to_factor(u64 flags) 7837 { 7838 const int index = btrfs_bg_flags_to_raid_index(flags); 7839 7840 return btrfs_raid_array[index].ncopies; 7841 } 7842 7843 7844 7845 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7846 u64 chunk_offset, u64 devid, 7847 u64 physical_offset, u64 physical_len) 7848 { 7849 struct btrfs_dev_lookup_args args = { .devid = devid }; 7850 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7851 struct extent_map *em; 7852 struct map_lookup *map; 7853 struct btrfs_device *dev; 7854 u64 stripe_len; 7855 bool found = false; 7856 int ret = 0; 7857 int i; 7858 7859 read_lock(&em_tree->lock); 7860 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7861 read_unlock(&em_tree->lock); 7862 7863 if (!em) { 7864 btrfs_err(fs_info, 7865 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 7866 physical_offset, devid); 7867 ret = -EUCLEAN; 7868 goto out; 7869 } 7870 7871 map = em->map_lookup; 7872 stripe_len = btrfs_calc_stripe_length(em); 7873 if (physical_len != stripe_len) { 7874 btrfs_err(fs_info, 7875 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 7876 physical_offset, devid, em->start, physical_len, 7877 stripe_len); 7878 ret = -EUCLEAN; 7879 goto out; 7880 } 7881 7882 /* 7883 * Very old mkfs.btrfs (before v4.1) will not respect the reserved 7884 * space. Although kernel can handle it without problem, better to warn 7885 * the users. 7886 */ 7887 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) 7888 btrfs_warn(fs_info, 7889 "devid %llu physical %llu len %llu inside the reserved space", 7890 devid, physical_offset, physical_len); 7891 7892 for (i = 0; i < map->num_stripes; i++) { 7893 if (map->stripes[i].dev->devid == devid && 7894 map->stripes[i].physical == physical_offset) { 7895 found = true; 7896 if (map->verified_stripes >= map->num_stripes) { 7897 btrfs_err(fs_info, 7898 "too many dev extents for chunk %llu found", 7899 em->start); 7900 ret = -EUCLEAN; 7901 goto out; 7902 } 7903 map->verified_stripes++; 7904 break; 7905 } 7906 } 7907 if (!found) { 7908 btrfs_err(fs_info, 7909 "dev extent physical offset %llu devid %llu has no corresponding chunk", 7910 physical_offset, devid); 7911 ret = -EUCLEAN; 7912 } 7913 7914 /* Make sure no dev extent is beyond device boundary */ 7915 dev = btrfs_find_device(fs_info->fs_devices, &args); 7916 if (!dev) { 7917 btrfs_err(fs_info, "failed to find devid %llu", devid); 7918 ret = -EUCLEAN; 7919 goto out; 7920 } 7921 7922 if (physical_offset + physical_len > dev->disk_total_bytes) { 7923 btrfs_err(fs_info, 7924 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 7925 devid, physical_offset, physical_len, 7926 dev->disk_total_bytes); 7927 ret = -EUCLEAN; 7928 goto out; 7929 } 7930 7931 if (dev->zone_info) { 7932 u64 zone_size = dev->zone_info->zone_size; 7933 7934 if (!IS_ALIGNED(physical_offset, zone_size) || 7935 !IS_ALIGNED(physical_len, zone_size)) { 7936 btrfs_err(fs_info, 7937 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 7938 devid, physical_offset, physical_len); 7939 ret = -EUCLEAN; 7940 goto out; 7941 } 7942 } 7943 7944 out: 7945 free_extent_map(em); 7946 return ret; 7947 } 7948 7949 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 7950 { 7951 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7952 struct extent_map *em; 7953 struct rb_node *node; 7954 int ret = 0; 7955 7956 read_lock(&em_tree->lock); 7957 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 7958 em = rb_entry(node, struct extent_map, rb_node); 7959 if (em->map_lookup->num_stripes != 7960 em->map_lookup->verified_stripes) { 7961 btrfs_err(fs_info, 7962 "chunk %llu has missing dev extent, have %d expect %d", 7963 em->start, em->map_lookup->verified_stripes, 7964 em->map_lookup->num_stripes); 7965 ret = -EUCLEAN; 7966 goto out; 7967 } 7968 } 7969 out: 7970 read_unlock(&em_tree->lock); 7971 return ret; 7972 } 7973 7974 /* 7975 * Ensure that all dev extents are mapped to correct chunk, otherwise 7976 * later chunk allocation/free would cause unexpected behavior. 7977 * 7978 * NOTE: This will iterate through the whole device tree, which should be of 7979 * the same size level as the chunk tree. This slightly increases mount time. 7980 */ 7981 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 7982 { 7983 struct btrfs_path *path; 7984 struct btrfs_root *root = fs_info->dev_root; 7985 struct btrfs_key key; 7986 u64 prev_devid = 0; 7987 u64 prev_dev_ext_end = 0; 7988 int ret = 0; 7989 7990 /* 7991 * We don't have a dev_root because we mounted with ignorebadroots and 7992 * failed to load the root, so we want to skip the verification in this 7993 * case for sure. 7994 * 7995 * However if the dev root is fine, but the tree itself is corrupted 7996 * we'd still fail to mount. This verification is only to make sure 7997 * writes can happen safely, so instead just bypass this check 7998 * completely in the case of IGNOREBADROOTS. 