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 38 static struct bio_set btrfs_bioset; 39 40 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ 41 BTRFS_BLOCK_GROUP_RAID10 | \ 42 BTRFS_BLOCK_GROUP_RAID56_MASK) 43 44 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { 45 [BTRFS_RAID_RAID10] = { 46 .sub_stripes = 2, 47 .dev_stripes = 1, 48 .devs_max = 0, /* 0 == as many as possible */ 49 .devs_min = 2, 50 .tolerated_failures = 1, 51 .devs_increment = 2, 52 .ncopies = 2, 53 .nparity = 0, 54 .raid_name = "raid10", 55 .bg_flag = BTRFS_BLOCK_GROUP_RAID10, 56 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, 57 }, 58 [BTRFS_RAID_RAID1] = { 59 .sub_stripes = 1, 60 .dev_stripes = 1, 61 .devs_max = 2, 62 .devs_min = 2, 63 .tolerated_failures = 1, 64 .devs_increment = 2, 65 .ncopies = 2, 66 .nparity = 0, 67 .raid_name = "raid1", 68 .bg_flag = BTRFS_BLOCK_GROUP_RAID1, 69 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, 70 }, 71 [BTRFS_RAID_RAID1C3] = { 72 .sub_stripes = 1, 73 .dev_stripes = 1, 74 .devs_max = 3, 75 .devs_min = 3, 76 .tolerated_failures = 2, 77 .devs_increment = 3, 78 .ncopies = 3, 79 .nparity = 0, 80 .raid_name = "raid1c3", 81 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, 82 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, 83 }, 84 [BTRFS_RAID_RAID1C4] = { 85 .sub_stripes = 1, 86 .dev_stripes = 1, 87 .devs_max = 4, 88 .devs_min = 4, 89 .tolerated_failures = 3, 90 .devs_increment = 4, 91 .ncopies = 4, 92 .nparity = 0, 93 .raid_name = "raid1c4", 94 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, 95 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, 96 }, 97 [BTRFS_RAID_DUP] = { 98 .sub_stripes = 1, 99 .dev_stripes = 2, 100 .devs_max = 1, 101 .devs_min = 1, 102 .tolerated_failures = 0, 103 .devs_increment = 1, 104 .ncopies = 2, 105 .nparity = 0, 106 .raid_name = "dup", 107 .bg_flag = BTRFS_BLOCK_GROUP_DUP, 108 .mindev_error = 0, 109 }, 110 [BTRFS_RAID_RAID0] = { 111 .sub_stripes = 1, 112 .dev_stripes = 1, 113 .devs_max = 0, 114 .devs_min = 1, 115 .tolerated_failures = 0, 116 .devs_increment = 1, 117 .ncopies = 1, 118 .nparity = 0, 119 .raid_name = "raid0", 120 .bg_flag = BTRFS_BLOCK_GROUP_RAID0, 121 .mindev_error = 0, 122 }, 123 [BTRFS_RAID_SINGLE] = { 124 .sub_stripes = 1, 125 .dev_stripes = 1, 126 .devs_max = 1, 127 .devs_min = 1, 128 .tolerated_failures = 0, 129 .devs_increment = 1, 130 .ncopies = 1, 131 .nparity = 0, 132 .raid_name = "single", 133 .bg_flag = 0, 134 .mindev_error = 0, 135 }, 136 [BTRFS_RAID_RAID5] = { 137 .sub_stripes = 1, 138 .dev_stripes = 1, 139 .devs_max = 0, 140 .devs_min = 2, 141 .tolerated_failures = 1, 142 .devs_increment = 1, 143 .ncopies = 1, 144 .nparity = 1, 145 .raid_name = "raid5", 146 .bg_flag = BTRFS_BLOCK_GROUP_RAID5, 147 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, 148 }, 149 [BTRFS_RAID_RAID6] = { 150 .sub_stripes = 1, 151 .dev_stripes = 1, 152 .devs_max = 0, 153 .devs_min = 3, 154 .tolerated_failures = 2, 155 .devs_increment = 1, 156 .ncopies = 1, 157 .nparity = 2, 158 .raid_name = "raid6", 159 .bg_flag = BTRFS_BLOCK_GROUP_RAID6, 160 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, 161 }, 162 }; 163 164 /* 165 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 166 * can be used as index to access btrfs_raid_array[]. 167 */ 168 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) 169 { 170 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK); 171 172 if (!profile) 173 return BTRFS_RAID_SINGLE; 174 175 return BTRFS_BG_FLAG_TO_INDEX(profile); 176 } 177 178 const char *btrfs_bg_type_to_raid_name(u64 flags) 179 { 180 const int index = btrfs_bg_flags_to_raid_index(flags); 181 182 if (index >= BTRFS_NR_RAID_TYPES) 183 return NULL; 184 185 return btrfs_raid_array[index].raid_name; 186 } 187 188 int btrfs_nr_parity_stripes(u64 type) 189 { 190 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type); 191 192 return btrfs_raid_array[index].nparity; 193 } 194 195 /* 196 * Fill @buf with textual description of @bg_flags, no more than @size_buf 197 * bytes including terminating null byte. 198 */ 199 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) 200 { 201 int i; 202 int ret; 203 char *bp = buf; 204 u64 flags = bg_flags; 205 u32 size_bp = size_buf; 206 207 if (!flags) { 208 strcpy(bp, "NONE"); 209 return; 210 } 211 212 #define DESCRIBE_FLAG(flag, desc) \ 213 do { \ 214 if (flags & (flag)) { \ 215 ret = snprintf(bp, size_bp, "%s|", (desc)); \ 216 if (ret < 0 || ret >= size_bp) \ 217 goto out_overflow; \ 218 size_bp -= ret; \ 219 bp += ret; \ 220 flags &= ~(flag); \ 221 } \ 222 } while (0) 223 224 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); 225 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); 226 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); 227 228 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); 229 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) 230 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, 231 btrfs_raid_array[i].raid_name); 232 #undef DESCRIBE_FLAG 233 234 if (flags) { 235 ret = snprintf(bp, size_bp, "0x%llx|", flags); 236 size_bp -= ret; 237 } 238 239 if (size_bp < size_buf) 240 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ 241 242 /* 243 * The text is trimmed, it's up to the caller to provide sufficiently 244 * large buffer 245 */ 246 out_overflow:; 247 } 248 249 static int init_first_rw_device(struct btrfs_trans_handle *trans); 250 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); 251 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); 252 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 253 enum btrfs_map_op op, u64 logical, u64 *length, 254 struct btrfs_io_context **bioc_ret, 255 struct btrfs_io_stripe *smap, 256 int *mirror_num_ret, int need_raid_map); 257 258 /* 259 * Device locking 260 * ============== 261 * 262 * There are several mutexes that protect manipulation of devices and low-level 263 * structures like chunks but not block groups, extents or files 264 * 265 * uuid_mutex (global lock) 266 * ------------------------ 267 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from 268 * the SCAN_DEV ioctl registration or from mount either implicitly (the first 269 * device) or requested by the device= mount option 270 * 271 * the mutex can be very coarse and can cover long-running operations 272 * 273 * protects: updates to fs_devices counters like missing devices, rw devices, 274 * seeding, structure cloning, opening/closing devices at mount/umount time 275 * 276 * global::fs_devs - add, remove, updates to the global list 277 * 278 * does not protect: manipulation of the fs_devices::devices list in general 279 * but in mount context it could be used to exclude list modifications by eg. 280 * scan ioctl 281 * 282 * btrfs_device::name - renames (write side), read is RCU 283 * 284 * fs_devices::device_list_mutex (per-fs, with RCU) 285 * ------------------------------------------------ 286 * protects updates to fs_devices::devices, ie. adding and deleting 287 * 288 * simple list traversal with read-only actions can be done with RCU protection 289 * 290 * may be used to exclude some operations from running concurrently without any 291 * modifications to the list (see write_all_supers) 292 * 293 * Is not required at mount and close times, because our device list is 294 * protected by the uuid_mutex at that point. 295 * 296 * balance_mutex 297 * ------------- 298 * protects balance structures (status, state) and context accessed from 299 * several places (internally, ioctl) 300 * 301 * chunk_mutex 302 * ----------- 303 * protects chunks, adding or removing during allocation, trim or when a new 304 * device is added/removed. Additionally it also protects post_commit_list of 305 * individual devices, since they can be added to the transaction's 306 * post_commit_list only with chunk_mutex held. 307 * 308 * cleaner_mutex 309 * ------------- 310 * a big lock that is held by the cleaner thread and prevents running subvolume 311 * cleaning together with relocation or delayed iputs 312 * 313 * 314 * Lock nesting 315 * ============ 316 * 317 * uuid_mutex 318 * device_list_mutex 319 * chunk_mutex 320 * balance_mutex 321 * 322 * 323 * Exclusive operations 324 * ==================== 325 * 326 * Maintains the exclusivity of the following operations that apply to the 327 * whole filesystem and cannot run in parallel. 328 * 329 * - Balance (*) 330 * - Device add 331 * - Device remove 332 * - Device replace (*) 333 * - Resize 334 * 335 * The device operations (as above) can be in one of the following states: 336 * 337 * - Running state 338 * - Paused state 339 * - Completed state 340 * 341 * Only device operations marked with (*) can go into the Paused state for the 342 * following reasons: 343 * 344 * - ioctl (only Balance can be Paused through ioctl) 345 * - filesystem remounted as read-only 346 * - filesystem unmounted and mounted as read-only 347 * - system power-cycle and filesystem mounted as read-only 348 * - filesystem or device errors leading to forced read-only 349 * 350 * The status of exclusive operation is set and cleared atomically. 351 * During the course of Paused state, fs_info::exclusive_operation remains set. 352 * A device operation in Paused or Running state can be canceled or resumed 353 * either by ioctl (Balance only) or when remounted as read-write. 354 * The exclusive status is cleared when the device operation is canceled or 355 * completed. 356 */ 357 358 DEFINE_MUTEX(uuid_mutex); 359 static LIST_HEAD(fs_uuids); 360 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) 361 { 362 return &fs_uuids; 363 } 364 365 /* 366 * alloc_fs_devices - allocate struct btrfs_fs_devices 367 * @fsid: if not NULL, copy the UUID to fs_devices::fsid 368 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid 369 * 370 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). 371 * The returned struct is not linked onto any lists and can be destroyed with 372 * kfree() right away. 373 */ 374 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, 375 const u8 *metadata_fsid) 376 { 377 struct btrfs_fs_devices *fs_devs; 378 379 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 380 if (!fs_devs) 381 return ERR_PTR(-ENOMEM); 382 383 mutex_init(&fs_devs->device_list_mutex); 384 385 INIT_LIST_HEAD(&fs_devs->devices); 386 INIT_LIST_HEAD(&fs_devs->alloc_list); 387 INIT_LIST_HEAD(&fs_devs->fs_list); 388 INIT_LIST_HEAD(&fs_devs->seed_list); 389 if (fsid) 390 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 391 392 if (metadata_fsid) 393 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); 394 else if (fsid) 395 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); 396 397 return fs_devs; 398 } 399 400 void btrfs_free_device(struct btrfs_device *device) 401 { 402 WARN_ON(!list_empty(&device->post_commit_list)); 403 rcu_string_free(device->name); 404 extent_io_tree_release(&device->alloc_state); 405 btrfs_destroy_dev_zone_info(device); 406 kfree(device); 407 } 408 409 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 410 { 411 struct btrfs_device *device; 412 WARN_ON(fs_devices->opened); 413 while (!list_empty(&fs_devices->devices)) { 414 device = list_entry(fs_devices->devices.next, 415 struct btrfs_device, dev_list); 416 list_del(&device->dev_list); 417 btrfs_free_device(device); 418 } 419 kfree(fs_devices); 420 } 421 422 void __exit btrfs_cleanup_fs_uuids(void) 423 { 424 struct btrfs_fs_devices *fs_devices; 425 426 while (!list_empty(&fs_uuids)) { 427 fs_devices = list_entry(fs_uuids.next, 428 struct btrfs_fs_devices, fs_list); 429 list_del(&fs_devices->fs_list); 430 free_fs_devices(fs_devices); 431 } 432 } 433 434 static noinline struct btrfs_fs_devices *find_fsid( 435 const u8 *fsid, const u8 *metadata_fsid) 436 { 437 struct btrfs_fs_devices *fs_devices; 438 439 ASSERT(fsid); 440 441 /* Handle non-split brain cases */ 442 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 443 if (metadata_fsid) { 444 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 445 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 446 BTRFS_FSID_SIZE) == 0) 447 return fs_devices; 448 } else { 449 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 450 return fs_devices; 451 } 452 } 453 return NULL; 454 } 455 456 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 457 struct btrfs_super_block *disk_super) 458 { 459 460 struct btrfs_fs_devices *fs_devices; 461 462 /* 463 * Handle scanned device having completed its fsid change but 464 * belonging to a fs_devices that was created by first scanning 465 * a device which didn't have its fsid/metadata_uuid changed 466 * at all and the CHANGING_FSID_V2 flag set. 467 */ 468 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 469 if (fs_devices->fsid_change && 470 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 471 BTRFS_FSID_SIZE) == 0 && 472 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 473 BTRFS_FSID_SIZE) == 0) { 474 return fs_devices; 475 } 476 } 477 /* 478 * Handle scanned device having completed its fsid change but 479 * belonging to a fs_devices that was created by a device that 480 * has an outdated pair of fsid/metadata_uuid and 481 * CHANGING_FSID_V2 flag set. 482 */ 483 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 484 if (fs_devices->fsid_change && 485 memcmp(fs_devices->metadata_uuid, 486 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 487 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 488 BTRFS_FSID_SIZE) == 0) { 489 return fs_devices; 490 } 491 } 492 493 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 494 } 495 496 497 static int 498 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 499 int flush, struct block_device **bdev, 500 struct btrfs_super_block **disk_super) 501 { 502 int ret; 503 504 *bdev = blkdev_get_by_path(device_path, flags, holder); 505 506 if (IS_ERR(*bdev)) { 507 ret = PTR_ERR(*bdev); 508 goto error; 509 } 510 511 if (flush) 512 sync_blockdev(*bdev); 513 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 514 if (ret) { 515 blkdev_put(*bdev, flags); 516 goto error; 517 } 518 invalidate_bdev(*bdev); 519 *disk_super = btrfs_read_dev_super(*bdev); 520 if (IS_ERR(*disk_super)) { 521 ret = PTR_ERR(*disk_super); 522 blkdev_put(*bdev, flags); 523 goto error; 524 } 525 526 return 0; 527 528 error: 529 *bdev = NULL; 530 return ret; 531 } 532 533 /** 534 * Search and remove all stale devices (which are not mounted). 535 * When both inputs are NULL, it will search and release all stale devices. 536 * 537 * @devt: Optional. When provided will it release all unmounted devices 538 * matching this devt only. 539 * @skip_device: Optional. Will skip this device when searching for the stale 540 * devices. 541 * 542 * Return: 0 for success or if @devt is 0. 543 * -EBUSY if @devt is a mounted device. 544 * -ENOENT if @devt does not match any device in the list. 545 */ 546 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) 547 { 548 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 549 struct btrfs_device *device, *tmp_device; 550 int ret = 0; 551 552 lockdep_assert_held(&uuid_mutex); 553 554 if (devt) 555 ret = -ENOENT; 556 557 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 558 559 mutex_lock(&fs_devices->device_list_mutex); 560 list_for_each_entry_safe(device, tmp_device, 561 &fs_devices->devices, dev_list) { 562 if (skip_device && skip_device == device) 563 continue; 564 if (devt && devt != device->devt) 565 continue; 566 if (fs_devices->opened) { 567 /* for an already deleted device return 0 */ 568 if (devt && ret != 0) 569 ret = -EBUSY; 570 break; 571 } 572 573 /* delete the stale device */ 574 fs_devices->num_devices--; 575 list_del(&device->dev_list); 576 btrfs_free_device(device); 577 578 ret = 0; 579 } 580 mutex_unlock(&fs_devices->device_list_mutex); 581 582 if (fs_devices->num_devices == 0) { 583 btrfs_sysfs_remove_fsid(fs_devices); 584 list_del(&fs_devices->fs_list); 585 free_fs_devices(fs_devices); 586 } 587 } 588 589 return ret; 590 } 591 592 /* 593 * This is only used on mount, and we are protected from competing things 594 * messing with our fs_devices by the uuid_mutex, thus we do not need the 595 * fs_devices->device_list_mutex here. 596 */ 597 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 598 struct btrfs_device *device, fmode_t flags, 599 void *holder) 600 { 601 struct block_device *bdev; 602 struct btrfs_super_block *disk_super; 603 u64 devid; 604 int ret; 605 606 if (device->bdev) 607 return -EINVAL; 608 if (!device->name) 609 return -EINVAL; 610 611 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 612 &bdev, &disk_super); 613 if (ret) 614 return ret; 615 616 devid = btrfs_stack_device_id(&disk_super->dev_item); 617 if (devid != device->devid) 618 goto error_free_page; 619 620 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 621 goto error_free_page; 622 623 device->generation = btrfs_super_generation(disk_super); 624 625 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 626 if (btrfs_super_incompat_flags(disk_super) & 627 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 628 pr_err( 629 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 630 goto error_free_page; 631 } 632 633 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 634 fs_devices->seeding = true; 635 } else { 636 if (bdev_read_only(bdev)) 637 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 638 else 639 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 640 } 641 642 if (!bdev_nonrot(bdev)) 643 fs_devices->rotating = true; 644 645 if (bdev_max_discard_sectors(bdev)) 646 fs_devices->discardable = true; 647 648 device->bdev = bdev; 649 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 650 device->mode = flags; 651 652 fs_devices->open_devices++; 653 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 654 device->devid != BTRFS_DEV_REPLACE_DEVID) { 655 fs_devices->rw_devices++; 656 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 657 } 658 btrfs_release_disk_super(disk_super); 659 660 return 0; 661 662 error_free_page: 663 btrfs_release_disk_super(disk_super); 664 blkdev_put(bdev, flags); 665 666 return -EINVAL; 667 } 668 669 /* 670 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 671 * being created with a disk that has already completed its fsid change. Such 672 * disk can belong to an fs which has its FSID changed or to one which doesn't. 673 * Handle both cases here. 674 */ 675 static struct btrfs_fs_devices *find_fsid_inprogress( 676 struct btrfs_super_block *disk_super) 677 { 678 struct btrfs_fs_devices *fs_devices; 679 680 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 681 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 682 BTRFS_FSID_SIZE) != 0 && 683 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 684 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 685 return fs_devices; 686 } 687 } 688 689 return find_fsid(disk_super->fsid, NULL); 690 } 691 692 693 static struct btrfs_fs_devices *find_fsid_changed( 694 struct btrfs_super_block *disk_super) 695 { 696 struct btrfs_fs_devices *fs_devices; 697 698 /* 699 * Handles the case where scanned device is part of an fs that had 700 * multiple successful changes of FSID but currently device didn't 701 * observe it. Meaning our fsid will be different than theirs. We need 702 * to handle two subcases : 703 * 1 - The fs still continues to have different METADATA/FSID uuids. 704 * 2 - The fs is switched back to its original FSID (METADATA/FSID 705 * are equal). 706 */ 707 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 708 /* Changed UUIDs */ 709 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 710 BTRFS_FSID_SIZE) != 0 && 711 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 712 BTRFS_FSID_SIZE) == 0 && 713 memcmp(fs_devices->fsid, disk_super->fsid, 714 BTRFS_FSID_SIZE) != 0) 715 return fs_devices; 716 717 /* Unchanged UUIDs */ 718 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 719 BTRFS_FSID_SIZE) == 0 && 720 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 721 BTRFS_FSID_SIZE) == 0) 722 return fs_devices; 723 } 724 725 return NULL; 726 } 727 728 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 729 struct btrfs_super_block *disk_super) 730 { 731 struct btrfs_fs_devices *fs_devices; 732 733 /* 734 * Handle the case where the scanned device is part of an fs whose last 735 * metadata UUID change reverted it to the original FSID. At the same 736 * time * fs_devices was first created by another constitutent device 737 * which didn't fully observe the operation. This results in an 738 * btrfs_fs_devices created with metadata/fsid different AND 739 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 740 * fs_devices equal to the FSID of the disk. 741 */ 742 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 743 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 744 BTRFS_FSID_SIZE) != 0 && 745 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 746 BTRFS_FSID_SIZE) == 0 && 747 fs_devices->fsid_change) 748 return fs_devices; 749 } 750 751 return NULL; 752 } 753 /* 754 * Add new device to list of registered devices 755 * 756 * Returns: 757 * device pointer which was just added or updated when successful 758 * error pointer when failed 759 */ 760 static noinline struct btrfs_device *device_list_add(const char *path, 761 struct btrfs_super_block *disk_super, 762 bool *new_device_added) 763 { 764 struct btrfs_device *device; 765 struct btrfs_fs_devices *fs_devices = NULL; 766 struct rcu_string *name; 767 u64 found_transid = btrfs_super_generation(disk_super); 768 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 769 dev_t path_devt; 770 int error; 771 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 772 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 773 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 774 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 775 776 error = lookup_bdev(path, &path_devt); 777 if (error) 778 return ERR_PTR(error); 779 780 if (fsid_change_in_progress) { 781 if (!has_metadata_uuid) 782 fs_devices = find_fsid_inprogress(disk_super); 783 else 784 fs_devices = find_fsid_changed(disk_super); 785 } else if (has_metadata_uuid) { 786 fs_devices = find_fsid_with_metadata_uuid(disk_super); 787 } else { 788 fs_devices = find_fsid_reverted_metadata(disk_super); 789 if (!fs_devices) 790 fs_devices = find_fsid(disk_super->fsid, NULL); 791 } 792 793 794 if (!fs_devices) { 795 if (has_metadata_uuid) 796 fs_devices = alloc_fs_devices(disk_super->fsid, 797 disk_super->metadata_uuid); 798 else 799 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 800 801 if (IS_ERR(fs_devices)) 802 return ERR_CAST(fs_devices); 803 804 fs_devices->fsid_change = fsid_change_in_progress; 805 806 mutex_lock(&fs_devices->device_list_mutex); 807 list_add(&fs_devices->fs_list, &fs_uuids); 808 809 device = NULL; 810 } else { 811 struct btrfs_dev_lookup_args args = { 812 .devid = devid, 813 .uuid = disk_super->dev_item.uuid, 814 }; 815 816 mutex_lock(&fs_devices->device_list_mutex); 817 device = btrfs_find_device(fs_devices, &args); 818 819 /* 820 * If this disk has been pulled into an fs devices created by 821 * a device which had the CHANGING_FSID_V2 flag then replace the 822 * metadata_uuid/fsid values of the fs_devices. 823 */ 824 if (fs_devices->fsid_change && 825 found_transid > fs_devices->latest_generation) { 826 memcpy(fs_devices->fsid, disk_super->fsid, 827 BTRFS_FSID_SIZE); 828 829 if (has_metadata_uuid) 830 memcpy(fs_devices->metadata_uuid, 831 disk_super->metadata_uuid, 832 BTRFS_FSID_SIZE); 833 else 834 memcpy(fs_devices->metadata_uuid, 835 disk_super->fsid, BTRFS_FSID_SIZE); 836 837 fs_devices->fsid_change = false; 838 } 839 } 840 841 if (!device) { 842 if (fs_devices->opened) { 843 mutex_unlock(&fs_devices->device_list_mutex); 844 return ERR_PTR(-EBUSY); 845 } 846 847 device = btrfs_alloc_device(NULL, &devid, 848 disk_super->dev_item.uuid); 849 if (IS_ERR(device)) { 850 mutex_unlock(&fs_devices->device_list_mutex); 851 /* we can safely leave the fs_devices entry around */ 852 return device; 853 } 854 855 name = rcu_string_strdup(path, GFP_NOFS); 856 if (!name) { 857 btrfs_free_device(device); 858 mutex_unlock(&fs_devices->device_list_mutex); 859 return ERR_PTR(-ENOMEM); 860 } 861 rcu_assign_pointer(device->name, name); 862 device->devt = path_devt; 863 864 list_add_rcu(&device->dev_list, &fs_devices->devices); 865 fs_devices->num_devices++; 866 867 device->fs_devices = fs_devices; 868 *new_device_added = true; 869 870 if (disk_super->label[0]) 871 pr_info( 872 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 873 disk_super->label, devid, found_transid, path, 874 current->comm, task_pid_nr(current)); 875 else 876 pr_info( 877 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 878 disk_super->fsid, devid, found_transid, path, 879 current->comm, task_pid_nr(current)); 880 881 } else if (!device->name || strcmp(device->name->str, path)) { 882 /* 883 * When FS is already mounted. 884 * 1. If you are here and if the device->name is NULL that 885 * means this device was missing at time of FS mount. 886 * 2. If you are here and if the device->name is different 887 * from 'path' that means either 888 * a. The same device disappeared and reappeared with 889 * different name. or 890 * b. The missing-disk-which-was-replaced, has 891 * reappeared now. 892 * 893 * We must allow 1 and 2a above. But 2b would be a spurious 894 * and unintentional. 895 * 896 * Further in case of 1 and 2a above, the disk at 'path' 897 * would have missed some transaction when it was away and 898 * in case of 2a the stale bdev has to be updated as well. 899 * 2b must not be allowed at all time. 900 */ 901 902 /* 903 * For now, we do allow update to btrfs_fs_device through the 904 * btrfs dev scan cli after FS has been mounted. We're still 905 * tracking a problem where systems fail mount by subvolume id 906 * when we reject replacement on a mounted FS. 907 */ 908 if (!fs_devices->opened && found_transid < device->generation) { 909 /* 910 * That is if the FS is _not_ mounted and if you 911 * are here, that means there is more than one 912 * disk with same uuid and devid.We keep the one 913 * with larger generation number or the last-in if 914 * generation are equal. 915 */ 916 mutex_unlock(&fs_devices->device_list_mutex); 917 return ERR_PTR(-EEXIST); 918 } 919 920 /* 921 * We are going to replace the device path for a given devid, 922 * make sure it's the same device if the device is mounted 923 * 924 * NOTE: the device->fs_info may not be reliable here so pass 925 * in a NULL to message helpers instead. This avoids a possible 926 * use-after-free when the fs_info and fs_info->sb are already 927 * torn down. 928 */ 929 if (device->bdev) { 930 if (device->devt != path_devt) { 931 mutex_unlock(&fs_devices->device_list_mutex); 932 btrfs_warn_in_rcu(NULL, 933 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 934 path, devid, found_transid, 935 current->comm, 936 task_pid_nr(current)); 937 return ERR_PTR(-EEXIST); 938 } 939 btrfs_info_in_rcu(NULL, 940 "devid %llu device path %s changed to %s scanned by %s (%d)", 941 devid, rcu_str_deref(device->name), 942 path, current->comm, 943 task_pid_nr(current)); 944 } 945 946 name = rcu_string_strdup(path, GFP_NOFS); 947 if (!name) { 948 mutex_unlock(&fs_devices->device_list_mutex); 949 return ERR_PTR(-ENOMEM); 950 } 951 rcu_string_free(device->name); 952 rcu_assign_pointer(device->name, name); 953 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 954 fs_devices->missing_devices--; 955 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 956 } 957 device->devt = path_devt; 958 } 959 960 /* 961 * Unmount does not free the btrfs_device struct but would zero 962 * generation along with most of the other members. So just update 963 * it back. We need it to pick the disk with largest generation 964 * (as above). 965 */ 966 if (!fs_devices->opened) { 967 device->generation = found_transid; 968 fs_devices->latest_generation = max_t(u64, found_transid, 969 fs_devices->latest_generation); 970 } 971 972 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 973 974 mutex_unlock(&fs_devices->device_list_mutex); 975 return device; 976 } 977 978 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 979 { 980 struct btrfs_fs_devices *fs_devices; 981 struct btrfs_device *device; 982 struct btrfs_device *orig_dev; 983 int ret = 0; 984 985 lockdep_assert_held(&uuid_mutex); 986 987 fs_devices = alloc_fs_devices(orig->fsid, NULL); 988 if (IS_ERR(fs_devices)) 989 return fs_devices; 990 991 fs_devices->total_devices = orig->total_devices; 992 993 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 994 struct rcu_string *name; 995 996 device = btrfs_alloc_device(NULL, &orig_dev->devid, 997 orig_dev->uuid); 998 if (IS_ERR(device)) { 999 ret = PTR_ERR(device); 1000 goto error; 1001 } 1002 1003 /* 1004 * This is ok to do without rcu read locked because we hold the 1005 * uuid mutex so nothing we touch in here is going to disappear. 1006 */ 1007 if (orig_dev->name) { 1008 name = rcu_string_strdup(orig_dev->name->str, 1009 GFP_KERNEL); 1010 if (!name) { 1011 btrfs_free_device(device); 1012 ret = -ENOMEM; 1013 goto error; 1014 } 1015 rcu_assign_pointer(device->name, name); 1016 } 1017 1018 if (orig_dev->zone_info) { 1019 struct btrfs_zoned_device_info *zone_info; 1020 1021 zone_info = btrfs_clone_dev_zone_info(orig_dev); 1022 if (!zone_info) { 1023 btrfs_free_device(device); 1024 ret = -ENOMEM; 1025 goto error; 1026 } 1027 device->zone_info = zone_info; 1028 } 1029 1030 list_add(&device->dev_list, &fs_devices->devices); 1031 device->fs_devices = fs_devices; 1032 fs_devices->num_devices++; 1033 } 1034 return fs_devices; 1035 error: 1036 free_fs_devices(fs_devices); 1037 return ERR_PTR(ret); 1038 } 1039 1040 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1041 struct btrfs_device **latest_dev) 1042 { 1043 struct btrfs_device *device, *next; 1044 1045 /* This is the initialized path, it is safe to release the devices. */ 1046 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1047 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1048 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1049 &device->dev_state) && 1050 !test_bit(BTRFS_DEV_STATE_MISSING, 1051 &device->dev_state) && 1052 (!*latest_dev || 1053 device->generation > (*latest_dev)->generation)) { 1054 *latest_dev = device; 1055 } 1056 continue; 1057 } 1058 1059 /* 1060 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1061 * in btrfs_init_dev_replace() so just continue. 1062 */ 1063 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1064 continue; 1065 1066 if (device->bdev) { 1067 blkdev_put(device->bdev, device->mode); 1068 device->bdev = NULL; 1069 fs_devices->open_devices--; 1070 } 1071 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1072 list_del_init(&device->dev_alloc_list); 1073 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1074 fs_devices->rw_devices--; 1075 } 1076 list_del_init(&device->dev_list); 1077 fs_devices->num_devices--; 1078 btrfs_free_device(device); 1079 } 1080 1081 } 1082 1083 /* 1084 * After we have read the system tree and know devids belonging to this 1085 * filesystem, remove the device which does not belong there. 