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 37 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ 38 BTRFS_BLOCK_GROUP_RAID10 | \ 39 BTRFS_BLOCK_GROUP_RAID56_MASK) 40 41 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { 42 [BTRFS_RAID_RAID10] = { 43 .sub_stripes = 2, 44 .dev_stripes = 1, 45 .devs_max = 0, /* 0 == as many as possible */ 46 .devs_min = 2, 47 .tolerated_failures = 1, 48 .devs_increment = 2, 49 .ncopies = 2, 50 .nparity = 0, 51 .raid_name = "raid10", 52 .bg_flag = BTRFS_BLOCK_GROUP_RAID10, 53 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, 54 }, 55 [BTRFS_RAID_RAID1] = { 56 .sub_stripes = 1, 57 .dev_stripes = 1, 58 .devs_max = 2, 59 .devs_min = 2, 60 .tolerated_failures = 1, 61 .devs_increment = 2, 62 .ncopies = 2, 63 .nparity = 0, 64 .raid_name = "raid1", 65 .bg_flag = BTRFS_BLOCK_GROUP_RAID1, 66 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, 67 }, 68 [BTRFS_RAID_RAID1C3] = { 69 .sub_stripes = 1, 70 .dev_stripes = 1, 71 .devs_max = 3, 72 .devs_min = 3, 73 .tolerated_failures = 2, 74 .devs_increment = 3, 75 .ncopies = 3, 76 .nparity = 0, 77 .raid_name = "raid1c3", 78 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, 79 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, 80 }, 81 [BTRFS_RAID_RAID1C4] = { 82 .sub_stripes = 1, 83 .dev_stripes = 1, 84 .devs_max = 4, 85 .devs_min = 4, 86 .tolerated_failures = 3, 87 .devs_increment = 4, 88 .ncopies = 4, 89 .nparity = 0, 90 .raid_name = "raid1c4", 91 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, 92 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, 93 }, 94 [BTRFS_RAID_DUP] = { 95 .sub_stripes = 1, 96 .dev_stripes = 2, 97 .devs_max = 1, 98 .devs_min = 1, 99 .tolerated_failures = 0, 100 .devs_increment = 1, 101 .ncopies = 2, 102 .nparity = 0, 103 .raid_name = "dup", 104 .bg_flag = BTRFS_BLOCK_GROUP_DUP, 105 .mindev_error = 0, 106 }, 107 [BTRFS_RAID_RAID0] = { 108 .sub_stripes = 1, 109 .dev_stripes = 1, 110 .devs_max = 0, 111 .devs_min = 1, 112 .tolerated_failures = 0, 113 .devs_increment = 1, 114 .ncopies = 1, 115 .nparity = 0, 116 .raid_name = "raid0", 117 .bg_flag = BTRFS_BLOCK_GROUP_RAID0, 118 .mindev_error = 0, 119 }, 120 [BTRFS_RAID_SINGLE] = { 121 .sub_stripes = 1, 122 .dev_stripes = 1, 123 .devs_max = 1, 124 .devs_min = 1, 125 .tolerated_failures = 0, 126 .devs_increment = 1, 127 .ncopies = 1, 128 .nparity = 0, 129 .raid_name = "single", 130 .bg_flag = 0, 131 .mindev_error = 0, 132 }, 133 [BTRFS_RAID_RAID5] = { 134 .sub_stripes = 1, 135 .dev_stripes = 1, 136 .devs_max = 0, 137 .devs_min = 2, 138 .tolerated_failures = 1, 139 .devs_increment = 1, 140 .ncopies = 1, 141 .nparity = 1, 142 .raid_name = "raid5", 143 .bg_flag = BTRFS_BLOCK_GROUP_RAID5, 144 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, 145 }, 146 [BTRFS_RAID_RAID6] = { 147 .sub_stripes = 1, 148 .dev_stripes = 1, 149 .devs_max = 0, 150 .devs_min = 3, 151 .tolerated_failures = 2, 152 .devs_increment = 1, 153 .ncopies = 1, 154 .nparity = 2, 155 .raid_name = "raid6", 156 .bg_flag = BTRFS_BLOCK_GROUP_RAID6, 157 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, 158 }, 159 }; 160 161 /* 162 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 163 * can be used as index to access btrfs_raid_array[]. 164 */ 165 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) 166 { 167 if (flags & BTRFS_BLOCK_GROUP_RAID10) 168 return BTRFS_RAID_RAID10; 169 else if (flags & BTRFS_BLOCK_GROUP_RAID1) 170 return BTRFS_RAID_RAID1; 171 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3) 172 return BTRFS_RAID_RAID1C3; 173 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4) 174 return BTRFS_RAID_RAID1C4; 175 else if (flags & BTRFS_BLOCK_GROUP_DUP) 176 return BTRFS_RAID_DUP; 177 else if (flags & BTRFS_BLOCK_GROUP_RAID0) 178 return BTRFS_RAID_RAID0; 179 else if (flags & BTRFS_BLOCK_GROUP_RAID5) 180 return BTRFS_RAID_RAID5; 181 else if (flags & BTRFS_BLOCK_GROUP_RAID6) 182 return BTRFS_RAID_RAID6; 183 184 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ 185 } 186 187 const char *btrfs_bg_type_to_raid_name(u64 flags) 188 { 189 const int index = btrfs_bg_flags_to_raid_index(flags); 190 191 if (index >= BTRFS_NR_RAID_TYPES) 192 return NULL; 193 194 return btrfs_raid_array[index].raid_name; 195 } 196 197 /* 198 * Fill @buf with textual description of @bg_flags, no more than @size_buf 199 * bytes including terminating null byte. 200 */ 201 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) 202 { 203 int i; 204 int ret; 205 char *bp = buf; 206 u64 flags = bg_flags; 207 u32 size_bp = size_buf; 208 209 if (!flags) { 210 strcpy(bp, "NONE"); 211 return; 212 } 213 214 #define DESCRIBE_FLAG(flag, desc) \ 215 do { \ 216 if (flags & (flag)) { \ 217 ret = snprintf(bp, size_bp, "%s|", (desc)); \ 218 if (ret < 0 || ret >= size_bp) \ 219 goto out_overflow; \ 220 size_bp -= ret; \ 221 bp += ret; \ 222 flags &= ~(flag); \ 223 } \ 224 } while (0) 225 226 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); 227 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); 228 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); 229 230 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); 231 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) 232 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, 233 btrfs_raid_array[i].raid_name); 234 #undef DESCRIBE_FLAG 235 236 if (flags) { 237 ret = snprintf(bp, size_bp, "0x%llx|", flags); 238 size_bp -= ret; 239 } 240 241 if (size_bp < size_buf) 242 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ 243 244 /* 245 * The text is trimmed, it's up to the caller to provide sufficiently 246 * large buffer 247 */ 248 out_overflow:; 249 } 250 251 static int init_first_rw_device(struct btrfs_trans_handle *trans); 252 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); 253 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); 254 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); 255 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 256 enum btrfs_map_op op, 257 u64 logical, u64 *length, 258 struct btrfs_io_context **bioc_ret, 259 int mirror_num, int need_raid_map); 260 261 /* 262 * Device locking 263 * ============== 264 * 265 * There are several mutexes that protect manipulation of devices and low-level 266 * structures like chunks but not block groups, extents or files 267 * 268 * uuid_mutex (global lock) 269 * ------------------------ 270 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from 271 * the SCAN_DEV ioctl registration or from mount either implicitly (the first 272 * device) or requested by the device= mount option 273 * 274 * the mutex can be