1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2007 Oracle. All rights reserved. 4 */ 5 6 #include <linux/sched.h> 7 #include <linux/sched/mm.h> 8 #include <linux/bio.h> 9 #include <linux/slab.h> 10 #include <linux/blkdev.h> 11 #include <linux/ratelimit.h> 12 #include <linux/kthread.h> 13 #include <linux/raid/pq.h> 14 #include <linux/semaphore.h> 15 #include <linux/uuid.h> 16 #include <linux/list_sort.h> 17 #include <linux/namei.h> 18 #include "misc.h" 19 #include "ctree.h" 20 #include "extent_map.h" 21 #include "disk-io.h" 22 #include "transaction.h" 23 #include "print-tree.h" 24 #include "volumes.h" 25 #include "raid56.h" 26 #include "async-thread.h" 27 #include "check-integrity.h" 28 #include "rcu-string.h" 29 #include "dev-replace.h" 30 #include "sysfs.h" 31 #include "tree-checker.h" 32 #include "space-info.h" 33 #include "block-group.h" 34 #include "discard.h" 35 #include "zoned.h" 36 #include "fs.h" 37 #include "accessors.h" 38 39 static struct bio_set btrfs_bioset; 40 41 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ 42 BTRFS_BLOCK_GROUP_RAID10 | \ 43 BTRFS_BLOCK_GROUP_RAID56_MASK) 44 45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { 46 [BTRFS_RAID_RAID10] = { 47 .sub_stripes = 2, 48 .dev_stripes = 1, 49 .devs_max = 0, /* 0 == as many as possible */ 50 .devs_min = 2, 51 .tolerated_failures = 1, 52 .devs_increment = 2, 53 .ncopies = 2, 54 .nparity = 0, 55 .raid_name = "raid10", 56 .bg_flag = BTRFS_BLOCK_GROUP_RAID10, 57 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, 58 }, 59 [BTRFS_RAID_RAID1] = { 60 .sub_stripes = 1, 61 .dev_stripes = 1, 62 .devs_max = 2, 63 .devs_min = 2, 64 .tolerated_failures = 1, 65 .devs_increment = 2, 66 .ncopies = 2, 67 .nparity = 0, 68 .raid_name = "raid1", 69 .bg_flag = BTRFS_BLOCK_GROUP_RAID1, 70 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, 71 }, 72 [BTRFS_RAID_RAID1C3] = { 73 .sub_stripes = 1, 74 .dev_stripes = 1, 75 .devs_max = 3, 76 .devs_min = 3, 77 .tolerated_failures = 2, 78 .devs_increment = 3, 79 .ncopies = 3, 80 .nparity = 0, 81 .raid_name = "raid1c3", 82 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, 83 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, 84 }, 85 [BTRFS_RAID_RAID1C4] = { 86 .sub_stripes = 1, 87 .dev_stripes = 1, 88 .devs_max = 4, 89 .devs_min = 4, 90 .tolerated_failures = 3, 91 .devs_increment = 4, 92 .ncopies = 4, 93 .nparity = 0, 94 .raid_name = "raid1c4", 95 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, 96 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, 97 }, 98 [BTRFS_RAID_DUP] = { 99 .sub_stripes = 1, 100 .dev_stripes = 2, 101 .devs_max = 1, 102 .devs_min = 1, 103 .tolerated_failures = 0, 104 .devs_increment = 1, 105 .ncopies = 2, 106 .nparity = 0, 107 .raid_name = "dup", 108 .bg_flag = BTRFS_BLOCK_GROUP_DUP, 109 .mindev_error = 0, 110 }, 111 [BTRFS_RAID_RAID0] = { 112 .sub_stripes = 1, 113 .dev_stripes = 1, 114 .devs_max = 0, 115 .devs_min = 1, 116 .tolerated_failures = 0, 117 .devs_increment = 1, 118 .ncopies = 1, 119 .nparity = 0, 120 .raid_name = "raid0", 121 .bg_flag = BTRFS_BLOCK_GROUP_RAID0, 122 .mindev_error = 0, 123 }, 124 [BTRFS_RAID_SINGLE] = { 125 .sub_stripes = 1, 126 .dev_stripes = 1, 127 .devs_max = 1, 128 .devs_min = 1, 129 .tolerated_failures = 0, 130 .devs_increment = 1, 131 .ncopies = 1, 132 .nparity = 0, 133 .raid_name = "single", 134 .bg_flag = 0, 135 .mindev_error = 0, 136 }, 137 [BTRFS_RAID_RAID5] = { 138 .sub_stripes = 1, 139 .dev_stripes = 1, 140 .devs_max = 0, 141 .devs_min = 2, 142 .tolerated_failures = 1, 143 .devs_increment = 1, 144 .ncopies = 1, 145 .nparity = 1, 146 .raid_name = "raid5", 147 .bg_flag = BTRFS_BLOCK_GROUP_RAID5, 148 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, 149 }, 150 [BTRFS_RAID_RAID6] = { 151 .sub_stripes = 1, 152 .dev_stripes = 1, 153 .devs_max = 0, 154 .devs_min = 3, 155 .tolerated_failures = 2, 156 .devs_increment = 1, 157 .ncopies = 1, 158 .nparity = 2, 159 .raid_name = "raid6", 160 .bg_flag = BTRFS_BLOCK_GROUP_RAID6, 161 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, 162 }, 163 }; 164 165 /* 166 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 167 * can be used as index to access btrfs_raid_array[]. 168 */ 169 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) 170 { 171 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK); 172 173 if (!profile) 174 return BTRFS_RAID_SINGLE; 175 176 return BTRFS_BG_FLAG_TO_INDEX(profile); 177 } 178 179 const char *btrfs_bg_type_to_raid_name(u64 flags) 180 { 181 const int index = btrfs_bg_flags_to_raid_index(flags); 182 183 if (index >= BTRFS_NR_RAID_TYPES) 184 return NULL; 185 186 return btrfs_raid_array[index].raid_name; 187 } 188 189 int btrfs_nr_parity_stripes(u64 type) 190 { 191 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type); 192 193 return btrfs_raid_array[index].nparity; 194 } 195 196 /* 197 * Fill @buf with textual description of @bg_flags, no more than @size_buf 198 * bytes including terminating null byte. 199 */ 200 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) 201 { 202 int i; 203 int ret; 204 char *bp = buf; 205 u64 flags = bg_flags; 206 u32 size_bp = size_buf; 207 208 if (!flags) { 209 strcpy(bp, "NONE"); 210 return; 211 } 212 213 #define DESCRIBE_FLAG(flag, desc) \ 214 do { \ 215 if (flags & (flag)) { \ 216 ret = snprintf(bp, size_bp, "%s|", (desc)); \ 217 if (ret < 0 || ret >= size_bp) \ 218 goto out_overflow; \ 219 size_bp -= ret; \ 220 bp += ret; \ 221 flags &= ~(flag); \ 222 } \ 223 } while (0) 224 225 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); 226 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); 227 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); 228 229 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); 230 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) 231 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, 232 btrfs_raid_array[i].raid_name); 233 #undef DESCRIBE_FLAG 234 235 if (flags) { 236 ret = snprintf(bp, size_bp, "0x%llx|", flags); 237 size_bp -= ret; 238 } 239 240 if (size_bp < size_buf) 241 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ 242 243 /* 244 * The text is trimmed, it's up to the caller to provide sufficiently 245 * large buffer 246 */ 247 out_overflow:; 248 } 249 250 static int init_first_rw_device(struct btrfs_trans_handle *trans); 251 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); 252 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); 253 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 254 enum btrfs_map_op op, u64 logical, u64 *length, 255 struct btrfs_io_context **bioc_ret, 256 struct btrfs_io_stripe *smap, 257 int *mirror_num_ret, int need_raid_map); 258 259 /* 260 * Device locking 261 * ============== 262 * 263 * There are several mutexes that protect manipulation of devices and low-level 264 * structures like chunks but not block groups, extents or files 265 * 266 * uuid_mutex (global lock) 267 * ------------------------ 268 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from 269 * the SCAN_DEV ioctl registration or from mount either implicitly (the first 270 * device) or requested by the device= mount option 271 * 272 * the mutex can be very coarse and can cover long-running operations 273 * 274 * protects: updates to fs_devices counters like missing devices, rw devices, 275 * seeding, structure cloning, opening/closing devices at mount/umount time 276 * 277 * global::fs_devs - add, remove, updates to the global list 278 * 279 * does not protect: manipulation of the fs_devices::devices list in general 280 * but in mount context it could be used to exclude list modifications by eg. 281 * scan ioctl 282 * 283 * btrfs_device::name - renames (write side), read is RCU 284 * 285 * fs_devices::device_list_mutex (per-fs, with RCU) 286 * ------------------------------------------------ 287 * protects updates to fs_devices::devices, ie. adding and deleting 288 * 289 * simple list traversal with read-only actions can be done with RCU protection 290 * 291 * may be used to exclude some operations from running concurrently without any 292 * modifications to the list (see write_all_supers) 293 * 294 * Is not required at mount and close times, because our device list is 295 * protected by the uuid_mutex at that point. 296 * 297 * balance_mutex 298 * ------------- 299 * protects balance structures (status, state) and context accessed from 300 * several places (internally, ioctl) 301 * 302 * chunk_mutex 303 * ----------- 304 * protects chunks, adding or removing during allocation, trim or when a new 305 * device is added/removed. Additionally it also protects post_commit_list of 306 * individual devices, since they can be added to the transaction's 307 * post_commit_list only with chunk_mutex held. 308 * 309 * cleaner_mutex 310 * ------------- 311 * a big lock that is held by the cleaner thread and prevents running subvolume 312 * cleaning together with relocation or delayed iputs 313 * 314 * 315 * Lock nesting 316 * ============ 317 * 318 * uuid_mutex 319 * device_list_mutex 320 * chunk_mutex 321 * balance_mutex 322 * 323 * 324 * Exclusive operations 325 * ==================== 326 * 327 * Maintains the exclusivity of the following operations that apply to the 328 * whole filesystem and cannot run in parallel. 329 * 330 * - Balance (*) 331 * - Device add 332 * - Device remove 333 * - Device replace (*) 334 * - Resize 335 * 336 * The device operations (as above) can be in one of the following states: 337 * 338 * - Running state 339 * - Paused state 340 * - Completed state 341 * 342 * Only device operations marked with (*) can go into the Paused state for the 343 * following reasons: 344 * 345 * - ioctl (only Balance can be Paused through ioctl) 346 * - filesystem remounted as read-only 347 * - filesystem unmounted and mounted as read-only 348 * - system power-cycle and filesystem mounted as read-only 349 * - filesystem or device errors leading to forced read-only 350 * 351 * The status of exclusive operation is set and cleared atomically. 352 * During the course of Paused state, fs_info::exclusive_operation remains set. 353 * A device operation in Paused or Running state can be canceled or resumed 354 * either by ioctl (Balance only) or when remounted as read-write. 355 * The exclusive status is cleared when the device operation is canceled or 356 * completed. 357 */ 358 359 DEFINE_MUTEX(uuid_mutex); 360 static LIST_HEAD(fs_uuids); 361 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) 362 { 363 return &fs_uuids; 364 } 365 366 /* 367 * alloc_fs_devices - allocate struct btrfs_fs_devices 368 * @fsid: if not NULL, copy the UUID to fs_devices::fsid 369 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid 370 * 371 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). 372 * The returned struct is not linked onto any lists and can be destroyed with 373 * kfree() right away. 374 */ 375 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, 376 const u8 *metadata_fsid) 377 { 378 struct btrfs_fs_devices *fs_devs; 379 380 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 381 if (!fs_devs) 382 return ERR_PTR(-ENOMEM); 383 384 mutex_init(&fs_devs->device_list_mutex); 385 386 INIT_LIST_HEAD(&fs_devs->devices); 387 INIT_LIST_HEAD(&fs_devs->alloc_list); 388 INIT_LIST_HEAD(&fs_devs->fs_list); 389 INIT_LIST_HEAD(&fs_devs->seed_list); 390 if (fsid) 391 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 392 393 if (metadata_fsid) 394 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); 395 else if (fsid) 396 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); 397 398 return fs_devs; 399 } 400 401 void btrfs_free_device(struct btrfs_device *device) 402 { 403 WARN_ON(!list_empty(&device->post_commit_list)); 404 rcu_string_free(device->name); 405 extent_io_tree_release(&device->alloc_state); 406 btrfs_destroy_dev_zone_info(device); 407 kfree(device); 408 } 409 410 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 411 { 412 struct btrfs_device *device; 413 WARN_ON(fs_devices->opened); 414 while (!list_empty(&fs_devices->devices)) { 415 device = list_entry(fs_devices->devices.next, 416 struct btrfs_device, dev_list); 417 list_del(&device->dev_list); 418 btrfs_free_device(device); 419 } 420 kfree(fs_devices); 421 } 422 423 void __exit btrfs_cleanup_fs_uuids(void) 424 { 425 struct btrfs_fs_devices *fs_devices; 426 427 while (!list_empty(&fs_uuids)) { 428 fs_devices = list_entry(fs_uuids.next, 429 struct btrfs_fs_devices, fs_list); 430 list_del(&fs_devices->fs_list); 431 free_fs_devices(fs_devices); 432 } 433 } 434 435 static noinline struct btrfs_fs_devices *find_fsid( 436 const u8 *fsid, const u8 *metadata_fsid) 437 { 438 struct btrfs_fs_devices *fs_devices; 439 440 ASSERT(fsid); 441 442 /* Handle non-split brain cases */ 443 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 444 if (metadata_fsid) { 445 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 446 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 447 BTRFS_FSID_SIZE) == 0) 448 return fs_devices; 449 } else { 450 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 451 return fs_devices; 452 } 453 } 454 return NULL; 455 } 456 457 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 458 struct btrfs_super_block *disk_super) 459 { 460 461 struct btrfs_fs_devices *fs_devices; 462 463 /* 464 * Handle scanned device having completed its fsid change but 465 * belonging to a fs_devices that was created by first scanning 466 * a device which didn't have its fsid/metadata_uuid changed 467 * at all and the CHANGING_FSID_V2 flag set. 468 */ 469 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 470 if (fs_devices->fsid_change && 471 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 472 BTRFS_FSID_SIZE) == 0 && 473 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 474 BTRFS_FSID_SIZE) == 0) { 475 return fs_devices; 476 } 477 } 478 /* 479 * Handle scanned device having completed its fsid change but 480 * belonging to a fs_devices that was created by a device that 481 * has an outdated pair of fsid/metadata_uuid and 482 * CHANGING_FSID_V2 flag set. 483 */ 484 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 485 if (fs_devices->fsid_change && 486 memcmp(fs_devices->metadata_uuid, 487 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 488 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 489 BTRFS_FSID_SIZE) == 0) { 490 return fs_devices; 491 } 492 } 493 494 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 495 } 496 497 498 static int 499 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 500 int flush, struct block_device **bdev, 501 struct btrfs_super_block **disk_super) 502 { 503 int ret; 504 505 *bdev = blkdev_get_by_path(device_path, flags, holder); 506 507 if (IS_ERR(*bdev)) { 508 ret = PTR_ERR(*bdev); 509 goto error; 510 } 511 512 if (flush) 513 sync_blockdev(*bdev); 514 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 515 if (ret) { 516 blkdev_put(*bdev, flags); 517 goto error; 518 } 519 invalidate_bdev(*bdev); 520 *disk_super = btrfs_read_dev_super(*bdev); 521 if (IS_ERR(*disk_super)) { 522 ret = PTR_ERR(*disk_super); 523 blkdev_put(*bdev, flags); 524 goto error; 525 } 526 527 return 0; 528 529 error: 530 *bdev = NULL; 531 return ret; 532 } 533 534 /* 535 * Search and remove all stale devices (which are not mounted). When both 536 * inputs are NULL, it will search and release all stale devices. 537 * 538 * @devt: Optional. When provided will it release all unmounted devices 539 * matching this devt only. 540 * @skip_device: Optional. Will skip this device when searching for the stale 541 * devices. 542 * 543 * Return: 0 for success or if @devt is 0. 544 * -EBUSY if @devt is a mounted device. 545 * -ENOENT if @devt does not match any device in the list. 546 */ 547 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) 548 { 549 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 550 struct btrfs_device *device, *tmp_device; 551 int ret = 0; 552 553 lockdep_assert_held(&uuid_mutex); 554 555 if (devt) 556 ret = -ENOENT; 557 558 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 559 560 mutex_lock(&fs_devices->device_list_mutex); 561 list_for_each_entry_safe(device, tmp_device, 562 &fs_devices->devices, dev_list) { 563 if (skip_device && skip_device == device) 564 continue; 565 if (devt && devt != device->devt) 566 continue; 567 if (fs_devices->opened) { 568 /* for an already deleted device return 0 */ 569 if (devt && ret != 0) 570 ret = -EBUSY; 571 break; 572 } 573 574 /* delete the stale device */ 575 fs_devices->num_devices--; 576 list_del(&device->dev_list); 577 btrfs_free_device(device); 578 579 ret = 0; 580 } 581 mutex_unlock(&fs_devices->device_list_mutex); 582 583 if (fs_devices->num_devices == 0) { 584 btrfs_sysfs_remove_fsid(fs_devices); 585 list_del(&fs_devices->fs_list); 586 free_fs_devices(fs_devices); 587 } 588 } 589 590 return ret; 591 } 592 593 /* 594 * This is only used on mount, and we are protected from competing things 595 * messing with our fs_devices by the uuid_mutex, thus we do not need the 596 * fs_devices->device_list_mutex here. 597 */ 598 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 599 struct btrfs_device *device, fmode_t flags, 600 void *holder) 601 { 602 struct block_device *bdev; 603 struct btrfs_super_block *disk_super; 604 u64 devid; 605 int ret; 606 607 if (device->bdev) 608 return -EINVAL; 609 if (!device->name) 610 return -EINVAL; 611 612 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 613 &bdev, &disk_super); 614 if (ret) 615 return ret; 616 617 devid = btrfs_stack_device_id(&disk_super->dev_item); 618 if (devid != device->devid) 619 goto error_free_page; 620 621 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 622 goto error_free_page; 623 624 device->generation = btrfs_super_generation(disk_super); 625 626 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 627 if (btrfs_super_incompat_flags(disk_super) & 628 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 629 pr_err( 630 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 631 goto error_free_page; 632 } 633 634 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 635 fs_devices->seeding = true; 636 } else { 637 if (bdev_read_only(bdev)) 638 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 639 else 640 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 641 } 642 643 if (!bdev_nonrot(bdev)) 644 fs_devices->rotating = true; 645 646 if (bdev_max_discard_sectors(bdev)) 647 fs_devices->discardable = true; 648 649 device->bdev = bdev; 650 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 651 device->mode = flags; 652 653 fs_devices->open_devices++; 654 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 655 device->devid != BTRFS_DEV_REPLACE_DEVID) { 656 fs_devices->rw_devices++; 657 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 658 } 659 btrfs_release_disk_super(disk_super); 660 661 return 0; 662 663 error_free_page: 664 btrfs_release_disk_super(disk_super); 665 blkdev_put(bdev, flags); 666 667 return -EINVAL; 668 } 669 670 /* 671 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 672 * being created with a disk that has already completed its fsid change. Such 673 * disk can belong to an fs which has its FSID changed or to one which doesn't. 674 * Handle both cases here. 675 */ 676 static struct btrfs_fs_devices *find_fsid_inprogress( 677 struct btrfs_super_block *disk_super) 678 { 679 struct btrfs_fs_devices *fs_devices; 680 681 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 682 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 683 BTRFS_FSID_SIZE) != 0 && 684 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 685 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 686 return fs_devices; 687 } 688 } 689 690 return find_fsid(disk_super->fsid, NULL); 691 } 692 693 694 static struct btrfs_fs_devices *find_fsid_changed( 695 struct btrfs_super_block *disk_super) 696 { 697 struct btrfs_fs_devices *fs_devices; 698 699 /* 700 * Handles the case where scanned device is part of an fs that had 701 * multiple successful changes of FSID but currently device didn't 702 * observe it. Meaning our fsid will be different than theirs. We need 703 * to handle two subcases : 704 * 1 - The fs still continues to have different METADATA/FSID uuids. 705 * 2 - The fs is switched back to its original FSID (METADATA/FSID 706 * are equal). 707 */ 708 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 709 /* Changed UUIDs */ 710 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 711 BTRFS_FSID_SIZE) != 0 && 712 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 713 BTRFS_FSID_SIZE) == 0 && 714 memcmp(fs_devices->fsid, disk_super->fsid, 715 BTRFS_FSID_SIZE) != 0) 716 return fs_devices; 717 718 /* Unchanged UUIDs */ 719 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 720 BTRFS_FSID_SIZE) == 0 && 721 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 722 BTRFS_FSID_SIZE) == 0) 723 return fs_devices; 724 } 725 726 return NULL; 727 } 728 729 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 730 struct btrfs_super_block *disk_super) 731 { 732 struct btrfs_fs_devices *fs_devices; 733 734 /* 735 * Handle the case where the scanned device is part of an fs whose last 736 * metadata UUID change reverted it to the original FSID. At the same 737 * time * fs_devices was first created by another constitutent device 738 * which didn't fully observe the operation. This results in an 739 * btrfs_fs_devices created with metadata/fsid different AND 740 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 741 * fs_devices equal to the FSID of the disk. 742 */ 743 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 744 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 745 BTRFS_FSID_SIZE) != 0 && 746 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 747 BTRFS_FSID_SIZE) == 0 && 748 fs_devices->fsid_change) 749 return fs_devices; 750 } 751 752 return NULL; 753 } 754 /* 755 * Add new device to list of registered devices 756 * 757 * Returns: 758 * device pointer which was just added or updated when successful 759 * error pointer when failed 760 */ 761 static noinline struct btrfs_device *device_list_add(const char *path, 762 struct btrfs_super_block *disk_super, 763 bool *new_device_added) 764 { 765 struct btrfs_device *device; 766 struct btrfs_fs_devices *fs_devices = NULL; 767 struct rcu_string *name; 768 u64 found_transid = btrfs_super_generation(disk_super); 769 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 770 dev_t path_devt; 771 int error; 772 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 773 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 774 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 775 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 776 777 error = lookup_bdev(path, &path_devt); 778 if (error) 779 return ERR_PTR(error); 780 781 if (fsid_change_in_progress) { 782 if (!has_metadata_uuid) 783 fs_devices = find_fsid_inprogress(disk_super); 784 else 785 fs_devices = find_fsid_changed(disk_super); 786 } else if (has_metadata_uuid) { 787 fs_devices = find_fsid_with_metadata_uuid(disk_super); 788 } else { 789 fs_devices = find_fsid_reverted_metadata(disk_super); 790 if (!fs_devices) 791 fs_devices = find_fsid(disk_super->fsid, NULL); 792 } 793 794 795 if (!fs_devices) { 796 if (has_metadata_uuid) 797 fs_devices = alloc_fs_devices(disk_super->fsid, 798 disk_super->metadata_uuid); 799 else 800 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 801 802 if (IS_ERR(fs_devices)) 803 return ERR_CAST(fs_devices); 804 805 fs_devices->fsid_change = fsid_change_in_progress; 806 807 mutex_lock(&fs_devices->device_list_mutex); 808 list_add(&fs_devices->fs_list, &fs_uuids); 809 810 device = NULL; 811 } else { 812 struct btrfs_dev_lookup_args args = { 813 .devid = devid, 814 .uuid = disk_super->dev_item.uuid, 815 }; 816 817 mutex_lock(&fs_devices->device_list_mutex); 818 device = btrfs_find_device(fs_devices, &args); 819 820 /* 821 * If this disk has been pulled into an fs devices created by 822 * a device which had the CHANGING_FSID_V2 flag then replace the 823 * metadata_uuid/fsid values of the fs_devices. 824 */ 825 if (fs_devices->fsid_change && 826 found_transid > fs_devices->latest_generation) { 827 memcpy(fs_devices->fsid, disk_super->fsid, 828 BTRFS_FSID_SIZE); 829 830 if (has_metadata_uuid) 831 memcpy(fs_devices->metadata_uuid, 832 disk_super->metadata_uuid, 833 BTRFS_FSID_SIZE); 834 else 835 memcpy(fs_devices->metadata_uuid, 836 disk_super->fsid, BTRFS_FSID_SIZE); 837 838 fs_devices->fsid_change = false; 839 } 840 } 841 842 if (!device) { 843 if (fs_devices->opened) { 844 mutex_unlock(&fs_devices->device_list_mutex); 845 return ERR_PTR(-EBUSY); 846 } 847 848 device = btrfs_alloc_device(NULL, &devid, 849 disk_super->dev_item.uuid); 850 if (IS_ERR(device)) { 851 mutex_unlock(&fs_devices->device_list_mutex); 852 /* we can safely leave the fs_devices entry around */ 853 return device; 854 } 855 856 name = rcu_string_strdup(path, GFP_NOFS); 857 if (!name) { 858 btrfs_free_device(device); 859 mutex_unlock(&fs_devices->device_list_mutex); 860 return ERR_PTR(-ENOMEM); 861 } 862 rcu_assign_pointer(device->name, name); 863 device->devt = path_devt; 864 865 list_add_rcu(&device->dev_list, &fs_devices->devices); 866 fs_devices->num_devices++; 867 868 device->fs_devices = fs_devices; 869 *new_device_added = true; 870 871 if (disk_super->label[0]) 872 pr_info( 873 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 874 disk_super->label, devid, found_transid, path, 875 current->comm, task_pid_nr(current)); 876 else 877 pr_info( 878 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 879 disk_super->fsid, devid, found_transid, path, 880 current->comm, task_pid_nr(current)); 881 882 } else if (!device->name || strcmp(device->name->str, path)) { 883 /* 884 * When FS is already mounted. 885 * 1. If you are here and if the device->name is NULL that 886 * means this device was missing at time of FS mount. 887 * 2. If you are here and if the device->name is different 888 * from 'path' that means either 889 * a. The same device disappeared and reappeared with 890 * different name. or 891 * b. The missing-disk-which-was-replaced, has 892 * reappeared now. 893 * 894 * We must allow 1 and 2a above. But 2b would be a spurious 895 * and unintentional. 896 * 897 * Further in case of 1 and 2a above, the disk at 'path' 898 * would have missed some transaction when it was away and 899 * in case of 2a the stale bdev has to be updated as well. 900 * 2b must not be allowed at all time. 901 */ 902 903 /* 904 * For now, we do allow update to btrfs_fs_device through the 905 * btrfs dev scan cli after FS has been mounted. We're still 906 * tracking a problem where systems fail mount by subvolume id 907 * when we reject replacement on a mounted FS. 908 */ 909 if (!fs_devices->opened && found_transid < device->generation) { 910 /* 911 * That is if the FS is _not_ mounted and if you 912 * are here, that means there is more than one 913 * disk with same uuid and devid.We keep the one 914 * with larger generation number or the last-in if 915 * generation are equal. 916 */ 917 mutex_unlock(&fs_devices->device_list_mutex); 918 return ERR_PTR(-EEXIST); 919 } 920 921 /* 922 * We are going to replace the device path for a given devid, 923 * make sure it's the same device if the device is mounted 924 * 925 * NOTE: the device->fs_info may not be reliable here so pass 926 * in a NULL to message helpers instead. This avoids a possible 927 * use-after-free when the fs_info and fs_info->sb are already 928 * torn down. 929 */ 930 if (device->bdev) { 931 if (device->devt != path_devt) { 932 mutex_unlock(&fs_devices->device_list_mutex); 933 btrfs_warn_in_rcu(NULL, 934 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 935 path, devid, found_transid, 936 current->comm, 937 task_pid_nr(current)); 938 return ERR_PTR(-EEXIST); 939 } 940 btrfs_info_in_rcu(NULL, 941 "devid %llu device path %s changed to %s scanned by %s (%d)", 942 devid, rcu_str_deref(device->name), 943 path, current->comm, 944 task_pid_nr(current)); 945 } 946 947 name = rcu_string_strdup(path, GFP_NOFS); 948 if (!name) { 949 mutex_unlock(&fs_devices->device_list_mutex); 950 return ERR_PTR(-ENOMEM); 951 } 952 rcu_string_free(device->name); 953 rcu_assign_pointer(device->name, name); 954 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 955 fs_devices->missing_devices--; 956 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 957 } 958 device->devt = path_devt; 959 } 960 961 /* 962 * Unmount does not free the btrfs_device struct but would zero 963 * generation along with most of the other members. So just update 964 * it back. We need it to pick the disk with largest generation 965 * (as above). 966 */ 967 if (!fs_devices->opened) { 968 device->generation = found_transid; 969 fs_devices->latest_generation = max_t(u64, found_transid, 970 fs_devices->latest_generation); 971 } 972 973 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 974 975 mutex_unlock(&fs_devices->device_list_mutex); 976 return device; 977 } 978 979 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 980 { 981 struct btrfs_fs_devices *fs_devices; 982 struct btrfs_device *device; 983 struct btrfs_device *orig_dev; 984 int ret = 0; 985 986 lockdep_assert_held(&uuid_mutex); 987 988 fs_devices = alloc_fs_devices(orig->fsid, NULL); 989 if (IS_ERR(fs_devices)) 990 return fs_devices; 991 992 fs_devices->total_devices = orig->total_devices; 993 994 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 995 struct rcu_string *name; 996 997 device = btrfs_alloc_device(NULL, &orig_dev->devid, 998 orig_dev->uuid); 999 if (IS_ERR(device)) { 1000 ret = PTR_ERR(device); 1001 goto error; 1002 } 1003 1004 /* 1005 * This is ok to do without rcu read locked because we hold the 1006 * uuid mutex so nothing we touch in here is going to disappear. 1007 */ 1008 if (orig_dev->name) { 1009 name = rcu_string_strdup(orig_dev->name->str, 1010 GFP_KERNEL); 1011 if (!name) { 1012 btrfs_free_device(device); 1013 ret = -ENOMEM; 1014 goto error; 1015 } 1016 rcu_assign_pointer(device->name, name); 1017 } 1018 1019 if (orig_dev->zone_info) { 1020 struct btrfs_zoned_device_info *zone_info; 1021 1022 zone_info = btrfs_clone_dev_zone_info(orig_dev); 1023 if (!zone_info) { 1024 btrfs_free_device(device); 1025 ret = -ENOMEM; 1026 goto error; 1027 } 1028 device->zone_info = zone_info; 1029 } 1030 1031 list_add(&device->dev_list, &fs_devices->devices); 1032 device->fs_devices = fs_devices; 1033 fs_devices->num_devices++; 1034 } 1035 return fs_devices; 1036 error: 1037 free_fs_devices(fs_devices); 1038 return ERR_PTR(ret); 1039 } 1040 1041 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1042 struct btrfs_device **latest_dev) 1043 { 1044 struct btrfs_device *device, *next; 1045 1046 /* This is the initialized path, it is safe to release the devices. */ 1047 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1048 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1049 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1050 &device->dev_state) && 1051 !test_bit(BTRFS_DEV_STATE_MISSING, 1052 &device->dev_state) && 1053 (!*latest_dev || 1054 device->generation > (*latest_dev)->generation)) { 1055 *latest_dev = device; 1056 } 1057 continue; 1058 } 1059 1060 /* 1061 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1062 * in btrfs_init_dev_replace() so just continue. 