1086 */ 1087 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1088 { 1089 struct btrfs_device *latest_dev = NULL; 1090 struct btrfs_fs_devices *seed_dev; 1091 1092 mutex_lock(&uuid_mutex); 1093 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1094 1095 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1096 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1097 1098 fs_devices->latest_dev = latest_dev; 1099 1100 mutex_unlock(&uuid_mutex); 1101 } 1102 1103 static void btrfs_close_bdev(struct btrfs_device *device) 1104 { 1105 if (!device->bdev) 1106 return; 1107 1108 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1109 sync_blockdev(device->bdev); 1110 invalidate_bdev(device->bdev); 1111 } 1112 1113 blkdev_put(device->bdev, device->mode); 1114 } 1115 1116 static void btrfs_close_one_device(struct btrfs_device *device) 1117 { 1118 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1119 1120 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1121 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1122 list_del_init(&device->dev_alloc_list); 1123 fs_devices->rw_devices--; 1124 } 1125 1126 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1127 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1128 1129 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 1130 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 1131 fs_devices->missing_devices--; 1132 } 1133 1134 btrfs_close_bdev(device); 1135 if (device->bdev) { 1136 fs_devices->open_devices--; 1137 device->bdev = NULL; 1138 } 1139 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1140 btrfs_destroy_dev_zone_info(device); 1141 1142 device->fs_info = NULL; 1143 atomic_set(&device->dev_stats_ccnt, 0); 1144 extent_io_tree_release(&device->alloc_state); 1145 1146 /* 1147 * Reset the flush error record. We might have a transient flush error 1148 * in this mount, and if so we aborted the current transaction and set 1149 * the fs to an error state, guaranteeing no super blocks can be further 1150 * committed. However that error might be transient and if we unmount the 1151 * filesystem and mount it again, we should allow the mount to succeed 1152 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1153 * filesystem again we still get flush errors, then we will again abort 1154 * any transaction and set the error state, guaranteeing no commits of 1155 * unsafe super blocks. 1156 */ 1157 device->last_flush_error = 0; 1158 1159 /* Verify the device is back in a pristine state */ 1160 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1161 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1162 ASSERT(list_empty(&device->dev_alloc_list)); 1163 ASSERT(list_empty(&device->post_commit_list)); 1164 } 1165 1166 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1167 { 1168 struct btrfs_device *device, *tmp; 1169 1170 lockdep_assert_held(&uuid_mutex); 1171 1172 if (--fs_devices->opened > 0) 1173 return; 1174 1175 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1176 btrfs_close_one_device(device); 1177 1178 WARN_ON(fs_devices->open_devices); 1179 WARN_ON(fs_devices->rw_devices); 1180 fs_devices->opened = 0; 1181 fs_devices->seeding = false; 1182 fs_devices->fs_info = NULL; 1183 } 1184 1185 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1186 { 1187 LIST_HEAD(list); 1188 struct btrfs_fs_devices *tmp; 1189 1190 mutex_lock(&uuid_mutex); 1191 close_fs_devices(fs_devices); 1192 if (!fs_devices->opened) 1193 list_splice_init(&fs_devices->seed_list, &list); 1194 1195 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1196 close_fs_devices(fs_devices); 1197 list_del(&fs_devices->seed_list); 1198 free_fs_devices(fs_devices); 1199 } 1200 mutex_unlock(&uuid_mutex); 1201 } 1202 1203 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1204 fmode_t flags, void *holder) 1205 { 1206 struct btrfs_device *device; 1207 struct btrfs_device *latest_dev = NULL; 1208 struct btrfs_device *tmp_device; 1209 1210 flags |= FMODE_EXCL; 1211 1212 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1213 dev_list) { 1214 int ret; 1215 1216 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1217 if (ret == 0 && 1218 (!latest_dev || device->generation > latest_dev->generation)) { 1219 latest_dev = device; 1220 } else if (ret == -ENODATA) { 1221 fs_devices->num_devices--; 1222 list_del(&device->dev_list); 1223 btrfs_free_device(device); 1224 } 1225 } 1226 if (fs_devices->open_devices == 0) 1227 return -EINVAL; 1228 1229 fs_devices->opened = 1; 1230 fs_devices->latest_dev = latest_dev; 1231 fs_devices->total_rw_bytes = 0; 1232 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1233 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1234 1235 return 0; 1236 } 1237 1238 static int devid_cmp(void *priv, const struct list_head *a, 1239 const struct list_head *b) 1240 { 1241 const struct btrfs_device *dev1, *dev2; 1242 1243 dev1 = list_entry(a, struct btrfs_device, dev_list); 1244 dev2 = list_entry(b, struct btrfs_device, dev_list); 1245 1246 if (dev1->devid < dev2->devid) 1247 return -1; 1248 else if (dev1->devid > dev2->devid) 1249 return 1; 1250 return 0; 1251 } 1252 1253 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1254 fmode_t flags, void *holder) 1255 { 1256 int ret; 1257 1258 lockdep_assert_held(&uuid_mutex); 1259 /* 1260 * The device_list_mutex cannot be taken here in case opening the 1261 * underlying device takes further locks like open_mutex. 1262 * 1263 * We also don't need the lock here as this is called during mount and 1264 * exclusion is provided by uuid_mutex 1265 */ 1266 1267 if (fs_devices->opened) { 1268 fs_devices->opened++; 1269 ret = 0; 1270 } else { 1271 list_sort(NULL, &fs_devices->devices, devid_cmp); 1272 ret = open_fs_devices(fs_devices, flags, holder); 1273 } 1274 1275 return ret; 1276 } 1277 1278 void btrfs_release_disk_super(struct btrfs_super_block *super) 1279 { 1280 struct page *page = virt_to_page(super); 1281 1282 put_page(page); 1283 } 1284 1285 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1286 u64 bytenr, u64 bytenr_orig) 1287 { 1288 struct btrfs_super_block *disk_super; 1289 struct page *page; 1290 void *p; 1291 pgoff_t index; 1292 1293 /* make sure our super fits in the device */ 1294 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) 1295 return ERR_PTR(-EINVAL); 1296 1297 /* make sure our super fits in the page */ 1298 if (sizeof(*disk_super) > PAGE_SIZE) 1299 return ERR_PTR(-EINVAL); 1300 1301 /* make sure our super doesn't straddle pages on disk */ 1302 index = bytenr >> PAGE_SHIFT; 1303 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1304 return ERR_PTR(-EINVAL); 1305 1306 /* pull in the page with our super */ 1307 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1308 1309 if (IS_ERR(page)) 1310 return ERR_CAST(page); 1311 1312 p = page_address(page); 1313 1314 /* align our pointer to the offset of the super block */ 1315 disk_super = p + offset_in_page(bytenr); 1316 1317 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1318 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1319 btrfs_release_disk_super(p); 1320 return ERR_PTR(-EINVAL); 1321 } 1322 1323 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1324 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1325 1326 return disk_super; 1327 } 1328 1329 int btrfs_forget_devices(dev_t devt) 1330 { 1331 int ret; 1332 1333 mutex_lock(&uuid_mutex); 1334 ret = btrfs_free_stale_devices(devt, NULL); 1335 mutex_unlock(&uuid_mutex); 1336 1337 return ret; 1338 } 1339 1340 /* 1341 * Look for a btrfs signature on a device. This may be called out of the mount path 1342 * and we are not allowed to call set_blocksize during the scan. The superblock 1343 * is read via pagecache 1344 */ 1345 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1346 void *holder) 1347 { 1348 struct btrfs_super_block *disk_super; 1349 bool new_device_added = false; 1350 struct btrfs_device *device = NULL; 1351 struct block_device *bdev; 1352 u64 bytenr, bytenr_orig; 1353 int ret; 1354 1355 lockdep_assert_held(&uuid_mutex); 1356 1357 /* 1358 * we would like to check all the supers, but that would make 1359 * a btrfs mount succeed after a mkfs from a different FS. 1360 * So, we need to add a special mount option to scan for 1361 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1362 */ 1363 flags |= FMODE_EXCL; 1364 1365 bdev = blkdev_get_by_path(path, flags, holder); 1366 if (IS_ERR(bdev)) 1367 return ERR_CAST(bdev); 1368 1369 bytenr_orig = btrfs_sb_offset(0); 1370 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1371 if (ret) { 1372 device = ERR_PTR(ret); 1373 goto error_bdev_put; 1374 } 1375 1376 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1377 if (IS_ERR(disk_super)) { 1378 device = ERR_CAST(disk_super); 1379 goto error_bdev_put; 1380 } 1381 1382 device = device_list_add(path, disk_super, &new_device_added); 1383 if (!IS_ERR(device) && new_device_added) 1384 btrfs_free_stale_devices(device->devt, device); 1385 1386 btrfs_release_disk_super(disk_super); 1387 1388 error_bdev_put: 1389 blkdev_put(bdev, flags); 1390 1391 return device; 1392 } 1393 1394 /* 1395 * Try to find a chunk that intersects [start, start + len] range and when one 1396 * such is found, record the end of it in *start 1397 */ 1398 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1399 u64 len) 1400 { 1401 u64 physical_start, physical_end; 1402 1403 lockdep_assert_held(&device->fs_info->chunk_mutex); 1404 1405 if (!find_first_extent_bit(&device->alloc_state, *start, 1406 &physical_start, &physical_end, 1407 CHUNK_ALLOCATED, NULL)) { 1408 1409 if (in_range(physical_start, *start, len) || 1410 in_range(*start, physical_start, 1411 physical_end - physical_start)) { 1412 *start = physical_end + 1; 1413 return true; 1414 } 1415 } 1416 return false; 1417 } 1418 1419 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1420 { 1421 switch (device->fs_devices->chunk_alloc_policy) { 1422 case BTRFS_CHUNK_ALLOC_REGULAR: 1423 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); 1424 case BTRFS_CHUNK_ALLOC_ZONED: 1425 /* 1426 * We don't care about the starting region like regular 1427 * allocator, because we anyway use/reserve the first two zones 1428 * for superblock logging. 1429 */ 1430 return ALIGN(start, device->zone_info->zone_size); 1431 default: 1432 BUG(); 1433 } 1434 } 1435 1436 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1437 u64 *hole_start, u64 *hole_size, 1438 u64 num_bytes) 1439 { 1440 u64 zone_size = device->zone_info->zone_size; 1441 u64 pos; 1442 int ret; 1443 bool changed = false; 1444 1445 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1446 1447 while (*hole_size > 0) { 1448 pos = btrfs_find_allocatable_zones(device, *hole_start, 1449 *hole_start + *hole_size, 1450 num_bytes); 1451 if (pos != *hole_start) { 1452 *hole_size = *hole_start + *hole_size - pos; 1453 *hole_start = pos; 1454 changed = true; 1455 if (*hole_size < num_bytes) 1456 break; 1457 } 1458 1459 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1460 1461 /* Range is ensured to be empty */ 1462 if (!ret) 1463 return changed; 1464 1465 /* Given hole range was invalid (outside of device) */ 1466 if (ret == -ERANGE) { 1467 *hole_start += *hole_size; 1468 *hole_size = 0; 1469 return true; 1470 } 1471 1472 *hole_start += zone_size; 1473 *hole_size -= zone_size; 1474 changed = true; 1475 } 1476 1477 return changed; 1478 } 1479 1480 /** 1481 * dev_extent_hole_check - check if specified hole is suitable for allocation 1482 * @device: the device which we have the hole 1483 * @hole_start: starting position of the hole 1484 * @hole_size: the size of the hole 1485 * @num_bytes: the size of the free space that we need 1486 * 1487 * This function may modify @hole_start and @hole_size to reflect the suitable 1488 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1489 */ 1490 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1491 u64 *hole_size, u64 num_bytes) 1492 { 1493 bool changed = false; 1494 u64 hole_end = *hole_start + *hole_size; 1495 1496 for (;;) { 1497 /* 1498 * Check before we set max_hole_start, otherwise we could end up 1499 * sending back this offset anyway. 1500 */ 1501 if (contains_pending_extent(device, hole_start, *hole_size)) { 1502 if (hole_end >= *hole_start) 1503 *hole_size = hole_end - *hole_start; 1504 else 1505 *hole_size = 0; 1506 changed = true; 1507 } 1508 1509 switch (device->fs_devices->chunk_alloc_policy) { 1510 case BTRFS_CHUNK_ALLOC_REGULAR: 1511 /* No extra check */ 1512 break; 1513 case BTRFS_CHUNK_ALLOC_ZONED: 1514 if (dev_extent_hole_check_zoned(device, hole_start, 1515 hole_size, num_bytes)) { 1516 changed = true; 1517 /* 1518 * The changed hole can contain pending extent. 1519 * Loop again to check that. 1520 */ 1521 continue; 1522 } 1523 break; 1524 default: 1525 BUG(); 1526 } 1527 1528 break; 1529 } 1530 1531 return changed; 1532 } 1533 1534 /* 1535 * find_free_dev_extent_start - find free space in the specified device 1536 * @device: the device which we search the free space in 1537 * @num_bytes: the size of the free space that we need 1538 * @search_start: the position from which to begin the search 1539 * @start: store the start of the free space. 1540 * @len: the size of the free space. that we find, or the size 1541 * of the max free space if we don't find suitable free space 1542 * 1543 * this uses a pretty simple search, the expectation is that it is 1544 * called very infrequently and that a given device has a small number 1545 * of extents 1546 * 1547 * @start is used to store the start of the free space if we find. But if we 1548 * don't find suitable free space, it will be used to store the start position 1549 * of the max free space. 1550 * 1551 * @len is used to store the size of the free space that we find. 1552 * But if we don't find suitable free space, it is used to store the size of 1553 * the max free space. 1554 * 1555 * NOTE: This function will search *commit* root of device tree, and does extra 1556 * check to ensure dev extents are not double allocated. 1557 * This makes the function safe to allocate dev extents but may not report 1558 * correct usable device space, as device extent freed in current transaction 1559 * is not reported as available. 1560 */ 1561 static int find_free_dev_extent_start(struct btrfs_device *device, 1562 u64 num_bytes, u64 search_start, u64 *start, 1563 u64 *len) 1564 { 1565 struct btrfs_fs_info *fs_info = device->fs_info; 1566 struct btrfs_root *root = fs_info->dev_root; 1567 struct btrfs_key key; 1568 struct btrfs_dev_extent *dev_extent; 1569 struct btrfs_path *path; 1570 u64 hole_size; 1571 u64 max_hole_start; 1572 u64 max_hole_size; 1573 u64 extent_end; 1574 u64 search_end = device->total_bytes; 1575 int ret; 1576 int slot; 1577 struct extent_buffer *l; 1578 1579 search_start = dev_extent_search_start(device, search_start); 1580 1581 WARN_ON(device->zone_info && 1582 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1583 1584 path = btrfs_alloc_path(); 1585 if (!path) 1586 return -ENOMEM; 1587 1588 max_hole_start = search_start; 1589 max_hole_size = 0; 1590 1591 again: 1592 if (search_start >= search_end || 1593 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1594 ret = -ENOSPC; 1595 goto out; 1596 } 1597 1598 path->reada = READA_FORWARD; 1599 path->search_commit_root = 1; 1600 path->skip_locking = 1; 1601 1602 key.objectid = device->devid; 1603 key.offset = search_start; 1604 key.type = BTRFS_DEV_EXTENT_KEY; 1605 1606 ret = btrfs_search_backwards(root, &key, path); 1607 if (ret < 0) 1608 goto out; 1609 1610 while (1) { 1611 l = path->nodes[0]; 1612 slot = path->slots[0]; 1613 if (slot >= btrfs_header_nritems(l)) { 1614 ret = btrfs_next_leaf(root, path); 1615 if (ret == 0) 1616 continue; 1617 if (ret < 0) 1618 goto out; 1619 1620 break; 1621 } 1622 btrfs_item_key_to_cpu(l, &key, slot); 1623 1624 if (key.objectid < device->devid) 1625 goto next; 1626 1627 if (key.objectid > device->devid) 1628 break; 1629 1630 if (key.type != BTRFS_DEV_EXTENT_KEY) 1631 goto next; 1632 1633 if (key.offset > search_start) { 1634 hole_size = key.offset - search_start; 1635 dev_extent_hole_check(device, &search_start, &hole_size, 1636 num_bytes); 1637 1638 if (hole_size > max_hole_size) { 1639 max_hole_start = search_start; 1640 max_hole_size = hole_size; 1641 } 1642 1643 /* 1644 * If this free space is greater than which we need, 1645 * it must be the max free space that we have found 1646 * until now, so max_hole_start must point to the start 1647 * of this free space and the length of this free space 1648 * is stored in max_hole_size. Thus, we return 1649 * max_hole_start and max_hole_size and go back to the 1650 * caller. 1651 */ 1652 if (hole_size >= num_bytes) { 1653 ret = 0; 1654 goto out; 1655 } 1656 } 1657 1658 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1659 extent_end = key.offset + btrfs_dev_extent_length(l, 1660 dev_extent); 1661 if (extent_end > search_start) 1662 search_start = extent_end; 1663 next: 1664 path->slots[0]++; 1665 cond_resched(); 1666 } 1667 1668 /* 1669 * At this point, search_start should be the end of 1670 * allocated dev extents, and when shrinking the device, 1671 * search_end may be smaller than search_start. 1672 */ 1673 if (search_end > search_start) { 1674 hole_size = search_end - search_start; 1675 if (dev_extent_hole_check(device, &search_start, &hole_size, 1676 num_bytes)) { 1677 btrfs_release_path(path); 1678 goto again; 1679 } 1680 1681 if (hole_size > max_hole_size) { 1682 max_hole_start = search_start; 1683 max_hole_size = hole_size; 1684 } 1685 } 1686 1687 /* See above. */ 1688 if (max_hole_size < num_bytes) 1689 ret = -ENOSPC; 1690 else 1691 ret = 0; 1692 1693 out: 1694 btrfs_free_path(path); 1695 *start = max_hole_start; 1696 if (len) 1697 *len = max_hole_size; 1698 return ret; 1699 } 1700 1701 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1702 u64 *start, u64 *len) 1703 { 1704 /* FIXME use last free of some kind */ 1705 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1706 } 1707 1708 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1709 struct btrfs_device *device, 1710 u64 start, u64 *dev_extent_len) 1711 { 1712 struct btrfs_fs_info *fs_info = device->fs_info; 1713 struct btrfs_root *root = fs_info->dev_root; 1714 int ret; 1715 struct btrfs_path *path; 1716 struct btrfs_key key; 1717 struct btrfs_key found_key; 1718 struct extent_buffer *leaf = NULL; 1719 struct btrfs_dev_extent *extent = NULL; 1720 1721 path = btrfs_alloc_path(); 1722 if (!path) 1723 return -ENOMEM; 1724 1725 key.objectid = device->devid; 1726 key.offset = start; 1727 key.type = BTRFS_DEV_EXTENT_KEY; 1728 again: 1729 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1730 if (ret > 0) { 1731 ret = btrfs_previous_item(root, path, key.objectid, 1732 BTRFS_DEV_EXTENT_KEY); 1733 if (ret) 1734 goto out; 1735 leaf = path->nodes[0]; 1736 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1737 extent = btrfs_item_ptr(leaf, path->slots[0], 1738 struct btrfs_dev_extent); 1739 BUG_ON(found_key.offset > start || found_key.offset + 1740 btrfs_dev_extent_length(leaf, extent) < start); 1741 key = found_key; 1742 btrfs_release_path(path); 1743 goto again; 1744 } else if (ret == 0) { 1745 leaf = path->nodes[0]; 1746 extent = btrfs_item_ptr(leaf, path->slots[0], 1747 struct btrfs_dev_extent); 1748 } else { 1749 goto out; 1750 } 1751 1752 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1753 1754 ret = btrfs_del_item(trans, root, path); 1755 if (ret == 0) 1756 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1757 out: 1758 btrfs_free_path(path); 1759 return ret; 1760 } 1761 1762 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1763 { 1764 struct extent_map_tree *em_tree; 1765 struct extent_map *em; 1766 struct rb_node *n; 1767 u64 ret = 0; 1768 1769 em_tree = &fs_info->mapping_tree; 1770 read_lock(&em_tree->lock); 1771 n = rb_last(&em_tree->map.rb_root); 1772 if (n) { 1773 em = rb_entry(n, struct extent_map, rb_node); 1774 ret = em->start + em->len; 1775 } 1776 read_unlock(&em_tree->lock); 1777 1778 return ret; 1779 } 1780 1781 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1782 u64 *devid_ret) 1783 { 1784 int ret; 1785 struct btrfs_key key; 1786 struct btrfs_key found_key; 1787 struct btrfs_path *path; 1788 1789 path = btrfs_alloc_path(); 1790 if (!path) 1791 return -ENOMEM; 1792 1793 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1794 key.type = BTRFS_DEV_ITEM_KEY; 1795 key.offset = (u64)-1; 1796 1797 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1798 if (ret < 0) 1799 goto error; 1800 1801 if (ret == 0) { 1802 /* Corruption */ 1803 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1804 ret = -EUCLEAN; 1805 goto error; 1806 } 1807 1808 ret = btrfs_previous_item(fs_info->chunk_root, path, 1809 BTRFS_DEV_ITEMS_OBJECTID, 1810 BTRFS_DEV_ITEM_KEY); 1811 if (ret) { 1812 *devid_ret = 1; 1813 } else { 1814 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1815 path->slots[0]); 1816 *devid_ret = found_key.offset + 1; 1817 } 1818 ret = 0; 1819 error: 1820 btrfs_free_path(path); 1821 return ret; 1822 } 1823 1824 /* 1825 * the device information is stored in the chunk root 1826 * the btrfs_device struct should be fully filled in 1827 */ 1828 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1829 struct btrfs_device *device) 1830 { 1831 int ret; 1832 struct btrfs_path *path; 1833 struct btrfs_dev_item *dev_item; 1834 struct extent_buffer *leaf; 1835 struct btrfs_key key; 1836 unsigned long ptr; 1837 1838 path = btrfs_alloc_path(); 1839 if (!path) 1840 return -ENOMEM; 1841 1842 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1843 key.type = BTRFS_DEV_ITEM_KEY; 1844 key.offset = device->devid; 1845 1846 btrfs_reserve_chunk_metadata(trans, true); 1847 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1848 &key, sizeof(*dev_item)); 1849 btrfs_trans_release_chunk_metadata(trans); 1850 if (ret) 1851 goto out; 1852 1853 leaf = path->nodes[0]; 1854 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1855 1856 btrfs_set_device_id(leaf, dev_item, device->devid); 1857 btrfs_set_device_generation(leaf, dev_item, 0); 1858 btrfs_set_device_type(leaf, dev_item, device->type); 1859 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1860 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1861 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1862 btrfs_set_device_total_bytes(leaf, dev_item, 1863 btrfs_device_get_disk_total_bytes(device)); 1864 btrfs_set_device_bytes_used(leaf, dev_item, 1865 btrfs_device_get_bytes_used(device)); 1866 btrfs_set_device_group(leaf, dev_item, 0); 1867 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1868 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1869 btrfs_set_device_start_offset(leaf, dev_item, 0); 1870 1871 ptr = btrfs_device_uuid(dev_item); 1872 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1873 ptr = btrfs_device_fsid(dev_item); 1874 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1875 ptr, BTRFS_FSID_SIZE); 1876 btrfs_mark_buffer_dirty(leaf); 1877 1878 ret = 0; 1879 out: 1880 btrfs_free_path(path); 1881 return ret; 1882 } 1883 1884 /* 1885 * Function to update ctime/mtime for a given device path. 1886 * Mainly used for ctime/mtime based probe like libblkid. 1887 * 1888 * We don't care about errors here, this is just to be kind to userspace. 1889 */ 1890 static void update_dev_time(const char *device_path) 1891 { 1892 struct path path; 1893 struct timespec64 now; 1894 int ret; 1895 1896 ret = kern_path(device_path, LOOKUP_FOLLOW, &path); 1897 if (ret) 1898 return; 1899 1900 now = current_time(d_inode(path.dentry)); 1901 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); 1902 path_put(&path); 1903 } 1904 1905 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, 1906 struct btrfs_device *device) 1907 { 1908 struct btrfs_root *root = device->fs_info->chunk_root; 1909 int ret; 1910 struct btrfs_path *path; 1911 struct btrfs_key key; 1912 1913 path = btrfs_alloc_path(); 1914 if (!path) 1915 return -ENOMEM; 1916 1917 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1918 key.type = BTRFS_DEV_ITEM_KEY; 1919 key.offset = device->devid; 1920 1921 btrfs_reserve_chunk_metadata(trans, false); 1922 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1923 btrfs_trans_release_chunk_metadata(trans); 1924 if (ret) { 1925 if (ret > 0) 1926 ret = -ENOENT; 1927 goto out; 1928 } 1929 1930 ret = btrfs_del_item(trans, root, path); 1931 out: 1932 btrfs_free_path(path); 1933 return ret; 1934 } 1935 1936 /* 1937 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1938 * filesystem. It's up to the caller to adjust that number regarding eg. device 1939 * replace. 1940 */ 1941 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1942 u64 num_devices) 1943 { 1944 u64 all_avail; 1945 unsigned seq; 1946 int i; 1947 1948 do { 1949 seq = read_seqbegin(&fs_info->profiles_lock); 1950 1951 all_avail = fs_info->avail_data_alloc_bits | 1952 fs_info->avail_system_alloc_bits | 1953 fs_info->avail_metadata_alloc_bits; 1954 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1955 1956 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1957 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1958 continue; 1959 1960 if (num_devices < btrfs_raid_array[i].devs_min) 1961 return btrfs_raid_array[i].mindev_error; 1962 } 1963 1964 return 0; 1965 } 1966 1967 static struct btrfs_device * btrfs_find_next_active_device( 1968 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1969 { 1970 struct btrfs_device *next_device; 1971 1972 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1973 if (next_device != device && 1974 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1975 && next_device->bdev) 1976 return next_device; 1977 } 1978 1979 return NULL; 1980 } 1981 1982 /* 1983 * Helper function to check if the given device is part of s_bdev / latest_dev 1984 * and replace it with the provided or the next active device, in the context 1985 * where this function called, there should be always be another device (or 1986 * this_dev) which is active. 1987 */ 1988 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 1989 struct btrfs_device *next_device) 1990 { 1991 struct btrfs_fs_info *fs_info = device->fs_info; 1992 1993 if (!next_device) 1994 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 1995 device); 1996 ASSERT(next_device); 1997 1998 if (fs_info->sb->s_bdev && 1999 (fs_info->sb->s_bdev == device->bdev)) 2000 fs_info->sb->s_bdev = next_device->bdev; 2001 2002 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2003 fs_info->fs_devices->latest_dev = next_device; 2004 } 2005 2006 /* 2007 * Return btrfs_fs_devices::num_devices excluding the device that's being 2008 * currently replaced. 2009 */ 2010 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2011 { 2012 u64 num_devices = fs_info->fs_devices->num_devices; 2013 2014 down_read(&fs_info->dev_replace.rwsem); 2015 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2016 ASSERT(num_devices > 1); 2017 num_devices--; 2018 } 2019 up_read(&fs_info->dev_replace.rwsem); 2020 2021 return num_devices; 2022 } 2023 2024 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2025 struct block_device *bdev, 2026 const char *device_path) 2027 { 2028 struct btrfs_super_block *disk_super; 2029 int copy_num; 2030 2031 if (!bdev) 2032 return; 2033 2034 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2035 struct page *page; 2036 int ret; 2037 2038 disk_super = btrfs_read_dev_one_super(bdev, copy_num, false); 2039 if (IS_ERR(disk_super)) 2040 continue; 2041 2042 if (bdev_is_zoned(bdev)) { 2043 btrfs_reset_sb_log_zones(bdev, copy_num); 2044 continue; 2045 } 2046 2047 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 2048 2049 page = virt_to_page(disk_super); 2050 set_page_dirty(page); 2051 lock_page(page); 2052 /* write_on_page() unlocks the page */ 2053 ret = write_one_page(page); 2054 if (ret) 2055 btrfs_warn(fs_info, 2056 "error clearing superblock number %d (%d)", 2057 copy_num, ret); 2058 btrfs_release_disk_super(disk_super); 2059 2060 } 2061 2062 /* Notify udev that device has changed */ 2063 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2064 2065 /* Update ctime/mtime for device path for libblkid */ 2066 update_dev_time(device_path); 2067 } 2068 2069 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2070 struct btrfs_dev_lookup_args *args, 2071 struct block_device **bdev, fmode_t *mode) 2072 { 2073 struct btrfs_trans_handle *trans; 2074 struct btrfs_device *device; 2075 struct btrfs_fs_devices *cur_devices; 2076 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2077 u64 num_devices; 2078 int ret = 0; 2079 2080 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2081 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); 2082 return -EINVAL; 2083 } 2084 2085 /* 2086 * The device list in fs_devices is accessed without locks (neither 2087 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2088 * filesystem and another device rm cannot run. 2089 */ 2090 num_devices = btrfs_num_devices(fs_info); 2091 2092 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2093 if (ret) 2094 return ret; 2095 2096 device = btrfs_find_device(fs_info->fs_devices, args); 2097 if (!device) { 2098 if (args->missing) 2099 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2100 else 2101 ret = -ENOENT; 2102 return ret; 2103 } 2104 2105 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2106 btrfs_warn_in_rcu(fs_info, 2107 "cannot remove device %s (devid %llu) due to active swapfile", 2108 rcu_str_deref(device->name), device->devid); 2109 return -ETXTBSY; 2110 } 2111 2112 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 2113 return BTRFS_ERROR_DEV_TGT_REPLACE; 2114 2115 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2116 fs_info->fs_devices->rw_devices == 1) 2117 return BTRFS_ERROR_DEV_ONLY_WRITABLE; 2118 2119 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2120 mutex_lock(&fs_info->chunk_mutex); 2121 list_del_init(&device->dev_alloc_list); 2122 device->fs_devices->rw_devices--; 2123 mutex_unlock(&fs_info->chunk_mutex); 2124 } 2125 2126 ret = btrfs_shrink_device(device, 0); 2127 if (ret) 2128 goto error_undo; 2129 2130 trans = btrfs_start_transaction(fs_info->chunk_root, 0); 2131 if (IS_ERR(trans)) { 2132 ret = PTR_ERR(trans); 2133 goto error_undo; 2134 } 2135 2136 ret = btrfs_rm_dev_item(trans, device); 2137 if (ret) { 2138 /* Any error in dev item removal is critical */ 2139 btrfs_crit(fs_info, 2140 "failed to remove device item for devid %llu: %d", 2141 device->devid, ret); 2142 btrfs_abort_transaction(trans, ret); 2143 btrfs_end_transaction(trans); 2144 return ret; 2145 } 2146 2147 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2148 btrfs_scrub_cancel_dev(device); 2149 2150 /* 2151 * the device list mutex makes sure that we don't change 2152 * the device list while someone else is writing out all 2153 * the device supers. Whoever is writing all supers, should 2154 * lock the device list mutex before getting the number of 2155 * devices in the super block (super_copy). Conversely, 2156 * whoever updates the number of devices in the super block 2157 * (super_copy) should hold the device list mutex. 2158 */ 2159 2160 /* 2161 * In normal cases the cur_devices == fs_devices. But in case 2162 * of deleting a seed device, the cur_devices should point to 2163 * its own fs_devices listed under the fs_devices->seed_list. 2164 */ 2165 cur_devices = device->fs_devices; 2166 mutex_lock(&fs_devices->device_list_mutex); 2167 list_del_rcu(&device->dev_list); 2168 2169 cur_devices->num_devices--; 2170 cur_devices->total_devices--; 2171 /* Update total_devices of the parent fs_devices if it's seed */ 2172 if (cur_devices != fs_devices) 2173 fs_devices->total_devices--; 2174 2175 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2176 cur_devices->missing_devices--; 2177 2178 btrfs_assign_next_active_device(device, NULL); 2179 2180 if (device->bdev) { 2181 cur_devices->open_devices--; 2182 /* remove sysfs entry */ 2183 btrfs_sysfs_remove_device(device); 2184 } 2185 2186 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2187 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2188 mutex_unlock(&fs_devices->device_list_mutex); 2189 2190 /* 2191 * At this point, the device is zero sized and detached from the 2192 * devices list. All that's left is to zero out the old supers and 2193 * free the device. 