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/bio.h> 9 #include <linux/slab.h> 10 #include <linux/blkdev.h> 11 #include <linux/ratelimit.h> 12 #include <linux/kthread.h> 13 #include <linux/raid/pq.h> 14 #include <linux/semaphore.h> 15 #include <linux/uuid.h> 16 #include <linux/list_sort.h> 17 #include <linux/namei.h> 18 #include "misc.h" 19 #include "ctree.h" 20 #include "extent_map.h" 21 #include "disk-io.h" 22 #include "transaction.h" 23 #include "print-tree.h" 24 #include "volumes.h" 25 #include "raid56.h" 26 #include "async-thread.h" 27 #include "check-integrity.h" 28 #include "rcu-string.h" 29 #include "dev-replace.h" 30 #include "sysfs.h" 31 #include "tree-checker.h" 32 #include "space-info.h" 33 #include "block-group.h" 34 #include "discard.h" 35 #include "zoned.h" 36 #include "fs.h" 37 #include "accessors.h" 38 #include "uuid-tree.h" 39 #include "ioctl.h" 40 #include "relocation.h" 41 #include "scrub.h" 42 #include "super.h" 43 44 static struct bio_set btrfs_bioset; 45 46 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ 47 BTRFS_BLOCK_GROUP_RAID10 | \ 48 BTRFS_BLOCK_GROUP_RAID56_MASK) 49 50 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { 51 [BTRFS_RAID_RAID10] = { 52 .sub_stripes = 2, 53 .dev_stripes = 1, 54 .devs_max = 0, /* 0 == as many as possible */ 55 .devs_min = 2, 56 .tolerated_failures = 1, 57 .devs_increment = 2, 58 .ncopies = 2, 59 .nparity = 0, 60 .raid_name = "raid10", 61 .bg_flag = BTRFS_BLOCK_GROUP_RAID10, 62 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, 63 }, 64 [BTRFS_RAID_RAID1] = { 65 .sub_stripes = 1, 66 .dev_stripes = 1, 67 .devs_max = 2, 68 .devs_min = 2, 69 .tolerated_failures = 1, 70 .devs_increment = 2, 71 .ncopies = 2, 72 .nparity = 0, 73 .raid_name = "raid1", 74 .bg_flag = BTRFS_BLOCK_GROUP_RAID1, 75 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, 76 }, 77 [BTRFS_RAID_RAID1C3] = { 78 .sub_stripes = 1, 79 .dev_stripes = 1, 80 .devs_max = 3, 81 .devs_min = 3, 82 .tolerated_failures = 2, 83 .devs_increment = 3, 84 .ncopies = 3, 85 .nparity = 0, 86 .raid_name = "raid1c3", 87 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, 88 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, 89 }, 90 [BTRFS_RAID_RAID1C4] = { 91 .sub_stripes = 1, 92 .dev_stripes = 1, 93 .devs_max = 4, 94 .devs_min = 4, 95 .tolerated_failures = 3, 96 .devs_increment = 4, 97 .ncopies = 4, 98 .nparity = 0, 99 .raid_name = "raid1c4", 100 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, 101 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, 102 }, 103 [BTRFS_RAID_DUP] = { 104 .sub_stripes = 1, 105 .dev_stripes = 2, 106 .devs_max = 1, 107 .devs_min = 1, 108 .tolerated_failures = 0, 109 .devs_increment = 1, 110 .ncopies = 2, 111 .nparity = 0, 112 .raid_name = "dup", 113 .bg_flag = BTRFS_BLOCK_GROUP_DUP, 114 .mindev_error = 0, 115 }, 116 [BTRFS_RAID_RAID0] = { 117 .sub_stripes = 1, 118 .dev_stripes = 1, 119 .devs_max = 0, 120 .devs_min = 1, 121 .tolerated_failures = 0, 122 .devs_increment = 1, 123 .ncopies = 1, 124 .nparity = 0, 125 .raid_name = "raid0", 126 .bg_flag = BTRFS_BLOCK_GROUP_RAID0, 127 .mindev_error = 0, 128 }, 129 [BTRFS_RAID_SINGLE] = { 130 .sub_stripes = 1, 131 .dev_stripes = 1, 132 .devs_max = 1, 133 .devs_min = 1, 134 .tolerated_failures = 0, 135 .devs_increment = 1, 136 .ncopies = 1, 137 .nparity = 0, 138 .raid_name = "single", 139 .bg_flag = 0, 140 .mindev_error = 0, 141 }, 142 [BTRFS_RAID_RAID5] = { 143 .sub_stripes = 1, 144 .dev_stripes = 1, 145 .devs_max = 0, 146 .devs_min = 2, 147 .tolerated_failures = 1, 148 .devs_increment = 1, 149 .ncopies = 1, 150 .nparity = 1, 151 .raid_name = "raid5", 152 .bg_flag = BTRFS_BLOCK_GROUP_RAID5, 153 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, 154 }, 155 [BTRFS_RAID_RAID6] = { 156 .sub_stripes = 1, 157 .dev_stripes = 1, 158 .devs_max = 0, 159 .devs_min = 3, 160 .tolerated_failures = 2, 161 .devs_increment = 1, 162 .ncopies = 1, 163 .nparity = 2, 164 .raid_name = "raid6", 165 .bg_flag = BTRFS_BLOCK_GROUP_RAID6, 166 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, 167 }, 168 }; 169 170 /* 171 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 172 * can be used as index to access btrfs_raid_array[]. 173 */ 174 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) 175 { 176 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK); 177 178 if (!profile) 179 return BTRFS_RAID_SINGLE; 180 181 return BTRFS_BG_FLAG_TO_INDEX(profile); 182 } 183 184 const char *btrfs_bg_type_to_raid_name(u64 flags) 185 { 186 const int index = btrfs_bg_flags_to_raid_index(flags); 187 188 if (index >= BTRFS_NR_RAID_TYPES) 189 return NULL; 190 191 return btrfs_raid_array[index].raid_name; 192 } 193 194 int btrfs_nr_parity_stripes(u64 type) 195 { 196 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type); 197 198 return btrfs_raid_array[index].nparity; 199 } 200 201 /* 202 * Fill @buf with textual description of @bg_flags, no more than @size_buf 203 * bytes including terminating null byte. 204 */ 205 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) 206 { 207 int i; 208 int ret; 209 char *bp = buf; 210 u64 flags = bg_flags; 211 u32 size_bp = size_buf; 212 213 if (!flags) { 214 strcpy(bp, "NONE"); 215 return; 216 } 217 218 #define DESCRIBE_FLAG(flag, desc) \ 219 do { \ 220 if (flags & (flag)) { \ 221 ret = snprintf(bp, size_bp, "%s|", (desc)); \ 222 if (ret < 0 || ret >= size_bp) \ 223 goto out_overflow; \ 224 size_bp -= ret; \ 225 bp += ret; \ 226 flags &= ~(flag); \ 227 } \ 228 } while (0) 229 230 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); 231 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); 232 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); 233 234 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); 235 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) 236 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, 237 btrfs_raid_array[i].raid_name); 238 #undef DESCRIBE_FLAG 239 240 if (flags) { 241 ret = snprintf(bp, size_bp, "0x%llx|", flags); 242 size_bp -= ret; 243 } 244 245 if (size_bp < size_buf) 246 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ 247 248 /* 249 * The text is trimmed, it's up to the caller to provide sufficiently 250 * large buffer 251 */ 252 out_overflow:; 253 } 254 255 static int init_first_rw_device(struct btrfs_trans_handle *trans); 256 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); 257 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); 258 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 259 enum btrfs_map_op op, u64 logical, u64 *length, 260 struct btrfs_io_context **bioc_ret, 261 struct btrfs_io_stripe *smap, 262 int *mirror_num_ret, int need_raid_map); 263 264 /* 265 * Device locking 266 * ============== 267 * 268 * There are several mutexes that protect manipulation of devices and low-level 269 * structures like chunks but not block groups, extents or files 270 * 271 * uuid_mutex (global lock) 272 * ------------------------ 273 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from 274 * the SCAN_DEV ioctl registration or from mount either implicitly (the first 275 * device) or requested by the device= mount option 276 * 277 * the mutex can be very coarse and can cover long-running operations 278 * 279 * protects: updates to fs_devices counters like missing devices, rw devices, 280 * seeding, structure cloning, opening/closing devices at mount/umount time 281 * 282 * global::fs_devs - add, remove, updates to the global list 283 * 284 * does not protect: manipulation of the fs_devices::devices list in general 285 * but in mount context it could be used to exclude list modifications by eg. 286 * scan ioctl 287 * 288 * btrfs_device::name - renames (write side), read is RCU 289 * 290 * fs_devices::device_list_mutex (per-fs, with RCU) 291 * ------------------------------------------------ 292 * protects updates to fs_devices::devices, ie. adding and deleting 293 * 294 * simple list traversal with read-only actions can be done with RCU protection 295 * 296 * may be used to exclude some operations from running concurrently without any 297 * modifications to the list (see write_all_supers) 298 * 299 * Is not required at mount and close times, because our device list is 300 * protected by the uuid_mutex at that point. 301 * 302 * balance_mutex 303 * ------------- 304 * protects balance structures (status, state) and context accessed from 305 * several places (internally, ioctl) 306 * 307 * chunk_mutex 308 * ----------- 309 * protects chunks, adding or removing during allocation, trim or when a new 310 * device is added/removed. Additionally it also protects post_commit_list of 311 * individual devices, since they can be added to the transaction's 312 * post_commit_list only with chunk_mutex held. 313 * 314 * cleaner_mutex 315 * ------------- 316 * a big lock that is held by the cleaner thread and prevents running subvolume 317 * cleaning together with relocation or delayed iputs 318 * 319 * 320 * Lock nesting 321 * ============ 322 * 323 * uuid_mutex 324 * device_list_mutex 325 * chunk_mutex 326 * balance_mutex 327 * 328 * 329 * Exclusive operations 330 * ==================== 331 * 332 * Maintains the exclusivity of the following operations that apply to the 333 * whole filesystem and cannot run in parallel. 334 * 335 * - Balance (*) 336 * - Device add 337 * - Device remove 338 * - Device replace (*) 339 * - Resize 340 * 341 * The device operations (as above) can be in one of the following states: 342 * 343 * - Running state 344 * - Paused state 345 * - Completed state 346 * 347 * Only device operations marked with (*) can go into the Paused state for the 348 * following reasons: 349 * 350 * - ioctl (only Balance can be Paused through ioctl) 351 * - filesystem remounted as read-only 352 * - filesystem unmounted and mounted as read-only 353 * - system power-cycle and filesystem mounted as read-only 354 * - filesystem or device errors leading to forced read-only 355 * 356 * The status of exclusive operation is set and cleared atomically. 357 * During the course of Paused state, fs_info::exclusive_operation remains set. 358 * A device operation in Paused or Running state can be canceled or resumed 359 * either by ioctl (Balance only) or when remounted as read-write. 360 * The exclusive status is cleared when the device operation is canceled or 361 * completed. 362 */ 363 364 DEFINE_MUTEX(uuid_mutex); 365 static LIST_HEAD(fs_uuids); 366 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) 367 { 368 return &fs_uuids; 369 } 370 371 /* 372 * alloc_fs_devices - allocate struct btrfs_fs_devices 373 * @fsid: if not NULL, copy the UUID to fs_devices::fsid 374 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid 375 * 376 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). 377 * The returned struct is not linked onto any lists and can be destroyed with 378 * kfree() right away. 379 */ 380 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, 381 const u8 *metadata_fsid) 382 { 383 struct btrfs_fs_devices *fs_devs; 384 385 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 386 if (!fs_devs) 387 return ERR_PTR(-ENOMEM); 388 389 mutex_init(&fs_devs->device_list_mutex); 390 391 INIT_LIST_HEAD(&fs_devs->devices); 392 INIT_LIST_HEAD(&fs_devs->alloc_list); 393 INIT_LIST_HEAD(&fs_devs->fs_list); 394 INIT_LIST_HEAD(&fs_devs->seed_list); 395 if (fsid) 396 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 397 398 if (metadata_fsid) 399 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); 400 else if (fsid) 401 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); 402 403 return fs_devs; 404 } 405 406 void btrfs_free_device(struct btrfs_device *device) 407 { 408 WARN_ON(!list_empty(&device->post_commit_list)); 409 rcu_string_free(device->name); 410 extent_io_tree_release(&device->alloc_state); 411 btrfs_destroy_dev_zone_info(device); 412 kfree(device); 413 } 414 415 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 416 { 417 struct btrfs_device *device; 418 WARN_ON(fs_devices->opened); 419 while (!list_empty(&fs_devices->devices)) { 420 device = list_entry(fs_devices->devices.next, 421 struct btrfs_device, dev_list); 422 list_del(&device->dev_list); 423 btrfs_free_device(device); 424 } 425 kfree(fs_devices); 426 } 427 428 void __exit btrfs_cleanup_fs_uuids(void) 429 { 430 struct btrfs_fs_devices *fs_devices; 431 432 while (!list_empty(&fs_uuids)) { 433 fs_devices = list_entry(fs_uuids.next, 434 struct btrfs_fs_devices, fs_list); 435 list_del(&fs_devices->fs_list); 436 free_fs_devices(fs_devices); 437 } 438 } 439 440 static noinline struct btrfs_fs_devices *find_fsid( 441 const u8 *fsid, const u8 *metadata_fsid) 442 { 443 struct btrfs_fs_devices *fs_devices; 444 445 ASSERT(fsid); 446 447 /* Handle non-split brain cases */ 448 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 449 if (metadata_fsid) { 450 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 451 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 452 BTRFS_FSID_SIZE) == 0) 453 return fs_devices; 454 } else { 455 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 456 return fs_devices; 457 } 458 } 459 return NULL; 460 } 461 462 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 463 struct btrfs_super_block *disk_super) 464 { 465 466 struct btrfs_fs_devices *fs_devices; 467 468 /* 469 * Handle scanned device having completed its fsid change but 470 * belonging to a fs_devices that was created by first scanning 471 * a device which didn't have its fsid/metadata_uuid changed 472 * at all and the CHANGING_FSID_V2 flag set. 473 */ 474 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 475 if (fs_devices->fsid_change && 476 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 477 BTRFS_FSID_SIZE) == 0 && 478 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 479 BTRFS_FSID_SIZE) == 0) { 480 return fs_devices; 481 } 482 } 483 /* 484 * Handle scanned device having completed its fsid change but 485 * belonging to a fs_devices that was created by a device that 486 * has an outdated pair of fsid/metadata_uuid and 487 * CHANGING_FSID_V2 flag set. 488 */ 489 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 490 if (fs_devices->fsid_change && 491 memcmp(fs_devices->metadata_uuid, 492 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 493 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 494 BTRFS_FSID_SIZE) == 0) { 495 return fs_devices; 496 } 497 } 498 499 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 500 } 501 502 503 static int 504 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 505 int flush, struct block_device **bdev, 506 struct btrfs_super_block **disk_super) 507 { 508 int ret; 509 510 *bdev = blkdev_get_by_path(device_path, flags, holder); 511 512 if (IS_ERR(*bdev)) { 513 ret = PTR_ERR(*bdev); 514 goto error; 515 } 516 517 if (flush) 518 sync_blockdev(*bdev); 519 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 520 if (ret) { 521 blkdev_put(*bdev, flags); 522 goto error; 523 } 524 invalidate_bdev(*bdev); 525 *disk_super = btrfs_read_dev_super(*bdev); 526 if (IS_ERR(*disk_super)) { 527 ret = PTR_ERR(*disk_super); 528 blkdev_put(*bdev, flags); 529 goto error; 530 } 531 532 return 0; 533 534 error: 535 *bdev = NULL; 536 return ret; 537 } 538 539 /* 540 * Search and remove all stale devices (which are not mounted). When both 541 * inputs are NULL, it will search and release all stale devices. 542 * 543 * @devt: Optional. When provided will it release all unmounted devices 544 * matching this devt only. 545 * @skip_device: Optional. Will skip this device when searching for the stale 546 * devices. 547 * 548 * Return: 0 for success or if @devt is 0. 549 * -EBUSY if @devt is a mounted device. 550 * -ENOENT if @devt does not match any device in the list. 551 */ 552 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) 553 { 554 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 555 struct btrfs_device *device, *tmp_device; 556 int ret = 0; 557 558 lockdep_assert_held(&uuid_mutex); 559 560 if (devt) 561 ret = -ENOENT; 562 563 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 564 565 mutex_lock(&fs_devices->device_list_mutex); 566 list_for_each_entry_safe(device, tmp_device, 567 &fs_devices->devices, dev_list) { 568 if (skip_device && skip_device == device) 569 continue; 570 if (devt && devt != device->devt) 571 continue; 572 if (fs_devices->opened) { 573 /* for an already deleted device return 0 */ 574 if (devt && ret != 0) 575 ret = -EBUSY; 576 break; 577 } 578 579 /* delete the stale device */ 580 fs_devices->num_devices--; 581 list_del(&device->dev_list); 582 btrfs_free_device(device); 583 584 ret = 0; 585 } 586 mutex_unlock(&fs_devices->device_list_mutex); 587 588 if (fs_devices->num_devices == 0) { 589 btrfs_sysfs_remove_fsid(fs_devices); 590 list_del(&fs_devices->fs_list); 591 free_fs_devices(fs_devices); 592 } 593 } 594 595 return ret; 596 } 597 598 /* 599 * This is only used on mount, and we are protected from competing things 600 * messing with our fs_devices by the uuid_mutex, thus we do not need the 601 * fs_devices->device_list_mutex here. 602 */ 603 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 604 struct btrfs_device *device, fmode_t flags, 605 void *holder) 606 { 607 struct block_device *bdev; 608 struct btrfs_super_block *disk_super; 609 u64 devid; 610 int ret; 611 612 if (device->bdev) 613 return -EINVAL; 614 if (!device->name) 615 return -EINVAL; 616 617 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 618 &bdev, &disk_super); 619 if (ret) 620 return ret; 621 622 devid = btrfs_stack_device_id(&disk_super->dev_item); 623 if (devid != device->devid) 624 goto error_free_page; 625 626 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 627 goto error_free_page; 628 629 device->generation = btrfs_super_generation(disk_super); 630 631 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 632 if (btrfs_super_incompat_flags(disk_super) & 633 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 634 pr_err( 635 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 636 goto error_free_page; 637 } 638 639 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 640 fs_devices->seeding = true; 641 } else { 642 if (bdev_read_only(bdev)) 643 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 644 else 645 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 646 } 647 648 if (!bdev_nonrot(bdev)) 649 fs_devices->rotating = true; 650 651 if (bdev_max_discard_sectors(bdev)) 652 fs_devices->discardable = true; 653 654 device->bdev = bdev; 655 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 656 device->mode = flags; 657 658 fs_devices->open_devices++; 659 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 660 device->devid != BTRFS_DEV_REPLACE_DEVID) { 661 fs_devices->rw_devices++; 662 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 663 } 664 btrfs_release_disk_super(disk_super); 665 666 return 0; 667 668 error_free_page: 669 btrfs_release_disk_super(disk_super); 670 blkdev_put(bdev, flags); 671 672 return -EINVAL; 673 } 674 675 /* 676 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 677 * being created with a disk that has already completed its fsid change. Such 678 * disk can belong to an fs which has its FSID changed or to one which doesn't. 679 * Handle both cases here. 680 */ 681 static struct btrfs_fs_devices *find_fsid_inprogress( 682 struct btrfs_super_block *disk_super) 683 { 684 struct btrfs_fs_devices *fs_devices; 685 686 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 687 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 688 BTRFS_FSID_SIZE) != 0 && 689 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 690 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 691 return fs_devices; 692 } 693 } 694 695 return find_fsid(disk_super->fsid, NULL); 696 } 697 698 699 static struct btrfs_fs_devices *find_fsid_changed( 700 struct btrfs_super_block *disk_super) 701 { 702 struct btrfs_fs_devices *fs_devices; 703 704 /* 705 * Handles the case where scanned device is part of an fs that had 706 * multiple successful changes of FSID but currently device didn't 707 * observe it. Meaning our fsid will be different than theirs. We need 708 * to handle two subcases : 709 * 1 - The fs still continues to have different METADATA/FSID uuids. 710 * 2 - The fs is switched back to its original FSID (METADATA/FSID 711 * are equal). 712 */ 713 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 714 /* Changed UUIDs */ 715 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 716 BTRFS_FSID_SIZE) != 0 && 717 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 718 BTRFS_FSID_SIZE) == 0 && 719 memcmp(fs_devices->fsid, disk_super->fsid, 720 BTRFS_FSID_SIZE) != 0) 721 return fs_devices; 722 723 /* Unchanged UUIDs */ 724 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 725 BTRFS_FSID_SIZE) == 0 && 726 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 727 BTRFS_FSID_SIZE) == 0) 728 return fs_devices; 729 } 730 731 return NULL; 732 } 733 734 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 735 struct btrfs_super_block *disk_super) 736 { 737 struct btrfs_fs_devices *fs_devices; 738 739 /* 740 * Handle the case where the scanned device is part of an fs whose last 741 * metadata UUID change reverted it to the original FSID. At the same 742 * time * fs_devices was first created by another constitutent device 743 * which didn't fully observe the operation. This results in an 744 * btrfs_fs_devices created with metadata/fsid different AND 745 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 746 * fs_devices equal to the FSID of the disk. 747 */ 748 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 749 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 750 BTRFS_FSID_SIZE) != 0 && 751 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 752 BTRFS_FSID_SIZE) == 0 && 753 fs_devices->fsid_change) 754 return fs_devices; 755 } 756 757 return NULL; 758 } 759 /* 760 * Add new device to list of registered devices 761 * 762 * Returns: 763 * device pointer which was just added or updated when successful 764 * error pointer when failed 765 */ 766 static noinline struct btrfs_device *device_list_add(const char *path, 767 struct btrfs_super_block *disk_super, 768 bool *new_device_added) 769 { 770 struct btrfs_device *device; 771 struct btrfs_fs_devices *fs_devices = NULL; 772 struct rcu_string *name; 773 u64 found_transid = btrfs_super_generation(disk_super); 774 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 775 dev_t path_devt; 776 int error; 777 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 778 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 779 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 780 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 781 782 error = lookup_bdev(path, &path_devt); 783 if (error) 784 return ERR_PTR(error); 785 786 if (fsid_change_in_progress) { 787 if (!has_metadata_uuid) 788 fs_devices = find_fsid_inprogress(disk_super); 789 else 790 fs_devices = find_fsid_changed(disk_super); 791 } else if (has_metadata_uuid) { 792 fs_devices = find_fsid_with_metadata_uuid(disk_super); 793 } else { 794 fs_devices = find_fsid_reverted_metadata(disk_super); 795 if (!fs_devices) 796 fs_devices = find_fsid(disk_super->fsid, NULL); 797 } 798 799 800 if (!fs_devices) { 801 if (has_metadata_uuid) 802 fs_devices = alloc_fs_devices(disk_super->fsid, 803 disk_super->metadata_uuid); 804 else 805 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 806 807 if (IS_ERR(fs_devices)) 808 return ERR_CAST(fs_devices); 809 810 fs_devices->fsid_change = fsid_change_in_progress; 811 812 mutex_lock(&fs_devices->device_list_mutex); 813 list_add(&fs_devices->fs_list, &fs_uuids); 814 815 device = NULL; 816 } else { 817 struct btrfs_dev_lookup_args args = { 818 .devid = devid, 819 .uuid = disk_super->dev_item.uuid, 820 }; 821 822 mutex_lock(&fs_devices->device_list_mutex); 823 device = btrfs_find_device(fs_devices, &args); 824 825 /* 826 * If this disk has been pulled into an fs devices created by 827 * a device which had the CHANGING_FSID_V2 flag then replace the 828 * metadata_uuid/fsid values of the fs_devices. 829 */ 830 if (fs_devices->fsid_change && 831 found_transid > fs_devices->latest_generation) { 832 memcpy(fs_devices->fsid, disk_super->fsid, 833 BTRFS_FSID_SIZE); 834 835 if (has_metadata_uuid) 836 memcpy(fs_devices->metadata_uuid, 837 disk_super->metadata_uuid, 838 BTRFS_FSID_SIZE); 839 else 840 memcpy(fs_devices->metadata_uuid, 841 disk_super->fsid, BTRFS_FSID_SIZE); 842 843 fs_devices->fsid_change = false; 844 } 845 } 846 847 if (!device) { 848 if (fs_devices->opened) { 849 mutex_unlock(&fs_devices->device_list_mutex); 850 return ERR_PTR(-EBUSY); 851 } 852 853 device = btrfs_alloc_device(NULL, &devid, 854 disk_super->dev_item.uuid); 855 if (IS_ERR(device)) { 856 mutex_unlock(&fs_devices->device_list_mutex); 857 /* we can safely leave the fs_devices entry around */ 858 return device; 859 } 860 861 name = rcu_string_strdup(path, GFP_NOFS); 862 if (!name) { 863 btrfs_free_device(device); 864 mutex_unlock(&fs_devices->device_list_mutex); 865 return ERR_PTR(-ENOMEM); 866 } 867 rcu_assign_pointer(device->name, name); 868 device->devt = path_devt; 869 870 list_add_rcu(&device->dev_list, &fs_devices->devices); 871 fs_devices->num_devices++; 872 873 device->fs_devices = fs_devices; 874 *new_device_added = true; 875 876 if (disk_super->label[0]) 877 pr_info( 878 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 879 disk_super->label, devid, found_transid, path, 880 current->comm, task_pid_nr(current)); 881 else 882 pr_info( 883 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 884 disk_super->fsid, devid, found_transid, path, 885 current->comm, task_pid_nr(current)); 886 887 } else if (!device->name || strcmp(device->name->str, path)) { 888 /* 889 * When FS is already mounted. 890 * 1. If you are here and if the device->name is NULL that 891 * means this device was missing at time of FS mount. 892 * 2. If you are here and if the device->name is different 893 * from 'path' that means either 894 * a. The same device disappeared and reappeared with 895 * different name. or 896 * b. The missing-disk-which-was-replaced, has 897 * reappeared now. 898 * 899 * We must allow 1 and 2a above. But 2b would be a spurious 900 * and unintentional. 901 * 902 * Further in case of 1 and 2a above, the disk at 'path' 903 * would have missed some transaction when it was away and 904 * in case of 2a the stale bdev has to be updated as well. 905 * 2b must not be allowed at all time. 906 */ 907 908 /* 909 * For now, we do allow update to btrfs_fs_device through the 910 * btrfs dev scan cli after FS has been mounted. We're still 911 * tracking a problem where systems fail mount by subvolume id 912 * when we reject replacement on a mounted FS. 913 */ 914 if (!fs_devices->opened && found_transid < device->generation) { 915 /* 916 * That is if the FS is _not_ mounted and if you 917 * are here, that means there is more than one 918 * disk with same uuid and devid.We keep the one 919 * with larger generation number or the last-in if 920 * generation are equal. 921 */ 922 mutex_unlock(&fs_devices->device_list_mutex); 923 return ERR_PTR(-EEXIST); 924 } 925 926 /* 927 * We are going to replace the device path for a given devid, 928 * make sure it's the same device if the device is mounted 929 * 930 * NOTE: the device->fs_info may not be reliable here so pass 931 * in a NULL to message helpers instead. This avoids a possible 932 * use-after-free when the fs_info and fs_info->sb are already 933 * torn down. 934 */ 935 if (device->bdev) { 936 if (device->devt != path_devt) { 937 mutex_unlock(&fs_devices->device_list_mutex); 938 btrfs_warn_in_rcu(NULL, 939 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 940 path, devid, found_transid, 941 current->comm, 942 task_pid_nr(current)); 943 return ERR_PTR(-EEXIST); 944 } 945 btrfs_info_in_rcu(NULL, 946 "devid %llu device path %s changed to %s scanned by %s (%d)", 947 devid, rcu_str_deref(device->name), 948 path, current->comm, 949 task_pid_nr(current)); 950 } 951 952 name = rcu_string_strdup(path, GFP_NOFS); 953 if (!name) { 954 mutex_unlock(&fs_devices->device_list_mutex); 955 return ERR_PTR(-ENOMEM); 956 } 957 rcu_string_free(device->name); 958 rcu_assign_pointer(device->name, name); 959 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 960 fs_devices->missing_devices--; 961 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 962 } 963 device->devt = path_devt; 964 } 965 966 /* 967 * Unmount does not free the btrfs_device struct but would zero 968 * generation along with most of the other members. So just update 969 * it back. We need it to pick the disk with largest generation 970 * (as above). 971 */ 972 if (!fs_devices->opened) { 973 device->generation = found_transid; 974 fs_devices->latest_generation = max_t(u64, found_transid, 975 fs_devices->latest_generation); 976 } 977 978 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 979 980 mutex_unlock(&fs_devices->device_list_mutex); 981 return device; 982 } 983 984 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 985 { 986 struct btrfs_fs_devices *fs_devices; 987 struct btrfs_device *device; 988 struct btrfs_device *orig_dev; 989 int ret = 0; 990 991 lockdep_assert_held(&uuid_mutex); 992 993 fs_devices = alloc_fs_devices(orig->fsid, NULL); 994 if (IS_ERR(fs_devices)) 995 return fs_devices; 996 997 fs_devices->total_devices = orig->total_devices; 998 999 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 1000 struct rcu_string *name; 1001 1002 device = btrfs_alloc_device(NULL, &orig_dev->devid, 1003 orig_dev->uuid); 1004 if (IS_ERR(device)) { 1005 ret = PTR_ERR(device); 1006 goto error; 1007 } 1008 1009 /* 1010 * This is ok to do without rcu read locked because we hold the 1011 * uuid mutex so nothing we touch in here is going to disappear. 1012 */ 1013 if (orig_dev->name) { 1014 name = rcu_string_strdup(orig_dev->name->str, 1015 GFP_KERNEL); 1016 if (!name) { 1017 btrfs_free_device(device); 1018 ret = -ENOMEM; 1019 goto error; 1020 } 1021 rcu_assign_pointer(device->name, name); 1022 } 1023 1024 if (orig_dev->zone_info) { 1025 struct btrfs_zoned_device_info *zone_info; 1026 1027 zone_info = btrfs_clone_dev_zone_info(orig_dev); 1028 if (!zone_info) { 1029 btrfs_free_device(device); 1030 ret = -ENOMEM; 1031 goto error; 1032 } 1033 device->zone_info = zone_info; 1034 } 1035 1036 list_add(&device->dev_list, &fs_devices->devices); 1037 device->fs_devices = fs_devices; 1038 fs_devices->num_devices++; 1039 } 1040 return fs_devices; 1041 error: 1042 free_fs_devices(fs_devices); 1043 return ERR_PTR(ret); 1044 } 1045 1046 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1047 struct btrfs_device **latest_dev) 1048 { 1049 struct btrfs_device *device, *next; 1050 1051 /* This is the initialized path, it is safe to release the devices. */ 1052 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1053 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1054 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1055 &device->dev_state) && 1056 !test_bit(BTRFS_DEV_STATE_MISSING, 1057 &device->dev_state) && 1058 (!