very coarse and can cover long-running operations 275 * 276 * protects: updates to fs_devices counters like missing devices, rw devices, 277 * seeding, structure cloning, opening/closing devices at mount/umount time 278 * 279 * global::fs_devs - add, remove, updates to the global list 280 * 281 * does not protect: manipulation of the fs_devices::devices list in general 282 * but in mount context it could be used to exclude list modifications by eg. 283 * scan ioctl 284 * 285 * btrfs_device::name - renames (write side), read is RCU 286 * 287 * fs_devices::device_list_mutex (per-fs, with RCU) 288 * ------------------------------------------------ 289 * protects updates to fs_devices::devices, ie. adding and deleting 290 * 291 * simple list traversal with read-only actions can be done with RCU protection 292 * 293 * may be used to exclude some operations from running concurrently without any 294 * modifications to the list (see write_all_supers) 295 * 296 * Is not required at mount and close times, because our device list is 297 * protected by the uuid_mutex at that point. 298 * 299 * balance_mutex 300 * ------------- 301 * protects balance structures (status, state) and context accessed from 302 * several places (internally, ioctl) 303 * 304 * chunk_mutex 305 * ----------- 306 * protects chunks, adding or removing during allocation, trim or when a new 307 * device is added/removed. Additionally it also protects post_commit_list of 308 * individual devices, since they can be added to the transaction's 309 * post_commit_list only with chunk_mutex held. 310 * 311 * cleaner_mutex 312 * ------------- 313 * a big lock that is held by the cleaner thread and prevents running subvolume 314 * cleaning together with relocation or delayed iputs 315 * 316 * 317 * Lock nesting 318 * ============ 319 * 320 * uuid_mutex 321 * device_list_mutex 322 * chunk_mutex 323 * balance_mutex 324 * 325 * 326 * Exclusive operations 327 * ==================== 328 * 329 * Maintains the exclusivity of the following operations that apply to the 330 * whole filesystem and cannot run in parallel. 331 * 332 * - Balance (*) 333 * - Device add 334 * - Device remove 335 * - Device replace (*) 336 * - Resize 337 * 338 * The device operations (as above) can be in one of the following states: 339 * 340 * - Running state 341 * - Paused state 342 * - Completed state 343 * 344 * Only device operations marked with (*) can go into the Paused state for the 345 * following reasons: 346 * 347 * - ioctl (only Balance can be Paused through ioctl) 348 * - filesystem remounted as read-only 349 * - filesystem unmounted and mounted as read-only 350 * - system power-cycle and filesystem mounted as read-only 351 * - filesystem or device errors leading to forced read-only 352 * 353 * The status of exclusive operation is set and cleared atomically. 354 * During the course of Paused state, fs_info::exclusive_operation remains set. 355 * A device operation in Paused or Running state can be canceled or resumed 356 * either by ioctl (Balance only) or when remounted as read-write. 357 * The exclusive status is cleared when the device operation is canceled or 358 * completed. 359 */ 360 361 DEFINE_MUTEX(uuid_mutex); 362 static LIST_HEAD(fs_uuids); 363 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) 364 { 365 return &fs_uuids; 366 } 367 368 /* 369 * alloc_fs_devices - allocate struct btrfs_fs_devices 370 * @fsid: if not NULL, copy the UUID to fs_devices::fsid 371 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid 372 * 373 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). 374 * The returned struct is not linked onto any lists and can be destroyed with 375 * kfree() right away. 376 */ 377 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, 378 const u8 *metadata_fsid) 379 { 380 struct btrfs_fs_devices *fs_devs; 381 382 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 383 if (!fs_devs) 384 return ERR_PTR(-ENOMEM); 385 386 mutex_init(&fs_devs->device_list_mutex); 387 388 INIT_LIST_HEAD(&fs_devs->devices); 389 INIT_LIST_HEAD(&fs_devs->alloc_list); 390 INIT_LIST_HEAD(&fs_devs->fs_list); 391 INIT_LIST_HEAD(&fs_devs->seed_list); 392 if (fsid) 393 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 394 395 if (metadata_fsid) 396 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); 397 else if (fsid) 398 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); 399 400 return fs_devs; 401 } 402 403 void btrfs_free_device(struct btrfs_device *device) 404 { 405 WARN_ON(!list_empty(&device->post_commit_list)); 406 rcu_string_free(device->name); 407 extent_io_tree_release(&device->alloc_state); 408 btrfs_destroy_dev_zone_info(device); 409 kfree(device); 410 } 411 412 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 413 { 414 struct btrfs_device *device; 415 WARN_ON(fs_devices->opened); 416 while (!list_empty(&fs_devices->devices)) { 417 device = list_entry(fs_devices->devices.next, 418 struct btrfs_device, dev_list); 419 list_del(&device->dev_list); 420 btrfs_free_device(device); 421 } 422 kfree(fs_devices); 423 } 424 425 void __exit btrfs_cleanup_fs_uuids(void) 426 { 427 struct btrfs_fs_devices *fs_devices; 428 429 while (!list_empty(&fs_uuids)) { 430 fs_devices = list_entry(fs_uuids.next, 431 struct btrfs_fs_devices, fs_list); 432 list_del(&fs_devices->fs_list); 433 free_fs_devices(fs_devices); 434 } 435 } 436 437 static noinline struct btrfs_fs_devices *find_fsid( 438 const u8 *fsid, const u8 *metadata_fsid) 439 { 440 struct btrfs_fs_devices *fs_devices; 441 442 ASSERT(fsid); 443 444 /* Handle non-split brain cases */ 445 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 446 if (metadata_fsid) { 447 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 448 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 449 BTRFS_FSID_SIZE) == 0) 450 return fs_devices; 451 } else { 452 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 453 return fs_devices; 454 } 455 } 456 return NULL; 457 } 458 459 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 460 struct btrfs_super_block *disk_super) 461 { 462 463 struct btrfs_fs_devices *fs_devices; 464 465 /* 466 * Handle scanned device having completed its fsid change but 467 * belonging to a fs_devices that was created by first scanning 468 * a device which didn't have its fsid/metadata_uuid changed 469 * at all and the CHANGING_FSID_V2 flag set. 470 */ 471 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 472 if (fs_devices->fsid_change && 473 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 474 BTRFS_FSID_SIZE) == 0 && 475 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 