7999 */ 8000 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8001 return 0; 8002 8003 key.objectid = 1; 8004 key.type = BTRFS_DEV_EXTENT_KEY; 8005 key.offset = 0; 8006 8007 path = btrfs_alloc_path(); 8008 if (!path) 8009 return -ENOMEM; 8010 8011 path->reada = READA_FORWARD; 8012 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8013 if (ret < 0) 8014 goto out; 8015 8016 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8017 ret = btrfs_next_leaf(root, path); 8018 if (ret < 0) 8019 goto out; 8020 /* No dev extents at all? Not good */ 8021 if (ret > 0) { 8022 ret = -EUCLEAN; 8023 goto out; 8024 } 8025 } 8026 while (1) { 8027 struct extent_buffer *leaf = path->nodes[0]; 8028 struct btrfs_dev_extent *dext; 8029 int slot = path->slots[0]; 8030 u64 chunk_offset; 8031 u64 physical_offset; 8032 u64 physical_len; 8033 u64 devid; 8034 8035 btrfs_item_key_to_cpu(leaf, &key, slot); 8036 if (key.type != BTRFS_DEV_EXTENT_KEY) 8037 break; 8038 devid = key.objectid; 8039 physical_offset = key.offset; 8040 8041 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8042 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8043 physical_len = btrfs_dev_extent_length(leaf, dext); 8044 8045 /* Check if this dev extent overlaps with the previous one */ 8046 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8047 btrfs_err(fs_info, 8048 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8049 devid, physical_offset, prev_dev_ext_end); 8050 ret = -EUCLEAN; 8051 goto out; 8052 } 8053 8054 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8055 physical_offset, physical_len); 8056 if (ret < 0) 8057 goto out; 8058 prev_devid = devid; 8059 prev_dev_ext_end = physical_offset + physical_len; 8060 8061 ret = btrfs_next_item(root, path); 8062 if (ret < 0) 8063 goto out; 8064 if (ret > 0) { 8065 ret = 0; 8066 break; 8067 } 8068 } 8069 8070 /* Ensure all chunks have corresponding dev extents */ 8071 ret = verify_chunk_dev_extent_mapping(fs_info); 8072 out: 8073 btrfs_free_path(path); 8074 return ret; 8075 } 8076 8077 /* 8078 * Check whether the given block group or device is pinned by any inode being 8079 * used as a swapfile. 8080 */ 8081 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8082 { 8083 struct btrfs_swapfile_pin *sp; 8084 struct rb_node *node; 8085 8086 spin_lock(&fs_info->swapfile_pins_lock); 8087 node = fs_info->swapfile_pins.rb_node; 8088 while (node) { 8089 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8090 if (ptr < sp->ptr) 8091 node = node->rb_left; 8092 else if (ptr > sp->ptr) 8093 node = node->rb_right; 8094 else 8095 break; 8096 } 8097 spin_unlock(&fs_info->swapfile_pins_lock); 8098 return node != NULL; 8099 } 8100 8101 static int relocating_repair_kthread(void *data) 8102 { 8103 struct btrfs_block_group *cache = data; 8104 struct btrfs_fs_info *fs_info = cache->fs_info; 8105 u64 target; 8106 int ret = 0; 8107 8108 target = cache->start; 8109 btrfs_put_block_group(cache); 8110 8111 sb_start_write(fs_info->sb); 8112 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8113 btrfs_info(fs_info, 8114 "zoned: skip relocating block group %llu to repair: EBUSY", 8115 target); 8116 sb_end_write(fs_info->sb); 8117 return -EBUSY; 8118 } 8119 8120 mutex_lock(&fs_info->reclaim_bgs_lock); 8121 8122 /* Ensure block group still exists */ 8123 cache = btrfs_lookup_block_group(fs_info, target); 8124 if (!cache) 8125 goto out; 8126 8127 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) 8128 goto out; 8129 8130 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8131 if (ret < 0) 8132 goto out; 8133 8134 btrfs_info(fs_info, 8135 "zoned: relocating block group %llu to repair IO failure", 8136 target); 8137 ret = btrfs_relocate_chunk(fs_info, target); 8138 8139 out: 8140 if (cache) 8141 btrfs_put_block_group(cache); 8142 mutex_unlock(&fs_info->reclaim_bgs_lock); 8143 btrfs_exclop_finish(fs_info); 8144 sb_end_write(fs_info->sb); 8145 8146 return ret; 8147 } 8148 8149 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8150 { 8151 struct btrfs_block_group *cache; 8152 8153 if (!btrfs_is_zoned(fs_info)) 8154 return false; 8155 8156 /* Do not attempt to repair in degraded state */ 8157 if (btrfs_test_opt(fs_info, DEGRADED)) 8158 return true; 8159 8160 cache = btrfs_lookup_block_group(fs_info, logical); 8161 if (!cache) 8162 return true; 8163 8164 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { 8165 btrfs_put_block_group(cache); 8166 return true; 8167 } 8168 8169 kthread_run(relocating_repair_kthread, cache, 8170 "btrfs-relocating-repair"); 8171 8172 return true; 8173 } 8174