*latest_dev || 1059 device->generation > (*latest_dev)->generation)) { 1060 *latest_dev = device; 1061 } 1062 continue; 1063 } 1064 1065 /* 1066 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1067 * in btrfs_init_dev_replace() so just continue. 1068 */ 1069 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1070 continue; 1071 1072 if (device->bdev) { 1073 blkdev_put(device->bdev, device->mode); 1074 device->bdev = NULL; 1075 fs_devices->open_devices--; 1076 } 1077 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1078 list_del_init(&device->dev_alloc_list); 1079 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1080 fs_devices->rw_devices--; 1081 } 1082 list_del_init(&device->dev_list); 1083 fs_devices->num_devices--; 1084 btrfs_free_device(device); 1085 } 1086 1087 } 1088 1089 /* 1090 * After we have read the system tree and know devids belonging to this 1091 * filesystem, remove the device which does not belong there. 1092 */ 1093 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1094 { 1095 struct btrfs_device *latest_dev = NULL; 1096 struct btrfs_fs_devices *seed_dev; 1097 1098 mutex_lock(&uuid_mutex); 1099 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1100 1101 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1102 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1103 1104 fs_devices->latest_dev = latest_dev; 1105 1106 mutex_unlock(&uuid_mutex); 1107 } 1108 1109 static void btrfs_close_bdev(struct btrfs_device *device) 1110 { 1111 if (!device->bdev) 1112 return; 1113 1114 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1115 sync_blockdev(device->bdev); 1116 invalidate_bdev(device->bdev); 1117 } 1118 1119 blkdev_put(device->bdev, device->mode); 1120 } 1121 1122 static void btrfs_close_one_device(struct btrfs_device *device) 1123 { 1124 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1125 1126 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1127 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1128 list_del_init(&device->dev_alloc_list); 1129 fs_devices->rw_devices--; 1130 } 1131 1132 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1133 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1134 1135 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 1136 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 1137 fs_devices->missing_devices--; 1138 } 1139 1140 btrfs_close_bdev(device); 1141 if (device->bdev) { 1142 fs_devices->open_devices--; 1143 device->bdev = NULL; 1144 } 1145 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1146 btrfs_destroy_dev_zone_info(device); 1147 1148 device->fs_info = NULL; 1149 atomic_set(&device->dev_stats_ccnt, 0); 1150 extent_io_tree_release(&device->alloc_state); 1151 1152 /* 1153 * Reset the flush error record. We might have a transient flush error 1154 * in this mount, and if so we aborted the current transaction and set 1155 * the fs to an error state, guaranteeing no super blocks can be further 1156 * committed. However that error might be transient and if we unmount the 1157 * filesystem and mount it again, we should allow the mount to succeed 1158 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1159 * filesystem again we still get flush errors, then we will again abort 1160 * any transaction and set the error state, guaranteeing no commits of 1161 * unsafe super blocks. 1162 */ 1163 device->last_flush_error = 0; 1164 1165 /* Verify the device is back in a pristine state */ 1166 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1167 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1168 ASSERT(list_empty(&device->dev_alloc_list)); 1169 ASSERT(list_empty(&device->post_commit_list)); 1170 } 1171 1172 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1173 { 1174 struct btrfs_device *device, *tmp; 1175 1176 lockdep_assert_held(&uuid_mutex); 1177 1178 if (--fs_devices->opened > 0) 1179 return; 1180 1181 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1182 btrfs_close_one_device(device); 1183 1184 WARN_ON(fs_devices->open_devices); 1185 WARN_ON(fs_devices->rw_devices); 1186 fs_devices->opened = 0; 1187 fs_devices->seeding = false; 1188 fs_devices->fs_info = NULL; 1189 } 1190 1191 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1192 { 1193 LIST_HEAD(list); 1194 struct btrfs_fs_devices *tmp; 1195 1196 mutex_lock(&uuid_mutex); 1197 close_fs_devices(fs_devices); 1198 if (!fs_devices->opened) 1199 list_splice_init(&fs_devices->seed_list, &list); 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 flags |= FMODE_EXCL; 1370 1371 bdev = blkdev_get_by_path(path, flags, holder); 1372 if (IS_ERR(bdev)) 1373 return ERR_CAST(bdev); 1374 1375 bytenr_orig = btrfs_sb_offset(0); 1376 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1377 if (ret) { 1378 device = ERR_PTR(ret); 1379 goto error_bdev_put; 1380 } 1381 1382 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1383 if (IS_ERR(disk_super)) { 1384 device = ERR_CAST(disk_super); 1385 goto error_bdev_put; 1386 } 1387 1388 device = device_list_add(path, disk_super, &new_device_added); 1389 if (!IS_ERR(device) && new_device_added) 1390 btrfs_free_stale_devices(device->devt, device); 1391 1392 btrfs_release_disk_super(disk_super); 1393 1394 error_bdev_put: 1395 blkdev_put(bdev, flags); 1396 1397 return device; 1398 } 1399 1400 /* 1401 * Try to find a chunk that intersects [start, start + len] range and when one 1402 * such is found, record the end of it in *start 1403 */ 1404 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1405 u64 len) 1406 { 1407 u64 physical_start, physical_end; 1408 1409 lockdep_assert_held(&device->fs_info->chunk_mutex); 1410 1411 if (!find_first_extent_bit(&device->alloc_state, *start, 1412 &physical_start, &physical_end, 1413 CHUNK_ALLOCATED, NULL)) { 1414 1415 if (in_range(physical_start, *start, len) || 1416 in_range(*start, physical_start, 1417 physical_end - physical_start)) { 1418 *start = physical_end + 1; 1419 return true; 1420 } 1421 } 1422 return false; 1423 } 1424 1425 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1426 { 1427 switch (device->fs_devices->chunk_alloc_policy) { 1428 case BTRFS_CHUNK_ALLOC_REGULAR: 1429 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); 1430 case BTRFS_CHUNK_ALLOC_ZONED: 1431 /* 1432 * We don't care about the starting region like regular 1433 * allocator, because we anyway use/reserve the first two zones 1434 * for superblock logging. 1435 */ 1436 return ALIGN(start, device->zone_info->zone_size); 1437 default: 1438 BUG(); 1439 } 1440 } 1441 1442 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1443 u64 *hole_start, u64 *hole_size, 1444 u64 num_bytes) 1445 { 1446 u64 zone_size = device->zone_info->zone_size; 1447 u64 pos; 1448 int ret; 1449 bool changed = false; 1450 1451 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1452 1453 while (*hole_size > 0) { 1454 pos = btrfs_find_allocatable_zones(device, *hole_start, 1455 *hole_start + *hole_size, 1456 num_bytes); 1457 if (pos != *hole_start) { 1458 *hole_size = *hole_start + *hole_size - pos; 1459 *hole_start = pos; 1460 changed = true; 1461 if (*hole_size < num_bytes) 1462 break; 1463 } 1464 1465 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1466 1467 /* Range is ensured to be empty */ 1468 if (!ret) 1469 return changed; 1470 1471 /* Given hole range was invalid (outside of device) */ 1472 if (ret == -ERANGE) { 1473 *hole_start += *hole_size; 1474 *hole_size = 0; 1475 return true; 1476 } 1477 1478 *hole_start += zone_size; 1479 *hole_size -= zone_size; 1480 changed = true; 1481 } 1482 1483 return changed; 1484 } 1485 1486 /* 1487 * Check if specified hole is suitable for allocation. 1488 * 1489 * @device: the device which we have the hole 1490 * @hole_start: starting position of the hole 1491 * @hole_size: the size of the hole 1492 * @num_bytes: the size of the free space that we need 1493 * 1494 * This function may modify @hole_start and @hole_size to reflect the suitable 1495 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1496 */ 1497 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1498 u64 *hole_size, u64 num_bytes) 1499 { 1500 bool changed = false; 1501 u64 hole_end = *hole_start + *hole_size; 1502 1503 for (;;) { 1504 /* 1505 * Check before we set max_hole_start, otherwise we could end up 1506 * sending back this offset anyway. 1507 */ 1508 if (contains_pending_extent(device, hole_start, *hole_size)) { 1509 if (hole_end >= *hole_start) 1510 *hole_size = hole_end - *hole_start; 1511 else 1512 *hole_size = 0; 1513 changed = true; 1514 } 1515 1516 switch (device->fs_devices->chunk_alloc_policy) { 1517 case BTRFS_CHUNK_ALLOC_REGULAR: 1518 /* No extra check */ 1519 break; 1520 case BTRFS_CHUNK_ALLOC_ZONED: 1521 if (dev_extent_hole_check_zoned(device, hole_start, 1522 hole_size, num_bytes)) { 1523 changed = true; 1524 /* 1525 * The changed hole can contain pending extent. 1526 * Loop again to check that. 1527 */ 1528 continue; 1529 } 1530 break; 1531 default: 1532 BUG(); 1533 } 1534 1535 break; 1536 } 1537 1538 return changed; 1539 } 1540 1541 /* 1542 * Find free space in the specified device. 1543 * 1544 * @device: the device which we search the free space in 1545 * @num_bytes: the size of the free space that we need 1546 * @search_start: the position from which to begin the search 1547 * @start: store the start of the free space. 1548 * @len: the size of the free space. that we find, or the size 1549 * of the max free space if we don't find suitable free space 1550 * 1551 * This does a pretty simple search, the expectation is that it is called very 1552 * infrequently and that a given device has a small number of extents. 1553 * 1554 * @start is used to store the start of the free space if we find. But if we 1555 * don't find suitable free space, it will be used to store the start position 1556 * of the max free space. 1557 * 1558 * @len is used to store the size of the free space that we find. 1559 * But if we don't find suitable free space, it is used to store the size of 1560 * the max free space. 1561 * 1562 * NOTE: This function will search *commit* root of device tree, and does extra 1563 * check to ensure dev extents are not double allocated. 1564 * This makes the function safe to allocate dev extents but may not report 1565 * correct usable device space, as device extent freed in current transaction 1566 * is not reported as available. 1567 */ 1568 static int find_free_dev_extent_start(struct btrfs_device *device, 1569 u64 num_bytes, u64 search_start, u64 *start, 1570 u64 *len) 1571 { 1572 struct btrfs_fs_info *fs_info = device->fs_info; 1573 struct btrfs_root *root = fs_info->dev_root; 1574 struct btrfs_key key; 1575 struct btrfs_dev_extent *dev_extent; 1576 struct btrfs_path *path; 1577 u64 hole_size; 1578 u64 max_hole_start; 1579 u64 max_hole_size; 1580 u64 extent_end; 1581 u64 search_end = device->total_bytes; 1582 int ret; 1583 int slot; 1584 struct extent_buffer *l; 1585 1586 search_start = dev_extent_search_start(device, search_start); 1587 1588 WARN_ON(device->zone_info && 1589 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1590 1591 path = btrfs_alloc_path(); 1592 if (!path) 1593 return -ENOMEM; 1594 1595 max_hole_start = search_start; 1596 max_hole_size = 0; 1597 1598 again: 1599 if (search_start >= search_end || 1600 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1601 ret = -ENOSPC; 1602 goto out; 1603 } 1604 1605 path->reada = READA_FORWARD; 1606 path->search_commit_root = 1; 1607 path->skip_locking = 1; 1608 1609 key.objectid = device->devid; 1610 key.offset = search_start; 1611 key.type = BTRFS_DEV_EXTENT_KEY; 1612 1613 ret = btrfs_search_backwards(root, &key, path); 1614 if (ret < 0) 1615 goto out; 1616 1617 while (1) { 1618 l = path->nodes[0]; 1619 slot = path->slots[0]; 1620 if (slot >= btrfs_header_nritems(l)) { 1621 ret = btrfs_next_leaf(root, path); 1622 if (ret == 0) 1623 continue; 1624 if (ret < 0) 1625 goto out; 1626 1627 break; 1628 } 1629 btrfs_item_key_to_cpu(l, &key, slot); 1630 1631 if (key.objectid < device->devid) 1632 goto next; 1633 1634 if (key.objectid > device->devid) 1635 break; 1636 1637 if (key.type != BTRFS_DEV_EXTENT_KEY) 1638 goto next; 1639 1640 if (key.offset > search_start) { 1641 hole_size = key.offset - search_start; 1642 dev_extent_hole_check(device, &search_start, &hole_size, 1643 num_bytes); 1644 1645 if (hole_size > max_hole_size) { 1646 max_hole_start = search_start; 1647 max_hole_size = hole_size; 1648 } 1649 1650 /* 1651 * If this free space is greater than which we need, 1652 * it must be the max free space that we have found 1653 * until now, so max_hole_start must point to the start 1654 * of this free space and the length of this free space 1655 * is stored in max_hole_size. Thus, we return 1656 * max_hole_start and max_hole_size and go back to the 1657 * caller. 1658 */ 1659 if (hole_size >= num_bytes) { 1660 ret = 0; 1661 goto out; 1662 } 1663 } 1664 1665 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1666 extent_end = key.offset + btrfs_dev_extent_length(l, 1667 dev_extent); 1668 if (extent_end > search_start) 1669 search_start = extent_end; 1670 next: 1671 path->slots[0]++; 1672 cond_resched(); 1673 } 1674 1675 /* 1676 * At this point, search_start should be the end of 1677 * allocated dev extents, and when shrinking the device, 1678 * search_end may be smaller than search_start. 1679 */ 1680 if (search_end > search_start) { 1681 hole_size = search_end - search_start; 1682 if (dev_extent_hole_check(device, &search_start, &hole_size, 1683 num_bytes)) { 1684 btrfs_release_path(path); 1685 goto again; 1686 } 1687 1688 if (hole_size > max_hole_size) { 1689 max_hole_start = search_start; 1690 max_hole_size = hole_size; 1691 } 1692 } 1693 1694 /* See above. */ 1695 if (max_hole_size < num_bytes) 1696 ret = -ENOSPC; 1697 else 1698 ret = 0; 1699 1700 out: 1701 btrfs_free_path(path); 1702 *start = max_hole_start; 1703 if (len) 1704 *len = max_hole_size; 1705 return ret; 1706 } 1707 1708 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1709 u64 *start, u64 *len) 1710 { 1711 /* FIXME use last free of some kind */ 1712 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1713 } 1714 1715 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1716 struct btrfs_device *device, 1717 u64 start, u64 *dev_extent_len) 1718 { 1719 struct btrfs_fs_info *fs_info = device->fs_info; 1720 struct btrfs_root *root = fs_info->dev_root; 1721 int ret; 1722 struct btrfs_path *path; 1723 struct btrfs_key key; 1724 struct btrfs_key found_key; 1725 struct extent_buffer *leaf = NULL; 1726 struct btrfs_dev_extent *extent = NULL; 1727 1728 path = btrfs_alloc_path(); 1729 if (!path) 1730 return -ENOMEM; 1731 1732 key.objectid = device->devid; 1733 key.offset = start; 1734 key.type = BTRFS_DEV_EXTENT_KEY; 1735 again: 1736 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1737 if (ret > 0) { 1738 ret = btrfs_previous_item(root, path, key.objectid, 1739 BTRFS_DEV_EXTENT_KEY); 1740 if (ret) 1741 goto out; 1742 leaf = path->nodes[0]; 1743 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1744 extent = btrfs_item_ptr(leaf, path->slots[0], 1745 struct btrfs_dev_extent); 1746 BUG_ON(found_key.offset > start || found_key.offset + 1747 btrfs_dev_extent_length(leaf, extent) < start); 1748 key = found_key; 1749 btrfs_release_path(path); 1750 goto again; 1751 } else if (ret == 0) { 1752 leaf = path->nodes[0]; 1753 extent = btrfs_item_ptr(leaf, path->slots[0], 1754 struct btrfs_dev_extent); 1755 } else { 1756 goto out; 1757 } 1758 1759 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1760 1761 ret = btrfs_del_item(trans, root, path); 1762 if (ret == 0) 1763 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1764 out: 1765 btrfs_free_path(path); 1766 return ret; 1767 } 1768 1769 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1770 { 1771 struct extent_map_tree *em_tree; 1772 struct extent_map *em; 1773 struct rb_node *n; 1774 u64 ret = 0; 1775 1776 em_tree = &fs_info->mapping_tree; 1777 read_lock(&em_tree->lock); 1778 n = rb_last(&em_tree->map.rb_root); 1779 if (n) { 1780 em = rb_entry(n, struct extent_map, rb_node); 1781 ret = em->start + em->len; 1782 } 1783 read_unlock(&em_tree->lock); 1784 1785 return ret; 1786 } 1787 1788 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1789 u64 *devid_ret) 1790 { 1791 int ret; 1792 struct btrfs_key key; 1793 struct btrfs_key found_key; 1794 struct btrfs_path *path; 1795 1796 path = btrfs_alloc_path(); 1797 if (!path) 1798 return -ENOMEM; 1799 1800 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1801 key.type = BTRFS_DEV_ITEM_KEY; 1802 key.offset = (u64)-1; 1803 1804 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1805 if (ret < 0) 1806 goto error; 1807 1808 if (ret == 0) { 1809 /* Corruption */ 1810 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1811 ret = -EUCLEAN; 1812 goto error; 1813 } 1814 1815 ret = btrfs_previous_item(fs_info->chunk_root, path, 1816 BTRFS_DEV_ITEMS_OBJECTID, 1817 BTRFS_DEV_ITEM_KEY); 1818 if (ret) { 1819 *devid_ret = 1; 1820 } else { 1821 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1822 path->slots[0]); 1823 *devid_ret = found_key.offset + 1; 1824 } 1825 ret = 0; 1826 error: 1827 btrfs_free_path(path); 1828 return ret; 1829 } 1830 1831 /* 1832 * the device information is stored in the chunk root 1833 * the btrfs_device struct should be fully filled in 1834 */ 1835 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1836 struct btrfs_device *device) 1837 { 1838 int ret; 1839 struct btrfs_path *path; 1840 struct btrfs_dev_item *dev_item; 1841 struct extent_buffer *leaf; 1842 struct btrfs_key key; 1843 unsigned long ptr; 1844 1845 path = btrfs_alloc_path(); 1846 if (!path) 1847 return -ENOMEM; 1848 1849 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1850 key.type = BTRFS_DEV_ITEM_KEY; 1851 key.offset = device->devid; 1852 1853 btrfs_reserve_chunk_metadata(trans, true); 1854 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1855 &key, sizeof(*dev_item)); 1856 btrfs_trans_release_chunk_metadata(trans); 1857 if (ret) 1858 goto out; 1859 1860 leaf = path->nodes[0]; 1861 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1862 1863 btrfs_set_device_id(leaf, dev_item, device->devid); 1864 btrfs_set_device_generation(leaf, dev_item, 0); 1865 btrfs_set_device_type(leaf, dev_item, device->type); 1866 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1867 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1868 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1869 btrfs_set_device_total_bytes(leaf, dev_item, 1870 btrfs_device_get_disk_total_bytes(device)); 1871 btrfs_set_device_bytes_used(leaf, dev_item, 1872 btrfs_device_get_bytes_used(device)); 1873 btrfs_set_device_group(leaf, dev_item, 0); 1874 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1875 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1876 btrfs_set_device_start_offset(leaf, dev_item, 0); 1877 1878 ptr = btrfs_device_uuid(dev_item); 1879 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1880 ptr = btrfs_device_fsid(dev_item); 1881 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1882 ptr, BTRFS_FSID_SIZE); 1883 btrfs_mark_buffer_dirty(leaf); 1884 1885 ret = 0; 1886 out: 1887 btrfs_free_path(path); 1888 return ret; 1889 } 1890 1891 /* 1892 * Function to update ctime/mtime for a given device path. 1893 * Mainly used for ctime/mtime based probe like libblkid. 1894 * 1895 * We don't care about errors here, this is just to be kind to userspace. 1896 */ 1897 static void update_dev_time(const char *device_path) 1898 { 1899 struct path path; 1900 struct timespec64 now; 1901 int ret; 1902 1903 ret = kern_path(device_path, LOOKUP_FOLLOW, &path); 1904 if (ret) 1905 return; 1906 1907 now = current_time(d_inode(path.dentry)); 1908 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); 1909 path_put(&path); 1910 } 1911 1912 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, 1913 struct btrfs_device *device) 1914 { 1915 struct btrfs_root *root = device->fs_info->chunk_root; 1916 int ret; 1917 struct btrfs_path *path; 1918 struct btrfs_key key; 1919 1920 path = btrfs_alloc_path(); 1921 if (!path) 1922 return -ENOMEM; 1923 1924 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1925 key.type = BTRFS_DEV_ITEM_KEY; 1926 key.offset = device->devid; 1927 1928 btrfs_reserve_chunk_metadata(trans, false); 1929 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1930 btrfs_trans_release_chunk_metadata(trans); 1931 if (ret) { 1932 if (ret > 0) 1933 ret = -ENOENT; 1934 goto out; 1935 } 1936 1937 ret = btrfs_del_item(trans, root, path); 1938 out: 1939 btrfs_free_path(path); 1940 return ret; 1941 } 1942 1943 /* 1944 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1945 * filesystem. It's up to the caller to adjust that number regarding eg. device 1946 * replace. 1947 */ 1948 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1949 u64 num_devices) 1950 { 1951 u64 all_avail; 1952 unsigned seq; 1953 int i; 1954 1955 do { 1956 seq = read_seqbegin(&fs_info->profiles_lock); 1957 1958 all_avail = fs_info->avail_data_alloc_bits | 1959 fs_info->avail_system_alloc_bits | 1960 fs_info->avail_metadata_alloc_bits; 1961 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1962 1963 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1964 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1965 continue; 1966 1967 if (num_devices < btrfs_raid_array[i].devs_min) 1968 return btrfs_raid_array[i].mindev_error; 1969 } 1970 1971 return 0; 1972 } 1973 1974 static struct btrfs_device * btrfs_find_next_active_device( 1975 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1976 { 1977 struct btrfs_device *next_device; 1978 1979 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1980 if (next_device != device && 1981 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1982 && next_device->bdev) 1983 return next_device; 1984 } 1985 1986 return NULL; 1987 } 1988 1989 /* 1990 * Helper function to check if the given device is part of s_bdev / latest_dev 1991 * and replace it with the provided or the next active device, in the context 1992 * where this function called, there should be always be another device (or 1993 * this_dev) which is active. 1994 */ 1995 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 1996 struct btrfs_device *next_device) 1997 { 1998 struct btrfs_fs_info *fs_info = device->fs_info; 1999 2000 if (!next_device) 2001 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 2002 device); 2003 ASSERT(next_device); 2004 2005 if (fs_info->sb->s_bdev && 2006 (fs_info->sb->s_bdev == device->bdev)) 2007 fs_info->sb->s_bdev = next_device->bdev; 2008 2009 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2010 fs_info->fs_devices->latest_dev = next_device; 2011 } 2012 2013 /* 2014 * Return btrfs_fs_devices::num_devices excluding the device that's being 2015 * currently replaced. 2016 */ 2017 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2018 { 2019 u64 num_devices = fs_info->fs_devices->num_devices; 2020 2021 down_read(&fs_info->dev_replace.rwsem); 2022 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2023 ASSERT(num_devices > 1); 2024 num_devices--; 2025 } 2026 up_read(&fs_info->dev_replace.rwsem); 2027 2028 return num_devices; 2029 } 2030 2031 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2032 struct block_device *bdev, 2033 const char *device_path) 2034 { 2035 struct btrfs_super_block *disk_super; 2036 int copy_num; 2037 2038 if (!bdev) 2039 return; 2040 2041 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2042 struct page *page; 2043 int ret; 2044 2045 disk_super = btrfs_read_dev_one_super(bdev, copy_num, false); 2046 if (IS_ERR(disk_super)) 2047 continue; 2048 2049 if (bdev_is_zoned(bdev)) { 2050 btrfs_reset_sb_log_zones(bdev, copy_num); 2051 continue; 2052 } 2053 2054 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 2055 2056 page = virt_to_page(disk_super); 2057 set_page_dirty(page); 2058 lock_page(page); 2059 /* write_on_page() unlocks the page */ 2060 ret = write_one_page(page); 2061 if (ret) 2062 btrfs_warn(fs_info, 2063 "error clearing superblock number %d (%d)", 2064 copy_num, ret); 2065 btrfs_release_disk_super(disk_super); 2066 2067 } 2068 2069 /* Notify udev that device has changed */ 2070 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2071 2072 /* Update ctime/mtime for device path for libblkid */ 2073 update_dev_time(device_path); 2074 } 2075 2076 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2077 struct btrfs_dev_lookup_args *args, 2078 struct block_device **bdev, fmode_t *mode) 2079 { 2080 struct btrfs_trans_handle *trans; 2081 struct btrfs_device *device; 2082 struct btrfs_fs_devices *cur_devices; 2083 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2084 u64 num_devices; 2085 int ret = 0; 2086 2087 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2088 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); 2089 return -EINVAL; 2090 } 2091 2092 /* 2093 * The device list in fs_devices is accessed without locks (neither 2094 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2095 * filesystem and another device rm cannot run. 2096 */ 2097 num_devices = btrfs_num_devices(fs_info); 2098 2099 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2100 if (ret) 2101 return ret; 2102 2103 device = btrfs_find_device(fs_info->fs_devices, args); 2104 if (!device) { 2105 if (args->missing) 2106 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2107 else 2108 ret = -ENOENT; 2109 return ret; 2110 } 2111 2112 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2113 btrfs_warn_in_rcu(fs_info, 2114 "cannot remove device %s (devid %llu) due to active swapfile", 2115 rcu_str_deref(device->name), device->devid); 2116 return -ETXTBSY; 2117 } 2118 2119 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 2120 return BTRFS_ERROR_DEV_TGT_REPLACE; 2121 2122 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2123 fs_info->fs_devices->rw_devices == 1) 2124 return BTRFS_ERROR_DEV_ONLY_WRITABLE; 2125 2126 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2127 mutex_lock(&fs_info->chunk_mutex); 2128 list_del_init(&device->dev_alloc_list); 2129 device->fs_devices->rw_devices--; 2130 mutex_unlock(&fs_info->chunk_mutex); 2131 } 2132 2133 ret = btrfs_shrink_device(device, 0); 2134 if (ret) 2135 goto error_undo; 2136 2137 trans = btrfs_start_transaction(fs_info->chunk_root, 0); 2138 if (IS_ERR(trans)) { 2139 ret = PTR_ERR(trans); 2140 goto error_undo; 2141 } 2142 2143 ret = btrfs_rm_dev_item(trans, device); 2144 if (ret) { 2145 /* Any error in dev item removal is critical */ 2146 btrfs_crit(fs_info, 2147 "failed to remove device item for devid %llu: %d", 2148 device->devid, ret); 2149 btrfs_abort_transaction(trans, ret); 2150 btrfs_end_transaction(trans); 2151 return ret; 2152 } 2153 2154 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2155 btrfs_scrub_cancel_dev(device); 2156 2157 /* 2158 * the device list mutex makes sure that we don't change 2159 * the device list while someone else is writing out all 2160 * the device supers. Whoever is writing all supers, should 2161 * lock the device list mutex before getting the number of 2162 * devices in the super block (super_copy). Conversely, 2163 * whoever updates the number of devices in the super block 2164 * (super_copy) should hold the device list mutex. 2165 */ 2166 2167 /* 2168 * In normal cases the cur_devices == fs_devices. But in case 2169 * of deleting a seed device, the cur_devices should point to 2170 * its own fs_devices listed under the fs_devices->seed_list. 2171 */ 2172 cur_devices = device->fs_devices; 2173 mutex_lock(&fs_devices->device_list_mutex); 2174 list_del_rcu(&device->dev_list); 2175 2176 cur_devices->num_devices--; 2177 cur_devices->total_devices--; 2178 /* Update total_devices of the parent fs_devices if it's seed */ 2179 if (cur_devices != fs_devices) 2180 fs_devices->total_devices--; 2181 2182 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2183 cur_devices->missing_devices--; 2184 2185 btrfs_assign_next_active_device(device, NULL); 2186 2187 if (device->bdev) { 2188 cur_devices->open_devices--; 2189 /* remove sysfs entry */ 2190 btrfs_sysfs_remove_device(device); 2191 } 2192 2193 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2194 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2195 mutex_unlock(&fs_devices->device_list_mutex); 2196 2197 /* 2198 * At this point, the device is zero sized and detached from the 2199 * devices list. All that's left is to zero out the old supers and 2200 * free the device. 2201 * 2202 * We cannot call btrfs_close_bdev() here because we're holding the sb 2203 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2204 * block device and it's dependencies. Instead just flush the device 2205 * and let the caller do the final blkdev_put. 2206 */ 2207 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2208 btrfs_scratch_superblocks(fs_info, device->bdev, 2209 device->name->str); 2210 if (device->bdev) { 2211 sync_blockdev(device->bdev); 2212 invalidate_bdev(device->bdev); 2213 } 2214 } 2215 2216 *bdev = device->bdev; 2217 *mode = device->mode; 2218 synchronize_rcu(); 2219 btrfs_free_device(device); 2220 2221 /* 2222 * This can happen if cur_devices is the private seed devices list. We 2223 * cannot call close_fs_devices() here because it expects the uuid_mutex 2224 * to be held, but in fact we don't need that for the private 2225 * seed_devices, we can simply decrement cur_devices->opened and then 2226 * remove it from our list and free the fs_devices. 2227 */ 2228 if (cur_devices->num_devices == 0) { 2229 list_del_init(&cur_devices->seed_list); 2230 ASSERT(cur_devices->opened == 1); 2231 cur_devices->opened--; 2232 free_fs_devices(cur_devices); 2233 } 2234 2235 ret = btrfs_commit_transaction(trans); 2236 2237 return ret; 2238 2239 error_undo: 2240 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2241 mutex_lock(&fs_info->chunk_mutex); 2242 list_add(&device->dev_alloc_list, 2243 &fs_devices->alloc_list); 2244 device->fs_devices->rw_devices++; 2245 mutex_unlock(&fs_info->chunk_mutex); 2246 } 2247 return ret; 2248 } 2249 2250 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2251 { 2252 struct btrfs_fs_devices *fs_devices; 2253 2254 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2255 2256 /* 2257 * in case of fs with no seed, srcdev->fs_devices will point 2258 * to fs_devices of fs_info. However when the dev being replaced is 2259 * a seed dev it will point to the seed's local fs_devices. In short 2260 * srcdev will have its correct fs_devices in both the cases. 