476 BTRFS_FSID_SIZE) == 0) { 477 return fs_devices; 478 } 479 } 480 /* 481 * Handle scanned device having completed its fsid change but 482 * belonging to a fs_devices that was created by a device that 483 * has an outdated pair of fsid/metadata_uuid and 484 * CHANGING_FSID_V2 flag set. 485 */ 486 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 487 if (fs_devices->fsid_change && 488 memcmp(fs_devices->metadata_uuid, 489 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 490 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 491 BTRFS_FSID_SIZE) == 0) { 492 return fs_devices; 493 } 494 } 495 496 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 497 } 498 499 500 static int 501 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 502 int flush, struct block_device **bdev, 503 struct btrfs_super_block **disk_super) 504 { 505 int ret; 506 507 *bdev = blkdev_get_by_path(device_path, flags, holder); 508 509 if (IS_ERR(*bdev)) { 510 ret = PTR_ERR(*bdev); 511 goto error; 512 } 513 514 if (flush) 515 sync_blockdev(*bdev); 516 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 517 if (ret) { 518 blkdev_put(*bdev, flags); 519 goto error; 520 } 521 invalidate_bdev(*bdev); 522 *disk_super = btrfs_read_dev_super(*bdev); 523 if (IS_ERR(*disk_super)) { 524 ret = PTR_ERR(*disk_super); 525 blkdev_put(*bdev, flags); 526 goto error; 527 } 528 529 return 0; 530 531 error: 532 *bdev = NULL; 533 return ret; 534 } 535 536 /** 537 * Search and remove all stale devices (which are not mounted). 538 * When both inputs are NULL, it will search and release all stale devices. 539 * 540 * @devt: Optional. When provided will it release all unmounted devices 541 * matching this devt only. 542 * @skip_device: Optional. Will skip this device when searching for the stale 543 * devices. 544 * 545 * Return: 0 for success or if @devt is 0. 546 * -EBUSY if @devt is a mounted device. 547 * -ENOENT if @devt does not match any device in the list. 548 */ 549 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) 550 { 551 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 552 struct btrfs_device *device, *tmp_device; 553 int ret = 0; 554 555 lockdep_assert_held(&uuid_mutex); 556 557 if (devt) 558 ret = -ENOENT; 559 560 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 561 562 mutex_lock(&fs_devices->device_list_mutex); 563 list_for_each_entry_safe(device, tmp_device, 564 &fs_devices->devices, dev_list) { 565 if (skip_device && skip_device == device) 566 continue; 567 if (devt && devt != device->devt) 568 continue; 569 if (fs_devices->opened) { 570 /* for an already deleted device return 0 */ 571 if (devt && ret != 0) 572 ret = -EBUSY; 573 break; 574 } 575 576 /* delete the stale device */ 577 fs_devices->num_devices--; 578 list_del(&device->dev_list); 579 btrfs_free_device(device); 580 581 ret = 0; 582 } 583 mutex_unlock(&fs_devices->device_list_mutex); 584 585 if (fs_devices->num_devices == 0) { 586 btrfs_sysfs_remove_fsid(fs_devices); 587 list_del(&fs_devices->fs_list); 588 free_fs_devices(fs_devices); 589 } 590 } 591 592 return ret; 593 } 594 595 /* 596 * This is only used on mount, and we are protected from competing things 597 * messing with our fs_devices by the uuid_mutex, thus we do not need the 598 * fs_devices->device_list_mutex here. 599 */ 600 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 601 struct btrfs_device *device, fmode_t flags, 602 void *holder) 603 { 604 struct block_device *bdev; 605 struct btrfs_super_block *disk_super; 606 u64 devid; 607 int ret; 608 609 if (device->bdev) 610 return -EINVAL; 611 if (!device->name) 612 return -EINVAL; 613 614 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 615 &bdev, &disk_super); 616 if (ret) 617 return ret; 618 619 devid = btrfs_stack_device_id(&disk_super->dev_item); 620 if (devid != device->devid) 621 goto error_free_page; 622 623 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 624 goto error_free_page; 625 626 device->generation = btrfs_super_generation(disk_super); 627 628 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 629 if (btrfs_super_incompat_flags(disk_super) & 630 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 631 pr_err( 632 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 633 goto error_free_page; 634 } 635 636 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 637 fs_devices->seeding = true; 638 } else { 639 if (bdev_read_only(bdev)) 640 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 641 else 642 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 643 } 644 645 if (!bdev_nonrot(bdev)) 646 fs_devices->rotating = true; 647 648 device->bdev = bdev; 649 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 650 device->mode = flags; 651 652 fs_devices->open_devices++; 653 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 654 device->devid != BTRFS_DEV_REPLACE_DEVID) { 655 fs_devices->rw_devices++; 656 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 657 } 658 btrfs_release_disk_super(disk_super); 659 660 return 0; 661 662 error_free_page: 663 btrfs_release_disk_super(disk_super); 664 blkdev_put(bdev, flags); 665 666 return -EINVAL; 667 } 668 669 /* 670 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 671 * being created with a disk that has already completed its fsid change. Such 672 * disk can belong to an fs which has its FSID changed or to one which doesn't. 673 * Handle both cases here. 674 */ 675 static struct btrfs_fs_devices *find_fsid_inprogress( 676 struct btrfs_super_block *disk_super) 677 { 678 struct btrfs_fs_devices *fs_devices; 679 680 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 681 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 682 BTRFS_FSID_SIZE) != 0 && 683 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 684 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 685 return fs_devices; 686 } 687 } 688 689 return find_fsid(disk_super->fsid, NULL); 690 } 691 692 693 static struct btrfs_fs_devices *find_fsid_changed( 694 struct btrfs_super_block *disk_super) 695 { 696 struct btrfs_fs_devices *fs_devices; 697 698 /* 699 * Handles the case where scanned device is part of an fs that had 700 * multiple successful changes of FSID but currently device didn't 701 * observe it. Meaning our fsid will be different than theirs. We need 702 * to handle two subcases : 703 * 1 - The fs still continues to have different METADATA/FSID uuids. 704 * 2 - The fs is switched back to its original FSID (METADATA/FSID 705 * are equal). 