2194 * 2195 * We cannot call btrfs_close_bdev() here because we're holding the sb 2196 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2197 * block device and it's dependencies. Instead just flush the device 2198 * and let the caller do the final blkdev_put. 2199 */ 2200 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2201 btrfs_scratch_superblocks(fs_info, device->bdev, 2202 device->name->str); 2203 if (device->bdev) { 2204 sync_blockdev(device->bdev); 2205 invalidate_bdev(device->bdev); 2206 } 2207 } 2208 2209 *bdev = device->bdev; 2210 *mode = device->mode; 2211 synchronize_rcu(); 2212 btrfs_free_device(device); 2213 2214 /* 2215 * This can happen if cur_devices is the private seed devices list. We 2216 * cannot call close_fs_devices() here because it expects the uuid_mutex 2217 * to be held, but in fact we don't need that for the private 2218 * seed_devices, we can simply decrement cur_devices->opened and then 2219 * remove it from our list and free the fs_devices. 2220 */ 2221 if (cur_devices->num_devices == 0) { 2222 list_del_init(&cur_devices->seed_list); 2223 ASSERT(cur_devices->opened == 1); 2224 cur_devices->opened--; 2225 free_fs_devices(cur_devices); 2226 } 2227 2228 ret = btrfs_commit_transaction(trans); 2229 2230 return ret; 2231 2232 error_undo: 2233 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2234 mutex_lock(&fs_info->chunk_mutex); 2235 list_add(&device->dev_alloc_list, 2236 &fs_devices->alloc_list); 2237 device->fs_devices->rw_devices++; 2238 mutex_unlock(&fs_info->chunk_mutex); 2239 } 2240 return ret; 2241 } 2242 2243 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2244 { 2245 struct btrfs_fs_devices *fs_devices; 2246 2247 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2248 2249 /* 2250 * in case of fs with no seed, srcdev->fs_devices will point 2251 * to fs_devices of fs_info. However when the dev being replaced is 2252 * a seed dev it will point to the seed's local fs_devices. In short 2253 * srcdev will have its correct fs_devices in both the cases. 2254 */ 2255 fs_devices = srcdev->fs_devices; 2256 2257 list_del_rcu(&srcdev->dev_list); 2258 list_del(&srcdev->dev_alloc_list); 2259 fs_devices->num_devices--; 2260 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2261 fs_devices->missing_devices--; 2262 2263 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2264 fs_devices->rw_devices--; 2265 2266 if (srcdev->bdev) 2267 fs_devices->open_devices--; 2268 } 2269 2270 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2271 { 2272 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2273 2274 mutex_lock(&uuid_mutex); 2275 2276 btrfs_close_bdev(srcdev); 2277 synchronize_rcu(); 2278 btrfs_free_device(srcdev); 2279 2280 /* if this is no devs we rather delete the fs_devices */ 2281 if (!fs_devices->num_devices) { 2282 /* 2283 * On a mounted FS, num_devices can't be zero unless it's a 2284 * seed. In case of a seed device being replaced, the replace 2285 * target added to the sprout FS, so there will be no more 2286 * device left under the seed FS. 2287 */ 2288 ASSERT(fs_devices->seeding); 2289 2290 list_del_init(&fs_devices->seed_list); 2291 close_fs_devices(fs_devices); 2292 free_fs_devices(fs_devices); 2293 } 2294 mutex_unlock(&uuid_mutex); 2295 } 2296 2297 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2298 { 2299 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2300 2301 mutex_lock(&fs_devices->device_list_mutex); 2302 2303 btrfs_sysfs_remove_device(tgtdev); 2304 2305 if (tgtdev->bdev) 2306 fs_devices->open_devices--; 2307 2308 fs_devices->num_devices--; 2309 2310 btrfs_assign_next_active_device(tgtdev, NULL); 2311 2312 list_del_rcu(&tgtdev->dev_list); 2313 2314 mutex_unlock(&fs_devices->device_list_mutex); 2315 2316 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2317 tgtdev->name->str); 2318 2319 btrfs_close_bdev(tgtdev); 2320 synchronize_rcu(); 2321 btrfs_free_device(tgtdev); 2322 } 2323 2324 /** 2325 * Populate args from device at path 2326 * 2327 * @fs_info: the filesystem 2328 * @args: the args to populate 2329 * @path: the path to the device 2330 * 2331 * This will read the super block of the device at @path and populate @args with 2332 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2333 * lookup a device to operate on, but need to do it before we take any locks. 2334 * This properly handles the special case of "missing" that a user may pass in, 2335 * and does some basic sanity checks. The caller must make sure that @path is 2336 * properly NUL terminated before calling in, and must call 2337 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2338 * uuid buffers. 2339 * 2340 * Return: 0 for success, -errno for failure 2341 */ 2342 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2343 struct btrfs_dev_lookup_args *args, 2344 const char *path) 2345 { 2346 struct btrfs_super_block *disk_super; 2347 struct block_device *bdev; 2348 int ret; 2349 2350 if (!path || !path[0]) 2351 return -EINVAL; 2352 if (!strcmp(path, "missing")) { 2353 args->missing = true; 2354 return 0; 2355 } 2356 2357 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2358 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2359 if (!args->uuid || !args->fsid) { 2360 btrfs_put_dev_args_from_path(args); 2361 return -ENOMEM; 2362 } 2363 2364 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2365 &bdev, &disk_super); 2366 if (ret) { 2367 btrfs_put_dev_args_from_path(args); 2368 return ret; 2369 } 2370 2371 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2372 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2373 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2374 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2375 else 2376 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2377 btrfs_release_disk_super(disk_super); 2378 blkdev_put(bdev, FMODE_READ); 2379 return 0; 2380 } 2381 2382 /* 2383 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2384 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2385 * that don't need to be freed. 2386 */ 2387 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2388 { 2389 kfree(args->uuid); 2390 kfree(args->fsid); 2391 args->uuid = NULL; 2392 args->fsid = NULL; 2393 } 2394 2395 struct btrfs_device *btrfs_find_device_by_devspec( 2396 struct btrfs_fs_info *fs_info, u64 devid, 2397 const char *device_path) 2398 { 2399 BTRFS_DEV_LOOKUP_ARGS(args); 2400 struct btrfs_device *device; 2401 int ret; 2402 2403 if (devid) { 2404 args.devid = devid; 2405 device = btrfs_find_device(fs_info->fs_devices, &args); 2406 if (!device) 2407 return ERR_PTR(-ENOENT); 2408 return device; 2409 } 2410 2411 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2412 if (ret) 2413 return ERR_PTR(ret); 2414 device = btrfs_find_device(fs_info->fs_devices, &args); 2415 btrfs_put_dev_args_from_path(&args); 2416 if (!device) 2417 return ERR_PTR(-ENOENT); 2418 return device; 2419 } 2420 2421 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) 2422 { 2423 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2424 struct btrfs_fs_devices *old_devices; 2425 struct btrfs_fs_devices *seed_devices; 2426 2427 lockdep_assert_held(&uuid_mutex); 2428 if (!fs_devices->seeding) 2429 return ERR_PTR(-EINVAL); 2430 2431 /* 2432 * Private copy of the seed devices, anchored at 2433 * fs_info->fs_devices->seed_list 2434 */ 2435 seed_devices = alloc_fs_devices(NULL, NULL); 2436 if (IS_ERR(seed_devices)) 2437 return seed_devices; 2438 2439 /* 2440 * It's necessary to retain a copy of the original seed fs_devices in 2441 * fs_uuids so that filesystems which have been seeded can successfully 2442 * reference the seed device from open_seed_devices. This also supports 2443 * multiple fs seed. 2444 */ 2445 old_devices = clone_fs_devices(fs_devices); 2446 if (IS_ERR(old_devices)) { 2447 kfree(seed_devices); 2448 return old_devices; 2449 } 2450 2451 list_add(&old_devices->fs_list, &fs_uuids); 2452 2453 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2454 seed_devices->opened = 1; 2455 INIT_LIST_HEAD(&seed_devices->devices); 2456 INIT_LIST_HEAD(&seed_devices->alloc_list); 2457 mutex_init(&seed_devices->device_list_mutex); 2458 2459 return seed_devices; 2460 } 2461 2462 /* 2463 * Splice seed devices into the sprout fs_devices. 2464 * Generate a new fsid for the sprouted read-write filesystem. 2465 */ 2466 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, 2467 struct btrfs_fs_devices *seed_devices) 2468 { 2469 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2470 struct btrfs_super_block *disk_super = fs_info->super_copy; 2471 struct btrfs_device *device; 2472 u64 super_flags; 2473 2474 /* 2475 * We are updating the fsid, the thread leading to device_list_add() 2476 * could race, so uuid_mutex is needed. 2477 */ 2478 lockdep_assert_held(&uuid_mutex); 2479 2480 /* 2481 * The threads listed below may traverse dev_list but can do that without 2482 * device_list_mutex: 2483 * - All device ops and balance - as we are in btrfs_exclop_start. 2484 * - Various dev_list readers - are using RCU. 2485 * - btrfs_ioctl_fitrim() - is using RCU. 2486 * 2487 * For-read threads as below are using device_list_mutex: 2488 * - Readonly scrub btrfs_scrub_dev() 2489 * - Readonly scrub btrfs_scrub_progress() 2490 * - btrfs_get_dev_stats() 2491 */ 2492 lockdep_assert_held(&fs_devices->device_list_mutex); 2493 2494 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2495 synchronize_rcu); 2496 list_for_each_entry(device, &seed_devices->devices, dev_list) 2497 device->fs_devices = seed_devices; 2498 2499 fs_devices->seeding = false; 2500 fs_devices->num_devices = 0; 2501 fs_devices->open_devices = 0; 2502 fs_devices->missing_devices = 0; 2503 fs_devices->rotating = false; 2504 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2505 2506 generate_random_uuid(fs_devices->fsid); 2507 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2508 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2509 2510 super_flags = btrfs_super_flags(disk_super) & 2511 ~BTRFS_SUPER_FLAG_SEEDING; 2512 btrfs_set_super_flags(disk_super, super_flags); 2513 } 2514 2515 /* 2516 * Store the expected generation for seed devices in device items. 2517 */ 2518 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2519 { 2520 BTRFS_DEV_LOOKUP_ARGS(args); 2521 struct btrfs_fs_info *fs_info = trans->fs_info; 2522 struct btrfs_root *root = fs_info->chunk_root; 2523 struct btrfs_path *path; 2524 struct extent_buffer *leaf; 2525 struct btrfs_dev_item *dev_item; 2526 struct btrfs_device *device; 2527 struct btrfs_key key; 2528 u8 fs_uuid[BTRFS_FSID_SIZE]; 2529 u8 dev_uuid[BTRFS_UUID_SIZE]; 2530 int ret; 2531 2532 path = btrfs_alloc_path(); 2533 if (!path) 2534 return -ENOMEM; 2535 2536 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2537 key.offset = 0; 2538 key.type = BTRFS_DEV_ITEM_KEY; 2539 2540 while (1) { 2541 btrfs_reserve_chunk_metadata(trans, false); 2542 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2543 btrfs_trans_release_chunk_metadata(trans); 2544 if (ret < 0) 2545 goto error; 2546 2547 leaf = path->nodes[0]; 2548 next_slot: 2549 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2550 ret = btrfs_next_leaf(root, path); 2551 if (ret > 0) 2552 break; 2553 if (ret < 0) 2554 goto error; 2555 leaf = path->nodes[0]; 2556 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2557 btrfs_release_path(path); 2558 continue; 2559 } 2560 2561 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2562 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2563 key.type != BTRFS_DEV_ITEM_KEY) 2564 break; 2565 2566 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2567 struct btrfs_dev_item); 2568 args.devid = btrfs_device_id(leaf, dev_item); 2569 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2570 BTRFS_UUID_SIZE); 2571 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2572 BTRFS_FSID_SIZE); 2573 args.uuid = dev_uuid; 2574 args.fsid = fs_uuid; 2575 device = btrfs_find_device(fs_info->fs_devices, &args); 2576 BUG_ON(!device); /* Logic error */ 2577 2578 if (device->fs_devices->seeding) { 2579 btrfs_set_device_generation(leaf, dev_item, 2580 device->generation); 2581 btrfs_mark_buffer_dirty(leaf); 2582 } 2583 2584 path->slots[0]++; 2585 goto next_slot; 2586 } 2587 ret = 0; 2588 error: 2589 btrfs_free_path(path); 2590 return ret; 2591 } 2592 2593 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2594 { 2595 struct btrfs_root *root = fs_info->dev_root; 2596 struct btrfs_trans_handle *trans; 2597 struct btrfs_device *device; 2598 struct block_device *bdev; 2599 struct super_block *sb = fs_info->sb; 2600 struct rcu_string *name; 2601 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2602 struct btrfs_fs_devices *seed_devices; 2603 u64 orig_super_total_bytes; 2604 u64 orig_super_num_devices; 2605 int ret = 0; 2606 bool seeding_dev = false; 2607 bool locked = false; 2608 2609 if (sb_rdonly(sb) && !fs_devices->seeding) 2610 return -EROFS; 2611 2612 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2613 fs_info->bdev_holder); 2614 if (IS_ERR(bdev)) 2615 return PTR_ERR(bdev); 2616 2617 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2618 ret = -EINVAL; 2619 goto error; 2620 } 2621 2622 if (fs_devices->seeding) { 2623 seeding_dev = true; 2624 down_write(&sb->s_umount); 2625 mutex_lock(&uuid_mutex); 2626 locked = true; 2627 } 2628 2629 sync_blockdev(bdev); 2630 2631 rcu_read_lock(); 2632 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2633 if (device->bdev == bdev) { 2634 ret = -EEXIST; 2635 rcu_read_unlock(); 2636 goto error; 2637 } 2638 } 2639 rcu_read_unlock(); 2640 2641 device = btrfs_alloc_device(fs_info, NULL, NULL); 2642 if (IS_ERR(device)) { 2643 /* we can safely leave the fs_devices entry around */ 2644 ret = PTR_ERR(device); 2645 goto error; 2646 } 2647 2648 name = rcu_string_strdup(device_path, GFP_KERNEL); 2649 if (!name) { 2650 ret = -ENOMEM; 2651 goto error_free_device; 2652 } 2653 rcu_assign_pointer(device->name, name); 2654 2655 device->fs_info = fs_info; 2656 device->bdev = bdev; 2657 ret = lookup_bdev(device_path, &device->devt); 2658 if (ret) 2659 goto error_free_device; 2660 2661 ret = btrfs_get_dev_zone_info(device, false); 2662 if (ret) 2663 goto error_free_device; 2664 2665 trans = btrfs_start_transaction(root, 0); 2666 if (IS_ERR(trans)) { 2667 ret = PTR_ERR(trans); 2668 goto error_free_zone; 2669 } 2670 2671 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2672 device->generation = trans->transid; 2673 device->io_width = fs_info->sectorsize; 2674 device->io_align = fs_info->sectorsize; 2675 device->sector_size = fs_info->sectorsize; 2676 device->total_bytes = 2677 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); 2678 device->disk_total_bytes = device->total_bytes; 2679 device->commit_total_bytes = device->total_bytes; 2680 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2681 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2682 device->mode = FMODE_EXCL; 2683 device->dev_stats_valid = 1; 2684 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2685 2686 if (seeding_dev) { 2687 btrfs_clear_sb_rdonly(sb); 2688 2689 /* GFP_KERNEL allocation must not be under device_list_mutex */ 2690 seed_devices = btrfs_init_sprout(fs_info); 2691 if (IS_ERR(seed_devices)) { 2692 ret = PTR_ERR(seed_devices); 2693 btrfs_abort_transaction(trans, ret); 2694 goto error_trans; 2695 } 2696 } 2697 2698 mutex_lock(&fs_devices->device_list_mutex); 2699 if (seeding_dev) { 2700 btrfs_setup_sprout(fs_info, seed_devices); 2701 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2702 device); 2703 } 2704 2705 device->fs_devices = fs_devices; 2706 2707 mutex_lock(&fs_info->chunk_mutex); 2708 list_add_rcu(&device->dev_list, &fs_devices->devices); 2709 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2710 fs_devices->num_devices++; 2711 fs_devices->open_devices++; 2712 fs_devices->rw_devices++; 2713 fs_devices->total_devices++; 2714 fs_devices->total_rw_bytes += device->total_bytes; 2715 2716 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2717 2718 if (!bdev_nonrot(bdev)) 2719 fs_devices->rotating = true; 2720 2721 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2722 btrfs_set_super_total_bytes(fs_info->super_copy, 2723 round_down(orig_super_total_bytes + device->total_bytes, 2724 fs_info->sectorsize)); 2725 2726 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2727 btrfs_set_super_num_devices(fs_info->super_copy, 2728 orig_super_num_devices + 1); 2729 2730 /* 2731 * we've got more storage, clear any full flags on the space 2732 * infos 2733 */ 2734 btrfs_clear_space_info_full(fs_info); 2735 2736 mutex_unlock(&fs_info->chunk_mutex); 2737 2738 /* Add sysfs device entry */ 2739 btrfs_sysfs_add_device(device); 2740 2741 mutex_unlock(&fs_devices->device_list_mutex); 2742 2743 if (seeding_dev) { 2744 mutex_lock(&fs_info->chunk_mutex); 2745 ret = init_first_rw_device(trans); 2746 mutex_unlock(&fs_info->chunk_mutex); 2747 if (ret) { 2748 btrfs_abort_transaction(trans, ret); 2749 goto error_sysfs; 2750 } 2751 } 2752 2753 ret = btrfs_add_dev_item(trans, device); 2754 if (ret) { 2755 btrfs_abort_transaction(trans, ret); 2756 goto error_sysfs; 2757 } 2758 2759 if (seeding_dev) { 2760 ret = btrfs_finish_sprout(trans); 2761 if (ret) { 2762 btrfs_abort_transaction(trans, ret); 2763 goto error_sysfs; 2764 } 2765 2766 /* 2767 * fs_devices now represents the newly sprouted filesystem and 2768 * its fsid has been changed by btrfs_sprout_splice(). 2769 */ 2770 btrfs_sysfs_update_sprout_fsid(fs_devices); 2771 } 2772 2773 ret = btrfs_commit_transaction(trans); 2774 2775 if (seeding_dev) { 2776 mutex_unlock(&uuid_mutex); 2777 up_write(&sb->s_umount); 2778 locked = false; 2779 2780 if (ret) /* transaction commit */ 2781 return ret; 2782 2783 ret = btrfs_relocate_sys_chunks(fs_info); 2784 if (ret < 0) 2785 btrfs_handle_fs_error(fs_info, ret, 2786 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2787 trans = btrfs_attach_transaction(root); 2788 if (IS_ERR(trans)) { 2789 if (PTR_ERR(trans) == -ENOENT) 2790 return 0; 2791 ret = PTR_ERR(trans); 2792 trans = NULL; 2793 goto error_sysfs; 2794 } 2795 ret = btrfs_commit_transaction(trans); 2796 } 2797 2798 /* 2799 * Now that we have written a new super block to this device, check all 2800 * other fs_devices list if device_path alienates any other scanned 2801 * device. 2802 * We can ignore the return value as it typically returns -EINVAL and 2803 * only succeeds if the device was an alien. 2804 */ 2805 btrfs_forget_devices(device->devt); 2806 2807 /* Update ctime/mtime for blkid or udev */ 2808 update_dev_time(device_path); 2809 2810 return ret; 2811 2812 error_sysfs: 2813 btrfs_sysfs_remove_device(device); 2814 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2815 mutex_lock(&fs_info->chunk_mutex); 2816 list_del_rcu(&device->dev_list); 2817 list_del(&device->dev_alloc_list); 2818 fs_info->fs_devices->num_devices--; 2819 fs_info->fs_devices->open_devices--; 2820 fs_info->fs_devices->rw_devices--; 2821 fs_info->fs_devices->total_devices--; 2822 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2823 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2824 btrfs_set_super_total_bytes(fs_info->super_copy, 2825 orig_super_total_bytes); 2826 btrfs_set_super_num_devices(fs_info->super_copy, 2827 orig_super_num_devices); 2828 mutex_unlock(&fs_info->chunk_mutex); 2829 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2830 error_trans: 2831 if (seeding_dev) 2832 btrfs_set_sb_rdonly(sb); 2833 if (trans) 2834 btrfs_end_transaction(trans); 2835 error_free_zone: 2836 btrfs_destroy_dev_zone_info(device); 2837 error_free_device: 2838 btrfs_free_device(device); 2839 error: 2840 blkdev_put(bdev, FMODE_EXCL); 2841 if (locked) { 2842 mutex_unlock(&uuid_mutex); 2843 up_write(&sb->s_umount); 2844 } 2845 return ret; 2846 } 2847 2848 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2849 struct btrfs_device *device) 2850 { 2851 int ret; 2852 struct btrfs_path *path; 2853 struct btrfs_root *root = device->fs_info->chunk_root; 2854 struct btrfs_dev_item *dev_item; 2855 struct extent_buffer *leaf; 2856 struct btrfs_key key; 2857 2858 path = btrfs_alloc_path(); 2859 if (!path) 2860 return -ENOMEM; 2861 2862 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2863 key.type = BTRFS_DEV_ITEM_KEY; 2864 key.offset = device->devid; 2865 2866 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2867 if (ret < 0) 2868 goto out; 2869 2870 if (ret > 0) { 2871 ret = -ENOENT; 2872 goto out; 2873 } 2874 2875 leaf = path->nodes[0]; 2876 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2877 2878 btrfs_set_device_id(leaf, dev_item, device->devid); 2879 btrfs_set_device_type(leaf, dev_item, device->type); 2880 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2881 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2882 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2883 btrfs_set_device_total_bytes(leaf, dev_item, 2884 btrfs_device_get_disk_total_bytes(device)); 2885 btrfs_set_device_bytes_used(leaf, dev_item, 2886 btrfs_device_get_bytes_used(device)); 2887 btrfs_mark_buffer_dirty(leaf); 2888 2889 out: 2890 btrfs_free_path(path); 2891 return ret; 2892 } 2893 2894 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2895 struct btrfs_device *device, u64 new_size) 2896 { 2897 struct btrfs_fs_info *fs_info = device->fs_info; 2898 struct btrfs_super_block *super_copy = fs_info->super_copy; 2899 u64 old_total; 2900 u64 diff; 2901 int ret; 2902 2903 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2904 return -EACCES; 2905 2906 new_size = round_down(new_size, fs_info->sectorsize); 2907 2908 mutex_lock(&fs_info->chunk_mutex); 2909 old_total = btrfs_super_total_bytes(super_copy); 2910 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2911 2912 if (new_size <= device->total_bytes || 2913 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2914 mutex_unlock(&fs_info->chunk_mutex); 2915 return -EINVAL; 2916 } 2917 2918 btrfs_set_super_total_bytes(super_copy, 2919 round_down(old_total + diff, fs_info->sectorsize)); 2920 device->fs_devices->total_rw_bytes += diff; 2921 2922 btrfs_device_set_total_bytes(device, new_size); 2923 btrfs_device_set_disk_total_bytes(device, new_size); 2924 btrfs_clear_space_info_full(device->fs_info); 2925 if (list_empty(&device->post_commit_list)) 2926 list_add_tail(&device->post_commit_list, 2927 &trans->transaction->dev_update_list); 2928 mutex_unlock(&fs_info->chunk_mutex); 2929 2930 btrfs_reserve_chunk_metadata(trans, false); 2931 ret = btrfs_update_device(trans, device); 2932 btrfs_trans_release_chunk_metadata(trans); 2933 2934 return ret; 2935 } 2936 2937 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2938 { 2939 struct btrfs_fs_info *fs_info = trans->fs_info; 2940 struct btrfs_root *root = fs_info->chunk_root; 2941 int ret; 2942 struct btrfs_path *path; 2943 struct btrfs_key key; 2944 2945 path = btrfs_alloc_path(); 2946 if (!path) 2947 return -ENOMEM; 2948 2949 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2950 key.offset = chunk_offset; 2951 key.type = BTRFS_CHUNK_ITEM_KEY; 2952 2953 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2954 if (ret < 0) 2955 goto out; 2956 else if (ret > 0) { /* Logic error or corruption */ 2957 btrfs_handle_fs_error(fs_info, -ENOENT, 2958 "Failed lookup while freeing chunk."); 2959 ret = -ENOENT; 2960 goto out; 2961 } 2962 2963 ret = btrfs_del_item(trans, root, path); 2964 if (ret < 0) 2965 btrfs_handle_fs_error(fs_info, ret, 2966 "Failed to delete chunk item."); 2967 out: 2968 btrfs_free_path(path); 2969 return ret; 2970 } 2971 2972 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2973 { 2974 struct btrfs_super_block *super_copy = fs_info->super_copy; 2975 struct btrfs_disk_key *disk_key; 2976 struct btrfs_chunk *chunk; 2977 u8 *ptr; 2978 int ret = 0; 2979 u32 num_stripes; 2980 u32 array_size; 2981 u32 len = 0; 2982 u32 cur; 2983 struct btrfs_key key; 2984 2985 lockdep_assert_held(&fs_info->chunk_mutex); 2986 array_size = btrfs_super_sys_array_size(super_copy); 2987 2988 ptr = super_copy->sys_chunk_array; 2989 cur = 0; 2990 2991 while (cur < array_size) { 2992 disk_key = (struct btrfs_disk_key *)ptr; 2993 btrfs_disk_key_to_cpu(&key, disk_key); 2994 2995 len = sizeof(*disk_key); 2996 2997 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 2998 chunk = (struct btrfs_chunk *)(ptr + len); 2999 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 3000 len += btrfs_chunk_item_size(num_stripes); 3001 } else { 3002 ret = -EIO; 3003 break; 3004 } 3005 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 3006 key.offset == chunk_offset) { 3007 memmove(ptr, ptr + len, array_size - (cur + len)); 3008 array_size -= len; 3009 btrfs_set_super_sys_array_size(super_copy, array_size); 3010 } else { 3011 ptr += len; 3012 cur += len; 3013 } 3014 } 3015 return ret; 3016 } 3017 3018 /* 3019 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 3020 * @logical: Logical block offset in bytes. 3021 * @length: Length of extent in bytes. 3022 * 3023 * Return: Chunk mapping or ERR_PTR. 3024 */ 3025 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 3026 u64 logical, u64 length) 3027 { 3028 struct extent_map_tree *em_tree; 3029 struct extent_map *em; 3030 3031 em_tree = &fs_info->mapping_tree; 3032 read_lock(&em_tree->lock); 3033 em = lookup_extent_mapping(em_tree, logical, length); 3034 read_unlock(&em_tree->lock); 3035 3036 if (!em) { 3037 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 3038 logical, length); 3039 return ERR_PTR(-EINVAL); 3040 } 3041 3042 if (em->start > logical || em->start + em->len < logical) { 3043 btrfs_crit(fs_info, 3044 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3045 logical, length, em->start, em->start + em->len); 3046 free_extent_map(em); 3047 return ERR_PTR(-EINVAL); 3048 } 3049 3050 /* callers are responsible for dropping em's ref. */ 3051 return em; 3052 } 3053 3054 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3055 struct map_lookup *map, u64 chunk_offset) 3056 { 3057 int i; 3058 3059 /* 3060 * Removing chunk items and updating the device items in the chunks btree 3061 * requires holding the chunk_mutex. 3062 * See the comment at btrfs_chunk_alloc() for the details. 3063 */ 3064 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3065 3066 for (i = 0; i < map->num_stripes; i++) { 3067 int ret; 3068 3069 ret = btrfs_update_device(trans, map->stripes[i].dev); 3070 if (ret) 3071 return ret; 3072 } 3073 3074 return btrfs_free_chunk(trans, chunk_offset); 3075 } 3076 3077 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3078 { 3079 struct btrfs_fs_info *fs_info = trans->fs_info; 3080 struct extent_map *em; 3081 struct map_lookup *map; 3082 u64 dev_extent_len = 0; 3083 int i, ret = 0; 3084 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3085 3086 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3087 if (IS_ERR(em)) { 3088 /* 3089 * This is a logic error, but we don't want to just rely on the 3090 * user having built with ASSERT enabled, so if ASSERT doesn't 3091 * do anything we still error out. 3092 */ 3093 ASSERT(0); 3094 return PTR_ERR(em); 3095 } 3096 map = em->map_lookup; 3097 3098 /* 3099 * First delete the device extent items from the devices btree. 3100 * We take the device_list_mutex to avoid racing with the finishing phase 3101 * of a device replace operation. See the comment below before acquiring 3102 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3103 * because that can result in a deadlock when deleting the device extent 3104 * items from the devices btree - COWing an extent buffer from the btree 3105 * may result in allocating a new metadata chunk, which would attempt to 3106 * lock again fs_info->chunk_mutex. 3107 */ 3108 mutex_lock(&fs_devices->device_list_mutex); 3109 for (i = 0; i < map->num_stripes; i++) { 3110 struct btrfs_device *device = map->stripes[i].dev; 3111 ret = btrfs_free_dev_extent(trans, device, 3112 map->stripes[i].physical, 3113 &dev_extent_len); 3114 if (ret) { 3115 mutex_unlock(&fs_devices->device_list_mutex); 3116 btrfs_abort_transaction(trans, ret); 3117 goto out; 3118 } 3119 3120 if (device->bytes_used > 0) { 3121 mutex_lock(&fs_info->chunk_mutex); 3122 btrfs_device_set_bytes_used(device, 3123 device->bytes_used - dev_extent_len); 3124 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3125 btrfs_clear_space_info_full(fs_info); 3126 mutex_unlock(&fs_info->chunk_mutex); 3127 } 3128 } 3129 mutex_unlock(&fs_devices->device_list_mutex); 3130 3131 /* 3132 * We acquire fs_info->chunk_mutex for 2 reasons: 3133 * 3134 * 1) Just like with the first phase of the chunk allocation, we must 3135 * reserve system space, do all chunk btree updates and deletions, and 3136 * update the system chunk array in the superblock while holding this 3137 * mutex. This is for similar reasons as explained on the comment at 3138 * the top of btrfs_chunk_alloc(); 3139 * 3140 * 2) Prevent races with the final phase of a device replace operation 3141 * that replaces the device object associated with the map's stripes, 3142 * because the device object's id can change at any time during that 3143 * final phase of the device replace operation 3144 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3145 * replaced device and then see it with an ID of 3146 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3147 * the device item, which does not exists on the chunk btree. 3148 * The finishing phase of device replace acquires both the 3149 * device_list_mutex and the chunk_mutex, in that order, so we are 3150 * safe by just acquiring the chunk_mutex. 3151 */ 3152 trans->removing_chunk = true; 3153 mutex_lock(&fs_info->chunk_mutex); 3154 3155 check_system_chunk(trans, map->type); 3156 3157 ret = remove_chunk_item(trans, map, chunk_offset); 3158 /* 3159 * Normally we should not get -ENOSPC since we reserved space before 3160 * through the call to check_system_chunk(). 3161 * 3162 * Despite our system space_info having enough free space, we may not 3163 * be able to allocate extents from its block groups, because all have 3164 * an incompatible profile, which will force us to allocate a new system 3165 * block group with the right profile, or right after we called 3166 * check_system_space() above, a scrub turned the only system block group 3167 * with enough free space into RO mode. 3168 * This is explained with more detail at do_chunk_alloc(). 