2261 */ 2262 fs_devices = srcdev->fs_devices; 2263 2264 list_del_rcu(&srcdev->dev_list); 2265 list_del(&srcdev->dev_alloc_list); 2266 fs_devices->num_devices--; 2267 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2268 fs_devices->missing_devices--; 2269 2270 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2271 fs_devices->rw_devices--; 2272 2273 if (srcdev->bdev) 2274 fs_devices->open_devices--; 2275 } 2276 2277 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2278 { 2279 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2280 2281 mutex_lock(&uuid_mutex); 2282 2283 btrfs_close_bdev(srcdev); 2284 synchronize_rcu(); 2285 btrfs_free_device(srcdev); 2286 2287 /* if this is no devs we rather delete the fs_devices */ 2288 if (!fs_devices->num_devices) { 2289 /* 2290 * On a mounted FS, num_devices can't be zero unless it's a 2291 * seed. In case of a seed device being replaced, the replace 2292 * target added to the sprout FS, so there will be no more 2293 * device left under the seed FS. 2294 */ 2295 ASSERT(fs_devices->seeding); 2296 2297 list_del_init(&fs_devices->seed_list); 2298 close_fs_devices(fs_devices); 2299 free_fs_devices(fs_devices); 2300 } 2301 mutex_unlock(&uuid_mutex); 2302 } 2303 2304 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2305 { 2306 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2307 2308 mutex_lock(&fs_devices->device_list_mutex); 2309 2310 btrfs_sysfs_remove_device(tgtdev); 2311 2312 if (tgtdev->bdev) 2313 fs_devices->open_devices--; 2314 2315 fs_devices->num_devices--; 2316 2317 btrfs_assign_next_active_device(tgtdev, NULL); 2318 2319 list_del_rcu(&tgtdev->dev_list); 2320 2321 mutex_unlock(&fs_devices->device_list_mutex); 2322 2323 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2324 tgtdev->name->str); 2325 2326 btrfs_close_bdev(tgtdev); 2327 synchronize_rcu(); 2328 btrfs_free_device(tgtdev); 2329 } 2330 2331 /* 2332 * Populate args from device at path. 2333 * 2334 * @fs_info: the filesystem 2335 * @args: the args to populate 2336 * @path: the path to the device 2337 * 2338 * This will read the super block of the device at @path and populate @args with 2339 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2340 * lookup a device to operate on, but need to do it before we take any locks. 2341 * This properly handles the special case of "missing" that a user may pass in, 2342 * and does some basic sanity checks. The caller must make sure that @path is 2343 * properly NUL terminated before calling in, and must call 2344 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2345 * uuid buffers. 2346 * 2347 * Return: 0 for success, -errno for failure 2348 */ 2349 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2350 struct btrfs_dev_lookup_args *args, 2351 const char *path) 2352 { 2353 struct btrfs_super_block *disk_super; 2354 struct block_device *bdev; 2355 int ret; 2356 2357 if (!path || !path[0]) 2358 return -EINVAL; 2359 if (!strcmp(path, "missing")) { 2360 args->missing = true; 2361 return 0; 2362 } 2363 2364 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2365 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2366 if (!args->uuid || !args->fsid) { 2367 btrfs_put_dev_args_from_path(args); 2368 return -ENOMEM; 2369 } 2370 2371 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2372 &bdev, &disk_super); 2373 if (ret) { 2374 btrfs_put_dev_args_from_path(args); 2375 return ret; 2376 } 2377 2378 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2379 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2380 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2381 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2382 else 2383 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2384 btrfs_release_disk_super(disk_super); 2385 blkdev_put(bdev, FMODE_READ); 2386 return 0; 2387 } 2388 2389 /* 2390 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2391 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2392 * that don't need to be freed. 2393 */ 2394 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2395 { 2396 kfree(args->uuid); 2397 kfree(args->fsid); 2398 args->uuid = NULL; 2399 args->fsid = NULL; 2400 } 2401 2402 struct btrfs_device *btrfs_find_device_by_devspec( 2403 struct btrfs_fs_info *fs_info, u64 devid, 2404 const char *device_path) 2405 { 2406 BTRFS_DEV_LOOKUP_ARGS(args); 2407 struct btrfs_device *device; 2408 int ret; 2409 2410 if (devid) { 2411 args.devid = devid; 2412 device = btrfs_find_device(fs_info->fs_devices, &args); 2413 if (!device) 2414 return ERR_PTR(-ENOENT); 2415 return device; 2416 } 2417 2418 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2419 if (ret) 2420 return ERR_PTR(ret); 2421 device = btrfs_find_device(fs_info->fs_devices, &args); 2422 btrfs_put_dev_args_from_path(&args); 2423 if (!device) 2424 return ERR_PTR(-ENOENT); 2425 return device; 2426 } 2427 2428 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) 2429 { 2430 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2431 struct btrfs_fs_devices *old_devices; 2432 struct btrfs_fs_devices *seed_devices; 2433 2434 lockdep_assert_held(&uuid_mutex); 2435 if (!fs_devices->seeding) 2436 return ERR_PTR(-EINVAL); 2437 2438 /* 2439 * Private copy of the seed devices, anchored at 2440 * fs_info->fs_devices->seed_list 2441 */ 2442 seed_devices = alloc_fs_devices(NULL, NULL); 2443 if (IS_ERR(seed_devices)) 2444 return seed_devices; 2445 2446 /* 2447 * It's necessary to retain a copy of the original seed fs_devices in 2448 * fs_uuids so that filesystems which have been seeded can successfully 2449 * reference the seed device from open_seed_devices. This also supports 2450 * multiple fs seed. 2451 */ 2452 old_devices = clone_fs_devices(fs_devices); 2453 if (IS_ERR(old_devices)) { 2454 kfree(seed_devices); 2455 return old_devices; 2456 } 2457 2458 list_add(&old_devices->fs_list, &fs_uuids); 2459 2460 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2461 seed_devices->opened = 1; 2462 INIT_LIST_HEAD(&seed_devices->devices); 2463 INIT_LIST_HEAD(&seed_devices->alloc_list); 2464 mutex_init(&seed_devices->device_list_mutex); 2465 2466 return seed_devices; 2467 } 2468 2469 /* 2470 * Splice seed devices into the sprout fs_devices. 2471 * Generate a new fsid for the sprouted read-write filesystem. 2472 */ 2473 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, 2474 struct btrfs_fs_devices *seed_devices) 2475 { 2476 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2477 struct btrfs_super_block *disk_super = fs_info->super_copy; 2478 struct btrfs_device *device; 2479 u64 super_flags; 2480 2481 /* 2482 * We are updating the fsid, the thread leading to device_list_add() 2483 * could race, so uuid_mutex is needed. 2484 */ 2485 lockdep_assert_held(&uuid_mutex); 2486 2487 /* 2488 * The threads listed below may traverse dev_list but can do that without 2489 * device_list_mutex: 2490 * - All device ops and balance - as we are in btrfs_exclop_start. 2491 * - Various dev_list readers - are using RCU. 2492 * - btrfs_ioctl_fitrim() - is using RCU. 2493 * 2494 * For-read threads as below are using device_list_mutex: 2495 * - Readonly scrub btrfs_scrub_dev() 2496 * - Readonly scrub btrfs_scrub_progress() 2497 * - btrfs_get_dev_stats() 2498 */ 2499 lockdep_assert_held(&fs_devices->device_list_mutex); 2500 2501 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2502 synchronize_rcu); 2503 list_for_each_entry(device, &seed_devices->devices, dev_list) 2504 device->fs_devices = seed_devices; 2505 2506 fs_devices->seeding = false; 2507 fs_devices->num_devices = 0; 2508 fs_devices->open_devices = 0; 2509 fs_devices->missing_devices = 0; 2510 fs_devices->rotating = false; 2511 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2512 2513 generate_random_uuid(fs_devices->fsid); 2514 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2515 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2516 2517 super_flags = btrfs_super_flags(disk_super) & 2518 ~BTRFS_SUPER_FLAG_SEEDING; 2519 btrfs_set_super_flags(disk_super, super_flags); 2520 } 2521 2522 /* 2523 * Store the expected generation for seed devices in device items. 2524 */ 2525 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2526 { 2527 BTRFS_DEV_LOOKUP_ARGS(args); 2528 struct btrfs_fs_info *fs_info = trans->fs_info; 2529 struct btrfs_root *root = fs_info->chunk_root; 2530 struct btrfs_path *path; 2531 struct extent_buffer *leaf; 2532 struct btrfs_dev_item *dev_item; 2533 struct btrfs_device *device; 2534 struct btrfs_key key; 2535 u8 fs_uuid[BTRFS_FSID_SIZE]; 2536 u8 dev_uuid[BTRFS_UUID_SIZE]; 2537 int ret; 2538 2539 path = btrfs_alloc_path(); 2540 if (!path) 2541 return -ENOMEM; 2542 2543 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2544 key.offset = 0; 2545 key.type = BTRFS_DEV_ITEM_KEY; 2546 2547 while (1) { 2548 btrfs_reserve_chunk_metadata(trans, false); 2549 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2550 btrfs_trans_release_chunk_metadata(trans); 2551 if (ret < 0) 2552 goto error; 2553 2554 leaf = path->nodes[0]; 2555 next_slot: 2556 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2557 ret = btrfs_next_leaf(root, path); 2558 if (ret > 0) 2559 break; 2560 if (ret < 0) 2561 goto error; 2562 leaf = path->nodes[0]; 2563 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2564 btrfs_release_path(path); 2565 continue; 2566 } 2567 2568 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2569 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2570 key.type != BTRFS_DEV_ITEM_KEY) 2571 break; 2572 2573 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2574 struct btrfs_dev_item); 2575 args.devid = btrfs_device_id(leaf, dev_item); 2576 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2577 BTRFS_UUID_SIZE); 2578 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2579 BTRFS_FSID_SIZE); 2580 args.uuid = dev_uuid; 2581 args.fsid = fs_uuid; 2582 device = btrfs_find_device(fs_info->fs_devices, &args); 2583 BUG_ON(!device); /* Logic error */ 2584 2585 if (device->fs_devices->seeding) { 2586 btrfs_set_device_generation(leaf, dev_item, 2587 device->generation); 2588 btrfs_mark_buffer_dirty(leaf); 2589 } 2590 2591 path->slots[0]++; 2592 goto next_slot; 2593 } 2594 ret = 0; 2595 error: 2596 btrfs_free_path(path); 2597 return ret; 2598 } 2599 2600 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2601 { 2602 struct btrfs_root *root = fs_info->dev_root; 2603 struct btrfs_trans_handle *trans; 2604 struct btrfs_device *device; 2605 struct block_device *bdev; 2606 struct super_block *sb = fs_info->sb; 2607 struct rcu_string *name; 2608 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2609 struct btrfs_fs_devices *seed_devices; 2610 u64 orig_super_total_bytes; 2611 u64 orig_super_num_devices; 2612 int ret = 0; 2613 bool seeding_dev = false; 2614 bool locked = false; 2615 2616 if (sb_rdonly(sb) && !fs_devices->seeding) 2617 return -EROFS; 2618 2619 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2620 fs_info->bdev_holder); 2621 if (IS_ERR(bdev)) 2622 return PTR_ERR(bdev); 2623 2624 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2625 ret = -EINVAL; 2626 goto error; 2627 } 2628 2629 if (fs_devices->seeding) { 2630 seeding_dev = true; 2631 down_write(&sb->s_umount); 2632 mutex_lock(&uuid_mutex); 2633 locked = true; 2634 } 2635 2636 sync_blockdev(bdev); 2637 2638 rcu_read_lock(); 2639 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2640 if (device->bdev == bdev) { 2641 ret = -EEXIST; 2642 rcu_read_unlock(); 2643 goto error; 2644 } 2645 } 2646 rcu_read_unlock(); 2647 2648 device = btrfs_alloc_device(fs_info, NULL, NULL); 2649 if (IS_ERR(device)) { 2650 /* we can safely leave the fs_devices entry around */ 2651 ret = PTR_ERR(device); 2652 goto error; 2653 } 2654 2655 name = rcu_string_strdup(device_path, GFP_KERNEL); 2656 if (!name) { 2657 ret = -ENOMEM; 2658 goto error_free_device; 2659 } 2660 rcu_assign_pointer(device->name, name); 2661 2662 device->fs_info = fs_info; 2663 device->bdev = bdev; 2664 ret = lookup_bdev(device_path, &device->devt); 2665 if (ret) 2666 goto error_free_device; 2667 2668 ret = btrfs_get_dev_zone_info(device, false); 2669 if (ret) 2670 goto error_free_device; 2671 2672 trans = btrfs_start_transaction(root, 0); 2673 if (IS_ERR(trans)) { 2674 ret = PTR_ERR(trans); 2675 goto error_free_zone; 2676 } 2677 2678 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2679 device->generation = trans->transid; 2680 device->io_width = fs_info->sectorsize; 2681 device->io_align = fs_info->sectorsize; 2682 device->sector_size = fs_info->sectorsize; 2683 device->total_bytes = 2684 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); 2685 device->disk_total_bytes = device->total_bytes; 2686 device->commit_total_bytes = device->total_bytes; 2687 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2688 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2689 device->mode = FMODE_EXCL; 2690 device->dev_stats_valid = 1; 2691 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2692 2693 if (seeding_dev) { 2694 btrfs_clear_sb_rdonly(sb); 2695 2696 /* GFP_KERNEL allocation must not be under device_list_mutex */ 2697 seed_devices = btrfs_init_sprout(fs_info); 2698 if (IS_ERR(seed_devices)) { 2699 ret = PTR_ERR(seed_devices); 2700 btrfs_abort_transaction(trans, ret); 2701 goto error_trans; 2702 } 2703 } 2704 2705 mutex_lock(&fs_devices->device_list_mutex); 2706 if (seeding_dev) { 2707 btrfs_setup_sprout(fs_info, seed_devices); 2708 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2709 device); 2710 } 2711 2712 device->fs_devices = fs_devices; 2713 2714 mutex_lock(&fs_info->chunk_mutex); 2715 list_add_rcu(&device->dev_list, &fs_devices->devices); 2716 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2717 fs_devices->num_devices++; 2718 fs_devices->open_devices++; 2719 fs_devices->rw_devices++; 2720 fs_devices->total_devices++; 2721 fs_devices->total_rw_bytes += device->total_bytes; 2722 2723 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2724 2725 if (!bdev_nonrot(bdev)) 2726 fs_devices->rotating = true; 2727 2728 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2729 btrfs_set_super_total_bytes(fs_info->super_copy, 2730 round_down(orig_super_total_bytes + device->total_bytes, 2731 fs_info->sectorsize)); 2732 2733 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2734 btrfs_set_super_num_devices(fs_info->super_copy, 2735 orig_super_num_devices + 1); 2736 2737 /* 2738 * we've got more storage, clear any full flags on the space 2739 * infos 2740 */ 2741 btrfs_clear_space_info_full(fs_info); 2742 2743 mutex_unlock(&fs_info->chunk_mutex); 2744 2745 /* Add sysfs device entry */ 2746 btrfs_sysfs_add_device(device); 2747 2748 mutex_unlock(&fs_devices->device_list_mutex); 2749 2750 if (seeding_dev) { 2751 mutex_lock(&fs_info->chunk_mutex); 2752 ret = init_first_rw_device(trans); 2753 mutex_unlock(&fs_info->chunk_mutex); 2754 if (ret) { 2755 btrfs_abort_transaction(trans, ret); 2756 goto error_sysfs; 2757 } 2758 } 2759 2760 ret = btrfs_add_dev_item(trans, device); 2761 if (ret) { 2762 btrfs_abort_transaction(trans, ret); 2763 goto error_sysfs; 2764 } 2765 2766 if (seeding_dev) { 2767 ret = btrfs_finish_sprout(trans); 2768 if (ret) { 2769 btrfs_abort_transaction(trans, ret); 2770 goto error_sysfs; 2771 } 2772 2773 /* 2774 * fs_devices now represents the newly sprouted filesystem and 2775 * its fsid has been changed by btrfs_sprout_splice(). 2776 */ 2777 btrfs_sysfs_update_sprout_fsid(fs_devices); 2778 } 2779 2780 ret = btrfs_commit_transaction(trans); 2781 2782 if (seeding_dev) { 2783 mutex_unlock(&uuid_mutex); 2784 up_write(&sb->s_umount); 2785 locked = false; 2786 2787 if (ret) /* transaction commit */ 2788 return ret; 2789 2790 ret = btrfs_relocate_sys_chunks(fs_info); 2791 if (ret < 0) 2792 btrfs_handle_fs_error(fs_info, ret, 2793 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2794 trans = btrfs_attach_transaction(root); 2795 if (IS_ERR(trans)) { 2796 if (PTR_ERR(trans) == -ENOENT) 2797 return 0; 2798 ret = PTR_ERR(trans); 2799 trans = NULL; 2800 goto error_sysfs; 2801 } 2802 ret = btrfs_commit_transaction(trans); 2803 } 2804 2805 /* 2806 * Now that we have written a new super block to this device, check all 2807 * other fs_devices list if device_path alienates any other scanned 2808 * device. 2809 * We can ignore the return value as it typically returns -EINVAL and 2810 * only succeeds if the device was an alien. 2811 */ 2812 btrfs_forget_devices(device->devt); 2813 2814 /* Update ctime/mtime for blkid or udev */ 2815 update_dev_time(device_path); 2816 2817 return ret; 2818 2819 error_sysfs: 2820 btrfs_sysfs_remove_device(device); 2821 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2822 mutex_lock(&fs_info->chunk_mutex); 2823 list_del_rcu(&device->dev_list); 2824 list_del(&device->dev_alloc_list); 2825 fs_info->fs_devices->num_devices--; 2826 fs_info->fs_devices->open_devices--; 2827 fs_info->fs_devices->rw_devices--; 2828 fs_info->fs_devices->total_devices--; 2829 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2830 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2831 btrfs_set_super_total_bytes(fs_info->super_copy, 2832 orig_super_total_bytes); 2833 btrfs_set_super_num_devices(fs_info->super_copy, 2834 orig_super_num_devices); 2835 mutex_unlock(&fs_info->chunk_mutex); 2836 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2837 error_trans: 2838 if (seeding_dev) 2839 btrfs_set_sb_rdonly(sb); 2840 if (trans) 2841 btrfs_end_transaction(trans); 2842 error_free_zone: 2843 btrfs_destroy_dev_zone_info(device); 2844 error_free_device: 2845 btrfs_free_device(device); 2846 error: 2847 blkdev_put(bdev, FMODE_EXCL); 2848 if (locked) { 2849 mutex_unlock(&uuid_mutex); 2850 up_write(&sb->s_umount); 2851 } 2852 return ret; 2853 } 2854 2855 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2856 struct btrfs_device *device) 2857 { 2858 int ret; 2859 struct btrfs_path *path; 2860 struct btrfs_root *root = device->fs_info->chunk_root; 2861 struct btrfs_dev_item *dev_item; 2862 struct extent_buffer *leaf; 2863 struct btrfs_key key; 2864 2865 path = btrfs_alloc_path(); 2866 if (!path) 2867 return -ENOMEM; 2868 2869 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2870 key.type = BTRFS_DEV_ITEM_KEY; 2871 key.offset = device->devid; 2872 2873 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2874 if (ret < 0) 2875 goto out; 2876 2877 if (ret > 0) { 2878 ret = -ENOENT; 2879 goto out; 2880 } 2881 2882 leaf = path->nodes[0]; 2883 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2884 2885 btrfs_set_device_id(leaf, dev_item, device->devid); 2886 btrfs_set_device_type(leaf, dev_item, device->type); 2887 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2888 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2889 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2890 btrfs_set_device_total_bytes(leaf, dev_item, 2891 btrfs_device_get_disk_total_bytes(device)); 2892 btrfs_set_device_bytes_used(leaf, dev_item, 2893 btrfs_device_get_bytes_used(device)); 2894 btrfs_mark_buffer_dirty(leaf); 2895 2896 out: 2897 btrfs_free_path(path); 2898 return ret; 2899 } 2900 2901 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2902 struct btrfs_device *device, u64 new_size) 2903 { 2904 struct btrfs_fs_info *fs_info = device->fs_info; 2905 struct btrfs_super_block *super_copy = fs_info->super_copy; 2906 u64 old_total; 2907 u64 diff; 2908 int ret; 2909 2910 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2911 return -EACCES; 2912 2913 new_size = round_down(new_size, fs_info->sectorsize); 2914 2915 mutex_lock(&fs_info->chunk_mutex); 2916 old_total = btrfs_super_total_bytes(super_copy); 2917 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2918 2919 if (new_size <= device->total_bytes || 2920 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2921 mutex_unlock(&fs_info->chunk_mutex); 2922 return -EINVAL; 2923 } 2924 2925 btrfs_set_super_total_bytes(super_copy, 2926 round_down(old_total + diff, fs_info->sectorsize)); 2927 device->fs_devices->total_rw_bytes += diff; 2928 2929 btrfs_device_set_total_bytes(device, new_size); 2930 btrfs_device_set_disk_total_bytes(device, new_size); 2931 btrfs_clear_space_info_full(device->fs_info); 2932 if (list_empty(&device->post_commit_list)) 2933 list_add_tail(&device->post_commit_list, 2934 &trans->transaction->dev_update_list); 2935 mutex_unlock(&fs_info->chunk_mutex); 2936 2937 btrfs_reserve_chunk_metadata(trans, false); 2938 ret = btrfs_update_device(trans, device); 2939 btrfs_trans_release_chunk_metadata(trans); 2940 2941 return ret; 2942 } 2943 2944 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2945 { 2946 struct btrfs_fs_info *fs_info = trans->fs_info; 2947 struct btrfs_root *root = fs_info->chunk_root; 2948 int ret; 2949 struct btrfs_path *path; 2950 struct btrfs_key key; 2951 2952 path = btrfs_alloc_path(); 2953 if (!path) 2954 return -ENOMEM; 2955 2956 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2957 key.offset = chunk_offset; 2958 key.type = BTRFS_CHUNK_ITEM_KEY; 2959 2960 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2961 if (ret < 0) 2962 goto out; 2963 else if (ret > 0) { /* Logic error or corruption */ 2964 btrfs_handle_fs_error(fs_info, -ENOENT, 2965 "Failed lookup while freeing chunk."); 2966 ret = -ENOENT; 2967 goto out; 2968 } 2969 2970 ret = btrfs_del_item(trans, root, path); 2971 if (ret < 0) 2972 btrfs_handle_fs_error(fs_info, ret, 2973 "Failed to delete chunk item."); 2974 out: 2975 btrfs_free_path(path); 2976 return ret; 2977 } 2978 2979 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2980 { 2981 struct btrfs_super_block *super_copy = fs_info->super_copy; 2982 struct btrfs_disk_key *disk_key; 2983 struct btrfs_chunk *chunk; 2984 u8 *ptr; 2985 int ret = 0; 2986 u32 num_stripes; 2987 u32 array_size; 2988 u32 len = 0; 2989 u32 cur; 2990 struct btrfs_key key; 2991 2992 lockdep_assert_held(&fs_info->chunk_mutex); 2993 array_size = btrfs_super_sys_array_size(super_copy); 2994 2995 ptr = super_copy->sys_chunk_array; 2996 cur = 0; 2997 2998 while (cur < array_size) { 2999 disk_key = (struct btrfs_disk_key *)ptr; 3000 btrfs_disk_key_to_cpu(&key, disk_key); 3001 3002 len = sizeof(*disk_key); 3003 3004 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 3005 chunk = (struct btrfs_chunk *)(ptr + len); 3006 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 3007 len += btrfs_chunk_item_size(num_stripes); 3008 } else { 3009 ret = -EIO; 3010 break; 3011 } 3012 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 3013 key.offset == chunk_offset) { 3014 memmove(ptr, ptr + len, array_size - (cur + len)); 3015 array_size -= len; 3016 btrfs_set_super_sys_array_size(super_copy, array_size); 3017 } else { 3018 ptr += len; 3019 cur += len; 3020 } 3021 } 3022 return ret; 3023 } 3024 3025 /* 3026 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 3027 * @logical: Logical block offset in bytes. 3028 * @length: Length of extent in bytes. 3029 * 3030 * Return: Chunk mapping or ERR_PTR. 3031 */ 3032 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 3033 u64 logical, u64 length) 3034 { 3035 struct extent_map_tree *em_tree; 3036 struct extent_map *em; 3037 3038 em_tree = &fs_info->mapping_tree; 3039 read_lock(&em_tree->lock); 3040 em = lookup_extent_mapping(em_tree, logical, length); 3041 read_unlock(&em_tree->lock); 3042 3043 if (!em) { 3044 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 3045 logical, length); 3046 return ERR_PTR(-EINVAL); 3047 } 3048 3049 if (em->start > logical || em->start + em->len < logical) { 3050 btrfs_crit(fs_info, 3051 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3052 logical, length, em->start, em->start + em->len); 3053 free_extent_map(em); 3054 return ERR_PTR(-EINVAL); 3055 } 3056 3057 /* callers are responsible for dropping em's ref. */ 3058 return em; 3059 } 3060 3061 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3062 struct map_lookup *map, u64 chunk_offset) 3063 { 3064 int i; 3065 3066 /* 3067 * Removing chunk items and updating the device items in the chunks btree 3068 * requires holding the chunk_mutex. 3069 * See the comment at btrfs_chunk_alloc() for the details. 3070 */ 3071 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3072 3073 for (i = 0; i < map->num_stripes; i++) { 3074 int ret; 3075 3076 ret = btrfs_update_device(trans, map->stripes[i].dev); 3077 if (ret) 3078 return ret; 3079 } 3080 3081 return btrfs_free_chunk(trans, chunk_offset); 3082 } 3083 3084 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3085 { 3086 struct btrfs_fs_info *fs_info = trans->fs_info; 3087 struct extent_map *em; 3088 struct map_lookup *map; 3089 u64 dev_extent_len = 0; 3090 int i, ret = 0; 3091 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3092 3093 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3094 if (IS_ERR(em)) { 3095 /* 3096 * This is a logic error, but we don't want to just rely on the 3097 * user having built with ASSERT enabled, so if ASSERT doesn't 3098 * do anything we still error out. 3099 */ 3100 ASSERT(0); 3101 return PTR_ERR(em); 3102 } 3103 map = em->map_lookup; 3104 3105 /* 3106 * First delete the device extent items from the devices btree. 3107 * We take the device_list_mutex to avoid racing with the finishing phase 3108 * of a device replace operation. See the comment below before acquiring 3109 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3110 * because that can result in a deadlock when deleting the device extent 3111 * items from the devices btree - COWing an extent buffer from the btree 3112 * may result in allocating a new metadata chunk, which would attempt to 3113 * lock again fs_info->chunk_mutex. 3114 */ 3115 mutex_lock(&fs_devices->device_list_mutex); 3116 for (i = 0; i < map->num_stripes; i++) { 3117 struct btrfs_device *device = map->stripes[i].dev; 3118 ret = btrfs_free_dev_extent(trans, device, 3119 map->stripes[i].physical, 3120 &dev_extent_len); 3121 if (ret) { 3122 mutex_unlock(&fs_devices->device_list_mutex); 3123 btrfs_abort_transaction(trans, ret); 3124 goto out; 3125 } 3126 3127 if (device->bytes_used > 0) { 3128 mutex_lock(&fs_info->chunk_mutex); 3129 btrfs_device_set_bytes_used(device, 3130 device->bytes_used - dev_extent_len); 3131 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3132 btrfs_clear_space_info_full(fs_info); 3133 mutex_unlock(&fs_info->chunk_mutex); 3134 } 3135 } 3136 mutex_unlock(&fs_devices->device_list_mutex); 3137 3138 /* 3139 * We acquire fs_info->chunk_mutex for 2 reasons: 3140 * 3141 * 1) Just like with the first phase of the chunk allocation, we must 3142 * reserve system space, do all chunk btree updates and deletions, and 3143 * update the system chunk array in the superblock while holding this 3144 * mutex. This is for similar reasons as explained on the comment at 3145 * the top of btrfs_chunk_alloc(); 3146 * 3147 * 2) Prevent races with the final phase of a device replace operation 3148 * that replaces the device object associated with the map's stripes, 3149 * because the device object's id can change at any time during that 3150 * final phase of the device replace operation 3151 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3152 * replaced device and then see it with an ID of 3153 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3154 * the device item, which does not exists on the chunk btree. 3155 * The finishing phase of device replace acquires both the 3156 * device_list_mutex and the chunk_mutex, in that order, so we are 3157 * safe by just acquiring the chunk_mutex. 3158 */ 3159 trans->removing_chunk = true; 3160 mutex_lock(&fs_info->chunk_mutex); 3161 3162 check_system_chunk(trans, map->type); 3163 3164 ret = remove_chunk_item(trans, map, chunk_offset); 3165 /* 3166 * Normally we should not get -ENOSPC since we reserved space before 3167 * through the call to check_system_chunk(). 3168 * 3169 * Despite our system space_info having enough free space, we may not 3170 * be able to allocate extents from its block groups, because all have 3171 * an incompatible profile, which will force us to allocate a new system 3172 * block group with the right profile, or right after we called 3173 * check_system_space() above, a scrub turned the only system block group 3174 * with enough free space into RO mode. 3175 * This is explained with more detail at do_chunk_alloc(). 3176 * 3177 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3178 */ 3179 if (ret == -ENOSPC) { 3180 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3181 struct btrfs_block_group *sys_bg; 3182 3183 sys_bg = btrfs_create_chunk(trans, sys_flags); 3184 if (IS_ERR(sys_bg)) { 3185 ret = PTR_ERR(sys_bg); 3186 btrfs_abort_transaction(trans, ret); 3187 goto out; 3188 } 3189 3190 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3191 if (ret) { 3192 btrfs_abort_transaction(trans, ret); 3193 goto out; 3194 } 3195 3196 ret = remove_chunk_item(trans, map, chunk_offset); 3197 if (ret) { 3198 btrfs_abort_transaction(trans, ret); 3199 goto out; 3200 } 3201 } else if (ret) { 3202 btrfs_abort_transaction(trans, ret); 3203 goto out; 3204 } 3205 3206 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3207 3208 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3209 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3210 if (ret) { 3211 btrfs_abort_transaction(trans, ret); 3212 goto out; 3213 } 3214 } 3215 3216 mutex_unlock(&fs_info->chunk_mutex); 3217 trans->removing_chunk = false; 3218 3219 /* 3220 * We are done with chunk btree updates and deletions, so release the 3221 * system space we previously reserved (with check_system_chunk()). 3222 */ 3223 btrfs_trans_release_chunk_metadata(trans); 3224 3225 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3226 if (ret) { 3227 btrfs_abort_transaction(trans, ret); 3228 goto out; 3229 } 3230 3231 out: 3232 if (trans->removing_chunk) { 3233 mutex_unlock(&fs_info->chunk_mutex); 3234 trans->removing_chunk = false; 3235 } 3236 /* once for us */ 3237 free_extent_map(em); 3238 return ret; 3239 } 3240 3241 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3242 { 3243 struct btrfs_root *root = fs_info->chunk_root; 3244 struct btrfs_trans_handle *trans; 3245 struct btrfs_block_group *block_group; 3246 u64 length; 3247 int ret; 3248 3249 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3250 btrfs_err(fs_info, 3251 "relocate: not supported on extent tree v2 yet"); 3252 return -EINVAL; 3253 } 3254 3255 /* 3256 * Prevent races with automatic removal of unused block groups. 3257 * After we relocate and before we remove the chunk with offset 3258 * chunk_offset, automatic removal of the block group can kick in, 3259 * resulting in a failure when calling btrfs_remove_chunk() below. 