706 */ 707 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 708 /* Changed UUIDs */ 709 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 710 BTRFS_FSID_SIZE) != 0 && 711 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 712 BTRFS_FSID_SIZE) == 0 && 713 memcmp(fs_devices->fsid, disk_super->fsid, 714 BTRFS_FSID_SIZE) != 0) 715 return fs_devices; 716 717 /* Unchanged UUIDs */ 718 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 719 BTRFS_FSID_SIZE) == 0 && 720 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 721 BTRFS_FSID_SIZE) == 0) 722 return fs_devices; 723 } 724 725 return NULL; 726 } 727 728 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 729 struct btrfs_super_block *disk_super) 730 { 731 struct btrfs_fs_devices *fs_devices; 732 733 /* 734 * Handle the case where the scanned device is part of an fs whose last 735 * metadata UUID change reverted it to the original FSID. At the same 736 * time * fs_devices was first created by another constitutent device 737 * which didn't fully observe the operation. This results in an 738 * btrfs_fs_devices created with metadata/fsid different AND 739 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 740 * fs_devices equal to the FSID of the disk. 741 */ 742 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 743 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 744 BTRFS_FSID_SIZE) != 0 && 745 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 746 BTRFS_FSID_SIZE) == 0 && 747 fs_devices->fsid_change) 748 return fs_devices; 749 } 750 751 return NULL; 752 } 753 /* 754 * Add new device to list of registered devices 755 * 756 * Returns: 757 * device pointer which was just added or updated when successful 758 * error pointer when failed 759 */ 760 static noinline struct btrfs_device *device_list_add(const char *path, 761 struct btrfs_super_block *disk_super, 762 bool *new_device_added) 763 { 764 struct btrfs_device *device; 765 struct btrfs_fs_devices *fs_devices = NULL; 766 struct rcu_string *name; 767 u64 found_transid = btrfs_super_generation(disk_super); 768 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 769 dev_t path_devt; 770 int error; 771 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 772 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 773 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 774 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 775 776 error = lookup_bdev(path, &path_devt); 777 if (error) 778 return ERR_PTR(error); 779 780 if (fsid_change_in_progress) { 781 if (!has_metadata_uuid) 782 fs_devices = find_fsid_inprogress(disk_super); 783 else 784 fs_devices = find_fsid_changed(disk_super); 785 } else if (has_metadata_uuid) { 786 fs_devices = find_fsid_with_metadata_uuid(disk_super); 787 } else { 788 fs_devices = find_fsid_reverted_metadata(disk_super); 789 if (!fs_devices) 790 fs_devices = find_fsid(disk_super->fsid, NULL); 791 } 792 793 794 if (!fs_devices) { 795 if (has_metadata_uuid) 796 fs_devices = alloc_fs_devices(disk_super->fsid, 797 disk_super->metadata_uuid); 798 else 799 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 800 801 if (IS_ERR(fs_devices)) 802 return ERR_CAST(fs_devices); 803 804 fs_devices->fsid_change = fsid_change_in_progress; 805 806 mutex_lock(&fs_devices->device_list_mutex); 807 list_add(&fs_devices->fs_list, &fs_uuids); 808 809 device = NULL; 810 } else { 811 struct btrfs_dev_lookup_args args = { 812 .devid = devid, 813 .uuid = disk_super->dev_item.uuid, 814 }; 815 816 mutex_lock(&fs_devices->device_list_mutex); 817 device = btrfs_find_device(fs_devices, &args); 818 819 /* 820 * If this disk has been pulled into an fs devices created by 821 * a device which had the CHANGING_FSID_V2 flag then replace the 822 * metadata_uuid/fsid values of the fs_devices. 823 */ 824 if (fs_devices->fsid_change && 825 found_transid > fs_devices->latest_generation) { 826 memcpy(fs_devices->fsid, disk_super->fsid, 827 BTRFS_FSID_SIZE); 828 829 if (has_metadata_uuid) 830 memcpy(fs_devices->metadata_uuid, 831 disk_super->metadata_uuid, 832 BTRFS_FSID_SIZE); 833 else 834 memcpy(fs_devices->metadata_uuid, 835 disk_super->fsid, BTRFS_FSID_SIZE); 836 837 fs_devices->fsid_change = false; 838 } 839 } 840 841 if (!device) { 842 if (fs_devices->opened) { 843 mutex_unlock(&fs_devices->device_list_mutex); 844 return ERR_PTR(-EBUSY); 845 } 846 847 device = btrfs_alloc_device(NULL, &devid, 848 disk_super->dev_item.uuid); 849 if (IS_ERR(device)) { 850 mutex_unlock(&fs_devices->device_list_mutex); 851 /* we can safely leave the fs_devices entry around */ 852 return device; 853 } 854 855 name = rcu_string_strdup(path, GFP_NOFS); 856 if (!name) { 857 btrfs_free_device(device); 858 mutex_unlock(&fs_devices->device_list_mutex); 859 return ERR_PTR(-ENOMEM); 860 } 861 rcu_assign_pointer(device->name, name); 862 device->devt = path_devt; 863 864 list_add_rcu(&device->dev_list, &fs_devices->devices); 865 fs_devices->num_devices++; 866 867 device->fs_devices = fs_devices; 868 *new_device_added = true; 869 870 if (disk_super->label[0]) 871 pr_info( 872 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 873 disk_super->label, devid, found_transid, path, 874 current->comm, task_pid_nr(current)); 875 else 876 pr_info( 877 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 878 disk_super->fsid, devid, found_transid, path, 879 current->comm, task_pid_nr(current)); 880 881 } else if (!device->name || strcmp(device->name->str, path)) { 882 /* 883 * When FS is already mounted. 884 * 1. If you are here and if the device->name is NULL that 885 * means this device was missing at time of FS mount. 886 * 2. If you are here and if the device->name is different 887 * from 'path' that means either 888 * a. The same device disappeared and reappeared with 889 * different name. or 890 * b. The missing-disk-which-was-replaced, has 891 * reappeared now. 892 * 893 * We must allow 1 and 2a above. But 2b would be a spurious 894 * and unintentional. 895 * 896 * Further in case of 1 and 2a above, the disk at 'path' 897 * would have missed some transaction when it was away and 898 * in case of 2a the stale bdev has to be updated as well. 899 * 2b must not be allowed at all time. 900 */ 901 902 /* 903 * For now, we do allow update to btrfs_fs_device through the 904 * btrfs dev scan cli after FS has been mounted. We're still 905 * tracking a problem where systems fail mount by subvolume id 906 * when we reject replacement on a mounted FS. 907 */ 908 if (!fs_devices->opened && found_transid < device->generation) { 909 /* 910 * That is if the FS is _not_ mounted and if you 911 * are here, that means there is more than one 912 * disk with same uuid and devid.We keep the one 913 * with larger generation number or the last-in if 914 * generation are equal. 