1063 */ 1064 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1065 continue; 1066 1067 if (device->bdev) { 1068 blkdev_put(device->bdev, device->mode); 1069 device->bdev = NULL; 1070 fs_devices->open_devices--; 1071 } 1072 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1073 list_del_init(&device->dev_alloc_list); 1074 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1075 fs_devices->rw_devices--; 1076 } 1077 list_del_init(&device->dev_list); 1078 fs_devices->num_devices--; 1079 btrfs_free_device(device); 1080 } 1081 1082 } 1083 1084 /* 1085 * After we have read the system tree and know devids belonging to this 1086 * filesystem, remove the device which does not belong there. 1087 */ 1088 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1089 { 1090 struct btrfs_device *latest_dev = NULL; 1091 struct btrfs_fs_devices *seed_dev; 1092 1093 mutex_lock(&uuid_mutex); 1094 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1095 1096 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1097 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1098 1099 fs_devices->latest_dev = latest_dev; 1100 1101 mutex_unlock(&uuid_mutex); 1102 } 1103 1104 static void btrfs_close_bdev(struct btrfs_device *device) 1105 { 1106 if (!device->bdev) 1107 return; 1108 1109 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1110 sync_blockdev(device->bdev); 1111 invalidate_bdev(device->bdev); 1112 } 1113 1114 blkdev_put(device->bdev, device->mode); 1115 } 1116 1117 static void btrfs_close_one_device(struct btrfs_device *device) 1118 { 1119 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1120 1121 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1122 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1123 list_del_init(&device->dev_alloc_list); 1124 fs_devices->rw_devices--; 1125 } 1126 1127 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1128 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1129 1130 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 1131 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 1132 fs_devices->missing_devices--; 1133 } 1134 1135 btrfs_close_bdev(device); 1136 if (device->bdev) { 1137 fs_devices->open_devices--; 1138 device->bdev = NULL; 1139 } 1140 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1141 btrfs_destroy_dev_zone_info(device); 1142 1143 device->fs_info = NULL; 1144 atomic_set(&device->dev_stats_ccnt, 0); 1145 extent_io_tree_release(&device->alloc_state); 1146 1147 /* 1148 * Reset the flush error record. We might have a transient flush error 1149 * in this mount, and if so we aborted the current transaction and set 1150 * the fs to an error state, guaranteeing no super blocks can be further 1151 * committed. However that error might be transient and if we unmount the 1152 * filesystem and mount it again, we should allow the mount to succeed 1153 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1154 * filesystem again we still get flush errors, then we will again abort 1155 * any transaction and set the error state, guaranteeing no commits of 1156 * unsafe super blocks. 1157 */ 1158 device->last_flush_error = 0; 1159 1160 /* Verify the device is back in a pristine state */ 1161 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1162 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1163 ASSERT(list_empty(&device->dev_alloc_list)); 1164 ASSERT(list_empty(&device->post_commit_list)); 1165 } 1166 1167 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1168 { 1169 struct btrfs_device *device, *tmp; 1170 1171 lockdep_assert_held(&uuid_mutex); 1172 1173 if (--fs_devices->opened > 0) 1174 return; 1175 1176 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1177 btrfs_close_one_device(device); 1178 1179 WARN_ON(fs_devices->open_devices); 1180 WARN_ON(fs_devices->rw_devices); 1181 fs_devices->opened = 0; 1182 fs_devices->seeding = false; 1183 fs_devices->fs_info = NULL; 1184 } 1185 1186 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1187 { 1188 LIST_HEAD(list); 1189 struct btrfs_fs_devices *tmp; 1190 1191 mutex_lock(&uuid_mutex); 1192 close_fs_devices(fs_devices); 1193 if (!fs_devices->opened) 1194 list_splice_init(&fs_devices->seed_list, &list); 1195 1196 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1197 close_fs_devices(fs_devices); 1198 list_del(&fs_devices->seed_list); 1199 free_fs_devices(fs_devices); 1200 } 1201 mutex_unlock(&uuid_mutex); 1202 } 1203 1204 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1205 fmode_t flags, void *holder) 1206 { 1207 struct btrfs_device *device; 1208 struct btrfs_device *latest_dev = NULL; 1209 struct btrfs_device *tmp_device; 1210 1211 flags |= FMODE_EXCL; 1212 1213 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1214 dev_list) { 1215 int ret; 1216 1217 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1218 if (ret == 0 && 1219 (!latest_dev || device->generation > latest_dev->generation)) { 1220 latest_dev = device; 1221 } else if (ret == -ENODATA) { 1222 fs_devices->num_devices--; 1223 list_del(&device->dev_list); 1224 btrfs_free_device(device); 1225 } 1226 } 1227 if (fs_devices->open_devices == 0) 1228 return -EINVAL; 1229 1230 fs_devices->opened = 1; 1231 fs_devices->latest_dev = latest_dev; 1232 fs_devices->total_rw_bytes = 0; 1233 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1234 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1235 1236 return 0; 1237 } 1238 1239 static int devid_cmp(void *priv, const struct list_head *a, 1240 const struct list_head *b) 1241 { 1242 const struct btrfs_device *dev1, *dev2; 1243 1244 dev1 = list_entry(a, struct btrfs_device, dev_list); 1245 dev2 = list_entry(b, struct btrfs_device, dev_list); 1246 1247 if (dev1->devid < dev2->devid) 1248 return -1; 1249 else if (dev1->devid > dev2->devid) 1250 return 1; 1251 return 0; 1252 } 1253 1254 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1255 fmode_t flags, void *holder) 1256 { 1257 int ret; 1258 1259 lockdep_assert_held(&uuid_mutex); 1260 /* 1261 * The device_list_mutex cannot be taken here in case opening the 1262 * underlying device takes further locks like open_mutex. 1263 * 1264 * We also don't need the lock here as this is called during mount and 1265 * exclusion is provided by uuid_mutex 1266 */ 1267 1268 if (fs_devices->opened) { 1269 fs_devices->opened++; 1270 ret = 0; 1271 } else { 1272 list_sort(NULL, &fs_devices->devices, devid_cmp); 1273 ret = open_fs_devices(fs_devices, flags, holder); 1274 } 1275 1276 return ret; 1277 } 1278 1279 void btrfs_release_disk_super(struct btrfs_super_block *super) 1280 { 1281 struct page *page = virt_to_page(super); 1282 1283 put_page(page); 1284 } 1285 1286 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1287 u64 bytenr, u64 bytenr_orig) 1288 { 1289 struct btrfs_super_block *disk_super; 1290 struct page *page; 1291 void *p; 1292 pgoff_t index; 1293 1294 /* make sure our super fits in the device */ 1295 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) 1296 return ERR_PTR(-EINVAL); 1297 1298 /* make sure our super fits in the page */ 1299 if (sizeof(*disk_super) > PAGE_SIZE) 1300 return ERR_PTR(-EINVAL); 1301 1302 /* make sure our super doesn't straddle pages on disk */ 1303 index = bytenr >> PAGE_SHIFT; 1304 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1305 return ERR_PTR(-EINVAL); 1306 1307 /* pull in the page with our super */ 1308 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1309 1310 if (IS_ERR(page)) 1311 return ERR_CAST(page); 1312 1313 p = page_address(page); 1314 1315 /* align our pointer to the offset of the super block */ 1316 disk_super = p + offset_in_page(bytenr); 1317 1318 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1319 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1320 btrfs_release_disk_super(p); 1321 return ERR_PTR(-EINVAL); 1322 } 1323 1324 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1325 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1326 1327 return disk_super; 1328 } 1329 1330 int btrfs_forget_devices(dev_t devt) 1331 { 1332 int ret; 1333 1334 mutex_lock(&uuid_mutex); 1335 ret = btrfs_free_stale_devices(devt, NULL); 1336 mutex_unlock(&uuid_mutex); 1337 1338 return ret; 1339 } 1340 1341 /* 1342 * Look for a btrfs signature on a device. This may be called out of the mount path 1343 * and we are not allowed to call set_blocksize during the scan. The superblock 1344 * is read via pagecache 1345 */ 1346 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1347 void *holder) 1348 { 1349 struct btrfs_super_block *disk_super; 1350 bool new_device_added = false; 1351 struct btrfs_device *device = NULL; 1352 struct block_device *bdev; 1353 u64 bytenr, bytenr_orig; 1354 int ret; 1355 1356 lockdep_assert_held(&uuid_mutex); 1357 1358 /* 1359 * we would like to check all the supers, but that would make 1360 * a btrfs mount succeed after a mkfs from a different FS. 1361 * So, we need to add a special mount option to scan for 1362 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1363 */ 1364 flags |= FMODE_EXCL; 1365 1366 bdev = blkdev_get_by_path(path, flags, holder); 1367 if (IS_ERR(bdev)) 1368 return ERR_CAST(bdev); 1369 1370 bytenr_orig = btrfs_sb_offset(0); 1371 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1372 if (ret) { 1373 device = ERR_PTR(ret); 1374 goto error_bdev_put; 1375 } 1376 1377 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1378 if (IS_ERR(disk_super)) { 1379 device = ERR_CAST(disk_super); 1380 goto error_bdev_put; 1381 } 1382 1383 device = device_list_add(path, disk_super, &new_device_added); 1384 if (!IS_ERR(device) && new_device_added) 1385 btrfs_free_stale_devices(device->devt, device); 1386 1387 btrfs_release_disk_super(disk_super); 1388 1389 error_bdev_put: 1390 blkdev_put(bdev, flags); 1391 1392 return device; 1393 } 1394 1395 /* 1396 * Try to find a chunk that intersects [start, start + len] range and when one 1397 * such is found, record the end of it in *start 1398 */ 1399 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1400 u64 len) 1401 { 1402 u64 physical_start, physical_end; 1403 1404 lockdep_assert_held(&device->fs_info->chunk_mutex); 1405 1406 if (!find_first_extent_bit(&device->alloc_state, *start, 1407 &physical_start, &physical_end, 1408 CHUNK_ALLOCATED, NULL)) { 1409 1410 if (in_range(physical_start, *start, len) || 1411 in_range(*start, physical_start, 1412 physical_end - physical_start)) { 1413 *start = physical_end + 1; 1414 return true; 1415 } 1416 } 1417 return false; 1418 } 1419 1420 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1421 { 1422 switch (device->fs_devices->chunk_alloc_policy) { 1423 case BTRFS_CHUNK_ALLOC_REGULAR: 1424 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); 1425 case BTRFS_CHUNK_ALLOC_ZONED: 1426 /* 1427 * We don't care about the starting region like regular 1428 * allocator, because we anyway use/reserve the first two zones 1429 * for superblock logging. 1430 */ 1431 return ALIGN(start, device->zone_info->zone_size); 1432 default: 1433 BUG(); 1434 } 1435 } 1436 1437 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1438 u64 *hole_start, u64 *hole_size, 1439 u64 num_bytes) 1440 { 1441 u64 zone_size = device->zone_info->zone_size; 1442 u64 pos; 1443 int ret; 1444 bool changed = false; 1445 1446 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1447 1448 while (*hole_size > 0) { 1449 pos = btrfs_find_allocatable_zones(device, *hole_start, 1450 *hole_start + *hole_size, 1451 num_bytes); 1452 if (pos != *hole_start) { 1453 *hole_size = *hole_start + *hole_size - pos; 1454 *hole_start = pos; 1455 changed = true; 1456 if (*hole_size < num_bytes) 1457 break; 1458 } 1459 1460 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1461 1462 /* Range is ensured to be empty */ 1463 if (!ret) 1464 return changed; 1465 1466 /* Given hole range was invalid (outside of device) */ 1467 if (ret == -ERANGE) { 1468 *hole_start += *hole_size; 1469 *hole_size = 0; 1470 return true; 1471 } 1472 1473 *hole_start += zone_size; 1474 *hole_size -= zone_size; 1475 changed = true; 1476 } 1477 1478 return changed; 1479 } 1480 1481 /* 1482 * Check if specified hole is suitable for allocation. 1483 * 1484 * @device: the device which we have the hole 1485 * @hole_start: starting position of the hole 1486 * @hole_size: the size of the hole 1487 * @num_bytes: the size of the free space that we need 1488 * 1489 * This function may modify @hole_start and @hole_size to reflect the suitable 1490 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1491 */ 1492 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1493 u64 *hole_size, u64 num_bytes) 1494 { 1495 bool changed = false; 1496 u64 hole_end = *hole_start + *hole_size; 1497 1498 for (;;) { 1499 /* 1500 * Check before we set max_hole_start, otherwise we could end up 1501 * sending back this offset anyway. 1502 */ 1503 if (contains_pending_extent(device, hole_start, *hole_size)) { 1504 if (hole_end >= *hole_start) 1505 *hole_size = hole_end - *hole_start; 1506 else 1507 *hole_size = 0; 1508 changed = true; 1509 } 1510 1511 switch (device->fs_devices->chunk_alloc_policy) { 1512 case BTRFS_CHUNK_ALLOC_REGULAR: 1513 /* No extra check */ 1514 break; 1515 case BTRFS_CHUNK_ALLOC_ZONED: 1516 if (dev_extent_hole_check_zoned(device, hole_start, 1517 hole_size, num_bytes)) { 1518 changed = true; 1519 /* 1520 * The changed hole can contain pending extent. 1521 * Loop again to check that. 1522 */ 1523 continue; 1524 } 1525 break; 1526 default: 1527 BUG(); 1528 } 1529 1530 break; 1531 } 1532 1533 return changed; 1534 } 1535 1536 /* 1537 * Find free space in the specified device. 1538 * 1539 * @device: the device which we search the free space in 1540 * @num_bytes: the size of the free space that we need 1541 * @search_start: the position from which to begin the search 1542 * @start: store the start of the free space. 1543 * @len: the size of the free space. that we find, or the size 1544 * of the max free space if we don't find suitable free space 1545 * 1546 * This does a pretty simple search, the expectation is that it is called very 1547 * infrequently and that a given device has a small number of extents. 1548 * 1549 * @start is used to store the start of the free space if we find. But if we 1550 * don't find suitable free space, it will be used to store the start position 1551 * of the max free space. 1552 * 1553 * @len is used to store the size of the free space that we find. 1554 * But if we don't find suitable free space, it is used to store the size of 1555 * the max free space. 1556 * 1557 * NOTE: This function will search *commit* root of device tree, and does extra 1558 * check to ensure dev extents are not double allocated. 1559 * This makes the function safe to allocate dev extents but may not report 1560 * correct usable device space, as device extent freed in current transaction 1561 * is not reported as available. 1562 */ 1563 static int find_free_dev_extent_start(struct btrfs_device *device, 1564 u64 num_bytes, u64 search_start, u64 *start, 1565 u64 *len) 1566 { 1567 struct btrfs_fs_info *fs_info = device->fs_info; 1568 struct btrfs_root *root = fs_info->dev_root; 1569 struct btrfs_key key; 1570 struct btrfs_dev_extent *dev_extent; 1571 struct btrfs_path *path; 1572 u64 hole_size; 1573 u64 max_hole_start; 1574 u64 max_hole_size; 1575 u64 extent_end; 1576 u64 search_end = device->total_bytes; 1577 int ret; 1578 int slot; 1579 struct extent_buffer *l; 1580 1581 search_start = dev_extent_search_start(device, search_start); 1582 1583 WARN_ON(device->zone_info && 1584 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1585 1586 path = btrfs_alloc_path(); 1587 if (!path) 1588 return -ENOMEM; 1589 1590 max_hole_start = search_start; 1591 max_hole_size = 0; 1592 1593 again: 1594 if (search_start >= search_end || 1595 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1596 ret = -ENOSPC; 1597 goto out; 1598 } 1599 1600 path->reada = READA_FORWARD; 1601 path->search_commit_root = 1; 1602 path->skip_locking = 1; 1603 1604 key.objectid = device->devid; 1605 key.offset = search_start; 1606 key.type = BTRFS_DEV_EXTENT_KEY; 1607 1608 ret = btrfs_search_backwards(root, &key, path); 1609 if (ret < 0) 1610 goto out; 1611 1612 while (1) { 1613 l = path->nodes[0]; 1614 slot = path->slots[0]; 1615 if (slot >= btrfs_header_nritems(l)) { 1616 ret = btrfs_next_leaf(root, path); 1617 if (ret == 0) 1618 continue; 1619 if (ret < 0) 1620 goto out; 1621 1622 break; 1623 } 1624 btrfs_item_key_to_cpu(l, &key, slot); 1625 1626 if (key.objectid < device->devid) 1627 goto next; 1628 1629 if (key.objectid > device->devid) 1630 break; 1631 1632 if (key.type != BTRFS_DEV_EXTENT_KEY) 1633 goto next; 1634 1635 if (key.offset > search_start) { 1636 hole_size = key.offset - search_start; 1637 dev_extent_hole_check(device, &search_start, &hole_size, 1638 num_bytes); 1639 1640 if (hole_size > max_hole_size) { 1641 max_hole_start = search_start; 1642 max_hole_size = hole_size; 1643 } 1644 1645 /* 1646 * If this free space is greater than which we need, 1647 * it must be the max free space that we have found 1648 * until now, so max_hole_start must point to the start 1649 * of this free space and the length of this free space 1650 * is stored in max_hole_size. Thus, we return 1651 * max_hole_start and max_hole_size and go back to the 1652 * caller. 1653 */ 1654 if (hole_size >= num_bytes) { 1655 ret = 0; 1656 goto out; 1657 } 1658 } 1659 1660 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1661 extent_end = key.offset + btrfs_dev_extent_length(l, 1662 dev_extent); 1663 if (extent_end > search_start) 1664 search_start = extent_end; 1665 next: 1666 path->slots[0]++; 1667 cond_resched(); 1668 } 1669 1670 /* 1671 * At this point, search_start should be the end of 1672 * allocated dev extents, and when shrinking the device, 1673 * search_end may be smaller than search_start. 1674 */ 1675 if (search_end > search_start) { 1676 hole_size = search_end - search_start; 1677 if (dev_extent_hole_check(device, &search_start, &hole_size, 1678 num_bytes)) { 1679 btrfs_release_path(path); 1680 goto again; 1681 } 1682 1683 if (hole_size > max_hole_size) { 1684 max_hole_start = search_start; 1685 max_hole_size = hole_size; 1686 } 1687 } 1688 1689 /* See above. */ 1690 if (max_hole_size < num_bytes) 1691 ret = -ENOSPC; 1692 else 1693 ret = 0; 1694 1695 out: 1696 btrfs_free_path(path); 1697 *start = max_hole_start; 1698 if (len) 1699 *len = max_hole_size; 1700 return ret; 1701 } 1702 1703 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1704 u64 *start, u64 *len) 1705 { 1706 /* FIXME use last free of some kind */ 1707 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1708 } 1709 1710 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1711 struct btrfs_device *device, 1712 u64 start, u64 *dev_extent_len) 1713 { 1714 struct btrfs_fs_info *fs_info = device->fs_info; 1715 struct btrfs_root *root = fs_info->dev_root; 1716 int ret; 1717 struct btrfs_path *path; 1718 struct btrfs_key key; 1719 struct btrfs_key found_key; 1720 struct extent_buffer *leaf = NULL; 1721 struct btrfs_dev_extent *extent = NULL; 1722 1723 path = btrfs_alloc_path(); 1724 if (!path) 1725 return -ENOMEM; 1726 1727 key.objectid = device->devid; 1728 key.offset = start; 1729 key.type = BTRFS_DEV_EXTENT_KEY; 1730 again: 1731 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1732 if (ret > 0) { 1733 ret = btrfs_previous_item(root, path, key.objectid, 1734 BTRFS_DEV_EXTENT_KEY); 1735 if (ret) 1736 goto out; 1737 leaf = path->nodes[0]; 1738 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1739 extent = btrfs_item_ptr(leaf, path->slots[0], 1740 struct btrfs_dev_extent); 1741 BUG_ON(found_key.offset > start || found_key.offset + 1742 btrfs_dev_extent_length(leaf, extent) < start); 1743 key = found_key; 1744 btrfs_release_path(path); 1745 goto again; 1746 } else if (ret == 0) { 1747 leaf = path->nodes[0]; 1748 extent = btrfs_item_ptr(leaf, path->slots[0], 1749 struct btrfs_dev_extent); 1750 } else { 1751 goto out; 1752 } 1753 1754 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1755 1756 ret = btrfs_del_item(trans, root, path); 1757 if (ret == 0) 1758 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1759 out: 1760 btrfs_free_path(path); 1761 return ret; 1762 } 1763 1764 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1765 { 1766 struct extent_map_tree *em_tree; 1767 struct extent_map *em; 1768 struct rb_node *n; 1769 u64 ret = 0; 1770 1771 em_tree = &fs_info->mapping_tree; 1772 read_lock(&em_tree->lock); 1773 n = rb_last(&em_tree->map.rb_root); 1774 if (n) { 1775 em = rb_entry(n, struct extent_map, rb_node); 1776 ret = em->start + em->len; 1777 } 1778 read_unlock(&em_tree->lock); 1779 1780 return ret; 1781 } 1782 1783 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1784 u64 *devid_ret) 1785 { 1786 int ret; 1787 struct btrfs_key key; 1788 struct btrfs_key found_key; 1789 struct btrfs_path *path; 1790 1791 path = btrfs_alloc_path(); 1792 if (!path) 1793 return -ENOMEM; 1794 1795 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1796 key.type = BTRFS_DEV_ITEM_KEY; 1797 key.offset = (u64)-1; 1798 1799 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1800 if (ret < 0) 1801 goto error; 1802 1803 if (ret == 0) { 1804 /* Corruption */ 1805 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1806 ret = -EUCLEAN; 1807 goto error; 1808 } 1809 1810 ret = btrfs_previous_item(fs_info->chunk_root, path, 1811 BTRFS_DEV_ITEMS_OBJECTID, 1812 BTRFS_DEV_ITEM_KEY); 1813 if (ret) { 1814 *devid_ret = 1; 1815 } else { 1816 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1817 path->slots[0]); 1818 *devid_ret = found_key.offset + 1; 1819 } 1820 ret = 0; 1821 error: 1822 btrfs_free_path(path); 1823 return ret; 1824 } 1825 1826 /* 1827 * the device information is stored in the chunk root 1828 * the btrfs_device struct should be fully filled in 1829 */ 1830 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1831 struct btrfs_device *device) 1832 { 1833 int ret; 1834 struct btrfs_path *path; 1835 struct btrfs_dev_item *dev_item; 1836 struct extent_buffer *leaf; 1837 struct btrfs_key key; 1838 unsigned long ptr; 1839 1840 path = btrfs_alloc_path(); 1841 if (!path) 1842 return -ENOMEM; 1843 1844 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1845 key.type = BTRFS_DEV_ITEM_KEY; 1846 key.offset = device->devid; 1847 1848 btrfs_reserve_chunk_metadata(trans, true); 1849 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1850 &key, sizeof(*dev_item)); 1851 btrfs_trans_release_chunk_metadata(trans); 1852 if (ret) 1853 goto out; 1854 1855 leaf = path->nodes[0]; 1856 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1857 1858 btrfs_set_device_id(leaf, dev_item, device->devid); 1859 btrfs_set_device_generation(leaf, dev_item, 0); 1860 btrfs_set_device_type(leaf, dev_item, device->type); 1861 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1862 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1863 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1864 btrfs_set_device_total_bytes(leaf, dev_item, 1865 btrfs_device_get_disk_total_bytes(device)); 1866 btrfs_set_device_bytes_used(leaf, dev_item, 1867 btrfs_device_get_bytes_used(device)); 1868 btrfs_set_device_group(leaf, dev_item, 0); 1869 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1870 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1871 btrfs_set_device_start_offset(leaf, dev_item, 0); 1872 1873 ptr = btrfs_device_uuid(dev_item); 1874 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1875 ptr = btrfs_device_fsid(dev_item); 1876 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1877 ptr, BTRFS_FSID_SIZE); 1878 btrfs_mark_buffer_dirty(leaf); 1879 1880 ret = 0; 1881 out: 1882 btrfs_free_path(path); 1883 return ret; 1884 } 1885 1886 /* 1887 * Function to update ctime/mtime for a given device path. 1888 * Mainly used for ctime/mtime based probe like libblkid. 1889 * 1890 * We don't care about errors here, this is just to be kind to userspace. 1891 */ 1892 static void update_dev_time(const char *device_path) 1893 { 1894 struct path path; 1895 struct timespec64 now; 1896 int ret; 1897 1898 ret = kern_path(device_path, LOOKUP_FOLLOW, &path); 1899 if (ret) 1900 return; 1901 1902 now = current_time(d_inode(path.dentry)); 1903 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); 1904 path_put(&path); 1905 } 1906 1907 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, 1908 struct btrfs_device *device) 1909 { 1910 struct btrfs_root *root = device->fs_info->chunk_root; 1911 int ret; 1912 struct btrfs_path *path; 1913 struct btrfs_key key; 1914 1915 path = btrfs_alloc_path(); 1916 if (!path) 1917 return -ENOMEM; 1918 1919 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1920 key.type = BTRFS_DEV_ITEM_KEY; 1921 key.offset = device->devid; 1922 1923 btrfs_reserve_chunk_metadata(trans, false); 1924 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1925 btrfs_trans_release_chunk_metadata(trans); 1926 if (ret) { 1927 if (ret > 0) 1928 ret = -ENOENT; 1929 goto out; 1930 } 1931 1932 ret = btrfs_del_item(trans, root, path); 1933 out: 1934 btrfs_free_path(path); 1935 return ret; 1936 } 1937 1938 /* 1939 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1940 * filesystem. It's up to the caller to adjust that number regarding eg. device 1941 * replace. 1942 */ 1943 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1944 u64 num_devices) 1945 { 1946 u64 all_avail; 1947 unsigned seq; 1948 int i; 1949 1950 do { 1951 seq = read_seqbegin(&fs_info->profiles_lock); 1952 1953 all_avail = fs_info->avail_data_alloc_bits | 1954 fs_info->avail_system_alloc_bits | 1955 fs_info->avail_metadata_alloc_bits; 1956 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1957 1958 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1959 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1960 continue; 1961 1962 if (num_devices < btrfs_raid_array[i].devs_min) 1963 return btrfs_raid_array[i].mindev_error; 1964 } 1965 1966 return 0; 1967 } 1968 1969 static struct btrfs_device * btrfs_find_next_active_device( 1970 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1971 { 1972 struct btrfs_device *next_device; 1973 1974 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1975 if (next_device != device && 1976 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1977 && next_device->bdev) 1978 return next_device; 1979 } 1980 1981 return NULL; 1982 } 1983 1984 /* 1985 * Helper function to check if the given device is part of s_bdev / latest_dev 1986 * and replace it with the provided or the next active device, in the context 1987 * where this function called, there should be always be another device (or 1988 * this_dev) which is active. 1989 */ 1990 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 1991 struct btrfs_device *next_device) 1992 { 1993 struct btrfs_fs_info *fs_info = device->fs_info; 1994 1995 if (!next_device) 1996 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 1997 device); 1998 ASSERT(next_device); 1999 2000 if (fs_info->sb->s_bdev && 2001 (fs_info->sb->s_bdev == device->bdev)) 2002 fs_info->sb->s_bdev = next_device->bdev; 2003 2004 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2005 fs_info->fs_devices->latest_dev = next_device; 2006 } 2007 2008 /* 2009 * Return btrfs_fs_devices::num_devices excluding the device that's being 2010 * currently replaced. 2011 */ 2012 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2013 { 2014 u64 num_devices = fs_info->fs_devices->num_devices; 2015 2016 down_read(&fs_info->dev_replace.rwsem); 2017 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2018 ASSERT(num_devices > 1); 2019 num_devices--; 2020 } 2021 up_read(&fs_info->dev_replace.rwsem); 2022 2023 return num_devices; 2024 } 2025 2026 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2027 struct block_device *bdev, 2028 const char *device_path) 2029 { 2030 struct btrfs_super_block *disk_super; 2031 int copy_num; 2032 2033 if (!bdev) 2034 return; 2035 2036 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2037 struct page *page; 2038 int ret; 2039 2040 disk_super = btrfs_read_dev_one_super(bdev, copy_num, false); 2041 if (IS_ERR(disk_super)) 2042 continue; 2043 2044 if (bdev_is_zoned(bdev)) { 2045 btrfs_reset_sb_log_zones(bdev, copy_num); 2046 continue; 2047 } 2048 2049 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 2050 2051 page = virt_to_page(disk_super); 2052 set_page_dirty(page); 2053 lock_page(page); 2054 /* write_on_page() unlocks the page */ 2055 ret = write_one_page(page); 2056 if (ret) 2057 btrfs_warn(fs_info, 2058 "error clearing superblock number %d (%d)", 2059 copy_num, ret); 2060 btrfs_release_disk_super(disk_super); 2061 2062 } 2063 2064 /* Notify udev that device has changed */ 2065 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2066 2067 /* Update ctime/mtime for device path for libblkid */ 2068 update_dev_time(device_path); 2069 } 2070 2071 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2072 struct btrfs_dev_lookup_args *args, 2073 struct block_device **bdev, fmode_t *mode) 2074 { 2075 struct btrfs_trans_handle *trans; 2076 struct btrfs_device *device; 2077 struct btrfs_fs_devices *cur_devices; 2078 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2079 u64 num_devices; 2080 int ret = 0; 2081 2082 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2083 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); 2084 return -EINVAL; 2085 } 2086 2087 /* 2088 * The device list in fs_devices is accessed without locks (neither 2089 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2090 * filesystem and another device rm cannot run. 2091 */ 2092 num_devices = btrfs_num_devices(fs_info); 2093 2094 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2095 if (ret) 2096 return ret; 2097 2098 device = btrfs_find_device(fs_info->fs_devices, args); 2099 if (!device) { 2100 if (args->missing) 2101 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2102 else 2103 ret = -ENOENT; 2104 return ret; 2105 } 2106 2107 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2108 btrfs_warn_in_rcu(fs_info, 2109 "cannot remove device %s (devid %llu) due to active swapfile", 2110 rcu_str_deref(device->name), device->devid); 2111 return -ETXTBSY; 2112 } 2113 2114 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 2115 return BTRFS_ERROR_DEV_TGT_REPLACE; 2116 2117 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2118 fs_info->fs_devices->rw_devices == 1) 2119 return BTRFS_ERROR_DEV_ONLY_WRITABLE; 2120 2121 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2122 mutex_lock(&fs_info->chunk_mutex); 2123 list_del_init(&device->dev_alloc_list); 2124 device->fs_devices->rw_devices--; 2125 mutex_unlock(&fs_info->chunk_mutex); 2126 } 2127 2128 ret = btrfs_shrink_device(device, 0); 2129 if (ret) 2130 goto error_undo; 2131 2132 trans = btrfs_start_transaction(fs_info->chunk_root, 0); 2133 if (IS_ERR(trans)) { 2134 ret = PTR_ERR(trans); 2135 goto error_undo; 2136 } 2137 2138 ret = btrfs_rm_dev_item(trans, device); 2139 if (ret) { 2140 /* Any error in dev item removal is critical */ 2141 btrfs_crit(fs_info, 2142 "failed to remove device item for devid %llu: %d", 2143 device->devid, ret); 2144 btrfs_abort_transaction(trans, ret); 2145 btrfs_end_transaction(trans); 2146 return ret; 2147 } 2148 2149 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2150 btrfs_scrub_cancel_dev(device); 2151 2152 /* 2153 * the device list mutex makes sure that we don't change 2154 * the device list while someone else is writing out all 2155 * the device supers. Whoever is writing all supers, should 2156 * lock the device list mutex before getting the number of 2157 * devices in the super block (super_copy). Conversely, 2158 * whoever updates the number of devices in the super block 2159 * (super_copy) should hold the device list mutex. 