3260 * 3261 * Make sure to acquire this mutex before doing a tree search (dev 3262 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3263 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3264 * we release the path used to search the chunk/dev tree and before 3265 * the current task acquires this mutex and calls us. 3266 */ 3267 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3268 3269 /* step one, relocate all the extents inside this chunk */ 3270 btrfs_scrub_pause(fs_info); 3271 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3272 btrfs_scrub_continue(fs_info); 3273 if (ret) 3274 return ret; 3275 3276 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3277 if (!block_group) 3278 return -ENOENT; 3279 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3280 length = block_group->length; 3281 btrfs_put_block_group(block_group); 3282 3283 /* 3284 * On a zoned file system, discard the whole block group, this will 3285 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3286 * resetting the zone fails, don't treat it as a fatal problem from the 3287 * filesystem's point of view. 3288 */ 3289 if (btrfs_is_zoned(fs_info)) { 3290 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3291 if (ret) 3292 btrfs_info(fs_info, 3293 "failed to reset zone %llu after relocation", 3294 chunk_offset); 3295 } 3296 3297 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3298 chunk_offset); 3299 if (IS_ERR(trans)) { 3300 ret = PTR_ERR(trans); 3301 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3302 return ret; 3303 } 3304 3305 /* 3306 * step two, delete the device extents and the 3307 * chunk tree entries 3308 */ 3309 ret = btrfs_remove_chunk(trans, chunk_offset); 3310 btrfs_end_transaction(trans); 3311 return ret; 3312 } 3313 3314 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3315 { 3316 struct btrfs_root *chunk_root = fs_info->chunk_root; 3317 struct btrfs_path *path; 3318 struct extent_buffer *leaf; 3319 struct btrfs_chunk *chunk; 3320 struct btrfs_key key; 3321 struct btrfs_key found_key; 3322 u64 chunk_type; 3323 bool retried = false; 3324 int failed = 0; 3325 int ret; 3326 3327 path = btrfs_alloc_path(); 3328 if (!path) 3329 return -ENOMEM; 3330 3331 again: 3332 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3333 key.offset = (u64)-1; 3334 key.type = BTRFS_CHUNK_ITEM_KEY; 3335 3336 while (1) { 3337 mutex_lock(&fs_info->reclaim_bgs_lock); 3338 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3339 if (ret < 0) { 3340 mutex_unlock(&fs_info->reclaim_bgs_lock); 3341 goto error; 3342 } 3343 BUG_ON(ret == 0); /* Corruption */ 3344 3345 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3346 key.type); 3347 if (ret) 3348 mutex_unlock(&fs_info->reclaim_bgs_lock); 3349 if (ret < 0) 3350 goto error; 3351 if (ret > 0) 3352 break; 3353 3354 leaf = path->nodes[0]; 3355 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3356 3357 chunk = btrfs_item_ptr(leaf, path->slots[0], 3358 struct btrfs_chunk); 3359 chunk_type = btrfs_chunk_type(leaf, chunk); 3360 btrfs_release_path(path); 3361 3362 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3363 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3364 if (ret == -ENOSPC) 3365 failed++; 3366 else 3367 BUG_ON(ret); 3368 } 3369 mutex_unlock(&fs_info->reclaim_bgs_lock); 3370 3371 if (found_key.offset == 0) 3372 break; 3373 key.offset = found_key.offset - 1; 3374 } 3375 ret = 0; 3376 if (failed && !retried) { 3377 failed = 0; 3378 retried = true; 3379 goto again; 3380 } else if (WARN_ON(failed && retried)) { 3381 ret = -ENOSPC; 3382 } 3383 error: 3384 btrfs_free_path(path); 3385 return ret; 3386 } 3387 3388 /* 3389 * return 1 : allocate a data chunk successfully, 3390 * return <0: errors during allocating a data chunk, 3391 * return 0 : no need to allocate a data chunk. 3392 */ 3393 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3394 u64 chunk_offset) 3395 { 3396 struct btrfs_block_group *cache; 3397 u64 bytes_used; 3398 u64 chunk_type; 3399 3400 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3401 ASSERT(cache); 3402 chunk_type = cache->flags; 3403 btrfs_put_block_group(cache); 3404 3405 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3406 return 0; 3407 3408 spin_lock(&fs_info->data_sinfo->lock); 3409 bytes_used = fs_info->data_sinfo->bytes_used; 3410 spin_unlock(&fs_info->data_sinfo->lock); 3411 3412 if (!bytes_used) { 3413 struct btrfs_trans_handle *trans; 3414 int ret; 3415 3416 trans = btrfs_join_transaction(fs_info->tree_root); 3417 if (IS_ERR(trans)) 3418 return PTR_ERR(trans); 3419 3420 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3421 btrfs_end_transaction(trans); 3422 if (ret < 0) 3423 return ret; 3424 return 1; 3425 } 3426 3427 return 0; 3428 } 3429 3430 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3431 struct btrfs_balance_control *bctl) 3432 { 3433 struct btrfs_root *root = fs_info->tree_root; 3434 struct btrfs_trans_handle *trans; 3435 struct btrfs_balance_item *item; 3436 struct btrfs_disk_balance_args disk_bargs; 3437 struct btrfs_path *path; 3438 struct extent_buffer *leaf; 3439 struct btrfs_key key; 3440 int ret, err; 3441 3442 path = btrfs_alloc_path(); 3443 if (!path) 3444 return -ENOMEM; 3445 3446 trans = btrfs_start_transaction(root, 0); 3447 if (IS_ERR(trans)) { 3448 btrfs_free_path(path); 3449 return PTR_ERR(trans); 3450 } 3451 3452 key.objectid = BTRFS_BALANCE_OBJECTID; 3453 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3454 key.offset = 0; 3455 3456 ret = btrfs_insert_empty_item(trans, root, path, &key, 3457 sizeof(*item)); 3458 if (ret) 3459 goto out; 3460 3461 leaf = path->nodes[0]; 3462 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3463 3464 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3465 3466 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3467 btrfs_set_balance_data(leaf, item, &disk_bargs); 3468 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3469 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3470 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3471 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3472 3473 btrfs_set_balance_flags(leaf, item, bctl->flags); 3474 3475 btrfs_mark_buffer_dirty(leaf); 3476 out: 3477 btrfs_free_path(path); 3478 err = btrfs_commit_transaction(trans); 3479 if (err && !ret) 3480 ret = err; 3481 return ret; 3482 } 3483 3484 static int del_balance_item(struct btrfs_fs_info *fs_info) 3485 { 3486 struct btrfs_root *root = fs_info->tree_root; 3487 struct btrfs_trans_handle *trans; 3488 struct btrfs_path *path; 3489 struct btrfs_key key; 3490 int ret, err; 3491 3492 path = btrfs_alloc_path(); 3493 if (!path) 3494 return -ENOMEM; 3495 3496 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3497 if (IS_ERR(trans)) { 3498 btrfs_free_path(path); 3499 return PTR_ERR(trans); 3500 } 3501 3502 key.objectid = BTRFS_BALANCE_OBJECTID; 3503 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3504 key.offset = 0; 3505 3506 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3507 if (ret < 0) 3508 goto out; 3509 if (ret > 0) { 3510 ret = -ENOENT; 3511 goto out; 3512 } 3513 3514 ret = btrfs_del_item(trans, root, path); 3515 out: 3516 btrfs_free_path(path); 3517 err = btrfs_commit_transaction(trans); 3518 if (err && !ret) 3519 ret = err; 3520 return ret; 3521 } 3522 3523 /* 3524 * This is a heuristic used to reduce the number of chunks balanced on 3525 * resume after balance was interrupted. 3526 */ 3527 static void update_balance_args(struct btrfs_balance_control *bctl) 3528 { 3529 /* 3530 * Turn on soft mode for chunk types that were being converted. 3531 */ 3532 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3533 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3534 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3535 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3536 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3537 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3538 3539 /* 3540 * Turn on usage filter if is not already used. The idea is 3541 * that chunks that we have already balanced should be 3542 * reasonably full. Don't do it for chunks that are being 3543 * converted - that will keep us from relocating unconverted 3544 * (albeit full) chunks. 3545 */ 3546 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3547 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3548 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3549 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3550 bctl->data.usage = 90; 3551 } 3552 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3553 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3554 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3555 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3556 bctl->sys.usage = 90; 3557 } 3558 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3559 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3560 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3561 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3562 bctl->meta.usage = 90; 3563 } 3564 } 3565 3566 /* 3567 * Clear the balance status in fs_info and delete the balance item from disk. 3568 */ 3569 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3570 { 3571 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3572 int ret; 3573 3574 BUG_ON(!fs_info->balance_ctl); 3575 3576 spin_lock(&fs_info->balance_lock); 3577 fs_info->balance_ctl = NULL; 3578 spin_unlock(&fs_info->balance_lock); 3579 3580 kfree(bctl); 3581 ret = del_balance_item(fs_info); 3582 if (ret) 3583 btrfs_handle_fs_error(fs_info, ret, NULL); 3584 } 3585 3586 /* 3587 * Balance filters. Return 1 if chunk should be filtered out 3588 * (should not be balanced). 3589 */ 3590 static int chunk_profiles_filter(u64 chunk_type, 3591 struct btrfs_balance_args *bargs) 3592 { 3593 chunk_type = chunk_to_extended(chunk_type) & 3594 BTRFS_EXTENDED_PROFILE_MASK; 3595 3596 if (bargs->profiles & chunk_type) 3597 return 0; 3598 3599 return 1; 3600 } 3601 3602 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3603 struct btrfs_balance_args *bargs) 3604 { 3605 struct btrfs_block_group *cache; 3606 u64 chunk_used; 3607 u64 user_thresh_min; 3608 u64 user_thresh_max; 3609 int ret = 1; 3610 3611 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3612 chunk_used = cache->used; 3613 3614 if (bargs->usage_min == 0) 3615 user_thresh_min = 0; 3616 else 3617 user_thresh_min = div_factor_fine(cache->length, 3618 bargs->usage_min); 3619 3620 if (bargs->usage_max == 0) 3621 user_thresh_max = 1; 3622 else if (bargs->usage_max > 100) 3623 user_thresh_max = cache->length; 3624 else 3625 user_thresh_max = div_factor_fine(cache->length, 3626 bargs->usage_max); 3627 3628 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3629 ret = 0; 3630 3631 btrfs_put_block_group(cache); 3632 return ret; 3633 } 3634 3635 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3636 u64 chunk_offset, struct btrfs_balance_args *bargs) 3637 { 3638 struct btrfs_block_group *cache; 3639 u64 chunk_used, user_thresh; 3640 int ret = 1; 3641 3642 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3643 chunk_used = cache->used; 3644 3645 if (bargs->usage_min == 0) 3646 user_thresh = 1; 3647 else if (bargs->usage > 100) 3648 user_thresh = cache->length; 3649 else 3650 user_thresh = div_factor_fine(cache->length, bargs->usage); 3651 3652 if (chunk_used < user_thresh) 3653 ret = 0; 3654 3655 btrfs_put_block_group(cache); 3656 return ret; 3657 } 3658 3659 static int chunk_devid_filter(struct extent_buffer *leaf, 3660 struct btrfs_chunk *chunk, 3661 struct btrfs_balance_args *bargs) 3662 { 3663 struct btrfs_stripe *stripe; 3664 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3665 int i; 3666 3667 for (i = 0; i < num_stripes; i++) { 3668 stripe = btrfs_stripe_nr(chunk, i); 3669 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3670 return 0; 3671 } 3672 3673 return 1; 3674 } 3675 3676 static u64 calc_data_stripes(u64 type, int num_stripes) 3677 { 3678 const int index = btrfs_bg_flags_to_raid_index(type); 3679 const int ncopies = btrfs_raid_array[index].ncopies; 3680 const int nparity = btrfs_raid_array[index].nparity; 3681 3682 return (num_stripes - nparity) / ncopies; 3683 } 3684 3685 /* [pstart, pend) */ 3686 static int chunk_drange_filter(struct extent_buffer *leaf, 3687 struct btrfs_chunk *chunk, 3688 struct btrfs_balance_args *bargs) 3689 { 3690 struct btrfs_stripe *stripe; 3691 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3692 u64 stripe_offset; 3693 u64 stripe_length; 3694 u64 type; 3695 int factor; 3696 int i; 3697 3698 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3699 return 0; 3700 3701 type = btrfs_chunk_type(leaf, chunk); 3702 factor = calc_data_stripes(type, num_stripes); 3703 3704 for (i = 0; i < num_stripes; i++) { 3705 stripe = btrfs_stripe_nr(chunk, i); 3706 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3707 continue; 3708 3709 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3710 stripe_length = btrfs_chunk_length(leaf, chunk); 3711 stripe_length = div_u64(stripe_length, factor); 3712 3713 if (stripe_offset < bargs->pend && 3714 stripe_offset + stripe_length > bargs->pstart) 3715 return 0; 3716 } 3717 3718 return 1; 3719 } 3720 3721 /* [vstart, vend) */ 3722 static int chunk_vrange_filter(struct extent_buffer *leaf, 3723 struct btrfs_chunk *chunk, 3724 u64 chunk_offset, 3725 struct btrfs_balance_args *bargs) 3726 { 3727 if (chunk_offset < bargs->vend && 3728 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3729 /* at least part of the chunk is inside this vrange */ 3730 return 0; 3731 3732 return 1; 3733 } 3734 3735 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3736 struct btrfs_chunk *chunk, 3737 struct btrfs_balance_args *bargs) 3738 { 3739 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3740 3741 if (bargs->stripes_min <= num_stripes 3742 && num_stripes <= bargs->stripes_max) 3743 return 0; 3744 3745 return 1; 3746 } 3747 3748 static int chunk_soft_convert_filter(u64 chunk_type, 3749 struct btrfs_balance_args *bargs) 3750 { 3751 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3752 return 0; 3753 3754 chunk_type = chunk_to_extended(chunk_type) & 3755 BTRFS_EXTENDED_PROFILE_MASK; 3756 3757 if (bargs->target == chunk_type) 3758 return 1; 3759 3760 return 0; 3761 } 3762 3763 static int should_balance_chunk(struct extent_buffer *leaf, 3764 struct btrfs_chunk *chunk, u64 chunk_offset) 3765 { 3766 struct btrfs_fs_info *fs_info = leaf->fs_info; 3767 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3768 struct btrfs_balance_args *bargs = NULL; 3769 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3770 3771 /* type filter */ 3772 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3773 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3774 return 0; 3775 } 3776 3777 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3778 bargs = &bctl->data; 3779 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3780 bargs = &bctl->sys; 3781 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3782 bargs = &bctl->meta; 3783 3784 /* profiles filter */ 3785 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3786 chunk_profiles_filter(chunk_type, bargs)) { 3787 return 0; 3788 } 3789 3790 /* usage filter */ 3791 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3792 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3793 return 0; 3794 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3795 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3796 return 0; 3797 } 3798 3799 /* devid filter */ 3800 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3801 chunk_devid_filter(leaf, chunk, bargs)) { 3802 return 0; 3803 } 3804 3805 /* drange filter, makes sense only with devid filter */ 3806 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3807 chunk_drange_filter(leaf, chunk, bargs)) { 3808 return 0; 3809 } 3810 3811 /* vrange filter */ 3812 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3813 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3814 return 0; 3815 } 3816 3817 /* stripes filter */ 3818 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3819 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3820 return 0; 3821 } 3822 3823 /* soft profile changing mode */ 3824 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3825 chunk_soft_convert_filter(chunk_type, bargs)) { 3826 return 0; 3827 } 3828 3829 /* 3830 * limited by count, must be the last filter 3831 */ 3832 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3833 if (bargs->limit == 0) 3834 return 0; 3835 else 3836 bargs->limit--; 3837 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3838 /* 3839 * Same logic as the 'limit' filter; the minimum cannot be 3840 * determined here because we do not have the global information 3841 * about the count of all chunks that satisfy the filters. 3842 */ 3843 if (bargs->limit_max == 0) 3844 return 0; 3845 else 3846 bargs->limit_max--; 3847 } 3848 3849 return 1; 3850 } 3851 3852 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3853 { 3854 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3855 struct btrfs_root *chunk_root = fs_info->chunk_root; 3856 u64 chunk_type; 3857 struct btrfs_chunk *chunk; 3858 struct btrfs_path *path = NULL; 3859 struct btrfs_key key; 3860 struct btrfs_key found_key; 3861 struct extent_buffer *leaf; 3862 int slot; 3863 int ret; 3864 int enospc_errors = 0; 3865 bool counting = true; 3866 /* The single value limit and min/max limits use the same bytes in the */ 3867 u64 limit_data = bctl->data.limit; 3868 u64 limit_meta = bctl->meta.limit; 3869 u64 limit_sys = bctl->sys.limit; 3870 u32 count_data = 0; 3871 u32 count_meta = 0; 3872 u32 count_sys = 0; 3873 int chunk_reserved = 0; 3874 3875 path = btrfs_alloc_path(); 3876 if (!path) { 3877 ret = -ENOMEM; 3878 goto error; 3879 } 3880 3881 /* zero out stat counters */ 3882 spin_lock(&fs_info->balance_lock); 3883 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3884 spin_unlock(&fs_info->balance_lock); 3885 again: 3886 if (!counting) { 3887 /* 3888 * The single value limit and min/max limits use the same bytes 3889 * in the 3890 */ 3891 bctl->data.limit = limit_data; 3892 bctl->meta.limit = limit_meta; 3893 bctl->sys.limit = limit_sys; 3894 } 3895 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3896 key.offset = (u64)-1; 3897 key.type = BTRFS_CHUNK_ITEM_KEY; 3898 3899 while (1) { 3900 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3901 atomic_read(&fs_info->balance_cancel_req)) { 3902 ret = -ECANCELED; 3903 goto error; 3904 } 3905 3906 mutex_lock(&fs_info->reclaim_bgs_lock); 3907 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3908 if (ret < 0) { 3909 mutex_unlock(&fs_info->reclaim_bgs_lock); 3910 goto error; 3911 } 3912 3913 /* 3914 * this shouldn't happen, it means the last relocate 3915 * failed 3916 */ 3917 if (ret == 0) 3918 BUG(); /* FIXME break ? */ 3919 3920 ret = btrfs_previous_item(chunk_root, path, 0, 3921 BTRFS_CHUNK_ITEM_KEY); 3922 if (ret) { 3923 mutex_unlock(&fs_info->reclaim_bgs_lock); 3924 ret = 0; 3925 break; 3926 } 3927 3928 leaf = path->nodes[0]; 3929 slot = path->slots[0]; 3930 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3931 3932 if (found_key.objectid != key.objectid) { 3933 mutex_unlock(&fs_info->reclaim_bgs_lock); 3934 break; 3935 } 3936 3937 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3938 chunk_type = btrfs_chunk_type(leaf, chunk); 3939 3940 if (!counting) { 3941 spin_lock(&fs_info->balance_lock); 3942 bctl->stat.considered++; 3943 spin_unlock(&fs_info->balance_lock); 3944 } 3945 3946 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3947 3948 btrfs_release_path(path); 3949 if (!ret) { 3950 mutex_unlock(&fs_info->reclaim_bgs_lock); 3951 goto loop; 3952 } 3953 3954 if (counting) { 3955 mutex_unlock(&fs_info->reclaim_bgs_lock); 3956 spin_lock(&fs_info->balance_lock); 3957 bctl->stat.expected++; 3958 spin_unlock(&fs_info->balance_lock); 3959 3960 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3961 count_data++; 3962 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3963 count_sys++; 3964 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3965 count_meta++; 3966 3967 goto loop; 3968 } 3969 3970 /* 3971 * Apply limit_min filter, no need to check if the LIMITS 3972 * filter is used, limit_min is 0 by default 3973 */ 3974 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3975 count_data < bctl->data.limit_min) 3976 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3977 count_meta < bctl->meta.limit_min) 3978 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3979 count_sys < bctl->sys.limit_min)) { 3980 mutex_unlock(&fs_info->reclaim_bgs_lock); 3981 goto loop; 3982 } 3983 3984 if (!chunk_reserved) { 3985 /* 3986 * We may be relocating the only data chunk we have, 3987 * which could potentially end up with losing data's 3988 * raid profile, so lets allocate an empty one in 3989 * advance. 3990 */ 3991 ret = btrfs_may_alloc_data_chunk(fs_info, 3992 found_key.offset); 3993 if (ret < 0) { 3994 mutex_unlock(&fs_info->reclaim_bgs_lock); 3995 goto error; 3996 } else if (ret == 1) { 3997 chunk_reserved = 1; 3998 } 3999 } 4000 4001 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 4002 mutex_unlock(&fs_info->reclaim_bgs_lock); 4003 if (ret == -ENOSPC) { 4004 enospc_errors++; 4005 } else if (ret == -ETXTBSY) { 4006 btrfs_info(fs_info, 4007 "skipping relocation of block group %llu due to active swapfile", 4008 found_key.offset); 4009 ret = 0; 4010 } else if (ret) { 4011 goto error; 4012 } else { 4013 spin_lock(&fs_info->balance_lock); 4014 bctl->stat.completed++; 4015 spin_unlock(&fs_info->balance_lock); 4016 } 4017 loop: 4018 if (found_key.offset == 0) 4019 break; 4020 key.offset = found_key.offset - 1; 4021 } 4022 4023 if (counting) { 4024 btrfs_release_path(path); 4025 counting = false; 4026 goto again; 4027 } 4028 error: 4029 btrfs_free_path(path); 4030 if (enospc_errors) { 4031 btrfs_info(fs_info, "%d enospc errors during balance", 4032 enospc_errors); 4033 if (!ret) 4034 ret = -ENOSPC; 4035 } 4036 4037 return ret; 4038 } 4039 4040 /* 4041 * See if a given profile is valid and reduced. 4042 * 4043 * @flags: profile to validate 4044 * @extended: if true @flags is treated as an extended profile 4045 */ 4046 static int alloc_profile_is_valid(u64 flags, int extended) 4047 { 4048 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 4049 BTRFS_BLOCK_GROUP_PROFILE_MASK); 4050 4051 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 4052 4053 /* 1) check that all other bits are zeroed */ 4054 if (flags & ~mask) 4055 return 0; 4056 4057 /* 2) see if profile is reduced */ 4058 if (flags == 0) 4059 return !extended; /* "0" is valid for usual profiles */ 4060 4061 return has_single_bit_set(flags); 4062 } 4063 4064 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4065 { 4066 /* cancel requested || normal exit path */ 4067 return atomic_read(&fs_info->balance_cancel_req) || 4068 (atomic_read(&fs_info->balance_pause_req) == 0 && 4069 atomic_read(&fs_info->balance_cancel_req) == 0); 4070 } 4071 4072 /* 4073 * Validate target profile against allowed profiles and return true if it's OK. 4074 * Otherwise print the error message and return false. 4075 */ 4076 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4077 const struct btrfs_balance_args *bargs, 4078 u64 allowed, const char *type) 4079 { 4080 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4081 return true; 4082 4083 /* Profile is valid and does not have bits outside of the allowed set */ 4084 if (alloc_profile_is_valid(bargs->target, 1) && 4085 (bargs->target & ~allowed) == 0) 4086 return true; 4087 4088 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4089 type, btrfs_bg_type_to_raid_name(bargs->target)); 4090 return false; 4091 } 4092 4093 /* 4094 * Fill @buf with textual description of balance filter flags @bargs, up to 4095 * @size_buf including the terminating null. The output may be trimmed if it 4096 * does not fit into the provided buffer. 4097 */ 4098 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4099 u32 size_buf) 4100 { 4101 int ret; 4102 u32 size_bp = size_buf; 4103 char *bp = buf; 4104 u64 flags = bargs->flags; 4105 char tmp_buf[128] = {'\0'}; 4106 4107 if (!flags) 4108 return; 4109 4110 #define CHECK_APPEND_NOARG(a) \ 4111 do { \ 4112 ret = snprintf(bp, size_bp, (a)); \ 4113 if (ret < 0 || ret >= size_bp) \ 4114 goto out_overflow; \ 4115 size_bp -= ret; \ 4116 bp += ret; \ 4117 } while (0) 4118 4119 #define CHECK_APPEND_1ARG(a, v1) \ 4120 do { \ 4121 ret = snprintf(bp, size_bp, (a), (v1)); \ 4122 if (ret < 0 || ret >= size_bp) \ 4123 goto out_overflow; \ 4124 size_bp -= ret; \ 4125 bp += ret; \ 4126 } while (0) 4127 4128 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4129 do { \ 4130 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4131 if (ret < 0 || ret >= size_bp) \ 4132 goto out_overflow; \ 4133 size_bp -= ret; \ 4134 bp += ret; \ 4135 } while (0) 4136 4137 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4138 CHECK_APPEND_1ARG("convert=%s,", 4139 btrfs_bg_type_to_raid_name(bargs->target)); 4140 4141 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4142 CHECK_APPEND_NOARG("soft,"); 4143 4144 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4145 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4146 sizeof(tmp_buf)); 4147 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4148 } 4149 4150 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4151 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4152 4153 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4154 CHECK_APPEND_2ARG("usage=%u..%u,", 4155 bargs->usage_min, bargs->usage_max); 4156 4157 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4158 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4159 4160 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4161 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4162 bargs->pstart, bargs->pend); 4163 4164 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4165 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4166 bargs->vstart, bargs->vend); 4167 4168 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4169 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4170 4171 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4172 CHECK_APPEND_2ARG("limit=%u..%u,", 4173 bargs->limit_min, bargs->limit_max); 4174 4175 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4176 CHECK_APPEND_2ARG("stripes=%u..%u,", 4177 bargs->stripes_min, bargs->stripes_max); 4178 4179 #undef CHECK_APPEND_2ARG 4180 #undef CHECK_APPEND_1ARG 4181 #undef CHECK_APPEND_NOARG 4182 4183 out_overflow: 4184 4185 if (size_bp < size_buf) 4186 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4187 else 4188 buf[0] = '\0'; 4189 } 4190 4191 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4192 { 4193 u32 size_buf = 1024; 4194 char tmp_buf[192] = {'\0'}; 4195 char *buf; 4196 char *bp; 4197 u32 size_bp = size_buf; 4198 int ret; 4199 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4200 4201 buf = kzalloc(size_buf, GFP_KERNEL); 4202 if (!buf) 4203 return; 4204 4205 bp = buf; 4206 4207 #define CHECK_APPEND_1ARG(a, v1) \ 4208 do { \ 4209 ret = snprintf(bp, size_bp, (a), (v1)); \ 4210 if (ret < 0 || ret >= size_bp) \ 4211 goto out_overflow; \ 4212 size_bp -= ret; \ 4213 bp += ret; \ 4214 } while (0) 4215 4216 if (bctl->flags & BTRFS_BALANCE_FORCE) 4217 CHECK_APPEND_1ARG("%s", "-f "); 4218 4219 if (bctl->flags & BTRFS_BALANCE_DATA) { 4220 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4221 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4222 } 4223 4224 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4225 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4226 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4227 } 4228 4229 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4230 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4231 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4232 } 4233 4234 #undef CHECK_APPEND_1ARG 4235 4236 out_overflow: 4237 4238 if (size_bp < size_buf) 4239 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4240 btrfs_info(fs_info, "balance: %s %s", 4241 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4242 "resume" : "start", buf); 4243 4244 kfree(buf); 4245 } 4246 4247 /* 4248 * Should be called with balance mutexe held 4249 */ 4250 int btrfs_balance(struct btrfs_fs_info *fs_info, 4251 struct btrfs_balance_control *bctl, 4252 struct btrfs_ioctl_balance_args *bargs) 4253 { 4254 u64 meta_target, data_target; 4255 u64 allowed; 4256 int mixed = 0; 4257 int ret; 4258 u64 num_devices; 4259 unsigned seq; 4260 bool reducing_redundancy; 4261 int i; 4262 4263 if (btrfs_fs_closing(fs_info) || 4264 atomic_read(&fs_info->balance_pause_req) || 4265 btrfs_should_cancel_balance(fs_info)) { 4266 ret = -EINVAL; 4267 goto out; 4268 } 4269 4270 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4271 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4272 mixed = 1; 4273 4274 /* 4275 * In case of mixed groups both data and meta should be picked, 4276 * and identical options should be given for both of them. 4277 */ 4278 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4279 if (mixed && (bctl->flags & allowed)) { 4280 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4281 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4282 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4283 btrfs_err(fs_info, 4284 "balance: mixed groups data and metadata options must be the same"); 4285 ret = -EINVAL; 4286 goto out; 4287 } 4288 } 4289 4290 /* 4291 * rw_devices will not change at the moment, device add/delete/replace 4292 * are exclusive 4293 */ 4294 num_devices = fs_info->fs_devices->rw_devices; 4295 4296 /* 4297 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4298 * special bit for it, to make it easier to distinguish. Thus we need 4299 * to set it manually, or balance would refuse the profile. 4300 */ 4301 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4302 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4303 if (num_devices >= btrfs_raid_array[i].devs_min) 4304 allowed |= btrfs_raid_array[i].bg_flag; 4305 4306 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4307 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4308 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4309 ret = -EINVAL; 4310 goto out; 4311 } 4312 4313 /* 4314 * Allow to reduce metadata or system integrity only if force set for 4315 * profiles with redundancy (copies, parity) 4316 */ 4317 allowed = 0; 4318 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4319 if (btrfs_raid_array[i].ncopies >= 2 || 4320 btrfs_raid_array[i].tolerated_failures >= 1) 4321 allowed |= btrfs_raid_array[i].bg_flag; 4322 } 4323 do { 4324 seq = read_seqbegin(&fs_info->profiles_lock); 4325 4326 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4327 (fs_info->avail_system_alloc_bits & allowed) && 4328 !(bctl->sys.target & allowed)) || 4329 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4330 (fs_info->avail_metadata_alloc_bits & allowed) && 4331 !