3169 * 3170 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3171 */ 3172 if (ret == -ENOSPC) { 3173 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3174 struct btrfs_block_group *sys_bg; 3175 3176 sys_bg = btrfs_create_chunk(trans, sys_flags); 3177 if (IS_ERR(sys_bg)) { 3178 ret = PTR_ERR(sys_bg); 3179 btrfs_abort_transaction(trans, ret); 3180 goto out; 3181 } 3182 3183 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3184 if (ret) { 3185 btrfs_abort_transaction(trans, ret); 3186 goto out; 3187 } 3188 3189 ret = remove_chunk_item(trans, map, chunk_offset); 3190 if (ret) { 3191 btrfs_abort_transaction(trans, ret); 3192 goto out; 3193 } 3194 } else if (ret) { 3195 btrfs_abort_transaction(trans, ret); 3196 goto out; 3197 } 3198 3199 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3200 3201 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3202 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3203 if (ret) { 3204 btrfs_abort_transaction(trans, ret); 3205 goto out; 3206 } 3207 } 3208 3209 mutex_unlock(&fs_info->chunk_mutex); 3210 trans->removing_chunk = false; 3211 3212 /* 3213 * We are done with chunk btree updates and deletions, so release the 3214 * system space we previously reserved (with check_system_chunk()). 3215 */ 3216 btrfs_trans_release_chunk_metadata(trans); 3217 3218 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3219 if (ret) { 3220 btrfs_abort_transaction(trans, ret); 3221 goto out; 3222 } 3223 3224 out: 3225 if (trans->removing_chunk) { 3226 mutex_unlock(&fs_info->chunk_mutex); 3227 trans->removing_chunk = false; 3228 } 3229 /* once for us */ 3230 free_extent_map(em); 3231 return ret; 3232 } 3233 3234 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3235 { 3236 struct btrfs_root *root = fs_info->chunk_root; 3237 struct btrfs_trans_handle *trans; 3238 struct btrfs_block_group *block_group; 3239 u64 length; 3240 int ret; 3241 3242 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3243 btrfs_err(fs_info, 3244 "relocate: not supported on extent tree v2 yet"); 3245 return -EINVAL; 3246 } 3247 3248 /* 3249 * Prevent races with automatic removal of unused block groups. 3250 * After we relocate and before we remove the chunk with offset 3251 * chunk_offset, automatic removal of the block group can kick in, 3252 * resulting in a failure when calling btrfs_remove_chunk() below. 3253 * 3254 * Make sure to acquire this mutex before doing a tree search (dev 3255 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3256 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3257 * we release the path used to search the chunk/dev tree and before 3258 * the current task acquires this mutex and calls us. 3259 */ 3260 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3261 3262 /* step one, relocate all the extents inside this chunk */ 3263 btrfs_scrub_pause(fs_info); 3264 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3265 btrfs_scrub_continue(fs_info); 3266 if (ret) 3267 return ret; 3268 3269 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3270 if (!block_group) 3271 return -ENOENT; 3272 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3273 length = block_group->length; 3274 btrfs_put_block_group(block_group); 3275 3276 /* 3277 * On a zoned file system, discard the whole block group, this will 3278 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3279 * resetting the zone fails, don't treat it as a fatal problem from the 3280 * filesystem's point of view. 3281 */ 3282 if (btrfs_is_zoned(fs_info)) { 3283 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3284 if (ret) 3285 btrfs_info(fs_info, 3286 "failed to reset zone %llu after relocation", 3287 chunk_offset); 3288 } 3289 3290 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3291 chunk_offset); 3292 if (IS_ERR(trans)) { 3293 ret = PTR_ERR(trans); 3294 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3295 return ret; 3296 } 3297 3298 /* 3299 * step two, delete the device extents and the 3300 * chunk tree entries 3301 */ 3302 ret = btrfs_remove_chunk(trans, chunk_offset); 3303 btrfs_end_transaction(trans); 3304 return ret; 3305 } 3306 3307 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3308 { 3309 struct btrfs_root *chunk_root = fs_info->chunk_root; 3310 struct btrfs_path *path; 3311 struct extent_buffer *leaf; 3312 struct btrfs_chunk *chunk; 3313 struct btrfs_key key; 3314 struct btrfs_key found_key; 3315 u64 chunk_type; 3316 bool retried = false; 3317 int failed = 0; 3318 int ret; 3319 3320 path = btrfs_alloc_path(); 3321 if (!path) 3322 return -ENOMEM; 3323 3324 again: 3325 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3326 key.offset = (u64)-1; 3327 key.type = BTRFS_CHUNK_ITEM_KEY; 3328 3329 while (1) { 3330 mutex_lock(&fs_info->reclaim_bgs_lock); 3331 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3332 if (ret < 0) { 3333 mutex_unlock(&fs_info->reclaim_bgs_lock); 3334 goto error; 3335 } 3336 BUG_ON(ret == 0); /* Corruption */ 3337 3338 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3339 key.type); 3340 if (ret) 3341 mutex_unlock(&fs_info->reclaim_bgs_lock); 3342 if (ret < 0) 3343 goto error; 3344 if (ret > 0) 3345 break; 3346 3347 leaf = path->nodes[0]; 3348 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3349 3350 chunk = btrfs_item_ptr(leaf, path->slots[0], 3351 struct btrfs_chunk); 3352 chunk_type = btrfs_chunk_type(leaf, chunk); 3353 btrfs_release_path(path); 3354 3355 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3356 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3357 if (ret == -ENOSPC) 3358 failed++; 3359 else 3360 BUG_ON(ret); 3361 } 3362 mutex_unlock(&fs_info->reclaim_bgs_lock); 3363 3364 if (found_key.offset == 0) 3365 break; 3366 key.offset = found_key.offset - 1; 3367 } 3368 ret = 0; 3369 if (failed && !retried) { 3370 failed = 0; 3371 retried = true; 3372 goto again; 3373 } else if (WARN_ON(failed && retried)) { 3374 ret = -ENOSPC; 3375 } 3376 error: 3377 btrfs_free_path(path); 3378 return ret; 3379 } 3380 3381 /* 3382 * return 1 : allocate a data chunk successfully, 3383 * return <0: errors during allocating a data chunk, 3384 * return 0 : no need to allocate a data chunk. 3385 */ 3386 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3387 u64 chunk_offset) 3388 { 3389 struct btrfs_block_group *cache; 3390 u64 bytes_used; 3391 u64 chunk_type; 3392 3393 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3394 ASSERT(cache); 3395 chunk_type = cache->flags; 3396 btrfs_put_block_group(cache); 3397 3398 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3399 return 0; 3400 3401 spin_lock(&fs_info->data_sinfo->lock); 3402 bytes_used = fs_info->data_sinfo->bytes_used; 3403 spin_unlock(&fs_info->data_sinfo->lock); 3404 3405 if (!bytes_used) { 3406 struct btrfs_trans_handle *trans; 3407 int ret; 3408 3409 trans = btrfs_join_transaction(fs_info->tree_root); 3410 if (IS_ERR(trans)) 3411 return PTR_ERR(trans); 3412 3413 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3414 btrfs_end_transaction(trans); 3415 if (ret < 0) 3416 return ret; 3417 return 1; 3418 } 3419 3420 return 0; 3421 } 3422 3423 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3424 struct btrfs_balance_control *bctl) 3425 { 3426 struct btrfs_root *root = fs_info->tree_root; 3427 struct btrfs_trans_handle *trans; 3428 struct btrfs_balance_item *item; 3429 struct btrfs_disk_balance_args disk_bargs; 3430 struct btrfs_path *path; 3431 struct extent_buffer *leaf; 3432 struct btrfs_key key; 3433 int ret, err; 3434 3435 path = btrfs_alloc_path(); 3436 if (!path) 3437 return -ENOMEM; 3438 3439 trans = btrfs_start_transaction(root, 0); 3440 if (IS_ERR(trans)) { 3441 btrfs_free_path(path); 3442 return PTR_ERR(trans); 3443 } 3444 3445 key.objectid = BTRFS_BALANCE_OBJECTID; 3446 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3447 key.offset = 0; 3448 3449 ret = btrfs_insert_empty_item(trans, root, path, &key, 3450 sizeof(*item)); 3451 if (ret) 3452 goto out; 3453 3454 leaf = path->nodes[0]; 3455 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3456 3457 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3458 3459 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3460 btrfs_set_balance_data(leaf, item, &disk_bargs); 3461 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3462 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3463 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3464 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3465 3466 btrfs_set_balance_flags(leaf, item, bctl->flags); 3467 3468 btrfs_mark_buffer_dirty(leaf); 3469 out: 3470 btrfs_free_path(path); 3471 err = btrfs_commit_transaction(trans); 3472 if (err && !ret) 3473 ret = err; 3474 return ret; 3475 } 3476 3477 static int del_balance_item(struct btrfs_fs_info *fs_info) 3478 { 3479 struct btrfs_root *root = fs_info->tree_root; 3480 struct btrfs_trans_handle *trans; 3481 struct btrfs_path *path; 3482 struct btrfs_key key; 3483 int ret, err; 3484 3485 path = btrfs_alloc_path(); 3486 if (!path) 3487 return -ENOMEM; 3488 3489 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3490 if (IS_ERR(trans)) { 3491 btrfs_free_path(path); 3492 return PTR_ERR(trans); 3493 } 3494 3495 key.objectid = BTRFS_BALANCE_OBJECTID; 3496 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3497 key.offset = 0; 3498 3499 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3500 if (ret < 0) 3501 goto out; 3502 if (ret > 0) { 3503 ret = -ENOENT; 3504 goto out; 3505 } 3506 3507 ret = btrfs_del_item(trans, root, path); 3508 out: 3509 btrfs_free_path(path); 3510 err = btrfs_commit_transaction(trans); 3511 if (err && !ret) 3512 ret = err; 3513 return ret; 3514 } 3515 3516 /* 3517 * This is a heuristic used to reduce the number of chunks balanced on 3518 * resume after balance was interrupted. 3519 */ 3520 static void update_balance_args(struct btrfs_balance_control *bctl) 3521 { 3522 /* 3523 * Turn on soft mode for chunk types that were being converted. 3524 */ 3525 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3526 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3527 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3528 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3529 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3530 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3531 3532 /* 3533 * Turn on usage filter if is not already used. The idea is 3534 * that chunks that we have already balanced should be 3535 * reasonably full. Don't do it for chunks that are being 3536 * converted - that will keep us from relocating unconverted 3537 * (albeit full) chunks. 3538 */ 3539 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3540 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3541 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3542 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3543 bctl->data.usage = 90; 3544 } 3545 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3546 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3547 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3548 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3549 bctl->sys.usage = 90; 3550 } 3551 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3552 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3553 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3554 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3555 bctl->meta.usage = 90; 3556 } 3557 } 3558 3559 /* 3560 * Clear the balance status in fs_info and delete the balance item from disk. 3561 */ 3562 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3563 { 3564 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3565 int ret; 3566 3567 BUG_ON(!fs_info->balance_ctl); 3568 3569 spin_lock(&fs_info->balance_lock); 3570 fs_info->balance_ctl = NULL; 3571 spin_unlock(&fs_info->balance_lock); 3572 3573 kfree(bctl); 3574 ret = del_balance_item(fs_info); 3575 if (ret) 3576 btrfs_handle_fs_error(fs_info, ret, NULL); 3577 } 3578 3579 /* 3580 * Balance filters. Return 1 if chunk should be filtered out 3581 * (should not be balanced). 3582 */ 3583 static int chunk_profiles_filter(u64 chunk_type, 3584 struct btrfs_balance_args *bargs) 3585 { 3586 chunk_type = chunk_to_extended(chunk_type) & 3587 BTRFS_EXTENDED_PROFILE_MASK; 3588 3589 if (bargs->profiles & chunk_type) 3590 return 0; 3591 3592 return 1; 3593 } 3594 3595 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3596 struct btrfs_balance_args *bargs) 3597 { 3598 struct btrfs_block_group *cache; 3599 u64 chunk_used; 3600 u64 user_thresh_min; 3601 u64 user_thresh_max; 3602 int ret = 1; 3603 3604 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3605 chunk_used = cache->used; 3606 3607 if (bargs->usage_min == 0) 3608 user_thresh_min = 0; 3609 else 3610 user_thresh_min = div_factor_fine(cache->length, 3611 bargs->usage_min); 3612 3613 if (bargs->usage_max == 0) 3614 user_thresh_max = 1; 3615 else if (bargs->usage_max > 100) 3616 user_thresh_max = cache->length; 3617 else 3618 user_thresh_max = div_factor_fine(cache->length, 3619 bargs->usage_max); 3620 3621 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3622 ret = 0; 3623 3624 btrfs_put_block_group(cache); 3625 return ret; 3626 } 3627 3628 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3629 u64 chunk_offset, struct btrfs_balance_args *bargs) 3630 { 3631 struct btrfs_block_group *cache; 3632 u64 chunk_used, user_thresh; 3633 int ret = 1; 3634 3635 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3636 chunk_used = cache->used; 3637 3638 if (bargs->usage_min == 0) 3639 user_thresh = 1; 3640 else if (bargs->usage > 100) 3641 user_thresh = cache->length; 3642 else 3643 user_thresh = div_factor_fine(cache->length, bargs->usage); 3644 3645 if (chunk_used < user_thresh) 3646 ret = 0; 3647 3648 btrfs_put_block_group(cache); 3649 return ret; 3650 } 3651 3652 static int chunk_devid_filter(struct extent_buffer *leaf, 3653 struct btrfs_chunk *chunk, 3654 struct btrfs_balance_args *bargs) 3655 { 3656 struct btrfs_stripe *stripe; 3657 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3658 int i; 3659 3660 for (i = 0; i < num_stripes; i++) { 3661 stripe = btrfs_stripe_nr(chunk, i); 3662 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3663 return 0; 3664 } 3665 3666 return 1; 3667 } 3668 3669 static u64 calc_data_stripes(u64 type, int num_stripes) 3670 { 3671 const int index = btrfs_bg_flags_to_raid_index(type); 3672 const int ncopies = btrfs_raid_array[index].ncopies; 3673 const int nparity = btrfs_raid_array[index].nparity; 3674 3675 return (num_stripes - nparity) / ncopies; 3676 } 3677 3678 /* [pstart, pend) */ 3679 static int chunk_drange_filter(struct extent_buffer *leaf, 3680 struct btrfs_chunk *chunk, 3681 struct btrfs_balance_args *bargs) 3682 { 3683 struct btrfs_stripe *stripe; 3684 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3685 u64 stripe_offset; 3686 u64 stripe_length; 3687 u64 type; 3688 int factor; 3689 int i; 3690 3691 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3692 return 0; 3693 3694 type = btrfs_chunk_type(leaf, chunk); 3695 factor = calc_data_stripes(type, num_stripes); 3696 3697 for (i = 0; i < num_stripes; i++) { 3698 stripe = btrfs_stripe_nr(chunk, i); 3699 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3700 continue; 3701 3702 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3703 stripe_length = btrfs_chunk_length(leaf, chunk); 3704 stripe_length = div_u64(stripe_length, factor); 3705 3706 if (stripe_offset < bargs->pend && 3707 stripe_offset + stripe_length > bargs->pstart) 3708 return 0; 3709 } 3710 3711 return 1; 3712 } 3713 3714 /* [vstart, vend) */ 3715 static int chunk_vrange_filter(struct extent_buffer *leaf, 3716 struct btrfs_chunk *chunk, 3717 u64 chunk_offset, 3718 struct btrfs_balance_args *bargs) 3719 { 3720 if (chunk_offset < bargs->vend && 3721 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3722 /* at least part of the chunk is inside this vrange */ 3723 return 0; 3724 3725 return 1; 3726 } 3727 3728 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3729 struct btrfs_chunk *chunk, 3730 struct btrfs_balance_args *bargs) 3731 { 3732 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3733 3734 if (bargs->stripes_min <= num_stripes 3735 && num_stripes <= bargs->stripes_max) 3736 return 0; 3737 3738 return 1; 3739 } 3740 3741 static int chunk_soft_convert_filter(u64 chunk_type, 3742 struct btrfs_balance_args *bargs) 3743 { 3744 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3745 return 0; 3746 3747 chunk_type = chunk_to_extended(chunk_type) & 3748 BTRFS_EXTENDED_PROFILE_MASK; 3749 3750 if (bargs->target == chunk_type) 3751 return 1; 3752 3753 return 0; 3754 } 3755 3756 static int should_balance_chunk(struct extent_buffer *leaf, 3757 struct btrfs_chunk *chunk, u64 chunk_offset) 3758 { 3759 struct btrfs_fs_info *fs_info = leaf->fs_info; 3760 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3761 struct btrfs_balance_args *bargs = NULL; 3762 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3763 3764 /* type filter */ 3765 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3766 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3767 return 0; 3768 } 3769 3770 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3771 bargs = &bctl->data; 3772 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3773 bargs = &bctl->sys; 3774 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3775 bargs = &bctl->meta; 3776 3777 /* profiles filter */ 3778 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3779 chunk_profiles_filter(chunk_type, bargs)) { 3780 return 0; 3781 } 3782 3783 /* usage filter */ 3784 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3785 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3786 return 0; 3787 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3788 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3789 return 0; 3790 } 3791 3792 /* devid filter */ 3793 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3794 chunk_devid_filter(leaf, chunk, bargs)) { 3795 return 0; 3796 } 3797 3798 /* drange filter, makes sense only with devid filter */ 3799 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3800 chunk_drange_filter(leaf, chunk, bargs)) { 3801 return 0; 3802 } 3803 3804 /* vrange filter */ 3805 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3806 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3807 return 0; 3808 } 3809 3810 /* stripes filter */ 3811 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3812 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3813 return 0; 3814 } 3815 3816 /* soft profile changing mode */ 3817 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3818 chunk_soft_convert_filter(chunk_type, bargs)) { 3819 return 0; 3820 } 3821 3822 /* 3823 * limited by count, must be the last filter 3824 */ 3825 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3826 if (bargs->limit == 0) 3827 return 0; 3828 else 3829 bargs->limit--; 3830 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3831 /* 3832 * Same logic as the 'limit' filter; the minimum cannot be 3833 * determined here because we do not have the global information 3834 * about the count of all chunks that satisfy the filters. 3835 */ 3836 if (bargs->limit_max == 0) 3837 return 0; 3838 else 3839 bargs->limit_max--; 3840 } 3841 3842 return 1; 3843 } 3844 3845 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3846 { 3847 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3848 struct btrfs_root *chunk_root = fs_info->chunk_root; 3849 u64 chunk_type; 3850 struct btrfs_chunk *chunk; 3851 struct btrfs_path *path = NULL; 3852 struct btrfs_key key; 3853 struct btrfs_key found_key; 3854 struct extent_buffer *leaf; 3855 int slot; 3856 int ret; 3857 int enospc_errors = 0; 3858 bool counting = true; 3859 /* The single value limit and min/max limits use the same bytes in the */ 3860 u64 limit_data = bctl->data.limit; 3861 u64 limit_meta = bctl->meta.limit; 3862 u64 limit_sys = bctl->sys.limit; 3863 u32 count_data = 0; 3864 u32 count_meta = 0; 3865 u32 count_sys = 0; 3866 int chunk_reserved = 0; 3867 3868 path = btrfs_alloc_path(); 3869 if (!path) { 3870 ret = -ENOMEM; 3871 goto error; 3872 } 3873 3874 /* zero out stat counters */ 3875 spin_lock(&fs_info->balance_lock); 3876 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3877 spin_unlock(&fs_info->balance_lock); 3878 again: 3879 if (!counting) { 3880 /* 3881 * The single value limit and min/max limits use the same bytes 3882 * in the 3883 */ 3884 bctl->data.limit = limit_data; 3885 bctl->meta.limit = limit_meta; 3886 bctl->sys.limit = limit_sys; 3887 } 3888 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3889 key.offset = (u64)-1; 3890 key.type = BTRFS_CHUNK_ITEM_KEY; 3891 3892 while (1) { 3893 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3894 atomic_read(&fs_info->balance_cancel_req)) { 3895 ret = -ECANCELED; 3896 goto error; 3897 } 3898 3899 mutex_lock(&fs_info->reclaim_bgs_lock); 3900 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3901 if (ret < 0) { 3902 mutex_unlock(&fs_info->reclaim_bgs_lock); 3903 goto error; 3904 } 3905 3906 /* 3907 * this shouldn't happen, it means the last relocate 3908 * failed 3909 */ 3910 if (ret == 0) 3911 BUG(); /* FIXME break ? */ 3912 3913 ret = btrfs_previous_item(chunk_root, path, 0, 3914 BTRFS_CHUNK_ITEM_KEY); 3915 if (ret) { 3916 mutex_unlock(&fs_info->reclaim_bgs_lock); 3917 ret = 0; 3918 break; 3919 } 3920 3921 leaf = path->nodes[0]; 3922 slot = path->slots[0]; 3923 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3924 3925 if (found_key.objectid != key.objectid) { 3926 mutex_unlock(&fs_info->reclaim_bgs_lock); 3927 break; 3928 } 3929 3930 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3931 chunk_type = btrfs_chunk_type(leaf, chunk); 3932 3933 if (!counting) { 3934 spin_lock(&fs_info->balance_lock); 3935 bctl->stat.considered++; 3936 spin_unlock(&fs_info->balance_lock); 3937 } 3938 3939 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3940 3941 btrfs_release_path(path); 3942 if (!ret) { 3943 mutex_unlock(&fs_info->reclaim_bgs_lock); 3944 goto loop; 3945 } 3946 3947 if (counting) { 3948 mutex_unlock(&fs_info->reclaim_bgs_lock); 3949 spin_lock(&fs_info->balance_lock); 3950 bctl->stat.expected++; 3951 spin_unlock(&fs_info->balance_lock); 3952 3953 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3954 count_data++; 3955 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3956 count_sys++; 3957 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3958 count_meta++; 3959 3960 goto loop; 3961 } 3962 3963 /* 3964 * Apply limit_min filter, no need to check if the LIMITS 3965 * filter is used, limit_min is 0 by default 3966 */ 3967 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3968 count_data < bctl->data.limit_min) 3969 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3970 count_meta < bctl->meta.limit_min) 3971 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3972 count_sys < bctl->sys.limit_min)) { 3973 mutex_unlock(&fs_info->reclaim_bgs_lock); 3974 goto loop; 3975 } 3976 3977 if (!chunk_reserved) { 3978 /* 3979 * We may be relocating the only data chunk we have, 3980 * which could potentially end up with losing data's 3981 * raid profile, so lets allocate an empty one in 3982 * advance. 3983 */ 3984 ret = btrfs_may_alloc_data_chunk(fs_info, 3985 found_key.offset); 3986 if (ret < 0) { 3987 mutex_unlock(&fs_info->reclaim_bgs_lock); 3988 goto error; 3989 } else if (ret == 1) { 3990 chunk_reserved = 1; 3991 } 3992 } 3993 3994 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3995 mutex_unlock(&fs_info->reclaim_bgs_lock); 3996 if (ret == -ENOSPC) { 3997 enospc_errors++; 3998 } else if (ret == -ETXTBSY) { 3999 btrfs_info(fs_info, 4000 "skipping relocation of block group %llu due to active swapfile", 4001 found_key.offset); 4002 ret = 0; 4003 } else if (ret) { 4004 goto error; 4005 } else { 4006 spin_lock(&fs_info->balance_lock); 4007 bctl->stat.completed++; 4008 spin_unlock(&fs_info->balance_lock); 4009 } 4010 loop: 4011 if (found_key.offset == 0) 4012 break; 4013 key.offset = found_key.offset - 1; 4014 } 4015 4016 if (counting) { 4017 btrfs_release_path(path); 4018 counting = false; 4019 goto again; 4020 } 4021 error: 4022 btrfs_free_path(path); 4023 if (enospc_errors) { 4024 btrfs_info(fs_info, "%d enospc errors during balance", 4025 enospc_errors); 4026 if (!ret) 4027 ret = -ENOSPC; 4028 } 4029 4030 return ret; 4031 } 4032 4033 /** 4034 * alloc_profile_is_valid - see if a given profile is valid and reduced 4035 * @flags: profile to validate 4036 * @extended: if true @flags is treated as an extended profile 4037 */ 4038 static int alloc_profile_is_valid(u64 flags, int extended) 4039 { 4040 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 4041 BTRFS_BLOCK_GROUP_PROFILE_MASK); 4042 4043 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 4044 4045 /* 1) check that all other bits are zeroed */ 4046 if (flags & ~mask) 4047 return 0; 4048 4049 /* 2) see if profile is reduced */ 4050 if (flags == 0) 4051 return !extended; /* "0" is valid for usual profiles */ 4052 4053 return has_single_bit_set(flags); 4054 } 4055 4056 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4057 { 4058 /* cancel requested || normal exit path */ 4059 return atomic_read(&fs_info->balance_cancel_req) || 4060 (atomic_read(&fs_info->balance_pause_req) == 0 && 4061 atomic_read(&fs_info->balance_cancel_req) == 0); 4062 } 4063 4064 /* 4065 * Validate target profile against allowed profiles and return true if it's OK. 4066 * Otherwise print the error message and return false. 4067 */ 4068 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4069 const struct btrfs_balance_args *bargs, 4070 u64 allowed, const char *type) 4071 { 4072 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4073 return true; 4074 4075 /* Profile is valid and does not have bits outside of the allowed set */ 4076 if (alloc_profile_is_valid(bargs->target, 1) && 4077 (bargs->target & ~allowed) == 0) 4078 return true; 4079 4080 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4081 type, btrfs_bg_type_to_raid_name(bargs->target)); 4082 return false; 4083 } 4084 4085 /* 4086 * Fill @buf with textual description of balance filter flags @bargs, up to 4087 * @size_buf including the terminating null. The output may be trimmed if it 4088 * does not fit into the provided buffer. 4089 */ 4090 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4091 u32 size_buf) 4092 { 4093 int ret; 4094 u32 size_bp = size_buf; 4095 char *bp = buf; 4096 u64 flags = bargs->flags; 4097 char tmp_buf[128] = {'\0'}; 4098 4099 if (!flags) 4100 return; 4101 4102 #define CHECK_APPEND_NOARG(a) \ 4103 do { \ 4104 ret = snprintf(bp, size_bp, (a)); \ 4105 if (ret < 0 || ret >= size_bp) \ 4106 goto out_overflow; \ 4107 size_bp -= ret; \ 4108 bp += ret; \ 4109 } while (0) 4110 4111 #define CHECK_APPEND_1ARG(a, v1) \ 4112 do { \ 4113 ret = snprintf(bp, size_bp, (a), (v1)); \ 4114 if (ret < 0 || ret >= size_bp) \ 4115 goto out_overflow; \ 4116 size_bp -= ret; \ 4117 bp += ret; \ 4118 } while (0) 4119 4120 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4121 do { \ 4122 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4123 if (ret < 0 || ret >= size_bp) \ 4124 goto out_overflow; \ 4125 size_bp -= ret; \ 4126 bp += ret; \ 4127 } while (0) 4128 4129 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4130 CHECK_APPEND_1ARG("convert=%s,", 4131 btrfs_bg_type_to_raid_name(bargs->target)); 4132 4133 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4134 CHECK_APPEND_NOARG("soft,"); 4135 4136 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4137 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4138 sizeof(tmp_buf)); 4139 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4140 } 4141 4142 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4143 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4144 4145 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4146 CHECK_APPEND_2ARG("usage=%u..%u,", 4147 bargs->usage_min, bargs->usage_max); 4148 4149 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4150 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4151 4152 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4153 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4154 bargs->pstart, bargs->pend); 4155 4156 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4157 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4158 bargs->vstart, bargs->vend); 4159 4160 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4161 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4162 4163 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4164 CHECK_APPEND_2ARG("limit=%u..%u,", 4165 bargs->limit_min, bargs->limit_max); 4166 4167 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4168 CHECK_APPEND_2ARG("stripes=%u..%u,", 4169 bargs->stripes_min, bargs->stripes_max); 4170 4171 #undef CHECK_APPEND_2ARG 4172 #undef CHECK_APPEND_1ARG 4173 #undef CHECK_APPEND_NOARG 4174 4175 out_overflow: 4176 4177 if (size_bp < size_buf) 4178 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4179 else 4180 buf[0] = '\0'; 4181 } 4182 4183 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4184 { 4185 u32 size_buf = 1024; 4186 char tmp_buf[192] = {'\0'}; 4187 char *buf; 4188 char *bp; 4189 u32 size_bp = size_buf; 4190 int ret; 4191 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4192 4193 buf = kzalloc(size_buf, GFP_KERNEL); 4194 if (!buf) 4195 return; 4196 4197 bp = buf; 4198 4199 #define CHECK_APPEND_1ARG(a, v1) \ 4200 do { \ 4201 ret = snprintf(bp, size_bp, (a), (v1)); \ 4202 if (ret < 0 || ret >= size_bp) \ 4203 goto out_overflow; \ 4204 size_bp -= ret; \ 4205 bp += ret; \ 4206 } while (0) 4207 4208 if (bctl->flags & BTRFS_BALANCE_FORCE) 4209 CHECK_APPEND_1ARG("%s", "-f "); 4210 4211 if (bctl->flags & BTRFS_BALANCE_DATA) { 4212 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4213 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4214 } 4215 4216 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4217 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4218 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4219 } 4220 4221 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4222 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4223 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4224 } 4225 4226 #undef CHECK_APPEND_1ARG 4227 4228 out_overflow: 4229 4230 if (size_bp < size_buf) 4231 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4232 btrfs_info(fs_info, "balance: %s %s", 4233 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4234 "resume" : "start", buf); 4235 4236 kfree(buf); 4237 } 4238 4239 /* 4240 * Should be called with balance mutexe held 4241 */ 4242 int btrfs_balance(struct btrfs_fs_info *fs_info, 4243 struct btrfs_balance_control *bctl, 4244 struct btrfs_ioctl_balance_args *bargs) 4245 { 4246 u64 meta_target, data_target; 4247 u64 allowed; 4248 int mixed = 0; 4249 int ret; 4250 u64 num_devices; 4251 unsigned seq; 4252 bool reducing_redundancy; 4253 int i; 4254 4255 if (btrfs_fs_closing(fs_info) || 4256 atomic_read(&fs_info->balance_pause_req) || 4257 btrfs_should_cancel_balance(fs_info)) { 4258 ret = -EINVAL; 4259 goto out; 4260 } 4261 4262 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4263 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4264 mixed = 1; 4265 4266 /* 4267 * In case of mixed groups both data and meta should be picked, 4268 * and identical options should be given for both of them. 4269 */ 4270 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4271 if (mixed && (bctl->flags & allowed)) { 4272 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4273 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4274 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4275 btrfs_err(fs_info, 4276 "balance: mixed groups data and metadata options must be the same"); 4277 ret = -EINVAL; 4278 goto out; 4279 } 4280 } 4281 4282 /* 4283 * rw_devices will not change at the moment, device add/delete/replace 4284 * are exclusive 4285 */ 4286 num_devices = fs_info->fs_devices->rw_devices; 4287 4288 /* 4289 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4290 * special bit for it, to make it easier to distinguish. Thus we need 4291 * to set it manually, or balance would refuse the profile. 