2160 */ 2161 2162 /* 2163 * In normal cases the cur_devices == fs_devices. But in case 2164 * of deleting a seed device, the cur_devices should point to 2165 * its own fs_devices listed under the fs_devices->seed_list. 2166 */ 2167 cur_devices = device->fs_devices; 2168 mutex_lock(&fs_devices->device_list_mutex); 2169 list_del_rcu(&device->dev_list); 2170 2171 cur_devices->num_devices--; 2172 cur_devices->total_devices--; 2173 /* Update total_devices of the parent fs_devices if it's seed */ 2174 if (cur_devices != fs_devices) 2175 fs_devices->total_devices--; 2176 2177 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2178 cur_devices->missing_devices--; 2179 2180 btrfs_assign_next_active_device(device, NULL); 2181 2182 if (device->bdev) { 2183 cur_devices->open_devices--; 2184 /* remove sysfs entry */ 2185 btrfs_sysfs_remove_device(device); 2186 } 2187 2188 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2189 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2190 mutex_unlock(&fs_devices->device_list_mutex); 2191 2192 /* 2193 * At this point, the device is zero sized and detached from the 2194 * devices list. All that's left is to zero out the old supers and 2195 * free the device. 2196 * 2197 * We cannot call btrfs_close_bdev() here because we're holding the sb 2198 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2199 * block device and it's dependencies. Instead just flush the device 2200 * and let the caller do the final blkdev_put. 2201 */ 2202 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2203 btrfs_scratch_superblocks(fs_info, device->bdev, 2204 device->name->str); 2205 if (device->bdev) { 2206 sync_blockdev(device->bdev); 2207 invalidate_bdev(device->bdev); 2208 } 2209 } 2210 2211 *bdev = device->bdev; 2212 *mode = device->mode; 2213 synchronize_rcu(); 2214 btrfs_free_device(device); 2215 2216 /* 2217 * This can happen if cur_devices is the private seed devices list. We 2218 * cannot call close_fs_devices() here because it expects the uuid_mutex 2219 * to be held, but in fact we don't need that for the private 2220 * seed_devices, we can simply decrement cur_devices->opened and then 2221 * remove it from our list and free the fs_devices. 2222 */ 2223 if (cur_devices->num_devices == 0) { 2224 list_del_init(&cur_devices->seed_list); 2225 ASSERT(cur_devices->opened == 1); 2226 cur_devices->opened--; 2227 free_fs_devices(cur_devices); 2228 } 2229 2230 ret = btrfs_commit_transaction(trans); 2231 2232 return ret; 2233 2234 error_undo: 2235 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2236 mutex_lock(&fs_info->chunk_mutex); 2237 list_add(&device->dev_alloc_list, 2238 &fs_devices->alloc_list); 2239 device->fs_devices->rw_devices++; 2240 mutex_unlock(&fs_info->chunk_mutex); 2241 } 2242 return ret; 2243 } 2244 2245 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2246 { 2247 struct btrfs_fs_devices *fs_devices; 2248 2249 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2250 2251 /* 2252 * in case of fs with no seed, srcdev->fs_devices will point 2253 * to fs_devices of fs_info. However when the dev being replaced is 2254 * a seed dev it will point to the seed's local fs_devices. In short 2255 * srcdev will have its correct fs_devices in both the cases. 2256 */ 2257 fs_devices = srcdev->fs_devices; 2258 2259 list_del_rcu(&srcdev->dev_list); 2260 list_del(&srcdev->dev_alloc_list); 2261 fs_devices->num_devices--; 2262 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2263 fs_devices->missing_devices--; 2264 2265 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2266 fs_devices->rw_devices--; 2267 2268 if (srcdev->bdev) 2269 fs_devices->open_devices--; 2270 } 2271 2272 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2273 { 2274 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2275 2276 mutex_lock(&uuid_mutex); 2277 2278 btrfs_close_bdev(srcdev); 2279 synchronize_rcu(); 2280 btrfs_free_device(srcdev); 2281 2282 /* if this is no devs we rather delete the fs_devices */ 2283 if (!fs_devices->num_devices) { 2284 /* 2285 * On a mounted FS, num_devices can't be zero unless it's a 2286 * seed. In case of a seed device being replaced, the replace 2287 * target added to the sprout FS, so there will be no more 2288 * device left under the seed FS. 2289 */ 2290 ASSERT(fs_devices->seeding); 2291 2292 list_del_init(&fs_devices->seed_list); 2293 close_fs_devices(fs_devices); 2294 free_fs_devices(fs_devices); 2295 } 2296 mutex_unlock(&uuid_mutex); 2297 } 2298 2299 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2300 { 2301 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2302 2303 mutex_lock(&fs_devices->device_list_mutex); 2304 2305 btrfs_sysfs_remove_device(tgtdev); 2306 2307 if (tgtdev->bdev) 2308 fs_devices->open_devices--; 2309 2310 fs_devices->num_devices--; 2311 2312 btrfs_assign_next_active_device(tgtdev, NULL); 2313 2314 list_del_rcu(&tgtdev->dev_list); 2315 2316 mutex_unlock(&fs_devices->device_list_mutex); 2317 2318 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2319 tgtdev->name->str); 2320 2321 btrfs_close_bdev(tgtdev); 2322 synchronize_rcu(); 2323 btrfs_free_device(tgtdev); 2324 } 2325 2326 /* 2327 * Populate args from device at path. 2328 * 2329 * @fs_info: the filesystem 2330 * @args: the args to populate 2331 * @path: the path to the device 2332 * 2333 * This will read the super block of the device at @path and populate @args with 2334 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2335 * lookup a device to operate on, but need to do it before we take any locks. 2336 * This properly handles the special case of "missing" that a user may pass in, 2337 * and does some basic sanity checks. The caller must make sure that @path is 2338 * properly NUL terminated before calling in, and must call 2339 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2340 * uuid buffers. 2341 * 2342 * Return: 0 for success, -errno for failure 2343 */ 2344 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2345 struct btrfs_dev_lookup_args *args, 2346 const char *path) 2347 { 2348 struct btrfs_super_block *disk_super; 2349 struct block_device *bdev; 2350 int ret; 2351 2352 if (!path || !path[0]) 2353 return -EINVAL; 2354 if (!strcmp(path, "missing")) { 2355 args->missing = true; 2356 return 0; 2357 } 2358 2359 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2360 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2361 if (!args->uuid || !args->fsid) { 2362 btrfs_put_dev_args_from_path(args); 2363 return -ENOMEM; 2364 } 2365 2366 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2367 &bdev, &disk_super); 2368 if (ret) { 2369 btrfs_put_dev_args_from_path(args); 2370 return ret; 2371 } 2372 2373 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2374 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2375 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2376 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2377 else 2378 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2379 btrfs_release_disk_super(disk_super); 2380 blkdev_put(bdev, FMODE_READ); 2381 return 0; 2382 } 2383 2384 /* 2385 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2386 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2387 * that don't need to be freed. 2388 */ 2389 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2390 { 2391 kfree(args->uuid); 2392 kfree(args->fsid); 2393 args->uuid = NULL; 2394 args->fsid = NULL; 2395 } 2396 2397 struct btrfs_device *btrfs_find_device_by_devspec( 2398 struct btrfs_fs_info *fs_info, u64 devid, 2399 const char *device_path) 2400 { 2401 BTRFS_DEV_LOOKUP_ARGS(args); 2402 struct btrfs_device *device; 2403 int ret; 2404 2405 if (devid) { 2406 args.devid = devid; 2407 device = btrfs_find_device(fs_info->fs_devices, &args); 2408 if (!device) 2409 return ERR_PTR(-ENOENT); 2410 return device; 2411 } 2412 2413 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2414 if (ret) 2415 return ERR_PTR(ret); 2416 device = btrfs_find_device(fs_info->fs_devices, &args); 2417 btrfs_put_dev_args_from_path(&args); 2418 if (!device) 2419 return ERR_PTR(-ENOENT); 2420 return device; 2421 } 2422 2423 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) 2424 { 2425 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2426 struct btrfs_fs_devices *old_devices; 2427 struct btrfs_fs_devices *seed_devices; 2428 2429 lockdep_assert_held(&uuid_mutex); 2430 if (!fs_devices->seeding) 2431 return ERR_PTR(-EINVAL); 2432 2433 /* 2434 * Private copy of the seed devices, anchored at 2435 * fs_info->fs_devices->seed_list 2436 */ 2437 seed_devices = alloc_fs_devices(NULL, NULL); 2438 if (IS_ERR(seed_devices)) 2439 return seed_devices; 2440 2441 /* 2442 * It's necessary to retain a copy of the original seed fs_devices in 2443 * fs_uuids so that filesystems which have been seeded can successfully 2444 * reference the seed device from open_seed_devices. This also supports 2445 * multiple fs seed. 2446 */ 2447 old_devices = clone_fs_devices(fs_devices); 2448 if (IS_ERR(old_devices)) { 2449 kfree(seed_devices); 2450 return old_devices; 2451 } 2452 2453 list_add(&old_devices->fs_list, &fs_uuids); 2454 2455 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2456 seed_devices->opened = 1; 2457 INIT_LIST_HEAD(&seed_devices->devices); 2458 INIT_LIST_HEAD(&seed_devices->alloc_list); 2459 mutex_init(&seed_devices->device_list_mutex); 2460 2461 return seed_devices; 2462 } 2463 2464 /* 2465 * Splice seed devices into the sprout fs_devices. 2466 * Generate a new fsid for the sprouted read-write filesystem. 2467 */ 2468 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, 2469 struct btrfs_fs_devices *seed_devices) 2470 { 2471 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2472 struct btrfs_super_block *disk_super = fs_info->super_copy; 2473 struct btrfs_device *device; 2474 u64 super_flags; 2475 2476 /* 2477 * We are updating the fsid, the thread leading to device_list_add() 2478 * could race, so uuid_mutex is needed. 2479 */ 2480 lockdep_assert_held(&uuid_mutex); 2481 2482 /* 2483 * The threads listed below may traverse dev_list but can do that without 2484 * device_list_mutex: 2485 * - All device ops and balance - as we are in btrfs_exclop_start. 2486 * - Various dev_list readers - are using RCU. 2487 * - btrfs_ioctl_fitrim() - is using RCU. 2488 * 2489 * For-read threads as below are using device_list_mutex: 2490 * - Readonly scrub btrfs_scrub_dev() 2491 * - Readonly scrub btrfs_scrub_progress() 2492 * - btrfs_get_dev_stats() 2493 */ 2494 lockdep_assert_held(&fs_devices->device_list_mutex); 2495 2496 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2497 synchronize_rcu); 2498 list_for_each_entry(device, &seed_devices->devices, dev_list) 2499 device->fs_devices = seed_devices; 2500 2501 fs_devices->seeding = false; 2502 fs_devices->num_devices = 0; 2503 fs_devices->open_devices = 0; 2504 fs_devices->missing_devices = 0; 2505 fs_devices->rotating = false; 2506 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2507 2508 generate_random_uuid(fs_devices->fsid); 2509 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2510 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2511 2512 super_flags = btrfs_super_flags(disk_super) & 2513 ~BTRFS_SUPER_FLAG_SEEDING; 2514 btrfs_set_super_flags(disk_super, super_flags); 2515 } 2516 2517 /* 2518 * Store the expected generation for seed devices in device items. 2519 */ 2520 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2521 { 2522 BTRFS_DEV_LOOKUP_ARGS(args); 2523 struct btrfs_fs_info *fs_info = trans->fs_info; 2524 struct btrfs_root *root = fs_info->chunk_root; 2525 struct btrfs_path *path; 2526 struct extent_buffer *leaf; 2527 struct btrfs_dev_item *dev_item; 2528 struct btrfs_device *device; 2529 struct btrfs_key key; 2530 u8 fs_uuid[BTRFS_FSID_SIZE]; 2531 u8 dev_uuid[BTRFS_UUID_SIZE]; 2532 int ret; 2533 2534 path = btrfs_alloc_path(); 2535 if (!path) 2536 return -ENOMEM; 2537 2538 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2539 key.offset = 0; 2540 key.type = BTRFS_DEV_ITEM_KEY; 2541 2542 while (1) { 2543 btrfs_reserve_chunk_metadata(trans, false); 2544 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2545 btrfs_trans_release_chunk_metadata(trans); 2546 if (ret < 0) 2547 goto error; 2548 2549 leaf = path->nodes[0]; 2550 next_slot: 2551 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2552 ret = btrfs_next_leaf(root, path); 2553 if (ret > 0) 2554 break; 2555 if (ret < 0) 2556 goto error; 2557 leaf = path->nodes[0]; 2558 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2559 btrfs_release_path(path); 2560 continue; 2561 } 2562 2563 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2564 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2565 key.type != BTRFS_DEV_ITEM_KEY) 2566 break; 2567 2568 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2569 struct btrfs_dev_item); 2570 args.devid = btrfs_device_id(leaf, dev_item); 2571 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2572 BTRFS_UUID_SIZE); 2573 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2574 BTRFS_FSID_SIZE); 2575 args.uuid = dev_uuid; 2576 args.fsid = fs_uuid; 2577 device = btrfs_find_device(fs_info->fs_devices, &args); 2578 BUG_ON(!device); /* Logic error */ 2579 2580 if (device->fs_devices->seeding) { 2581 btrfs_set_device_generation(leaf, dev_item, 2582 device->generation); 2583 btrfs_mark_buffer_dirty(leaf); 2584 } 2585 2586 path->slots[0]++; 2587 goto next_slot; 2588 } 2589 ret = 0; 2590 error: 2591 btrfs_free_path(path); 2592 return ret; 2593 } 2594 2595 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2596 { 2597 struct btrfs_root *root = fs_info->dev_root; 2598 struct btrfs_trans_handle *trans; 2599 struct btrfs_device *device; 2600 struct block_device *bdev; 2601 struct super_block *sb = fs_info->sb; 2602 struct rcu_string *name; 2603 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2604 struct btrfs_fs_devices *seed_devices; 2605 u64 orig_super_total_bytes; 2606 u64 orig_super_num_devices; 2607 int ret = 0; 2608 bool seeding_dev = false; 2609 bool locked = false; 2610 2611 if (sb_rdonly(sb) && !fs_devices->seeding) 2612 return -EROFS; 2613 2614 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2615 fs_info->bdev_holder); 2616 if (IS_ERR(bdev)) 2617 return PTR_ERR(bdev); 2618 2619 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2620 ret = -EINVAL; 2621 goto error; 2622 } 2623 2624 if (fs_devices->seeding) { 2625 seeding_dev = true; 2626 down_write(&sb->s_umount); 2627 mutex_lock(&uuid_mutex); 2628 locked = true; 2629 } 2630 2631 sync_blockdev(bdev); 2632 2633 rcu_read_lock(); 2634 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2635 if (device->bdev == bdev) { 2636 ret = -EEXIST; 2637 rcu_read_unlock(); 2638 goto error; 2639 } 2640 } 2641 rcu_read_unlock(); 2642 2643 device = btrfs_alloc_device(fs_info, NULL, NULL); 2644 if (IS_ERR(device)) { 2645 /* we can safely leave the fs_devices entry around */ 2646 ret = PTR_ERR(device); 2647 goto error; 2648 } 2649 2650 name = rcu_string_strdup(device_path, GFP_KERNEL); 2651 if (!name) { 2652 ret = -ENOMEM; 2653 goto error_free_device; 2654 } 2655 rcu_assign_pointer(device->name, name); 2656 2657 device->fs_info = fs_info; 2658 device->bdev = bdev; 2659 ret = lookup_bdev(device_path, &device->devt); 2660 if (ret) 2661 goto error_free_device; 2662 2663 ret = btrfs_get_dev_zone_info(device, false); 2664 if (ret) 2665 goto error_free_device; 2666 2667 trans = btrfs_start_transaction(root, 0); 2668 if (IS_ERR(trans)) { 2669 ret = PTR_ERR(trans); 2670 goto error_free_zone; 2671 } 2672 2673 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2674 device->generation = trans->transid; 2675 device->io_width = fs_info->sectorsize; 2676 device->io_align = fs_info->sectorsize; 2677 device->sector_size = fs_info->sectorsize; 2678 device->total_bytes = 2679 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); 2680 device->disk_total_bytes = device->total_bytes; 2681 device->commit_total_bytes = device->total_bytes; 2682 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2683 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2684 device->mode = FMODE_EXCL; 2685 device->dev_stats_valid = 1; 2686 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2687 2688 if (seeding_dev) { 2689 btrfs_clear_sb_rdonly(sb); 2690 2691 /* GFP_KERNEL allocation must not be under device_list_mutex */ 2692 seed_devices = btrfs_init_sprout(fs_info); 2693 if (IS_ERR(seed_devices)) { 2694 ret = PTR_ERR(seed_devices); 2695 btrfs_abort_transaction(trans, ret); 2696 goto error_trans; 2697 } 2698 } 2699 2700 mutex_lock(&fs_devices->device_list_mutex); 2701 if (seeding_dev) { 2702 btrfs_setup_sprout(fs_info, seed_devices); 2703 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2704 device); 2705 } 2706 2707 device->fs_devices = fs_devices; 2708 2709 mutex_lock(&fs_info->chunk_mutex); 2710 list_add_rcu(&device->dev_list, &fs_devices->devices); 2711 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2712 fs_devices->num_devices++; 2713 fs_devices->open_devices++; 2714 fs_devices->rw_devices++; 2715 fs_devices->total_devices++; 2716 fs_devices->total_rw_bytes += device->total_bytes; 2717 2718 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2719 2720 if (!bdev_nonrot(bdev)) 2721 fs_devices->rotating = true; 2722 2723 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2724 btrfs_set_super_total_bytes(fs_info->super_copy, 2725 round_down(orig_super_total_bytes + device->total_bytes, 2726 fs_info->sectorsize)); 2727 2728 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2729 btrfs_set_super_num_devices(fs_info->super_copy, 2730 orig_super_num_devices + 1); 2731 2732 /* 2733 * we've got more storage, clear any full flags on the space 2734 * infos 2735 */ 2736 btrfs_clear_space_info_full(fs_info); 2737 2738 mutex_unlock(&fs_info->chunk_mutex); 2739 2740 /* Add sysfs device entry */ 2741 btrfs_sysfs_add_device(device); 2742 2743 mutex_unlock(&fs_devices->device_list_mutex); 2744 2745 if (seeding_dev) { 2746 mutex_lock(&fs_info->chunk_mutex); 2747 ret = init_first_rw_device(trans); 2748 mutex_unlock(&fs_info->chunk_mutex); 2749 if (ret) { 2750 btrfs_abort_transaction(trans, ret); 2751 goto error_sysfs; 2752 } 2753 } 2754 2755 ret = btrfs_add_dev_item(trans, device); 2756 if (ret) { 2757 btrfs_abort_transaction(trans, ret); 2758 goto error_sysfs; 2759 } 2760 2761 if (seeding_dev) { 2762 ret = btrfs_finish_sprout(trans); 2763 if (ret) { 2764 btrfs_abort_transaction(trans, ret); 2765 goto error_sysfs; 2766 } 2767 2768 /* 2769 * fs_devices now represents the newly sprouted filesystem and 2770 * its fsid has been changed by btrfs_sprout_splice(). 2771 */ 2772 btrfs_sysfs_update_sprout_fsid(fs_devices); 2773 } 2774 2775 ret = btrfs_commit_transaction(trans); 2776 2777 if (seeding_dev) { 2778 mutex_unlock(&uuid_mutex); 2779 up_write(&sb->s_umount); 2780 locked = false; 2781 2782 if (ret) /* transaction commit */ 2783 return ret; 2784 2785 ret = btrfs_relocate_sys_chunks(fs_info); 2786 if (ret < 0) 2787 btrfs_handle_fs_error(fs_info, ret, 2788 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2789 trans = btrfs_attach_transaction(root); 2790 if (IS_ERR(trans)) { 2791 if (PTR_ERR(trans) == -ENOENT) 2792 return 0; 2793 ret = PTR_ERR(trans); 2794 trans = NULL; 2795 goto error_sysfs; 2796 } 2797 ret = btrfs_commit_transaction(trans); 2798 } 2799 2800 /* 2801 * Now that we have written a new super block to this device, check all 2802 * other fs_devices list if device_path alienates any other scanned 2803 * device. 2804 * We can ignore the return value as it typically returns -EINVAL and 2805 * only succeeds if the device was an alien. 2806 */ 2807 btrfs_forget_devices(device->devt); 2808 2809 /* Update ctime/mtime for blkid or udev */ 2810 update_dev_time(device_path); 2811 2812 return ret; 2813 2814 error_sysfs: 2815 btrfs_sysfs_remove_device(device); 2816 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2817 mutex_lock(&fs_info->chunk_mutex); 2818 list_del_rcu(&device->dev_list); 2819 list_del(&device->dev_alloc_list); 2820 fs_info->fs_devices->num_devices--; 2821 fs_info->fs_devices->open_devices--; 2822 fs_info->fs_devices->rw_devices--; 2823 fs_info->fs_devices->total_devices--; 2824 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2825 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2826 btrfs_set_super_total_bytes(fs_info->super_copy, 2827 orig_super_total_bytes); 2828 btrfs_set_super_num_devices(fs_info->super_copy, 2829 orig_super_num_devices); 2830 mutex_unlock(&fs_info->chunk_mutex); 2831 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2832 error_trans: 2833 if (seeding_dev) 2834 btrfs_set_sb_rdonly(sb); 2835 if (trans) 2836 btrfs_end_transaction(trans); 2837 error_free_zone: 2838 btrfs_destroy_dev_zone_info(device); 2839 error_free_device: 2840 btrfs_free_device(device); 2841 error: 2842 blkdev_put(bdev, FMODE_EXCL); 2843 if (locked) { 2844 mutex_unlock(&uuid_mutex); 2845 up_write(&sb->s_umount); 2846 } 2847 return ret; 2848 } 2849 2850 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2851 struct btrfs_device *device) 2852 { 2853 int ret; 2854 struct btrfs_path *path; 2855 struct btrfs_root *root = device->fs_info->chunk_root; 2856 struct btrfs_dev_item *dev_item; 2857 struct extent_buffer *leaf; 2858 struct btrfs_key key; 2859 2860 path = btrfs_alloc_path(); 2861 if (!path) 2862 return -ENOMEM; 2863 2864 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2865 key.type = BTRFS_DEV_ITEM_KEY; 2866 key.offset = device->devid; 2867 2868 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2869 if (ret < 0) 2870 goto out; 2871 2872 if (ret > 0) { 2873 ret = -ENOENT; 2874 goto out; 2875 } 2876 2877 leaf = path->nodes[0]; 2878 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2879 2880 btrfs_set_device_id(leaf, dev_item, device->devid); 2881 btrfs_set_device_type(leaf, dev_item, device->type); 2882 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2883 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2884 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2885 btrfs_set_device_total_bytes(leaf, dev_item, 2886 btrfs_device_get_disk_total_bytes(device)); 2887 btrfs_set_device_bytes_used(leaf, dev_item, 2888 btrfs_device_get_bytes_used(device)); 2889 btrfs_mark_buffer_dirty(leaf); 2890 2891 out: 2892 btrfs_free_path(path); 2893 return ret; 2894 } 2895 2896 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2897 struct btrfs_device *device, u64 new_size) 2898 { 2899 struct btrfs_fs_info *fs_info = device->fs_info; 2900 struct btrfs_super_block *super_copy = fs_info->super_copy; 2901 u64 old_total; 2902 u64 diff; 2903 int ret; 2904 2905 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2906 return -EACCES; 2907 2908 new_size = round_down(new_size, fs_info->sectorsize); 2909 2910 mutex_lock(&fs_info->chunk_mutex); 2911 old_total = btrfs_super_total_bytes(super_copy); 2912 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2913 2914 if (new_size <= device->total_bytes || 2915 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2916 mutex_unlock(&fs_info->chunk_mutex); 2917 return -EINVAL; 2918 } 2919 2920 btrfs_set_super_total_bytes(super_copy, 2921 round_down(old_total + diff, fs_info->sectorsize)); 2922 device->fs_devices->total_rw_bytes += diff; 2923 2924 btrfs_device_set_total_bytes(device, new_size); 2925 btrfs_device_set_disk_total_bytes(device, new_size); 2926 btrfs_clear_space_info_full(device->fs_info); 2927 if (list_empty(&device->post_commit_list)) 2928 list_add_tail(&device->post_commit_list, 2929 &trans->transaction->dev_update_list); 2930 mutex_unlock(&fs_info->chunk_mutex); 2931 2932 btrfs_reserve_chunk_metadata(trans, false); 2933 ret = btrfs_update_device(trans, device); 2934 btrfs_trans_release_chunk_metadata(trans); 2935 2936 return ret; 2937 } 2938 2939 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2940 { 2941 struct btrfs_fs_info *fs_info = trans->fs_info; 2942 struct btrfs_root *root = fs_info->chunk_root; 2943 int ret; 2944 struct btrfs_path *path; 2945 struct btrfs_key key; 2946 2947 path = btrfs_alloc_path(); 2948 if (!path) 2949 return -ENOMEM; 2950 2951 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2952 key.offset = chunk_offset; 2953 key.type = BTRFS_CHUNK_ITEM_KEY; 2954 2955 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2956 if (ret < 0) 2957 goto out; 2958 else if (ret > 0) { /* Logic error or corruption */ 2959 btrfs_handle_fs_error(fs_info, -ENOENT, 2960 "Failed lookup while freeing chunk."); 2961 ret = -ENOENT; 2962 goto out; 2963 } 2964 2965 ret = btrfs_del_item(trans, root, path); 2966 if (ret < 0) 2967 btrfs_handle_fs_error(fs_info, ret, 2968 "Failed to delete chunk item."); 2969 out: 2970 btrfs_free_path(path); 2971 return ret; 2972 } 2973 2974 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2975 { 2976 struct btrfs_super_block *super_copy = fs_info->super_copy; 2977 struct btrfs_disk_key *disk_key; 2978 struct btrfs_chunk *chunk; 2979 u8 *ptr; 2980 int ret = 0; 2981 u32 num_stripes; 2982 u32 array_size; 2983 u32 len = 0; 2984 u32 cur; 2985 struct btrfs_key key; 2986 2987 lockdep_assert_held(&fs_info->chunk_mutex); 2988 array_size = btrfs_super_sys_array_size(super_copy); 2989 2990 ptr = super_copy->sys_chunk_array; 2991 cur = 0; 2992 2993 while (cur < array_size) { 2994 disk_key = (struct btrfs_disk_key *)ptr; 2995 btrfs_disk_key_to_cpu(&key, disk_key); 2996 2997 len = sizeof(*disk_key); 2998 2999 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 3000 chunk = (struct btrfs_chunk *)(ptr + len); 3001 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 3002 len += btrfs_chunk_item_size(num_stripes); 3003 } else { 3004 ret = -EIO; 3005 break; 3006 } 3007 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 3008 key.offset == chunk_offset) { 3009 memmove(ptr, ptr + len, array_size - (cur + len)); 3010 array_size -= len; 3011 btrfs_set_super_sys_array_size(super_copy, array_size); 3012 } else { 3013 ptr += len; 3014 cur += len; 3015 } 3016 } 3017 return ret; 3018 } 3019 3020 /* 3021 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 3022 * @logical: Logical block offset in bytes. 3023 * @length: Length of extent in bytes. 3024 * 3025 * Return: Chunk mapping or ERR_PTR. 3026 */ 3027 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 3028 u64 logical, u64 length) 3029 { 3030 struct extent_map_tree *em_tree; 3031 struct extent_map *em; 3032 3033 em_tree = &fs_info->mapping_tree; 3034 read_lock(&em_tree->lock); 3035 em = lookup_extent_mapping(em_tree, logical, length); 3036 read_unlock(&em_tree->lock); 3037 3038 if (!em) { 3039 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 3040 logical, length); 3041 return ERR_PTR(-EINVAL); 3042 } 3043 3044 if (em->start > logical || em->start + em->len < logical) { 3045 btrfs_crit(fs_info, 3046 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3047 logical, length, em->start, em->start + em->len); 3048 free_extent_map(em); 3049 return ERR_PTR(-EINVAL); 3050 } 3051 3052 /* callers are responsible for dropping em's ref. */ 3053 return em; 3054 } 3055 3056 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3057 struct map_lookup *map, u64 chunk_offset) 3058 { 3059 int i; 3060 3061 /* 3062 * Removing chunk items and updating the device items in the chunks btree 3063 * requires holding the chunk_mutex. 3064 * See the comment at btrfs_chunk_alloc() for the details. 3065 */ 3066 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3067 3068 for (i = 0; i < map->num_stripes; i++) { 3069 int ret; 3070 3071 ret = btrfs_update_device(trans, map->stripes[i].dev); 3072 if (ret) 3073 return ret; 3074 } 3075 3076 return btrfs_free_chunk(trans, chunk_offset); 3077 } 3078 3079 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3080 { 3081 struct btrfs_fs_info *fs_info = trans->fs_info; 3082 struct extent_map *em; 3083 struct map_lookup *map; 3084 u64 dev_extent_len = 0; 3085 int i, ret = 0; 3086 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3087 3088 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3089 if (IS_ERR(em)) { 3090 /* 3091 * This is a logic error, but we don't want to just rely on the 3092 * user having built with ASSERT enabled, so if ASSERT doesn't 3093 * do anything we still error out. 3094 */ 3095 ASSERT(0); 3096 return PTR_ERR(em); 3097 } 3098 map = em->map_lookup; 3099 3100 /* 3101 * First delete the device extent items from the devices btree. 3102 * We take the device_list_mutex to avoid racing with the finishing phase 3103 * of a device replace operation. See the comment below before acquiring 3104 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3105 * because that can result in a deadlock when deleting the device extent 3106 * items from the devices btree - COWing an extent buffer from the btree 3107 * may result in allocating a new metadata chunk, which would attempt to 3108 * lock again fs_info->chunk_mutex. 3109 */ 3110 mutex_lock(&fs_devices->device_list_mutex); 3111 for (i = 0; i < map->num_stripes; i++) { 3112 struct btrfs_device *device = map->stripes[i].dev; 3113 ret = btrfs_free_dev_extent(trans, device, 3114 map->stripes[i].physical, 3115 &dev_extent_len); 3116 if (ret) { 3117 mutex_unlock(&fs_devices->device_list_mutex); 3118 btrfs_abort_transaction(trans, ret); 3119 goto out; 3120 } 3121 3122 if (device->bytes_used > 0) { 3123 mutex_lock(&fs_info->chunk_mutex); 3124 btrfs_device_set_bytes_used(device, 3125 device->bytes_used - dev_extent_len); 3126 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3127 btrfs_clear_space_info_full(fs_info); 3128 mutex_unlock(&fs_info->chunk_mutex); 3129 } 3130 } 3131 mutex_unlock(&fs_devices->device_list_mutex); 3132 3133 /* 3134 * We acquire fs_info->chunk_mutex for 2 reasons: 3135 * 3136 * 1) Just like with the first phase of the chunk allocation, we must 3137 * reserve system space, do all chunk btree updates and deletions, and 3138 * update the system chunk array in the superblock while holding this 3139 * mutex. This is for similar reasons as explained on the comment at 3140 * the top of btrfs_chunk_alloc(); 3141 * 3142 * 2) Prevent races with the final phase of a device replace operation 3143 * that replaces the device object associated with the map's stripes, 3144 * because the device object's id can change at any time during that 3145 * final phase of the device replace operation 3146 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3147 * replaced device and then see it with an ID of 3148 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3149 * the device item, which does not exists on the chunk btree. 3150 * The finishing phase of device replace acquires both the 3151 * device_list_mutex and the chunk_mutex, in that order, so we are 3152 * safe by just acquiring the chunk_mutex. 3153 */ 3154 trans->removing_chunk = true; 3155 mutex_lock(&fs_info->chunk_mutex); 3156 3157 check_system_chunk(trans, map->type); 3158 3159 ret = remove_chunk_item(trans, map, chunk_offset); 3160 /* 3161 * Normally we should not get -ENOSPC since we reserved space before 3162 * through the call to check_system_chunk(). 3163 * 3164 * Despite our system space_info having enough free space, we may not 3165 * be able to allocate extents from its block groups, because all have 3166 * an incompatible profile, which will force us to allocate a new system 3167 * block group with the right profile, or right after we called 3168 * check_system_space() above, a scrub turned the only system block group 3169 * with enough free space into RO mode. 3170 * This is explained with more detail at do_chunk_alloc(). 