(bctl->meta.target & allowed))) 4332 reducing_redundancy = true; 4333 else 4334 reducing_redundancy = false; 4335 4336 /* if we're not converting, the target field is uninitialized */ 4337 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4338 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4339 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4340 bctl->data.target : fs_info->avail_data_alloc_bits; 4341 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4342 4343 if (reducing_redundancy) { 4344 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4345 btrfs_info(fs_info, 4346 "balance: force reducing metadata redundancy"); 4347 } else { 4348 btrfs_err(fs_info, 4349 "balance: reduces metadata redundancy, use --force if you want this"); 4350 ret = -EINVAL; 4351 goto out; 4352 } 4353 } 4354 4355 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4356 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4357 btrfs_warn(fs_info, 4358 "balance: metadata profile %s has lower redundancy than data profile %s", 4359 btrfs_bg_type_to_raid_name(meta_target), 4360 btrfs_bg_type_to_raid_name(data_target)); 4361 } 4362 4363 ret = insert_balance_item(fs_info, bctl); 4364 if (ret && ret != -EEXIST) 4365 goto out; 4366 4367 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4368 BUG_ON(ret == -EEXIST); 4369 BUG_ON(fs_info->balance_ctl); 4370 spin_lock(&fs_info->balance_lock); 4371 fs_info->balance_ctl = bctl; 4372 spin_unlock(&fs_info->balance_lock); 4373 } else { 4374 BUG_ON(ret != -EEXIST); 4375 spin_lock(&fs_info->balance_lock); 4376 update_balance_args(bctl); 4377 spin_unlock(&fs_info->balance_lock); 4378 } 4379 4380 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4381 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4382 describe_balance_start_or_resume(fs_info); 4383 mutex_unlock(&fs_info->balance_mutex); 4384 4385 ret = __btrfs_balance(fs_info); 4386 4387 mutex_lock(&fs_info->balance_mutex); 4388 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4389 btrfs_info(fs_info, "balance: paused"); 4390 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4391 } 4392 /* 4393 * Balance can be canceled by: 4394 * 4395 * - Regular cancel request 4396 * Then ret == -ECANCELED and balance_cancel_req > 0 4397 * 4398 * - Fatal signal to "btrfs" process 4399 * Either the signal caught by wait_reserve_ticket() and callers 4400 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4401 * got -ECANCELED. 4402 * Either way, in this case balance_cancel_req = 0, and 4403 * ret == -EINTR or ret == -ECANCELED. 4404 * 4405 * So here we only check the return value to catch canceled balance. 4406 */ 4407 else if (ret == -ECANCELED || ret == -EINTR) 4408 btrfs_info(fs_info, "balance: canceled"); 4409 else 4410 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4411 4412 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4413 4414 if (bargs) { 4415 memset(bargs, 0, sizeof(*bargs)); 4416 btrfs_update_ioctl_balance_args(fs_info, bargs); 4417 } 4418 4419 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4420 balance_need_close(fs_info)) { 4421 reset_balance_state(fs_info); 4422 btrfs_exclop_finish(fs_info); 4423 } 4424 4425 wake_up(&fs_info->balance_wait_q); 4426 4427 return ret; 4428 out: 4429 if (bctl->flags & BTRFS_BALANCE_RESUME) 4430 reset_balance_state(fs_info); 4431 else 4432 kfree(bctl); 4433 btrfs_exclop_finish(fs_info); 4434 4435 return ret; 4436 } 4437 4438 static int balance_kthread(void *data) 4439 { 4440 struct btrfs_fs_info *fs_info = data; 4441 int ret = 0; 4442 4443 sb_start_write(fs_info->sb); 4444 mutex_lock(&fs_info->balance_mutex); 4445 if (fs_info->balance_ctl) 4446 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4447 mutex_unlock(&fs_info->balance_mutex); 4448 sb_end_write(fs_info->sb); 4449 4450 return ret; 4451 } 4452 4453 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4454 { 4455 struct task_struct *tsk; 4456 4457 mutex_lock(&fs_info->balance_mutex); 4458 if (!fs_info->balance_ctl) { 4459 mutex_unlock(&fs_info->balance_mutex); 4460 return 0; 4461 } 4462 mutex_unlock(&fs_info->balance_mutex); 4463 4464 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4465 btrfs_info(fs_info, "balance: resume skipped"); 4466 return 0; 4467 } 4468 4469 spin_lock(&fs_info->super_lock); 4470 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4471 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4472 spin_unlock(&fs_info->super_lock); 4473 /* 4474 * A ro->rw remount sequence should continue with the paused balance 4475 * regardless of who pauses it, system or the user as of now, so set 4476 * the resume flag. 4477 */ 4478 spin_lock(&fs_info->balance_lock); 4479 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4480 spin_unlock(&fs_info->balance_lock); 4481 4482 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4483 return PTR_ERR_OR_ZERO(tsk); 4484 } 4485 4486 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4487 { 4488 struct btrfs_balance_control *bctl; 4489 struct btrfs_balance_item *item; 4490 struct btrfs_disk_balance_args disk_bargs; 4491 struct btrfs_path *path; 4492 struct extent_buffer *leaf; 4493 struct btrfs_key key; 4494 int ret; 4495 4496 path = btrfs_alloc_path(); 4497 if (!path) 4498 return -ENOMEM; 4499 4500 key.objectid = BTRFS_BALANCE_OBJECTID; 4501 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4502 key.offset = 0; 4503 4504 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4505 if (ret < 0) 4506 goto out; 4507 if (ret > 0) { /* ret = -ENOENT; */ 4508 ret = 0; 4509 goto out; 4510 } 4511 4512 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4513 if (!bctl) { 4514 ret = -ENOMEM; 4515 goto out; 4516 } 4517 4518 leaf = path->nodes[0]; 4519 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4520 4521 bctl->flags = btrfs_balance_flags(leaf, item); 4522 bctl->flags |= BTRFS_BALANCE_RESUME; 4523 4524 btrfs_balance_data(leaf, item, &disk_bargs); 4525 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4526 btrfs_balance_meta(leaf, item, &disk_bargs); 4527 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4528 btrfs_balance_sys(leaf, item, &disk_bargs); 4529 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4530 4531 /* 4532 * This should never happen, as the paused balance state is recovered 4533 * during mount without any chance of other exclusive ops to collide. 4534 * 4535 * This gives the exclusive op status to balance and keeps in paused 4536 * state until user intervention (cancel or umount). If the ownership 4537 * cannot be assigned, show a message but do not fail. The balance 4538 * is in a paused state and must have fs_info::balance_ctl properly 4539 * set up. 4540 */ 4541 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4542 btrfs_warn(fs_info, 4543 "balance: cannot set exclusive op status, resume manually"); 4544 4545 btrfs_release_path(path); 4546 4547 mutex_lock(&fs_info->balance_mutex); 4548 BUG_ON(fs_info->balance_ctl); 4549 spin_lock(&fs_info->balance_lock); 4550 fs_info->balance_ctl = bctl; 4551 spin_unlock(&fs_info->balance_lock); 4552 mutex_unlock(&fs_info->balance_mutex); 4553 out: 4554 btrfs_free_path(path); 4555 return ret; 4556 } 4557 4558 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4559 { 4560 int ret = 0; 4561 4562 mutex_lock(&fs_info->balance_mutex); 4563 if (!fs_info->balance_ctl) { 4564 mutex_unlock(&fs_info->balance_mutex); 4565 return -ENOTCONN; 4566 } 4567 4568 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4569 atomic_inc(&fs_info->balance_pause_req); 4570 mutex_unlock(&fs_info->balance_mutex); 4571 4572 wait_event(fs_info->balance_wait_q, 4573 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4574 4575 mutex_lock(&fs_info->balance_mutex); 4576 /* we are good with balance_ctl ripped off from under us */ 4577 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4578 atomic_dec(&fs_info->balance_pause_req); 4579 } else { 4580 ret = -ENOTCONN; 4581 } 4582 4583 mutex_unlock(&fs_info->balance_mutex); 4584 return ret; 4585 } 4586 4587 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4588 { 4589 mutex_lock(&fs_info->balance_mutex); 4590 if (!fs_info->balance_ctl) { 4591 mutex_unlock(&fs_info->balance_mutex); 4592 return -ENOTCONN; 4593 } 4594 4595 /* 4596 * A paused balance with the item stored on disk can be resumed at 4597 * mount time if the mount is read-write. Otherwise it's still paused 4598 * and we must not allow cancelling as it deletes the item. 4599 */ 4600 if (sb_rdonly(fs_info->sb)) { 4601 mutex_unlock(&fs_info->balance_mutex); 4602 return -EROFS; 4603 } 4604 4605 atomic_inc(&fs_info->balance_cancel_req); 4606 /* 4607 * if we are running just wait and return, balance item is 4608 * deleted in btrfs_balance in this case 4609 */ 4610 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4611 mutex_unlock(&fs_info->balance_mutex); 4612 wait_event(fs_info->balance_wait_q, 4613 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4614 mutex_lock(&fs_info->balance_mutex); 4615 } else { 4616 mutex_unlock(&fs_info->balance_mutex); 4617 /* 4618 * Lock released to allow other waiters to continue, we'll 4619 * reexamine the status again. 4620 */ 4621 mutex_lock(&fs_info->balance_mutex); 4622 4623 if (fs_info->balance_ctl) { 4624 reset_balance_state(fs_info); 4625 btrfs_exclop_finish(fs_info); 4626 btrfs_info(fs_info, "balance: canceled"); 4627 } 4628 } 4629 4630 BUG_ON(fs_info->balance_ctl || 4631 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4632 atomic_dec(&fs_info->balance_cancel_req); 4633 mutex_unlock(&fs_info->balance_mutex); 4634 return 0; 4635 } 4636 4637 int btrfs_uuid_scan_kthread(void *data) 4638 { 4639 struct btrfs_fs_info *fs_info = data; 4640 struct btrfs_root *root = fs_info->tree_root; 4641 struct btrfs_key key; 4642 struct btrfs_path *path = NULL; 4643 int ret = 0; 4644 struct extent_buffer *eb; 4645 int slot; 4646 struct btrfs_root_item root_item; 4647 u32 item_size; 4648 struct btrfs_trans_handle *trans = NULL; 4649 bool closing = false; 4650 4651 path = btrfs_alloc_path(); 4652 if (!path) { 4653 ret = -ENOMEM; 4654 goto out; 4655 } 4656 4657 key.objectid = 0; 4658 key.type = BTRFS_ROOT_ITEM_KEY; 4659 key.offset = 0; 4660 4661 while (1) { 4662 if (btrfs_fs_closing(fs_info)) { 4663 closing = true; 4664 break; 4665 } 4666 ret = btrfs_search_forward(root, &key, path, 4667 BTRFS_OLDEST_GENERATION); 4668 if (ret) { 4669 if (ret > 0) 4670 ret = 0; 4671 break; 4672 } 4673 4674 if (key.type != BTRFS_ROOT_ITEM_KEY || 4675 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4676 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4677 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4678 goto skip; 4679 4680 eb = path->nodes[0]; 4681 slot = path->slots[0]; 4682 item_size = btrfs_item_size(eb, slot); 4683 if (item_size < sizeof(root_item)) 4684 goto skip; 4685 4686 read_extent_buffer(eb, &root_item, 4687 btrfs_item_ptr_offset(eb, slot), 4688 (int)sizeof(root_item)); 4689 if (btrfs_root_refs(&root_item) == 0) 4690 goto skip; 4691 4692 if (!btrfs_is_empty_uuid(root_item.uuid) || 4693 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4694 if (trans) 4695 goto update_tree; 4696 4697 btrfs_release_path(path); 4698 /* 4699 * 1 - subvol uuid item 4700 * 1 - received_subvol uuid item 4701 */ 4702 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4703 if (IS_ERR(trans)) { 4704 ret = PTR_ERR(trans); 4705 break; 4706 } 4707 continue; 4708 } else { 4709 goto skip; 4710 } 4711 update_tree: 4712 btrfs_release_path(path); 4713 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4714 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4715 BTRFS_UUID_KEY_SUBVOL, 4716 key.objectid); 4717 if (ret < 0) { 4718 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4719 ret); 4720 break; 4721 } 4722 } 4723 4724 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4725 ret = btrfs_uuid_tree_add(trans, 4726 root_item.received_uuid, 4727 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4728 key.objectid); 4729 if (ret < 0) { 4730 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4731 ret); 4732 break; 4733 } 4734 } 4735 4736 skip: 4737 btrfs_release_path(path); 4738 if (trans) { 4739 ret = btrfs_end_transaction(trans); 4740 trans = NULL; 4741 if (ret) 4742 break; 4743 } 4744 4745 if (key.offset < (u64)-1) { 4746 key.offset++; 4747 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4748 key.offset = 0; 4749 key.type = BTRFS_ROOT_ITEM_KEY; 4750 } else if (key.objectid < (u64)-1) { 4751 key.offset = 0; 4752 key.type = BTRFS_ROOT_ITEM_KEY; 4753 key.objectid++; 4754 } else { 4755 break; 4756 } 4757 cond_resched(); 4758 } 4759 4760 out: 4761 btrfs_free_path(path); 4762 if (trans && !IS_ERR(trans)) 4763 btrfs_end_transaction(trans); 4764 if (ret) 4765 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4766 else if (!closing) 4767 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4768 up(&fs_info->uuid_tree_rescan_sem); 4769 return 0; 4770 } 4771 4772 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4773 { 4774 struct btrfs_trans_handle *trans; 4775 struct btrfs_root *tree_root = fs_info->tree_root; 4776 struct btrfs_root *uuid_root; 4777 struct task_struct *task; 4778 int ret; 4779 4780 /* 4781 * 1 - root node 4782 * 1 - root item 4783 */ 4784 trans = btrfs_start_transaction(tree_root, 2); 4785 if (IS_ERR(trans)) 4786 return PTR_ERR(trans); 4787 4788 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4789 if (IS_ERR(uuid_root)) { 4790 ret = PTR_ERR(uuid_root); 4791 btrfs_abort_transaction(trans, ret); 4792 btrfs_end_transaction(trans); 4793 return ret; 4794 } 4795 4796 fs_info->uuid_root = uuid_root; 4797 4798 ret = btrfs_commit_transaction(trans); 4799 if (ret) 4800 return ret; 4801 4802 down(&fs_info->uuid_tree_rescan_sem); 4803 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4804 if (IS_ERR(task)) { 4805 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4806 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4807 up(&fs_info->uuid_tree_rescan_sem); 4808 return PTR_ERR(task); 4809 } 4810 4811 return 0; 4812 } 4813 4814 /* 4815 * shrinking a device means finding all of the device extents past 4816 * the new size, and then following the back refs to the chunks. 4817 * The chunk relocation code actually frees the device extent 4818 */ 4819 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4820 { 4821 struct btrfs_fs_info *fs_info = device->fs_info; 4822 struct btrfs_root *root = fs_info->dev_root; 4823 struct btrfs_trans_handle *trans; 4824 struct btrfs_dev_extent *dev_extent = NULL; 4825 struct btrfs_path *path; 4826 u64 length; 4827 u64 chunk_offset; 4828 int ret; 4829 int slot; 4830 int failed = 0; 4831 bool retried = false; 4832 struct extent_buffer *l; 4833 struct btrfs_key key; 4834 struct btrfs_super_block *super_copy = fs_info->super_copy; 4835 u64 old_total = btrfs_super_total_bytes(super_copy); 4836 u64 old_size = btrfs_device_get_total_bytes(device); 4837 u64 diff; 4838 u64 start; 4839 4840 new_size = round_down(new_size, fs_info->sectorsize); 4841 start = new_size; 4842 diff = round_down(old_size - new_size, fs_info->sectorsize); 4843 4844 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4845 return -EINVAL; 4846 4847 path = btrfs_alloc_path(); 4848 if (!path) 4849 return -ENOMEM; 4850 4851 path->reada = READA_BACK; 4852 4853 trans = btrfs_start_transaction(root, 0); 4854 if (IS_ERR(trans)) { 4855 btrfs_free_path(path); 4856 return PTR_ERR(trans); 4857 } 4858 4859 mutex_lock(&fs_info->chunk_mutex); 4860 4861 btrfs_device_set_total_bytes(device, new_size); 4862 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4863 device->fs_devices->total_rw_bytes -= diff; 4864 atomic64_sub(diff, &fs_info->free_chunk_space); 4865 } 4866 4867 /* 4868 * Once the device's size has been set to the new size, ensure all 4869 * in-memory chunks are synced to disk so that the loop below sees them 4870 * and relocates them accordingly. 4871 */ 4872 if (contains_pending_extent(device, &start, diff)) { 4873 mutex_unlock(&fs_info->chunk_mutex); 4874 ret = btrfs_commit_transaction(trans); 4875 if (ret) 4876 goto done; 4877 } else { 4878 mutex_unlock(&fs_info->chunk_mutex); 4879 btrfs_end_transaction(trans); 4880 } 4881 4882 again: 4883 key.objectid = device->devid; 4884 key.offset = (u64)-1; 4885 key.type = BTRFS_DEV_EXTENT_KEY; 4886 4887 do { 4888 mutex_lock(&fs_info->reclaim_bgs_lock); 4889 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4890 if (ret < 0) { 4891 mutex_unlock(&fs_info->reclaim_bgs_lock); 4892 goto done; 4893 } 4894 4895 ret = btrfs_previous_item(root, path, 0, key.type); 4896 if (ret) { 4897 mutex_unlock(&fs_info->reclaim_bgs_lock); 4898 if (ret < 0) 4899 goto done; 4900 ret = 0; 4901 btrfs_release_path(path); 4902 break; 4903 } 4904 4905 l = path->nodes[0]; 4906 slot = path->slots[0]; 4907 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4908 4909 if (key.objectid != device->devid) { 4910 mutex_unlock(&fs_info->reclaim_bgs_lock); 4911 btrfs_release_path(path); 4912 break; 4913 } 4914 4915 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4916 length = btrfs_dev_extent_length(l, dev_extent); 4917 4918 if (key.offset + length <= new_size) { 4919 mutex_unlock(&fs_info->reclaim_bgs_lock); 4920 btrfs_release_path(path); 4921 break; 4922 } 4923 4924 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4925 btrfs_release_path(path); 4926 4927 /* 4928 * We may be relocating the only data chunk we have, 4929 * which could potentially end up with losing data's 4930 * raid profile, so lets allocate an empty one in 4931 * advance. 4932 */ 4933 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4934 if (ret < 0) { 4935 mutex_unlock(&fs_info->reclaim_bgs_lock); 4936 goto done; 4937 } 4938 4939 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4940 mutex_unlock(&fs_info->reclaim_bgs_lock); 4941 if (ret == -ENOSPC) { 4942 failed++; 4943 } else if (ret) { 4944 if (ret == -ETXTBSY) { 4945 btrfs_warn(fs_info, 4946 "could not shrink block group %llu due to active swapfile", 4947 chunk_offset); 4948 } 4949 goto done; 4950 } 4951 } while (key.offset-- > 0); 4952 4953 if (failed && !retried) { 4954 failed = 0; 4955 retried = true; 4956 goto again; 4957 } else if (failed && retried) { 4958 ret = -ENOSPC; 4959 goto done; 4960 } 4961 4962 /* Shrinking succeeded, else we would be at "done". */ 4963 trans = btrfs_start_transaction(root, 0); 4964 if (IS_ERR(trans)) { 4965 ret = PTR_ERR(trans); 4966 goto done; 4967 } 4968 4969 mutex_lock(&fs_info->chunk_mutex); 4970 /* Clear all state bits beyond the shrunk device size */ 4971 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4972 CHUNK_STATE_MASK); 4973 4974 btrfs_device_set_disk_total_bytes(device, new_size); 4975 if (list_empty(&device->post_commit_list)) 4976 list_add_tail(&device->post_commit_list, 4977 &trans->transaction->dev_update_list); 4978 4979 WARN_ON(diff > old_total); 4980 btrfs_set_super_total_bytes(super_copy, 4981 round_down(old_total - diff, fs_info->sectorsize)); 4982 mutex_unlock(&fs_info->chunk_mutex); 4983 4984 btrfs_reserve_chunk_metadata(trans, false); 4985 /* Now btrfs_update_device() will change the on-disk size. */ 4986 ret = btrfs_update_device(trans, device); 4987 btrfs_trans_release_chunk_metadata(trans); 4988 if (ret < 0) { 4989 btrfs_abort_transaction(trans, ret); 4990 btrfs_end_transaction(trans); 4991 } else { 4992 ret = btrfs_commit_transaction(trans); 4993 } 4994 done: 4995 btrfs_free_path(path); 4996 if (ret) { 4997 mutex_lock(&fs_info->chunk_mutex); 4998 btrfs_device_set_total_bytes(device, old_size); 4999 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 5000 device->fs_devices->total_rw_bytes += diff; 5001 atomic64_add(diff, &fs_info->free_chunk_space); 5002 mutex_unlock(&fs_info->chunk_mutex); 5003 } 5004 return ret; 5005 } 5006 5007 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 5008 struct btrfs_key *key, 5009 struct btrfs_chunk *chunk, int item_size) 5010 { 5011 struct btrfs_super_block *super_copy = fs_info->super_copy; 5012 struct btrfs_disk_key disk_key; 5013 u32 array_size; 5014 u8 *ptr; 5015 5016 lockdep_assert_held(&fs_info->chunk_mutex); 5017 5018 array_size = btrfs_super_sys_array_size(super_copy); 5019 if (array_size + item_size + sizeof(disk_key) 5020 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5021 return -EFBIG; 5022 5023 ptr = super_copy->sys_chunk_array + array_size; 5024 btrfs_cpu_key_to_disk(&disk_key, key); 5025 memcpy(ptr, &disk_key, sizeof(disk_key)); 5026 ptr += sizeof(disk_key); 5027 memcpy(ptr, chunk, item_size); 5028 item_size += sizeof(disk_key); 5029 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5030 5031 return 0; 5032 } 5033 5034 /* 5035 * sort the devices in descending order by max_avail, total_avail 5036 */ 5037 static int btrfs_cmp_device_info(const void *a, const void *b) 5038 { 5039 const struct btrfs_device_info *di_a = a; 5040 const struct btrfs_device_info *di_b = b; 5041 5042 if (di_a->max_avail > di_b->max_avail) 5043 return -1; 5044 if (di_a->max_avail < di_b->max_avail) 5045 return 1; 5046 if (di_a->total_avail > di_b->total_avail) 5047 return -1; 5048 if (di_a->total_avail < di_b->total_avail) 5049 return 1; 5050 return 0; 5051 } 5052 5053 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5054 { 5055 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5056 return; 5057 5058 btrfs_set_fs_incompat(info, RAID56); 5059 } 5060 5061 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5062 { 5063 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5064 return; 5065 5066 btrfs_set_fs_incompat(info, RAID1C34); 5067 } 5068 5069 /* 5070 * Structure used internally for btrfs_create_chunk() function. 5071 * Wraps needed parameters. 5072 */ 5073 struct alloc_chunk_ctl { 5074 u64 start; 5075 u64 type; 5076 /* Total number of stripes to allocate */ 5077 int num_stripes; 5078 /* sub_stripes info for map */ 5079 int sub_stripes; 5080 /* Stripes per device */ 5081 int dev_stripes; 5082 /* Maximum number of devices to use */ 5083 int devs_max; 5084 /* Minimum number of devices to use */ 5085 int devs_min; 5086 /* ndevs has to be a multiple of this */ 5087 int devs_increment; 5088 /* Number of copies */ 5089 int ncopies; 5090 /* Number of stripes worth of bytes to store parity information */ 5091 int nparity; 5092 u64 max_stripe_size; 5093 u64 max_chunk_size; 5094 u64 dev_extent_min; 5095 u64 stripe_size; 5096 u64 chunk_size; 5097 int ndevs; 5098 }; 5099 5100 static void init_alloc_chunk_ctl_policy_regular( 5101 struct btrfs_fs_devices *fs_devices, 5102 struct alloc_chunk_ctl *ctl) 5103 { 5104 struct btrfs_space_info *space_info; 5105 5106 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); 5107 ASSERT(space_info); 5108 5109 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); 5110 ctl->max_stripe_size = ctl->max_chunk_size; 5111 5112 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) 5113 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); 5114 5115 /* We don't want a chunk larger than 10% of writable space */ 5116 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5117 ctl->max_chunk_size); 5118 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5119 } 5120 5121 static void init_alloc_chunk_ctl_policy_zoned( 5122 struct btrfs_fs_devices *fs_devices, 5123 struct alloc_chunk_ctl *ctl) 5124 { 5125 u64 zone_size = fs_devices->fs_info->zone_size; 5126 u64 limit; 5127 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5128 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5129 u64 min_chunk_size = min_data_stripes * zone_size; 5130 u64 type = ctl->type; 5131 5132 ctl->max_stripe_size = zone_size; 5133 if (type & BTRFS_BLOCK_GROUP_DATA) { 5134 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5135 zone_size); 5136 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5137 ctl->max_chunk_size = ctl->max_stripe_size; 5138 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5139 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5140 ctl->devs_max = min_t(int, ctl->devs_max, 5141 BTRFS_MAX_DEVS_SYS_CHUNK); 5142 } else { 5143 BUG(); 5144 } 5145 5146 /* We don't want a chunk larger than 10% of writable space */ 5147 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5148 zone_size), 5149 min_chunk_size); 5150 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5151 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5152 } 5153 5154 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5155 struct alloc_chunk_ctl *ctl) 5156 { 5157 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5158 5159 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5160 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5161 ctl->devs_max = btrfs_raid_array[index].devs_max; 5162 if (!ctl->devs_max) 5163 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5164 ctl->devs_min = btrfs_raid_array[index].devs_min; 5165 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5166 ctl->ncopies = btrfs_raid_array[index].ncopies; 5167 ctl->nparity = btrfs_raid_array[index].nparity; 5168 ctl->ndevs = 0; 5169 5170 switch (fs_devices->chunk_alloc_policy) { 5171 case BTRFS_CHUNK_ALLOC_REGULAR: 5172 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5173 break; 5174 case BTRFS_CHUNK_ALLOC_ZONED: 5175 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5176 break; 5177 default: 5178 BUG(); 5179 } 5180 } 5181 5182 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5183 struct alloc_chunk_ctl *ctl, 5184 struct btrfs_device_info *devices_info) 5185 { 5186 struct btrfs_fs_info *info = fs_devices->fs_info; 5187 struct btrfs_device *device; 5188 u64 total_avail; 5189 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5190 int ret; 5191 int ndevs = 0; 5192 u64 max_avail; 5193 u64 dev_offset; 5194 5195 /* 5196 * in the first pass through the devices list, we gather information 5197 * about the available holes on each device. 5198 */ 5199 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5200 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5201 WARN(1, KERN_ERR 5202 "BTRFS: read-only device in alloc_list\n"); 5203 continue; 5204 } 5205 5206 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5207 &device->dev_state) || 5208 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5209 continue; 5210 5211 if (device->total_bytes > device->bytes_used) 5212 total_avail = device->total_bytes - device->bytes_used; 5213 else 5214 total_avail = 0; 5215 5216 /* If there is no space on this device, skip it. */ 5217 if (total_avail < ctl->dev_extent_min) 5218 continue; 5219 5220 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5221 &max_avail); 5222 if (ret && ret != -ENOSPC) 5223 return ret; 5224 5225 if (ret == 0) 5226 max_avail = dev_extent_want; 5227 5228 if (max_avail < ctl->dev_extent_min) { 5229 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5230 btrfs_debug(info, 5231 "%s: devid %llu has no free space, have=%llu want=%llu", 5232 __func__, device->devid, max_avail, 5233 ctl->dev_extent_min); 5234 continue; 5235 } 5236 5237 if (ndevs == fs_devices->rw_devices) { 5238 WARN(1, "%s: found more than %llu devices\n", 5239 __func__, fs_devices->rw_devices); 5240 break; 5241 } 5242 devices_info[ndevs].dev_offset = dev_offset; 5243 devices_info[ndevs].max_avail = max_avail; 5244 devices_info[ndevs].total_avail = total_avail; 5245 devices_info[ndevs].dev = device; 5246 ++ndevs; 5247 } 5248 ctl->ndevs = ndevs; 5249 5250 /* 5251 * now sort the devices by hole size / available space 5252 */ 5253 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5254 btrfs_cmp_device_info, NULL); 5255 5256 return 0; 5257 } 5258 5259 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5260 struct btrfs_device_info *devices_info) 5261 { 5262 /* Number of stripes that count for block group size */ 5263 int data_stripes; 5264 5265 /* 5266 * The primary goal is to maximize the number of stripes, so use as 5267 * many devices as possible, even if the stripes are not maximum sized. 5268 * 5269 * The DUP profile stores more than one stripe per device, the 5270 * max_avail is the total size so we have to adjust. 5271 */ 5272 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5273 ctl->dev_stripes); 5274 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5275 5276 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5277 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5278 5279 /* 5280 * Use the number of data stripes to figure out how big this chunk is 5281 * really going to be in terms of logical address space, and compare 5282 * that answer with the max chunk size. If it's higher, we try to 5283 * reduce stripe_size. 5284 */ 5285 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5286 /* 5287 * Reduce stripe_size, round it up to a 16MB boundary again and 5288 * then use it, unless it ends up being even bigger than the 5289 * previous value we had already. 