4292 */ 4293 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4294 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4295 if (num_devices >= btrfs_raid_array[i].devs_min) 4296 allowed |= btrfs_raid_array[i].bg_flag; 4297 4298 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4299 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4300 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4301 ret = -EINVAL; 4302 goto out; 4303 } 4304 4305 /* 4306 * Allow to reduce metadata or system integrity only if force set for 4307 * profiles with redundancy (copies, parity) 4308 */ 4309 allowed = 0; 4310 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4311 if (btrfs_raid_array[i].ncopies >= 2 || 4312 btrfs_raid_array[i].tolerated_failures >= 1) 4313 allowed |= btrfs_raid_array[i].bg_flag; 4314 } 4315 do { 4316 seq = read_seqbegin(&fs_info->profiles_lock); 4317 4318 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4319 (fs_info->avail_system_alloc_bits & allowed) && 4320 !(bctl->sys.target & allowed)) || 4321 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4322 (fs_info->avail_metadata_alloc_bits & allowed) && 4323 !(bctl->meta.target & allowed))) 4324 reducing_redundancy = true; 4325 else 4326 reducing_redundancy = false; 4327 4328 /* if we're not converting, the target field is uninitialized */ 4329 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4330 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4331 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4332 bctl->data.target : fs_info->avail_data_alloc_bits; 4333 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4334 4335 if (reducing_redundancy) { 4336 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4337 btrfs_info(fs_info, 4338 "balance: force reducing metadata redundancy"); 4339 } else { 4340 btrfs_err(fs_info, 4341 "balance: reduces metadata redundancy, use --force if you want this"); 4342 ret = -EINVAL; 4343 goto out; 4344 } 4345 } 4346 4347 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4348 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4349 btrfs_warn(fs_info, 4350 "balance: metadata profile %s has lower redundancy than data profile %s", 4351 btrfs_bg_type_to_raid_name(meta_target), 4352 btrfs_bg_type_to_raid_name(data_target)); 4353 } 4354 4355 ret = insert_balance_item(fs_info, bctl); 4356 if (ret && ret != -EEXIST) 4357 goto out; 4358 4359 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4360 BUG_ON(ret == -EEXIST); 4361 BUG_ON(fs_info->balance_ctl); 4362 spin_lock(&fs_info->balance_lock); 4363 fs_info->balance_ctl = bctl; 4364 spin_unlock(&fs_info->balance_lock); 4365 } else { 4366 BUG_ON(ret != -EEXIST); 4367 spin_lock(&fs_info->balance_lock); 4368 update_balance_args(bctl); 4369 spin_unlock(&fs_info->balance_lock); 4370 } 4371 4372 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4373 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4374 describe_balance_start_or_resume(fs_info); 4375 mutex_unlock(&fs_info->balance_mutex); 4376 4377 ret = __btrfs_balance(fs_info); 4378 4379 mutex_lock(&fs_info->balance_mutex); 4380 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4381 btrfs_info(fs_info, "balance: paused"); 4382 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4383 } 4384 /* 4385 * Balance can be canceled by: 4386 * 4387 * - Regular cancel request 4388 * Then ret == -ECANCELED and balance_cancel_req > 0 4389 * 4390 * - Fatal signal to "btrfs" process 4391 * Either the signal caught by wait_reserve_ticket() and callers 4392 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4393 * got -ECANCELED. 4394 * Either way, in this case balance_cancel_req = 0, and 4395 * ret == -EINTR or ret == -ECANCELED. 4396 * 4397 * So here we only check the return value to catch canceled balance. 4398 */ 4399 else if (ret == -ECANCELED || ret == -EINTR) 4400 btrfs_info(fs_info, "balance: canceled"); 4401 else 4402 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4403 4404 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4405 4406 if (bargs) { 4407 memset(bargs, 0, sizeof(*bargs)); 4408 btrfs_update_ioctl_balance_args(fs_info, bargs); 4409 } 4410 4411 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4412 balance_need_close(fs_info)) { 4413 reset_balance_state(fs_info); 4414 btrfs_exclop_finish(fs_info); 4415 } 4416 4417 wake_up(&fs_info->balance_wait_q); 4418 4419 return ret; 4420 out: 4421 if (bctl->flags & BTRFS_BALANCE_RESUME) 4422 reset_balance_state(fs_info); 4423 else 4424 kfree(bctl); 4425 btrfs_exclop_finish(fs_info); 4426 4427 return ret; 4428 } 4429 4430 static int balance_kthread(void *data) 4431 { 4432 struct btrfs_fs_info *fs_info = data; 4433 int ret = 0; 4434 4435 sb_start_write(fs_info->sb); 4436 mutex_lock(&fs_info->balance_mutex); 4437 if (fs_info->balance_ctl) 4438 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4439 mutex_unlock(&fs_info->balance_mutex); 4440 sb_end_write(fs_info->sb); 4441 4442 return ret; 4443 } 4444 4445 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4446 { 4447 struct task_struct *tsk; 4448 4449 mutex_lock(&fs_info->balance_mutex); 4450 if (!fs_info->balance_ctl) { 4451 mutex_unlock(&fs_info->balance_mutex); 4452 return 0; 4453 } 4454 mutex_unlock(&fs_info->balance_mutex); 4455 4456 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4457 btrfs_info(fs_info, "balance: resume skipped"); 4458 return 0; 4459 } 4460 4461 spin_lock(&fs_info->super_lock); 4462 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4463 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4464 spin_unlock(&fs_info->super_lock); 4465 /* 4466 * A ro->rw remount sequence should continue with the paused balance 4467 * regardless of who pauses it, system or the user as of now, so set 4468 * the resume flag. 4469 */ 4470 spin_lock(&fs_info->balance_lock); 4471 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4472 spin_unlock(&fs_info->balance_lock); 4473 4474 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4475 return PTR_ERR_OR_ZERO(tsk); 4476 } 4477 4478 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4479 { 4480 struct btrfs_balance_control *bctl; 4481 struct btrfs_balance_item *item; 4482 struct btrfs_disk_balance_args disk_bargs; 4483 struct btrfs_path *path; 4484 struct extent_buffer *leaf; 4485 struct btrfs_key key; 4486 int ret; 4487 4488 path = btrfs_alloc_path(); 4489 if (!path) 4490 return -ENOMEM; 4491 4492 key.objectid = BTRFS_BALANCE_OBJECTID; 4493 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4494 key.offset = 0; 4495 4496 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4497 if (ret < 0) 4498 goto out; 4499 if (ret > 0) { /* ret = -ENOENT; */ 4500 ret = 0; 4501 goto out; 4502 } 4503 4504 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4505 if (!bctl) { 4506 ret = -ENOMEM; 4507 goto out; 4508 } 4509 4510 leaf = path->nodes[0]; 4511 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4512 4513 bctl->flags = btrfs_balance_flags(leaf, item); 4514 bctl->flags |= BTRFS_BALANCE_RESUME; 4515 4516 btrfs_balance_data(leaf, item, &disk_bargs); 4517 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4518 btrfs_balance_meta(leaf, item, &disk_bargs); 4519 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4520 btrfs_balance_sys(leaf, item, &disk_bargs); 4521 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4522 4523 /* 4524 * This should never happen, as the paused balance state is recovered 4525 * during mount without any chance of other exclusive ops to collide. 4526 * 4527 * This gives the exclusive op status to balance and keeps in paused 4528 * state until user intervention (cancel or umount). If the ownership 4529 * cannot be assigned, show a message but do not fail. The balance 4530 * is in a paused state and must have fs_info::balance_ctl properly 4531 * set up. 4532 */ 4533 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4534 btrfs_warn(fs_info, 4535 "balance: cannot set exclusive op status, resume manually"); 4536 4537 btrfs_release_path(path); 4538 4539 mutex_lock(&fs_info->balance_mutex); 4540 BUG_ON(fs_info->balance_ctl); 4541 spin_lock(&fs_info->balance_lock); 4542 fs_info->balance_ctl = bctl; 4543 spin_unlock(&fs_info->balance_lock); 4544 mutex_unlock(&fs_info->balance_mutex); 4545 out: 4546 btrfs_free_path(path); 4547 return ret; 4548 } 4549 4550 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4551 { 4552 int ret = 0; 4553 4554 mutex_lock(&fs_info->balance_mutex); 4555 if (!fs_info->balance_ctl) { 4556 mutex_unlock(&fs_info->balance_mutex); 4557 return -ENOTCONN; 4558 } 4559 4560 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4561 atomic_inc(&fs_info->balance_pause_req); 4562 mutex_unlock(&fs_info->balance_mutex); 4563 4564 wait_event(fs_info->balance_wait_q, 4565 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4566 4567 mutex_lock(&fs_info->balance_mutex); 4568 /* we are good with balance_ctl ripped off from under us */ 4569 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4570 atomic_dec(&fs_info->balance_pause_req); 4571 } else { 4572 ret = -ENOTCONN; 4573 } 4574 4575 mutex_unlock(&fs_info->balance_mutex); 4576 return ret; 4577 } 4578 4579 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4580 { 4581 mutex_lock(&fs_info->balance_mutex); 4582 if (!fs_info->balance_ctl) { 4583 mutex_unlock(&fs_info->balance_mutex); 4584 return -ENOTCONN; 4585 } 4586 4587 /* 4588 * A paused balance with the item stored on disk can be resumed at 4589 * mount time if the mount is read-write. Otherwise it's still paused 4590 * and we must not allow cancelling as it deletes the item. 4591 */ 4592 if (sb_rdonly(fs_info->sb)) { 4593 mutex_unlock(&fs_info->balance_mutex); 4594 return -EROFS; 4595 } 4596 4597 atomic_inc(&fs_info->balance_cancel_req); 4598 /* 4599 * if we are running just wait and return, balance item is 4600 * deleted in btrfs_balance in this case 4601 */ 4602 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4603 mutex_unlock(&fs_info->balance_mutex); 4604 wait_event(fs_info->balance_wait_q, 4605 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4606 mutex_lock(&fs_info->balance_mutex); 4607 } else { 4608 mutex_unlock(&fs_info->balance_mutex); 4609 /* 4610 * Lock released to allow other waiters to continue, we'll 4611 * reexamine the status again. 4612 */ 4613 mutex_lock(&fs_info->balance_mutex); 4614 4615 if (fs_info->balance_ctl) { 4616 reset_balance_state(fs_info); 4617 btrfs_exclop_finish(fs_info); 4618 btrfs_info(fs_info, "balance: canceled"); 4619 } 4620 } 4621 4622 BUG_ON(fs_info->balance_ctl || 4623 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4624 atomic_dec(&fs_info->balance_cancel_req); 4625 mutex_unlock(&fs_info->balance_mutex); 4626 return 0; 4627 } 4628 4629 int btrfs_uuid_scan_kthread(void *data) 4630 { 4631 struct btrfs_fs_info *fs_info = data; 4632 struct btrfs_root *root = fs_info->tree_root; 4633 struct btrfs_key key; 4634 struct btrfs_path *path = NULL; 4635 int ret = 0; 4636 struct extent_buffer *eb; 4637 int slot; 4638 struct btrfs_root_item root_item; 4639 u32 item_size; 4640 struct btrfs_trans_handle *trans = NULL; 4641 bool closing = false; 4642 4643 path = btrfs_alloc_path(); 4644 if (!path) { 4645 ret = -ENOMEM; 4646 goto out; 4647 } 4648 4649 key.objectid = 0; 4650 key.type = BTRFS_ROOT_ITEM_KEY; 4651 key.offset = 0; 4652 4653 while (1) { 4654 if (btrfs_fs_closing(fs_info)) { 4655 closing = true; 4656 break; 4657 } 4658 ret = btrfs_search_forward(root, &key, path, 4659 BTRFS_OLDEST_GENERATION); 4660 if (ret) { 4661 if (ret > 0) 4662 ret = 0; 4663 break; 4664 } 4665 4666 if (key.type != BTRFS_ROOT_ITEM_KEY || 4667 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4668 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4669 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4670 goto skip; 4671 4672 eb = path->nodes[0]; 4673 slot = path->slots[0]; 4674 item_size = btrfs_item_size(eb, slot); 4675 if (item_size < sizeof(root_item)) 4676 goto skip; 4677 4678 read_extent_buffer(eb, &root_item, 4679 btrfs_item_ptr_offset(eb, slot), 4680 (int)sizeof(root_item)); 4681 if (btrfs_root_refs(&root_item) == 0) 4682 goto skip; 4683 4684 if (!btrfs_is_empty_uuid(root_item.uuid) || 4685 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4686 if (trans) 4687 goto update_tree; 4688 4689 btrfs_release_path(path); 4690 /* 4691 * 1 - subvol uuid item 4692 * 1 - received_subvol uuid item 4693 */ 4694 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4695 if (IS_ERR(trans)) { 4696 ret = PTR_ERR(trans); 4697 break; 4698 } 4699 continue; 4700 } else { 4701 goto skip; 4702 } 4703 update_tree: 4704 btrfs_release_path(path); 4705 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4706 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4707 BTRFS_UUID_KEY_SUBVOL, 4708 key.objectid); 4709 if (ret < 0) { 4710 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4711 ret); 4712 break; 4713 } 4714 } 4715 4716 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4717 ret = btrfs_uuid_tree_add(trans, 4718 root_item.received_uuid, 4719 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4720 key.objectid); 4721 if (ret < 0) { 4722 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4723 ret); 4724 break; 4725 } 4726 } 4727 4728 skip: 4729 btrfs_release_path(path); 4730 if (trans) { 4731 ret = btrfs_end_transaction(trans); 4732 trans = NULL; 4733 if (ret) 4734 break; 4735 } 4736 4737 if (key.offset < (u64)-1) { 4738 key.offset++; 4739 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4740 key.offset = 0; 4741 key.type = BTRFS_ROOT_ITEM_KEY; 4742 } else if (key.objectid < (u64)-1) { 4743 key.offset = 0; 4744 key.type = BTRFS_ROOT_ITEM_KEY; 4745 key.objectid++; 4746 } else { 4747 break; 4748 } 4749 cond_resched(); 4750 } 4751 4752 out: 4753 btrfs_free_path(path); 4754 if (trans && !IS_ERR(trans)) 4755 btrfs_end_transaction(trans); 4756 if (ret) 4757 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4758 else if (!closing) 4759 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4760 up(&fs_info->uuid_tree_rescan_sem); 4761 return 0; 4762 } 4763 4764 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4765 { 4766 struct btrfs_trans_handle *trans; 4767 struct btrfs_root *tree_root = fs_info->tree_root; 4768 struct btrfs_root *uuid_root; 4769 struct task_struct *task; 4770 int ret; 4771 4772 /* 4773 * 1 - root node 4774 * 1 - root item 4775 */ 4776 trans = btrfs_start_transaction(tree_root, 2); 4777 if (IS_ERR(trans)) 4778 return PTR_ERR(trans); 4779 4780 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4781 if (IS_ERR(uuid_root)) { 4782 ret = PTR_ERR(uuid_root); 4783 btrfs_abort_transaction(trans, ret); 4784 btrfs_end_transaction(trans); 4785 return ret; 4786 } 4787 4788 fs_info->uuid_root = uuid_root; 4789 4790 ret = btrfs_commit_transaction(trans); 4791 if (ret) 4792 return ret; 4793 4794 down(&fs_info->uuid_tree_rescan_sem); 4795 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4796 if (IS_ERR(task)) { 4797 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4798 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4799 up(&fs_info->uuid_tree_rescan_sem); 4800 return PTR_ERR(task); 4801 } 4802 4803 return 0; 4804 } 4805 4806 /* 4807 * shrinking a device means finding all of the device extents past 4808 * the new size, and then following the back refs to the chunks. 4809 * The chunk relocation code actually frees the device extent 4810 */ 4811 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4812 { 4813 struct btrfs_fs_info *fs_info = device->fs_info; 4814 struct btrfs_root *root = fs_info->dev_root; 4815 struct btrfs_trans_handle *trans; 4816 struct btrfs_dev_extent *dev_extent = NULL; 4817 struct btrfs_path *path; 4818 u64 length; 4819 u64 chunk_offset; 4820 int ret; 4821 int slot; 4822 int failed = 0; 4823 bool retried = false; 4824 struct extent_buffer *l; 4825 struct btrfs_key key; 4826 struct btrfs_super_block *super_copy = fs_info->super_copy; 4827 u64 old_total = btrfs_super_total_bytes(super_copy); 4828 u64 old_size = btrfs_device_get_total_bytes(device); 4829 u64 diff; 4830 u64 start; 4831 4832 new_size = round_down(new_size, fs_info->sectorsize); 4833 start = new_size; 4834 diff = round_down(old_size - new_size, fs_info->sectorsize); 4835 4836 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4837 return -EINVAL; 4838 4839 path = btrfs_alloc_path(); 4840 if (!path) 4841 return -ENOMEM; 4842 4843 path->reada = READA_BACK; 4844 4845 trans = btrfs_start_transaction(root, 0); 4846 if (IS_ERR(trans)) { 4847 btrfs_free_path(path); 4848 return PTR_ERR(trans); 4849 } 4850 4851 mutex_lock(&fs_info->chunk_mutex); 4852 4853 btrfs_device_set_total_bytes(device, new_size); 4854 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4855 device->fs_devices->total_rw_bytes -= diff; 4856 atomic64_sub(diff, &fs_info->free_chunk_space); 4857 } 4858 4859 /* 4860 * Once the device's size has been set to the new size, ensure all 4861 * in-memory chunks are synced to disk so that the loop below sees them 4862 * and relocates them accordingly. 4863 */ 4864 if (contains_pending_extent(device, &start, diff)) { 4865 mutex_unlock(&fs_info->chunk_mutex); 4866 ret = btrfs_commit_transaction(trans); 4867 if (ret) 4868 goto done; 4869 } else { 4870 mutex_unlock(&fs_info->chunk_mutex); 4871 btrfs_end_transaction(trans); 4872 } 4873 4874 again: 4875 key.objectid = device->devid; 4876 key.offset = (u64)-1; 4877 key.type = BTRFS_DEV_EXTENT_KEY; 4878 4879 do { 4880 mutex_lock(&fs_info->reclaim_bgs_lock); 4881 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4882 if (ret < 0) { 4883 mutex_unlock(&fs_info->reclaim_bgs_lock); 4884 goto done; 4885 } 4886 4887 ret = btrfs_previous_item(root, path, 0, key.type); 4888 if (ret) { 4889 mutex_unlock(&fs_info->reclaim_bgs_lock); 4890 if (ret < 0) 4891 goto done; 4892 ret = 0; 4893 btrfs_release_path(path); 4894 break; 4895 } 4896 4897 l = path->nodes[0]; 4898 slot = path->slots[0]; 4899 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4900 4901 if (key.objectid != device->devid) { 4902 mutex_unlock(&fs_info->reclaim_bgs_lock); 4903 btrfs_release_path(path); 4904 break; 4905 } 4906 4907 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4908 length = btrfs_dev_extent_length(l, dev_extent); 4909 4910 if (key.offset + length <= new_size) { 4911 mutex_unlock(&fs_info->reclaim_bgs_lock); 4912 btrfs_release_path(path); 4913 break; 4914 } 4915 4916 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4917 btrfs_release_path(path); 4918 4919 /* 4920 * We may be relocating the only data chunk we have, 4921 * which could potentially end up with losing data's 4922 * raid profile, so lets allocate an empty one in 4923 * advance. 4924 */ 4925 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4926 if (ret < 0) { 4927 mutex_unlock(&fs_info->reclaim_bgs_lock); 4928 goto done; 4929 } 4930 4931 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4932 mutex_unlock(&fs_info->reclaim_bgs_lock); 4933 if (ret == -ENOSPC) { 4934 failed++; 4935 } else if (ret) { 4936 if (ret == -ETXTBSY) { 4937 btrfs_warn(fs_info, 4938 "could not shrink block group %llu due to active swapfile", 4939 chunk_offset); 4940 } 4941 goto done; 4942 } 4943 } while (key.offset-- > 0); 4944 4945 if (failed && !retried) { 4946 failed = 0; 4947 retried = true; 4948 goto again; 4949 } else if (failed && retried) { 4950 ret = -ENOSPC; 4951 goto done; 4952 } 4953 4954 /* Shrinking succeeded, else we would be at "done". */ 4955 trans = btrfs_start_transaction(root, 0); 4956 if (IS_ERR(trans)) { 4957 ret = PTR_ERR(trans); 4958 goto done; 4959 } 4960 4961 mutex_lock(&fs_info->chunk_mutex); 4962 /* Clear all state bits beyond the shrunk device size */ 4963 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4964 CHUNK_STATE_MASK); 4965 4966 btrfs_device_set_disk_total_bytes(device, new_size); 4967 if (list_empty(&device->post_commit_list)) 4968 list_add_tail(&device->post_commit_list, 4969 &trans->transaction->dev_update_list); 4970 4971 WARN_ON(diff > old_total); 4972 btrfs_set_super_total_bytes(super_copy, 4973 round_down(old_total - diff, fs_info->sectorsize)); 4974 mutex_unlock(&fs_info->chunk_mutex); 4975 4976 btrfs_reserve_chunk_metadata(trans, false); 4977 /* Now btrfs_update_device() will change the on-disk size. */ 4978 ret = btrfs_update_device(trans, device); 4979 btrfs_trans_release_chunk_metadata(trans); 4980 if (ret < 0) { 4981 btrfs_abort_transaction(trans, ret); 4982 btrfs_end_transaction(trans); 4983 } else { 4984 ret = btrfs_commit_transaction(trans); 4985 } 4986 done: 4987 btrfs_free_path(path); 4988 if (ret) { 4989 mutex_lock(&fs_info->chunk_mutex); 4990 btrfs_device_set_total_bytes(device, old_size); 4991 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 4992 device->fs_devices->total_rw_bytes += diff; 4993 atomic64_add(diff, &fs_info->free_chunk_space); 4994 mutex_unlock(&fs_info->chunk_mutex); 4995 } 4996 return ret; 4997 } 4998 4999 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 5000 struct btrfs_key *key, 5001 struct btrfs_chunk *chunk, int item_size) 5002 { 5003 struct btrfs_super_block *super_copy = fs_info->super_copy; 5004 struct btrfs_disk_key disk_key; 5005 u32 array_size; 5006 u8 *ptr; 5007 5008 lockdep_assert_held(&fs_info->chunk_mutex); 5009 5010 array_size = btrfs_super_sys_array_size(super_copy); 5011 if (array_size + item_size + sizeof(disk_key) 5012 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5013 return -EFBIG; 5014 5015 ptr = super_copy->sys_chunk_array + array_size; 5016 btrfs_cpu_key_to_disk(&disk_key, key); 5017 memcpy(ptr, &disk_key, sizeof(disk_key)); 5018 ptr += sizeof(disk_key); 5019 memcpy(ptr, chunk, item_size); 5020 item_size += sizeof(disk_key); 5021 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5022 5023 return 0; 5024 } 5025 5026 /* 5027 * sort the devices in descending order by max_avail, total_avail 5028 */ 5029 static int btrfs_cmp_device_info(const void *a, const void *b) 5030 { 5031 const struct btrfs_device_info *di_a = a; 5032 const struct btrfs_device_info *di_b = b; 5033 5034 if (di_a->max_avail > di_b->max_avail) 5035 return -1; 5036 if (di_a->max_avail < di_b->max_avail) 5037 return 1; 5038 if (di_a->total_avail > di_b->total_avail) 5039 return -1; 5040 if (di_a->total_avail < di_b->total_avail) 5041 return 1; 5042 return 0; 5043 } 5044 5045 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5046 { 5047 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5048 return; 5049 5050 btrfs_set_fs_incompat(info, RAID56); 5051 } 5052 5053 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5054 { 5055 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5056 return; 5057 5058 btrfs_set_fs_incompat(info, RAID1C34); 5059 } 5060 5061 /* 5062 * Structure used internally for btrfs_create_chunk() function. 5063 * Wraps needed parameters. 5064 */ 5065 struct alloc_chunk_ctl { 5066 u64 start; 5067 u64 type; 5068 /* Total number of stripes to allocate */ 5069 int num_stripes; 5070 /* sub_stripes info for map */ 5071 int sub_stripes; 5072 /* Stripes per device */ 5073 int dev_stripes; 5074 /* Maximum number of devices to use */ 5075 int devs_max; 5076 /* Minimum number of devices to use */ 5077 int devs_min; 5078 /* ndevs has to be a multiple of this */ 5079 int devs_increment; 5080 /* Number of copies */ 5081 int ncopies; 5082 /* Number of stripes worth of bytes to store parity information */ 5083 int nparity; 5084 u64 max_stripe_size; 5085 u64 max_chunk_size; 5086 u64 dev_extent_min; 5087 u64 stripe_size; 5088 u64 chunk_size; 5089 int ndevs; 5090 }; 5091 5092 static void init_alloc_chunk_ctl_policy_regular( 5093 struct btrfs_fs_devices *fs_devices, 5094 struct alloc_chunk_ctl *ctl) 5095 { 5096 struct btrfs_space_info *space_info; 5097 5098 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); 5099 ASSERT(space_info); 5100 5101 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); 5102 ctl->max_stripe_size = ctl->max_chunk_size; 5103 5104 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) 5105 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); 5106 5107 /* We don't want a chunk larger than 10% of writable space */ 5108 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5109 ctl->max_chunk_size); 5110 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5111 } 5112 5113 static void init_alloc_chunk_ctl_policy_zoned( 5114 struct btrfs_fs_devices *fs_devices, 5115 struct alloc_chunk_ctl *ctl) 5116 { 5117 u64 zone_size = fs_devices->fs_info->zone_size; 5118 u64 limit; 5119 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5120 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5121 u64 min_chunk_size = min_data_stripes * zone_size; 5122 u64 type = ctl->type; 5123 5124 ctl->max_stripe_size = zone_size; 5125 if (type & BTRFS_BLOCK_GROUP_DATA) { 5126 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5127 zone_size); 5128 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5129 ctl->max_chunk_size = ctl->max_stripe_size; 5130 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5131 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5132 ctl->devs_max = min_t(int, ctl->devs_max, 5133 BTRFS_MAX_DEVS_SYS_CHUNK); 5134 } else { 5135 BUG(); 5136 } 5137 5138 /* We don't want a chunk larger than 10% of writable space */ 5139 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5140 zone_size), 5141 min_chunk_size); 5142 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5143 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5144 } 5145 5146 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5147 struct alloc_chunk_ctl *ctl) 5148 { 5149 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5150 5151 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5152 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5153 ctl->devs_max = btrfs_raid_array[index].devs_max; 5154 if (!ctl->devs_max) 5155 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5156 ctl->devs_min = btrfs_raid_array[index].devs_min; 5157 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5158 ctl->ncopies = btrfs_raid_array[index].ncopies; 5159 ctl->nparity = btrfs_raid_array[index].nparity; 5160 ctl->ndevs = 0; 5161 5162 switch (fs_devices->chunk_alloc_policy) { 5163 case BTRFS_CHUNK_ALLOC_REGULAR: 5164 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5165 break; 5166 case BTRFS_CHUNK_ALLOC_ZONED: 5167 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5168 break; 5169 default: 5170 BUG(); 5171 } 5172 } 5173 5174 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5175 struct alloc_chunk_ctl *ctl, 5176 struct btrfs_device_info *devices_info) 5177 { 5178 struct btrfs_fs_info *info = fs_devices->fs_info; 5179 struct btrfs_device *device; 5180 u64 total_avail; 5181 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5182 int ret; 5183 int ndevs = 0; 5184 u64 max_avail; 5185 u64 dev_offset; 5186 5187 /* 5188 * in the first pass through the devices list, we gather information 5189 * about the available holes on each device. 5190 */ 5191 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5192 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5193 WARN(1, KERN_ERR 5194 "BTRFS: read-only device in alloc_list\n"); 5195 continue; 5196 } 5197 5198 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5199 &device->dev_state) || 5200 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5201 continue; 5202 5203 if (device->total_bytes > device->bytes_used) 5204 total_avail = device->total_bytes - device->bytes_used; 5205 else 5206 total_avail = 0; 5207 5208 /* If there is no space on this device, skip it. */ 5209 if (total_avail < ctl->dev_extent_min) 5210 continue; 5211 5212 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5213 &max_avail); 5214 if (ret && ret != -ENOSPC) 5215 return ret; 5216 5217 if (ret == 0) 5218 max_avail = dev_extent_want; 5219 5220 if (max_avail < ctl->dev_extent_min) { 5221 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5222 btrfs_debug(info, 5223 "%s: devid %llu has no free space, have=%llu want=%llu", 5224 __func__, device->devid, max_avail, 5225 ctl->dev_extent_min); 5226 continue; 5227 } 5228 5229 if (ndevs == fs_devices->rw_devices) { 5230 WARN(1, "%s: found more than %llu devices\n", 5231 __func__, fs_devices->rw_devices); 5232 break; 5233 } 5234 devices_info[ndevs].dev_offset = dev_offset; 5235 devices_info[ndevs].max_avail = max_avail; 5236 devices_info[ndevs].total_avail = total_avail; 5237 devices_info[ndevs].dev = device; 5238 ++ndevs; 5239 } 5240 ctl->ndevs = ndevs; 5241 5242 /* 5243 * now sort the devices by hole size / available space 5244 */ 5245 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5246 btrfs_cmp_device_info, NULL); 5247 5248 return 0; 5249 } 5250 5251 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5252 struct btrfs_device_info *devices_info) 5253 { 5254 /* Number of stripes that count for block group size */ 5255 int data_stripes; 5256 5257 /* 5258 * The primary goal is to maximize the number of stripes, so use as 5259 * many devices as possible, even if the stripes are not maximum sized. 5260 * 5261 * The DUP profile stores more than one stripe per device, the 5262 * max_avail is the total size so we have to adjust. 5263 */ 5264 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5265 ctl->dev_stripes); 5266 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5267 5268 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5269 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5270 5271 /* 5272 * Use the number of data stripes to figure out how big this chunk is 5273 * really going to be in terms of logical address space, and compare 5274 * that answer with the max chunk size. If it's higher, we try to 5275 * reduce stripe_size. 5276 */ 5277 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5278 /* 5279 * Reduce stripe_size, round it up to a 16MB boundary again and 5280 * then use it, unless it ends up being even bigger than the 5281 * previous value we had already. 