3171 * 3172 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3173 */ 3174 if (ret == -ENOSPC) { 3175 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3176 struct btrfs_block_group *sys_bg; 3177 3178 sys_bg = btrfs_create_chunk(trans, sys_flags); 3179 if (IS_ERR(sys_bg)) { 3180 ret = PTR_ERR(sys_bg); 3181 btrfs_abort_transaction(trans, ret); 3182 goto out; 3183 } 3184 3185 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3186 if (ret) { 3187 btrfs_abort_transaction(trans, ret); 3188 goto out; 3189 } 3190 3191 ret = remove_chunk_item(trans, map, chunk_offset); 3192 if (ret) { 3193 btrfs_abort_transaction(trans, ret); 3194 goto out; 3195 } 3196 } else if (ret) { 3197 btrfs_abort_transaction(trans, ret); 3198 goto out; 3199 } 3200 3201 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3202 3203 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3204 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3205 if (ret) { 3206 btrfs_abort_transaction(trans, ret); 3207 goto out; 3208 } 3209 } 3210 3211 mutex_unlock(&fs_info->chunk_mutex); 3212 trans->removing_chunk = false; 3213 3214 /* 3215 * We are done with chunk btree updates and deletions, so release the 3216 * system space we previously reserved (with check_system_chunk()). 3217 */ 3218 btrfs_trans_release_chunk_metadata(trans); 3219 3220 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3221 if (ret) { 3222 btrfs_abort_transaction(trans, ret); 3223 goto out; 3224 } 3225 3226 out: 3227 if (trans->removing_chunk) { 3228 mutex_unlock(&fs_info->chunk_mutex); 3229 trans->removing_chunk = false; 3230 } 3231 /* once for us */ 3232 free_extent_map(em); 3233 return ret; 3234 } 3235 3236 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3237 { 3238 struct btrfs_root *root = fs_info->chunk_root; 3239 struct btrfs_trans_handle *trans; 3240 struct btrfs_block_group *block_group; 3241 u64 length; 3242 int ret; 3243 3244 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3245 btrfs_err(fs_info, 3246 "relocate: not supported on extent tree v2 yet"); 3247 return -EINVAL; 3248 } 3249 3250 /* 3251 * Prevent races with automatic removal of unused block groups. 3252 * After we relocate and before we remove the chunk with offset 3253 * chunk_offset, automatic removal of the block group can kick in, 3254 * resulting in a failure when calling btrfs_remove_chunk() below. 3255 * 3256 * Make sure to acquire this mutex before doing a tree search (dev 3257 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3258 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3259 * we release the path used to search the chunk/dev tree and before 3260 * the current task acquires this mutex and calls us. 3261 */ 3262 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3263 3264 /* step one, relocate all the extents inside this chunk */ 3265 btrfs_scrub_pause(fs_info); 3266 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3267 btrfs_scrub_continue(fs_info); 3268 if (ret) 3269 return ret; 3270 3271 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3272 if (!block_group) 3273 return -ENOENT; 3274 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3275 length = block_group->length; 3276 btrfs_put_block_group(block_group); 3277 3278 /* 3279 * On a zoned file system, discard the whole block group, this will 3280 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3281 * resetting the zone fails, don't treat it as a fatal problem from the 3282 * filesystem's point of view. 3283 */ 3284 if (btrfs_is_zoned(fs_info)) { 3285 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3286 if (ret) 3287 btrfs_info(fs_info, 3288 "failed to reset zone %llu after relocation", 3289 chunk_offset); 3290 } 3291 3292 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3293 chunk_offset); 3294 if (IS_ERR(trans)) { 3295 ret = PTR_ERR(trans); 3296 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3297 return ret; 3298 } 3299 3300 /* 3301 * step two, delete the device extents and the 3302 * chunk tree entries 3303 */ 3304 ret = btrfs_remove_chunk(trans, chunk_offset); 3305 btrfs_end_transaction(trans); 3306 return ret; 3307 } 3308 3309 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3310 { 3311 struct btrfs_root *chunk_root = fs_info->chunk_root; 3312 struct btrfs_path *path; 3313 struct extent_buffer *leaf; 3314 struct btrfs_chunk *chunk; 3315 struct btrfs_key key; 3316 struct btrfs_key found_key; 3317 u64 chunk_type; 3318 bool retried = false; 3319 int failed = 0; 3320 int ret; 3321 3322 path = btrfs_alloc_path(); 3323 if (!path) 3324 return -ENOMEM; 3325 3326 again: 3327 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3328 key.offset = (u64)-1; 3329 key.type = BTRFS_CHUNK_ITEM_KEY; 3330 3331 while (1) { 3332 mutex_lock(&fs_info->reclaim_bgs_lock); 3333 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3334 if (ret < 0) { 3335 mutex_unlock(&fs_info->reclaim_bgs_lock); 3336 goto error; 3337 } 3338 BUG_ON(ret == 0); /* Corruption */ 3339 3340 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3341 key.type); 3342 if (ret) 3343 mutex_unlock(&fs_info->reclaim_bgs_lock); 3344 if (ret < 0) 3345 goto error; 3346 if (ret > 0) 3347 break; 3348 3349 leaf = path->nodes[0]; 3350 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3351 3352 chunk = btrfs_item_ptr(leaf, path->slots[0], 3353 struct btrfs_chunk); 3354 chunk_type = btrfs_chunk_type(leaf, chunk); 3355 btrfs_release_path(path); 3356 3357 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3358 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3359 if (ret == -ENOSPC) 3360 failed++; 3361 else 3362 BUG_ON(ret); 3363 } 3364 mutex_unlock(&fs_info->reclaim_bgs_lock); 3365 3366 if (found_key.offset == 0) 3367 break; 3368 key.offset = found_key.offset - 1; 3369 } 3370 ret = 0; 3371 if (failed && !retried) { 3372 failed = 0; 3373 retried = true; 3374 goto again; 3375 } else if (WARN_ON(failed && retried)) { 3376 ret = -ENOSPC; 3377 } 3378 error: 3379 btrfs_free_path(path); 3380 return ret; 3381 } 3382 3383 /* 3384 * return 1 : allocate a data chunk successfully, 3385 * return <0: errors during allocating a data chunk, 3386 * return 0 : no need to allocate a data chunk. 3387 */ 3388 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3389 u64 chunk_offset) 3390 { 3391 struct btrfs_block_group *cache; 3392 u64 bytes_used; 3393 u64 chunk_type; 3394 3395 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3396 ASSERT(cache); 3397 chunk_type = cache->flags; 3398 btrfs_put_block_group(cache); 3399 3400 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3401 return 0; 3402 3403 spin_lock(&fs_info->data_sinfo->lock); 3404 bytes_used = fs_info->data_sinfo->bytes_used; 3405 spin_unlock(&fs_info->data_sinfo->lock); 3406 3407 if (!bytes_used) { 3408 struct btrfs_trans_handle *trans; 3409 int ret; 3410 3411 trans = btrfs_join_transaction(fs_info->tree_root); 3412 if (IS_ERR(trans)) 3413 return PTR_ERR(trans); 3414 3415 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3416 btrfs_end_transaction(trans); 3417 if (ret < 0) 3418 return ret; 3419 return 1; 3420 } 3421 3422 return 0; 3423 } 3424 3425 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3426 struct btrfs_balance_control *bctl) 3427 { 3428 struct btrfs_root *root = fs_info->tree_root; 3429 struct btrfs_trans_handle *trans; 3430 struct btrfs_balance_item *item; 3431 struct btrfs_disk_balance_args disk_bargs; 3432 struct btrfs_path *path; 3433 struct extent_buffer *leaf; 3434 struct btrfs_key key; 3435 int ret, err; 3436 3437 path = btrfs_alloc_path(); 3438 if (!path) 3439 return -ENOMEM; 3440 3441 trans = btrfs_start_transaction(root, 0); 3442 if (IS_ERR(trans)) { 3443 btrfs_free_path(path); 3444 return PTR_ERR(trans); 3445 } 3446 3447 key.objectid = BTRFS_BALANCE_OBJECTID; 3448 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3449 key.offset = 0; 3450 3451 ret = btrfs_insert_empty_item(trans, root, path, &key, 3452 sizeof(*item)); 3453 if (ret) 3454 goto out; 3455 3456 leaf = path->nodes[0]; 3457 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3458 3459 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3460 3461 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3462 btrfs_set_balance_data(leaf, item, &disk_bargs); 3463 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3464 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3465 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3466 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3467 3468 btrfs_set_balance_flags(leaf, item, bctl->flags); 3469 3470 btrfs_mark_buffer_dirty(leaf); 3471 out: 3472 btrfs_free_path(path); 3473 err = btrfs_commit_transaction(trans); 3474 if (err && !ret) 3475 ret = err; 3476 return ret; 3477 } 3478 3479 static int del_balance_item(struct btrfs_fs_info *fs_info) 3480 { 3481 struct btrfs_root *root = fs_info->tree_root; 3482 struct btrfs_trans_handle *trans; 3483 struct btrfs_path *path; 3484 struct btrfs_key key; 3485 int ret, err; 3486 3487 path = btrfs_alloc_path(); 3488 if (!path) 3489 return -ENOMEM; 3490 3491 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3492 if (IS_ERR(trans)) { 3493 btrfs_free_path(path); 3494 return PTR_ERR(trans); 3495 } 3496 3497 key.objectid = BTRFS_BALANCE_OBJECTID; 3498 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3499 key.offset = 0; 3500 3501 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3502 if (ret < 0) 3503 goto out; 3504 if (ret > 0) { 3505 ret = -ENOENT; 3506 goto out; 3507 } 3508 3509 ret = btrfs_del_item(trans, root, path); 3510 out: 3511 btrfs_free_path(path); 3512 err = btrfs_commit_transaction(trans); 3513 if (err && !ret) 3514 ret = err; 3515 return ret; 3516 } 3517 3518 /* 3519 * This is a heuristic used to reduce the number of chunks balanced on 3520 * resume after balance was interrupted. 3521 */ 3522 static void update_balance_args(struct btrfs_balance_control *bctl) 3523 { 3524 /* 3525 * Turn on soft mode for chunk types that were being converted. 3526 */ 3527 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3528 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3529 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3530 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3531 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3532 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3533 3534 /* 3535 * Turn on usage filter if is not already used. The idea is 3536 * that chunks that we have already balanced should be 3537 * reasonably full. Don't do it for chunks that are being 3538 * converted - that will keep us from relocating unconverted 3539 * (albeit full) chunks. 3540 */ 3541 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3542 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3543 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3544 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3545 bctl->data.usage = 90; 3546 } 3547 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3548 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3549 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3550 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3551 bctl->sys.usage = 90; 3552 } 3553 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3554 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3555 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3556 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3557 bctl->meta.usage = 90; 3558 } 3559 } 3560 3561 /* 3562 * Clear the balance status in fs_info and delete the balance item from disk. 3563 */ 3564 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3565 { 3566 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3567 int ret; 3568 3569 BUG_ON(!fs_info->balance_ctl); 3570 3571 spin_lock(&fs_info->balance_lock); 3572 fs_info->balance_ctl = NULL; 3573 spin_unlock(&fs_info->balance_lock); 3574 3575 kfree(bctl); 3576 ret = del_balance_item(fs_info); 3577 if (ret) 3578 btrfs_handle_fs_error(fs_info, ret, NULL); 3579 } 3580 3581 /* 3582 * Balance filters. Return 1 if chunk should be filtered out 3583 * (should not be balanced). 3584 */ 3585 static int chunk_profiles_filter(u64 chunk_type, 3586 struct btrfs_balance_args *bargs) 3587 { 3588 chunk_type = chunk_to_extended(chunk_type) & 3589 BTRFS_EXTENDED_PROFILE_MASK; 3590 3591 if (bargs->profiles & chunk_type) 3592 return 0; 3593 3594 return 1; 3595 } 3596 3597 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3598 struct btrfs_balance_args *bargs) 3599 { 3600 struct btrfs_block_group *cache; 3601 u64 chunk_used; 3602 u64 user_thresh_min; 3603 u64 user_thresh_max; 3604 int ret = 1; 3605 3606 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3607 chunk_used = cache->used; 3608 3609 if (bargs->usage_min == 0) 3610 user_thresh_min = 0; 3611 else 3612 user_thresh_min = div_factor_fine(cache->length, 3613 bargs->usage_min); 3614 3615 if (bargs->usage_max == 0) 3616 user_thresh_max = 1; 3617 else if (bargs->usage_max > 100) 3618 user_thresh_max = cache->length; 3619 else 3620 user_thresh_max = div_factor_fine(cache->length, 3621 bargs->usage_max); 3622 3623 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3624 ret = 0; 3625 3626 btrfs_put_block_group(cache); 3627 return ret; 3628 } 3629 3630 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3631 u64 chunk_offset, struct btrfs_balance_args *bargs) 3632 { 3633 struct btrfs_block_group *cache; 3634 u64 chunk_used, user_thresh; 3635 int ret = 1; 3636 3637 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3638 chunk_used = cache->used; 3639 3640 if (bargs->usage_min == 0) 3641 user_thresh = 1; 3642 else if (bargs->usage > 100) 3643 user_thresh = cache->length; 3644 else 3645 user_thresh = div_factor_fine(cache->length, bargs->usage); 3646 3647 if (chunk_used < user_thresh) 3648 ret = 0; 3649 3650 btrfs_put_block_group(cache); 3651 return ret; 3652 } 3653 3654 static int chunk_devid_filter(struct extent_buffer *leaf, 3655 struct btrfs_chunk *chunk, 3656 struct btrfs_balance_args *bargs) 3657 { 3658 struct btrfs_stripe *stripe; 3659 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3660 int i; 3661 3662 for (i = 0; i < num_stripes; i++) { 3663 stripe = btrfs_stripe_nr(chunk, i); 3664 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3665 return 0; 3666 } 3667 3668 return 1; 3669 } 3670 3671 static u64 calc_data_stripes(u64 type, int num_stripes) 3672 { 3673 const int index = btrfs_bg_flags_to_raid_index(type); 3674 const int ncopies = btrfs_raid_array[index].ncopies; 3675 const int nparity = btrfs_raid_array[index].nparity; 3676 3677 return (num_stripes - nparity) / ncopies; 3678 } 3679 3680 /* [pstart, pend) */ 3681 static int chunk_drange_filter(struct extent_buffer *leaf, 3682 struct btrfs_chunk *chunk, 3683 struct btrfs_balance_args *bargs) 3684 { 3685 struct btrfs_stripe *stripe; 3686 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3687 u64 stripe_offset; 3688 u64 stripe_length; 3689 u64 type; 3690 int factor; 3691 int i; 3692 3693 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3694 return 0; 3695 3696 type = btrfs_chunk_type(leaf, chunk); 3697 factor = calc_data_stripes(type, num_stripes); 3698 3699 for (i = 0; i < num_stripes; i++) { 3700 stripe = btrfs_stripe_nr(chunk, i); 3701 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3702 continue; 3703 3704 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3705 stripe_length = btrfs_chunk_length(leaf, chunk); 3706 stripe_length = div_u64(stripe_length, factor); 3707 3708 if (stripe_offset < bargs->pend && 3709 stripe_offset + stripe_length > bargs->pstart) 3710 return 0; 3711 } 3712 3713 return 1; 3714 } 3715 3716 /* [vstart, vend) */ 3717 static int chunk_vrange_filter(struct extent_buffer *leaf, 3718 struct btrfs_chunk *chunk, 3719 u64 chunk_offset, 3720 struct btrfs_balance_args *bargs) 3721 { 3722 if (chunk_offset < bargs->vend && 3723 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3724 /* at least part of the chunk is inside this vrange */ 3725 return 0; 3726 3727 return 1; 3728 } 3729 3730 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3731 struct btrfs_chunk *chunk, 3732 struct btrfs_balance_args *bargs) 3733 { 3734 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3735 3736 if (bargs->stripes_min <= num_stripes 3737 && num_stripes <= bargs->stripes_max) 3738 return 0; 3739 3740 return 1; 3741 } 3742 3743 static int chunk_soft_convert_filter(u64 chunk_type, 3744 struct btrfs_balance_args *bargs) 3745 { 3746 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3747 return 0; 3748 3749 chunk_type = chunk_to_extended(chunk_type) & 3750 BTRFS_EXTENDED_PROFILE_MASK; 3751 3752 if (bargs->target == chunk_type) 3753 return 1; 3754 3755 return 0; 3756 } 3757 3758 static int should_balance_chunk(struct extent_buffer *leaf, 3759 struct btrfs_chunk *chunk, u64 chunk_offset) 3760 { 3761 struct btrfs_fs_info *fs_info = leaf->fs_info; 3762 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3763 struct btrfs_balance_args *bargs = NULL; 3764 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3765 3766 /* type filter */ 3767 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3768 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3769 return 0; 3770 } 3771 3772 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3773 bargs = &bctl->data; 3774 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3775 bargs = &bctl->sys; 3776 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3777 bargs = &bctl->meta; 3778 3779 /* profiles filter */ 3780 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3781 chunk_profiles_filter(chunk_type, bargs)) { 3782 return 0; 3783 } 3784 3785 /* usage filter */ 3786 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3787 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3788 return 0; 3789 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3790 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3791 return 0; 3792 } 3793 3794 /* devid filter */ 3795 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3796 chunk_devid_filter(leaf, chunk, bargs)) { 3797 return 0; 3798 } 3799 3800 /* drange filter, makes sense only with devid filter */ 3801 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3802 chunk_drange_filter(leaf, chunk, bargs)) { 3803 return 0; 3804 } 3805 3806 /* vrange filter */ 3807 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3808 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3809 return 0; 3810 } 3811 3812 /* stripes filter */ 3813 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3814 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3815 return 0; 3816 } 3817 3818 /* soft profile changing mode */ 3819 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3820 chunk_soft_convert_filter(chunk_type, bargs)) { 3821 return 0; 3822 } 3823 3824 /* 3825 * limited by count, must be the last filter 3826 */ 3827 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3828 if (bargs->limit == 0) 3829 return 0; 3830 else 3831 bargs->limit--; 3832 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3833 /* 3834 * Same logic as the 'limit' filter; the minimum cannot be 3835 * determined here because we do not have the global information 3836 * about the count of all chunks that satisfy the filters. 3837 */ 3838 if (bargs->limit_max == 0) 3839 return 0; 3840 else 3841 bargs->limit_max--; 3842 } 3843 3844 return 1; 3845 } 3846 3847 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3848 { 3849 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3850 struct btrfs_root *chunk_root = fs_info->chunk_root; 3851 u64 chunk_type; 3852 struct btrfs_chunk *chunk; 3853 struct btrfs_path *path = NULL; 3854 struct btrfs_key key; 3855 struct btrfs_key found_key; 3856 struct extent_buffer *leaf; 3857 int slot; 3858 int ret; 3859 int enospc_errors = 0; 3860 bool counting = true; 3861 /* The single value limit and min/max limits use the same bytes in the */ 3862 u64 limit_data = bctl->data.limit; 3863 u64 limit_meta = bctl->meta.limit; 3864 u64 limit_sys = bctl->sys.limit; 3865 u32 count_data = 0; 3866 u32 count_meta = 0; 3867 u32 count_sys = 0; 3868 int chunk_reserved = 0; 3869 3870 path = btrfs_alloc_path(); 3871 if (!path) { 3872 ret = -ENOMEM; 3873 goto error; 3874 } 3875 3876 /* zero out stat counters */ 3877 spin_lock(&fs_info->balance_lock); 3878 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3879 spin_unlock(&fs_info->balance_lock); 3880 again: 3881 if (!counting) { 3882 /* 3883 * The single value limit and min/max limits use the same bytes 3884 * in the 3885 */ 3886 bctl->data.limit = limit_data; 3887 bctl->meta.limit = limit_meta; 3888 bctl->sys.limit = limit_sys; 3889 } 3890 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3891 key.offset = (u64)-1; 3892 key.type = BTRFS_CHUNK_ITEM_KEY; 3893 3894 while (1) { 3895 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3896 atomic_read(&fs_info->balance_cancel_req)) { 3897 ret = -ECANCELED; 3898 goto error; 3899 } 3900 3901 mutex_lock(&fs_info->reclaim_bgs_lock); 3902 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3903 if (ret < 0) { 3904 mutex_unlock(&fs_info->reclaim_bgs_lock); 3905 goto error; 3906 } 3907 3908 /* 3909 * this shouldn't happen, it means the last relocate 3910 * failed 3911 */ 3912 if (ret == 0) 3913 BUG(); /* FIXME break ? */ 3914 3915 ret = btrfs_previous_item(chunk_root, path, 0, 3916 BTRFS_CHUNK_ITEM_KEY); 3917 if (ret) { 3918 mutex_unlock(&fs_info->reclaim_bgs_lock); 3919 ret = 0; 3920 break; 3921 } 3922 3923 leaf = path->nodes[0]; 3924 slot = path->slots[0]; 3925 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3926 3927 if (found_key.objectid != key.objectid) { 3928 mutex_unlock(&fs_info->reclaim_bgs_lock); 3929 break; 3930 } 3931 3932 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3933 chunk_type = btrfs_chunk_type(leaf, chunk); 3934 3935 if (!counting) { 3936 spin_lock(&fs_info->balance_lock); 3937 bctl->stat.considered++; 3938 spin_unlock(&fs_info->balance_lock); 3939 } 3940 3941 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3942 3943 btrfs_release_path(path); 3944 if (!ret) { 3945 mutex_unlock(&fs_info->reclaim_bgs_lock); 3946 goto loop; 3947 } 3948 3949 if (counting) { 3950 mutex_unlock(&fs_info->reclaim_bgs_lock); 3951 spin_lock(&fs_info->balance_lock); 3952 bctl->stat.expected++; 3953 spin_unlock(&fs_info->balance_lock); 3954 3955 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3956 count_data++; 3957 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3958 count_sys++; 3959 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3960 count_meta++; 3961 3962 goto loop; 3963 } 3964 3965 /* 3966 * Apply limit_min filter, no need to check if the LIMITS 3967 * filter is used, limit_min is 0 by default 3968 */ 3969 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3970 count_data < bctl->data.limit_min) 3971 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3972 count_meta < bctl->meta.limit_min) 3973 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3974 count_sys < bctl->sys.limit_min)) { 3975 mutex_unlock(&fs_info->reclaim_bgs_lock); 3976 goto loop; 3977 } 3978 3979 if (!chunk_reserved) { 3980 /* 3981 * We may be relocating the only data chunk we have, 3982 * which could potentially end up with losing data's 3983 * raid profile, so lets allocate an empty one in 3984 * advance. 3985 */ 3986 ret = btrfs_may_alloc_data_chunk(fs_info, 3987 found_key.offset); 3988 if (ret < 0) { 3989 mutex_unlock(&fs_info->reclaim_bgs_lock); 3990 goto error; 3991 } else if (ret == 1) { 3992 chunk_reserved = 1; 3993 } 3994 } 3995 3996 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3997 mutex_unlock(&fs_info->reclaim_bgs_lock); 3998 if (ret == -ENOSPC) { 3999 enospc_errors++; 4000 } else if (ret == -ETXTBSY) { 4001 btrfs_info(fs_info, 4002 "skipping relocation of block group %llu due to active swapfile", 4003 found_key.offset); 4004 ret = 0; 4005 } else if (ret) { 4006 goto error; 4007 } else { 4008 spin_lock(&fs_info->balance_lock); 4009 bctl->stat.completed++; 4010 spin_unlock(&fs_info->balance_lock); 4011 } 4012 loop: 4013 if (found_key.offset == 0) 4014 break; 4015 key.offset = found_key.offset - 1; 4016 } 4017 4018 if (counting) { 4019 btrfs_release_path(path); 4020 counting = false; 4021 goto again; 4022 } 4023 error: 4024 btrfs_free_path(path); 4025 if (enospc_errors) { 4026 btrfs_info(fs_info, "%d enospc errors during balance", 4027 enospc_errors); 4028 if (!ret) 4029 ret = -ENOSPC; 4030 } 4031 4032 return ret; 4033 } 4034 4035 /* 4036 * See if a given profile is valid and reduced. 4037 * 4038 * @flags: profile to validate 4039 * @extended: if true @flags is treated as an extended profile 4040 */ 4041 static int alloc_profile_is_valid(u64 flags, int extended) 4042 { 4043 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 4044 BTRFS_BLOCK_GROUP_PROFILE_MASK); 4045 4046 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 4047 4048 /* 1) check that all other bits are zeroed */ 4049 if (flags & ~mask) 4050 return 0; 4051 4052 /* 2) see if profile is reduced */ 4053 if (flags == 0) 4054 return !extended; /* "0" is valid for usual profiles */ 4055 4056 return has_single_bit_set(flags); 4057 } 4058 4059 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4060 { 4061 /* cancel requested || normal exit path */ 4062 return atomic_read(&fs_info->balance_cancel_req) || 4063 (atomic_read(&fs_info->balance_pause_req) == 0 && 4064 atomic_read(&fs_info->balance_cancel_req) == 0); 4065 } 4066 4067 /* 4068 * Validate target profile against allowed profiles and return true if it's OK. 4069 * Otherwise print the error message and return false. 4070 */ 4071 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4072 const struct btrfs_balance_args *bargs, 4073 u64 allowed, const char *type) 4074 { 4075 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4076 return true; 4077 4078 /* Profile is valid and does not have bits outside of the allowed set */ 4079 if (alloc_profile_is_valid(bargs->target, 1) && 4080 (bargs->target & ~allowed) == 0) 4081 return true; 4082 4083 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4084 type, btrfs_bg_type_to_raid_name(bargs->target)); 4085 return false; 4086 } 4087 4088 /* 4089 * Fill @buf with textual description of balance filter flags @bargs, up to 4090 * @size_buf including the terminating null. The output may be trimmed if it 4091 * does not fit into the provided buffer. 4092 */ 4093 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4094 u32 size_buf) 4095 { 4096 int ret; 4097 u32 size_bp = size_buf; 4098 char *bp = buf; 4099 u64 flags = bargs->flags; 4100 char tmp_buf[128] = {'\0'}; 4101 4102 if (!flags) 4103 return; 4104 4105 #define CHECK_APPEND_NOARG(a) \ 4106 do { \ 4107 ret = snprintf(bp, size_bp, (a)); \ 4108 if (ret < 0 || ret >= size_bp) \ 4109 goto out_overflow; \ 4110 size_bp -= ret; \ 4111 bp += ret; \ 4112 } while (0) 4113 4114 #define CHECK_APPEND_1ARG(a, v1) \ 4115 do { \ 4116 ret = snprintf(bp, size_bp, (a), (v1)); \ 4117 if (ret < 0 || ret >= size_bp) \ 4118 goto out_overflow; \ 4119 size_bp -= ret; \ 4120 bp += ret; \ 4121 } while (0) 4122 4123 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4124 do { \ 4125 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4126 if (ret < 0 || ret >= size_bp) \ 4127 goto out_overflow; \ 4128 size_bp -= ret; \ 4129 bp += ret; \ 4130 } while (0) 4131 4132 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4133 CHECK_APPEND_1ARG("convert=%s,", 4134 btrfs_bg_type_to_raid_name(bargs->target)); 4135 4136 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4137 CHECK_APPEND_NOARG("soft,"); 4138 4139 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4140 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4141 sizeof(tmp_buf)); 4142 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4143 } 4144 4145 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4146 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4147 4148 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4149 CHECK_APPEND_2ARG("usage=%u..%u,", 4150 bargs->usage_min, bargs->usage_max); 4151 4152 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4153 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4154 4155 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4156 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4157 bargs->pstart, bargs->pend); 4158 4159 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4160 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4161 bargs->vstart, bargs->vend); 4162 4163 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4164 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4165 4166 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4167 CHECK_APPEND_2ARG("limit=%u..%u,", 4168 bargs->limit_min, bargs->limit_max); 4169 4170 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4171 CHECK_APPEND_2ARG("stripes=%u..%u,", 4172 bargs->stripes_min, bargs->stripes_max); 4173 4174 #undef CHECK_APPEND_2ARG 4175 #undef CHECK_APPEND_1ARG 4176 #undef CHECK_APPEND_NOARG 4177 4178 out_overflow: 4179 4180 if (size_bp < size_buf) 4181 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4182 else 4183 buf[0] = '\0'; 4184 } 4185 4186 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4187 { 4188 u32 size_buf = 1024; 4189 char tmp_buf[192] = {'\0'}; 4190 char *buf; 4191 char *bp; 4192 u32 size_bp = size_buf; 4193 int ret; 4194 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4195 4196 buf = kzalloc(size_buf, GFP_KERNEL); 4197 if (!buf) 4198 return; 4199 4200 bp = buf; 4201 4202 #define CHECK_APPEND_1ARG(a, v1) \ 4203 do { \ 4204 ret = snprintf(bp, size_bp, (a), (v1)); \ 4205 if (ret < 0 || ret >= size_bp) \ 4206 goto out_overflow; \ 4207 size_bp -= ret; \ 4208 bp += ret; \ 4209 } while (0) 4210 4211 if (bctl->flags & BTRFS_BALANCE_FORCE) 4212 CHECK_APPEND_1ARG("%s", "-f "); 4213 4214 if (bctl->flags & BTRFS_BALANCE_DATA) { 4215 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4216 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4217 } 4218 4219 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4220 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4221 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4222 } 4223 4224 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4225 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4226 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4227 } 4228 4229 #undef CHECK_APPEND_1ARG 4230 4231 out_overflow: 4232 4233 if (size_bp < size_buf) 4234 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4235 btrfs_info(fs_info, "balance: %s %s", 4236 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4237 "resume" : "start", buf); 4238 4239 kfree(buf); 4240 } 4241 4242 /* 4243 * Should be called with balance mutexe held 4244 */ 4245 int btrfs_balance(struct btrfs_fs_info *fs_info, 4246 struct btrfs_balance_control *bctl, 4247 struct btrfs_ioctl_balance_args *bargs) 4248 { 4249 u64 meta_target, data_target; 4250 u64 allowed; 4251 int mixed = 0; 4252 int ret; 4253 u64 num_devices; 4254 unsigned seq; 4255 bool reducing_redundancy; 4256 int i; 4257 4258 if (btrfs_fs_closing(fs_info) || 4259 atomic_read(&fs_info->balance_pause_req) || 4260 btrfs_should_cancel_balance(fs_info)) { 4261 ret = -EINVAL; 4262 goto out; 4263 } 4264 4265 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4266 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4267 mixed = 1; 4268 4269 /* 4270 * In case of mixed groups both data and meta should be picked, 4271 * and identical options should be given for both of them. 4272 */ 4273 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4274 if (mixed && (bctl->flags & allowed)) { 4275 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4276 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4277 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4278 btrfs_err(fs_info, 4279 "balance: mixed groups data and metadata options must be the same"); 4280 ret = -EINVAL; 4281 goto out; 4282 } 4283 } 4284 4285 /* 4286 * rw_devices will not change at the moment, device add/delete/replace 4287 * are exclusive 4288 */ 4289 num_devices = fs_info->fs_devices->rw_devices; 4290 4291 /* 4292 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4293 * special bit for it, to make it easier to distinguish. Thus we need 4294 * to set it manually, or balance would refuse the profile. 