5290 */ 5291 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5292 data_stripes), SZ_16M), 5293 ctl->stripe_size); 5294 } 5295 5296 /* Stripe size should not go beyond 1G. */ 5297 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); 5298 5299 /* Align to BTRFS_STRIPE_LEN */ 5300 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5301 ctl->chunk_size = ctl->stripe_size * data_stripes; 5302 5303 return 0; 5304 } 5305 5306 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5307 struct btrfs_device_info *devices_info) 5308 { 5309 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5310 /* Number of stripes that count for block group size */ 5311 int data_stripes; 5312 5313 /* 5314 * It should hold because: 5315 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5316 */ 5317 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5318 5319 ctl->stripe_size = zone_size; 5320 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5321 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5322 5323 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5324 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5325 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5326 ctl->stripe_size) + ctl->nparity, 5327 ctl->dev_stripes); 5328 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5329 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5330 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5331 } 5332 5333 ctl->chunk_size = ctl->stripe_size * data_stripes; 5334 5335 return 0; 5336 } 5337 5338 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5339 struct alloc_chunk_ctl *ctl, 5340 struct btrfs_device_info *devices_info) 5341 { 5342 struct btrfs_fs_info *info = fs_devices->fs_info; 5343 5344 /* 5345 * Round down to number of usable stripes, devs_increment can be any 5346 * number so we can't use round_down() that requires power of 2, while 5347 * rounddown is safe. 5348 */ 5349 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5350 5351 if (ctl->ndevs < ctl->devs_min) { 5352 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5353 btrfs_debug(info, 5354 "%s: not enough devices with free space: have=%d minimum required=%d", 5355 __func__, ctl->ndevs, ctl->devs_min); 5356 } 5357 return -ENOSPC; 5358 } 5359 5360 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5361 5362 switch (fs_devices->chunk_alloc_policy) { 5363 case BTRFS_CHUNK_ALLOC_REGULAR: 5364 return decide_stripe_size_regular(ctl, devices_info); 5365 case BTRFS_CHUNK_ALLOC_ZONED: 5366 return decide_stripe_size_zoned(ctl, devices_info); 5367 default: 5368 BUG(); 5369 } 5370 } 5371 5372 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5373 struct alloc_chunk_ctl *ctl, 5374 struct btrfs_device_info *devices_info) 5375 { 5376 struct btrfs_fs_info *info = trans->fs_info; 5377 struct map_lookup *map = NULL; 5378 struct extent_map_tree *em_tree; 5379 struct btrfs_block_group *block_group; 5380 struct extent_map *em; 5381 u64 start = ctl->start; 5382 u64 type = ctl->type; 5383 int ret; 5384 int i; 5385 int j; 5386 5387 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5388 if (!map) 5389 return ERR_PTR(-ENOMEM); 5390 map->num_stripes = ctl->num_stripes; 5391 5392 for (i = 0; i < ctl->ndevs; ++i) { 5393 for (j = 0; j < ctl->dev_stripes; ++j) { 5394 int s = i * ctl->dev_stripes + j; 5395 map->stripes[s].dev = devices_info[i].dev; 5396 map->stripes[s].physical = devices_info[i].dev_offset + 5397 j * ctl->stripe_size; 5398 } 5399 } 5400 map->stripe_len = BTRFS_STRIPE_LEN; 5401 map->io_align = BTRFS_STRIPE_LEN; 5402 map->io_width = BTRFS_STRIPE_LEN; 5403 map->type = type; 5404 map->sub_stripes = ctl->sub_stripes; 5405 5406 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5407 5408 em = alloc_extent_map(); 5409 if (!em) { 5410 kfree(map); 5411 return ERR_PTR(-ENOMEM); 5412 } 5413 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5414 em->map_lookup = map; 5415 em->start = start; 5416 em->len = ctl->chunk_size; 5417 em->block_start = 0; 5418 em->block_len = em->len; 5419 em->orig_block_len = ctl->stripe_size; 5420 5421 em_tree = &info->mapping_tree; 5422 write_lock(&em_tree->lock); 5423 ret = add_extent_mapping(em_tree, em, 0); 5424 if (ret) { 5425 write_unlock(&em_tree->lock); 5426 free_extent_map(em); 5427 return ERR_PTR(ret); 5428 } 5429 write_unlock(&em_tree->lock); 5430 5431 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5432 if (IS_ERR(block_group)) 5433 goto error_del_extent; 5434 5435 for (i = 0; i < map->num_stripes; i++) { 5436 struct btrfs_device *dev = map->stripes[i].dev; 5437 5438 btrfs_device_set_bytes_used(dev, 5439 dev->bytes_used + ctl->stripe_size); 5440 if (list_empty(&dev->post_commit_list)) 5441 list_add_tail(&dev->post_commit_list, 5442 &trans->transaction->dev_update_list); 5443 } 5444 5445 atomic64_sub(ctl->stripe_size * map->num_stripes, 5446 &info->free_chunk_space); 5447 5448 free_extent_map(em); 5449 check_raid56_incompat_flag(info, type); 5450 check_raid1c34_incompat_flag(info, type); 5451 5452 return block_group; 5453 5454 error_del_extent: 5455 write_lock(&em_tree->lock); 5456 remove_extent_mapping(em_tree, em); 5457 write_unlock(&em_tree->lock); 5458 5459 /* One for our allocation */ 5460 free_extent_map(em); 5461 /* One for the tree reference */ 5462 free_extent_map(em); 5463 5464 return block_group; 5465 } 5466 5467 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5468 u64 type) 5469 { 5470 struct btrfs_fs_info *info = trans->fs_info; 5471 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5472 struct btrfs_device_info *devices_info = NULL; 5473 struct alloc_chunk_ctl ctl; 5474 struct btrfs_block_group *block_group; 5475 int ret; 5476 5477 lockdep_assert_held(&info->chunk_mutex); 5478 5479 if (!alloc_profile_is_valid(type, 0)) { 5480 ASSERT(0); 5481 return ERR_PTR(-EINVAL); 5482 } 5483 5484 if (list_empty(&fs_devices->alloc_list)) { 5485 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5486 btrfs_debug(info, "%s: no writable device", __func__); 5487 return ERR_PTR(-ENOSPC); 5488 } 5489 5490 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5491 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5492 ASSERT(0); 5493 return ERR_PTR(-EINVAL); 5494 } 5495 5496 ctl.start = find_next_chunk(info); 5497 ctl.type = type; 5498 init_alloc_chunk_ctl(fs_devices, &ctl); 5499 5500 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5501 GFP_NOFS); 5502 if (!devices_info) 5503 return ERR_PTR(-ENOMEM); 5504 5505 ret = gather_device_info(fs_devices, &ctl, devices_info); 5506 if (ret < 0) { 5507 block_group = ERR_PTR(ret); 5508 goto out; 5509 } 5510 5511 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5512 if (ret < 0) { 5513 block_group = ERR_PTR(ret); 5514 goto out; 5515 } 5516 5517 block_group = create_chunk(trans, &ctl, devices_info); 5518 5519 out: 5520 kfree(devices_info); 5521 return block_group; 5522 } 5523 5524 /* 5525 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5526 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5527 * chunks. 5528 * 5529 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5530 * phases. 5531 */ 5532 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5533 struct btrfs_block_group *bg) 5534 { 5535 struct btrfs_fs_info *fs_info = trans->fs_info; 5536 struct btrfs_root *chunk_root = fs_info->chunk_root; 5537 struct btrfs_key key; 5538 struct btrfs_chunk *chunk; 5539 struct btrfs_stripe *stripe; 5540 struct extent_map *em; 5541 struct map_lookup *map; 5542 size_t item_size; 5543 int i; 5544 int ret; 5545 5546 /* 5547 * We take the chunk_mutex for 2 reasons: 5548 * 5549 * 1) Updates and insertions in the chunk btree must be done while holding 5550 * the chunk_mutex, as well as updating the system chunk array in the 5551 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5552 * details; 5553 * 5554 * 2) To prevent races with the final phase of a device replace operation 5555 * that replaces the device object associated with the map's stripes, 5556 * because the device object's id can change at any time during that 5557 * final phase of the device replace operation 5558 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5559 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5560 * which would cause a failure when updating the device item, which does 5561 * not exists, or persisting a stripe of the chunk item with such ID. 5562 * Here we can't use the device_list_mutex because our caller already 5563 * has locked the chunk_mutex, and the final phase of device replace 5564 * acquires both mutexes - first the device_list_mutex and then the 5565 * chunk_mutex. Using any of those two mutexes protects us from a 5566 * concurrent device replace. 5567 */ 5568 lockdep_assert_held(&fs_info->chunk_mutex); 5569 5570 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5571 if (IS_ERR(em)) { 5572 ret = PTR_ERR(em); 5573 btrfs_abort_transaction(trans, ret); 5574 return ret; 5575 } 5576 5577 map = em->map_lookup; 5578 item_size = btrfs_chunk_item_size(map->num_stripes); 5579 5580 chunk = kzalloc(item_size, GFP_NOFS); 5581 if (!chunk) { 5582 ret = -ENOMEM; 5583 btrfs_abort_transaction(trans, ret); 5584 goto out; 5585 } 5586 5587 for (i = 0; i < map->num_stripes; i++) { 5588 struct btrfs_device *device = map->stripes[i].dev; 5589 5590 ret = btrfs_update_device(trans, device); 5591 if (ret) 5592 goto out; 5593 } 5594 5595 stripe = &chunk->stripe; 5596 for (i = 0; i < map->num_stripes; i++) { 5597 struct btrfs_device *device = map->stripes[i].dev; 5598 const u64 dev_offset = map->stripes[i].physical; 5599 5600 btrfs_set_stack_stripe_devid(stripe, device->devid); 5601 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5602 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5603 stripe++; 5604 } 5605 5606 btrfs_set_stack_chunk_length(chunk, bg->length); 5607 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5608 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5609 btrfs_set_stack_chunk_type(chunk, map->type); 5610 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5611 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5612 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5613 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5614 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5615 5616 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5617 key.type = BTRFS_CHUNK_ITEM_KEY; 5618 key.offset = bg->start; 5619 5620 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5621 if (ret) 5622 goto out; 5623 5624 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); 5625 5626 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5627 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5628 if (ret) 5629 goto out; 5630 } 5631 5632 out: 5633 kfree(chunk); 5634 free_extent_map(em); 5635 return ret; 5636 } 5637 5638 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5639 { 5640 struct btrfs_fs_info *fs_info = trans->fs_info; 5641 u64 alloc_profile; 5642 struct btrfs_block_group *meta_bg; 5643 struct btrfs_block_group *sys_bg; 5644 5645 /* 5646 * When adding a new device for sprouting, the seed device is read-only 5647 * so we must first allocate a metadata and a system chunk. But before 5648 * adding the block group items to the extent, device and chunk btrees, 5649 * we must first: 5650 * 5651 * 1) Create both chunks without doing any changes to the btrees, as 5652 * otherwise we would get -ENOSPC since the block groups from the 5653 * seed device are read-only; 5654 * 5655 * 2) Add the device item for the new sprout device - finishing the setup 5656 * of a new block group requires updating the device item in the chunk 5657 * btree, so it must exist when we attempt to do it. The previous step 5658 * ensures this does not fail with -ENOSPC. 5659 * 5660 * After that we can add the block group items to their btrees: 5661 * update existing device item in the chunk btree, add a new block group 5662 * item to the extent btree, add a new chunk item to the chunk btree and 5663 * finally add the new device extent items to the devices btree. 5664 */ 5665 5666 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5667 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5668 if (IS_ERR(meta_bg)) 5669 return PTR_ERR(meta_bg); 5670 5671 alloc_profile = btrfs_system_alloc_profile(fs_info); 5672 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5673 if (IS_ERR(sys_bg)) 5674 return PTR_ERR(sys_bg); 5675 5676 return 0; 5677 } 5678 5679 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5680 { 5681 const int index = btrfs_bg_flags_to_raid_index(map->type); 5682 5683 return btrfs_raid_array[index].tolerated_failures; 5684 } 5685 5686 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5687 { 5688 struct extent_map *em; 5689 struct map_lookup *map; 5690 int miss_ndevs = 0; 5691 int i; 5692 bool ret = true; 5693 5694 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5695 if (IS_ERR(em)) 5696 return false; 5697 5698 map = em->map_lookup; 5699 for (i = 0; i < map->num_stripes; i++) { 5700 if (test_bit(BTRFS_DEV_STATE_MISSING, 5701 &map->stripes[i].dev->dev_state)) { 5702 miss_ndevs++; 5703 continue; 5704 } 5705 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5706 &map->stripes[i].dev->dev_state)) { 5707 ret = false; 5708 goto end; 5709 } 5710 } 5711 5712 /* 5713 * If the number of missing devices is larger than max errors, we can 5714 * not write the data into that chunk successfully. 5715 */ 5716 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5717 ret = false; 5718 end: 5719 free_extent_map(em); 5720 return ret; 5721 } 5722 5723 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5724 { 5725 struct extent_map *em; 5726 5727 while (1) { 5728 write_lock(&tree->lock); 5729 em = lookup_extent_mapping(tree, 0, (u64)-1); 5730 if (em) 5731 remove_extent_mapping(tree, em); 5732 write_unlock(&tree->lock); 5733 if (!em) 5734 break; 5735 /* once for us */ 5736 free_extent_map(em); 5737 /* once for the tree */ 5738 free_extent_map(em); 5739 } 5740 } 5741 5742 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5743 { 5744 struct extent_map *em; 5745 struct map_lookup *map; 5746 enum btrfs_raid_types index; 5747 int ret = 1; 5748 5749 em = btrfs_get_chunk_map(fs_info, logical, len); 5750 if (IS_ERR(em)) 5751 /* 5752 * We could return errors for these cases, but that could get 5753 * ugly and we'd probably do the same thing which is just not do 5754 * anything else and exit, so return 1 so the callers don't try 5755 * to use other copies. 5756 */ 5757 return 1; 5758 5759 map = em->map_lookup; 5760 index = btrfs_bg_flags_to_raid_index(map->type); 5761 5762 /* Non-RAID56, use their ncopies from btrfs_raid_array. */ 5763 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5764 ret = btrfs_raid_array[index].ncopies; 5765 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5766 ret = 2; 5767 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5768 /* 5769 * There could be two corrupted data stripes, we need 5770 * to loop retry in order to rebuild the correct data. 5771 * 5772 * Fail a stripe at a time on every retry except the 5773 * stripe under reconstruction. 5774 */ 5775 ret = map->num_stripes; 5776 free_extent_map(em); 5777 5778 down_read(&fs_info->dev_replace.rwsem); 5779 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5780 fs_info->dev_replace.tgtdev) 5781 ret++; 5782 up_read(&fs_info->dev_replace.rwsem); 5783 5784 return ret; 5785 } 5786 5787 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5788 u64 logical) 5789 { 5790 struct extent_map *em; 5791 struct map_lookup *map; 5792 unsigned long len = fs_info->sectorsize; 5793 5794 if (!btrfs_fs_incompat(fs_info, RAID56)) 5795 return len; 5796 5797 em = btrfs_get_chunk_map(fs_info, logical, len); 5798 5799 if (!WARN_ON(IS_ERR(em))) { 5800 map = em->map_lookup; 5801 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5802 len = map->stripe_len * nr_data_stripes(map); 5803 free_extent_map(em); 5804 } 5805 return len; 5806 } 5807 5808 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5809 { 5810 struct extent_map *em; 5811 struct map_lookup *map; 5812 int ret = 0; 5813 5814 if (!btrfs_fs_incompat(fs_info, RAID56)) 5815 return 0; 5816 5817 em = btrfs_get_chunk_map(fs_info, logical, len); 5818 5819 if(!WARN_ON(IS_ERR(em))) { 5820 map = em->map_lookup; 5821 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5822 ret = 1; 5823 free_extent_map(em); 5824 } 5825 return ret; 5826 } 5827 5828 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5829 struct map_lookup *map, int first, 5830 int dev_replace_is_ongoing) 5831 { 5832 int i; 5833 int num_stripes; 5834 int preferred_mirror; 5835 int tolerance; 5836 struct btrfs_device *srcdev; 5837 5838 ASSERT((map->type & 5839 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5840 5841 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5842 num_stripes = map->sub_stripes; 5843 else 5844 num_stripes = map->num_stripes; 5845 5846 switch (fs_info->fs_devices->read_policy) { 5847 default: 5848 /* Shouldn't happen, just warn and use pid instead of failing */ 5849 btrfs_warn_rl(fs_info, 5850 "unknown read_policy type %u, reset to pid", 5851 fs_info->fs_devices->read_policy); 5852 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5853 fallthrough; 5854 case BTRFS_READ_POLICY_PID: 5855 preferred_mirror = first + (current->pid % num_stripes); 5856 break; 5857 } 5858 5859 if (dev_replace_is_ongoing && 5860 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5861 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5862 srcdev = fs_info->dev_replace.srcdev; 5863 else 5864 srcdev = NULL; 5865 5866 /* 5867 * try to avoid the drive that is the source drive for a 5868 * dev-replace procedure, only choose it if no other non-missing 5869 * mirror is available 5870 */ 5871 for (tolerance = 0; tolerance < 2; tolerance++) { 5872 if (map->stripes[preferred_mirror].dev->bdev && 5873 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5874 return preferred_mirror; 5875 for (i = first; i < first + num_stripes; i++) { 5876 if (map->stripes[i].dev->bdev && 5877 (tolerance || map->stripes[i].dev != srcdev)) 5878 return i; 5879 } 5880 } 5881 5882 /* we couldn't find one that doesn't fail. Just return something 5883 * and the io error handling code will clean up eventually 5884 */ 5885 return preferred_mirror; 5886 } 5887 5888 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5889 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5890 { 5891 int i; 5892 int again = 1; 5893 5894 while (again) { 5895 again = 0; 5896 for (i = 0; i < num_stripes - 1; i++) { 5897 /* Swap if parity is on a smaller index */ 5898 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5899 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5900 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5901 again = 1; 5902 } 5903 } 5904 } 5905 } 5906 5907 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5908 int total_stripes, 5909 int real_stripes) 5910 { 5911 struct btrfs_io_context *bioc = kzalloc( 5912 /* The size of btrfs_io_context */ 5913 sizeof(struct btrfs_io_context) + 5914 /* Plus the variable array for the stripes */ 5915 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5916 /* Plus the variable array for the tgt dev */ 5917 sizeof(int) * (real_stripes) + 5918 /* 5919 * Plus the raid_map, which includes both the tgt dev 5920 * and the stripes. 5921 */ 5922 sizeof(u64) * (total_stripes), 5923 GFP_NOFS|__GFP_NOFAIL); 5924 5925 refcount_set(&bioc->refs, 1); 5926 5927 bioc->fs_info = fs_info; 5928 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5929 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5930 5931 return bioc; 5932 } 5933 5934 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5935 { 5936 WARN_ON(!refcount_read(&bioc->refs)); 5937 refcount_inc(&bioc->refs); 5938 } 5939 5940 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5941 { 5942 if (!bioc) 5943 return; 5944 if (refcount_dec_and_test(&bioc->refs)) 5945 kfree(bioc); 5946 } 5947 5948 /* 5949 * Please note that, discard won't be sent to target device of device 5950 * replace. 5951 */ 5952 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, 5953 u64 logical, u64 *length_ret, 5954 u32 *num_stripes) 5955 { 5956 struct extent_map *em; 5957 struct map_lookup *map; 5958 struct btrfs_discard_stripe *stripes; 5959 u64 length = *length_ret; 5960 u64 offset; 5961 u64 stripe_nr; 5962 u64 stripe_nr_end; 5963 u64 stripe_end_offset; 5964 u64 stripe_cnt; 5965 u64 stripe_len; 5966 u64 stripe_offset; 5967 u32 stripe_index; 5968 u32 factor = 0; 5969 u32 sub_stripes = 0; 5970 u64 stripes_per_dev = 0; 5971 u32 remaining_stripes = 0; 5972 u32 last_stripe = 0; 5973 int ret; 5974 int i; 5975 5976 em = btrfs_get_chunk_map(fs_info, logical, length); 5977 if (IS_ERR(em)) 5978 return ERR_CAST(em); 5979 5980 map = em->map_lookup; 5981 5982 /* we don't discard raid56 yet */ 5983 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5984 ret = -EOPNOTSUPP; 5985 goto out_free_map; 5986 } 5987 5988 offset = logical - em->start; 5989 length = min_t(u64, em->start + em->len - logical, length); 5990 *length_ret = length; 5991 5992 stripe_len = map->stripe_len; 5993 /* 5994 * stripe_nr counts the total number of stripes we have to stride 5995 * to get to this block 5996 */ 5997 stripe_nr = div64_u64(offset, stripe_len); 5998 5999 /* stripe_offset is the offset of this block in its stripe */ 6000 stripe_offset = offset - stripe_nr * stripe_len; 6001 6002 stripe_nr_end = round_up(offset + length, map->stripe_len); 6003 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); 6004 stripe_cnt = stripe_nr_end - stripe_nr; 6005 stripe_end_offset = stripe_nr_end * map->stripe_len - 6006 (offset + length); 6007 /* 6008 * after this, stripe_nr is the number of stripes on this 6009 * device we have to walk to find the data, and stripe_index is 6010 * the number of our device in the stripe array 6011 */ 6012 *num_stripes = 1; 6013 stripe_index = 0; 6014 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6015 BTRFS_BLOCK_GROUP_RAID10)) { 6016 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 6017 sub_stripes = 1; 6018 else 6019 sub_stripes = map->sub_stripes; 6020 6021 factor = map->num_stripes / sub_stripes; 6022 *num_stripes = min_t(u64, map->num_stripes, 6023 sub_stripes * stripe_cnt); 6024 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6025 stripe_index *= sub_stripes; 6026 stripes_per_dev = div_u64_rem(stripe_cnt, factor, 6027 &remaining_stripes); 6028 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); 6029 last_stripe *= sub_stripes; 6030 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 6031 BTRFS_BLOCK_GROUP_DUP)) { 6032 *num_stripes = map->num_stripes; 6033 } else { 6034 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6035 &stripe_index); 6036 } 6037 6038 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); 6039 if (!stripes) { 6040 ret = -ENOMEM; 6041 goto out_free_map; 6042 } 6043 6044 for (i = 0; i < *num_stripes; i++) { 6045 stripes[i].physical = 6046 map->stripes[stripe_index].physical + 6047 stripe_offset + stripe_nr * map->stripe_len; 6048 stripes[i].dev = map->stripes[stripe_index].dev; 6049 6050 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6051 BTRFS_BLOCK_GROUP_RAID10)) { 6052 stripes[i].length = stripes_per_dev * map->stripe_len; 6053 6054 if (i / sub_stripes < remaining_stripes) 6055 stripes[i].length += map->stripe_len; 6056 6057 /* 6058 * Special for the first stripe and 6059 * the last stripe: 6060 * 6061 * |-------|...|-------| 6062 * |----------| 6063 * off end_off 6064 */ 6065 if (i < sub_stripes) 6066 stripes[i].length -= stripe_offset; 6067 6068 if (stripe_index >= last_stripe && 6069 stripe_index <= (last_stripe + 6070 sub_stripes - 1)) 6071 stripes[i].length -= stripe_end_offset; 6072 6073 if (i == sub_stripes - 1) 6074 stripe_offset = 0; 6075 } else { 6076 stripes[i].length = length; 6077 } 6078 6079 stripe_index++; 6080 if (stripe_index == map->num_stripes) { 6081 stripe_index = 0; 6082 stripe_nr++; 6083 } 6084 } 6085 6086 free_extent_map(em); 6087 return stripes; 6088 out_free_map: 6089 free_extent_map(em); 6090 return ERR_PTR(ret); 6091 } 6092 6093 /* 6094 * In dev-replace case, for repair case (that's the only case where the mirror 6095 * is selected explicitly when calling btrfs_map_block), blocks left of the 6096 * left cursor can also be read from the target drive. 6097 * 6098 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6099 * array of stripes. 6100 * For READ, it also needs to be supported using the same mirror number. 6101 * 6102 * If the requested block is not left of the left cursor, EIO is returned. This 6103 * can happen because btrfs_num_copies() returns one more in the dev-replace 6104 * case. 6105 */ 6106 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6107 u64 logical, u64 length, 6108 u64 srcdev_devid, int *mirror_num, 6109 u64 *physical) 6110 { 6111 struct btrfs_io_context *bioc = NULL; 6112 int num_stripes; 6113 int index_srcdev = 0; 6114 int found = 0; 6115 u64 physical_of_found = 0; 6116 int i; 6117 int ret = 0; 6118 6119 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6120 logical, &length, &bioc, NULL, NULL, 0); 6121 if (ret) { 6122 ASSERT(bioc == NULL); 6123 return ret; 6124 } 6125 6126 num_stripes = bioc->num_stripes; 6127 if (*mirror_num > num_stripes) { 6128 /* 6129 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6130 * that means that the requested area is not left of the left 6131 * cursor 6132 */ 6133 btrfs_put_bioc(bioc); 6134 return -EIO; 6135 } 6136 6137 /* 6138 * process the rest of the function using the mirror_num of the source 6139 * drive. Therefore look it up first. At the end, patch the device 6140 * pointer to the one of the target drive. 6141 */ 6142 for (i = 0; i < num_stripes; i++) { 6143 if (bioc->stripes[i].dev->devid != srcdev_devid) 6144 continue; 6145 6146 /* 6147 * In case of DUP, in order to keep it simple, only add the 6148 * mirror with the lowest physical address 6149 */ 6150 if (found && 6151 physical_of_found <= bioc->stripes[i].physical) 6152 continue; 6153 6154 index_srcdev = i; 6155 found = 1; 6156 physical_of_found = bioc->stripes[i].physical; 6157 } 6158 6159 btrfs_put_bioc(bioc); 6160 6161 ASSERT(found); 6162 if (!found) 6163 return -EIO; 6164 6165 *mirror_num = index_srcdev + 1; 6166 *physical = physical_of_found; 6167 return ret; 6168 } 6169 6170 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6171 { 6172 struct btrfs_block_group *cache; 6173 bool ret; 6174 6175 /* Non zoned filesystem does not use "to_copy" flag */ 6176 if (!btrfs_is_zoned(fs_info)) 6177 return false; 6178 6179 cache = btrfs_lookup_block_group(fs_info, logical); 6180 6181 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 6182 6183 btrfs_put_block_group(cache); 6184 return ret; 6185 } 6186 6187 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6188 struct btrfs_io_context **bioc_ret, 6189 struct btrfs_dev_replace *dev_replace, 6190 u64 logical, 6191 int *num_stripes_ret, int *max_errors_ret) 6192 { 6193 struct btrfs_io_context *bioc = *bioc_ret; 6194 u64 srcdev_devid = dev_replace->srcdev->devid; 6195 int tgtdev_indexes = 0; 6196 int num_stripes = *num_stripes_ret; 6197 int max_errors = *max_errors_ret; 6198 int i; 6199 6200 if (op == BTRFS_MAP_WRITE) { 6201 int index_where_to_add; 6202 6203 /* 6204 * A block group which have "to_copy" set will eventually 6205 * copied by dev-replace process. We can avoid cloning IO here. 6206 */ 6207 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6208 return; 6209 6210 /* 6211 * duplicate the write operations while the dev replace 6212 * procedure is running. Since the copying of the old disk to 6213 * the new disk takes place at run time while the filesystem is 6214 * mounted writable, the regular write operations to the old 6215 * disk have to be duplicated to go to the new disk as well. 6216 * 6217 * Note that device->missing is handled by the caller, and that 6218 * the write to the old disk is already set up in the stripes 6219 * array. 6220 */ 6221 index_where_to_add = num_stripes; 6222 for (i = 0; i < num_stripes; i++) { 6223 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6224 /* write to new disk, too */ 6225 struct btrfs_io_stripe *new = 6226 bioc->stripes + index_where_to_add; 6227 struct btrfs_io_stripe *old = 6228 bioc->stripes + i; 6229 6230 new->physical = old->physical; 6231 new->dev = dev_replace->tgtdev; 6232 bioc->tgtdev_map[i] = index_where_to_add; 6233 index_where_to_add++; 6234 max_errors++; 6235 tgtdev_indexes++; 6236 } 6237 } 6238 num_stripes = index_where_to_add; 6239 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6240 int index_srcdev = 0; 6241 int found = 0; 6242 u64 physical_of_found = 0; 6243 6244 /* 6245 * During the dev-replace procedure, the target drive can also 6246 * be used to read data in case it is needed to repair a corrupt 6247 * block elsewhere. This is possible if the requested area is 6248 * left of the left cursor. In this area, the target drive is a 6249 * full copy of the source drive. 6250 */ 6251 for (i = 0; i < num_stripes; i++) { 6252 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6253 /* 6254 * In case of DUP, in order to keep it simple, 6255 * only add the mirror with the lowest physical 6256 * address 6257 */ 6258 if (found && 6259 physical_of_found <= bioc->stripes[i].physical) 6260 continue; 6261 index_srcdev = i; 6262 found = 1; 6263 physical_of_found = bioc->stripes[i].physical; 6264 } 6265 } 6266 if (found) { 6267 struct btrfs_io_stripe *tgtdev_stripe = 6268 bioc->stripes + num_stripes; 6269 6270 tgtdev_stripe->physical = physical_of_found; 6271 tgtdev_stripe->dev = dev_replace->tgtdev; 6272 bioc->tgtdev_map[index_srcdev] = num_stripes; 6273 6274 tgtdev_indexes++; 6275 num_stripes++; 6276 } 6277 } 6278 6279 *num_stripes_ret = num_stripes; 6280 *max_errors_ret = max_errors; 6281 bioc->num_tgtdevs = tgtdev_indexes; 6282 *bioc_ret = bioc; 6283 } 6284 6285 static bool need_full_stripe(enum btrfs_map_op op) 6286 { 6287 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6288 } 6289 6290 /* 6291 * Calculate the geometry of a particular (address, len) tuple. This 6292 * information is used to calculate how big a particular bio can get before it 6293 * straddles a stripe. 6294 * 6295 * @fs_info: the filesystem 6296 * @em: mapping containing the logical extent 6297 * @op: type of operation - write or read 6298 * @logical: address that we want to figure out the geometry of 6299 * @io_geom: pointer used to return values 6300 * 6301 * Returns < 0 in case a chunk for the given logical address cannot be found, 6302 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6303 */ 6304 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6305 enum btrfs_map_op op, u64 logical, 6306 struct btrfs_io_geometry *io_geom) 6307 { 6308 struct map_lookup *map; 6309 u64 len; 6310 u64 offset; 6311 u64 stripe_offset; 6312 u64 stripe_nr; 6313 u32 stripe_len; 6314 u64 raid56_full_stripe_start = (u64)-1; 6315 int data_stripes; 6316 6317 ASSERT(op != BTRFS_MAP_DISCARD); 6318 6319 map = em->map_lookup; 6320 /* Offset of this logical address in the chunk */ 6321 offset = logical - em->start; 6322 /* Len of a stripe in a chunk */ 6323 stripe_len = map->stripe_len; 6324 /* 6325 * Stripe_nr is where this block falls in 6326 * stripe_offset is the offset of this block in its stripe. 6327 */ 6328 stripe_nr = div64_u64_rem(offset, stripe_len, &stripe_offset); 6329 ASSERT(stripe_offset < U32_MAX); 6330 6331 data_stripes = nr_data_stripes(map); 6332 6333 /* Only stripe based profiles needs to check against stripe length. */ 6334 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) { 6335 u64 max_len = stripe_len - stripe_offset; 6336 6337 /* 6338 * In case of raid56, we need to know the stripe aligned start 6339 */ 6340 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6341 unsigned long full_stripe_len = stripe_len * data_stripes; 6342 raid56_full_stripe_start = offset; 6343 6344 /* 6345 * Allow a write of a full stripe, but make sure we 6346 * don't allow straddling of stripes 6347 */ 6348 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6349 full_stripe_len); 6350 raid56_full_stripe_start *= full_stripe_len; 6351 6352 /* 6353 * For writes to RAID[56], allow a full stripeset across 6354 * all disks. For other RAID types and for RAID[56] 6355 * reads, just allow a single stripe (on a single disk). 