915 */ 916 mutex_unlock(&fs_devices->device_list_mutex); 917 return ERR_PTR(-EEXIST); 918 } 919 920 /* 921 * We are going to replace the device path for a given devid, 922 * make sure it's the same device if the device is mounted 923 * 924 * NOTE: the device->fs_info may not be reliable here so pass 925 * in a NULL to message helpers instead. This avoids a possible 926 * use-after-free when the fs_info and fs_info->sb are already 927 * torn down. 928 */ 929 if (device->bdev) { 930 if (device->devt != path_devt) { 931 mutex_unlock(&fs_devices->device_list_mutex); 932 btrfs_warn_in_rcu(NULL, 933 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 934 path, devid, found_transid, 935 current->comm, 936 task_pid_nr(current)); 937 return ERR_PTR(-EEXIST); 938 } 939 btrfs_info_in_rcu(NULL, 940 "devid %llu device path %s changed to %s scanned by %s (%d)", 941 devid, rcu_str_deref(device->name), 942 path, current->comm, 943 task_pid_nr(current)); 944 } 945 946 name = rcu_string_strdup(path, GFP_NOFS); 947 if (!name) { 948 mutex_unlock(&fs_devices->device_list_mutex); 949 return ERR_PTR(-ENOMEM); 950 } 951 rcu_string_free(device->name); 952 rcu_assign_pointer(device->name, name); 953 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 954 fs_devices->missing_devices--; 955 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 956 } 957 device->devt = path_devt; 958 } 959 960 /* 961 * Unmount does not free the btrfs_device struct but would zero 962 * generation along with most of the other members. So just update 963 * it back. We need it to pick the disk with largest generation 964 * (as above). 965 */ 966 if (!fs_devices->opened) { 967 device->generation = found_transid; 968 fs_devices->latest_generation = max_t(u64, found_transid, 969 fs_devices->latest_generation); 970 } 971 972 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 973 974 mutex_unlock(&fs_devices->device_list_mutex); 975 return device; 976 } 977 978 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 979 { 980 struct btrfs_fs_devices *fs_devices; 981 struct btrfs_device *device; 982 struct btrfs_device *orig_dev; 983 int ret = 0; 984 985 lockdep_assert_held(&uuid_mutex); 986 987 fs_devices = alloc_fs_devices(orig->fsid, NULL); 988 if (IS_ERR(fs_devices)) 989 return fs_devices; 990 991 fs_devices->total_devices = orig->total_devices; 992 993 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 994 struct rcu_string *name; 995 996 device = btrfs_alloc_device(NULL, &orig_dev->devid, 997 orig_dev->uuid); 998 if (IS_ERR(device)) { 999 ret = PTR_ERR(device); 1000 goto error; 1001 } 1002 1003 /* 1004 * This is ok to do without rcu read locked because we hold the 1005 * uuid mutex so nothing we touch in here is going to disappear. 1006 */ 1007 if (orig_dev->name) { 1008 name = rcu_string_strdup(orig_dev->name->str, 1009 GFP_KERNEL); 1010 if (!name) { 1011 btrfs_free_device(device); 1012 ret = -ENOMEM; 1013 goto error; 1014 } 1015 rcu_assign_pointer(device->name, name); 1016 } 1017 1018 list_add(&device->dev_list, &fs_devices->devices); 1019 device->fs_devices = fs_devices; 1020 fs_devices->num_devices++; 1021 } 1022 return fs_devices; 1023 error: 1024 free_fs_devices(fs_devices); 1025 return ERR_PTR(ret); 1026 } 1027 1028 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1029 struct btrfs_device **latest_dev) 1030 { 1031 struct btrfs_device *device, *next; 1032 1033 /* This is the initialized path, it is safe to release the devices. */ 1034 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1035 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1036 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1037 &device->dev_state) && 1038 !test_bit(BTRFS_DEV_STATE_MISSING, 1039 &device->dev_state) && 1040 (!*latest_dev || 1041 device->generation > (*latest_dev)->generation)) { 1042 *latest_dev = device; 1043 } 1044 continue; 1045 } 1046 1047 /* 1048 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1049 * in btrfs_init_dev_replace() so just continue. 1050 */ 1051 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1052 continue; 1053 1054 if (device->bdev) { 1055 blkdev_put(device->bdev, device->mode); 1056 device->bdev = NULL; 1057 fs_devices->open_devices--; 1058 } 1059 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1060 list_del_init(&device->dev_alloc_list); 1061 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1062 fs_devices->rw_devices--; 1063 } 1064 list_del_init(&device->dev_list); 1065 fs_devices->num_devices--; 1066 btrfs_free_device(device); 1067 } 1068 1069 } 1070 1071 /* 1072 * After we have read the system tree and know devids belonging to this 1073 * filesystem, remove the device which does not belong there. 1074 */ 1075 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1076 { 1077 struct btrfs_device *latest_dev = NULL; 1078 struct btrfs_fs_devices *seed_dev; 1079 1080 mutex_lock(&uuid_mutex); 1081 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1082 1083 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1084 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1085 1086 fs_devices->latest_dev = latest_dev; 1087 1088 mutex_unlock(&uuid_mutex); 1089 } 1090 1091 static void btrfs_close_bdev(struct btrfs_device *device) 1092 { 1093 if (!device->bdev) 1094 return; 1095 1096 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1097 sync_blockdev(device->bdev); 1098 invalidate_bdev(device->bdev); 1099 } 1100 1101 blkdev_put(device->bdev, device->mode); 1102 } 1103 1104 static void btrfs_close_one_device(struct btrfs_device *device) 1105 { 1106 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1107 1108 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1109 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1110 list_del_init(&device->dev_alloc_list); 1111 fs_devices->rw_devices--; 1112 } 1113 1114 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1115 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1116 1117 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 1118 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 1119 fs_devices->missing_devices--; 1120 } 1121 1122 btrfs_close_bdev(device); 1123 if (device->bdev) { 1124 fs_devices->open_devices--; 1125 device->bdev = NULL; 1126 } 1127 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1128 btrfs_destroy_dev_zone_info(device); 1129 1130 device->fs_info = NULL; 1131 atomic_set(&device->dev_stats_ccnt, 0); 1132 extent_io_tree_release(&device->alloc_state); 1133 1134 /* 1135 * Reset the flush error record. We might have a transient flush error 1136 * in this mount, and if so we aborted the current transaction and set 1137 * the fs to an error state, guaranteeing no super blocks can be further 1138 * committed. However that error might be transient and if we unmount the 1139 * filesystem and mount it again, we should allow the mount to succeed 1140 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1141 * filesystem again we still get flush errors, then we will again abort 1142 * any transaction and set the error state, guaranteeing no commits of 1143 * unsafe super blocks. 