4295 */ 4296 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4297 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4298 if (num_devices >= btrfs_raid_array[i].devs_min) 4299 allowed |= btrfs_raid_array[i].bg_flag; 4300 4301 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4302 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4303 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4304 ret = -EINVAL; 4305 goto out; 4306 } 4307 4308 /* 4309 * Allow to reduce metadata or system integrity only if force set for 4310 * profiles with redundancy (copies, parity) 4311 */ 4312 allowed = 0; 4313 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4314 if (btrfs_raid_array[i].ncopies >= 2 || 4315 btrfs_raid_array[i].tolerated_failures >= 1) 4316 allowed |= btrfs_raid_array[i].bg_flag; 4317 } 4318 do { 4319 seq = read_seqbegin(&fs_info->profiles_lock); 4320 4321 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4322 (fs_info->avail_system_alloc_bits & allowed) && 4323 !(bctl->sys.target & allowed)) || 4324 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4325 (fs_info->avail_metadata_alloc_bits & allowed) && 4326 !(bctl->meta.target & allowed))) 4327 reducing_redundancy = true; 4328 else 4329 reducing_redundancy = false; 4330 4331 /* if we're not converting, the target field is uninitialized */ 4332 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4333 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4334 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4335 bctl->data.target : fs_info->avail_data_alloc_bits; 4336 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4337 4338 if (reducing_redundancy) { 4339 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4340 btrfs_info(fs_info, 4341 "balance: force reducing metadata redundancy"); 4342 } else { 4343 btrfs_err(fs_info, 4344 "balance: reduces metadata redundancy, use --force if you want this"); 4345 ret = -EINVAL; 4346 goto out; 4347 } 4348 } 4349 4350 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4351 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4352 btrfs_warn(fs_info, 4353 "balance: metadata profile %s has lower redundancy than data profile %s", 4354 btrfs_bg_type_to_raid_name(meta_target), 4355 btrfs_bg_type_to_raid_name(data_target)); 4356 } 4357 4358 ret = insert_balance_item(fs_info, bctl); 4359 if (ret && ret != -EEXIST) 4360 goto out; 4361 4362 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4363 BUG_ON(ret == -EEXIST); 4364 BUG_ON(fs_info->balance_ctl); 4365 spin_lock(&fs_info->balance_lock); 4366 fs_info->balance_ctl = bctl; 4367 spin_unlock(&fs_info->balance_lock); 4368 } else { 4369 BUG_ON(ret != -EEXIST); 4370 spin_lock(&fs_info->balance_lock); 4371 update_balance_args(bctl); 4372 spin_unlock(&fs_info->balance_lock); 4373 } 4374 4375 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4376 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4377 describe_balance_start_or_resume(fs_info); 4378 mutex_unlock(&fs_info->balance_mutex); 4379 4380 ret = __btrfs_balance(fs_info); 4381 4382 mutex_lock(&fs_info->balance_mutex); 4383 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4384 btrfs_info(fs_info, "balance: paused"); 4385 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4386 } 4387 /* 4388 * Balance can be canceled by: 4389 * 4390 * - Regular cancel request 4391 * Then ret == -ECANCELED and balance_cancel_req > 0 4392 * 4393 * - Fatal signal to "btrfs" process 4394 * Either the signal caught by wait_reserve_ticket() and callers 4395 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4396 * got -ECANCELED. 4397 * Either way, in this case balance_cancel_req = 0, and 4398 * ret == -EINTR or ret == -ECANCELED. 4399 * 4400 * So here we only check the return value to catch canceled balance. 4401 */ 4402 else if (ret == -ECANCELED || ret == -EINTR) 4403 btrfs_info(fs_info, "balance: canceled"); 4404 else 4405 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4406 4407 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4408 4409 if (bargs) { 4410 memset(bargs, 0, sizeof(*bargs)); 4411 btrfs_update_ioctl_balance_args(fs_info, bargs); 4412 } 4413 4414 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4415 balance_need_close(fs_info)) { 4416 reset_balance_state(fs_info); 4417 btrfs_exclop_finish(fs_info); 4418 } 4419 4420 wake_up(&fs_info->balance_wait_q); 4421 4422 return ret; 4423 out: 4424 if (bctl->flags & BTRFS_BALANCE_RESUME) 4425 reset_balance_state(fs_info); 4426 else 4427 kfree(bctl); 4428 btrfs_exclop_finish(fs_info); 4429 4430 return ret; 4431 } 4432 4433 static int balance_kthread(void *data) 4434 { 4435 struct btrfs_fs_info *fs_info = data; 4436 int ret = 0; 4437 4438 sb_start_write(fs_info->sb); 4439 mutex_lock(&fs_info->balance_mutex); 4440 if (fs_info->balance_ctl) 4441 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4442 mutex_unlock(&fs_info->balance_mutex); 4443 sb_end_write(fs_info->sb); 4444 4445 return ret; 4446 } 4447 4448 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4449 { 4450 struct task_struct *tsk; 4451 4452 mutex_lock(&fs_info->balance_mutex); 4453 if (!fs_info->balance_ctl) { 4454 mutex_unlock(&fs_info->balance_mutex); 4455 return 0; 4456 } 4457 mutex_unlock(&fs_info->balance_mutex); 4458 4459 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4460 btrfs_info(fs_info, "balance: resume skipped"); 4461 return 0; 4462 } 4463 4464 spin_lock(&fs_info->super_lock); 4465 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4466 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4467 spin_unlock(&fs_info->super_lock); 4468 /* 4469 * A ro->rw remount sequence should continue with the paused balance 4470 * regardless of who pauses it, system or the user as of now, so set 4471 * the resume flag. 4472 */ 4473 spin_lock(&fs_info->balance_lock); 4474 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4475 spin_unlock(&fs_info->balance_lock); 4476 4477 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4478 return PTR_ERR_OR_ZERO(tsk); 4479 } 4480 4481 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4482 { 4483 struct btrfs_balance_control *bctl; 4484 struct btrfs_balance_item *item; 4485 struct btrfs_disk_balance_args disk_bargs; 4486 struct btrfs_path *path; 4487 struct extent_buffer *leaf; 4488 struct btrfs_key key; 4489 int ret; 4490 4491 path = btrfs_alloc_path(); 4492 if (!path) 4493 return -ENOMEM; 4494 4495 key.objectid = BTRFS_BALANCE_OBJECTID; 4496 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4497 key.offset = 0; 4498 4499 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4500 if (ret < 0) 4501 goto out; 4502 if (ret > 0) { /* ret = -ENOENT; */ 4503 ret = 0; 4504 goto out; 4505 } 4506 4507 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4508 if (!bctl) { 4509 ret = -ENOMEM; 4510 goto out; 4511 } 4512 4513 leaf = path->nodes[0]; 4514 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4515 4516 bctl->flags = btrfs_balance_flags(leaf, item); 4517 bctl->flags |= BTRFS_BALANCE_RESUME; 4518 4519 btrfs_balance_data(leaf, item, &disk_bargs); 4520 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4521 btrfs_balance_meta(leaf, item, &disk_bargs); 4522 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4523 btrfs_balance_sys(leaf, item, &disk_bargs); 4524 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4525 4526 /* 4527 * This should never happen, as the paused balance state is recovered 4528 * during mount without any chance of other exclusive ops to collide. 4529 * 4530 * This gives the exclusive op status to balance and keeps in paused 4531 * state until user intervention (cancel or umount). If the ownership 4532 * cannot be assigned, show a message but do not fail. The balance 4533 * is in a paused state and must have fs_info::balance_ctl properly 4534 * set up. 4535 */ 4536 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4537 btrfs_warn(fs_info, 4538 "balance: cannot set exclusive op status, resume manually"); 4539 4540 btrfs_release_path(path); 4541 4542 mutex_lock(&fs_info->balance_mutex); 4543 BUG_ON(fs_info->balance_ctl); 4544 spin_lock(&fs_info->balance_lock); 4545 fs_info->balance_ctl = bctl; 4546 spin_unlock(&fs_info->balance_lock); 4547 mutex_unlock(&fs_info->balance_mutex); 4548 out: 4549 btrfs_free_path(path); 4550 return ret; 4551 } 4552 4553 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4554 { 4555 int ret = 0; 4556 4557 mutex_lock(&fs_info->balance_mutex); 4558 if (!fs_info->balance_ctl) { 4559 mutex_unlock(&fs_info->balance_mutex); 4560 return -ENOTCONN; 4561 } 4562 4563 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4564 atomic_inc(&fs_info->balance_pause_req); 4565 mutex_unlock(&fs_info->balance_mutex); 4566 4567 wait_event(fs_info->balance_wait_q, 4568 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4569 4570 mutex_lock(&fs_info->balance_mutex); 4571 /* we are good with balance_ctl ripped off from under us */ 4572 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4573 atomic_dec(&fs_info->balance_pause_req); 4574 } else { 4575 ret = -ENOTCONN; 4576 } 4577 4578 mutex_unlock(&fs_info->balance_mutex); 4579 return ret; 4580 } 4581 4582 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4583 { 4584 mutex_lock(&fs_info->balance_mutex); 4585 if (!fs_info->balance_ctl) { 4586 mutex_unlock(&fs_info->balance_mutex); 4587 return -ENOTCONN; 4588 } 4589 4590 /* 4591 * A paused balance with the item stored on disk can be resumed at 4592 * mount time if the mount is read-write. Otherwise it's still paused 4593 * and we must not allow cancelling as it deletes the item. 4594 */ 4595 if (sb_rdonly(fs_info->sb)) { 4596 mutex_unlock(&fs_info->balance_mutex); 4597 return -EROFS; 4598 } 4599 4600 atomic_inc(&fs_info->balance_cancel_req); 4601 /* 4602 * if we are running just wait and return, balance item is 4603 * deleted in btrfs_balance in this case 4604 */ 4605 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4606 mutex_unlock(&fs_info->balance_mutex); 4607 wait_event(fs_info->balance_wait_q, 4608 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4609 mutex_lock(&fs_info->balance_mutex); 4610 } else { 4611 mutex_unlock(&fs_info->balance_mutex); 4612 /* 4613 * Lock released to allow other waiters to continue, we'll 4614 * reexamine the status again. 4615 */ 4616 mutex_lock(&fs_info->balance_mutex); 4617 4618 if (fs_info->balance_ctl) { 4619 reset_balance_state(fs_info); 4620 btrfs_exclop_finish(fs_info); 4621 btrfs_info(fs_info, "balance: canceled"); 4622 } 4623 } 4624 4625 BUG_ON(fs_info->balance_ctl || 4626 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4627 atomic_dec(&fs_info->balance_cancel_req); 4628 mutex_unlock(&fs_info->balance_mutex); 4629 return 0; 4630 } 4631 4632 int btrfs_uuid_scan_kthread(void *data) 4633 { 4634 struct btrfs_fs_info *fs_info = data; 4635 struct btrfs_root *root = fs_info->tree_root; 4636 struct btrfs_key key; 4637 struct btrfs_path *path = NULL; 4638 int ret = 0; 4639 struct extent_buffer *eb; 4640 int slot; 4641 struct btrfs_root_item root_item; 4642 u32 item_size; 4643 struct btrfs_trans_handle *trans = NULL; 4644 bool closing = false; 4645 4646 path = btrfs_alloc_path(); 4647 if (!path) { 4648 ret = -ENOMEM; 4649 goto out; 4650 } 4651 4652 key.objectid = 0; 4653 key.type = BTRFS_ROOT_ITEM_KEY; 4654 key.offset = 0; 4655 4656 while (1) { 4657 if (btrfs_fs_closing(fs_info)) { 4658 closing = true; 4659 break; 4660 } 4661 ret = btrfs_search_forward(root, &key, path, 4662 BTRFS_OLDEST_GENERATION); 4663 if (ret) { 4664 if (ret > 0) 4665 ret = 0; 4666 break; 4667 } 4668 4669 if (key.type != BTRFS_ROOT_ITEM_KEY || 4670 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4671 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4672 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4673 goto skip; 4674 4675 eb = path->nodes[0]; 4676 slot = path->slots[0]; 4677 item_size = btrfs_item_size(eb, slot); 4678 if (item_size < sizeof(root_item)) 4679 goto skip; 4680 4681 read_extent_buffer(eb, &root_item, 4682 btrfs_item_ptr_offset(eb, slot), 4683 (int)sizeof(root_item)); 4684 if (btrfs_root_refs(&root_item) == 0) 4685 goto skip; 4686 4687 if (!btrfs_is_empty_uuid(root_item.uuid) || 4688 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4689 if (trans) 4690 goto update_tree; 4691 4692 btrfs_release_path(path); 4693 /* 4694 * 1 - subvol uuid item 4695 * 1 - received_subvol uuid item 4696 */ 4697 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4698 if (IS_ERR(trans)) { 4699 ret = PTR_ERR(trans); 4700 break; 4701 } 4702 continue; 4703 } else { 4704 goto skip; 4705 } 4706 update_tree: 4707 btrfs_release_path(path); 4708 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4709 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4710 BTRFS_UUID_KEY_SUBVOL, 4711 key.objectid); 4712 if (ret < 0) { 4713 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4714 ret); 4715 break; 4716 } 4717 } 4718 4719 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4720 ret = btrfs_uuid_tree_add(trans, 4721 root_item.received_uuid, 4722 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4723 key.objectid); 4724 if (ret < 0) { 4725 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4726 ret); 4727 break; 4728 } 4729 } 4730 4731 skip: 4732 btrfs_release_path(path); 4733 if (trans) { 4734 ret = btrfs_end_transaction(trans); 4735 trans = NULL; 4736 if (ret) 4737 break; 4738 } 4739 4740 if (key.offset < (u64)-1) { 4741 key.offset++; 4742 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4743 key.offset = 0; 4744 key.type = BTRFS_ROOT_ITEM_KEY; 4745 } else if (key.objectid < (u64)-1) { 4746 key.offset = 0; 4747 key.type = BTRFS_ROOT_ITEM_KEY; 4748 key.objectid++; 4749 } else { 4750 break; 4751 } 4752 cond_resched(); 4753 } 4754 4755 out: 4756 btrfs_free_path(path); 4757 if (trans && !IS_ERR(trans)) 4758 btrfs_end_transaction(trans); 4759 if (ret) 4760 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4761 else if (!closing) 4762 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4763 up(&fs_info->uuid_tree_rescan_sem); 4764 return 0; 4765 } 4766 4767 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4768 { 4769 struct btrfs_trans_handle *trans; 4770 struct btrfs_root *tree_root = fs_info->tree_root; 4771 struct btrfs_root *uuid_root; 4772 struct task_struct *task; 4773 int ret; 4774 4775 /* 4776 * 1 - root node 4777 * 1 - root item 4778 */ 4779 trans = btrfs_start_transaction(tree_root, 2); 4780 if (IS_ERR(trans)) 4781 return PTR_ERR(trans); 4782 4783 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4784 if (IS_ERR(uuid_root)) { 4785 ret = PTR_ERR(uuid_root); 4786 btrfs_abort_transaction(trans, ret); 4787 btrfs_end_transaction(trans); 4788 return ret; 4789 } 4790 4791 fs_info->uuid_root = uuid_root; 4792 4793 ret = btrfs_commit_transaction(trans); 4794 if (ret) 4795 return ret; 4796 4797 down(&fs_info->uuid_tree_rescan_sem); 4798 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4799 if (IS_ERR(task)) { 4800 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4801 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4802 up(&fs_info->uuid_tree_rescan_sem); 4803 return PTR_ERR(task); 4804 } 4805 4806 return 0; 4807 } 4808 4809 /* 4810 * shrinking a device means finding all of the device extents past 4811 * the new size, and then following the back refs to the chunks. 4812 * The chunk relocation code actually frees the device extent 4813 */ 4814 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4815 { 4816 struct btrfs_fs_info *fs_info = device->fs_info; 4817 struct btrfs_root *root = fs_info->dev_root; 4818 struct btrfs_trans_handle *trans; 4819 struct btrfs_dev_extent *dev_extent = NULL; 4820 struct btrfs_path *path; 4821 u64 length; 4822 u64 chunk_offset; 4823 int ret; 4824 int slot; 4825 int failed = 0; 4826 bool retried = false; 4827 struct extent_buffer *l; 4828 struct btrfs_key key; 4829 struct btrfs_super_block *super_copy = fs_info->super_copy; 4830 u64 old_total = btrfs_super_total_bytes(super_copy); 4831 u64 old_size = btrfs_device_get_total_bytes(device); 4832 u64 diff; 4833 u64 start; 4834 4835 new_size = round_down(new_size, fs_info->sectorsize); 4836 start = new_size; 4837 diff = round_down(old_size - new_size, fs_info->sectorsize); 4838 4839 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4840 return -EINVAL; 4841 4842 path = btrfs_alloc_path(); 4843 if (!path) 4844 return -ENOMEM; 4845 4846 path->reada = READA_BACK; 4847 4848 trans = btrfs_start_transaction(root, 0); 4849 if (IS_ERR(trans)) { 4850 btrfs_free_path(path); 4851 return PTR_ERR(trans); 4852 } 4853 4854 mutex_lock(&fs_info->chunk_mutex); 4855 4856 btrfs_device_set_total_bytes(device, new_size); 4857 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4858 device->fs_devices->total_rw_bytes -= diff; 4859 atomic64_sub(diff, &fs_info->free_chunk_space); 4860 } 4861 4862 /* 4863 * Once the device's size has been set to the new size, ensure all 4864 * in-memory chunks are synced to disk so that the loop below sees them 4865 * and relocates them accordingly. 4866 */ 4867 if (contains_pending_extent(device, &start, diff)) { 4868 mutex_unlock(&fs_info->chunk_mutex); 4869 ret = btrfs_commit_transaction(trans); 4870 if (ret) 4871 goto done; 4872 } else { 4873 mutex_unlock(&fs_info->chunk_mutex); 4874 btrfs_end_transaction(trans); 4875 } 4876 4877 again: 4878 key.objectid = device->devid; 4879 key.offset = (u64)-1; 4880 key.type = BTRFS_DEV_EXTENT_KEY; 4881 4882 do { 4883 mutex_lock(&fs_info->reclaim_bgs_lock); 4884 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4885 if (ret < 0) { 4886 mutex_unlock(&fs_info->reclaim_bgs_lock); 4887 goto done; 4888 } 4889 4890 ret = btrfs_previous_item(root, path, 0, key.type); 4891 if (ret) { 4892 mutex_unlock(&fs_info->reclaim_bgs_lock); 4893 if (ret < 0) 4894 goto done; 4895 ret = 0; 4896 btrfs_release_path(path); 4897 break; 4898 } 4899 4900 l = path->nodes[0]; 4901 slot = path->slots[0]; 4902 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4903 4904 if (key.objectid != device->devid) { 4905 mutex_unlock(&fs_info->reclaim_bgs_lock); 4906 btrfs_release_path(path); 4907 break; 4908 } 4909 4910 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4911 length = btrfs_dev_extent_length(l, dev_extent); 4912 4913 if (key.offset + length <= new_size) { 4914 mutex_unlock(&fs_info->reclaim_bgs_lock); 4915 btrfs_release_path(path); 4916 break; 4917 } 4918 4919 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4920 btrfs_release_path(path); 4921 4922 /* 4923 * We may be relocating the only data chunk we have, 4924 * which could potentially end up with losing data's 4925 * raid profile, so lets allocate an empty one in 4926 * advance. 4927 */ 4928 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4929 if (ret < 0) { 4930 mutex_unlock(&fs_info->reclaim_bgs_lock); 4931 goto done; 4932 } 4933 4934 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4935 mutex_unlock(&fs_info->reclaim_bgs_lock); 4936 if (ret == -ENOSPC) { 4937 failed++; 4938 } else if (ret) { 4939 if (ret == -ETXTBSY) { 4940 btrfs_warn(fs_info, 4941 "could not shrink block group %llu due to active swapfile", 4942 chunk_offset); 4943 } 4944 goto done; 4945 } 4946 } while (key.offset-- > 0); 4947 4948 if (failed && !retried) { 4949 failed = 0; 4950 retried = true; 4951 goto again; 4952 } else if (failed && retried) { 4953 ret = -ENOSPC; 4954 goto done; 4955 } 4956 4957 /* Shrinking succeeded, else we would be at "done". */ 4958 trans = btrfs_start_transaction(root, 0); 4959 if (IS_ERR(trans)) { 4960 ret = PTR_ERR(trans); 4961 goto done; 4962 } 4963 4964 mutex_lock(&fs_info->chunk_mutex); 4965 /* Clear all state bits beyond the shrunk device size */ 4966 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4967 CHUNK_STATE_MASK); 4968 4969 btrfs_device_set_disk_total_bytes(device, new_size); 4970 if (list_empty(&device->post_commit_list)) 4971 list_add_tail(&device->post_commit_list, 4972 &trans->transaction->dev_update_list); 4973 4974 WARN_ON(diff > old_total); 4975 btrfs_set_super_total_bytes(super_copy, 4976 round_down(old_total - diff, fs_info->sectorsize)); 4977 mutex_unlock(&fs_info->chunk_mutex); 4978 4979 btrfs_reserve_chunk_metadata(trans, false); 4980 /* Now btrfs_update_device() will change the on-disk size. */ 4981 ret = btrfs_update_device(trans, device); 4982 btrfs_trans_release_chunk_metadata(trans); 4983 if (ret < 0) { 4984 btrfs_abort_transaction(trans, ret); 4985 btrfs_end_transaction(trans); 4986 } else { 4987 ret = btrfs_commit_transaction(trans); 4988 } 4989 done: 4990 btrfs_free_path(path); 4991 if (ret) { 4992 mutex_lock(&fs_info->chunk_mutex); 4993 btrfs_device_set_total_bytes(device, old_size); 4994 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 4995 device->fs_devices->total_rw_bytes += diff; 4996 atomic64_add(diff, &fs_info->free_chunk_space); 4997 mutex_unlock(&fs_info->chunk_mutex); 4998 } 4999 return ret; 5000 } 5001 5002 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 5003 struct btrfs_key *key, 5004 struct btrfs_chunk *chunk, int item_size) 5005 { 5006 struct btrfs_super_block *super_copy = fs_info->super_copy; 5007 struct btrfs_disk_key disk_key; 5008 u32 array_size; 5009 u8 *ptr; 5010 5011 lockdep_assert_held(&fs_info->chunk_mutex); 5012 5013 array_size = btrfs_super_sys_array_size(super_copy); 5014 if (array_size + item_size + sizeof(disk_key) 5015 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5016 return -EFBIG; 5017 5018 ptr = super_copy->sys_chunk_array + array_size; 5019 btrfs_cpu_key_to_disk(&disk_key, key); 5020 memcpy(ptr, &disk_key, sizeof(disk_key)); 5021 ptr += sizeof(disk_key); 5022 memcpy(ptr, chunk, item_size); 5023 item_size += sizeof(disk_key); 5024 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5025 5026 return 0; 5027 } 5028 5029 /* 5030 * sort the devices in descending order by max_avail, total_avail 5031 */ 5032 static int btrfs_cmp_device_info(const void *a, const void *b) 5033 { 5034 const struct btrfs_device_info *di_a = a; 5035 const struct btrfs_device_info *di_b = b; 5036 5037 if (di_a->max_avail > di_b->max_avail) 5038 return -1; 5039 if (di_a->max_avail < di_b->max_avail) 5040 return 1; 5041 if (di_a->total_avail > di_b->total_avail) 5042 return -1; 5043 if (di_a->total_avail < di_b->total_avail) 5044 return 1; 5045 return 0; 5046 } 5047 5048 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5049 { 5050 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5051 return; 5052 5053 btrfs_set_fs_incompat(info, RAID56); 5054 } 5055 5056 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5057 { 5058 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5059 return; 5060 5061 btrfs_set_fs_incompat(info, RAID1C34); 5062 } 5063 5064 /* 5065 * Structure used internally for btrfs_create_chunk() function. 5066 * Wraps needed parameters. 5067 */ 5068 struct alloc_chunk_ctl { 5069 u64 start; 5070 u64 type; 5071 /* Total number of stripes to allocate */ 5072 int num_stripes; 5073 /* sub_stripes info for map */ 5074 int sub_stripes; 5075 /* Stripes per device */ 5076 int dev_stripes; 5077 /* Maximum number of devices to use */ 5078 int devs_max; 5079 /* Minimum number of devices to use */ 5080 int devs_min; 5081 /* ndevs has to be a multiple of this */ 5082 int devs_increment; 5083 /* Number of copies */ 5084 int ncopies; 5085 /* Number of stripes worth of bytes to store parity information */ 5086 int nparity; 5087 u64 max_stripe_size; 5088 u64 max_chunk_size; 5089 u64 dev_extent_min; 5090 u64 stripe_size; 5091 u64 chunk_size; 5092 int ndevs; 5093 }; 5094 5095 static void init_alloc_chunk_ctl_policy_regular( 5096 struct btrfs_fs_devices *fs_devices, 5097 struct alloc_chunk_ctl *ctl) 5098 { 5099 struct btrfs_space_info *space_info; 5100 5101 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); 5102 ASSERT(space_info); 5103 5104 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); 5105 ctl->max_stripe_size = ctl->max_chunk_size; 5106 5107 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) 5108 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); 5109 5110 /* We don't want a chunk larger than 10% of writable space */ 5111 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5112 ctl->max_chunk_size); 5113 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5114 } 5115 5116 static void init_alloc_chunk_ctl_policy_zoned( 5117 struct btrfs_fs_devices *fs_devices, 5118 struct alloc_chunk_ctl *ctl) 5119 { 5120 u64 zone_size = fs_devices->fs_info->zone_size; 5121 u64 limit; 5122 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5123 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5124 u64 min_chunk_size = min_data_stripes * zone_size; 5125 u64 type = ctl->type; 5126 5127 ctl->max_stripe_size = zone_size; 5128 if (type & BTRFS_BLOCK_GROUP_DATA) { 5129 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5130 zone_size); 5131 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5132 ctl->max_chunk_size = ctl->max_stripe_size; 5133 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5134 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5135 ctl->devs_max = min_t(int, ctl->devs_max, 5136 BTRFS_MAX_DEVS_SYS_CHUNK); 5137 } else { 5138 BUG(); 5139 } 5140 5141 /* We don't want a chunk larger than 10% of writable space */ 5142 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5143 zone_size), 5144 min_chunk_size); 5145 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5146 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5147 } 5148 5149 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5150 struct alloc_chunk_ctl *ctl) 5151 { 5152 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5153 5154 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5155 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5156 ctl->devs_max = btrfs_raid_array[index].devs_max; 5157 if (!ctl->devs_max) 5158 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5159 ctl->devs_min = btrfs_raid_array[index].devs_min; 5160 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5161 ctl->ncopies = btrfs_raid_array[index].ncopies; 5162 ctl->nparity = btrfs_raid_array[index].nparity; 5163 ctl->ndevs = 0; 5164 5165 switch (fs_devices->chunk_alloc_policy) { 5166 case BTRFS_CHUNK_ALLOC_REGULAR: 5167 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5168 break; 5169 case BTRFS_CHUNK_ALLOC_ZONED: 5170 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5171 break; 5172 default: 5173 BUG(); 5174 } 5175 } 5176 5177 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5178 struct alloc_chunk_ctl *ctl, 5179 struct btrfs_device_info *devices_info) 5180 { 5181 struct btrfs_fs_info *info = fs_devices->fs_info; 5182 struct btrfs_device *device; 5183 u64 total_avail; 5184 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5185 int ret; 5186 int ndevs = 0; 5187 u64 max_avail; 5188 u64 dev_offset; 5189 5190 /* 5191 * in the first pass through the devices list, we gather information 5192 * about the available holes on each device. 5193 */ 5194 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5195 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5196 WARN(1, KERN_ERR 5197 "BTRFS: read-only device in alloc_list\n"); 5198 continue; 5199 } 5200 5201 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5202 &device->dev_state) || 5203 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5204 continue; 5205 5206 if (device->total_bytes > device->bytes_used) 5207 total_avail = device->total_bytes - device->bytes_used; 5208 else 5209 total_avail = 0; 5210 5211 /* If there is no space on this device, skip it. */ 5212 if (total_avail < ctl->dev_extent_min) 5213 continue; 5214 5215 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5216 &max_avail); 5217 if (ret && ret != -ENOSPC) 5218 return ret; 5219 5220 if (ret == 0) 5221 max_avail = dev_extent_want; 5222 5223 if (max_avail < ctl->dev_extent_min) { 5224 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5225 btrfs_debug(info, 5226 "%s: devid %llu has no free space, have=%llu want=%llu", 5227 __func__, device->devid, max_avail, 5228 ctl->dev_extent_min); 5229 continue; 5230 } 5231 5232 if (ndevs == fs_devices->rw_devices) { 5233 WARN(1, "%s: found more than %llu devices\n", 5234 __func__, fs_devices->rw_devices); 5235 break; 5236 } 5237 devices_info[ndevs].dev_offset = dev_offset; 5238 devices_info[ndevs].max_avail = max_avail; 5239 devices_info[ndevs].total_avail = total_avail; 5240 devices_info[ndevs].dev = device; 5241 ++ndevs; 5242 } 5243 ctl->ndevs = ndevs; 5244 5245 /* 5246 * now sort the devices by hole size / available space 5247 */ 5248 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5249 btrfs_cmp_device_info, NULL); 5250 5251 return 0; 5252 } 5253 5254 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5255 struct btrfs_device_info *devices_info) 5256 { 5257 /* Number of stripes that count for block group size */ 5258 int data_stripes; 5259 5260 /* 5261 * The primary goal is to maximize the number of stripes, so use as 5262 * many devices as possible, even if the stripes are not maximum sized. 5263 * 5264 * The DUP profile stores more than one stripe per device, the 5265 * max_avail is the total size so we have to adjust. 5266 */ 5267 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5268 ctl->dev_stripes); 5269 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5270 5271 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5272 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5273 5274 /* 5275 * Use the number of data stripes to figure out how big this chunk is 5276 * really going to be in terms of logical address space, and compare 5277 * that answer with the max chunk size. If it's higher, we try to 5278 * reduce stripe_size. 5279 */ 5280 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5281 /* 5282 * Reduce stripe_size, round it up to a 16MB boundary again and 5283 * then use it, unless it ends up being even bigger than the 5284 * previous value we had already. 