6356 */ 6357 if (op == BTRFS_MAP_WRITE) { 6358 max_len = stripe_len * data_stripes - 6359 (offset - raid56_full_stripe_start); 6360 } 6361 } 6362 len = min_t(u64, em->len - offset, max_len); 6363 } else { 6364 len = em->len - offset; 6365 } 6366 6367 io_geom->len = len; 6368 io_geom->offset = offset; 6369 io_geom->stripe_len = stripe_len; 6370 io_geom->stripe_nr = stripe_nr; 6371 io_geom->stripe_offset = stripe_offset; 6372 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6373 6374 return 0; 6375 } 6376 6377 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, 6378 u32 stripe_index, u64 stripe_offset, u64 stripe_nr) 6379 { 6380 dst->dev = map->stripes[stripe_index].dev; 6381 dst->physical = map->stripes[stripe_index].physical + 6382 stripe_offset + stripe_nr * map->stripe_len; 6383 } 6384 6385 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6386 enum btrfs_map_op op, u64 logical, u64 *length, 6387 struct btrfs_io_context **bioc_ret, 6388 struct btrfs_io_stripe *smap, 6389 int *mirror_num_ret, int need_raid_map) 6390 { 6391 struct extent_map *em; 6392 struct map_lookup *map; 6393 u64 stripe_offset; 6394 u64 stripe_nr; 6395 u64 stripe_len; 6396 u32 stripe_index; 6397 int data_stripes; 6398 int i; 6399 int ret = 0; 6400 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); 6401 int num_stripes; 6402 int max_errors = 0; 6403 int tgtdev_indexes = 0; 6404 struct btrfs_io_context *bioc = NULL; 6405 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6406 int dev_replace_is_ongoing = 0; 6407 int num_alloc_stripes; 6408 int patch_the_first_stripe_for_dev_replace = 0; 6409 u64 physical_to_patch_in_first_stripe = 0; 6410 u64 raid56_full_stripe_start = (u64)-1; 6411 struct btrfs_io_geometry geom; 6412 6413 ASSERT(bioc_ret); 6414 ASSERT(op != BTRFS_MAP_DISCARD); 6415 6416 em = btrfs_get_chunk_map(fs_info, logical, *length); 6417 ASSERT(!IS_ERR(em)); 6418 6419 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6420 if (ret < 0) 6421 return ret; 6422 6423 map = em->map_lookup; 6424 6425 *length = geom.len; 6426 stripe_len = geom.stripe_len; 6427 stripe_nr = geom.stripe_nr; 6428 stripe_offset = geom.stripe_offset; 6429 raid56_full_stripe_start = geom.raid56_stripe_offset; 6430 data_stripes = nr_data_stripes(map); 6431 6432 down_read(&dev_replace->rwsem); 6433 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6434 /* 6435 * Hold the semaphore for read during the whole operation, write is 6436 * requested at commit time but must wait. 6437 */ 6438 if (!dev_replace_is_ongoing) 6439 up_read(&dev_replace->rwsem); 6440 6441 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6442 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6443 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6444 dev_replace->srcdev->devid, 6445 &mirror_num, 6446 &physical_to_patch_in_first_stripe); 6447 if (ret) 6448 goto out; 6449 else 6450 patch_the_first_stripe_for_dev_replace = 1; 6451 } else if (mirror_num > map->num_stripes) { 6452 mirror_num = 0; 6453 } 6454 6455 num_stripes = 1; 6456 stripe_index = 0; 6457 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6458 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6459 &stripe_index); 6460 if (!need_full_stripe(op)) 6461 mirror_num = 1; 6462 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6463 if (need_full_stripe(op)) 6464 num_stripes = map->num_stripes; 6465 else if (mirror_num) 6466 stripe_index = mirror_num - 1; 6467 else { 6468 stripe_index = find_live_mirror(fs_info, map, 0, 6469 dev_replace_is_ongoing); 6470 mirror_num = stripe_index + 1; 6471 } 6472 6473 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6474 if (need_full_stripe(op)) { 6475 num_stripes = map->num_stripes; 6476 } else if (mirror_num) { 6477 stripe_index = mirror_num - 1; 6478 } else { 6479 mirror_num = 1; 6480 } 6481 6482 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6483 u32 factor = map->num_stripes / map->sub_stripes; 6484 6485 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6486 stripe_index *= map->sub_stripes; 6487 6488 if (need_full_stripe(op)) 6489 num_stripes = map->sub_stripes; 6490 else if (mirror_num) 6491 stripe_index += mirror_num - 1; 6492 else { 6493 int old_stripe_index = stripe_index; 6494 stripe_index = find_live_mirror(fs_info, map, 6495 stripe_index, 6496 dev_replace_is_ongoing); 6497 mirror_num = stripe_index - old_stripe_index + 1; 6498 } 6499 6500 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6501 ASSERT(map->stripe_len == BTRFS_STRIPE_LEN); 6502 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6503 /* push stripe_nr back to the start of the full stripe */ 6504 stripe_nr = div64_u64(raid56_full_stripe_start, 6505 stripe_len * data_stripes); 6506 6507 /* RAID[56] write or recovery. Return all stripes */ 6508 num_stripes = map->num_stripes; 6509 max_errors = btrfs_chunk_max_errors(map); 6510 6511 /* Return the length to the full stripe end */ 6512 *length = min(logical + *length, 6513 raid56_full_stripe_start + em->start + 6514 data_stripes * stripe_len) - logical; 6515 stripe_index = 0; 6516 stripe_offset = 0; 6517 } else { 6518 /* 6519 * Mirror #0 or #1 means the original data block. 6520 * Mirror #2 is RAID5 parity block. 6521 * Mirror #3 is RAID6 Q block. 6522 */ 6523 stripe_nr = div_u64_rem(stripe_nr, 6524 data_stripes, &stripe_index); 6525 if (mirror_num > 1) 6526 stripe_index = data_stripes + mirror_num - 2; 6527 6528 /* We distribute the parity blocks across stripes */ 6529 div_u64_rem(stripe_nr + stripe_index, map->num_stripes, 6530 &stripe_index); 6531 if (!need_full_stripe(op) && mirror_num <= 1) 6532 mirror_num = 1; 6533 } 6534 } else { 6535 /* 6536 * after this, stripe_nr is the number of stripes on this 6537 * device we have to walk to find the data, and stripe_index is 6538 * the number of our device in the stripe array 6539 */ 6540 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6541 &stripe_index); 6542 mirror_num = stripe_index + 1; 6543 } 6544 if (stripe_index >= map->num_stripes) { 6545 btrfs_crit(fs_info, 6546 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6547 stripe_index, map->num_stripes); 6548 ret = -EINVAL; 6549 goto out; 6550 } 6551 6552 num_alloc_stripes = num_stripes; 6553 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6554 if (op == BTRFS_MAP_WRITE) 6555 num_alloc_stripes <<= 1; 6556 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6557 num_alloc_stripes++; 6558 tgtdev_indexes = num_stripes; 6559 } 6560 6561 /* 6562 * If this I/O maps to a single device, try to return the device and 6563 * physical block information on the stack instead of allocating an 6564 * I/O context structure. 6565 */ 6566 if (smap && num_alloc_stripes == 1 && 6567 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && 6568 (!need_full_stripe(op) || !dev_replace_is_ongoing || 6569 !dev_replace->tgtdev)) { 6570 if (patch_the_first_stripe_for_dev_replace) { 6571 smap->dev = dev_replace->tgtdev; 6572 smap->physical = physical_to_patch_in_first_stripe; 6573 *mirror_num_ret = map->num_stripes + 1; 6574 } else { 6575 set_io_stripe(smap, map, stripe_index, stripe_offset, 6576 stripe_nr); 6577 *mirror_num_ret = mirror_num; 6578 } 6579 *bioc_ret = NULL; 6580 ret = 0; 6581 goto out; 6582 } 6583 6584 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6585 if (!bioc) { 6586 ret = -ENOMEM; 6587 goto out; 6588 } 6589 6590 for (i = 0; i < num_stripes; i++) { 6591 set_io_stripe(&bioc->stripes[i], map, stripe_index, stripe_offset, 6592 stripe_nr); 6593 stripe_index++; 6594 } 6595 6596 /* Build raid_map */ 6597 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6598 (need_full_stripe(op) || mirror_num > 1)) { 6599 u64 tmp; 6600 unsigned rot; 6601 6602 /* Work out the disk rotation on this stripe-set */ 6603 div_u64_rem(stripe_nr, num_stripes, &rot); 6604 6605 /* Fill in the logical address of each stripe */ 6606 tmp = stripe_nr * data_stripes; 6607 for (i = 0; i < data_stripes; i++) 6608 bioc->raid_map[(i + rot) % num_stripes] = 6609 em->start + (tmp + i) * map->stripe_len; 6610 6611 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6612 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6613 bioc->raid_map[(i + rot + 1) % num_stripes] = 6614 RAID6_Q_STRIPE; 6615 6616 sort_parity_stripes(bioc, num_stripes); 6617 } 6618 6619 if (need_full_stripe(op)) 6620 max_errors = btrfs_chunk_max_errors(map); 6621 6622 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6623 need_full_stripe(op)) { 6624 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6625 &num_stripes, &max_errors); 6626 } 6627 6628 *bioc_ret = bioc; 6629 bioc->map_type = map->type; 6630 bioc->num_stripes = num_stripes; 6631 bioc->max_errors = max_errors; 6632 bioc->mirror_num = mirror_num; 6633 6634 /* 6635 * this is the case that REQ_READ && dev_replace_is_ongoing && 6636 * mirror_num == num_stripes + 1 && dev_replace target drive is 6637 * available as a mirror 6638 */ 6639 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6640 WARN_ON(num_stripes > 1); 6641 bioc->stripes[0].dev = dev_replace->tgtdev; 6642 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6643 bioc->mirror_num = map->num_stripes + 1; 6644 } 6645 out: 6646 if (dev_replace_is_ongoing) { 6647 lockdep_assert_held(&dev_replace->rwsem); 6648 /* Unlock and let waiting writers proceed */ 6649 up_read(&dev_replace->rwsem); 6650 } 6651 free_extent_map(em); 6652 return ret; 6653 } 6654 6655 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6656 u64 logical, u64 *length, 6657 struct btrfs_io_context **bioc_ret, int mirror_num) 6658 { 6659 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6660 NULL, &mirror_num, 0); 6661 } 6662 6663 /* For Scrub/replace */ 6664 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6665 u64 logical, u64 *length, 6666 struct btrfs_io_context **bioc_ret) 6667 { 6668 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6669 NULL, NULL, 1); 6670 } 6671 6672 /* 6673 * Initialize a btrfs_bio structure. This skips the embedded bio itself as it 6674 * is already initialized by the block layer. 6675 */ 6676 static inline void btrfs_bio_init(struct btrfs_bio *bbio, 6677 btrfs_bio_end_io_t end_io, void *private) 6678 { 6679 memset(bbio, 0, offsetof(struct btrfs_bio, bio)); 6680 bbio->end_io = end_io; 6681 bbio->private = private; 6682 } 6683 6684 /* 6685 * Allocate a btrfs_bio structure. The btrfs_bio is the main I/O container for 6686 * btrfs, and is used for all I/O submitted through btrfs_submit_bio. 6687 * 6688 * Just like the underlying bio_alloc_bioset it will not fail as it is backed by 6689 * a mempool. 6690 */ 6691 struct bio *btrfs_bio_alloc(unsigned int nr_vecs, blk_opf_t opf, 6692 btrfs_bio_end_io_t end_io, void *private) 6693 { 6694 struct bio *bio; 6695 6696 bio = bio_alloc_bioset(NULL, nr_vecs, opf, GFP_NOFS, &btrfs_bioset); 6697 btrfs_bio_init(btrfs_bio(bio), end_io, private); 6698 return bio; 6699 } 6700 6701 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size, 6702 btrfs_bio_end_io_t end_io, void *private) 6703 { 6704 struct bio *bio; 6705 struct btrfs_bio *bbio; 6706 6707 ASSERT(offset <= UINT_MAX && size <= UINT_MAX); 6708 6709 bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset); 6710 bbio = btrfs_bio(bio); 6711 btrfs_bio_init(bbio, end_io, private); 6712 6713 bio_trim(bio, offset >> 9, size >> 9); 6714 bbio->iter = bio->bi_iter; 6715 return bio; 6716 } 6717 6718 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev) 6719 { 6720 if (!dev || !dev->bdev) 6721 return; 6722 if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET) 6723 return; 6724 6725 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6726 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); 6727 if (!(bio->bi_opf & REQ_RAHEAD)) 6728 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); 6729 if (bio->bi_opf & REQ_PREFLUSH) 6730 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS); 6731 } 6732 6733 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info, 6734 struct bio *bio) 6735 { 6736 if (bio->bi_opf & REQ_META) 6737 return fs_info->endio_meta_workers; 6738 return fs_info->endio_workers; 6739 } 6740 6741 static void btrfs_end_bio_work(struct work_struct *work) 6742 { 6743 struct btrfs_bio *bbio = 6744 container_of(work, struct btrfs_bio, end_io_work); 6745 6746 bbio->end_io(bbio); 6747 } 6748 6749 static void btrfs_simple_end_io(struct bio *bio) 6750 { 6751 struct btrfs_fs_info *fs_info = bio->bi_private; 6752 struct btrfs_bio *bbio = btrfs_bio(bio); 6753 6754 btrfs_bio_counter_dec(fs_info); 6755 6756 if (bio->bi_status) 6757 btrfs_log_dev_io_error(bio, bbio->device); 6758 6759 if (bio_op(bio) == REQ_OP_READ) { 6760 INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work); 6761 queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work); 6762 } else { 6763 bbio->end_io(bbio); 6764 } 6765 } 6766 6767 static void btrfs_raid56_end_io(struct bio *bio) 6768 { 6769 struct btrfs_io_context *bioc = bio->bi_private; 6770 struct btrfs_bio *bbio = btrfs_bio(bio); 6771 6772 btrfs_bio_counter_dec(bioc->fs_info); 6773 bbio->mirror_num = bioc->mirror_num; 6774 bbio->end_io(bbio); 6775 6776 btrfs_put_bioc(bioc); 6777 } 6778 6779 static void btrfs_orig_write_end_io(struct bio *bio) 6780 { 6781 struct btrfs_io_stripe *stripe = bio->bi_private; 6782 struct btrfs_io_context *bioc = stripe->bioc; 6783 struct btrfs_bio *bbio = btrfs_bio(bio); 6784 6785 btrfs_bio_counter_dec(bioc->fs_info); 6786 6787 if (bio->bi_status) { 6788 atomic_inc(&bioc->error); 6789 btrfs_log_dev_io_error(bio, stripe->dev); 6790 } 6791 6792 /* 6793 * Only send an error to the higher layers if it is beyond the tolerance 6794 * threshold. 6795 */ 6796 if (atomic_read(&bioc->error) > bioc->max_errors) 6797 bio->bi_status = BLK_STS_IOERR; 6798 else 6799 bio->bi_status = BLK_STS_OK; 6800 6801 bbio->end_io(bbio); 6802 btrfs_put_bioc(bioc); 6803 } 6804 6805 static void btrfs_clone_write_end_io(struct bio *bio) 6806 { 6807 struct btrfs_io_stripe *stripe = bio->bi_private; 6808 6809 if (bio->bi_status) { 6810 atomic_inc(&stripe->bioc->error); 6811 btrfs_log_dev_io_error(bio, stripe->dev); 6812 } 6813 6814 /* Pass on control to the original bio this one was cloned from */ 6815 bio_endio(stripe->bioc->orig_bio); 6816 bio_put(bio); 6817 } 6818 6819 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio) 6820 { 6821 if (!dev || !dev->bdev || 6822 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 6823 (btrfs_op(bio) == BTRFS_MAP_WRITE && 6824 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6825 bio_io_error(bio); 6826 return; 6827 } 6828 6829 bio_set_dev(bio, dev->bdev); 6830 6831 /* 6832 * For zone append writing, bi_sector must point the beginning of the 6833 * zone 6834 */ 6835 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6836 u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; 6837 6838 if (btrfs_dev_is_sequential(dev, physical)) { 6839 u64 zone_start = round_down(physical, 6840 dev->fs_info->zone_size); 6841 6842 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6843 } else { 6844 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6845 bio->bi_opf |= REQ_OP_WRITE; 6846 } 6847 } 6848 btrfs_debug_in_rcu(dev->fs_info, 6849 "%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6850 __func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6851 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6852 dev->devid, bio->bi_iter.bi_size); 6853 6854 btrfsic_check_bio(bio); 6855 submit_bio(bio); 6856 } 6857 6858 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr) 6859 { 6860 struct bio *orig_bio = bioc->orig_bio, *bio; 6861 6862 ASSERT(bio_op(orig_bio) != REQ_OP_READ); 6863 6864 /* Reuse the bio embedded into the btrfs_bio for the last mirror */ 6865 if (dev_nr == bioc->num_stripes - 1) { 6866 bio = orig_bio; 6867 bio->bi_end_io = btrfs_orig_write_end_io; 6868 } else { 6869 bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set); 6870 bio_inc_remaining(orig_bio); 6871 bio->bi_end_io = btrfs_clone_write_end_io; 6872 } 6873 6874 bio->bi_private = &bioc->stripes[dev_nr]; 6875 bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT; 6876 bioc->stripes[dev_nr].bioc = bioc; 6877 btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio); 6878 } 6879 6880 void btrfs_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, int mirror_num) 6881 { 6882 u64 logical = bio->bi_iter.bi_sector << 9; 6883 u64 length = bio->bi_iter.bi_size; 6884 u64 map_length = length; 6885 struct btrfs_io_context *bioc = NULL; 6886 struct btrfs_io_stripe smap; 6887 int ret; 6888 6889 btrfs_bio_counter_inc_blocked(fs_info); 6890 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, 6891 &bioc, &smap, &mirror_num, 1); 6892 if (ret) { 6893 btrfs_bio_counter_dec(fs_info); 6894 btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret)); 6895 return; 6896 } 6897 6898 if (map_length < length) { 6899 btrfs_crit(fs_info, 6900 "mapping failed logical %llu bio len %llu len %llu", 6901 logical, length, map_length); 6902 BUG(); 6903 } 6904 6905 if (!bioc) { 6906 /* Single mirror read/write fast path */ 6907 btrfs_bio(bio)->mirror_num = mirror_num; 6908 btrfs_bio(bio)->device = smap.dev; 6909 bio->bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT; 6910 bio->bi_private = fs_info; 6911 bio->bi_end_io = btrfs_simple_end_io; 6912 btrfs_submit_dev_bio(smap.dev, bio); 6913 } else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6914 /* Parity RAID write or read recovery */ 6915 bio->bi_private = bioc; 6916 bio->bi_end_io = btrfs_raid56_end_io; 6917 if (bio_op(bio) == REQ_OP_READ) 6918 raid56_parity_recover(bio, bioc, mirror_num); 6919 else 6920 raid56_parity_write(bio, bioc); 6921 } else { 6922 /* Write to multiple mirrors */ 6923 int total_devs = bioc->num_stripes; 6924 int dev_nr; 6925 6926 bioc->orig_bio = bio; 6927 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) 6928 btrfs_submit_mirrored_bio(bioc, dev_nr); 6929 } 6930 } 6931 6932 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6933 const struct btrfs_fs_devices *fs_devices) 6934 { 6935 if (args->fsid == NULL) 6936 return true; 6937 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6938 return true; 6939 return false; 6940 } 6941 6942 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6943 const struct btrfs_device *device) 6944 { 6945 if (args->missing) { 6946 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6947 !device->bdev) 6948 return true; 6949 return false; 6950 } 6951 6952 if (device->devid != args->devid) 6953 return false; 6954 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6955 return false; 6956 return true; 6957 } 6958 6959 /* 6960 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6961 * return NULL. 6962 * 6963 * If devid and uuid are both specified, the match must be exact, otherwise 6964 * only devid is used. 6965 */ 6966 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6967 const struct btrfs_dev_lookup_args *args) 6968 { 6969 struct btrfs_device *device; 6970 struct btrfs_fs_devices *seed_devs; 6971 6972 if (dev_args_match_fs_devices(args, fs_devices)) { 6973 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6974 if (dev_args_match_device(args, device)) 6975 return device; 6976 } 6977 } 6978 6979 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6980 if (!dev_args_match_fs_devices(args, seed_devs)) 6981 continue; 6982 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6983 if (dev_args_match_device(args, device)) 6984 return device; 6985 } 6986 } 6987 6988 return NULL; 6989 } 6990 6991 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6992 u64 devid, u8 *dev_uuid) 6993 { 6994 struct btrfs_device *device; 6995 unsigned int nofs_flag; 6996 6997 /* 6998 * We call this under the chunk_mutex, so we want to use NOFS for this 6999 * allocation, however we don't want to change btrfs_alloc_device() to 7000 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 7001 * places. 7002 */ 7003 nofs_flag = memalloc_nofs_save(); 7004 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 7005 memalloc_nofs_restore(nofs_flag); 7006 if (IS_ERR(device)) 7007 return device; 7008 7009 list_add(&device->dev_list, &fs_devices->devices); 7010 device->fs_devices = fs_devices; 7011 fs_devices->num_devices++; 7012 7013 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7014 fs_devices->missing_devices++; 7015 7016 return device; 7017 } 7018 7019 /* 7020 * Allocate new device struct, set up devid and UUID. 7021 * 7022 * @fs_info: used only for generating a new devid, can be NULL if 7023 * devid is provided (i.e. @devid != NULL). 7024 * @devid: a pointer to devid for this device. If NULL a new devid 7025 * is generated. 7026 * @uuid: a pointer to UUID for this device. If NULL a new UUID 7027 * is generated. 7028 * 7029 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 7030 * on error. Returned struct is not linked onto any lists and must be 7031 * destroyed with btrfs_free_device. 7032 */ 7033 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 7034 const u64 *devid, 7035 const u8 *uuid) 7036 { 7037 struct btrfs_device *dev; 7038 u64 tmp; 7039 7040 if (WARN_ON(!devid && !fs_info)) 7041 return ERR_PTR(-EINVAL); 7042 7043 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 7044 if (!dev) 7045 return ERR_PTR(-ENOMEM); 7046 7047 INIT_LIST_HEAD(&dev->dev_list); 7048 INIT_LIST_HEAD(&dev->dev_alloc_list); 7049 INIT_LIST_HEAD(&dev->post_commit_list); 7050 7051 atomic_set(&dev->dev_stats_ccnt, 0); 7052 btrfs_device_data_ordered_init(dev); 7053 extent_io_tree_init(fs_info, &dev->alloc_state, 7054 IO_TREE_DEVICE_ALLOC_STATE, NULL); 7055 7056 if (devid) 7057 tmp = *devid; 7058 else { 7059 int ret; 7060 7061 ret = find_next_devid(fs_info, &tmp); 7062 if (ret) { 7063 btrfs_free_device(dev); 7064 return ERR_PTR(ret); 7065 } 7066 } 7067 dev->devid = tmp; 7068 7069 if (uuid) 7070 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 7071 else 7072 generate_random_uuid(dev->uuid); 7073 7074 return dev; 7075 } 7076 7077 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 7078 u64 devid, u8 *uuid, bool error) 7079 { 7080 if (error) 7081 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 7082 devid, uuid); 7083 else 7084 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 7085 devid, uuid); 7086 } 7087 7088 u64 btrfs_calc_stripe_length(const struct extent_map *em) 7089 { 7090 const struct map_lookup *map = em->map_lookup; 7091 const int data_stripes = calc_data_stripes(map->type, map->num_stripes); 7092 7093 return div_u64(em->len, data_stripes); 7094 } 7095 7096 #if BITS_PER_LONG == 32 7097 /* 7098 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 7099 * can't be accessed on 32bit systems. 7100 * 7101 * This function do mount time check to reject the fs if it already has 7102 * metadata chunk beyond that limit. 7103 */ 7104 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7105 u64 logical, u64 length, u64 type) 7106 { 7107 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7108 return 0; 7109 7110 if (logical + length < MAX_LFS_FILESIZE) 7111 return 0; 7112 7113 btrfs_err_32bit_limit(fs_info); 7114 return -EOVERFLOW; 7115 } 7116 7117 /* 7118 * This is to give early warning for any metadata chunk reaching 7119 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 7120 * Although we can still access the metadata, it's not going to be possible 7121 * once the limit is reached. 7122 */ 7123 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7124 u64 logical, u64 length, u64 type) 7125 { 7126 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7127 return; 7128 7129 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 7130 return; 7131 7132 btrfs_warn_32bit_limit(fs_info); 7133 } 7134 #endif 7135 7136 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 7137 u64 devid, u8 *uuid) 7138 { 7139 struct btrfs_device *dev; 7140 7141 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7142 btrfs_report_missing_device(fs_info, devid, uuid, true); 7143 return ERR_PTR(-ENOENT); 7144 } 7145 7146 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 7147 if (IS_ERR(dev)) { 7148 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 7149 devid, PTR_ERR(dev)); 7150 return dev; 7151 } 7152 btrfs_report_missing_device(fs_info, devid, uuid, false); 7153 7154 return dev; 7155 } 7156 7157 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 7158 struct btrfs_chunk *chunk) 7159 { 7160 BTRFS_DEV_LOOKUP_ARGS(args); 7161 struct btrfs_fs_info *fs_info = leaf->fs_info; 7162 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7163 struct map_lookup *map; 7164 struct extent_map *em; 7165 u64 logical; 7166 u64 length; 7167 u64 devid; 7168 u64 type; 7169 u8 uuid[BTRFS_UUID_SIZE]; 7170 int index; 7171 int num_stripes; 7172 int ret; 7173 int i; 7174 7175 logical = key->offset; 7176 length = btrfs_chunk_length(leaf, chunk); 7177 type = btrfs_chunk_type(leaf, chunk); 7178 index = btrfs_bg_flags_to_raid_index(type); 7179 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 7180 7181 #if BITS_PER_LONG == 32 7182 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 7183 if (ret < 0) 7184 return ret; 7185 warn_32bit_meta_chunk(fs_info, logical, length, type); 7186 #endif 7187 7188 /* 7189 * Only need to verify chunk item if we're reading from sys chunk array, 7190 * as chunk item in tree block is already verified by tree-checker. 7191 */ 7192 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 7193 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 7194 if (ret) 7195 return ret; 7196 } 7197 7198 read_lock(&map_tree->lock); 7199 em = lookup_extent_mapping(map_tree, logical, 1); 7200 read_unlock(&map_tree->lock); 7201 7202 /* already mapped? */ 7203 if (em && em->start <= logical && em->start + em->len > logical) { 7204 free_extent_map(em); 7205 return 0; 7206 } else if (em) { 7207 free_extent_map(em); 7208 } 7209 7210 em = alloc_extent_map(); 7211 if (!em) 7212 return -ENOMEM; 7213 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7214 if (!map) { 7215 free_extent_map(em); 7216 return -ENOMEM; 7217 } 7218 7219 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7220 em->map_lookup = map; 7221 em->start = logical; 7222 em->len = length; 7223 em->orig_start = 0; 7224 em->block_start = 0; 7225 em->block_len = em->len; 7226 7227 map->num_stripes = num_stripes; 7228 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7229 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7230 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7231 map->type = type; 7232 /* 7233 * We can't use the sub_stripes value, as for profiles other than 7234 * RAID10, they may have 0 as sub_stripes for filesystems created by 7235 * older mkfs (<v5.4). 7236 * In that case, it can cause divide-by-zero errors later. 7237 * Since currently sub_stripes is fixed for each profile, let's 7238 * use the trusted value instead. 7239 */ 7240 map->sub_stripes = btrfs_raid_array[index].sub_stripes; 7241 map->verified_stripes = 0; 7242 em->orig_block_len = btrfs_calc_stripe_length(em); 7243 for (i = 0; i < num_stripes; i++) { 7244 map->stripes[i].physical = 7245 btrfs_stripe_offset_nr(leaf, chunk, i); 7246 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7247 args.devid = devid; 7248 read_extent_buffer(leaf, uuid, (unsigned long) 7249 btrfs_stripe_dev_uuid_nr(chunk, i), 7250 BTRFS_UUID_SIZE); 7251 args.uuid = uuid; 7252 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 7253 if (!map->stripes[i].dev) { 7254 map->stripes[i].dev = handle_missing_device(fs_info, 7255 devid, uuid); 7256 if (IS_ERR(map->stripes[i].dev)) { 7257 free_extent_map(em); 7258 return PTR_ERR(map->stripes[i].dev); 7259 } 7260 } 7261 7262 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7263 &(map->stripes[i].dev->dev_state)); 7264 } 7265 7266 write_lock(&map_tree->lock); 7267 ret = add_extent_mapping(map_tree, em, 0); 7268 write_unlock(&map_tree->lock); 7269 if (ret < 0) { 7270 btrfs_err(fs_info, 7271 "failed to add chunk map, start=%llu len=%llu: %d", 7272 em->start, em->len, ret); 7273 } 7274 free_extent_map(em); 7275 7276 return ret; 7277 } 7278 7279 static void fill_device_from_item(struct extent_buffer *leaf, 7280 struct btrfs_dev_item *dev_item, 7281 struct btrfs_device *device) 7282 { 7283 unsigned long ptr; 7284 7285 device->devid = btrfs_device_id(leaf, dev_item); 7286 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7287 device->total_bytes = device->disk_total_bytes; 7288 device->commit_total_bytes = device->disk_total_bytes; 7289 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7290 device->commit_bytes_used = device->bytes_used; 7291 device->type = btrfs_device_type(leaf, dev_item); 7292 device->io_align = btrfs_device_io_align(leaf, dev_item); 7293 device->io_width = btrfs_device_io_width(leaf, dev_item); 7294 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7295 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7296 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7297 7298 ptr = btrfs_device_uuid(dev_item); 7299 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7300 } 7301 7302 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7303 u8 *fsid) 7304 { 7305 struct btrfs_fs_devices *fs_devices; 7306 int ret; 7307 7308 lockdep_assert_held(&uuid_mutex); 7309 ASSERT(fsid); 7310 7311 /* This will match only for multi-device seed fs */ 7312 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7313 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7314 return fs_devices; 7315 7316 7317 fs_devices = find_fsid(fsid, NULL); 7318 if (!fs_devices) { 7319 if (!btrfs_test_opt(fs_info, DEGRADED)) 7320 return ERR_PTR(-ENOENT); 7321 7322 fs_devices = alloc_fs_devices(fsid, NULL); 7323 if (IS_ERR(fs_devices)) 7324 return fs_devices; 7325 7326 fs_devices->seeding = true; 7327 fs_devices->opened = 1; 7328 return fs_devices; 7329 } 7330 7331 /* 7332 * Upon first call for a seed fs fsid, just create a private copy of the 7333 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7334 */ 7335 fs_devices = clone_fs_devices(fs_devices); 7336 if (IS_ERR(fs_devices)) 7337 return fs_devices; 7338 7339 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7340 if (ret) { 7341 free_fs_devices(fs_devices); 7342 return ERR_PTR(ret); 7343 } 7344 7345 if (!fs_devices->seeding) { 7346 close_fs_devices(fs_devices); 7347 free_fs_devices(fs_devices); 7348 return ERR_PTR(-EINVAL); 7349 } 7350 7351 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7352 7353 return fs_devices; 7354 } 7355 7356 static int read_one_dev(struct extent_buffer *leaf, 7357 struct btrfs_dev_item *dev_item) 7358 { 7359 BTRFS_DEV_LOOKUP_ARGS(args); 7360 struct btrfs_fs_info *fs_info = leaf->fs_info; 7361 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7362 struct btrfs_device *device; 7363 u64 devid; 7364 int ret; 7365 u8 fs_uuid[BTRFS_FSID_SIZE]; 7366 u8 dev_uuid[BTRFS_UUID_SIZE]; 7367 7368 devid = btrfs_device_id(leaf, dev_item); 7369 args.devid = devid; 7370 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7371 BTRFS_UUID_SIZE); 7372 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7373 BTRFS_FSID_SIZE); 7374 args.uuid = dev_uuid; 7375 args.fsid = fs_uuid; 7376 7377 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7378 fs_devices = open_seed_devices(fs_info, fs_uuid); 7379 if (IS_ERR(fs_devices)) 7380 return PTR_ERR(fs_devices); 7381 } 7382 7383 device = btrfs_find_device(fs_info->fs_devices, &args); 7384 if (!device) { 7385 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7386 btrfs_report_missing_device(fs_info, devid, 7387 dev_uuid, true); 7388 return -ENOENT; 7389 } 7390 7391 device = add_missing_dev(fs_devices, devid, dev_uuid); 7392 if (IS_ERR(device)) { 7393 btrfs_err(fs_info, 7394 "failed to add missing dev %llu: %ld", 7395 devid, PTR_ERR(device)); 7396 return PTR_ERR(device); 7397 } 7398 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7399 } else { 7400 if (!device->bdev) { 7401 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7402 btrfs_report_missing_device(fs_info, 7403 devid, dev_uuid, true); 7404 return -ENOENT; 7405 } 7406 btrfs_report_missing_device(fs_info, devid, 7407 dev_uuid, false); 7408 } 7409 7410 if (!device->bdev && 7411 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7412 /* 7413 * this happens when a device that was properly setup 7414 * in the device info lists suddenly goes bad. 7415 * device->bdev is NULL, and so we have to set 7416 * device->missing to one here 7417 */ 7418 device->fs_devices->missing_devices++; 7419 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7420 } 7421 7422 /* Move the device to its own fs_devices */ 7423 if (device->fs_devices != fs_devices) { 7424 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7425 &device->dev_state)); 7426 7427 list_move(&device->dev_list, &fs_devices->devices); 7428 device->fs_devices->num_devices--; 7429 fs_devices->num_devices++; 7430 7431 device->fs_devices->missing_devices--; 7432 fs_devices->missing_devices++; 7433 7434 device->fs_devices = fs_devices; 7435 } 7436 } 7437 7438 if (device->fs_devices != fs_info->fs_devices) { 7439 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7440 if (device->generation != 7441 btrfs_device_generation(leaf, dev_item)) 7442 return -EINVAL; 7443 } 7444 7445 fill_device_from_item(leaf, dev_item, device); 7446 if (device->bdev) { 7447 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7448 7449 if (device->total_bytes > max_total_bytes) { 7450 btrfs_err(fs_info, 7451 "device total_bytes should be at most %llu but found %llu", 7452 max_total_bytes, device->total_bytes); 7453 return -EINVAL; 7454 } 7455 } 7456 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7457 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7458 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7459 device->fs_devices->total_rw_bytes += device->total_bytes; 7460 atomic64_add(device->total_bytes - device->bytes_used, 7461 &fs_info->free_chunk_space); 7462 } 7463 ret = 0; 7464 return ret; 7465 } 7466 7467 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7468 { 7469 struct btrfs_super_block *super_copy = fs_info->super_copy; 7470 struct extent_buffer *sb; 7471 struct btrfs_disk_key *disk_key; 7472 struct btrfs_chunk *chunk; 7473 u8 *array_ptr; 7474 unsigned long sb_array_offset; 7475 int ret = 0; 7476 u32 num_stripes; 7477 u32 array_size; 7478 u32 len = 0; 7479 u32 cur_offset; 7480 u64 type; 7481 struct btrfs_key key; 7482 7483 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7484 7485 /* 7486 * We allocated a dummy extent, just to use extent buffer accessors. 