5282 */ 5283 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5284 data_stripes), SZ_16M), 5285 ctl->stripe_size); 5286 } 5287 5288 /* Stripe size should not go beyond 1G. */ 5289 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); 5290 5291 /* Align to BTRFS_STRIPE_LEN */ 5292 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5293 ctl->chunk_size = ctl->stripe_size * data_stripes; 5294 5295 return 0; 5296 } 5297 5298 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5299 struct btrfs_device_info *devices_info) 5300 { 5301 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5302 /* Number of stripes that count for block group size */ 5303 int data_stripes; 5304 5305 /* 5306 * It should hold because: 5307 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5308 */ 5309 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5310 5311 ctl->stripe_size = zone_size; 5312 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5313 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5314 5315 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5316 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5317 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5318 ctl->stripe_size) + ctl->nparity, 5319 ctl->dev_stripes); 5320 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5321 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5322 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5323 } 5324 5325 ctl->chunk_size = ctl->stripe_size * data_stripes; 5326 5327 return 0; 5328 } 5329 5330 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5331 struct alloc_chunk_ctl *ctl, 5332 struct btrfs_device_info *devices_info) 5333 { 5334 struct btrfs_fs_info *info = fs_devices->fs_info; 5335 5336 /* 5337 * Round down to number of usable stripes, devs_increment can be any 5338 * number so we can't use round_down() that requires power of 2, while 5339 * rounddown is safe. 5340 */ 5341 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5342 5343 if (ctl->ndevs < ctl->devs_min) { 5344 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5345 btrfs_debug(info, 5346 "%s: not enough devices with free space: have=%d minimum required=%d", 5347 __func__, ctl->ndevs, ctl->devs_min); 5348 } 5349 return -ENOSPC; 5350 } 5351 5352 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5353 5354 switch (fs_devices->chunk_alloc_policy) { 5355 case BTRFS_CHUNK_ALLOC_REGULAR: 5356 return decide_stripe_size_regular(ctl, devices_info); 5357 case BTRFS_CHUNK_ALLOC_ZONED: 5358 return decide_stripe_size_zoned(ctl, devices_info); 5359 default: 5360 BUG(); 5361 } 5362 } 5363 5364 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5365 struct alloc_chunk_ctl *ctl, 5366 struct btrfs_device_info *devices_info) 5367 { 5368 struct btrfs_fs_info *info = trans->fs_info; 5369 struct map_lookup *map = NULL; 5370 struct extent_map_tree *em_tree; 5371 struct btrfs_block_group *block_group; 5372 struct extent_map *em; 5373 u64 start = ctl->start; 5374 u64 type = ctl->type; 5375 int ret; 5376 int i; 5377 int j; 5378 5379 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5380 if (!map) 5381 return ERR_PTR(-ENOMEM); 5382 map->num_stripes = ctl->num_stripes; 5383 5384 for (i = 0; i < ctl->ndevs; ++i) { 5385 for (j = 0; j < ctl->dev_stripes; ++j) { 5386 int s = i * ctl->dev_stripes + j; 5387 map->stripes[s].dev = devices_info[i].dev; 5388 map->stripes[s].physical = devices_info[i].dev_offset + 5389 j * ctl->stripe_size; 5390 } 5391 } 5392 map->stripe_len = BTRFS_STRIPE_LEN; 5393 map->io_align = BTRFS_STRIPE_LEN; 5394 map->io_width = BTRFS_STRIPE_LEN; 5395 map->type = type; 5396 map->sub_stripes = ctl->sub_stripes; 5397 5398 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5399 5400 em = alloc_extent_map(); 5401 if (!em) { 5402 kfree(map); 5403 return ERR_PTR(-ENOMEM); 5404 } 5405 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5406 em->map_lookup = map; 5407 em->start = start; 5408 em->len = ctl->chunk_size; 5409 em->block_start = 0; 5410 em->block_len = em->len; 5411 em->orig_block_len = ctl->stripe_size; 5412 5413 em_tree = &info->mapping_tree; 5414 write_lock(&em_tree->lock); 5415 ret = add_extent_mapping(em_tree, em, 0); 5416 if (ret) { 5417 write_unlock(&em_tree->lock); 5418 free_extent_map(em); 5419 return ERR_PTR(ret); 5420 } 5421 write_unlock(&em_tree->lock); 5422 5423 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5424 if (IS_ERR(block_group)) 5425 goto error_del_extent; 5426 5427 for (i = 0; i < map->num_stripes; i++) { 5428 struct btrfs_device *dev = map->stripes[i].dev; 5429 5430 btrfs_device_set_bytes_used(dev, 5431 dev->bytes_used + ctl->stripe_size); 5432 if (list_empty(&dev->post_commit_list)) 5433 list_add_tail(&dev->post_commit_list, 5434 &trans->transaction->dev_update_list); 5435 } 5436 5437 atomic64_sub(ctl->stripe_size * map->num_stripes, 5438 &info->free_chunk_space); 5439 5440 free_extent_map(em); 5441 check_raid56_incompat_flag(info, type); 5442 check_raid1c34_incompat_flag(info, type); 5443 5444 return block_group; 5445 5446 error_del_extent: 5447 write_lock(&em_tree->lock); 5448 remove_extent_mapping(em_tree, em); 5449 write_unlock(&em_tree->lock); 5450 5451 /* One for our allocation */ 5452 free_extent_map(em); 5453 /* One for the tree reference */ 5454 free_extent_map(em); 5455 5456 return block_group; 5457 } 5458 5459 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5460 u64 type) 5461 { 5462 struct btrfs_fs_info *info = trans->fs_info; 5463 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5464 struct btrfs_device_info *devices_info = NULL; 5465 struct alloc_chunk_ctl ctl; 5466 struct btrfs_block_group *block_group; 5467 int ret; 5468 5469 lockdep_assert_held(&info->chunk_mutex); 5470 5471 if (!alloc_profile_is_valid(type, 0)) { 5472 ASSERT(0); 5473 return ERR_PTR(-EINVAL); 5474 } 5475 5476 if (list_empty(&fs_devices->alloc_list)) { 5477 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5478 btrfs_debug(info, "%s: no writable device", __func__); 5479 return ERR_PTR(-ENOSPC); 5480 } 5481 5482 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5483 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5484 ASSERT(0); 5485 return ERR_PTR(-EINVAL); 5486 } 5487 5488 ctl.start = find_next_chunk(info); 5489 ctl.type = type; 5490 init_alloc_chunk_ctl(fs_devices, &ctl); 5491 5492 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5493 GFP_NOFS); 5494 if (!devices_info) 5495 return ERR_PTR(-ENOMEM); 5496 5497 ret = gather_device_info(fs_devices, &ctl, devices_info); 5498 if (ret < 0) { 5499 block_group = ERR_PTR(ret); 5500 goto out; 5501 } 5502 5503 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5504 if (ret < 0) { 5505 block_group = ERR_PTR(ret); 5506 goto out; 5507 } 5508 5509 block_group = create_chunk(trans, &ctl, devices_info); 5510 5511 out: 5512 kfree(devices_info); 5513 return block_group; 5514 } 5515 5516 /* 5517 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5518 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5519 * chunks. 5520 * 5521 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5522 * phases. 5523 */ 5524 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5525 struct btrfs_block_group *bg) 5526 { 5527 struct btrfs_fs_info *fs_info = trans->fs_info; 5528 struct btrfs_root *chunk_root = fs_info->chunk_root; 5529 struct btrfs_key key; 5530 struct btrfs_chunk *chunk; 5531 struct btrfs_stripe *stripe; 5532 struct extent_map *em; 5533 struct map_lookup *map; 5534 size_t item_size; 5535 int i; 5536 int ret; 5537 5538 /* 5539 * We take the chunk_mutex for 2 reasons: 5540 * 5541 * 1) Updates and insertions in the chunk btree must be done while holding 5542 * the chunk_mutex, as well as updating the system chunk array in the 5543 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5544 * details; 5545 * 5546 * 2) To prevent races with the final phase of a device replace operation 5547 * that replaces the device object associated with the map's stripes, 5548 * because the device object's id can change at any time during that 5549 * final phase of the device replace operation 5550 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5551 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5552 * which would cause a failure when updating the device item, which does 5553 * not exists, or persisting a stripe of the chunk item with such ID. 5554 * Here we can't use the device_list_mutex because our caller already 5555 * has locked the chunk_mutex, and the final phase of device replace 5556 * acquires both mutexes - first the device_list_mutex and then the 5557 * chunk_mutex. Using any of those two mutexes protects us from a 5558 * concurrent device replace. 5559 */ 5560 lockdep_assert_held(&fs_info->chunk_mutex); 5561 5562 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5563 if (IS_ERR(em)) { 5564 ret = PTR_ERR(em); 5565 btrfs_abort_transaction(trans, ret); 5566 return ret; 5567 } 5568 5569 map = em->map_lookup; 5570 item_size = btrfs_chunk_item_size(map->num_stripes); 5571 5572 chunk = kzalloc(item_size, GFP_NOFS); 5573 if (!chunk) { 5574 ret = -ENOMEM; 5575 btrfs_abort_transaction(trans, ret); 5576 goto out; 5577 } 5578 5579 for (i = 0; i < map->num_stripes; i++) { 5580 struct btrfs_device *device = map->stripes[i].dev; 5581 5582 ret = btrfs_update_device(trans, device); 5583 if (ret) 5584 goto out; 5585 } 5586 5587 stripe = &chunk->stripe; 5588 for (i = 0; i < map->num_stripes; i++) { 5589 struct btrfs_device *device = map->stripes[i].dev; 5590 const u64 dev_offset = map->stripes[i].physical; 5591 5592 btrfs_set_stack_stripe_devid(stripe, device->devid); 5593 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5594 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5595 stripe++; 5596 } 5597 5598 btrfs_set_stack_chunk_length(chunk, bg->length); 5599 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5600 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5601 btrfs_set_stack_chunk_type(chunk, map->type); 5602 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5603 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5604 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5605 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5606 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5607 5608 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5609 key.type = BTRFS_CHUNK_ITEM_KEY; 5610 key.offset = bg->start; 5611 5612 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5613 if (ret) 5614 goto out; 5615 5616 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); 5617 5618 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5619 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5620 if (ret) 5621 goto out; 5622 } 5623 5624 out: 5625 kfree(chunk); 5626 free_extent_map(em); 5627 return ret; 5628 } 5629 5630 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5631 { 5632 struct btrfs_fs_info *fs_info = trans->fs_info; 5633 u64 alloc_profile; 5634 struct btrfs_block_group *meta_bg; 5635 struct btrfs_block_group *sys_bg; 5636 5637 /* 5638 * When adding a new device for sprouting, the seed device is read-only 5639 * so we must first allocate a metadata and a system chunk. But before 5640 * adding the block group items to the extent, device and chunk btrees, 5641 * we must first: 5642 * 5643 * 1) Create both chunks without doing any changes to the btrees, as 5644 * otherwise we would get -ENOSPC since the block groups from the 5645 * seed device are read-only; 5646 * 5647 * 2) Add the device item for the new sprout device - finishing the setup 5648 * of a new block group requires updating the device item in the chunk 5649 * btree, so it must exist when we attempt to do it. The previous step 5650 * ensures this does not fail with -ENOSPC. 5651 * 5652 * After that we can add the block group items to their btrees: 5653 * update existing device item in the chunk btree, add a new block group 5654 * item to the extent btree, add a new chunk item to the chunk btree and 5655 * finally add the new device extent items to the devices btree. 5656 */ 5657 5658 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5659 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5660 if (IS_ERR(meta_bg)) 5661 return PTR_ERR(meta_bg); 5662 5663 alloc_profile = btrfs_system_alloc_profile(fs_info); 5664 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5665 if (IS_ERR(sys_bg)) 5666 return PTR_ERR(sys_bg); 5667 5668 return 0; 5669 } 5670 5671 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5672 { 5673 const int index = btrfs_bg_flags_to_raid_index(map->type); 5674 5675 return btrfs_raid_array[index].tolerated_failures; 5676 } 5677 5678 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5679 { 5680 struct extent_map *em; 5681 struct map_lookup *map; 5682 int miss_ndevs = 0; 5683 int i; 5684 bool ret = true; 5685 5686 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5687 if (IS_ERR(em)) 5688 return false; 5689 5690 map = em->map_lookup; 5691 for (i = 0; i < map->num_stripes; i++) { 5692 if (test_bit(BTRFS_DEV_STATE_MISSING, 5693 &map->stripes[i].dev->dev_state)) { 5694 miss_ndevs++; 5695 continue; 5696 } 5697 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5698 &map->stripes[i].dev->dev_state)) { 5699 ret = false; 5700 goto end; 5701 } 5702 } 5703 5704 /* 5705 * If the number of missing devices is larger than max errors, we can 5706 * not write the data into that chunk successfully. 5707 */ 5708 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5709 ret = false; 5710 end: 5711 free_extent_map(em); 5712 return ret; 5713 } 5714 5715 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5716 { 5717 struct extent_map *em; 5718 5719 while (1) { 5720 write_lock(&tree->lock); 5721 em = lookup_extent_mapping(tree, 0, (u64)-1); 5722 if (em) 5723 remove_extent_mapping(tree, em); 5724 write_unlock(&tree->lock); 5725 if (!em) 5726 break; 5727 /* once for us */ 5728 free_extent_map(em); 5729 /* once for the tree */ 5730 free_extent_map(em); 5731 } 5732 } 5733 5734 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5735 { 5736 struct extent_map *em; 5737 struct map_lookup *map; 5738 enum btrfs_raid_types index; 5739 int ret = 1; 5740 5741 em = btrfs_get_chunk_map(fs_info, logical, len); 5742 if (IS_ERR(em)) 5743 /* 5744 * We could return errors for these cases, but that could get 5745 * ugly and we'd probably do the same thing which is just not do 5746 * anything else and exit, so return 1 so the callers don't try 5747 * to use other copies. 5748 */ 5749 return 1; 5750 5751 map = em->map_lookup; 5752 index = btrfs_bg_flags_to_raid_index(map->type); 5753 5754 /* Non-RAID56, use their ncopies from btrfs_raid_array. */ 5755 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5756 ret = btrfs_raid_array[index].ncopies; 5757 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5758 ret = 2; 5759 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5760 /* 5761 * There could be two corrupted data stripes, we need 5762 * to loop retry in order to rebuild the correct data. 5763 * 5764 * Fail a stripe at a time on every retry except the 5765 * stripe under reconstruction. 5766 */ 5767 ret = map->num_stripes; 5768 free_extent_map(em); 5769 5770 down_read(&fs_info->dev_replace.rwsem); 5771 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5772 fs_info->dev_replace.tgtdev) 5773 ret++; 5774 up_read(&fs_info->dev_replace.rwsem); 5775 5776 return ret; 5777 } 5778 5779 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5780 u64 logical) 5781 { 5782 struct extent_map *em; 5783 struct map_lookup *map; 5784 unsigned long len = fs_info->sectorsize; 5785 5786 if (!btrfs_fs_incompat(fs_info, RAID56)) 5787 return len; 5788 5789 em = btrfs_get_chunk_map(fs_info, logical, len); 5790 5791 if (!WARN_ON(IS_ERR(em))) { 5792 map = em->map_lookup; 5793 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5794 len = map->stripe_len * nr_data_stripes(map); 5795 free_extent_map(em); 5796 } 5797 return len; 5798 } 5799 5800 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5801 { 5802 struct extent_map *em; 5803 struct map_lookup *map; 5804 int ret = 0; 5805 5806 if (!btrfs_fs_incompat(fs_info, RAID56)) 5807 return 0; 5808 5809 em = btrfs_get_chunk_map(fs_info, logical, len); 5810 5811 if(!WARN_ON(IS_ERR(em))) { 5812 map = em->map_lookup; 5813 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5814 ret = 1; 5815 free_extent_map(em); 5816 } 5817 return ret; 5818 } 5819 5820 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5821 struct map_lookup *map, int first, 5822 int dev_replace_is_ongoing) 5823 { 5824 int i; 5825 int num_stripes; 5826 int preferred_mirror; 5827 int tolerance; 5828 struct btrfs_device *srcdev; 5829 5830 ASSERT((map->type & 5831 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5832 5833 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5834 num_stripes = map->sub_stripes; 5835 else 5836 num_stripes = map->num_stripes; 5837 5838 switch (fs_info->fs_devices->read_policy) { 5839 default: 5840 /* Shouldn't happen, just warn and use pid instead of failing */ 5841 btrfs_warn_rl(fs_info, 5842 "unknown read_policy type %u, reset to pid", 5843 fs_info->fs_devices->read_policy); 5844 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5845 fallthrough; 5846 case BTRFS_READ_POLICY_PID: 5847 preferred_mirror = first + (current->pid % num_stripes); 5848 break; 5849 } 5850 5851 if (dev_replace_is_ongoing && 5852 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5853 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5854 srcdev = fs_info->dev_replace.srcdev; 5855 else 5856 srcdev = NULL; 5857 5858 /* 5859 * try to avoid the drive that is the source drive for a 5860 * dev-replace procedure, only choose it if no other non-missing 5861 * mirror is available 5862 */ 5863 for (tolerance = 0; tolerance < 2; tolerance++) { 5864 if (map->stripes[preferred_mirror].dev->bdev && 5865 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5866 return preferred_mirror; 5867 for (i = first; i < first + num_stripes; i++) { 5868 if (map->stripes[i].dev->bdev && 5869 (tolerance || map->stripes[i].dev != srcdev)) 5870 return i; 5871 } 5872 } 5873 5874 /* we couldn't find one that doesn't fail. Just return something 5875 * and the io error handling code will clean up eventually 5876 */ 5877 return preferred_mirror; 5878 } 5879 5880 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5881 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5882 { 5883 int i; 5884 int again = 1; 5885 5886 while (again) { 5887 again = 0; 5888 for (i = 0; i < num_stripes - 1; i++) { 5889 /* Swap if parity is on a smaller index */ 5890 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5891 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5892 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5893 again = 1; 5894 } 5895 } 5896 } 5897 } 5898 5899 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5900 int total_stripes, 5901 int real_stripes) 5902 { 5903 struct btrfs_io_context *bioc = kzalloc( 5904 /* The size of btrfs_io_context */ 5905 sizeof(struct btrfs_io_context) + 5906 /* Plus the variable array for the stripes */ 5907 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5908 /* Plus the variable array for the tgt dev */ 5909 sizeof(int) * (real_stripes) + 5910 /* 5911 * Plus the raid_map, which includes both the tgt dev 5912 * and the stripes. 5913 */ 5914 sizeof(u64) * (total_stripes), 5915 GFP_NOFS|__GFP_NOFAIL); 5916 5917 refcount_set(&bioc->refs, 1); 5918 5919 bioc->fs_info = fs_info; 5920 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5921 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5922 5923 return bioc; 5924 } 5925 5926 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5927 { 5928 WARN_ON(!refcount_read(&bioc->refs)); 5929 refcount_inc(&bioc->refs); 5930 } 5931 5932 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5933 { 5934 if (!bioc) 5935 return; 5936 if (refcount_dec_and_test(&bioc->refs)) 5937 kfree(bioc); 5938 } 5939 5940 /* 5941 * Please note that, discard won't be sent to target device of device 5942 * replace. 5943 */ 5944 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, 5945 u64 logical, u64 *length_ret, 5946 u32 *num_stripes) 5947 { 5948 struct extent_map *em; 5949 struct map_lookup *map; 5950 struct btrfs_discard_stripe *stripes; 5951 u64 length = *length_ret; 5952 u64 offset; 5953 u64 stripe_nr; 5954 u64 stripe_nr_end; 5955 u64 stripe_end_offset; 5956 u64 stripe_cnt; 5957 u64 stripe_len; 5958 u64 stripe_offset; 5959 u32 stripe_index; 5960 u32 factor = 0; 5961 u32 sub_stripes = 0; 5962 u64 stripes_per_dev = 0; 5963 u32 remaining_stripes = 0; 5964 u32 last_stripe = 0; 5965 int ret; 5966 int i; 5967 5968 em = btrfs_get_chunk_map(fs_info, logical, length); 5969 if (IS_ERR(em)) 5970 return ERR_CAST(em); 5971 5972 map = em->map_lookup; 5973 5974 /* we don't discard raid56 yet */ 5975 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5976 ret = -EOPNOTSUPP; 5977 goto out_free_map; 5978 } 5979 5980 offset = logical - em->start; 5981 length = min_t(u64, em->start + em->len - logical, length); 5982 *length_ret = length; 5983 5984 stripe_len = map->stripe_len; 5985 /* 5986 * stripe_nr counts the total number of stripes we have to stride 5987 * to get to this block 5988 */ 5989 stripe_nr = div64_u64(offset, stripe_len); 5990 5991 /* stripe_offset is the offset of this block in its stripe */ 5992 stripe_offset = offset - stripe_nr * stripe_len; 5993 5994 stripe_nr_end = round_up(offset + length, map->stripe_len); 5995 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); 5996 stripe_cnt = stripe_nr_end - stripe_nr; 5997 stripe_end_offset = stripe_nr_end * map->stripe_len - 5998 (offset + length); 5999 /* 6000 * after this, stripe_nr is the number of stripes on this 6001 * device we have to walk to find the data, and stripe_index is 6002 * the number of our device in the stripe array 6003 */ 6004 *num_stripes = 1; 6005 stripe_index = 0; 6006 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6007 BTRFS_BLOCK_GROUP_RAID10)) { 6008 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 6009 sub_stripes = 1; 6010 else 6011 sub_stripes = map->sub_stripes; 6012 6013 factor = map->num_stripes / sub_stripes; 6014 *num_stripes = min_t(u64, map->num_stripes, 6015 sub_stripes * stripe_cnt); 6016 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6017 stripe_index *= sub_stripes; 6018 stripes_per_dev = div_u64_rem(stripe_cnt, factor, 6019 &remaining_stripes); 6020 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); 6021 last_stripe *= sub_stripes; 6022 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 6023 BTRFS_BLOCK_GROUP_DUP)) { 6024 *num_stripes = map->num_stripes; 6025 } else { 6026 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6027 &stripe_index); 6028 } 6029 6030 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); 6031 if (!stripes) { 6032 ret = -ENOMEM; 6033 goto out_free_map; 6034 } 6035 6036 for (i = 0; i < *num_stripes; i++) { 6037 stripes[i].physical = 6038 map->stripes[stripe_index].physical + 6039 stripe_offset + stripe_nr * map->stripe_len; 6040 stripes[i].dev = map->stripes[stripe_index].dev; 6041 6042 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6043 BTRFS_BLOCK_GROUP_RAID10)) { 6044 stripes[i].length = stripes_per_dev * map->stripe_len; 6045 6046 if (i / sub_stripes < remaining_stripes) 6047 stripes[i].length += map->stripe_len; 6048 6049 /* 6050 * Special for the first stripe and 6051 * the last stripe: 6052 * 6053 * |-------|...|-------| 6054 * |----------| 6055 * off end_off 6056 */ 6057 if (i < sub_stripes) 6058 stripes[i].length -= stripe_offset; 6059 6060 if (stripe_index >= last_stripe && 6061 stripe_index <= (last_stripe + 6062 sub_stripes - 1)) 6063 stripes[i].length -= stripe_end_offset; 6064 6065 if (i == sub_stripes - 1) 6066 stripe_offset = 0; 6067 } else { 6068 stripes[i].length = length; 6069 } 6070 6071 stripe_index++; 6072 if (stripe_index == map->num_stripes) { 6073 stripe_index = 0; 6074 stripe_nr++; 6075 } 6076 } 6077 6078 free_extent_map(em); 6079 return stripes; 6080 out_free_map: 6081 free_extent_map(em); 6082 return ERR_PTR(ret); 6083 } 6084 6085 /* 6086 * In dev-replace case, for repair case (that's the only case where the mirror 6087 * is selected explicitly when calling btrfs_map_block), blocks left of the 6088 * left cursor can also be read from the target drive. 6089 * 6090 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6091 * array of stripes. 6092 * For READ, it also needs to be supported using the same mirror number. 6093 * 6094 * If the requested block is not left of the left cursor, EIO is returned. This 6095 * can happen because btrfs_num_copies() returns one more in the dev-replace 6096 * case. 6097 */ 6098 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6099 u64 logical, u64 length, 6100 u64 srcdev_devid, int *mirror_num, 6101 u64 *physical) 6102 { 6103 struct btrfs_io_context *bioc = NULL; 6104 int num_stripes; 6105 int index_srcdev = 0; 6106 int found = 0; 6107 u64 physical_of_found = 0; 6108 int i; 6109 int ret = 0; 6110 6111 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6112 logical, &length, &bioc, NULL, NULL, 0); 6113 if (ret) { 6114 ASSERT(bioc == NULL); 6115 return ret; 6116 } 6117 6118 num_stripes = bioc->num_stripes; 6119 if (*mirror_num > num_stripes) { 6120 /* 6121 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6122 * that means that the requested area is not left of the left 6123 * cursor 6124 */ 6125 btrfs_put_bioc(bioc); 6126 return -EIO; 6127 } 6128 6129 /* 6130 * process the rest of the function using the mirror_num of the source 6131 * drive. Therefore look it up first. At the end, patch the device 6132 * pointer to the one of the target drive. 6133 */ 6134 for (i = 0; i < num_stripes; i++) { 6135 if (bioc->stripes[i].dev->devid != srcdev_devid) 6136 continue; 6137 6138 /* 6139 * In case of DUP, in order to keep it simple, only add the 6140 * mirror with the lowest physical address 6141 */ 6142 if (found && 6143 physical_of_found <= bioc->stripes[i].physical) 6144 continue; 6145 6146 index_srcdev = i; 6147 found = 1; 6148 physical_of_found = bioc->stripes[i].physical; 6149 } 6150 6151 btrfs_put_bioc(bioc); 6152 6153 ASSERT(found); 6154 if (!found) 6155 return -EIO; 6156 6157 *mirror_num = index_srcdev + 1; 6158 *physical = physical_of_found; 6159 return ret; 6160 } 6161 6162 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6163 { 6164 struct btrfs_block_group *cache; 6165 bool ret; 6166 6167 /* Non zoned filesystem does not use "to_copy" flag */ 6168 if (!btrfs_is_zoned(fs_info)) 6169 return false; 6170 6171 cache = btrfs_lookup_block_group(fs_info, logical); 6172 6173 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 6174 6175 btrfs_put_block_group(cache); 6176 return ret; 6177 } 6178 6179 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6180 struct btrfs_io_context **bioc_ret, 6181 struct btrfs_dev_replace *dev_replace, 6182 u64 logical, 6183 int *num_stripes_ret, int *max_errors_ret) 6184 { 6185 struct btrfs_io_context *bioc = *bioc_ret; 6186 u64 srcdev_devid = dev_replace->srcdev->devid; 6187 int tgtdev_indexes = 0; 6188 int num_stripes = *num_stripes_ret; 6189 int max_errors = *max_errors_ret; 6190 int i; 6191 6192 if (op == BTRFS_MAP_WRITE) { 6193 int index_where_to_add; 6194 6195 /* 6196 * A block group which have "to_copy" set will eventually 6197 * copied by dev-replace process. We can avoid cloning IO here. 6198 */ 6199 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6200 return; 6201 6202 /* 6203 * duplicate the write operations while the dev replace 6204 * procedure is running. Since the copying of the old disk to 6205 * the new disk takes place at run time while the filesystem is 6206 * mounted writable, the regular write operations to the old 6207 * disk have to be duplicated to go to the new disk as well. 6208 * 6209 * Note that device->missing is handled by the caller, and that 6210 * the write to the old disk is already set up in the stripes 6211 * array. 6212 */ 6213 index_where_to_add = num_stripes; 6214 for (i = 0; i < num_stripes; i++) { 6215 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6216 /* write to new disk, too */ 6217 struct btrfs_io_stripe *new = 6218 bioc->stripes + index_where_to_add; 6219 struct btrfs_io_stripe *old = 6220 bioc->stripes + i; 6221 6222 new->physical = old->physical; 6223 new->dev = dev_replace->tgtdev; 6224 bioc->tgtdev_map[i] = index_where_to_add; 6225 index_where_to_add++; 6226 max_errors++; 6227 tgtdev_indexes++; 6228 } 6229 } 6230 num_stripes = index_where_to_add; 6231 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6232 int index_srcdev = 0; 6233 int found = 0; 6234 u64 physical_of_found = 0; 6235 6236 /* 6237 * During the dev-replace procedure, the target drive can also 6238 * be used to read data in case it is needed to repair a corrupt 6239 * block elsewhere. This is possible if the requested area is 6240 * left of the left cursor. In this area, the target drive is a 6241 * full copy of the source drive. 6242 */ 6243 for (i = 0; i < num_stripes; i++) { 6244 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6245 /* 6246 * In case of DUP, in order to keep it simple, 6247 * only add the mirror with the lowest physical 6248 * address 6249 */ 6250 if (found && 6251 physical_of_found <= bioc->stripes[i].physical) 6252 continue; 6253 index_srcdev = i; 6254 found = 1; 6255 physical_of_found = bioc->stripes[i].physical; 6256 } 6257 } 6258 if (found) { 6259 struct btrfs_io_stripe *tgtdev_stripe = 6260 bioc->stripes + num_stripes; 6261 6262 tgtdev_stripe->physical = physical_of_found; 6263 tgtdev_stripe->dev = dev_replace->tgtdev; 6264 bioc->tgtdev_map[index_srcdev] = num_stripes; 6265 6266 tgtdev_indexes++; 6267 num_stripes++; 6268 } 6269 } 6270 6271 *num_stripes_ret = num_stripes; 6272 *max_errors_ret = max_errors; 6273 bioc->num_tgtdevs = tgtdev_indexes; 6274 *bioc_ret = bioc; 6275 } 6276 6277 static bool need_full_stripe(enum btrfs_map_op op) 6278 { 6279 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6280 } 6281 6282 /* 6283 * Calculate the geometry of a particular (address, len) tuple. This 6284 * information is used to calculate how big a particular bio can get before it 6285 * straddles a stripe. 6286 * 6287 * @fs_info: the filesystem 6288 * @em: mapping containing the logical extent 6289 * @op: type of operation - write or read 6290 * @logical: address that we want to figure out the geometry of 6291 * @io_geom: pointer used to return values 6292 * 6293 * Returns < 0 in case a chunk for the given logical address cannot be found, 6294 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6295 */ 6296 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6297 enum btrfs_map_op op, u64 logical, 6298 struct btrfs_io_geometry *io_geom) 6299 { 6300 struct map_lookup *map; 6301 u64 len; 6302 u64 offset; 6303 u64 stripe_offset; 6304 u64 stripe_nr; 6305 u32 stripe_len; 6306 u64 raid56_full_stripe_start = (u64)-1; 6307 int data_stripes; 6308 6309 ASSERT(op != BTRFS_MAP_DISCARD); 6310 6311 map = em->map_lookup; 6312 /* Offset of this logical address in the chunk */ 6313 offset = logical - em->start; 6314 /* Len of a stripe in a chunk */ 6315 stripe_len = map->stripe_len; 6316 /* 6317 * Stripe_nr is where this block falls in 6318 * stripe_offset is the offset of this block in its stripe. 6319 */ 6320 stripe_nr = div64_u64_rem(offset, stripe_len, &stripe_offset); 6321 ASSERT(stripe_offset < U32_MAX); 6322 6323 data_stripes = nr_data_stripes(map); 6324 6325 /* Only stripe based profiles needs to check against stripe length. */ 6326 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) { 6327 u64 max_len = stripe_len - stripe_offset; 6328 6329 /* 6330 * In case of raid56, we need to know the stripe aligned start 6331 */ 6332 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6333 unsigned long full_stripe_len = stripe_len * data_stripes; 6334 raid56_full_stripe_start = offset; 6335 6336 /* 6337 * Allow a write of a full stripe, but make sure we 6338 * don't allow straddling of stripes 6339 */ 6340 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6341 full_stripe_len); 6342 raid56_full_stripe_start *= full_stripe_len; 6343 6344 /* 6345 * For writes to RAID[56], allow a full stripeset across 6346 * all disks. For other RAID types and for RAID[56] 6347 * reads, just allow a single stripe (on a single disk). 