5285 */ 5286 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5287 data_stripes), SZ_16M), 5288 ctl->stripe_size); 5289 } 5290 5291 /* Stripe size should not go beyond 1G. */ 5292 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); 5293 5294 /* Align to BTRFS_STRIPE_LEN */ 5295 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5296 ctl->chunk_size = ctl->stripe_size * data_stripes; 5297 5298 return 0; 5299 } 5300 5301 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5302 struct btrfs_device_info *devices_info) 5303 { 5304 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5305 /* Number of stripes that count for block group size */ 5306 int data_stripes; 5307 5308 /* 5309 * It should hold because: 5310 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5311 */ 5312 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5313 5314 ctl->stripe_size = zone_size; 5315 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5316 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5317 5318 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5319 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5320 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5321 ctl->stripe_size) + ctl->nparity, 5322 ctl->dev_stripes); 5323 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5324 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5325 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5326 } 5327 5328 ctl->chunk_size = ctl->stripe_size * data_stripes; 5329 5330 return 0; 5331 } 5332 5333 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5334 struct alloc_chunk_ctl *ctl, 5335 struct btrfs_device_info *devices_info) 5336 { 5337 struct btrfs_fs_info *info = fs_devices->fs_info; 5338 5339 /* 5340 * Round down to number of usable stripes, devs_increment can be any 5341 * number so we can't use round_down() that requires power of 2, while 5342 * rounddown is safe. 5343 */ 5344 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5345 5346 if (ctl->ndevs < ctl->devs_min) { 5347 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5348 btrfs_debug(info, 5349 "%s: not enough devices with free space: have=%d minimum required=%d", 5350 __func__, ctl->ndevs, ctl->devs_min); 5351 } 5352 return -ENOSPC; 5353 } 5354 5355 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5356 5357 switch (fs_devices->chunk_alloc_policy) { 5358 case BTRFS_CHUNK_ALLOC_REGULAR: 5359 return decide_stripe_size_regular(ctl, devices_info); 5360 case BTRFS_CHUNK_ALLOC_ZONED: 5361 return decide_stripe_size_zoned(ctl, devices_info); 5362 default: 5363 BUG(); 5364 } 5365 } 5366 5367 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5368 struct alloc_chunk_ctl *ctl, 5369 struct btrfs_device_info *devices_info) 5370 { 5371 struct btrfs_fs_info *info = trans->fs_info; 5372 struct map_lookup *map = NULL; 5373 struct extent_map_tree *em_tree; 5374 struct btrfs_block_group *block_group; 5375 struct extent_map *em; 5376 u64 start = ctl->start; 5377 u64 type = ctl->type; 5378 int ret; 5379 int i; 5380 int j; 5381 5382 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5383 if (!map) 5384 return ERR_PTR(-ENOMEM); 5385 map->num_stripes = ctl->num_stripes; 5386 5387 for (i = 0; i < ctl->ndevs; ++i) { 5388 for (j = 0; j < ctl->dev_stripes; ++j) { 5389 int s = i * ctl->dev_stripes + j; 5390 map->stripes[s].dev = devices_info[i].dev; 5391 map->stripes[s].physical = devices_info[i].dev_offset + 5392 j * ctl->stripe_size; 5393 } 5394 } 5395 map->stripe_len = BTRFS_STRIPE_LEN; 5396 map->io_align = BTRFS_STRIPE_LEN; 5397 map->io_width = BTRFS_STRIPE_LEN; 5398 map->type = type; 5399 map->sub_stripes = ctl->sub_stripes; 5400 5401 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5402 5403 em = alloc_extent_map(); 5404 if (!em) { 5405 kfree(map); 5406 return ERR_PTR(-ENOMEM); 5407 } 5408 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5409 em->map_lookup = map; 5410 em->start = start; 5411 em->len = ctl->chunk_size; 5412 em->block_start = 0; 5413 em->block_len = em->len; 5414 em->orig_block_len = ctl->stripe_size; 5415 5416 em_tree = &info->mapping_tree; 5417 write_lock(&em_tree->lock); 5418 ret = add_extent_mapping(em_tree, em, 0); 5419 if (ret) { 5420 write_unlock(&em_tree->lock); 5421 free_extent_map(em); 5422 return ERR_PTR(ret); 5423 } 5424 write_unlock(&em_tree->lock); 5425 5426 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5427 if (IS_ERR(block_group)) 5428 goto error_del_extent; 5429 5430 for (i = 0; i < map->num_stripes; i++) { 5431 struct btrfs_device *dev = map->stripes[i].dev; 5432 5433 btrfs_device_set_bytes_used(dev, 5434 dev->bytes_used + ctl->stripe_size); 5435 if (list_empty(&dev->post_commit_list)) 5436 list_add_tail(&dev->post_commit_list, 5437 &trans->transaction->dev_update_list); 5438 } 5439 5440 atomic64_sub(ctl->stripe_size * map->num_stripes, 5441 &info->free_chunk_space); 5442 5443 free_extent_map(em); 5444 check_raid56_incompat_flag(info, type); 5445 check_raid1c34_incompat_flag(info, type); 5446 5447 return block_group; 5448 5449 error_del_extent: 5450 write_lock(&em_tree->lock); 5451 remove_extent_mapping(em_tree, em); 5452 write_unlock(&em_tree->lock); 5453 5454 /* One for our allocation */ 5455 free_extent_map(em); 5456 /* One for the tree reference */ 5457 free_extent_map(em); 5458 5459 return block_group; 5460 } 5461 5462 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5463 u64 type) 5464 { 5465 struct btrfs_fs_info *info = trans->fs_info; 5466 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5467 struct btrfs_device_info *devices_info = NULL; 5468 struct alloc_chunk_ctl ctl; 5469 struct btrfs_block_group *block_group; 5470 int ret; 5471 5472 lockdep_assert_held(&info->chunk_mutex); 5473 5474 if (!alloc_profile_is_valid(type, 0)) { 5475 ASSERT(0); 5476 return ERR_PTR(-EINVAL); 5477 } 5478 5479 if (list_empty(&fs_devices->alloc_list)) { 5480 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5481 btrfs_debug(info, "%s: no writable device", __func__); 5482 return ERR_PTR(-ENOSPC); 5483 } 5484 5485 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5486 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5487 ASSERT(0); 5488 return ERR_PTR(-EINVAL); 5489 } 5490 5491 ctl.start = find_next_chunk(info); 5492 ctl.type = type; 5493 init_alloc_chunk_ctl(fs_devices, &ctl); 5494 5495 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5496 GFP_NOFS); 5497 if (!devices_info) 5498 return ERR_PTR(-ENOMEM); 5499 5500 ret = gather_device_info(fs_devices, &ctl, devices_info); 5501 if (ret < 0) { 5502 block_group = ERR_PTR(ret); 5503 goto out; 5504 } 5505 5506 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5507 if (ret < 0) { 5508 block_group = ERR_PTR(ret); 5509 goto out; 5510 } 5511 5512 block_group = create_chunk(trans, &ctl, devices_info); 5513 5514 out: 5515 kfree(devices_info); 5516 return block_group; 5517 } 5518 5519 /* 5520 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5521 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5522 * chunks. 5523 * 5524 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5525 * phases. 5526 */ 5527 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5528 struct btrfs_block_group *bg) 5529 { 5530 struct btrfs_fs_info *fs_info = trans->fs_info; 5531 struct btrfs_root *chunk_root = fs_info->chunk_root; 5532 struct btrfs_key key; 5533 struct btrfs_chunk *chunk; 5534 struct btrfs_stripe *stripe; 5535 struct extent_map *em; 5536 struct map_lookup *map; 5537 size_t item_size; 5538 int i; 5539 int ret; 5540 5541 /* 5542 * We take the chunk_mutex for 2 reasons: 5543 * 5544 * 1) Updates and insertions in the chunk btree must be done while holding 5545 * the chunk_mutex, as well as updating the system chunk array in the 5546 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5547 * details; 5548 * 5549 * 2) To prevent races with the final phase of a device replace operation 5550 * that replaces the device object associated with the map's stripes, 5551 * because the device object's id can change at any time during that 5552 * final phase of the device replace operation 5553 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5554 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5555 * which would cause a failure when updating the device item, which does 5556 * not exists, or persisting a stripe of the chunk item with such ID. 5557 * Here we can't use the device_list_mutex because our caller already 5558 * has locked the chunk_mutex, and the final phase of device replace 5559 * acquires both mutexes - first the device_list_mutex and then the 5560 * chunk_mutex. Using any of those two mutexes protects us from a 5561 * concurrent device replace. 5562 */ 5563 lockdep_assert_held(&fs_info->chunk_mutex); 5564 5565 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5566 if (IS_ERR(em)) { 5567 ret = PTR_ERR(em); 5568 btrfs_abort_transaction(trans, ret); 5569 return ret; 5570 } 5571 5572 map = em->map_lookup; 5573 item_size = btrfs_chunk_item_size(map->num_stripes); 5574 5575 chunk = kzalloc(item_size, GFP_NOFS); 5576 if (!chunk) { 5577 ret = -ENOMEM; 5578 btrfs_abort_transaction(trans, ret); 5579 goto out; 5580 } 5581 5582 for (i = 0; i < map->num_stripes; i++) { 5583 struct btrfs_device *device = map->stripes[i].dev; 5584 5585 ret = btrfs_update_device(trans, device); 5586 if (ret) 5587 goto out; 5588 } 5589 5590 stripe = &chunk->stripe; 5591 for (i = 0; i < map->num_stripes; i++) { 5592 struct btrfs_device *device = map->stripes[i].dev; 5593 const u64 dev_offset = map->stripes[i].physical; 5594 5595 btrfs_set_stack_stripe_devid(stripe, device->devid); 5596 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5597 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5598 stripe++; 5599 } 5600 5601 btrfs_set_stack_chunk_length(chunk, bg->length); 5602 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5603 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5604 btrfs_set_stack_chunk_type(chunk, map->type); 5605 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5606 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5607 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5608 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5609 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5610 5611 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5612 key.type = BTRFS_CHUNK_ITEM_KEY; 5613 key.offset = bg->start; 5614 5615 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5616 if (ret) 5617 goto out; 5618 5619 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); 5620 5621 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5622 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5623 if (ret) 5624 goto out; 5625 } 5626 5627 out: 5628 kfree(chunk); 5629 free_extent_map(em); 5630 return ret; 5631 } 5632 5633 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5634 { 5635 struct btrfs_fs_info *fs_info = trans->fs_info; 5636 u64 alloc_profile; 5637 struct btrfs_block_group *meta_bg; 5638 struct btrfs_block_group *sys_bg; 5639 5640 /* 5641 * When adding a new device for sprouting, the seed device is read-only 5642 * so we must first allocate a metadata and a system chunk. But before 5643 * adding the block group items to the extent, device and chunk btrees, 5644 * we must first: 5645 * 5646 * 1) Create both chunks without doing any changes to the btrees, as 5647 * otherwise we would get -ENOSPC since the block groups from the 5648 * seed device are read-only; 5649 * 5650 * 2) Add the device item for the new sprout device - finishing the setup 5651 * of a new block group requires updating the device item in the chunk 5652 * btree, so it must exist when we attempt to do it. The previous step 5653 * ensures this does not fail with -ENOSPC. 5654 * 5655 * After that we can add the block group items to their btrees: 5656 * update existing device item in the chunk btree, add a new block group 5657 * item to the extent btree, add a new chunk item to the chunk btree and 5658 * finally add the new device extent items to the devices btree. 5659 */ 5660 5661 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5662 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5663 if (IS_ERR(meta_bg)) 5664 return PTR_ERR(meta_bg); 5665 5666 alloc_profile = btrfs_system_alloc_profile(fs_info); 5667 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5668 if (IS_ERR(sys_bg)) 5669 return PTR_ERR(sys_bg); 5670 5671 return 0; 5672 } 5673 5674 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5675 { 5676 const int index = btrfs_bg_flags_to_raid_index(map->type); 5677 5678 return btrfs_raid_array[index].tolerated_failures; 5679 } 5680 5681 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5682 { 5683 struct extent_map *em; 5684 struct map_lookup *map; 5685 int miss_ndevs = 0; 5686 int i; 5687 bool ret = true; 5688 5689 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5690 if (IS_ERR(em)) 5691 return false; 5692 5693 map = em->map_lookup; 5694 for (i = 0; i < map->num_stripes; i++) { 5695 if (test_bit(BTRFS_DEV_STATE_MISSING, 5696 &map->stripes[i].dev->dev_state)) { 5697 miss_ndevs++; 5698 continue; 5699 } 5700 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5701 &map->stripes[i].dev->dev_state)) { 5702 ret = false; 5703 goto end; 5704 } 5705 } 5706 5707 /* 5708 * If the number of missing devices is larger than max errors, we can 5709 * not write the data into that chunk successfully. 5710 */ 5711 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5712 ret = false; 5713 end: 5714 free_extent_map(em); 5715 return ret; 5716 } 5717 5718 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5719 { 5720 struct extent_map *em; 5721 5722 while (1) { 5723 write_lock(&tree->lock); 5724 em = lookup_extent_mapping(tree, 0, (u64)-1); 5725 if (em) 5726 remove_extent_mapping(tree, em); 5727 write_unlock(&tree->lock); 5728 if (!em) 5729 break; 5730 /* once for us */ 5731 free_extent_map(em); 5732 /* once for the tree */ 5733 free_extent_map(em); 5734 } 5735 } 5736 5737 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5738 { 5739 struct extent_map *em; 5740 struct map_lookup *map; 5741 enum btrfs_raid_types index; 5742 int ret = 1; 5743 5744 em = btrfs_get_chunk_map(fs_info, logical, len); 5745 if (IS_ERR(em)) 5746 /* 5747 * We could return errors for these cases, but that could get 5748 * ugly and we'd probably do the same thing which is just not do 5749 * anything else and exit, so return 1 so the callers don't try 5750 * to use other copies. 5751 */ 5752 return 1; 5753 5754 map = em->map_lookup; 5755 index = btrfs_bg_flags_to_raid_index(map->type); 5756 5757 /* Non-RAID56, use their ncopies from btrfs_raid_array. */ 5758 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5759 ret = btrfs_raid_array[index].ncopies; 5760 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5761 ret = 2; 5762 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5763 /* 5764 * There could be two corrupted data stripes, we need 5765 * to loop retry in order to rebuild the correct data. 5766 * 5767 * Fail a stripe at a time on every retry except the 5768 * stripe under reconstruction. 5769 */ 5770 ret = map->num_stripes; 5771 free_extent_map(em); 5772 5773 down_read(&fs_info->dev_replace.rwsem); 5774 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5775 fs_info->dev_replace.tgtdev) 5776 ret++; 5777 up_read(&fs_info->dev_replace.rwsem); 5778 5779 return ret; 5780 } 5781 5782 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5783 u64 logical) 5784 { 5785 struct extent_map *em; 5786 struct map_lookup *map; 5787 unsigned long len = fs_info->sectorsize; 5788 5789 if (!btrfs_fs_incompat(fs_info, RAID56)) 5790 return len; 5791 5792 em = btrfs_get_chunk_map(fs_info, logical, len); 5793 5794 if (!WARN_ON(IS_ERR(em))) { 5795 map = em->map_lookup; 5796 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5797 len = map->stripe_len * nr_data_stripes(map); 5798 free_extent_map(em); 5799 } 5800 return len; 5801 } 5802 5803 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5804 { 5805 struct extent_map *em; 5806 struct map_lookup *map; 5807 int ret = 0; 5808 5809 if (!btrfs_fs_incompat(fs_info, RAID56)) 5810 return 0; 5811 5812 em = btrfs_get_chunk_map(fs_info, logical, len); 5813 5814 if(!WARN_ON(IS_ERR(em))) { 5815 map = em->map_lookup; 5816 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5817 ret = 1; 5818 free_extent_map(em); 5819 } 5820 return ret; 5821 } 5822 5823 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5824 struct map_lookup *map, int first, 5825 int dev_replace_is_ongoing) 5826 { 5827 int i; 5828 int num_stripes; 5829 int preferred_mirror; 5830 int tolerance; 5831 struct btrfs_device *srcdev; 5832 5833 ASSERT((map->type & 5834 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5835 5836 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5837 num_stripes = map->sub_stripes; 5838 else 5839 num_stripes = map->num_stripes; 5840 5841 switch (fs_info->fs_devices->read_policy) { 5842 default: 5843 /* Shouldn't happen, just warn and use pid instead of failing */ 5844 btrfs_warn_rl(fs_info, 5845 "unknown read_policy type %u, reset to pid", 5846 fs_info->fs_devices->read_policy); 5847 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5848 fallthrough; 5849 case BTRFS_READ_POLICY_PID: 5850 preferred_mirror = first + (current->pid % num_stripes); 5851 break; 5852 } 5853 5854 if (dev_replace_is_ongoing && 5855 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5856 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5857 srcdev = fs_info->dev_replace.srcdev; 5858 else 5859 srcdev = NULL; 5860 5861 /* 5862 * try to avoid the drive that is the source drive for a 5863 * dev-replace procedure, only choose it if no other non-missing 5864 * mirror is available 5865 */ 5866 for (tolerance = 0; tolerance < 2; tolerance++) { 5867 if (map->stripes[preferred_mirror].dev->bdev && 5868 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5869 return preferred_mirror; 5870 for (i = first; i < first + num_stripes; i++) { 5871 if (map->stripes[i].dev->bdev && 5872 (tolerance || map->stripes[i].dev != srcdev)) 5873 return i; 5874 } 5875 } 5876 5877 /* we couldn't find one that doesn't fail. Just return something 5878 * and the io error handling code will clean up eventually 5879 */ 5880 return preferred_mirror; 5881 } 5882 5883 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5884 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5885 { 5886 int i; 5887 int again = 1; 5888 5889 while (again) { 5890 again = 0; 5891 for (i = 0; i < num_stripes - 1; i++) { 5892 /* Swap if parity is on a smaller index */ 5893 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5894 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5895 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5896 again = 1; 5897 } 5898 } 5899 } 5900 } 5901 5902 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5903 int total_stripes, 5904 int real_stripes) 5905 { 5906 struct btrfs_io_context *bioc = kzalloc( 5907 /* The size of btrfs_io_context */ 5908 sizeof(struct btrfs_io_context) + 5909 /* Plus the variable array for the stripes */ 5910 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5911 /* Plus the variable array for the tgt dev */ 5912 sizeof(int) * (real_stripes) + 5913 /* 5914 * Plus the raid_map, which includes both the tgt dev 5915 * and the stripes. 5916 */ 5917 sizeof(u64) * (total_stripes), 5918 GFP_NOFS|__GFP_NOFAIL); 5919 5920 refcount_set(&bioc->refs, 1); 5921 5922 bioc->fs_info = fs_info; 5923 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5924 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5925 5926 return bioc; 5927 } 5928 5929 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5930 { 5931 WARN_ON(!refcount_read(&bioc->refs)); 5932 refcount_inc(&bioc->refs); 5933 } 5934 5935 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5936 { 5937 if (!bioc) 5938 return; 5939 if (refcount_dec_and_test(&bioc->refs)) 5940 kfree(bioc); 5941 } 5942 5943 /* 5944 * Please note that, discard won't be sent to target device of device 5945 * replace. 5946 */ 5947 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, 5948 u64 logical, u64 *length_ret, 5949 u32 *num_stripes) 5950 { 5951 struct extent_map *em; 5952 struct map_lookup *map; 5953 struct btrfs_discard_stripe *stripes; 5954 u64 length = *length_ret; 5955 u64 offset; 5956 u64 stripe_nr; 5957 u64 stripe_nr_end; 5958 u64 stripe_end_offset; 5959 u64 stripe_cnt; 5960 u64 stripe_len; 5961 u64 stripe_offset; 5962 u32 stripe_index; 5963 u32 factor = 0; 5964 u32 sub_stripes = 0; 5965 u64 stripes_per_dev = 0; 5966 u32 remaining_stripes = 0; 5967 u32 last_stripe = 0; 5968 int ret; 5969 int i; 5970 5971 em = btrfs_get_chunk_map(fs_info, logical, length); 5972 if (IS_ERR(em)) 5973 return ERR_CAST(em); 5974 5975 map = em->map_lookup; 5976 5977 /* we don't discard raid56 yet */ 5978 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5979 ret = -EOPNOTSUPP; 5980 goto out_free_map; 5981 } 5982 5983 offset = logical - em->start; 5984 length = min_t(u64, em->start + em->len - logical, length); 5985 *length_ret = length; 5986 5987 stripe_len = map->stripe_len; 5988 /* 5989 * stripe_nr counts the total number of stripes we have to stride 5990 * to get to this block 5991 */ 5992 stripe_nr = div64_u64(offset, stripe_len); 5993 5994 /* stripe_offset is the offset of this block in its stripe */ 5995 stripe_offset = offset - stripe_nr * stripe_len; 5996 5997 stripe_nr_end = round_up(offset + length, map->stripe_len); 5998 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); 5999 stripe_cnt = stripe_nr_end - stripe_nr; 6000 stripe_end_offset = stripe_nr_end * map->stripe_len - 6001 (offset + length); 6002 /* 6003 * after this, stripe_nr is the number of stripes on this 6004 * device we have to walk to find the data, and stripe_index is 6005 * the number of our device in the stripe array 6006 */ 6007 *num_stripes = 1; 6008 stripe_index = 0; 6009 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6010 BTRFS_BLOCK_GROUP_RAID10)) { 6011 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 6012 sub_stripes = 1; 6013 else 6014 sub_stripes = map->sub_stripes; 6015 6016 factor = map->num_stripes / sub_stripes; 6017 *num_stripes = min_t(u64, map->num_stripes, 6018 sub_stripes * stripe_cnt); 6019 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6020 stripe_index *= sub_stripes; 6021 stripes_per_dev = div_u64_rem(stripe_cnt, factor, 6022 &remaining_stripes); 6023 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); 6024 last_stripe *= sub_stripes; 6025 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 6026 BTRFS_BLOCK_GROUP_DUP)) { 6027 *num_stripes = map->num_stripes; 6028 } else { 6029 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6030 &stripe_index); 6031 } 6032 6033 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); 6034 if (!stripes) { 6035 ret = -ENOMEM; 6036 goto out_free_map; 6037 } 6038 6039 for (i = 0; i < *num_stripes; i++) { 6040 stripes[i].physical = 6041 map->stripes[stripe_index].physical + 6042 stripe_offset + stripe_nr * map->stripe_len; 6043 stripes[i].dev = map->stripes[stripe_index].dev; 6044 6045 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6046 BTRFS_BLOCK_GROUP_RAID10)) { 6047 stripes[i].length = stripes_per_dev * map->stripe_len; 6048 6049 if (i / sub_stripes < remaining_stripes) 6050 stripes[i].length += map->stripe_len; 6051 6052 /* 6053 * Special for the first stripe and 6054 * the last stripe: 6055 * 6056 * |-------|...|-------| 6057 * |----------| 6058 * off end_off 6059 */ 6060 if (i < sub_stripes) 6061 stripes[i].length -= stripe_offset; 6062 6063 if (stripe_index >= last_stripe && 6064 stripe_index <= (last_stripe + 6065 sub_stripes - 1)) 6066 stripes[i].length -= stripe_end_offset; 6067 6068 if (i == sub_stripes - 1) 6069 stripe_offset = 0; 6070 } else { 6071 stripes[i].length = length; 6072 } 6073 6074 stripe_index++; 6075 if (stripe_index == map->num_stripes) { 6076 stripe_index = 0; 6077 stripe_nr++; 6078 } 6079 } 6080 6081 free_extent_map(em); 6082 return stripes; 6083 out_free_map: 6084 free_extent_map(em); 6085 return ERR_PTR(ret); 6086 } 6087 6088 /* 6089 * In dev-replace case, for repair case (that's the only case where the mirror 6090 * is selected explicitly when calling btrfs_map_block), blocks left of the 6091 * left cursor can also be read from the target drive. 6092 * 6093 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6094 * array of stripes. 6095 * For READ, it also needs to be supported using the same mirror number. 6096 * 6097 * If the requested block is not left of the left cursor, EIO is returned. This 6098 * can happen because btrfs_num_copies() returns one more in the dev-replace 6099 * case. 6100 */ 6101 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6102 u64 logical, u64 length, 6103 u64 srcdev_devid, int *mirror_num, 6104 u64 *physical) 6105 { 6106 struct btrfs_io_context *bioc = NULL; 6107 int num_stripes; 6108 int index_srcdev = 0; 6109 int found = 0; 6110 u64 physical_of_found = 0; 6111 int i; 6112 int ret = 0; 6113 6114 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6115 logical, &length, &bioc, NULL, NULL, 0); 6116 if (ret) { 6117 ASSERT(bioc == NULL); 6118 return ret; 6119 } 6120 6121 num_stripes = bioc->num_stripes; 6122 if (*mirror_num > num_stripes) { 6123 /* 6124 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6125 * that means that the requested area is not left of the left 6126 * cursor 6127 */ 6128 btrfs_put_bioc(bioc); 6129 return -EIO; 6130 } 6131 6132 /* 6133 * process the rest of the function using the mirror_num of the source 6134 * drive. Therefore look it up first. At the end, patch the device 6135 * pointer to the one of the target drive. 6136 */ 6137 for (i = 0; i < num_stripes; i++) { 6138 if (bioc->stripes[i].dev->devid != srcdev_devid) 6139 continue; 6140 6141 /* 6142 * In case of DUP, in order to keep it simple, only add the 6143 * mirror with the lowest physical address 6144 */ 6145 if (found && 6146 physical_of_found <= bioc->stripes[i].physical) 6147 continue; 6148 6149 index_srcdev = i; 6150 found = 1; 6151 physical_of_found = bioc->stripes[i].physical; 6152 } 6153 6154 btrfs_put_bioc(bioc); 6155 6156 ASSERT(found); 6157 if (!found) 6158 return -EIO; 6159 6160 *mirror_num = index_srcdev + 1; 6161 *physical = physical_of_found; 6162 return ret; 6163 } 6164 6165 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6166 { 6167 struct btrfs_block_group *cache; 6168 bool ret; 6169 6170 /* Non zoned filesystem does not use "to_copy" flag */ 6171 if (!btrfs_is_zoned(fs_info)) 6172 return false; 6173 6174 cache = btrfs_lookup_block_group(fs_info, logical); 6175 6176 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); 6177 6178 btrfs_put_block_group(cache); 6179 return ret; 6180 } 6181 6182 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6183 struct btrfs_io_context **bioc_ret, 6184 struct btrfs_dev_replace *dev_replace, 6185 u64 logical, 6186 int *num_stripes_ret, int *max_errors_ret) 6187 { 6188 struct btrfs_io_context *bioc = *bioc_ret; 6189 u64 srcdev_devid = dev_replace->srcdev->devid; 6190 int tgtdev_indexes = 0; 6191 int num_stripes = *num_stripes_ret; 6192 int max_errors = *max_errors_ret; 6193 int i; 6194 6195 if (op == BTRFS_MAP_WRITE) { 6196 int index_where_to_add; 6197 6198 /* 6199 * A block group which have "to_copy" set will eventually 6200 * copied by dev-replace process. We can avoid cloning IO here. 6201 */ 6202 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6203 return; 6204 6205 /* 6206 * duplicate the write operations while the dev replace 6207 * procedure is running. Since the copying of the old disk to 6208 * the new disk takes place at run time while the filesystem is 6209 * mounted writable, the regular write operations to the old 6210 * disk have to be duplicated to go to the new disk as well. 6211 * 6212 * Note that device->missing is handled by the caller, and that 6213 * the write to the old disk is already set up in the stripes 6214 * array. 6215 */ 6216 index_where_to_add = num_stripes; 6217 for (i = 0; i < num_stripes; i++) { 6218 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6219 /* write to new disk, too */ 6220 struct btrfs_io_stripe *new = 6221 bioc->stripes + index_where_to_add; 6222 struct btrfs_io_stripe *old = 6223 bioc->stripes + i; 6224 6225 new->physical = old->physical; 6226 new->dev = dev_replace->tgtdev; 6227 bioc->tgtdev_map[i] = index_where_to_add; 6228 index_where_to_add++; 6229 max_errors++; 6230 tgtdev_indexes++; 6231 } 6232 } 6233 num_stripes = index_where_to_add; 6234 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6235 int index_srcdev = 0; 6236 int found = 0; 6237 u64 physical_of_found = 0; 6238 6239 /* 6240 * During the dev-replace procedure, the target drive can also 6241 * be used to read data in case it is needed to repair a corrupt 6242 * block elsewhere. This is possible if the requested area is 6243 * left of the left cursor. In this area, the target drive is a 6244 * full copy of the source drive. 6245 */ 6246 for (i = 0; i < num_stripes; i++) { 6247 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6248 /* 6249 * In case of DUP, in order to keep it simple, 6250 * only add the mirror with the lowest physical 6251 * address 6252 */ 6253 if (found && 6254 physical_of_found <= bioc->stripes[i].physical) 6255 continue; 6256 index_srcdev = i; 6257 found = 1; 6258 physical_of_found = bioc->stripes[i].physical; 6259 } 6260 } 6261 if (found) { 6262 struct btrfs_io_stripe *tgtdev_stripe = 6263 bioc->stripes + num_stripes; 6264 6265 tgtdev_stripe->physical = physical_of_found; 6266 tgtdev_stripe->dev = dev_replace->tgtdev; 6267 bioc->tgtdev_map[index_srcdev] = num_stripes; 6268 6269 tgtdev_indexes++; 6270 num_stripes++; 6271 } 6272 } 6273 6274 *num_stripes_ret = num_stripes; 6275 *max_errors_ret = max_errors; 6276 bioc->num_tgtdevs = tgtdev_indexes; 6277 *bioc_ret = bioc; 6278 } 6279 6280 static bool need_full_stripe(enum btrfs_map_op op) 6281 { 6282 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6283 } 6284 6285 /* 6286 * Calculate the geometry of a particular (address, len) tuple. This 6287 * information is used to calculate how big a particular bio can get before it 6288 * straddles a stripe. 6289 * 6290 * @fs_info: the filesystem 6291 * @em: mapping containing the logical extent 6292 * @op: type of operation - write or read 6293 * @logical: address that we want to figure out the geometry of 6294 * @io_geom: pointer used to return values 6295 * 6296 * Returns < 0 in case a chunk for the given logical address cannot be found, 6297 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6298 */ 6299 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6300 enum btrfs_map_op op, u64 logical, 6301 struct btrfs_io_geometry *io_geom) 6302 { 6303 struct map_lookup *map; 6304 u64 len; 6305 u64 offset; 6306 u64 stripe_offset; 6307 u64 stripe_nr; 6308 u32 stripe_len; 6309 u64 raid56_full_stripe_start = (u64)-1; 6310 int data_stripes; 6311 6312 ASSERT(op != BTRFS_MAP_DISCARD); 6313 6314 map = em->map_lookup; 6315 /* Offset of this logical address in the chunk */ 6316 offset = logical - em->start; 6317 /* Len of a stripe in a chunk */ 6318 stripe_len = map->stripe_len; 6319 /* 6320 * Stripe_nr is where this block falls in 6321 * stripe_offset is the offset of this block in its stripe. 6322 */ 6323 stripe_nr = div64_u64_rem(offset, stripe_len, &stripe_offset); 6324 ASSERT(stripe_offset < U32_MAX); 6325 6326 data_stripes = nr_data_stripes(map); 6327 6328 /* Only stripe based profiles needs to check against stripe length. */ 6329 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) { 6330 u64 max_len = stripe_len - stripe_offset; 6331 6332 /* 6333 * In case of raid56, we need to know the stripe aligned start 6334 */ 6335 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6336 unsigned long full_stripe_len = stripe_len * data_stripes; 6337 raid56_full_stripe_start = offset; 6338 6339 /* 6340 * Allow a write of a full stripe, but make sure we 6341 * don't allow straddling of stripes 6342 */ 6343 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6344 full_stripe_len); 6345 raid56_full_stripe_start *= full_stripe_len; 6346 6347 /* 6348 * For writes to RAID[56], allow a full stripeset across 6349 * all disks. For other RAID types and for RAID[56] 6350 * reads, just allow a single stripe (on a single disk). 