1144 */ 1145 device->last_flush_error = 0; 1146 1147 /* Verify the device is back in a pristine state */ 1148 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1149 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1150 ASSERT(list_empty(&device->dev_alloc_list)); 1151 ASSERT(list_empty(&device->post_commit_list)); 1152 } 1153 1154 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1155 { 1156 struct btrfs_device *device, *tmp; 1157 1158 lockdep_assert_held(&uuid_mutex); 1159 1160 if (--fs_devices->opened > 0) 1161 return; 1162 1163 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1164 btrfs_close_one_device(device); 1165 1166 WARN_ON(fs_devices->open_devices); 1167 WARN_ON(fs_devices->rw_devices); 1168 fs_devices->opened = 0; 1169 fs_devices->seeding = false; 1170 fs_devices->fs_info = NULL; 1171 } 1172 1173 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1174 { 1175 LIST_HEAD(list); 1176 struct btrfs_fs_devices *tmp; 1177 1178 mutex_lock(&uuid_mutex); 1179 close_fs_devices(fs_devices); 1180 if (!fs_devices->opened) 1181 list_splice_init(&fs_devices->seed_list, &list); 1182 1183 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1184 close_fs_devices(fs_devices); 1185 list_del(&fs_devices->seed_list); 1186 free_fs_devices(fs_devices); 1187 } 1188 mutex_unlock(&uuid_mutex); 1189 } 1190 1191 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1192 fmode_t flags, void *holder) 1193 { 1194 struct btrfs_device *device; 1195 struct btrfs_device *latest_dev = NULL; 1196 struct btrfs_device *tmp_device; 1197 1198 flags |= FMODE_EXCL; 1199 1200 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1201 dev_list) { 1202 int ret; 1203 1204 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1205 if (ret == 0 && 1206 (!latest_dev || device->generation > latest_dev->generation)) { 1207 latest_dev = device; 1208 } else if (ret == -ENODATA) { 1209 fs_devices->num_devices--; 1210 list_del(&device->dev_list); 1211 btrfs_free_device(device); 1212 } 1213 } 1214 if (fs_devices->open_devices == 0) 1215 return -EINVAL; 1216 1217 fs_devices->opened = 1; 1218 fs_devices->latest_dev = latest_dev; 1219 fs_devices->total_rw_bytes = 0; 1220 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1221 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1222 1223 return 0; 1224 } 1225 1226 static int devid_cmp(void *priv, const struct list_head *a, 1227 const struct list_head *b) 1228 { 1229 const struct btrfs_device *dev1, *dev2; 1230 1231 dev1 = list_entry(a, struct btrfs_device, dev_list); 1232 dev2 = list_entry(b, struct btrfs_device, dev_list); 1233 1234 if (dev1->devid < dev2->devid) 1235 return -1; 1236 else if (dev1->devid > dev2->devid) 1237 return 1; 1238 return 0; 1239 } 1240 1241 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1242 fmode_t flags, void *holder) 1243 { 1244 int ret; 1245 1246 lockdep_assert_held(&uuid_mutex); 1247 /* 1248 * The device_list_mutex cannot be taken here in case opening the 1249 * underlying device takes further locks like open_mutex. 1250 * 1251 * We also don't need the lock here as this is called during mount and 1252 * exclusion is provided by uuid_mutex 1253 */ 1254 1255 if (fs_devices->opened) { 1256 fs_devices->opened++; 1257 ret = 0; 1258 } else { 1259 list_sort(NULL, &fs_devices->devices, devid_cmp); 1260 ret = open_fs_devices(fs_devices, flags, holder); 1261 } 1262 1263 return ret; 1264 } 1265 1266 void btrfs_release_disk_super(struct btrfs_super_block *super) 1267 { 1268 struct page *page = virt_to_page(super); 1269 1270 put_page(page); 1271 } 1272 1273 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1274 u64 bytenr, u64 bytenr_orig) 1275 { 1276 struct btrfs_super_block *disk_super; 1277 struct page *page; 1278 void *p; 1279 pgoff_t index; 1280 1281 /* make sure our super fits in the device */ 1282 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) 1283 return ERR_PTR(-EINVAL); 1284 1285 /* make sure our super fits in the page */ 1286 if (sizeof(*disk_super) > PAGE_SIZE) 1287 return ERR_PTR(-EINVAL); 1288 1289 /* make sure our super doesn't straddle pages on disk */ 1290 index = bytenr >> PAGE_SHIFT; 1291 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1292 return ERR_PTR(-EINVAL); 1293 1294 /* pull in the page with our super */ 1295 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1296 1297 if (IS_ERR(page)) 1298 return ERR_CAST(page); 1299 1300 p = page_address(page); 1301 1302 /* align our pointer to the offset of the super block */ 1303 disk_super = p + offset_in_page(bytenr); 1304 1305 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1306 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1307 btrfs_release_disk_super(p); 1308 return ERR_PTR(-EINVAL); 1309 } 1310 1311 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1312 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1313 1314 return disk_super; 1315 } 1316 1317 int btrfs_forget_devices(dev_t devt) 1318 { 1319 int ret; 1320 1321 mutex_lock(&uuid_mutex); 1322 ret = btrfs_free_stale_devices(devt, NULL); 1323 mutex_unlock(&uuid_mutex); 1324 1325 return ret; 1326 } 1327 1328 /* 1329 * Look for a btrfs signature on a device. This may be called out of the mount path 1330 * and we are not allowed to call set_blocksize during the scan. The superblock 1331 * is read via pagecache 1332 */ 1333 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1334 void *holder) 1335 { 1336 struct btrfs_super_block *disk_super; 1337 bool new_device_added = false; 1338 struct btrfs_device *device = NULL; 1339 struct block_device *bdev; 1340 u64 bytenr, bytenr_orig; 1341 int ret; 1342 1343 lockdep_assert_held(&uuid_mutex); 1344 1345 /* 1346 * we would like to check all the supers, but that would make 1347 * a btrfs mount succeed after a mkfs from a different FS. 1348 * So, we need to add a special mount option to scan for 1349 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1350 */ 1351 flags |= FMODE_EXCL; 1352 1353 bdev = blkdev_get_by_path(path, flags, holder); 1354 if (IS_ERR(bdev)) 1355 return ERR_CAST(bdev); 1356 1357 