6348 */ 6349 if (op == BTRFS_MAP_WRITE) { 6350 max_len = stripe_len * data_stripes - 6351 (offset - raid56_full_stripe_start); 6352 } 6353 } 6354 len = min_t(u64, em->len - offset, max_len); 6355 } else { 6356 len = em->len - offset; 6357 } 6358 6359 io_geom->len = len; 6360 io_geom->offset = offset; 6361 io_geom->stripe_len = stripe_len; 6362 io_geom->stripe_nr = stripe_nr; 6363 io_geom->stripe_offset = stripe_offset; 6364 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6365 6366 return 0; 6367 } 6368 6369 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, 6370 u32 stripe_index, u64 stripe_offset, u64 stripe_nr) 6371 { 6372 dst->dev = map->stripes[stripe_index].dev; 6373 dst->physical = map->stripes[stripe_index].physical + 6374 stripe_offset + stripe_nr * map->stripe_len; 6375 } 6376 6377 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6378 enum btrfs_map_op op, u64 logical, u64 *length, 6379 struct btrfs_io_context **bioc_ret, 6380 struct btrfs_io_stripe *smap, 6381 int *mirror_num_ret, int need_raid_map) 6382 { 6383 struct extent_map *em; 6384 struct map_lookup *map; 6385 u64 stripe_offset; 6386 u64 stripe_nr; 6387 u64 stripe_len; 6388 u32 stripe_index; 6389 int data_stripes; 6390 int i; 6391 int ret = 0; 6392 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); 6393 int num_stripes; 6394 int max_errors = 0; 6395 int tgtdev_indexes = 0; 6396 struct btrfs_io_context *bioc = NULL; 6397 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6398 int dev_replace_is_ongoing = 0; 6399 int num_alloc_stripes; 6400 int patch_the_first_stripe_for_dev_replace = 0; 6401 u64 physical_to_patch_in_first_stripe = 0; 6402 u64 raid56_full_stripe_start = (u64)-1; 6403 struct btrfs_io_geometry geom; 6404 6405 ASSERT(bioc_ret); 6406 ASSERT(op != BTRFS_MAP_DISCARD); 6407 6408 em = btrfs_get_chunk_map(fs_info, logical, *length); 6409 ASSERT(!IS_ERR(em)); 6410 6411 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6412 if (ret < 0) 6413 return ret; 6414 6415 map = em->map_lookup; 6416 6417 *length = geom.len; 6418 stripe_len = geom.stripe_len; 6419 stripe_nr = geom.stripe_nr; 6420 stripe_offset = geom.stripe_offset; 6421 raid56_full_stripe_start = geom.raid56_stripe_offset; 6422 data_stripes = nr_data_stripes(map); 6423 6424 down_read(&dev_replace->rwsem); 6425 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6426 /* 6427 * Hold the semaphore for read during the whole operation, write is 6428 * requested at commit time but must wait. 6429 */ 6430 if (!dev_replace_is_ongoing) 6431 up_read(&dev_replace->rwsem); 6432 6433 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6434 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6435 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6436 dev_replace->srcdev->devid, 6437 &mirror_num, 6438 &physical_to_patch_in_first_stripe); 6439 if (ret) 6440 goto out; 6441 else 6442 patch_the_first_stripe_for_dev_replace = 1; 6443 } else if (mirror_num > map->num_stripes) { 6444 mirror_num = 0; 6445 } 6446 6447 num_stripes = 1; 6448 stripe_index = 0; 6449 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6450 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6451 &stripe_index); 6452 if (!need_full_stripe(op)) 6453 mirror_num = 1; 6454 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6455 if (need_full_stripe(op)) 6456 num_stripes = map->num_stripes; 6457 else if (mirror_num) 6458 stripe_index = mirror_num - 1; 6459 else { 6460 stripe_index = find_live_mirror(fs_info, map, 0, 6461 dev_replace_is_ongoing); 6462 mirror_num = stripe_index + 1; 6463 } 6464 6465 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6466 if (need_full_stripe(op)) { 6467 num_stripes = map->num_stripes; 6468 } else if (mirror_num) { 6469 stripe_index = mirror_num - 1; 6470 } else { 6471 mirror_num = 1; 6472 } 6473 6474 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6475 u32 factor = map->num_stripes / map->sub_stripes; 6476 6477 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6478 stripe_index *= map->sub_stripes; 6479 6480 if (need_full_stripe(op)) 6481 num_stripes = map->sub_stripes; 6482 else if (mirror_num) 6483 stripe_index += mirror_num - 1; 6484 else { 6485 int old_stripe_index = stripe_index; 6486 stripe_index = find_live_mirror(fs_info, map, 6487 stripe_index, 6488 dev_replace_is_ongoing); 6489 mirror_num = stripe_index - old_stripe_index + 1; 6490 } 6491 6492 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6493 ASSERT(map->stripe_len == BTRFS_STRIPE_LEN); 6494 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6495 /* push stripe_nr back to the start of the full stripe */ 6496 stripe_nr = div64_u64(raid56_full_stripe_start, 6497 stripe_len * data_stripes); 6498 6499 /* RAID[56] write or recovery. Return all stripes */ 6500 num_stripes = map->num_stripes; 6501 max_errors = btrfs_chunk_max_errors(map); 6502 6503 /* Return the length to the full stripe end */ 6504 *length = min(logical + *length, 6505 raid56_full_stripe_start + em->start + 6506 data_stripes * stripe_len) - logical; 6507 stripe_index = 0; 6508 stripe_offset = 0; 6509 } else { 6510 /* 6511 * Mirror #0 or #1 means the original data block. 6512 * Mirror #2 is RAID5 parity block. 6513 * Mirror #3 is RAID6 Q block. 6514 */ 6515 stripe_nr = div_u64_rem(stripe_nr, 6516 data_stripes, &stripe_index); 6517 if (mirror_num > 1) 6518 stripe_index = data_stripes + mirror_num - 2; 6519 6520 /* We distribute the parity blocks across stripes */ 6521 div_u64_rem(stripe_nr + stripe_index, map->num_stripes, 6522 &stripe_index); 6523 if (!need_full_stripe(op) && mirror_num <= 1) 6524 mirror_num = 1; 6525 } 6526 } else { 6527 /* 6528 * after this, stripe_nr is the number of stripes on this 6529 * device we have to walk to find the data, and stripe_index is 6530 * the number of our device in the stripe array 6531 */ 6532 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6533 &stripe_index); 6534 mirror_num = stripe_index + 1; 6535 } 6536 if (stripe_index >= map->num_stripes) { 6537 btrfs_crit(fs_info, 6538 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6539 stripe_index, map->num_stripes); 6540 ret = -EINVAL; 6541 goto out; 6542 } 6543 6544 num_alloc_stripes = num_stripes; 6545 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6546 if (op == BTRFS_MAP_WRITE) 6547 num_alloc_stripes <<= 1; 6548 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6549 num_alloc_stripes++; 6550 tgtdev_indexes = num_stripes; 6551 } 6552 6553 /* 6554 * If this I/O maps to a single device, try to return the device and 6555 * physical block information on the stack instead of allocating an 6556 * I/O context structure. 6557 */ 6558 if (smap && num_alloc_stripes == 1 && 6559 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && 6560 (!need_full_stripe(op) || !dev_replace_is_ongoing || 6561 !dev_replace->tgtdev)) { 6562 if (patch_the_first_stripe_for_dev_replace) { 6563 smap->dev = dev_replace->tgtdev; 6564 smap->physical = physical_to_patch_in_first_stripe; 6565 *mirror_num_ret = map->num_stripes + 1; 6566 } else { 6567 set_io_stripe(smap, map, stripe_index, stripe_offset, 6568 stripe_nr); 6569 *mirror_num_ret = mirror_num; 6570 } 6571 *bioc_ret = NULL; 6572 ret = 0; 6573 goto out; 6574 } 6575 6576 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6577 if (!bioc) { 6578 ret = -ENOMEM; 6579 goto out; 6580 } 6581 6582 for (i = 0; i < num_stripes; i++) { 6583 set_io_stripe(&bioc->stripes[i], map, stripe_index, stripe_offset, 6584 stripe_nr); 6585 stripe_index++; 6586 } 6587 6588 /* Build raid_map */ 6589 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6590 (need_full_stripe(op) || mirror_num > 1)) { 6591 u64 tmp; 6592 unsigned rot; 6593 6594 /* Work out the disk rotation on this stripe-set */ 6595 div_u64_rem(stripe_nr, num_stripes, &rot); 6596 6597 /* Fill in the logical address of each stripe */ 6598 tmp = stripe_nr * data_stripes; 6599 for (i = 0; i < data_stripes; i++) 6600 bioc->raid_map[(i + rot) % num_stripes] = 6601 em->start + (tmp + i) * map->stripe_len; 6602 6603 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6604 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6605 bioc->raid_map[(i + rot + 1) % num_stripes] = 6606 RAID6_Q_STRIPE; 6607 6608 sort_parity_stripes(bioc, num_stripes); 6609 } 6610 6611 if (need_full_stripe(op)) 6612 max_errors = btrfs_chunk_max_errors(map); 6613 6614 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6615 need_full_stripe(op)) { 6616 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6617 &num_stripes, &max_errors); 6618 } 6619 6620 *bioc_ret = bioc; 6621 bioc->map_type = map->type; 6622 bioc->num_stripes = num_stripes; 6623 bioc->max_errors = max_errors; 6624 bioc->mirror_num = mirror_num; 6625 6626 /* 6627 * this is the case that REQ_READ && dev_replace_is_ongoing && 6628 * mirror_num == num_stripes + 1 && dev_replace target drive is 6629 * available as a mirror 6630 */ 6631 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6632 WARN_ON(num_stripes > 1); 6633 bioc->stripes[0].dev = dev_replace->tgtdev; 6634 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6635 bioc->mirror_num = map->num_stripes + 1; 6636 } 6637 out: 6638 if (dev_replace_is_ongoing) { 6639 lockdep_assert_held(&dev_replace->rwsem); 6640 /* Unlock and let waiting writers proceed */ 6641 up_read(&dev_replace->rwsem); 6642 } 6643 free_extent_map(em); 6644 return ret; 6645 } 6646 6647 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6648 u64 logical, u64 *length, 6649 struct btrfs_io_context **bioc_ret, int mirror_num) 6650 { 6651 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6652 NULL, &mirror_num, 0); 6653 } 6654 6655 /* For Scrub/replace */ 6656 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6657 u64 logical, u64 *length, 6658 struct btrfs_io_context **bioc_ret) 6659 { 6660 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6661 NULL, NULL, 1); 6662 } 6663 6664 /* 6665 * Initialize a btrfs_bio structure. This skips the embedded bio itself as it 6666 * is already initialized by the block layer. 6667 */ 6668 static inline void btrfs_bio_init(struct btrfs_bio *bbio, 6669 btrfs_bio_end_io_t end_io, void *private) 6670 { 6671 memset(bbio, 0, offsetof(struct btrfs_bio, bio)); 6672 bbio->end_io = end_io; 6673 bbio->private = private; 6674 } 6675 6676 /* 6677 * Allocate a btrfs_bio structure. The btrfs_bio is the main I/O container for 6678 * btrfs, and is used for all I/O submitted through btrfs_submit_bio. 6679 * 6680 * Just like the underlying bio_alloc_bioset it will not fail as it is backed by 6681 * a mempool. 6682 */ 6683 struct bio *btrfs_bio_alloc(unsigned int nr_vecs, blk_opf_t opf, 6684 btrfs_bio_end_io_t end_io, void *private) 6685 { 6686 struct bio *bio; 6687 6688 bio = bio_alloc_bioset(NULL, nr_vecs, opf, GFP_NOFS, &btrfs_bioset); 6689 btrfs_bio_init(btrfs_bio(bio), end_io, private); 6690 return bio; 6691 } 6692 6693 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size, 6694 btrfs_bio_end_io_t end_io, void *private) 6695 { 6696 struct bio *bio; 6697 struct btrfs_bio *bbio; 6698 6699 ASSERT(offset <= UINT_MAX && size <= UINT_MAX); 6700 6701 bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset); 6702 bbio = btrfs_bio(bio); 6703 btrfs_bio_init(bbio, end_io, private); 6704 6705 bio_trim(bio, offset >> 9, size >> 9); 6706 bbio->iter = bio->bi_iter; 6707 return bio; 6708 } 6709 6710 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev) 6711 { 6712 if (!dev || !dev->bdev) 6713 return; 6714 if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET) 6715 return; 6716 6717 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6718 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); 6719 if (!(bio->bi_opf & REQ_RAHEAD)) 6720 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); 6721 if (bio->bi_opf & REQ_PREFLUSH) 6722 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS); 6723 } 6724 6725 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info, 6726 struct bio *bio) 6727 { 6728 if (bio->bi_opf & REQ_META) 6729 return fs_info->endio_meta_workers; 6730 return fs_info->endio_workers; 6731 } 6732 6733 static void btrfs_end_bio_work(struct work_struct *work) 6734 { 6735 struct btrfs_bio *bbio = 6736 container_of(work, struct btrfs_bio, end_io_work); 6737 6738 bbio->end_io(bbio); 6739 } 6740 6741 static void btrfs_simple_end_io(struct bio *bio) 6742 { 6743 struct btrfs_fs_info *fs_info = bio->bi_private; 6744 struct btrfs_bio *bbio = btrfs_bio(bio); 6745 6746 btrfs_bio_counter_dec(fs_info); 6747 6748 if (bio->bi_status) 6749 btrfs_log_dev_io_error(bio, bbio->device); 6750 6751 if (bio_op(bio) == REQ_OP_READ) { 6752 INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work); 6753 queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work); 6754 } else { 6755 bbio->end_io(bbio); 6756 } 6757 } 6758 6759 static void btrfs_raid56_end_io(struct bio *bio) 6760 { 6761 struct btrfs_io_context *bioc = bio->bi_private; 6762 struct btrfs_bio *bbio = btrfs_bio(bio); 6763 6764 btrfs_bio_counter_dec(bioc->fs_info); 6765 bbio->mirror_num = bioc->mirror_num; 6766 bbio->end_io(bbio); 6767 6768 btrfs_put_bioc(bioc); 6769 } 6770 6771 static void btrfs_orig_write_end_io(struct bio *bio) 6772 { 6773 struct btrfs_io_stripe *stripe = bio->bi_private; 6774 struct btrfs_io_context *bioc = stripe->bioc; 6775 struct btrfs_bio *bbio = btrfs_bio(bio); 6776 6777 btrfs_bio_counter_dec(bioc->fs_info); 6778 6779 if (bio->bi_status) { 6780 atomic_inc(&bioc->error); 6781 btrfs_log_dev_io_error(bio, stripe->dev); 6782 } 6783 6784 /* 6785 * Only send an error to the higher layers if it is beyond the tolerance 6786 * threshold. 6787 */ 6788 if (atomic_read(&bioc->error) > bioc->max_errors) 6789 bio->bi_status = BLK_STS_IOERR; 6790 else 6791 bio->bi_status = BLK_STS_OK; 6792 6793 bbio->end_io(bbio); 6794 btrfs_put_bioc(bioc); 6795 } 6796 6797 static void btrfs_clone_write_end_io(struct bio *bio) 6798 { 6799 struct btrfs_io_stripe *stripe = bio->bi_private; 6800 6801 if (bio->bi_status) { 6802 atomic_inc(&stripe->bioc->error); 6803 btrfs_log_dev_io_error(bio, stripe->dev); 6804 } 6805 6806 /* Pass on control to the original bio this one was cloned from */ 6807 bio_endio(stripe->bioc->orig_bio); 6808 bio_put(bio); 6809 } 6810 6811 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio) 6812 { 6813 if (!dev || !dev->bdev || 6814 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 6815 (btrfs_op(bio) == BTRFS_MAP_WRITE && 6816 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6817 bio_io_error(bio); 6818 return; 6819 } 6820 6821 bio_set_dev(bio, dev->bdev); 6822 6823 /* 6824 * For zone append writing, bi_sector must point the beginning of the 6825 * zone 6826 */ 6827 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6828 u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; 6829 6830 if (btrfs_dev_is_sequential(dev, physical)) { 6831 u64 zone_start = round_down(physical, 6832 dev->fs_info->zone_size); 6833 6834 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6835 } else { 6836 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6837 bio->bi_opf |= REQ_OP_WRITE; 6838 } 6839 } 6840 btrfs_debug_in_rcu(dev->fs_info, 6841 "%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6842 __func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6843 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6844 dev->devid, bio->bi_iter.bi_size); 6845 6846 btrfsic_check_bio(bio); 6847 submit_bio(bio); 6848 } 6849 6850 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr) 6851 { 6852 struct bio *orig_bio = bioc->orig_bio, *bio; 6853 6854 ASSERT(bio_op(orig_bio) != REQ_OP_READ); 6855 6856 /* Reuse the bio embedded into the btrfs_bio for the last mirror */ 6857 if (dev_nr == bioc->num_stripes - 1) { 6858 bio = orig_bio; 6859 bio->bi_end_io = btrfs_orig_write_end_io; 6860 } else { 6861 bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set); 6862 bio_inc_remaining(orig_bio); 6863 bio->bi_end_io = btrfs_clone_write_end_io; 6864 } 6865 6866 bio->bi_private = &bioc->stripes[dev_nr]; 6867 bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT; 6868 bioc->stripes[dev_nr].bioc = bioc; 6869 btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio); 6870 } 6871 6872 void btrfs_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, int mirror_num) 6873 { 6874 u64 logical = bio->bi_iter.bi_sector << 9; 6875 u64 length = bio->bi_iter.bi_size; 6876 u64 map_length = length; 6877 struct btrfs_io_context *bioc = NULL; 6878 struct btrfs_io_stripe smap; 6879 int ret; 6880 6881 btrfs_bio_counter_inc_blocked(fs_info); 6882 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, 6883 &bioc, &smap, &mirror_num, 1); 6884 if (ret) { 6885 btrfs_bio_counter_dec(fs_info); 6886 btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret)); 6887 return; 6888 } 6889 6890 if (map_length < length) { 6891 btrfs_crit(fs_info, 6892 "mapping failed logical %llu bio len %llu len %llu", 6893 logical, length, map_length); 6894 BUG(); 6895 } 6896 6897 if (!bioc) { 6898 /* Single mirror read/write fast path */ 6899 btrfs_bio(bio)->mirror_num = mirror_num; 6900 btrfs_bio(bio)->device = smap.dev; 6901 bio->bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT; 6902 bio->bi_private = fs_info; 6903 bio->bi_end_io = btrfs_simple_end_io; 6904 btrfs_submit_dev_bio(smap.dev, bio); 6905 } else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6906 /* Parity RAID write or read recovery */ 6907 bio->bi_private = bioc; 6908 bio->bi_end_io = btrfs_raid56_end_io; 6909 if (bio_op(bio) == REQ_OP_READ) 6910 raid56_parity_recover(bio, bioc, mirror_num); 6911 else 6912 raid56_parity_write(bio, bioc); 6913 } else { 6914 /* Write to multiple mirrors */ 6915 int total_devs = bioc->num_stripes; 6916 int dev_nr; 6917 6918 bioc->orig_bio = bio; 6919 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) 6920 btrfs_submit_mirrored_bio(bioc, dev_nr); 6921 } 6922 } 6923 6924 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6925 const struct btrfs_fs_devices *fs_devices) 6926 { 6927 if (args->fsid == NULL) 6928 return true; 6929 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6930 return true; 6931 return false; 6932 } 6933 6934 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6935 const struct btrfs_device *device) 6936 { 6937 if (args->missing) { 6938 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6939 !device->bdev) 6940 return true; 6941 return false; 6942 } 6943 6944 if (device->devid != args->devid) 6945 return false; 6946 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6947 return false; 6948 return true; 6949 } 6950 6951 /* 6952 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6953 * return NULL. 6954 * 6955 * If devid and uuid are both specified, the match must be exact, otherwise 6956 * only devid is used. 6957 */ 6958 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6959 const struct btrfs_dev_lookup_args *args) 6960 { 6961 struct btrfs_device *device; 6962 struct btrfs_fs_devices *seed_devs; 6963 6964 if (dev_args_match_fs_devices(args, fs_devices)) { 6965 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6966 if (dev_args_match_device(args, device)) 6967 return device; 6968 } 6969 } 6970 6971 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6972 if (!dev_args_match_fs_devices(args, seed_devs)) 6973 continue; 6974 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6975 if (dev_args_match_device(args, device)) 6976 return device; 6977 } 6978 } 6979 6980 return NULL; 6981 } 6982 6983 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6984 u64 devid, u8 *dev_uuid) 6985 { 6986 struct btrfs_device *device; 6987 unsigned int nofs_flag; 6988 6989 /* 6990 * We call this under the chunk_mutex, so we want to use NOFS for this 6991 * allocation, however we don't want to change btrfs_alloc_device() to 6992 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6993 * places. 6994 */ 6995 nofs_flag = memalloc_nofs_save(); 6996 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 6997 memalloc_nofs_restore(nofs_flag); 6998 if (IS_ERR(device)) 6999 return device; 7000 7001 list_add(&device->dev_list, &fs_devices->devices); 7002 device->fs_devices = fs_devices; 7003 fs_devices->num_devices++; 7004 7005 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7006 fs_devices->missing_devices++; 7007 7008 return device; 7009 } 7010 7011 /** 7012 * btrfs_alloc_device - allocate struct btrfs_device 7013 * @fs_info: used only for generating a new devid, can be NULL if 7014 * devid is provided (i.e. @devid != NULL). 7015 * @devid: a pointer to devid for this device. If NULL a new devid 7016 * is generated. 7017 * @uuid: a pointer to UUID for this device. If NULL a new UUID 7018 * is generated. 7019 * 7020 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 7021 * on error. Returned struct is not linked onto any lists and must be 7022 * destroyed with btrfs_free_device. 7023 */ 7024 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 7025 const u64 *devid, 7026 const u8 *uuid) 7027 { 7028 struct btrfs_device *dev; 7029 u64 tmp; 7030 7031 if (WARN_ON(!devid && !fs_info)) 7032 return ERR_PTR(-EINVAL); 7033 7034 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 7035 if (!dev) 7036 return ERR_PTR(-ENOMEM); 7037 7038 INIT_LIST_HEAD(&dev->dev_list); 7039 INIT_LIST_HEAD(&dev->dev_alloc_list); 7040 INIT_LIST_HEAD(&dev->post_commit_list); 7041 7042 atomic_set(&dev->dev_stats_ccnt, 0); 7043 btrfs_device_data_ordered_init(dev); 7044 extent_io_tree_init(fs_info, &dev->alloc_state, 7045 IO_TREE_DEVICE_ALLOC_STATE, NULL); 7046 7047 if (devid) 7048 tmp = *devid; 7049 else { 7050 int ret; 7051 7052 ret = find_next_devid(fs_info, &tmp); 7053 if (ret) { 7054 btrfs_free_device(dev); 7055 return ERR_PTR(ret); 7056 } 7057 } 7058 dev->devid = tmp; 7059 7060 if (uuid) 7061 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 7062 else 7063 generate_random_uuid(dev->uuid); 7064 7065 return dev; 7066 } 7067 7068 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 7069 u64 devid, u8 *uuid, bool error) 7070 { 7071 if (error) 7072 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 7073 devid, uuid); 7074 else 7075 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 7076 devid, uuid); 7077 } 7078 7079 u64 btrfs_calc_stripe_length(const struct extent_map *em) 7080 { 7081 const struct map_lookup *map = em->map_lookup; 7082 const int data_stripes = calc_data_stripes(map->type, map->num_stripes); 7083 7084 return div_u64(em->len, data_stripes); 7085 } 7086 7087 #if BITS_PER_LONG == 32 7088 /* 7089 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 7090 * can't be accessed on 32bit systems. 7091 * 7092 * This function do mount time check to reject the fs if it already has 7093 * metadata chunk beyond that limit. 7094 */ 7095 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7096 u64 logical, u64 length, u64 type) 7097 { 7098 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7099 return 0; 7100 7101 if (logical + length < MAX_LFS_FILESIZE) 7102 return 0; 7103 7104 btrfs_err_32bit_limit(fs_info); 7105 return -EOVERFLOW; 7106 } 7107 7108 /* 7109 * This is to give early warning for any metadata chunk reaching 7110 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 7111 * Although we can still access the metadata, it's not going to be possible 7112 * once the limit is reached. 7113 */ 7114 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7115 u64 logical, u64 length, u64 type) 7116 { 7117 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7118 return; 7119 7120 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 7121 return; 7122 7123 btrfs_warn_32bit_limit(fs_info); 7124 } 7125 #endif 7126 7127 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 7128 u64 devid, u8 *uuid) 7129 { 7130 struct btrfs_device *dev; 7131 7132 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7133 btrfs_report_missing_device(fs_info, devid, uuid, true); 7134 return ERR_PTR(-ENOENT); 7135 } 7136 7137 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 7138 if (IS_ERR(dev)) { 7139 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 7140 devid, PTR_ERR(dev)); 7141 return dev; 7142 } 7143 btrfs_report_missing_device(fs_info, devid, uuid, false); 7144 7145 return dev; 7146 } 7147 7148 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 7149 struct btrfs_chunk *chunk) 7150 { 7151 BTRFS_DEV_LOOKUP_ARGS(args); 7152 struct btrfs_fs_info *fs_info = leaf->fs_info; 7153 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7154 struct map_lookup *map; 7155 struct extent_map *em; 7156 u64 logical; 7157 u64 length; 7158 u64 devid; 7159 u64 type; 7160 u8 uuid[BTRFS_UUID_SIZE]; 7161 int index; 7162 int num_stripes; 7163 int ret; 7164 int i; 7165 7166 logical = key->offset; 7167 length = btrfs_chunk_length(leaf, chunk); 7168 type = btrfs_chunk_type(leaf, chunk); 7169 index = btrfs_bg_flags_to_raid_index(type); 7170 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 7171 7172 #if BITS_PER_LONG == 32 7173 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 7174 if (ret < 0) 7175 return ret; 7176 warn_32bit_meta_chunk(fs_info, logical, length, type); 7177 #endif 7178 7179 /* 7180 * Only need to verify chunk item if we're reading from sys chunk array, 7181 * as chunk item in tree block is already verified by tree-checker. 7182 */ 7183 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 7184 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 7185 if (ret) 7186 return ret; 7187 } 7188 7189 read_lock(&map_tree->lock); 7190 em = lookup_extent_mapping(map_tree, logical, 1); 7191 read_unlock(&map_tree->lock); 7192 7193 /* already mapped? */ 7194 if (em && em->start <= logical && em->start + em->len > logical) { 7195 free_extent_map(em); 7196 return 0; 7197 } else if (em) { 7198 free_extent_map(em); 7199 } 7200 7201 em = alloc_extent_map(); 7202 if (!em) 7203 return -ENOMEM; 7204 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7205 if (!map) { 7206 free_extent_map(em); 7207 return -ENOMEM; 7208 } 7209 7210 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7211 em->map_lookup = map; 7212 em->start = logical; 7213 em->len = length; 7214 em->orig_start = 0; 7215 em->block_start = 0; 7216 em->block_len = em->len; 7217 7218 map->num_stripes = num_stripes; 7219 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7220 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7221 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7222 map->type = type; 7223 /* 7224 * We can't use the sub_stripes value, as for profiles other than 7225 * RAID10, they may have 0 as sub_stripes for filesystems created by 7226 * older mkfs (<v5.4). 7227 * In that case, it can cause divide-by-zero errors later. 7228 * Since currently sub_stripes is fixed for each profile, let's 7229 * use the trusted value instead. 7230 */ 7231 map->sub_stripes = btrfs_raid_array[index].sub_stripes; 7232 map->verified_stripes = 0; 7233 em->orig_block_len = btrfs_calc_stripe_length(em); 7234 for (i = 0; i < num_stripes; i++) { 7235 map->stripes[i].physical = 7236 btrfs_stripe_offset_nr(leaf, chunk, i); 7237 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7238 args.devid = devid; 7239 read_extent_buffer(leaf, uuid, (unsigned long) 7240 btrfs_stripe_dev_uuid_nr(chunk, i), 7241 BTRFS_UUID_SIZE); 7242 args.uuid = uuid; 7243 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 7244 if (!map->stripes[i].dev) { 7245 map->stripes[i].dev = handle_missing_device(fs_info, 7246 devid, uuid); 7247 if (IS_ERR(map->stripes[i].dev)) { 7248 free_extent_map(em); 7249 return PTR_ERR(map->stripes[i].dev); 7250 } 7251 } 7252 7253 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7254 &(map->stripes[i].dev->dev_state)); 7255 } 7256 7257 write_lock(&map_tree->lock); 7258 ret = add_extent_mapping(map_tree, em, 0); 7259 write_unlock(&map_tree->lock); 7260 if (ret < 0) { 7261 btrfs_err(fs_info, 7262 "failed to add chunk map, start=%llu len=%llu: %d", 7263 em->start, em->len, ret); 7264 } 7265 free_extent_map(em); 7266 7267 return ret; 7268 } 7269 7270 static void fill_device_from_item(struct extent_buffer *leaf, 7271 struct btrfs_dev_item *dev_item, 7272 struct btrfs_device *device) 7273 { 7274 unsigned long ptr; 7275 7276 device->devid = btrfs_device_id(leaf, dev_item); 7277 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7278 device->total_bytes = device->disk_total_bytes; 7279 device->commit_total_bytes = device->disk_total_bytes; 7280 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7281 device->commit_bytes_used = device->bytes_used; 7282 device->type = btrfs_device_type(leaf, dev_item); 7283 device->io_align = btrfs_device_io_align(leaf, dev_item); 7284 device->io_width = btrfs_device_io_width(leaf, dev_item); 7285 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7286 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7287 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7288 7289 ptr = btrfs_device_uuid(dev_item); 7290 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7291 } 7292 7293 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7294 u8 *fsid) 7295 { 7296 struct btrfs_fs_devices *fs_devices; 7297 int ret; 7298 7299 lockdep_assert_held(&uuid_mutex); 7300 ASSERT(fsid); 7301 7302 /* This will match only for multi-device seed fs */ 7303 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7304 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7305 return fs_devices; 7306 7307 7308 fs_devices = find_fsid(fsid, NULL); 7309 if (!fs_devices) { 7310 if (!btrfs_test_opt(fs_info, DEGRADED)) 7311 return ERR_PTR(-ENOENT); 7312 7313 fs_devices = alloc_fs_devices(fsid, NULL); 7314 if (IS_ERR(fs_devices)) 7315 return fs_devices; 7316 7317 fs_devices->seeding = true; 7318 fs_devices->opened = 1; 7319 return fs_devices; 7320 } 7321 7322 /* 7323 * Upon first call for a seed fs fsid, just create a private copy of the 7324 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7325 */ 7326 fs_devices = clone_fs_devices(fs_devices); 7327 if (IS_ERR(fs_devices)) 7328 return fs_devices; 7329 7330 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7331 if (ret) { 7332 free_fs_devices(fs_devices); 7333 return ERR_PTR(ret); 7334 } 7335 7336 if (!fs_devices->seeding) { 7337 close_fs_devices(fs_devices); 7338 free_fs_devices(fs_devices); 7339 return ERR_PTR(-EINVAL); 7340 } 7341 7342 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7343 7344 return fs_devices; 7345 } 7346 7347 static int read_one_dev(struct extent_buffer *leaf, 7348 struct btrfs_dev_item *dev_item) 7349 { 7350 BTRFS_DEV_LOOKUP_ARGS(args); 7351 struct btrfs_fs_info *fs_info = leaf->fs_info; 7352 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7353 struct btrfs_device *device; 7354 u64 devid; 7355 int ret; 7356 u8 fs_uuid[BTRFS_FSID_SIZE]; 7357 u8 dev_uuid[BTRFS_UUID_SIZE]; 7358 7359 devid = btrfs_device_id(leaf, dev_item); 7360 args.devid = devid; 7361 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7362 BTRFS_UUID_SIZE); 7363 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7364 BTRFS_FSID_SIZE); 7365 args.uuid = dev_uuid; 7366 args.fsid = fs_uuid; 7367 7368 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7369 fs_devices = open_seed_devices(fs_info, fs_uuid); 7370 if (IS_ERR(fs_devices)) 7371 return PTR_ERR(fs_devices); 7372 } 7373 7374 device = btrfs_find_device(fs_info->fs_devices, &args); 7375 if (!device) { 7376 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7377 btrfs_report_missing_device(fs_info, devid, 7378 dev_uuid, true); 7379 return -ENOENT; 7380 } 7381 7382 device = add_missing_dev(fs_devices, devid, dev_uuid); 7383 if (IS_ERR(device)) { 7384 btrfs_err(fs_info, 7385 "failed to add missing dev %llu: %ld", 7386 devid, PTR_ERR(device)); 7387 return PTR_ERR(device); 7388 } 7389 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7390 } else { 7391 if (!device->bdev) { 7392 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7393 btrfs_report_missing_device(fs_info, 7394 devid, dev_uuid, true); 7395 return -ENOENT; 7396 } 7397 btrfs_report_missing_device(fs_info, devid, 7398 dev_uuid, false); 7399 } 7400 7401 if (!device->bdev && 7402 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7403 /* 7404 * this happens when a device that was properly setup 7405 * in the device info lists suddenly goes bad. 7406 * device->bdev is NULL, and so we have to set 7407 * device->missing to one here 7408 */ 7409 device->fs_devices->missing_devices++; 7410 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7411 } 7412 7413 /* Move the device to its own fs_devices */ 7414 if (device->fs_devices != fs_devices) { 7415 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7416 &device->dev_state)); 7417 7418 list_move(&device->dev_list, &fs_devices->devices); 7419 device->fs_devices->num_devices--; 7420 fs_devices->num_devices++; 7421 7422 device->fs_devices->missing_devices--; 7423 fs_devices->missing_devices++; 7424 7425 device->fs_devices = fs_devices; 7426 } 7427 } 7428 7429 if (device->fs_devices != fs_info->fs_devices) { 7430 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7431 if (device->generation != 7432 btrfs_device_generation(leaf, dev_item)) 7433 return -EINVAL; 7434 } 7435 7436 fill_device_from_item(leaf, dev_item, device); 7437 if (device->bdev) { 7438 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7439 7440 if (device->total_bytes > max_total_bytes) { 7441 btrfs_err(fs_info, 7442 "device total_bytes should be at most %llu but found %llu", 7443 max_total_bytes, device->total_bytes); 7444 return -EINVAL; 7445 } 7446 } 7447 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7448 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7449 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7450 device->fs_devices->total_rw_bytes += device->total_bytes; 7451 atomic64_add(device->total_bytes - device->bytes_used, 7452 &fs_info->free_chunk_space); 7453 } 7454 ret = 0; 7455 return ret; 7456 } 7457 7458 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7459 { 7460 struct btrfs_super_block *super_copy = fs_info->super_copy; 7461 struct extent_buffer *sb; 7462 struct btrfs_disk_key *disk_key; 7463 struct btrfs_chunk *chunk; 7464 u8 *array_ptr; 7465 unsigned long sb_array_offset; 7466 int ret = 0; 7467 u32 num_stripes; 7468 u32 array_size; 7469 u32 len = 0; 7470 u32 cur_offset; 7471 u64 type; 7472 struct btrfs_key key; 7473 7474 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7475 7476 /* 7477 * We allocated a dummy extent, just to use extent buffer accessors. 