6351 */ 6352 if (op == BTRFS_MAP_WRITE) { 6353 max_len = stripe_len * data_stripes - 6354 (offset - raid56_full_stripe_start); 6355 } 6356 } 6357 len = min_t(u64, em->len - offset, max_len); 6358 } else { 6359 len = em->len - offset; 6360 } 6361 6362 io_geom->len = len; 6363 io_geom->offset = offset; 6364 io_geom->stripe_len = stripe_len; 6365 io_geom->stripe_nr = stripe_nr; 6366 io_geom->stripe_offset = stripe_offset; 6367 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6368 6369 return 0; 6370 } 6371 6372 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, 6373 u32 stripe_index, u64 stripe_offset, u64 stripe_nr) 6374 { 6375 dst->dev = map->stripes[stripe_index].dev; 6376 dst->physical = map->stripes[stripe_index].physical + 6377 stripe_offset + stripe_nr * map->stripe_len; 6378 } 6379 6380 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6381 enum btrfs_map_op op, u64 logical, u64 *length, 6382 struct btrfs_io_context **bioc_ret, 6383 struct btrfs_io_stripe *smap, 6384 int *mirror_num_ret, int need_raid_map) 6385 { 6386 struct extent_map *em; 6387 struct map_lookup *map; 6388 u64 stripe_offset; 6389 u64 stripe_nr; 6390 u64 stripe_len; 6391 u32 stripe_index; 6392 int data_stripes; 6393 int i; 6394 int ret = 0; 6395 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); 6396 int num_stripes; 6397 int max_errors = 0; 6398 int tgtdev_indexes = 0; 6399 struct btrfs_io_context *bioc = NULL; 6400 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6401 int dev_replace_is_ongoing = 0; 6402 int num_alloc_stripes; 6403 int patch_the_first_stripe_for_dev_replace = 0; 6404 u64 physical_to_patch_in_first_stripe = 0; 6405 u64 raid56_full_stripe_start = (u64)-1; 6406 struct btrfs_io_geometry geom; 6407 6408 ASSERT(bioc_ret); 6409 ASSERT(op != BTRFS_MAP_DISCARD); 6410 6411 em = btrfs_get_chunk_map(fs_info, logical, *length); 6412 ASSERT(!IS_ERR(em)); 6413 6414 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6415 if (ret < 0) 6416 return ret; 6417 6418 map = em->map_lookup; 6419 6420 *length = geom.len; 6421 stripe_len = geom.stripe_len; 6422 stripe_nr = geom.stripe_nr; 6423 stripe_offset = geom.stripe_offset; 6424 raid56_full_stripe_start = geom.raid56_stripe_offset; 6425 data_stripes = nr_data_stripes(map); 6426 6427 down_read(&dev_replace->rwsem); 6428 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6429 /* 6430 * Hold the semaphore for read during the whole operation, write is 6431 * requested at commit time but must wait. 6432 */ 6433 if (!dev_replace_is_ongoing) 6434 up_read(&dev_replace->rwsem); 6435 6436 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6437 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6438 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6439 dev_replace->srcdev->devid, 6440 &mirror_num, 6441 &physical_to_patch_in_first_stripe); 6442 if (ret) 6443 goto out; 6444 else 6445 patch_the_first_stripe_for_dev_replace = 1; 6446 } else if (mirror_num > map->num_stripes) { 6447 mirror_num = 0; 6448 } 6449 6450 num_stripes = 1; 6451 stripe_index = 0; 6452 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6453 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6454 &stripe_index); 6455 if (!need_full_stripe(op)) 6456 mirror_num = 1; 6457 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6458 if (need_full_stripe(op)) 6459 num_stripes = map->num_stripes; 6460 else if (mirror_num) 6461 stripe_index = mirror_num - 1; 6462 else { 6463 stripe_index = find_live_mirror(fs_info, map, 0, 6464 dev_replace_is_ongoing); 6465 mirror_num = stripe_index + 1; 6466 } 6467 6468 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6469 if (need_full_stripe(op)) { 6470 num_stripes = map->num_stripes; 6471 } else if (mirror_num) { 6472 stripe_index = mirror_num - 1; 6473 } else { 6474 mirror_num = 1; 6475 } 6476 6477 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6478 u32 factor = map->num_stripes / map->sub_stripes; 6479 6480 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6481 stripe_index *= map->sub_stripes; 6482 6483 if (need_full_stripe(op)) 6484 num_stripes = map->sub_stripes; 6485 else if (mirror_num) 6486 stripe_index += mirror_num - 1; 6487 else { 6488 int old_stripe_index = stripe_index; 6489 stripe_index = find_live_mirror(fs_info, map, 6490 stripe_index, 6491 dev_replace_is_ongoing); 6492 mirror_num = stripe_index - old_stripe_index + 1; 6493 } 6494 6495 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6496 ASSERT(map->stripe_len == BTRFS_STRIPE_LEN); 6497 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6498 /* push stripe_nr back to the start of the full stripe */ 6499 stripe_nr = div64_u64(raid56_full_stripe_start, 6500 stripe_len * data_stripes); 6501 6502 /* RAID[56] write or recovery. Return all stripes */ 6503 num_stripes = map->num_stripes; 6504 max_errors = btrfs_chunk_max_errors(map); 6505 6506 /* Return the length to the full stripe end */ 6507 *length = min(logical + *length, 6508 raid56_full_stripe_start + em->start + 6509 data_stripes * stripe_len) - logical; 6510 stripe_index = 0; 6511 stripe_offset = 0; 6512 } else { 6513 /* 6514 * Mirror #0 or #1 means the original data block. 6515 * Mirror #2 is RAID5 parity block. 6516 * Mirror #3 is RAID6 Q block. 6517 */ 6518 stripe_nr = div_u64_rem(stripe_nr, 6519 data_stripes, &stripe_index); 6520 if (mirror_num > 1) 6521 stripe_index = data_stripes + mirror_num - 2; 6522 6523 /* We distribute the parity blocks across stripes */ 6524 div_u64_rem(stripe_nr + stripe_index, map->num_stripes, 6525 &stripe_index); 6526 if (!need_full_stripe(op) && mirror_num <= 1) 6527 mirror_num = 1; 6528 } 6529 } else { 6530 /* 6531 * after this, stripe_nr is the number of stripes on this 6532 * device we have to walk to find the data, and stripe_index is 6533 * the number of our device in the stripe array 6534 */ 6535 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6536 &stripe_index); 6537 mirror_num = stripe_index + 1; 6538 } 6539 if (stripe_index >= map->num_stripes) { 6540 btrfs_crit(fs_info, 6541 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6542 stripe_index, map->num_stripes); 6543 ret = -EINVAL; 6544 goto out; 6545 } 6546 6547 num_alloc_stripes = num_stripes; 6548 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6549 if (op == BTRFS_MAP_WRITE) 6550 num_alloc_stripes <<= 1; 6551 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6552 num_alloc_stripes++; 6553 tgtdev_indexes = num_stripes; 6554 } 6555 6556 /* 6557 * If this I/O maps to a single device, try to return the device and 6558 * physical block information on the stack instead of allocating an 6559 * I/O context structure. 6560 */ 6561 if (smap && num_alloc_stripes == 1 && 6562 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && 6563 (!need_full_stripe(op) || !dev_replace_is_ongoing || 6564 !dev_replace->tgtdev)) { 6565 if (patch_the_first_stripe_for_dev_replace) { 6566 smap->dev = dev_replace->tgtdev; 6567 smap->physical = physical_to_patch_in_first_stripe; 6568 *mirror_num_ret = map->num_stripes + 1; 6569 } else { 6570 set_io_stripe(smap, map, stripe_index, stripe_offset, 6571 stripe_nr); 6572 *mirror_num_ret = mirror_num; 6573 } 6574 *bioc_ret = NULL; 6575 ret = 0; 6576 goto out; 6577 } 6578 6579 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6580 if (!bioc) { 6581 ret = -ENOMEM; 6582 goto out; 6583 } 6584 6585 for (i = 0; i < num_stripes; i++) { 6586 set_io_stripe(&bioc->stripes[i], map, stripe_index, stripe_offset, 6587 stripe_nr); 6588 stripe_index++; 6589 } 6590 6591 /* Build raid_map */ 6592 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6593 (need_full_stripe(op) || mirror_num > 1)) { 6594 u64 tmp; 6595 unsigned rot; 6596 6597 /* Work out the disk rotation on this stripe-set */ 6598 div_u64_rem(stripe_nr, num_stripes, &rot); 6599 6600 /* Fill in the logical address of each stripe */ 6601 tmp = stripe_nr * data_stripes; 6602 for (i = 0; i < data_stripes; i++) 6603 bioc->raid_map[(i + rot) % num_stripes] = 6604 em->start + (tmp + i) * map->stripe_len; 6605 6606 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6607 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6608 bioc->raid_map[(i + rot + 1) % num_stripes] = 6609 RAID6_Q_STRIPE; 6610 6611 sort_parity_stripes(bioc, num_stripes); 6612 } 6613 6614 if (need_full_stripe(op)) 6615 max_errors = btrfs_chunk_max_errors(map); 6616 6617 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6618 need_full_stripe(op)) { 6619 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6620 &num_stripes, &max_errors); 6621 } 6622 6623 *bioc_ret = bioc; 6624 bioc->map_type = map->type; 6625 bioc->num_stripes = num_stripes; 6626 bioc->max_errors = max_errors; 6627 bioc->mirror_num = mirror_num; 6628 6629 /* 6630 * this is the case that REQ_READ && dev_replace_is_ongoing && 6631 * mirror_num == num_stripes + 1 && dev_replace target drive is 6632 * available as a mirror 6633 */ 6634 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6635 WARN_ON(num_stripes > 1); 6636 bioc->stripes[0].dev = dev_replace->tgtdev; 6637 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6638 bioc->mirror_num = map->num_stripes + 1; 6639 } 6640 out: 6641 if (dev_replace_is_ongoing) { 6642 lockdep_assert_held(&dev_replace->rwsem); 6643 /* Unlock and let waiting writers proceed */ 6644 up_read(&dev_replace->rwsem); 6645 } 6646 free_extent_map(em); 6647 return ret; 6648 } 6649 6650 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6651 u64 logical, u64 *length, 6652 struct btrfs_io_context **bioc_ret, int mirror_num) 6653 { 6654 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6655 NULL, &mirror_num, 0); 6656 } 6657 6658 /* For Scrub/replace */ 6659 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6660 u64 logical, u64 *length, 6661 struct btrfs_io_context **bioc_ret) 6662 { 6663 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6664 NULL, NULL, 1); 6665 } 6666 6667 /* 6668 * Initialize a btrfs_bio structure. This skips the embedded bio itself as it 6669 * is already initialized by the block layer. 6670 */ 6671 static inline void btrfs_bio_init(struct btrfs_bio *bbio, 6672 btrfs_bio_end_io_t end_io, void *private) 6673 { 6674 memset(bbio, 0, offsetof(struct btrfs_bio, bio)); 6675 bbio->end_io = end_io; 6676 bbio->private = private; 6677 } 6678 6679 /* 6680 * Allocate a btrfs_bio structure. The btrfs_bio is the main I/O container for 6681 * btrfs, and is used for all I/O submitted through btrfs_submit_bio. 6682 * 6683 * Just like the underlying bio_alloc_bioset it will not fail as it is backed by 6684 * a mempool. 6685 */ 6686 struct bio *btrfs_bio_alloc(unsigned int nr_vecs, blk_opf_t opf, 6687 btrfs_bio_end_io_t end_io, void *private) 6688 { 6689 struct bio *bio; 6690 6691 bio = bio_alloc_bioset(NULL, nr_vecs, opf, GFP_NOFS, &btrfs_bioset); 6692 btrfs_bio_init(btrfs_bio(bio), end_io, private); 6693 return bio; 6694 } 6695 6696 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size, 6697 btrfs_bio_end_io_t end_io, void *private) 6698 { 6699 struct bio *bio; 6700 struct btrfs_bio *bbio; 6701 6702 ASSERT(offset <= UINT_MAX && size <= UINT_MAX); 6703 6704 bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset); 6705 bbio = btrfs_bio(bio); 6706 btrfs_bio_init(bbio, end_io, private); 6707 6708 bio_trim(bio, offset >> 9, size >> 9); 6709 bbio->iter = bio->bi_iter; 6710 return bio; 6711 } 6712 6713 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev) 6714 { 6715 if (!dev || !dev->bdev) 6716 return; 6717 if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET) 6718 return; 6719 6720 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6721 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); 6722 if (!(bio->bi_opf & REQ_RAHEAD)) 6723 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); 6724 if (bio->bi_opf & REQ_PREFLUSH) 6725 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS); 6726 } 6727 6728 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info, 6729 struct bio *bio) 6730 { 6731 if (bio->bi_opf & REQ_META) 6732 return fs_info->endio_meta_workers; 6733 return fs_info->endio_workers; 6734 } 6735 6736 static void btrfs_end_bio_work(struct work_struct *work) 6737 { 6738 struct btrfs_bio *bbio = 6739 container_of(work, struct btrfs_bio, end_io_work); 6740 6741 bbio->end_io(bbio); 6742 } 6743 6744 static void btrfs_simple_end_io(struct bio *bio) 6745 { 6746 struct btrfs_fs_info *fs_info = bio->bi_private; 6747 struct btrfs_bio *bbio = btrfs_bio(bio); 6748 6749 btrfs_bio_counter_dec(fs_info); 6750 6751 if (bio->bi_status) 6752 btrfs_log_dev_io_error(bio, bbio->device); 6753 6754 if (bio_op(bio) == REQ_OP_READ) { 6755 INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work); 6756 queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work); 6757 } else { 6758 bbio->end_io(bbio); 6759 } 6760 } 6761 6762 static void btrfs_raid56_end_io(struct bio *bio) 6763 { 6764 struct btrfs_io_context *bioc = bio->bi_private; 6765 struct btrfs_bio *bbio = btrfs_bio(bio); 6766 6767 btrfs_bio_counter_dec(bioc->fs_info); 6768 bbio->mirror_num = bioc->mirror_num; 6769 bbio->end_io(bbio); 6770 6771 btrfs_put_bioc(bioc); 6772 } 6773 6774 static void btrfs_orig_write_end_io(struct bio *bio) 6775 { 6776 struct btrfs_io_stripe *stripe = bio->bi_private; 6777 struct btrfs_io_context *bioc = stripe->bioc; 6778 struct btrfs_bio *bbio = btrfs_bio(bio); 6779 6780 btrfs_bio_counter_dec(bioc->fs_info); 6781 6782 if (bio->bi_status) { 6783 atomic_inc(&bioc->error); 6784 btrfs_log_dev_io_error(bio, stripe->dev); 6785 } 6786 6787 /* 6788 * Only send an error to the higher layers if it is beyond the tolerance 6789 * threshold. 6790 */ 6791 if (atomic_read(&bioc->error) > bioc->max_errors) 6792 bio->bi_status = BLK_STS_IOERR; 6793 else 6794 bio->bi_status = BLK_STS_OK; 6795 6796 bbio->end_io(bbio); 6797 btrfs_put_bioc(bioc); 6798 } 6799 6800 static void btrfs_clone_write_end_io(struct bio *bio) 6801 { 6802 struct btrfs_io_stripe *stripe = bio->bi_private; 6803 6804 if (bio->bi_status) { 6805 atomic_inc(&stripe->bioc->error); 6806 btrfs_log_dev_io_error(bio, stripe->dev); 6807 } 6808 6809 /* Pass on control to the original bio this one was cloned from */ 6810 bio_endio(stripe->bioc->orig_bio); 6811 bio_put(bio); 6812 } 6813 6814 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio) 6815 { 6816 if (!dev || !dev->bdev || 6817 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 6818 (btrfs_op(bio) == BTRFS_MAP_WRITE && 6819 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6820 bio_io_error(bio); 6821 return; 6822 } 6823 6824 bio_set_dev(bio, dev->bdev); 6825 6826 /* 6827 * For zone append writing, bi_sector must point the beginning of the 6828 * zone 6829 */ 6830 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6831 u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; 6832 6833 if (btrfs_dev_is_sequential(dev, physical)) { 6834 u64 zone_start = round_down(physical, 6835 dev->fs_info->zone_size); 6836 6837 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6838 } else { 6839 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6840 bio->bi_opf |= REQ_OP_WRITE; 6841 } 6842 } 6843 btrfs_debug_in_rcu(dev->fs_info, 6844 "%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6845 __func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6846 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6847 dev->devid, bio->bi_iter.bi_size); 6848 6849 btrfsic_check_bio(bio); 6850 submit_bio(bio); 6851 } 6852 6853 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr) 6854 { 6855 struct bio *orig_bio = bioc->orig_bio, *bio; 6856 6857 ASSERT(bio_op(orig_bio) != REQ_OP_READ); 6858 6859 /* Reuse the bio embedded into the btrfs_bio for the last mirror */ 6860 if (dev_nr == bioc->num_stripes - 1) { 6861 bio = orig_bio; 6862 bio->bi_end_io = btrfs_orig_write_end_io; 6863 } else { 6864 bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set); 6865 bio_inc_remaining(orig_bio); 6866 bio->bi_end_io = btrfs_clone_write_end_io; 6867 } 6868 6869 bio->bi_private = &bioc->stripes[dev_nr]; 6870 bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT; 6871 bioc->stripes[dev_nr].bioc = bioc; 6872 btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio); 6873 } 6874 6875 void btrfs_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, int mirror_num) 6876 { 6877 u64 logical = bio->bi_iter.bi_sector << 9; 6878 u64 length = bio->bi_iter.bi_size; 6879 u64 map_length = length; 6880 struct btrfs_io_context *bioc = NULL; 6881 struct btrfs_io_stripe smap; 6882 int ret; 6883 6884 btrfs_bio_counter_inc_blocked(fs_info); 6885 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, 6886 &bioc, &smap, &mirror_num, 1); 6887 if (ret) { 6888 btrfs_bio_counter_dec(fs_info); 6889 btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret)); 6890 return; 6891 } 6892 6893 if (map_length < length) { 6894 btrfs_crit(fs_info, 6895 "mapping failed logical %llu bio len %llu len %llu", 6896 logical, length, map_length); 6897 BUG(); 6898 } 6899 6900 if (!bioc) { 6901 /* Single mirror read/write fast path */ 6902 btrfs_bio(bio)->mirror_num = mirror_num; 6903 btrfs_bio(bio)->device = smap.dev; 6904 bio->bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT; 6905 bio->bi_private = fs_info; 6906 bio->bi_end_io = btrfs_simple_end_io; 6907 btrfs_submit_dev_bio(smap.dev, bio); 6908 } else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6909 /* Parity RAID write or read recovery */ 6910 bio->bi_private = bioc; 6911 bio->bi_end_io = btrfs_raid56_end_io; 6912 if (bio_op(bio) == REQ_OP_READ) 6913 raid56_parity_recover(bio, bioc, mirror_num); 6914 else 6915 raid56_parity_write(bio, bioc); 6916 } else { 6917 /* Write to multiple mirrors */ 6918 int total_devs = bioc->num_stripes; 6919 int dev_nr; 6920 6921 bioc->orig_bio = bio; 6922 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) 6923 btrfs_submit_mirrored_bio(bioc, dev_nr); 6924 } 6925 } 6926 6927 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6928 const struct btrfs_fs_devices *fs_devices) 6929 { 6930 if (args->fsid == NULL) 6931 return true; 6932 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6933 return true; 6934 return false; 6935 } 6936 6937 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6938 const struct btrfs_device *device) 6939 { 6940 if (args->missing) { 6941 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6942 !device->bdev) 6943 return true; 6944 return false; 6945 } 6946 6947 if (device->devid != args->devid) 6948 return false; 6949 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6950 return false; 6951 return true; 6952 } 6953 6954 /* 6955 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6956 * return NULL. 6957 * 6958 * If devid and uuid are both specified, the match must be exact, otherwise 6959 * only devid is used. 6960 */ 6961 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6962 const struct btrfs_dev_lookup_args *args) 6963 { 6964 struct btrfs_device *device; 6965 struct btrfs_fs_devices *seed_devs; 6966 6967 if (dev_args_match_fs_devices(args, fs_devices)) { 6968 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6969 if (dev_args_match_device(args, device)) 6970 return device; 6971 } 6972 } 6973 6974 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6975 if (!dev_args_match_fs_devices(args, seed_devs)) 6976 continue; 6977 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6978 if (dev_args_match_device(args, device)) 6979 return device; 6980 } 6981 } 6982 6983 return NULL; 6984 } 6985 6986 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6987 u64 devid, u8 *dev_uuid) 6988 { 6989 struct btrfs_device *device; 6990 unsigned int nofs_flag; 6991 6992 /* 6993 * We call this under the chunk_mutex, so we want to use NOFS for this 6994 * allocation, however we don't want to change btrfs_alloc_device() to 6995 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6996 * places. 6997 */ 6998 nofs_flag = memalloc_nofs_save(); 6999 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 7000 memalloc_nofs_restore(nofs_flag); 7001 if (IS_ERR(device)) 7002 return device; 7003 7004 list_add(&device->dev_list, &fs_devices->devices); 7005 device->fs_devices = fs_devices; 7006 fs_devices->num_devices++; 7007 7008 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7009 fs_devices->missing_devices++; 7010 7011 return device; 7012 } 7013 7014 /* 7015 * Allocate new device struct, set up devid and UUID. 7016 * 7017 * @fs_info: used only for generating a new devid, can be NULL if 7018 * devid is provided (i.e. @devid != NULL). 7019 * @devid: a pointer to devid for this device. If NULL a new devid 7020 * is generated. 7021 * @uuid: a pointer to UUID for this device. If NULL a new UUID 7022 * is generated. 7023 * 7024 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 7025 * on error. Returned struct is not linked onto any lists and must be 7026 * destroyed with btrfs_free_device. 7027 */ 7028 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 7029 const u64 *devid, 7030 const u8 *uuid) 7031 { 7032 struct btrfs_device *dev; 7033 u64 tmp; 7034 7035 if (WARN_ON(!devid && !fs_info)) 7036 return ERR_PTR(-EINVAL); 7037 7038 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 7039 if (!dev) 7040 return ERR_PTR(-ENOMEM); 7041 7042 INIT_LIST_HEAD(&dev->dev_list); 7043 INIT_LIST_HEAD(&dev->dev_alloc_list); 7044 INIT_LIST_HEAD(&dev->post_commit_list); 7045 7046 atomic_set(&dev->dev_stats_ccnt, 0); 7047 btrfs_device_data_ordered_init(dev); 7048 extent_io_tree_init(fs_info, &dev->alloc_state, 7049 IO_TREE_DEVICE_ALLOC_STATE, NULL); 7050 7051 if (devid) 7052 tmp = *devid; 7053 else { 7054 int ret; 7055 7056 ret = find_next_devid(fs_info, &tmp); 7057 if (ret) { 7058 btrfs_free_device(dev); 7059 return ERR_PTR(ret); 7060 } 7061 } 7062 dev->devid = tmp; 7063 7064 if (uuid) 7065 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 7066 else 7067 generate_random_uuid(dev->uuid); 7068 7069 return dev; 7070 } 7071 7072 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 7073 u64 devid, u8 *uuid, bool error) 7074 { 7075 if (error) 7076 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 7077 devid, uuid); 7078 else 7079 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 7080 devid, uuid); 7081 } 7082 7083 u64 btrfs_calc_stripe_length(const struct extent_map *em) 7084 { 7085 const struct map_lookup *map = em->map_lookup; 7086 const int data_stripes = calc_data_stripes(map->type, map->num_stripes); 7087 7088 return div_u64(em->len, data_stripes); 7089 } 7090 7091 #if BITS_PER_LONG == 32 7092 /* 7093 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 7094 * can't be accessed on 32bit systems. 7095 * 7096 * This function do mount time check to reject the fs if it already has 7097 * metadata chunk beyond that limit. 7098 */ 7099 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7100 u64 logical, u64 length, u64 type) 7101 { 7102 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7103 return 0; 7104 7105 if (logical + length < MAX_LFS_FILESIZE) 7106 return 0; 7107 7108 btrfs_err_32bit_limit(fs_info); 7109 return -EOVERFLOW; 7110 } 7111 7112 /* 7113 * This is to give early warning for any metadata chunk reaching 7114 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 7115 * Although we can still access the metadata, it's not going to be possible 7116 * once the limit is reached. 7117 */ 7118 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7119 u64 logical, u64 length, u64 type) 7120 { 7121 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7122 return; 7123 7124 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 7125 return; 7126 7127 btrfs_warn_32bit_limit(fs_info); 7128 } 7129 #endif 7130 7131 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 7132 u64 devid, u8 *uuid) 7133 { 7134 struct btrfs_device *dev; 7135 7136 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7137 btrfs_report_missing_device(fs_info, devid, uuid, true); 7138 return ERR_PTR(-ENOENT); 7139 } 7140 7141 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 7142 if (IS_ERR(dev)) { 7143 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 7144 devid, PTR_ERR(dev)); 7145 return dev; 7146 } 7147 btrfs_report_missing_device(fs_info, devid, uuid, false); 7148 7149 return dev; 7150 } 7151 7152 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 7153 struct btrfs_chunk *chunk) 7154 { 7155 BTRFS_DEV_LOOKUP_ARGS(args); 7156 struct btrfs_fs_info *fs_info = leaf->fs_info; 7157 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7158 struct map_lookup *map; 7159 struct extent_map *em; 7160 u64 logical; 7161 u64 length; 7162 u64 devid; 7163 u64 type; 7164 u8 uuid[BTRFS_UUID_SIZE]; 7165 int index; 7166 int num_stripes; 7167 int ret; 7168 int i; 7169 7170 logical = key->offset; 7171 length = btrfs_chunk_length(leaf, chunk); 7172 type = btrfs_chunk_type(leaf, chunk); 7173 index = btrfs_bg_flags_to_raid_index(type); 7174 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 7175 7176 #if BITS_PER_LONG == 32 7177 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 7178 if (ret < 0) 7179 return ret; 7180 warn_32bit_meta_chunk(fs_info, logical, length, type); 7181 #endif 7182 7183 /* 7184 * Only need to verify chunk item if we're reading from sys chunk array, 7185 * as chunk item in tree block is already verified by tree-checker. 7186 */ 7187 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 7188 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 7189 if (ret) 7190 return ret; 7191 } 7192 7193 read_lock(&map_tree->lock); 7194 em = lookup_extent_mapping(map_tree, logical, 1); 7195 read_unlock(&map_tree->lock); 7196 7197 /* already mapped? */ 7198 if (em && em->start <= logical && em->start + em->len > logical) { 7199 free_extent_map(em); 7200 return 0; 7201 } else if (em) { 7202 free_extent_map(em); 7203 } 7204 7205 em = alloc_extent_map(); 7206 if (!em) 7207 return -ENOMEM; 7208 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7209 if (!map) { 7210 free_extent_map(em); 7211 return -ENOMEM; 7212 } 7213 7214 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7215 em->map_lookup = map; 7216 em->start = logical; 7217 em->len = length; 7218 em->orig_start = 0; 7219 em->block_start = 0; 7220 em->block_len = em->len; 7221 7222 map->num_stripes = num_stripes; 7223 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7224 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7225 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7226 map->type = type; 7227 /* 7228 * We can't use the sub_stripes value, as for profiles other than 7229 * RAID10, they may have 0 as sub_stripes for filesystems created by 7230 * older mkfs (<v5.4). 7231 * In that case, it can cause divide-by-zero errors later. 7232 * Since currently sub_stripes is fixed for each profile, let's 7233 * use the trusted value instead. 7234 */ 7235 map->sub_stripes = btrfs_raid_array[index].sub_stripes; 7236 map->verified_stripes = 0; 7237 em->orig_block_len = btrfs_calc_stripe_length(em); 7238 for (i = 0; i < num_stripes; i++) { 7239 map->stripes[i].physical = 7240 btrfs_stripe_offset_nr(leaf, chunk, i); 7241 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7242 args.devid = devid; 7243 read_extent_buffer(leaf, uuid, (unsigned long) 7244 btrfs_stripe_dev_uuid_nr(chunk, i), 7245 BTRFS_UUID_SIZE); 7246 args.uuid = uuid; 7247 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 7248 if (!map->stripes[i].dev) { 7249 map->stripes[i].dev = handle_missing_device(fs_info, 7250 devid, uuid); 7251 if (IS_ERR(map->stripes[i].dev)) { 7252 free_extent_map(em); 7253 return PTR_ERR(map->stripes[i].dev); 7254 } 7255 } 7256 7257 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7258 &(map->stripes[i].dev->dev_state)); 7259 } 7260 7261 write_lock(&map_tree->lock); 7262 ret = add_extent_mapping(map_tree, em, 0); 7263 write_unlock(&map_tree->lock); 7264 if (ret < 0) { 7265 btrfs_err(fs_info, 7266 "failed to add chunk map, start=%llu len=%llu: %d", 7267 em->start, em->len, ret); 7268 } 7269 free_extent_map(em); 7270 7271 return ret; 7272 } 7273 7274 static void fill_device_from_item(struct extent_buffer *leaf, 7275 struct btrfs_dev_item *dev_item, 7276 struct btrfs_device *device) 7277 { 7278 unsigned long ptr; 7279 7280 device->devid = btrfs_device_id(leaf, dev_item); 7281 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7282 device->total_bytes = device->disk_total_bytes; 7283 device->commit_total_bytes = device->disk_total_bytes; 7284 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7285 device->commit_bytes_used = device->bytes_used; 7286 device->type = btrfs_device_type(leaf, dev_item); 7287 device->io_align = btrfs_device_io_align(leaf, dev_item); 7288 device->io_width = btrfs_device_io_width(leaf, dev_item); 7289 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7290 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7291 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7292 7293 ptr = btrfs_device_uuid(dev_item); 7294 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7295 } 7296 7297 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7298 u8 *fsid) 7299 { 7300 struct btrfs_fs_devices *fs_devices; 7301 int ret; 7302 7303 lockdep_assert_held(&uuid_mutex); 7304 ASSERT(fsid); 7305 7306 /* This will match only for multi-device seed fs */ 7307 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7308 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7309 return fs_devices; 7310 7311 7312 fs_devices = find_fsid(fsid, NULL); 7313 if (!fs_devices) { 7314 if (!btrfs_test_opt(fs_info, DEGRADED)) 7315 return ERR_PTR(-ENOENT); 7316 7317 fs_devices = alloc_fs_devices(fsid, NULL); 7318 if (IS_ERR(fs_devices)) 7319 return fs_devices; 7320 7321 fs_devices->seeding = true; 7322 fs_devices->opened = 1; 7323 return fs_devices; 7324 } 7325 7326 /* 7327 * Upon first call for a seed fs fsid, just create a private copy of the 7328 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7329 */ 7330 fs_devices = clone_fs_devices(fs_devices); 7331 if (IS_ERR(fs_devices)) 7332 return fs_devices; 7333 7334 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7335 if (ret) { 7336 free_fs_devices(fs_devices); 7337 return ERR_PTR(ret); 7338 } 7339 7340 if (!fs_devices->seeding) { 7341 close_fs_devices(fs_devices); 7342 free_fs_devices(fs_devices); 7343 return ERR_PTR(-EINVAL); 7344 } 7345 7346 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7347 7348 return fs_devices; 7349 } 7350 7351 static int read_one_dev(struct extent_buffer *leaf, 7352 struct btrfs_dev_item *dev_item) 7353 { 7354 BTRFS_DEV_LOOKUP_ARGS(args); 7355 struct btrfs_fs_info *fs_info = leaf->fs_info; 7356 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7357 struct btrfs_device *device; 7358 u64 devid; 7359 int ret; 7360 u8 fs_uuid[BTRFS_FSID_SIZE]; 7361 u8 dev_uuid[BTRFS_UUID_SIZE]; 7362 7363 devid = btrfs_device_id(leaf, dev_item); 7364 args.devid = devid; 7365 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7366 BTRFS_UUID_SIZE); 7367 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7368 BTRFS_FSID_SIZE); 7369 args.uuid = dev_uuid; 7370 args.fsid = fs_uuid; 7371 7372 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7373 fs_devices = open_seed_devices(fs_info, fs_uuid); 7374 if (IS_ERR(fs_devices)) 7375 return PTR_ERR(fs_devices); 7376 } 7377 7378 device = btrfs_find_device(fs_info->fs_devices, &args); 7379 if (!device) { 7380 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7381 btrfs_report_missing_device(fs_info, devid, 7382 dev_uuid, true); 7383 return -ENOENT; 7384 } 7385 7386 device = add_missing_dev(fs_devices, devid, dev_uuid); 7387 if (IS_ERR(device)) { 7388 btrfs_err(fs_info, 7389 "failed to add missing dev %llu: %ld", 7390 devid, PTR_ERR(device)); 7391 return PTR_ERR(device); 7392 } 7393 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7394 } else { 7395 if (!device->bdev) { 7396 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7397 btrfs_report_missing_device(fs_info, 7398 devid, dev_uuid, true); 7399 return -ENOENT; 7400 } 7401 btrfs_report_missing_device(fs_info, devid, 7402 dev_uuid, false); 7403 } 7404 7405 if (!device->bdev && 7406 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7407 /* 7408 * this happens when a device that was properly setup 7409 * in the device info lists suddenly goes bad. 7410 * device->bdev is NULL, and so we have to set 7411 * device->missing to one here 7412 */ 7413 device->fs_devices->missing_devices++; 7414 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7415 } 7416 7417 /* Move the device to its own fs_devices */ 7418 if (device->fs_devices != fs_devices) { 7419 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7420 &device->dev_state)); 7421 7422 list_move(&device->dev_list, &fs_devices->devices); 7423 device->fs_devices->num_devices--; 7424 fs_devices->num_devices++; 7425 7426 device->fs_devices->missing_devices--; 7427 fs_devices->missing_devices++; 7428 7429 device->fs_devices = fs_devices; 7430 } 7431 } 7432 7433 if (device->fs_devices != fs_info->fs_devices) { 7434 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7435 if (device->generation != 7436 btrfs_device_generation(leaf, dev_item)) 7437 return -EINVAL; 7438 } 7439 7440 fill_device_from_item(leaf, dev_item, device); 7441 if (device->bdev) { 7442 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7443 7444 if (device->total_bytes > max_total_bytes) { 7445 btrfs_err(fs_info, 7446 "device total_bytes should be at most %llu but found %llu", 7447 max_total_bytes, device->total_bytes); 7448 return -EINVAL; 7449 } 7450 } 7451 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7452 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7453 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7454 device->fs_devices->total_rw_bytes += device->total_bytes; 7455 atomic64_add(device->total_bytes - device->bytes_used, 7456 &fs_info->free_chunk_space); 7457 } 7458 ret = 0; 7459 return ret; 7460 } 7461 7462 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7463 { 7464 struct btrfs_super_block *super_copy = fs_info->super_copy; 7465 struct extent_buffer *sb; 7466 struct btrfs_disk_key *disk_key; 7467 struct btrfs_chunk *chunk; 7468 u8 *array_ptr; 7469 unsigned long sb_array_offset; 7470 int ret = 0; 7471 u32 num_stripes; 7472 u32 array_size; 7473 u32 len = 0; 7474 u32 cur_offset; 7475 u64 type; 7476 struct btrfs_key key; 7477 7478 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7479 7480 /* 7481 * We allocated a dummy extent, just to use extent buffer accessors. 