bytenr_orig = btrfs_sb_offset(0); 1358 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1359 if (ret) { 1360 device = ERR_PTR(ret); 1361 goto error_bdev_put; 1362 } 1363 1364 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1365 if (IS_ERR(disk_super)) { 1366 device = ERR_CAST(disk_super); 1367 goto error_bdev_put; 1368 } 1369 1370 device = device_list_add(path, disk_super, &new_device_added); 1371 if (!IS_ERR(device) && new_device_added) 1372 btrfs_free_stale_devices(device->devt, device); 1373 1374 btrfs_release_disk_super(disk_super); 1375 1376 error_bdev_put: 1377 blkdev_put(bdev, flags); 1378 1379 return device; 1380 } 1381 1382 /* 1383 * Try to find a chunk that intersects [start, start + len] range and when one 1384 * such is found, record the end of it in *start 1385 */ 1386 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1387 u64 len) 1388 { 1389 u64 physical_start, physical_end; 1390 1391 lockdep_assert_held(&device->fs_info->chunk_mutex); 1392 1393 if (!find_first_extent_bit(&device->alloc_state, *start, 1394 &physical_start, &physical_end, 1395 CHUNK_ALLOCATED, NULL)) { 1396 1397 if (in_range(physical_start, *start, len) || 1398 in_range(*start, physical_start, 1399 physical_end - physical_start)) { 1400 *start = physical_end + 1; 1401 return true; 1402 } 1403 } 1404 return false; 1405 } 1406 1407 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1408 { 1409 switch (device->fs_devices->chunk_alloc_policy) { 1410 case BTRFS_CHUNK_ALLOC_REGULAR: 1411 /* 1412 * We don't want to overwrite the superblock on the drive nor 1413 * any area used by the boot loader (grub for example), so we 1414 * make sure to start at an offset of at least 1MB. 1415 */ 1416 return max_t(u64, start, SZ_1M); 1417 case BTRFS_CHUNK_ALLOC_ZONED: 1418 /* 1419 * We don't care about the starting region like regular 1420 * allocator, because we anyway use/reserve the first two zones 1421 * for superblock logging. 1422 */ 1423 return ALIGN(start, device->zone_info->zone_size); 1424 default: 1425 BUG(); 1426 } 1427 } 1428 1429 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1430 u64 *hole_start, u64 *hole_size, 1431 u64 num_bytes) 1432 { 1433 u64 zone_size = device->zone_info->zone_size; 1434 u64 pos; 1435 int ret; 1436 bool changed = false; 1437 1438 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1439 1440 while (*hole_size > 0) { 1441 pos = btrfs_find_allocatable_zones(device, *hole_start, 1442 *hole_start + *hole_size, 1443 num_bytes); 1444 if (pos != *hole_start) { 1445 *hole_size = *hole_start + *hole_size - pos; 1446 *hole_start = pos; 1447 changed = true; 1448 if (*hole_size < num_bytes) 1449 break; 1450 } 1451 1452 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1453 1454 /* Range is ensured to be empty */ 1455 if (!ret) 1456 return changed; 1457 1458 /* Given hole range was invalid (outside of device) */ 1459 if (ret == -ERANGE) { 1460 *hole_start += *hole_size; 1461 *hole_size = 0; 1462 return true; 1463 } 1464 1465 *hole_start += zone_size; 1466 *hole_size -= zone_size; 1467 changed = true; 1468 } 1469 1470 return changed; 1471 } 1472 1473 /** 1474 * dev_extent_hole_check - check if specified hole is suitable for allocation 1475 * @device: the device which we have the hole 1476 * @hole_start: starting position of the hole 1477 * @hole_size: the size of the hole 1478 * @num_bytes: the size of the free space that we need 1479 * 1480 * This function may modify @hole_start and @hole_size to reflect the suitable 1481 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1482 */ 1483 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1484 u64 *hole_size, u64 num_bytes) 1485 { 1486 bool changed = false; 1487 u64 hole_end = *hole_start + *hole_size; 1488 1489 for (;;) { 1490 /* 1491 * Check before we set max_hole_start, otherwise we could end up 1492 * sending back this offset anyway. 1493 */ 1494 if (contains_pending_extent(device, hole_start, *hole_size)) { 1495 if (hole_end >= *hole_start) 1496 *hole_size = hole_end - *hole_start; 1497 else 1498 *hole_size = 0; 1499 changed = true; 1500 } 1501 1502 switch (device->fs_devices->chunk_alloc_policy) { 1503 case BTRFS_CHUNK_ALLOC_REGULAR: 1504 /* No extra check */ 1505 break; 1506 case BTRFS_CHUNK_ALLOC_ZONED: 1507 if (dev_extent_hole_check_zoned(device, hole_start, 1508 hole_size, num_bytes)) { 1509 changed = true; 1510 /* 1511 * The changed hole can contain pending extent. 1512 * Loop again to check that. 1513 */ 1514 continue; 1515 } 1516 break; 1517 default: 1518 BUG(); 1519 } 1520 1521 break; 1522 } 1523 1524 return changed; 1525 } 1526 1527 /* 1528 * find_free_dev_extent_start - find free space in the specified device 1529 * @device: the device which we search the free space in 1530 * @num_bytes: the size of the free space that we need 1531 * @search_start: the position from which to begin the search 1532 * @start: store the start of the free space. 1533 * @len: the size of the free space. that we find, or the size 1534 * of the max free space if we don't find suitable free space 1535 * 1536 * this uses a pretty simple search, the expectation is that it is 1537 * called very infrequently and that a given device has a small number 1538 * of extents 1539 * 1540 * @start is used to store the start of the free space if we find. But if we 1541 * don't find suitable free space, it will be used to store the start position 1542 * of the max free space. 1543 * 1544 * @len is used to store the size of the free space that we find. 1545 * But if we don't find suitable free space, it is used to store the size of 1546 * the max free space. 1547 * 1548 * NOTE: This function will search *commit* root of device tree, and does extra 1549 * check to ensure dev extents are not double allocated. 1550 * This makes the function safe to allocate dev extents but may not report 1551 * correct usable device space, as device extent freed in current transaction 1552 * is not reported as available. 