7478 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but 7479 * that's fine, we will not go beyond system chunk array anyway. 7480 */ 7481 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); 7482 if (!sb) 7483 return -ENOMEM; 7484 set_extent_buffer_uptodate(sb); 7485 7486 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7487 array_size = btrfs_super_sys_array_size(super_copy); 7488 7489 array_ptr = super_copy->sys_chunk_array; 7490 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7491 cur_offset = 0; 7492 7493 while (cur_offset < array_size) { 7494 disk_key = (struct btrfs_disk_key *)array_ptr; 7495 len = sizeof(*disk_key); 7496 if (cur_offset + len > array_size) 7497 goto out_short_read; 7498 7499 btrfs_disk_key_to_cpu(&key, disk_key); 7500 7501 array_ptr += len; 7502 sb_array_offset += len; 7503 cur_offset += len; 7504 7505 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7506 btrfs_err(fs_info, 7507 "unexpected item type %u in sys_array at offset %u", 7508 (u32)key.type, cur_offset); 7509 ret = -EIO; 7510 break; 7511 } 7512 7513 chunk = (struct btrfs_chunk *)sb_array_offset; 7514 /* 7515 * At least one btrfs_chunk with one stripe must be present, 7516 * exact stripe count check comes afterwards 7517 */ 7518 len = btrfs_chunk_item_size(1); 7519 if (cur_offset + len > array_size) 7520 goto out_short_read; 7521 7522 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7523 if (!num_stripes) { 7524 btrfs_err(fs_info, 7525 "invalid number of stripes %u in sys_array at offset %u", 7526 num_stripes, cur_offset); 7527 ret = -EIO; 7528 break; 7529 } 7530 7531 type = btrfs_chunk_type(sb, chunk); 7532 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7533 btrfs_err(fs_info, 7534 "invalid chunk type %llu in sys_array at offset %u", 7535 type, cur_offset); 7536 ret = -EIO; 7537 break; 7538 } 7539 7540 len = btrfs_chunk_item_size(num_stripes); 7541 if (cur_offset + len > array_size) 7542 goto out_short_read; 7543 7544 ret = read_one_chunk(&key, sb, chunk); 7545 if (ret) 7546 break; 7547 7548 array_ptr += len; 7549 sb_array_offset += len; 7550 cur_offset += len; 7551 } 7552 clear_extent_buffer_uptodate(sb); 7553 free_extent_buffer_stale(sb); 7554 return ret; 7555 7556 out_short_read: 7557 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7558 len, cur_offset); 7559 clear_extent_buffer_uptodate(sb); 7560 free_extent_buffer_stale(sb); 7561 return -EIO; 7562 } 7563 7564 /* 7565 * Check if all chunks in the fs are OK for read-write degraded mount 7566 * 7567 * If the @failing_dev is specified, it's accounted as missing. 7568 * 7569 * Return true if all chunks meet the minimal RW mount requirements. 7570 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7571 */ 7572 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7573 struct btrfs_device *failing_dev) 7574 { 7575 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7576 struct extent_map *em; 7577 u64 next_start = 0; 7578 bool ret = true; 7579 7580 read_lock(&map_tree->lock); 7581 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7582 read_unlock(&map_tree->lock); 7583 /* No chunk at all? Return false anyway */ 7584 if (!em) { 7585 ret = false; 7586 goto out; 7587 } 7588 while (em) { 7589 struct map_lookup *map; 7590 int missing = 0; 7591 int max_tolerated; 7592 int i; 7593 7594 map = em->map_lookup; 7595 max_tolerated = 7596 btrfs_get_num_tolerated_disk_barrier_failures( 7597 map->type); 7598 for (i = 0; i < map->num_stripes; i++) { 7599 struct btrfs_device *dev = map->stripes[i].dev; 7600 7601 if (!dev || !dev->bdev || 7602 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7603 dev->last_flush_error) 7604 missing++; 7605 else if (failing_dev && failing_dev == dev) 7606 missing++; 7607 } 7608 if (missing > max_tolerated) { 7609 if (!failing_dev) 7610 btrfs_warn(fs_info, 7611 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7612 em->start, missing, max_tolerated); 7613 free_extent_map(em); 7614 ret = false; 7615 goto out; 7616 } 7617 next_start = extent_map_end(em); 7618 free_extent_map(em); 7619 7620 read_lock(&map_tree->lock); 7621 em = lookup_extent_mapping(map_tree, next_start, 7622 (u64)(-1) - next_start); 7623 read_unlock(&map_tree->lock); 7624 } 7625 out: 7626 return ret; 7627 } 7628 7629 static void readahead_tree_node_children(struct extent_buffer *node) 7630 { 7631 int i; 7632 const int nr_items = btrfs_header_nritems(node); 7633 7634 for (i = 0; i < nr_items; i++) 7635 btrfs_readahead_node_child(node, i); 7636 } 7637 7638 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7639 { 7640 struct btrfs_root *root = fs_info->chunk_root; 7641 struct btrfs_path *path; 7642 struct extent_buffer *leaf; 7643 struct btrfs_key key; 7644 struct btrfs_key found_key; 7645 int ret; 7646 int slot; 7647 int iter_ret = 0; 7648 u64 total_dev = 0; 7649 u64 last_ra_node = 0; 7650 7651 path = btrfs_alloc_path(); 7652 if (!path) 7653 return -ENOMEM; 7654 7655 /* 7656 * uuid_mutex is needed only if we are mounting a sprout FS 7657 * otherwise we don't need it. 7658 */ 7659 mutex_lock(&uuid_mutex); 7660 7661 /* 7662 * It is possible for mount and umount to race in such a way that 7663 * we execute this code path, but open_fs_devices failed to clear 7664 * total_rw_bytes. We certainly want it cleared before reading the 7665 * device items, so clear it here. 7666 */ 7667 fs_info->fs_devices->total_rw_bytes = 0; 7668 7669 /* 7670 * Lockdep complains about possible circular locking dependency between 7671 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7672 * used for freeze procection of a fs (struct super_block.s_writers), 7673 * which we take when starting a transaction, and extent buffers of the 7674 * chunk tree if we call read_one_dev() while holding a lock on an 7675 * extent buffer of the chunk tree. Since we are mounting the filesystem 7676 * and at this point there can't be any concurrent task modifying the 7677 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7678 */ 7679 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7680 path->skip_locking = 1; 7681 7682 /* 7683 * Read all device items, and then all the chunk items. All 7684 * device items are found before any chunk item (their object id 7685 * is smaller than the lowest possible object id for a chunk 7686 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7687 */ 7688 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7689 key.offset = 0; 7690 key.type = 0; 7691 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 7692 struct extent_buffer *node = path->nodes[1]; 7693 7694 leaf = path->nodes[0]; 7695 slot = path->slots[0]; 7696 7697 if (node) { 7698 if (last_ra_node != node->start) { 7699 readahead_tree_node_children(node); 7700 last_ra_node = node->start; 7701 } 7702 } 7703 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7704 struct btrfs_dev_item *dev_item; 7705 dev_item = btrfs_item_ptr(leaf, slot, 7706 struct btrfs_dev_item); 7707 ret = read_one_dev(leaf, dev_item); 7708 if (ret) 7709 goto error; 7710 total_dev++; 7711 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7712 struct btrfs_chunk *chunk; 7713 7714 /* 7715 * We are only called at mount time, so no need to take 7716 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7717 * we always lock first fs_info->chunk_mutex before 7718 * acquiring any locks on the chunk tree. This is a 7719 * requirement for chunk allocation, see the comment on 7720 * top of btrfs_chunk_alloc() for details. 7721 */ 7722 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7723 ret = read_one_chunk(&found_key, leaf, chunk); 7724 if (ret) 7725 goto error; 7726 } 7727 } 7728 /* Catch error found during iteration */ 7729 if (iter_ret < 0) { 7730 ret = iter_ret; 7731 goto error; 7732 } 7733 7734 /* 7735 * After loading chunk tree, we've got all device information, 7736 * do another round of validation checks. 7737 */ 7738 if (total_dev != fs_info->fs_devices->total_devices) { 7739 btrfs_warn(fs_info, 7740 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", 7741 btrfs_super_num_devices(fs_info->super_copy), 7742 total_dev); 7743 fs_info->fs_devices->total_devices = total_dev; 7744 btrfs_set_super_num_devices(fs_info->super_copy, total_dev); 7745 } 7746 if (btrfs_super_total_bytes(fs_info->super_copy) < 7747 fs_info->fs_devices->total_rw_bytes) { 7748 btrfs_err(fs_info, 7749 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7750 btrfs_super_total_bytes(fs_info->super_copy), 7751 fs_info->fs_devices->total_rw_bytes); 7752 ret = -EINVAL; 7753 goto error; 7754 } 7755 ret = 0; 7756 error: 7757 mutex_unlock(&uuid_mutex); 7758 7759 btrfs_free_path(path); 7760 return ret; 7761 } 7762 7763 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7764 { 7765 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7766 struct btrfs_device *device; 7767 int ret = 0; 7768 7769 fs_devices->fs_info = fs_info; 7770 7771 mutex_lock(&fs_devices->device_list_mutex); 7772 list_for_each_entry(device, &fs_devices->devices, dev_list) 7773 device->fs_info = fs_info; 7774 7775 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7776 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7777 device->fs_info = fs_info; 7778 ret = btrfs_get_dev_zone_info(device, false); 7779 if (ret) 7780 break; 7781 } 7782 7783 seed_devs->fs_info = fs_info; 7784 } 7785 mutex_unlock(&fs_devices->device_list_mutex); 7786 7787 return ret; 7788 } 7789 7790 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7791 const struct btrfs_dev_stats_item *ptr, 7792 int index) 7793 { 7794 u64 val; 7795 7796 read_extent_buffer(eb, &val, 7797 offsetof(struct btrfs_dev_stats_item, values) + 7798 ((unsigned long)ptr) + (index * sizeof(u64)), 7799 sizeof(val)); 7800 return val; 7801 } 7802 7803 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7804 struct btrfs_dev_stats_item *ptr, 7805 int index, u64 val) 7806 { 7807 write_extent_buffer(eb, &val, 7808 offsetof(struct btrfs_dev_stats_item, values) + 7809 ((unsigned long)ptr) + (index * sizeof(u64)), 7810 sizeof(val)); 7811 } 7812 7813 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7814 struct btrfs_path *path) 7815 { 7816 struct btrfs_dev_stats_item *ptr; 7817 struct extent_buffer *eb; 7818 struct btrfs_key key; 7819 int item_size; 7820 int i, ret, slot; 7821 7822 if (!device->fs_info->dev_root) 7823 return 0; 7824 7825 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7826 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7827 key.offset = device->devid; 7828 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7829 if (ret) { 7830 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7831 btrfs_dev_stat_set(device, i, 0); 7832 device->dev_stats_valid = 1; 7833 btrfs_release_path(path); 7834 return ret < 0 ? ret : 0; 7835 } 7836 slot = path->slots[0]; 7837 eb = path->nodes[0]; 7838 item_size = btrfs_item_size(eb, slot); 7839 7840 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7841 7842 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7843 if (item_size >= (1 + i) * sizeof(__le64)) 7844 btrfs_dev_stat_set(device, i, 7845 btrfs_dev_stats_value(eb, ptr, i)); 7846 else 7847 btrfs_dev_stat_set(device, i, 0); 7848 } 7849 7850 device->dev_stats_valid = 1; 7851 btrfs_dev_stat_print_on_load(device); 7852 btrfs_release_path(path); 7853 7854 return 0; 7855 } 7856 7857 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7858 { 7859 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7860 struct btrfs_device *device; 7861 struct btrfs_path *path = NULL; 7862 int ret = 0; 7863 7864 path = btrfs_alloc_path(); 7865 if (!path) 7866 return -ENOMEM; 7867 7868 mutex_lock(&fs_devices->device_list_mutex); 7869 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7870 ret = btrfs_device_init_dev_stats(device, path); 7871 if (ret) 7872 goto out; 7873 } 7874 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7875 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7876 ret = btrfs_device_init_dev_stats(device, path); 7877 if (ret) 7878 goto out; 7879 } 7880 } 7881 out: 7882 mutex_unlock(&fs_devices->device_list_mutex); 7883 7884 btrfs_free_path(path); 7885 return ret; 7886 } 7887 7888 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7889 struct btrfs_device *device) 7890 { 7891 struct btrfs_fs_info *fs_info = trans->fs_info; 7892 struct btrfs_root *dev_root = fs_info->dev_root; 7893 struct btrfs_path *path; 7894 struct btrfs_key key; 7895 struct extent_buffer *eb; 7896 struct btrfs_dev_stats_item *ptr; 7897 int ret; 7898 int i; 7899 7900 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7901 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7902 key.offset = device->devid; 7903 7904 path = btrfs_alloc_path(); 7905 if (!path) 7906 return -ENOMEM; 7907 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7908 if (ret < 0) { 7909 btrfs_warn_in_rcu(fs_info, 7910 "error %d while searching for dev_stats item for device %s", 7911 ret, rcu_str_deref(device->name)); 7912 goto out; 7913 } 7914 7915 if (ret == 0 && 7916 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7917 /* need to delete old one and insert a new one */ 7918 ret = btrfs_del_item(trans, dev_root, path); 7919 if (ret != 0) { 7920 btrfs_warn_in_rcu(fs_info, 7921 "delete too small dev_stats item for device %s failed %d", 7922 rcu_str_deref(device->name), ret); 7923 goto out; 7924 } 7925 ret = 1; 7926 } 7927 7928 if (ret == 1) { 7929 /* need to insert a new item */ 7930 btrfs_release_path(path); 7931 ret = btrfs_insert_empty_item(trans, dev_root, path, 7932 &key, sizeof(*ptr)); 7933 if (ret < 0) { 7934 btrfs_warn_in_rcu(fs_info, 7935 "insert dev_stats item for device %s failed %d", 7936 rcu_str_deref(device->name), ret); 7937 goto out; 7938 } 7939 } 7940 7941 eb = path->nodes[0]; 7942 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7943 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7944 btrfs_set_dev_stats_value(eb, ptr, i, 7945 btrfs_dev_stat_read(device, i)); 7946 btrfs_mark_buffer_dirty(eb); 7947 7948 out: 7949 btrfs_free_path(path); 7950 return ret; 7951 } 7952 7953 /* 7954 * called from commit_transaction. Writes all changed device stats to disk. 7955 */ 7956 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7957 { 7958 struct btrfs_fs_info *fs_info = trans->fs_info; 7959 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7960 struct btrfs_device *device; 7961 int stats_cnt; 7962 int ret = 0; 7963 7964 mutex_lock(&fs_devices->device_list_mutex); 7965 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7966 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7967 if (!device->dev_stats_valid || stats_cnt == 0) 7968 continue; 7969 7970 7971 /* 7972 * There is a LOAD-LOAD control dependency between the value of 7973 * dev_stats_ccnt and updating the on-disk values which requires 7974 * reading the in-memory counters. Such control dependencies 7975 * require explicit read memory barriers. 7976 * 7977 * This memory barriers pairs with smp_mb__before_atomic in 7978 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7979 * barrier implied by atomic_xchg in 7980 * btrfs_dev_stats_read_and_reset 7981 */ 7982 smp_rmb(); 7983 7984 ret = update_dev_stat_item(trans, device); 7985 if (!ret) 7986 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7987 } 7988 mutex_unlock(&fs_devices->device_list_mutex); 7989 7990 return ret; 7991 } 7992 7993 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7994 { 7995 btrfs_dev_stat_inc(dev, index); 7996 7997 if (!dev->dev_stats_valid) 7998 return; 7999 btrfs_err_rl_in_rcu(dev->fs_info, 8000 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 8001 rcu_str_deref(dev->name), 8002 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 8003 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 8004 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 8005 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 8006 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 8007 } 8008 8009 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 8010 { 8011 int i; 8012 8013 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 8014 if (btrfs_dev_stat_read(dev, i) != 0) 8015 break; 8016 if (i == BTRFS_DEV_STAT_VALUES_MAX) 8017 return; /* all values == 0, suppress message */ 8018 8019 btrfs_info_in_rcu(dev->fs_info, 8020 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 8021 rcu_str_deref(dev->name), 8022 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 8023 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 8024 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 8025 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 8026 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 8027 } 8028 8029 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 8030 struct btrfs_ioctl_get_dev_stats *stats) 8031 { 8032 BTRFS_DEV_LOOKUP_ARGS(args); 8033 struct btrfs_device *dev; 8034 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 8035 int i; 8036 8037 mutex_lock(&fs_devices->device_list_mutex); 8038 args.devid = stats->devid; 8039 dev = btrfs_find_device(fs_info->fs_devices, &args); 8040 mutex_unlock(&fs_devices->device_list_mutex); 8041 8042 if (!dev) { 8043 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 8044 return -ENODEV; 8045 } else if (!dev->dev_stats_valid) { 8046 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 8047 return -ENODEV; 8048 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 8049 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 8050 if (stats->nr_items > i) 8051 stats->values[i] = 8052 btrfs_dev_stat_read_and_reset(dev, i); 8053 else 8054 btrfs_dev_stat_set(dev, i, 0); 8055 } 8056 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 8057 current->comm, task_pid_nr(current)); 8058 } else { 8059 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 8060 if (stats->nr_items > i) 8061 stats->values[i] = btrfs_dev_stat_read(dev, i); 8062 } 8063 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 8064 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 8065 return 0; 8066 } 8067 8068 /* 8069 * Update the size and bytes used for each device where it changed. This is 8070 * delayed since we would otherwise get errors while writing out the 8071 * superblocks. 8072 * 8073 * Must be invoked during transaction commit. 8074 */ 8075 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 8076 { 8077 struct btrfs_device *curr, *next; 8078 8079 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 8080 8081 if (list_empty(&trans->dev_update_list)) 8082 return; 8083 8084 /* 8085 * We don't need the device_list_mutex here. This list is owned by the 8086 * transaction and the transaction must complete before the device is 8087 * released. 8088 */ 8089 mutex_lock(&trans->fs_info->chunk_mutex); 8090 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 8091 post_commit_list) { 8092 list_del_init(&curr->post_commit_list); 8093 curr->commit_total_bytes = curr->disk_total_bytes; 8094 curr->commit_bytes_used = curr->bytes_used; 8095 } 8096 mutex_unlock(&trans->fs_info->chunk_mutex); 8097 } 8098 8099 /* 8100 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 8101 */ 8102 int btrfs_bg_type_to_factor(u64 flags) 8103 { 8104 const int index = btrfs_bg_flags_to_raid_index(flags); 8105 8106 return btrfs_raid_array[index].ncopies; 8107 } 8108 8109 8110 8111 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 8112 u64 chunk_offset, u64 devid, 8113 u64 physical_offset, u64 physical_len) 8114 { 8115 struct btrfs_dev_lookup_args args = { .devid = devid }; 8116 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8117 struct extent_map *em; 8118 struct map_lookup *map; 8119 struct btrfs_device *dev; 8120 u64 stripe_len; 8121 bool found = false; 8122 int ret = 0; 8123 int i; 8124 8125 read_lock(&em_tree->lock); 8126 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 8127 read_unlock(&em_tree->lock); 8128 8129 if (!em) { 8130 btrfs_err(fs_info, 8131 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 8132 physical_offset, devid); 8133 ret = -EUCLEAN; 8134 goto out; 8135 } 8136 8137 map = em->map_lookup; 8138 stripe_len = btrfs_calc_stripe_length(em); 8139 if (physical_len != stripe_len) { 8140 btrfs_err(fs_info, 8141 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 8142 physical_offset, devid, em->start, physical_len, 8143 stripe_len); 8144 ret = -EUCLEAN; 8145 goto out; 8146 } 8147 8148 /* 8149 * Very old mkfs.btrfs (before v4.1) will not respect the reserved 8150 * space. Although kernel can handle it without problem, better to warn 8151 * the users. 8152 */ 8153 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) 8154 btrfs_warn(fs_info, 8155 "devid %llu physical %llu len %llu inside the reserved space", 8156 devid, physical_offset, physical_len); 8157 8158 for (i = 0; i < map->num_stripes; i++) { 8159 if (map->stripes[i].dev->devid == devid && 8160 map->stripes[i].physical == physical_offset) { 8161 found = true; 8162 if (map->verified_stripes >= map->num_stripes) { 8163 btrfs_err(fs_info, 8164 "too many dev extents for chunk %llu found", 8165 em->start); 8166 ret = -EUCLEAN; 8167 goto out; 8168 } 8169 map->verified_stripes++; 8170 break; 8171 } 8172 } 8173 if (!found) { 8174 btrfs_err(fs_info, 8175 "dev extent physical offset %llu devid %llu has no corresponding chunk", 8176 physical_offset, devid); 8177 ret = -EUCLEAN; 8178 } 8179 8180 /* Make sure no dev extent is beyond device boundary */ 8181 dev = btrfs_find_device(fs_info->fs_devices, &args); 8182 if (!dev) { 8183 btrfs_err(fs_info, "failed to find devid %llu", devid); 8184 ret = -EUCLEAN; 8185 goto out; 8186 } 8187 8188 if (physical_offset + physical_len > dev->disk_total_bytes) { 8189 btrfs_err(fs_info, 8190 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 8191 devid, physical_offset, physical_len, 8192 dev->disk_total_bytes); 8193 ret = -EUCLEAN; 8194 goto out; 8195 } 8196 8197 if (dev->zone_info) { 8198 u64 zone_size = dev->zone_info->zone_size; 8199 8200 if (!IS_ALIGNED(physical_offset, zone_size) || 8201 !IS_ALIGNED(physical_len, zone_size)) { 8202 btrfs_err(fs_info, 8203 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8204 devid, physical_offset, physical_len); 8205 ret = -EUCLEAN; 8206 goto out; 8207 } 8208 } 8209 8210 out: 8211 free_extent_map(em); 8212 return ret; 8213 } 8214 8215 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8216 { 8217 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8218 struct extent_map *em; 8219 struct rb_node *node; 8220 int ret = 0; 8221 8222 read_lock(&em_tree->lock); 8223 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8224 em = rb_entry(node, struct extent_map, rb_node); 8225 if (em->map_lookup->num_stripes != 8226 em->map_lookup->verified_stripes) { 8227 btrfs_err(fs_info, 8228 "chunk %llu has missing dev extent, have %d expect %d", 8229 em->start, em->map_lookup->verified_stripes, 8230 em->map_lookup->num_stripes); 8231 ret = -EUCLEAN; 8232 goto out; 8233 } 8234 } 8235 out: 8236 read_unlock(&em_tree->lock); 8237 return ret; 8238 } 8239 8240 /* 8241 * Ensure that all dev extents are mapped to correct chunk, otherwise 8242 * later chunk allocation/free would cause unexpected behavior. 8243 * 8244 * NOTE: This will iterate through the whole device tree, which should be of 8245 * the same size level as the chunk tree. This slightly increases mount time. 8246 */ 8247 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8248 { 8249 struct btrfs_path *path; 8250 struct btrfs_root *root = fs_info->dev_root; 8251 struct btrfs_key key; 8252 u64 prev_devid = 0; 8253 u64 prev_dev_ext_end = 0; 8254 int ret = 0; 8255 8256 /* 8257 * We don't have a dev_root because we mounted with ignorebadroots and 8258 * failed to load the root, so we want to skip the verification in this 8259 * case for sure. 8260 * 8261 * However if the dev root is fine, but the tree itself is corrupted 8262 * we'd still fail to mount. This verification is only to make sure 8263 * writes can happen safely, so instead just bypass this check 8264 * completely in the case of IGNOREBADROOTS. 8265 */ 8266 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8267 return 0; 8268 8269 key.objectid = 1; 8270 key.type = BTRFS_DEV_EXTENT_KEY; 8271 key.offset = 0; 8272 8273 path = btrfs_alloc_path(); 8274 if (!path) 8275 return -ENOMEM; 8276 8277 path->reada = READA_FORWARD; 8278 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8279 if (ret < 0) 8280 goto out; 8281 8282 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8283 ret = btrfs_next_leaf(root, path); 8284 if (ret < 0) 8285 goto out; 8286 /* No dev extents at all? Not good */ 8287 if (ret > 0) { 8288 ret = -EUCLEAN; 8289 goto out; 8290 } 8291 } 8292 while (1) { 8293 struct extent_buffer *leaf = path->nodes[0]; 8294 struct btrfs_dev_extent *dext; 8295 int slot = path->slots[0]; 8296 u64 chunk_offset; 8297 u64 physical_offset; 8298 u64 physical_len; 8299 u64 devid; 8300 8301 btrfs_item_key_to_cpu(leaf, &key, slot); 8302 if (key.type != BTRFS_DEV_EXTENT_KEY) 8303 break; 8304 devid = key.objectid; 8305 physical_offset = key.offset; 8306 8307 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8308 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8309 physical_len = btrfs_dev_extent_length(leaf, dext); 8310 8311 /* Check if this dev extent overlaps with the previous one */ 8312 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8313 btrfs_err(fs_info, 8314 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8315 devid, physical_offset, prev_dev_ext_end); 8316 ret = -EUCLEAN; 8317 goto out; 8318 } 8319 8320 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8321 physical_offset, physical_len); 8322 if (ret < 0) 8323 goto out; 8324 prev_devid = devid; 8325 prev_dev_ext_end = physical_offset + physical_len; 8326 8327 ret = btrfs_next_item(root, path); 8328 if (ret < 0) 8329 goto out; 8330 if (ret > 0) { 8331 ret = 0; 8332 break; 8333 } 8334 } 8335 8336 /* Ensure all chunks have corresponding dev extents */ 8337 ret = verify_chunk_dev_extent_mapping(fs_info); 8338 out: 8339 btrfs_free_path(path); 8340 return ret; 8341 } 8342 8343 /* 8344 * Check whether the given block group or device is pinned by any inode being 8345 * used as a swapfile. 8346 */ 8347 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8348 { 8349 struct btrfs_swapfile_pin *sp; 8350 struct rb_node *node; 8351 8352 spin_lock(&fs_info->swapfile_pins_lock); 8353 node = fs_info->swapfile_pins.rb_node; 8354 while (node) { 8355 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8356 if (ptr < sp->ptr) 8357 node = node->rb_left; 8358 else if (ptr > sp->ptr) 8359 node = node->rb_right; 8360 else 8361 break; 8362 } 8363 spin_unlock(&fs_info->swapfile_pins_lock); 8364 return node != NULL; 8365 } 8366 8367 static int relocating_repair_kthread(void *data) 8368 { 8369 struct btrfs_block_group *cache = data; 8370 struct btrfs_fs_info *fs_info = cache->fs_info; 8371 u64 target; 8372 int ret = 0; 8373 8374 target = cache->start; 8375 btrfs_put_block_group(cache); 8376 8377 sb_start_write(fs_info->sb); 8378 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8379 btrfs_info(fs_info, 8380 "zoned: skip relocating block group %llu to repair: EBUSY", 8381 target); 8382 sb_end_write(fs_info->sb); 8383 return -EBUSY; 8384 } 8385 8386 mutex_lock(&fs_info->reclaim_bgs_lock); 8387 8388 /* Ensure block group still exists */ 8389 cache = btrfs_lookup_block_group(fs_info, target); 8390 if (!cache) 8391 goto out; 8392 8393 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) 8394 goto out; 8395 8396 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8397 if (ret < 0) 8398 goto out; 8399 8400 btrfs_info(fs_info, 8401 "zoned: relocating block group %llu to repair IO failure", 8402 target); 8403 ret = btrfs_relocate_chunk(fs_info, target); 8404 8405 out: 8406 if (cache) 8407 btrfs_put_block_group(cache); 8408 mutex_unlock(&fs_info->reclaim_bgs_lock); 8409 btrfs_exclop_finish(fs_info); 8410 sb_end_write(fs_info->sb); 8411 8412 return ret; 8413 } 8414 8415 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8416 { 8417 struct btrfs_block_group *cache; 8418 8419 if (!btrfs_is_zoned(fs_info)) 8420 return false; 8421 8422 /* Do not attempt to repair in degraded state */ 8423 if (btrfs_test_opt(fs_info, DEGRADED)) 8424 return true; 8425 8426 cache = btrfs_lookup_block_group(fs_info, logical); 8427 if (!cache) 8428 return true; 8429 8430 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { 8431 btrfs_put_block_group(cache); 8432 return true; 8433 } 8434 8435 kthread_run(relocating_repair_kthread, cache, 8436 "btrfs-relocating-repair"); 8437 8438 return true; 8439 } 8440 8441 int __init btrfs_bioset_init(void) 8442 { 8443 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, 8444 offsetof(struct btrfs_bio, bio), 8445 BIOSET_NEED_BVECS)) 8446 return -ENOMEM; 8447 return 0; 8448 } 8449 8450 void __cold btrfs_bioset_exit(void) 8451 { 8452 bioset_exit(&btrfs_bioset); 8453 } 8454