7482 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but 7483 * that's fine, we will not go beyond system chunk array anyway. 7484 */ 7485 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); 7486 if (!sb) 7487 return -ENOMEM; 7488 set_extent_buffer_uptodate(sb); 7489 7490 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7491 array_size = btrfs_super_sys_array_size(super_copy); 7492 7493 array_ptr = super_copy->sys_chunk_array; 7494 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7495 cur_offset = 0; 7496 7497 while (cur_offset < array_size) { 7498 disk_key = (struct btrfs_disk_key *)array_ptr; 7499 len = sizeof(*disk_key); 7500 if (cur_offset + len > array_size) 7501 goto out_short_read; 7502 7503 btrfs_disk_key_to_cpu(&key, disk_key); 7504 7505 array_ptr += len; 7506 sb_array_offset += len; 7507 cur_offset += len; 7508 7509 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7510 btrfs_err(fs_info, 7511 "unexpected item type %u in sys_array at offset %u", 7512 (u32)key.type, cur_offset); 7513 ret = -EIO; 7514 break; 7515 } 7516 7517 chunk = (struct btrfs_chunk *)sb_array_offset; 7518 /* 7519 * At least one btrfs_chunk with one stripe must be present, 7520 * exact stripe count check comes afterwards 7521 */ 7522 len = btrfs_chunk_item_size(1); 7523 if (cur_offset + len > array_size) 7524 goto out_short_read; 7525 7526 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7527 if (!num_stripes) { 7528 btrfs_err(fs_info, 7529 "invalid number of stripes %u in sys_array at offset %u", 7530 num_stripes, cur_offset); 7531 ret = -EIO; 7532 break; 7533 } 7534 7535 type = btrfs_chunk_type(sb, chunk); 7536 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7537 btrfs_err(fs_info, 7538 "invalid chunk type %llu in sys_array at offset %u", 7539 type, cur_offset); 7540 ret = -EIO; 7541 break; 7542 } 7543 7544 len = btrfs_chunk_item_size(num_stripes); 7545 if (cur_offset + len > array_size) 7546 goto out_short_read; 7547 7548 ret = read_one_chunk(&key, sb, chunk); 7549 if (ret) 7550 break; 7551 7552 array_ptr += len; 7553 sb_array_offset += len; 7554 cur_offset += len; 7555 } 7556 clear_extent_buffer_uptodate(sb); 7557 free_extent_buffer_stale(sb); 7558 return ret; 7559 7560 out_short_read: 7561 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7562 len, cur_offset); 7563 clear_extent_buffer_uptodate(sb); 7564 free_extent_buffer_stale(sb); 7565 return -EIO; 7566 } 7567 7568 /* 7569 * Check if all chunks in the fs are OK for read-write degraded mount 7570 * 7571 * If the @failing_dev is specified, it's accounted as missing. 7572 * 7573 * Return true if all chunks meet the minimal RW mount requirements. 7574 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7575 */ 7576 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7577 struct btrfs_device *failing_dev) 7578 { 7579 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7580 struct extent_map *em; 7581 u64 next_start = 0; 7582 bool ret = true; 7583 7584 read_lock(&map_tree->lock); 7585 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7586 read_unlock(&map_tree->lock); 7587 /* No chunk at all? Return false anyway */ 7588 if (!em) { 7589 ret = false; 7590 goto out; 7591 } 7592 while (em) { 7593 struct map_lookup *map; 7594 int missing = 0; 7595 int max_tolerated; 7596 int i; 7597 7598 map = em->map_lookup; 7599 max_tolerated = 7600 btrfs_get_num_tolerated_disk_barrier_failures( 7601 map->type); 7602 for (i = 0; i < map->num_stripes; i++) { 7603 struct btrfs_device *dev = map->stripes[i].dev; 7604 7605 if (!dev || !dev->bdev || 7606 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7607 dev->last_flush_error) 7608 missing++; 7609 else if (failing_dev && failing_dev == dev) 7610 missing++; 7611 } 7612 if (missing > max_tolerated) { 7613 if (!failing_dev) 7614 btrfs_warn(fs_info, 7615 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7616 em->start, missing, max_tolerated); 7617 free_extent_map(em); 7618 ret = false; 7619 goto out; 7620 } 7621 next_start = extent_map_end(em); 7622 free_extent_map(em); 7623 7624 read_lock(&map_tree->lock); 7625 em = lookup_extent_mapping(map_tree, next_start, 7626 (u64)(-1) - next_start); 7627 read_unlock(&map_tree->lock); 7628 } 7629 out: 7630 return ret; 7631 } 7632 7633 static void readahead_tree_node_children(struct extent_buffer *node) 7634 { 7635 int i; 7636 const int nr_items = btrfs_header_nritems(node); 7637 7638 for (i = 0; i < nr_items; i++) 7639 btrfs_readahead_node_child(node, i); 7640 } 7641 7642 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7643 { 7644 struct btrfs_root *root = fs_info->chunk_root; 7645 struct btrfs_path *path; 7646 struct extent_buffer *leaf; 7647 struct btrfs_key key; 7648 struct btrfs_key found_key; 7649 int ret; 7650 int slot; 7651 int iter_ret = 0; 7652 u64 total_dev = 0; 7653 u64 last_ra_node = 0; 7654 7655 path = btrfs_alloc_path(); 7656 if (!path) 7657 return -ENOMEM; 7658 7659 /* 7660 * uuid_mutex is needed only if we are mounting a sprout FS 7661 * otherwise we don't need it. 7662 */ 7663 mutex_lock(&uuid_mutex); 7664 7665 /* 7666 * It is possible for mount and umount to race in such a way that 7667 * we execute this code path, but open_fs_devices failed to clear 7668 * total_rw_bytes. We certainly want it cleared before reading the 7669 * device items, so clear it here. 7670 */ 7671 fs_info->fs_devices->total_rw_bytes = 0; 7672 7673 /* 7674 * Lockdep complains about possible circular locking dependency between 7675 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7676 * used for freeze procection of a fs (struct super_block.s_writers), 7677 * which we take when starting a transaction, and extent buffers of the 7678 * chunk tree if we call read_one_dev() while holding a lock on an 7679 * extent buffer of the chunk tree. Since we are mounting the filesystem 7680 * and at this point there can't be any concurrent task modifying the 7681 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7682 */ 7683 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7684 path->skip_locking = 1; 7685 7686 /* 7687 * Read all device items, and then all the chunk items. All 7688 * device items are found before any chunk item (their object id 7689 * is smaller than the lowest possible object id for a chunk 7690 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7691 */ 7692 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7693 key.offset = 0; 7694 key.type = 0; 7695 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { 7696 struct extent_buffer *node = path->nodes[1]; 7697 7698 leaf = path->nodes[0]; 7699 slot = path->slots[0]; 7700 7701 if (node) { 7702 if (last_ra_node != node->start) { 7703 readahead_tree_node_children(node); 7704 last_ra_node = node->start; 7705 } 7706 } 7707 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7708 struct btrfs_dev_item *dev_item; 7709 dev_item = btrfs_item_ptr(leaf, slot, 7710 struct btrfs_dev_item); 7711 ret = read_one_dev(leaf, dev_item); 7712 if (ret) 7713 goto error; 7714 total_dev++; 7715 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7716 struct btrfs_chunk *chunk; 7717 7718 /* 7719 * We are only called at mount time, so no need to take 7720 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7721 * we always lock first fs_info->chunk_mutex before 7722 * acquiring any locks on the chunk tree. This is a 7723 * requirement for chunk allocation, see the comment on 7724 * top of btrfs_chunk_alloc() for details. 7725 */ 7726 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7727 ret = read_one_chunk(&found_key, leaf, chunk); 7728 if (ret) 7729 goto error; 7730 } 7731 } 7732 /* Catch error found during iteration */ 7733 if (iter_ret < 0) { 7734 ret = iter_ret; 7735 goto error; 7736 } 7737 7738 /* 7739 * After loading chunk tree, we've got all device information, 7740 * do another round of validation checks. 7741 */ 7742 if (total_dev != fs_info->fs_devices->total_devices) { 7743 btrfs_warn(fs_info, 7744 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", 7745 btrfs_super_num_devices(fs_info->super_copy), 7746 total_dev); 7747 fs_info->fs_devices->total_devices = total_dev; 7748 btrfs_set_super_num_devices(fs_info->super_copy, total_dev); 7749 } 7750 if (btrfs_super_total_bytes(fs_info->super_copy) < 7751 fs_info->fs_devices->total_rw_bytes) { 7752 btrfs_err(fs_info, 7753 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7754 btrfs_super_total_bytes(fs_info->super_copy), 7755 fs_info->fs_devices->total_rw_bytes); 7756 ret = -EINVAL; 7757 goto error; 7758 } 7759 ret = 0; 7760 error: 7761 mutex_unlock(&uuid_mutex); 7762 7763 btrfs_free_path(path); 7764 return ret; 7765 } 7766 7767 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7768 { 7769 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7770 struct btrfs_device *device; 7771 int ret = 0; 7772 7773 fs_devices->fs_info = fs_info; 7774 7775 mutex_lock(&fs_devices->device_list_mutex); 7776 list_for_each_entry(device, &fs_devices->devices, dev_list) 7777 device->fs_info = fs_info; 7778 7779 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7780 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7781 device->fs_info = fs_info; 7782 ret = btrfs_get_dev_zone_info(device, false); 7783 if (ret) 7784 break; 7785 } 7786 7787 seed_devs->fs_info = fs_info; 7788 } 7789 mutex_unlock(&fs_devices->device_list_mutex); 7790 7791 return ret; 7792 } 7793 7794 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7795 const struct btrfs_dev_stats_item *ptr, 7796 int index) 7797 { 7798 u64 val; 7799 7800 read_extent_buffer(eb, &val, 7801 offsetof(struct btrfs_dev_stats_item, values) + 7802 ((unsigned long)ptr) + (index * sizeof(u64)), 7803 sizeof(val)); 7804 return val; 7805 } 7806 7807 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7808 struct btrfs_dev_stats_item *ptr, 7809 int index, u64 val) 7810 { 7811 write_extent_buffer(eb, &val, 7812 offsetof(struct btrfs_dev_stats_item, values) + 7813 ((unsigned long)ptr) + (index * sizeof(u64)), 7814 sizeof(val)); 7815 } 7816 7817 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7818 struct btrfs_path *path) 7819 { 7820 struct btrfs_dev_stats_item *ptr; 7821 struct extent_buffer *eb; 7822 struct btrfs_key key; 7823 int item_size; 7824 int i, ret, slot; 7825 7826 if (!device->fs_info->dev_root) 7827 return 0; 7828 7829 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7830 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7831 key.offset = device->devid; 7832 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7833 if (ret) { 7834 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7835 btrfs_dev_stat_set(device, i, 0); 7836 device->dev_stats_valid = 1; 7837 btrfs_release_path(path); 7838 return ret < 0 ? ret : 0; 7839 } 7840 slot = path->slots[0]; 7841 eb = path->nodes[0]; 7842 item_size = btrfs_item_size(eb, slot); 7843 7844 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7845 7846 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7847 if (item_size >= (1 + i) * sizeof(__le64)) 7848 btrfs_dev_stat_set(device, i, 7849 btrfs_dev_stats_value(eb, ptr, i)); 7850 else 7851 btrfs_dev_stat_set(device, i, 0); 7852 } 7853 7854 device->dev_stats_valid = 1; 7855 btrfs_dev_stat_print_on_load(device); 7856 btrfs_release_path(path); 7857 7858 return 0; 7859 } 7860 7861 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7862 { 7863 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7864 struct btrfs_device *device; 7865 struct btrfs_path *path = NULL; 7866 int ret = 0; 7867 7868 path = btrfs_alloc_path(); 7869 if (!path) 7870 return -ENOMEM; 7871 7872 mutex_lock(&fs_devices->device_list_mutex); 7873 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7874 ret = btrfs_device_init_dev_stats(device, path); 7875 if (ret) 7876 goto out; 7877 } 7878 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7879 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7880 ret = btrfs_device_init_dev_stats(device, path); 7881 if (ret) 7882 goto out; 7883 } 7884 } 7885 out: 7886 mutex_unlock(&fs_devices->device_list_mutex); 7887 7888 btrfs_free_path(path); 7889 return ret; 7890 } 7891 7892 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7893 struct btrfs_device *device) 7894 { 7895 struct btrfs_fs_info *fs_info = trans->fs_info; 7896 struct btrfs_root *dev_root = fs_info->dev_root; 7897 struct btrfs_path *path; 7898 struct btrfs_key key; 7899 struct extent_buffer *eb; 7900 struct btrfs_dev_stats_item *ptr; 7901 int ret; 7902 int i; 7903 7904 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7905 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7906 key.offset = device->devid; 7907 7908 path = btrfs_alloc_path(); 7909 if (!path) 7910 return -ENOMEM; 7911 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7912 if (ret < 0) { 7913 btrfs_warn_in_rcu(fs_info, 7914 "error %d while searching for dev_stats item for device %s", 7915 ret, rcu_str_deref(device->name)); 7916 goto out; 7917 } 7918 7919 if (ret == 0 && 7920 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7921 /* need to delete old one and insert a new one */ 7922 ret = btrfs_del_item(trans, dev_root, path); 7923 if (ret != 0) { 7924 btrfs_warn_in_rcu(fs_info, 7925 "delete too small dev_stats item for device %s failed %d", 7926 rcu_str_deref(device->name), ret); 7927 goto out; 7928 } 7929 ret = 1; 7930 } 7931 7932 if (ret == 1) { 7933 /* need to insert a new item */ 7934 btrfs_release_path(path); 7935 ret = btrfs_insert_empty_item(trans, dev_root, path, 7936 &key, sizeof(*ptr)); 7937 if (ret < 0) { 7938 btrfs_warn_in_rcu(fs_info, 7939 "insert dev_stats item for device %s failed %d", 7940 rcu_str_deref(device->name), ret); 7941 goto out; 7942 } 7943 } 7944 7945 eb = path->nodes[0]; 7946 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7947 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7948 btrfs_set_dev_stats_value(eb, ptr, i, 7949 btrfs_dev_stat_read(device, i)); 7950 btrfs_mark_buffer_dirty(eb); 7951 7952 out: 7953 btrfs_free_path(path); 7954 return ret; 7955 } 7956 7957 /* 7958 * called from commit_transaction. Writes all changed device stats to disk. 7959 */ 7960 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7961 { 7962 struct btrfs_fs_info *fs_info = trans->fs_info; 7963 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7964 struct btrfs_device *device; 7965 int stats_cnt; 7966 int ret = 0; 7967 7968 mutex_lock(&fs_devices->device_list_mutex); 7969 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7970 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7971 if (!device->dev_stats_valid || stats_cnt == 0) 7972 continue; 7973 7974 7975 /* 7976 * There is a LOAD-LOAD control dependency between the value of 7977 * dev_stats_ccnt and updating the on-disk values which requires 7978 * reading the in-memory counters. Such control dependencies 7979 * require explicit read memory barriers. 7980 * 7981 * This memory barriers pairs with smp_mb__before_atomic in 7982 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7983 * barrier implied by atomic_xchg in 7984 * btrfs_dev_stats_read_and_reset 7985 */ 7986 smp_rmb(); 7987 7988 ret = update_dev_stat_item(trans, device); 7989 if (!ret) 7990 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7991 } 7992 mutex_unlock(&fs_devices->device_list_mutex); 7993 7994 return ret; 7995 } 7996 7997 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7998 { 7999 btrfs_dev_stat_inc(dev, index); 8000 8001 if (!dev->dev_stats_valid) 8002 return; 8003 btrfs_err_rl_in_rcu(dev->fs_info, 8004 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 8005 rcu_str_deref(dev->name), 8006 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 8007 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 8008 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 8009 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 8010 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 8011 } 8012 8013 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 8014 { 8015 int i; 8016 8017 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 8018 if (btrfs_dev_stat_read(dev, i) != 0) 8019 break; 8020 if (i == BTRFS_DEV_STAT_VALUES_MAX) 8021 return; /* all values == 0, suppress message */ 8022 8023 btrfs_info_in_rcu(dev->fs_info, 8024 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 8025 rcu_str_deref(dev->name), 8026 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 8027 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 8028 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 8029 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 8030 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 8031 } 8032 8033 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 8034 struct btrfs_ioctl_get_dev_stats *stats) 8035 { 8036 BTRFS_DEV_LOOKUP_ARGS(args); 8037 struct btrfs_device *dev; 8038 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 8039 int i; 8040 8041 mutex_lock(&fs_devices->device_list_mutex); 8042 args.devid = stats->devid; 8043 dev = btrfs_find_device(fs_info->fs_devices, &args); 8044 mutex_unlock(&fs_devices->device_list_mutex); 8045 8046 if (!dev) { 8047 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 8048 return -ENODEV; 8049 } else if (!dev->dev_stats_valid) { 8050 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 8051 return -ENODEV; 8052 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 8053 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 8054 if (stats->nr_items > i) 8055 stats->values[i] = 8056 btrfs_dev_stat_read_and_reset(dev, i); 8057 else 8058 btrfs_dev_stat_set(dev, i, 0); 8059 } 8060 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 8061 current->comm, task_pid_nr(current)); 8062 } else { 8063 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 8064 if (stats->nr_items > i) 8065 stats->values[i] = btrfs_dev_stat_read(dev, i); 8066 } 8067 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 8068 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 8069 return 0; 8070 } 8071 8072 /* 8073 * Update the size and bytes used for each device where it changed. This is 8074 * delayed since we would otherwise get errors while writing out the 8075 * superblocks. 8076 * 8077 * Must be invoked during transaction commit. 8078 */ 8079 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 8080 { 8081 struct btrfs_device *curr, *next; 8082 8083 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 8084 8085 if (list_empty(&trans->dev_update_list)) 8086 return; 8087 8088 /* 8089 * We don't need the device_list_mutex here. This list is owned by the 8090 * transaction and the transaction must complete before the device is 8091 * released. 8092 */ 8093 mutex_lock(&trans->fs_info->chunk_mutex); 8094 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 8095 post_commit_list) { 8096 list_del_init(&curr->post_commit_list); 8097 curr->commit_total_bytes = curr->disk_total_bytes; 8098 curr->commit_bytes_used = curr->bytes_used; 8099 } 8100 mutex_unlock(&trans->fs_info->chunk_mutex); 8101 } 8102 8103 /* 8104 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 8105 */ 8106 int btrfs_bg_type_to_factor(u64 flags) 8107 { 8108 const int index = btrfs_bg_flags_to_raid_index(flags); 8109 8110 return btrfs_raid_array[index].ncopies; 8111 } 8112 8113 8114 8115 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 8116 u64 chunk_offset, u64 devid, 8117 u64 physical_offset, u64 physical_len) 8118 { 8119 struct btrfs_dev_lookup_args args = { .devid = devid }; 8120 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8121 struct extent_map *em; 8122 struct map_lookup *map; 8123 struct btrfs_device *dev; 8124 u64 stripe_len; 8125 bool found = false; 8126 int ret = 0; 8127 int i; 8128 8129 read_lock(&em_tree->lock); 8130 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 8131 read_unlock(&em_tree->lock); 8132 8133 if (!em) { 8134 btrfs_err(fs_info, 8135 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 8136 physical_offset, devid); 8137 ret = -EUCLEAN; 8138 goto out; 8139 } 8140 8141 map = em->map_lookup; 8142 stripe_len = btrfs_calc_stripe_length(em); 8143 if (physical_len != stripe_len) { 8144 btrfs_err(fs_info, 8145 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 8146 physical_offset, devid, em->start, physical_len, 8147 stripe_len); 8148 ret = -EUCLEAN; 8149 goto out; 8150 } 8151 8152 /* 8153 * Very old mkfs.btrfs (before v4.1) will not respect the reserved 8154 * space. Although kernel can handle it without problem, better to warn 8155 * the users. 8156 */ 8157 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) 8158 btrfs_warn(fs_info, 8159 "devid %llu physical %llu len %llu inside the reserved space", 8160 devid, physical_offset, physical_len); 8161 8162 for (i = 0; i < map->num_stripes; i++) { 8163 if (map->stripes[i].dev->devid == devid && 8164 map->stripes[i].physical == physical_offset) { 8165 found = true; 8166 if (map->verified_stripes >= map->num_stripes) { 8167 btrfs_err(fs_info, 8168 "too many dev extents for chunk %llu found", 8169 em->start); 8170 ret = -EUCLEAN; 8171 goto out; 8172 } 8173 map->verified_stripes++; 8174 break; 8175 } 8176 } 8177 if (!found) { 8178 btrfs_err(fs_info, 8179 "dev extent physical offset %llu devid %llu has no corresponding chunk", 8180 physical_offset, devid); 8181 ret = -EUCLEAN; 8182 } 8183 8184 /* Make sure no dev extent is beyond device boundary */ 8185 dev = btrfs_find_device(fs_info->fs_devices, &args); 8186 if (!dev) { 8187 btrfs_err(fs_info, "failed to find devid %llu", devid); 8188 ret = -EUCLEAN; 8189 goto out; 8190 } 8191 8192 if (physical_offset + physical_len > dev->disk_total_bytes) { 8193 btrfs_err(fs_info, 8194 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 8195 devid, physical_offset, physical_len, 8196 dev->disk_total_bytes); 8197 ret = -EUCLEAN; 8198 goto out; 8199 } 8200 8201 if (dev->zone_info) { 8202 u64 zone_size = dev->zone_info->zone_size; 8203 8204 if (!IS_ALIGNED(physical_offset, zone_size) || 8205 !IS_ALIGNED(physical_len, zone_size)) { 8206 btrfs_err(fs_info, 8207 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8208 devid, physical_offset, physical_len); 8209 ret = -EUCLEAN; 8210 goto out; 8211 } 8212 } 8213 8214 out: 8215 free_extent_map(em); 8216 return ret; 8217 } 8218 8219 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8220 { 8221 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8222 struct extent_map *em; 8223 struct rb_node *node; 8224 int ret = 0; 8225 8226 read_lock(&em_tree->lock); 8227 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8228 em = rb_entry(node, struct extent_map, rb_node); 8229 if (em->map_lookup->num_stripes != 8230 em->map_lookup->verified_stripes) { 8231 btrfs_err(fs_info, 8232 "chunk %llu has missing dev extent, have %d expect %d", 8233 em->start, em->map_lookup->verified_stripes, 8234 em->map_lookup->num_stripes); 8235 ret = -EUCLEAN; 8236 goto out; 8237 } 8238 } 8239 out: 8240 read_unlock(&em_tree->lock); 8241 return ret; 8242 } 8243 8244 /* 8245 * Ensure that all dev extents are mapped to correct chunk, otherwise 8246 * later chunk allocation/free would cause unexpected behavior. 8247 * 8248 * NOTE: This will iterate through the whole device tree, which should be of 8249 * the same size level as the chunk tree. This slightly increases mount time. 8250 */ 8251 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8252 { 8253 struct btrfs_path *path; 8254 struct btrfs_root *root = fs_info->dev_root; 8255 struct btrfs_key key; 8256 u64 prev_devid = 0; 8257 u64 prev_dev_ext_end = 0; 8258 int ret = 0; 8259 8260 /* 8261 * We don't have a dev_root because we mounted with ignorebadroots and 8262 * failed to load the root, so we want to skip the verification in this 8263 * case for sure. 8264 * 8265 * However if the dev root is fine, but the tree itself is corrupted 8266 * we'd still fail to mount. This verification is only to make sure 8267 * writes can happen safely, so instead just bypass this check 8268 * completely in the case of IGNOREBADROOTS. 8269 */ 8270 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8271 return 0; 8272 8273 key.objectid = 1; 8274 key.type = BTRFS_DEV_EXTENT_KEY; 8275 key.offset = 0; 8276 8277 path = btrfs_alloc_path(); 8278 if (!path) 8279 return -ENOMEM; 8280 8281 path->reada = READA_FORWARD; 8282 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8283 if (ret < 0) 8284 goto out; 8285 8286 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8287 ret = btrfs_next_leaf(root, path); 8288 if (ret < 0) 8289 goto out; 8290 /* No dev extents at all? Not good */ 8291 if (ret > 0) { 8292 ret = -EUCLEAN; 8293 goto out; 8294 } 8295 } 8296 while (1) { 8297 struct extent_buffer *leaf = path->nodes[0]; 8298 struct btrfs_dev_extent *dext; 8299 int slot = path->slots[0]; 8300 u64 chunk_offset; 8301 u64 physical_offset; 8302 u64 physical_len; 8303 u64 devid; 8304 8305 btrfs_item_key_to_cpu(leaf, &key, slot); 8306 if (key.type != BTRFS_DEV_EXTENT_KEY) 8307 break; 8308 devid = key.objectid; 8309 physical_offset = key.offset; 8310 8311 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8312 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8313 physical_len = btrfs_dev_extent_length(leaf, dext); 8314 8315 /* Check if this dev extent overlaps with the previous one */ 8316 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8317 btrfs_err(fs_info, 8318 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8319 devid, physical_offset, prev_dev_ext_end); 8320 ret = -EUCLEAN; 8321 goto out; 8322 } 8323 8324 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8325 physical_offset, physical_len); 8326 if (ret < 0) 8327 goto out; 8328 prev_devid = devid; 8329 prev_dev_ext_end = physical_offset + physical_len; 8330 8331 ret = btrfs_next_item(root, path); 8332 if (ret < 0) 8333 goto out; 8334 if (ret > 0) { 8335 ret = 0; 8336 break; 8337 } 8338 } 8339 8340 /* Ensure all chunks have corresponding dev extents */ 8341 ret = verify_chunk_dev_extent_mapping(fs_info); 8342 out: 8343 btrfs_free_path(path); 8344 return ret; 8345 } 8346 8347 /* 8348 * Check whether the given block group or device is pinned by any inode being 8349 * used as a swapfile. 8350 */ 8351 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8352 { 8353 struct btrfs_swapfile_pin *sp; 8354 struct rb_node *node; 8355 8356 spin_lock(&fs_info->swapfile_pins_lock); 8357 node = fs_info->swapfile_pins.rb_node; 8358 while (node) { 8359 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8360 if (ptr < sp->ptr) 8361 node = node->rb_left; 8362 else if (ptr > sp->ptr) 8363 node = node->rb_right; 8364 else 8365 break; 8366 } 8367 spin_unlock(&fs_info->swapfile_pins_lock); 8368 return node != NULL; 8369 } 8370 8371 static int relocating_repair_kthread(void *data) 8372 { 8373 struct btrfs_block_group *cache = data; 8374 struct btrfs_fs_info *fs_info = cache->fs_info; 8375 u64 target; 8376 int ret = 0; 8377 8378 target = cache->start; 8379 btrfs_put_block_group(cache); 8380 8381 sb_start_write(fs_info->sb); 8382 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8383 btrfs_info(fs_info, 8384 "zoned: skip relocating block group %llu to repair: EBUSY", 8385 target); 8386 sb_end_write(fs_info->sb); 8387 return -EBUSY; 8388 } 8389 8390 mutex_lock(&fs_info->reclaim_bgs_lock); 8391 8392 /* Ensure block group still exists */ 8393 cache = btrfs_lookup_block_group(fs_info, target); 8394 if (!cache) 8395 goto out; 8396 8397 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) 8398 goto out; 8399 8400 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8401 if (ret < 0) 8402 goto out; 8403 8404 btrfs_info(fs_info, 8405 "zoned: relocating block group %llu to repair IO failure", 8406 target); 8407 ret = btrfs_relocate_chunk(fs_info, target); 8408 8409 out: 8410 if (cache) 8411 btrfs_put_block_group(cache); 8412 mutex_unlock(&fs_info->reclaim_bgs_lock); 8413 btrfs_exclop_finish(fs_info); 8414 sb_end_write(fs_info->sb); 8415 8416 return ret; 8417 } 8418 8419 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8420 { 8421 struct btrfs_block_group *cache; 8422 8423 if (!btrfs_is_zoned(fs_info)) 8424 return false; 8425 8426 /* Do not attempt to repair in degraded state */ 8427 if (btrfs_test_opt(fs_info, DEGRADED)) 8428 return true; 8429 8430 cache = btrfs_lookup_block_group(fs_info, logical); 8431 if (!cache) 8432 return true; 8433 8434 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { 8435 btrfs_put_block_group(cache); 8436 return true; 8437 } 8438 8439 kthread_run(relocating_repair_kthread, cache, 8440 "btrfs-relocating-repair"); 8441 8442 return true; 8443 } 8444 8445 int __init btrfs_bioset_init(void) 8446 { 8447 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, 8448 offsetof(struct btrfs_bio, bio), 8449 BIOSET_NEED_BVECS)) 8450 return -ENOMEM; 8451 return 0; 8452 } 8453 8454 void __cold btrfs_bioset_exit(void) 8455 { 8456 bioset_exit(&btrfs_bioset); 8457 } 8458