7487 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but 7488 * that's fine, we will not go beyond system chunk array anyway. 7489 */ 7490 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); 7491 if (!sb) 7492 return -ENOMEM; 7493 set_extent_buffer_uptodate(sb); 7494 7495 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7496 array_size = btrfs_super_sys_array_size(super_copy); 7497 7498 array_ptr = super_copy->sys_chunk_array; 7499 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7500 cur_offset = 0; 7501 7502 while (cur_offset < array_size) { 7503 disk_key = (struct btrfs_disk_key *)array_ptr; 7504 len = sizeof(*disk_key); 7505 if (cur_offset + len > array_size) 7506 goto out_short_read; 7507 7508 btrfs_disk_key_to_cpu(&key, disk_key); 7509 7510 array_ptr += len; 7511 sb_array_offset += len; 7512 cur_offset += len; 7513 7514 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7515 btrfs_err(fs_info, 7516 "unexpected item type %u in sys_array at offset %u", 7517 (u32)key.type, cur_offset); 7518 ret = -EIO; 7519 break; 7520 } 7521 7522 chunk = (struct btrfs_chunk *)sb_array_offset; 7523 /* 7524 * At least one btrfs_chunk with one stripe must be present, 7525 * exact stripe count check comes afterwards 7526 */ 7527 len = btrfs_chunk_item_size(1); 7528 if (cur_offset + len > array_size) 7529 goto out_short_read; 7530 7531 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7532 if (!num_stripes) { 7533 btrfs_err(fs_info, 7534 "invalid number of stripes %u in sys_array at offset %u", 7535 num_stripes, cur_offset); 7536 ret = -EIO; 7537 break; 7538 } 7539 7540 type = btrfs_chunk_type(sb, chunk); 7541 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7542 btrfs_err(fs_info, 7543 "invalid chunk type %llu in sys_array at offset %u", 7544 type, cur_offset); 7545 ret = -EIO; 7546 break; 7547 } 7548 7549 len = btrfs_chunk_item_size(num_stripes); 7550 if (cur_offset + len > array_size) 7551 goto out_short_read; 7552 7553 ret = read_one_chunk(&key, sb, chunk); 7554 if (ret) 7555 break; 7556 7557 array_ptr += len; 7558 sb_array_offset += len; 7559 cur_offset += len; 7560 } 7561 clear_extent_buffer_uptodate(sb); 7562 free_extent_buffer_stale(sb); 7563 return ret; 7564 7565 out_short_read: 7566 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7567 len, cur_offset); 7568 clear_extent_buffer_uptodate(sb); 7569 free_extent_buffer_stale(sb); 7570 return -EIO; 7571 } 7572 7573 /* 7574 * Check if all chunks in the fs are OK for read-write degraded mount 7575 * 7576 * If the @failing_dev is specified, it's accounted as missing. 7577 * 7578 * Return true if all chunks meet the minimal RW mount requirements. 7579 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7580 */ 7581 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7582 struct btrfs_device *failing_dev) 7583 { 7584 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7585 struct extent_map *em; 7586 u64 next_start = 0; 7587 bool ret = true; 7588 7589 read_lock(&map_tree->lock); 7590 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7591 read_unlock(&map_tree->lock); 7592 /* No chunk at all? Return false anyway */ 7593 if (!em) { 7594 ret = false; 7595 goto out; 7596 } 7597 while (em) { 7598 struct map_lookup *map; 7599 int missing = 0; 7600 int max_tolerated; 7601 int i; 7602 7603 map = em->map_lookup; 7604 max_tolerated = 7605 btrfs_get_num_tolerated_disk_barrier_failures( 7606 map->type); 7607 for (i = 0; i < map->num_stripes; i++) { 7608 struct btrfs_device *dev = map->stripes[i].dev; 7609 7610 if (!dev || !dev->bdev || 7611 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7612 dev->last_flush_error) 7613 missing++; 7614 else if (failing_dev && failing_dev == dev) 7615 missing++; 7616 } 7617 if (missing > max_tolerated) { 7618 if (!failing_dev) 7619 btrfs_warn(fs_info, 7620 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7621 em->start, missing, max_tolerated); 7622 free_extent_map(em); 7623 ret = false; 7624 goto out; 7625 } 7626 next_start = extent_map_end(em); 7627 free_extent_map(em); 7628 7629 read_lock(&map_tree->lock); 7630 em = lookup_extent_mapping(map_tree, next_start, 7631 (u64)(-1) - next_start); 7632 read_unlock(&map_tree->lock); 7633 } 7634 out: 7635 return ret; 7636 } 7637 7638 static void readahead_tree_node_children(struct extent_buffer *node) 7639 { 7640 int i; 7641 const int nr_items = btrfs_header_nritems(node); 7642 7643 for (i = 0; i < nr_items; i++) 7644 btrfs_readahead_node_child(node, i); 7645 } 7646 7647 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7648 { 7649 struct btrfs_root *root = fs_info->chunk_root; 7650 struct btrfs_path *path; 7651 struct extent_buffer *leaf; 7652 struct btrfs_key key; 7653 struct btrfs_key found_key; 7654 int ret; 7655 int slot; 7656 int iter_ret = 0; 7657 u64 total_dev = 0; 7658 u64 last_ra_node = 0; 7659 7660 path = btrfs_alloc_path(); 7661 if (!path) 7662 return -ENOMEM; 7663 7664 /* 7665 * uuid_mutex is needed only if we are mounting a sprout FS 7666 * otherwise we don't need it. 7667 */ 7668 mutex_lock(&uuid_mutex); 7669 7670 /* 7671 * It is possible for mount and umount to race in such a way that 7672 * we execute this code path, but open_fs_devices failed to clear 7673 * total_rw_bytes. We certainly want it cleared before reading the 7674 * device items, so clear it here. 7675 */ 7676 fs_info->fs_devices->total_rw_bytes = 0; 7677 7678 /* 7679 * Lockdep complains about possible circular locking dependency between 7680 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7681 * used for freeze procection of a fs (struct super_block.s_writers), 7682 * which we take when starting a transaction, and extent buffers of the 7683 * chunk tree if we call read_one_dev() while holding a lock on an 7684 * extent buffer of the chunk tree. Since we are mounting the filesystem 7685 * and at this point there can't be any concurrent task modifying the 7686 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7687 */ 7688 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7689 path->skip_locking = 1; 7690 7691 /* 7692 * Read all device items, and then all the chunk items. All 7693 * device items are found before any chunk item (their object id 7694 * is smaller than the lowest possible object id for a chunk 7695 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7696 */ 7697 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7698 key.offset = 0; 7699 key.type = 0; 7700 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 7701 struct extent_buffer *node = path->nodes[1]; 7702 7703 leaf = path->nodes[0]; 7704 slot = path->slots[0]; 7705 7706 if (node) { 7707 if (last_ra_node != node->start) { 7708 readahead_tree_node_children(node); 7709 last_ra_node = node->start; 7710 } 7711 } 7712 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7713 struct btrfs_dev_item *dev_item; 7714 dev_item = btrfs_item_ptr(leaf, slot, 7715 struct btrfs_dev_item); 7716 ret = read_one_dev(leaf, dev_item); 7717 if (ret) 7718 goto error; 7719 total_dev++; 7720 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7721 struct btrfs_chunk *chunk; 7722 7723 /* 7724 * We are only called at mount time, so no need to take 7725 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7726 * we always lock first fs_info->chunk_mutex before 7727 * acquiring any locks on the chunk tree. This is a 7728 * requirement for chunk allocation, see the comment on 7729 * top of btrfs_chunk_alloc() for details. 7730 */ 7731 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7732 ret = read_one_chunk(&found_key, leaf, chunk); 7733 if (ret) 7734 goto error; 7735 } 7736 } 7737 /* Catch error found during iteration */ 7738 if (iter_ret < 0) { 7739 ret = iter_ret; 7740 goto error; 7741 } 7742 7743 /* 7744 * After loading chunk tree, we've got all device information, 7745 * do another round of validation checks. 7746 */ 7747 if (total_dev != fs_info->fs_devices->total_devices) { 7748 btrfs_warn(fs_info, 7749 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", 7750 btrfs_super_num_devices(fs_info->super_copy), 7751 total_dev); 7752 fs_info->fs_devices->total_devices = total_dev; 7753 btrfs_set_super_num_devices(fs_info->super_copy, total_dev); 7754 } 7755 if (btrfs_super_total_bytes(fs_info->super_copy) < 7756 fs_info->fs_devices->total_rw_bytes) { 7757 btrfs_err(fs_info, 7758 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7759 btrfs_super_total_bytes(fs_info->super_copy), 7760 fs_info->fs_devices->total_rw_bytes); 7761 ret = -EINVAL; 7762 goto error; 7763 } 7764 ret = 0; 7765 error: 7766 mutex_unlock(&uuid_mutex); 7767 7768 btrfs_free_path(path); 7769 return ret; 7770 } 7771 7772 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7773 { 7774 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7775 struct btrfs_device *device; 7776 int ret = 0; 7777 7778 fs_devices->fs_info = fs_info; 7779 7780 mutex_lock(&fs_devices->device_list_mutex); 7781 list_for_each_entry(device, &fs_devices->devices, dev_list) 7782 device->fs_info = fs_info; 7783 7784 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7785 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7786 device->fs_info = fs_info; 7787 ret = btrfs_get_dev_zone_info(device, false); 7788 if (ret) 7789 break; 7790 } 7791 7792 seed_devs->fs_info = fs_info; 7793 } 7794 mutex_unlock(&fs_devices->device_list_mutex); 7795 7796 return ret; 7797 } 7798 7799 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7800 const struct btrfs_dev_stats_item *ptr, 7801 int index) 7802 { 7803 u64 val; 7804 7805 read_extent_buffer(eb, &val, 7806 offsetof(struct btrfs_dev_stats_item, values) + 7807 ((unsigned long)ptr) + (index * sizeof(u64)), 7808 sizeof(val)); 7809 return val; 7810 } 7811 7812 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7813 struct btrfs_dev_stats_item *ptr, 7814 int index, u64 val) 7815 { 7816 write_extent_buffer(eb, &val, 7817 offsetof(struct btrfs_dev_stats_item, values) + 7818 ((unsigned long)ptr) + (index * sizeof(u64)), 7819 sizeof(val)); 7820 } 7821 7822 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7823 struct btrfs_path *path) 7824 { 7825 struct btrfs_dev_stats_item *ptr; 7826 struct extent_buffer *eb; 7827 struct btrfs_key key; 7828 int item_size; 7829 int i, ret, slot; 7830 7831 if (!device->fs_info->dev_root) 7832 return 0; 7833 7834 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7835 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7836 key.offset = device->devid; 7837 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7838 if (ret) { 7839 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7840 btrfs_dev_stat_set(device, i, 0); 7841 device->dev_stats_valid = 1; 7842 btrfs_release_path(path); 7843 return ret < 0 ? ret : 0; 7844 } 7845 slot = path->slots[0]; 7846 eb = path->nodes[0]; 7847 item_size = btrfs_item_size(eb, slot); 7848 7849 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7850 7851 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7852 if (item_size >= (1 + i) * sizeof(__le64)) 7853 btrfs_dev_stat_set(device, i, 7854 btrfs_dev_stats_value(eb, ptr, i)); 7855 else 7856 btrfs_dev_stat_set(device, i, 0); 7857 } 7858 7859 device->dev_stats_valid = 1; 7860 btrfs_dev_stat_print_on_load(device); 7861 btrfs_release_path(path); 7862 7863 return 0; 7864 } 7865 7866 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7867 { 7868 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7869 struct btrfs_device *device; 7870 struct btrfs_path *path = NULL; 7871 int ret = 0; 7872 7873 path = btrfs_alloc_path(); 7874 if (!path) 7875 return -ENOMEM; 7876 7877 mutex_lock(&fs_devices->device_list_mutex); 7878 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7879 ret = btrfs_device_init_dev_stats(device, path); 7880 if (ret) 7881 goto out; 7882 } 7883 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7884 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7885 ret = btrfs_device_init_dev_stats(device, path); 7886 if (ret) 7887 goto out; 7888 } 7889 } 7890 out: 7891 mutex_unlock(&fs_devices->device_list_mutex); 7892 7893 btrfs_free_path(path); 7894 return ret; 7895 } 7896 7897 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7898 struct btrfs_device *device) 7899 { 7900 struct btrfs_fs_info *fs_info = trans->fs_info; 7901 struct btrfs_root *dev_root = fs_info->dev_root; 7902 struct btrfs_path *path; 7903 struct btrfs_key key; 7904 struct extent_buffer *eb; 7905 struct btrfs_dev_stats_item *ptr; 7906 int ret; 7907 int i; 7908 7909 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7910 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7911 key.offset = device->devid; 7912 7913 path = btrfs_alloc_path(); 7914 if (!path) 7915 return -ENOMEM; 7916 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7917 if (ret < 0) { 7918 btrfs_warn_in_rcu(fs_info, 7919 "error %d while searching for dev_stats item for device %s", 7920 ret, rcu_str_deref(device->name)); 7921 goto out; 7922 } 7923 7924 if (ret == 0 && 7925 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7926 /* need to delete old one and insert a new one */ 7927 ret = btrfs_del_item(trans, dev_root, path); 7928 if (ret != 0) { 7929 btrfs_warn_in_rcu(fs_info, 7930 "delete too small dev_stats item for device %s failed %d", 7931 rcu_str_deref(device->name), ret); 7932 goto out; 7933 } 7934 ret = 1; 7935 } 7936 7937 if (ret == 1) { 7938 /* need to insert a new item */ 7939 btrfs_release_path(path); 7940 ret = btrfs_insert_empty_item(trans, dev_root, path, 7941 &key, sizeof(*ptr)); 7942 if (ret < 0) { 7943 btrfs_warn_in_rcu(fs_info, 7944 "insert dev_stats item for device %s failed %d", 7945 rcu_str_deref(device->name), ret); 7946 goto out; 7947 } 7948 } 7949 7950 eb = path->nodes[0]; 7951 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7952 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7953 btrfs_set_dev_stats_value(eb, ptr, i, 7954 btrfs_dev_stat_read(device, i)); 7955 btrfs_mark_buffer_dirty(eb); 7956 7957 out: 7958 btrfs_free_path(path); 7959 return ret; 7960 } 7961 7962 /* 7963 * called from commit_transaction. Writes all changed device stats to disk. 7964 */ 7965 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7966 { 7967 struct btrfs_fs_info *fs_info = trans->fs_info; 7968 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7969 struct btrfs_device *device; 7970 int stats_cnt; 7971 int ret = 0; 7972 7973 mutex_lock(&fs_devices->device_list_mutex); 7974 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7975 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7976 if (!device->dev_stats_valid || stats_cnt == 0) 7977 continue; 7978 7979 7980 /* 7981 * There is a LOAD-LOAD control dependency between the value of 7982 * dev_stats_ccnt and updating the on-disk values which requires 7983 * reading the in-memory counters. Such control dependencies 7984 * require explicit read memory barriers. 7985 * 7986 * This memory barriers pairs with smp_mb__before_atomic in 7987 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7988 * barrier implied by atomic_xchg in 7989 * btrfs_dev_stats_read_and_reset 7990 */ 7991 smp_rmb(); 7992 7993 ret = update_dev_stat_item(trans, device); 7994 if (!ret) 7995 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7996 } 7997 mutex_unlock(&fs_devices->device_list_mutex); 7998 7999 return ret; 8000 } 8001 8002 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 8003 { 8004 btrfs_dev_stat_inc(dev, index); 8005 8006 if (!dev->dev_stats_valid) 8007 return; 8008 btrfs_err_rl_in_rcu(dev->fs_info, 8009 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 8010 rcu_str_deref(dev->name), 8011 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 8012 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 8013 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 8014 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 8015 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 8016 } 8017 8018 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 8019 { 8020 int i; 8021 8022 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 8023 if (btrfs_dev_stat_read(dev, i) != 0) 8024 break; 8025 if (i == BTRFS_DEV_STAT_VALUES_MAX) 8026 return; /* all values == 0, suppress message */ 8027 8028 btrfs_info_in_rcu(dev->fs_info, 8029 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 8030 rcu_str_deref(dev->name), 8031 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 8032 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 8033 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 8034 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 8035 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 8036 } 8037 8038 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 8039 struct btrfs_ioctl_get_dev_stats *stats) 8040 { 8041 BTRFS_DEV_LOOKUP_ARGS(args); 8042 struct btrfs_device *dev; 8043 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 8044 int i; 8045 8046 mutex_lock(&fs_devices->device_list_mutex); 8047 args.devid = stats->devid; 8048 dev = btrfs_find_device(fs_info->fs_devices, &args); 8049 mutex_unlock(&fs_devices->device_list_mutex); 8050 8051 if (!dev) { 8052 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 8053 return -ENODEV; 8054 } else if (!dev->dev_stats_valid) { 8055 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 8056 return -ENODEV; 8057 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 8058 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 8059 if (stats->nr_items > i) 8060 stats->values[i] = 8061 btrfs_dev_stat_read_and_reset(dev, i); 8062 else 8063 btrfs_dev_stat_set(dev, i, 0); 8064 } 8065 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 8066 current->comm, task_pid_nr(current)); 8067 } else { 8068 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 8069 if (stats->nr_items > i) 8070 stats->values[i] = btrfs_dev_stat_read(dev, i); 8071 } 8072 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 8073 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 8074 return 0; 8075 } 8076 8077 /* 8078 * Update the size and bytes used for each device where it changed. This is 8079 * delayed since we would otherwise get errors while writing out the 8080 * superblocks. 8081 * 8082 * Must be invoked during transaction commit. 8083 */ 8084 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 8085 { 8086 struct btrfs_device *curr, *next; 8087 8088 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 8089 8090 if (list_empty(&trans->dev_update_list)) 8091 return; 8092 8093 /* 8094 * We don't need the device_list_mutex here. This list is owned by the 8095 * transaction and the transaction must complete before the device is 8096 * released. 8097 */ 8098 mutex_lock(&trans->fs_info->chunk_mutex); 8099 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 8100 post_commit_list) { 8101 list_del_init(&curr->post_commit_list); 8102 curr->commit_total_bytes = curr->disk_total_bytes; 8103 curr->commit_bytes_used = curr->bytes_used; 8104 } 8105 mutex_unlock(&trans->fs_info->chunk_mutex); 8106 } 8107 8108 /* 8109 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 8110 */ 8111 int btrfs_bg_type_to_factor(u64 flags) 8112 { 8113 const int index = btrfs_bg_flags_to_raid_index(flags); 8114 8115 return btrfs_raid_array[index].ncopies; 8116 } 8117 8118 8119 8120 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 8121 u64 chunk_offset, u64 devid, 8122 u64 physical_offset, u64 physical_len) 8123 { 8124 struct btrfs_dev_lookup_args args = { .devid = devid }; 8125 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8126 struct extent_map *em; 8127 struct map_lookup *map; 8128 struct btrfs_device *dev; 8129 u64 stripe_len; 8130 bool found = false; 8131 int ret = 0; 8132 int i; 8133 8134 read_lock(&em_tree->lock); 8135 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 8136 read_unlock(&em_tree->lock); 8137 8138 if (!em) { 8139 btrfs_err(fs_info, 8140 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 8141 physical_offset, devid); 8142 ret = -EUCLEAN; 8143 goto out; 8144 } 8145 8146 map = em->map_lookup; 8147 stripe_len = btrfs_calc_stripe_length(em); 8148 if (physical_len != stripe_len) { 8149 btrfs_err(fs_info, 8150 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 8151 physical_offset, devid, em->start, physical_len, 8152 stripe_len); 8153 ret = -EUCLEAN; 8154 goto out; 8155 } 8156 8157 /* 8158 * Very old mkfs.btrfs (before v4.1) will not respect the reserved 8159 * space. Although kernel can handle it without problem, better to warn 8160 * the users. 8161 */ 8162 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) 8163 btrfs_warn(fs_info, 8164 "devid %llu physical %llu len %llu inside the reserved space", 8165 devid, physical_offset, physical_len); 8166 8167 for (i = 0; i < map->num_stripes; i++) { 8168 if (map->stripes[i].dev->devid == devid && 8169 map->stripes[i].physical == physical_offset) { 8170 found = true; 8171 if (map->verified_stripes >= map->num_stripes) { 8172 btrfs_err(fs_info, 8173 "too many dev extents for chunk %llu found", 8174 em->start); 8175 ret = -EUCLEAN; 8176 goto out; 8177 } 8178 map->verified_stripes++; 8179 break; 8180 } 8181 } 8182 if (!found) { 8183 btrfs_err(fs_info, 8184 "dev extent physical offset %llu devid %llu has no corresponding chunk", 8185 physical_offset, devid); 8186 ret = -EUCLEAN; 8187 } 8188 8189 /* Make sure no dev extent is beyond device boundary */ 8190 dev = btrfs_find_device(fs_info->fs_devices, &args); 8191 if (!dev) { 8192 btrfs_err(fs_info, "failed to find devid %llu", devid); 8193 ret = -EUCLEAN; 8194 goto out; 8195 } 8196 8197 if (physical_offset + physical_len > dev->disk_total_bytes) { 8198 btrfs_err(fs_info, 8199 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 8200 devid, physical_offset, physical_len, 8201 dev->disk_total_bytes); 8202 ret = -EUCLEAN; 8203 goto out; 8204 } 8205 8206 if (dev->zone_info) { 8207 u64 zone_size = dev->zone_info->zone_size; 8208 8209 if (!IS_ALIGNED(physical_offset, zone_size) || 8210 !IS_ALIGNED(physical_len, zone_size)) { 8211 btrfs_err(fs_info, 8212 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8213 devid, physical_offset, physical_len); 8214 ret = -EUCLEAN; 8215 goto out; 8216 } 8217 } 8218 8219 out: 8220 free_extent_map(em); 8221 return ret; 8222 } 8223 8224 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8225 { 8226 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8227 struct extent_map *em; 8228 struct rb_node *node; 8229 int ret = 0; 8230 8231 read_lock(&em_tree->lock); 8232 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8233 em = rb_entry(node, struct extent_map, rb_node); 8234 if (em->map_lookup->num_stripes != 8235 em->map_lookup->verified_stripes) { 8236 btrfs_err(fs_info, 8237 "chunk %llu has missing dev extent, have %d expect %d", 8238 em->start, em->map_lookup->verified_stripes, 8239 em->map_lookup->num_stripes); 8240 ret = -EUCLEAN; 8241 goto out; 8242 } 8243 } 8244 out: 8245 read_unlock(&em_tree->lock); 8246 return ret; 8247 } 8248 8249 /* 8250 * Ensure that all dev extents are mapped to correct chunk, otherwise 8251 * later chunk allocation/free would cause unexpected behavior. 8252 * 8253 * NOTE: This will iterate through the whole device tree, which should be of 8254 * the same size level as the chunk tree. This slightly increases mount time. 8255 */ 8256 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8257 { 8258 struct btrfs_path *path; 8259 struct btrfs_root *root = fs_info->dev_root; 8260 struct btrfs_key key; 8261 u64 prev_devid = 0; 8262 u64 prev_dev_ext_end = 0; 8263 int ret = 0; 8264 8265 /* 8266 * We don't have a dev_root because we mounted with ignorebadroots and 8267 * failed to load the root, so we want to skip the verification in this 8268 * case for sure. 8269 * 8270 * However if the dev root is fine, but the tree itself is corrupted 8271 * we'd still fail to mount. This verification is only to make sure 8272 * writes can happen safely, so instead just bypass this check 8273 * completely in the case of IGNOREBADROOTS. 8274 */ 8275 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8276 return 0; 8277 8278 key.objectid = 1; 8279 key.type = BTRFS_DEV_EXTENT_KEY; 8280 key.offset = 0; 8281 8282 path = btrfs_alloc_path(); 8283 if (!path) 8284 return -ENOMEM; 8285 8286 path->reada = READA_FORWARD; 8287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8288 if (ret < 0) 8289 goto out; 8290 8291 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8292 ret = btrfs_next_leaf(root, path); 8293 if (ret < 0) 8294 goto out; 8295 /* No dev extents at all? Not good */ 8296 if (ret > 0) { 8297 ret = -EUCLEAN; 8298 goto out; 8299 } 8300 } 8301 while (1) { 8302 struct extent_buffer *leaf = path->nodes[0]; 8303 struct btrfs_dev_extent *dext; 8304 int slot = path->slots[0]; 8305 u64 chunk_offset; 8306 u64 physical_offset; 8307 u64 physical_len; 8308 u64 devid; 8309 8310 btrfs_item_key_to_cpu(leaf, &key, slot); 8311 if (key.type != BTRFS_DEV_EXTENT_KEY) 8312 break; 8313 devid = key.objectid; 8314 physical_offset = key.offset; 8315 8316 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8317 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8318 physical_len = btrfs_dev_extent_length(leaf, dext); 8319 8320 /* Check if this dev extent overlaps with the previous one */ 8321 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8322 btrfs_err(fs_info, 8323 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8324 devid, physical_offset, prev_dev_ext_end); 8325 ret = -EUCLEAN; 8326 goto out; 8327 } 8328 8329 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8330 physical_offset, physical_len); 8331 if (ret < 0) 8332 goto out; 8333 prev_devid = devid; 8334 prev_dev_ext_end = physical_offset + physical_len; 8335 8336 ret = btrfs_next_item(root, path); 8337 if (ret < 0) 8338 goto out; 8339 if (ret > 0) { 8340 ret = 0; 8341 break; 8342 } 8343 } 8344 8345 /* Ensure all chunks have corresponding dev extents */ 8346 ret = verify_chunk_dev_extent_mapping(fs_info); 8347 out: 8348 btrfs_free_path(path); 8349 return ret; 8350 } 8351 8352 /* 8353 * Check whether the given block group or device is pinned by any inode being 8354 * used as a swapfile. 8355 */ 8356 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8357 { 8358 struct btrfs_swapfile_pin *sp; 8359 struct rb_node *node; 8360 8361 spin_lock(&fs_info->swapfile_pins_lock); 8362 node = fs_info->swapfile_pins.rb_node; 8363 while (node) { 8364 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8365 if (ptr < sp->ptr) 8366 node = node->rb_left; 8367 else if (ptr > sp->ptr) 8368 node = node->rb_right; 8369 else 8370 break; 8371 } 8372 spin_unlock(&fs_info->swapfile_pins_lock); 8373 return node != NULL; 8374 } 8375 8376 static int relocating_repair_kthread(void *data) 8377 { 8378 struct btrfs_block_group *cache = data; 8379 struct btrfs_fs_info *fs_info = cache->fs_info; 8380 u64 target; 8381 int ret = 0; 8382 8383 target = cache->start; 8384 btrfs_put_block_group(cache); 8385 8386 sb_start_write(fs_info->sb); 8387 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8388 btrfs_info(fs_info, 8389 "zoned: skip relocating block group %llu to repair: EBUSY", 8390 target); 8391 sb_end_write(fs_info->sb); 8392 return -EBUSY; 8393 } 8394 8395 mutex_lock(&fs_info->reclaim_bgs_lock); 8396 8397 /* Ensure block group still exists */ 8398 cache = btrfs_lookup_block_group(fs_info, target); 8399 if (!cache) 8400 goto out; 8401 8402 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) 8403 goto out; 8404 8405 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8406 if (ret < 0) 8407 goto out; 8408 8409 btrfs_info(fs_info, 8410 "zoned: relocating block group %llu to repair IO failure", 8411 target); 8412 ret = btrfs_relocate_chunk(fs_info, target); 8413 8414 out: 8415 if (cache) 8416 btrfs_put_block_group(cache); 8417 mutex_unlock(&fs_info->reclaim_bgs_lock); 8418 btrfs_exclop_finish(fs_info); 8419 sb_end_write(fs_info->sb); 8420 8421 return ret; 8422 } 8423 8424 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8425 { 8426 struct btrfs_block_group *cache; 8427 8428 if (!btrfs_is_zoned(fs_info)) 8429 return false; 8430 8431 /* Do not attempt to repair in degraded state */ 8432 if (btrfs_test_opt(fs_info, DEGRADED)) 8433 return true; 8434 8435 cache = btrfs_lookup_block_group(fs_info, logical); 8436 if (!cache) 8437 return true; 8438 8439 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { 8440 btrfs_put_block_group(cache); 8441 return true; 8442 } 8443 8444 kthread_run(relocating_repair_kthread, cache, 8445 "btrfs-relocating-repair"); 8446 8447 return true; 8448 } 8449 8450 int __init btrfs_bioset_init(void) 8451 { 8452 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, 8453 offsetof(struct btrfs_bio, bio), 8454 BIOSET_NEED_BVECS)) 8455 return -ENOMEM; 8456 return 0; 8457 } 8458 8459 void __cold btrfs_bioset_exit(void) 8460 { 8461 bioset_exit(&btrfs_bioset); 8462 } 8463