1553 */ 1554 static int find_free_dev_extent_start(struct btrfs_device *device, 1555 u64 num_bytes, u64 search_start, u64 *start, 1556 u64 *len) 1557 { 1558 struct btrfs_fs_info *fs_info = device->fs_info; 1559 struct btrfs_root *root = fs_info->dev_root; 1560 struct btrfs_key key; 1561 struct btrfs_dev_extent *dev_extent; 1562 struct btrfs_path *path; 1563 u64 hole_size; 1564 u64 max_hole_start; 1565 u64 max_hole_size; 1566 u64 extent_end; 1567 u64 search_end = device->total_bytes; 1568 int ret; 1569 int slot; 1570 struct extent_buffer *l; 1571 1572 search_start = dev_extent_search_start(device, search_start); 1573 1574 WARN_ON(device->zone_info && 1575 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1576 1577 path = btrfs_alloc_path(); 1578 if (!path) 1579 return -ENOMEM; 1580 1581 max_hole_start = search_start; 1582 max_hole_size = 0; 1583 1584 again: 1585 if (search_start >= search_end || 1586 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1587 ret = -ENOSPC; 1588 goto out; 1589 } 1590 1591 path->reada = READA_FORWARD; 1592 path->search_commit_root = 1; 1593 path->skip_locking = 1; 1594 1595 key.objectid = device->devid; 1596 key.offset = search_start; 1597 key.type = BTRFS_DEV_EXTENT_KEY; 1598 1599 ret = btrfs_search_backwards(root, &key, path); 1600 if (ret < 0) 1601 goto out; 1602 1603 while (1) { 1604 l = path->nodes[0]; 1605 slot = path->slots[0]; 1606 if (slot >= btrfs_header_nritems(l)) { 1607 ret = btrfs_next_leaf(root, path); 1608 if (ret == 0) 1609 continue; 1610 if (ret < 0) 1611 goto out; 1612 1613 break; 1614 } 1615 btrfs_item_key_to_cpu(l, &key, slot); 1616 1617 if (key.objectid < device->devid) 1618 goto next; 1619 1620 if (key.objectid > device->devid) 1621 break; 1622 1623 if (key.type != BTRFS_DEV_EXTENT_KEY) 1624 goto next; 1625 1626 if (key.offset > search_start) { 1627 hole_size = key.offset - search_start; 1628 dev_extent_hole_check(device, &search_start, &hole_size, 1629 num_bytes); 1630 1631 if (hole_size > max_hole_size) { 1632 max_hole_start = search_start; 1633 max_hole_size = hole_size; 1634 } 1635 1636 /* 1637 * If this free space is greater than which we need, 1638 * it must be the max free space that we have found 1639 * until now, so max_hole_start must point to the start 1640 * of this free space and the length of this free space 1641 * is stored in max_hole_size. Thus, we return 1642 * max_hole_start and max_hole_size and go back to the 1643 * caller. 1644 */ 1645 if (hole_size >= num_bytes) { 1646 ret = 0; 1647 goto out; 1648 } 1649 } 1650 1651 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1652 extent_end = key.offset + btrfs_dev_extent_length(l, 1653 dev_extent); 1654 if (extent_end > search_start) 1655 search_start = extent_end; 1656 next: 1657 path->slots[0]++; 1658 cond_resched(); 1659 } 1660 1661 /* 1662 * At this point, search_start should be the end of 1663 * allocated dev extents, and when shrinking the device, 1664 * search_end may be smaller than search_start. 1665 */ 1666 if (search_end > search_start) { 1667 hole_size = search_end - search_start; 1668 if (dev_extent_hole_check(device, &search_start, &hole_size, 1669 num_bytes)) { 1670 btrfs_release_path(path); 1671 goto again; 1672 } 1673 1674 if (hole_size > max_hole_size) { 1675 max_hole_start = search_start; 1676 max_hole_size = hole_size; 1677 } 1678 } 1679 1680 /* See above. */ 1681 if (max_hole_size < num_bytes) 1682 ret = -ENOSPC; 1683 else 1684 ret = 0; 1685 1686 out: 1687 btrfs_free_path(path); 1688 *start = max_hole_start; 1689 if (len) 1690 *len = max_hole_size; 1691 return ret; 1692 } 1693 1694 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1695 u64 *start, u64 *len) 1696 { 1697 /* FIXME use last free of some kind */ 1698 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1699 } 1700 1701 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1702 struct btrfs_device *device, 1703 u64 start, u64 *dev_extent_len) 1704 { 1705 struct btrfs_fs_info *fs_info = device->fs_info; 1706 struct btrfs_root *root = fs_info->dev_root; 1707 int ret; 1708 struct btrfs_path *path; 1709 struct btrfs_key key; 1710 struct btrfs_key found_key; 1711 struct extent_buffer *leaf = NULL; 1712 struct btrfs_dev_extent *extent = NULL; 1713 1714 path = btrfs_alloc_path(); 1715 if (!path) 1716 return -ENOMEM; 1717 1718 key.objectid = device->devid; 1719 key.offset = start; 1720 key.type = BTRFS_DEV_EXTENT_KEY; 1721 again: 1722 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1723 if (ret > 0) { 1724 ret = btrfs_previous_item(root, path, key.objectid, 1725 BTRFS_DEV_EXTENT_KEY); 1726 if (ret) 1727 goto out; 1728 leaf = path->nodes[0]; 1729 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1730 extent = btrfs_item_ptr(leaf, path->slots[0], 1731 struct btrfs_dev_extent); 1732 BUG_ON(found_key.offset > start || found_key.offset + 1733 btrfs_dev_extent_length(leaf, extent) < start); 1734 key = found_key; 1735 btrfs_release_path(path); 1736 goto again; 1737 } else if (ret == 0) { 1738 leaf = path->nodes[0]; 1739 extent = btrfs_item_ptr(leaf, path->slots[0], 1740 struct btrfs_dev_extent); 1741 } else { 1742 goto out; 1743 } 1744 1745 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1746 1747 ret = btrfs_del_item(trans, root, path); 1748 if (ret == 0) 1749 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1750 out: 1751 btrfs_free_path(path); 1752 return ret; 1753 } 1754 1755 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1756 { 1757 struct extent_map_tree *em_tree; 1758 struct extent_map *em; 1759 struct rb_node *n; 1760 u64 ret = 0; 1761 1762 em_tree = &fs_info->mapping_tree; 1763 read_lock(&em_tree->lock); 1764 n = rb_last(&em_tree->map.rb_root); 1765 if (n) { 1766 em = rb_entry(n, struct extent_map, rb_node); 1767 ret = em->start + em->len; 1768 } 1769 read_unlock(&em_tree->lock); 1770 1771 return ret; 1772 } 1773 1774 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1775 u64 *devid_ret) 1776 { 1777 int ret; 1778 struct btrfs_key key; 1779 struct btrfs_key found_key; 1780 struct btrfs_path *path; 1781 1782 path = btrfs_alloc_path(); 1783 if (!path) 1784 return -ENOMEM; 1785 1786 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1787 key.type = BTRFS_DEV_ITEM_KEY; 1788 key.offset = (u64)-1; 1789 1790 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1791 if (ret < 0) 1792 goto error; 1793 1794 if (ret == 0) { 1795 /* Corruption */ 1796 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1797 ret = -EUCLEAN; 1798 goto error; 1799 } 1800 1801 ret = btrfs_previous_item(fs_info->chunk_root, path, 1802 BTRFS_DEV_ITEMS_OBJECTID, 1803 BTRFS_DEV_ITEM_KEY); 1804 if (ret) { 1805 *devid_ret = 1; 1806 } else { 1807 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1808 path->slots[0]); 1809 *devid_ret = found_key.offset + 1; 1810 } 1811 ret = 0; 1812 error: 1813 btrfs_free_path(path); 1814 return ret; 1815 } 1816 1817 /* 1818 * the device information is stored in the chunk root 1819 * the btrfs_device struct should be fully filled in 1820 */ 1821 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1822 struct btrfs_device *device) 1823 { 1824 int ret; 1825 struct btrfs_path *path; 1826 struct btrfs_dev_item *dev_item; 1827 struct extent_buffer *leaf; 1828 struct btrfs_key key; 1829 unsigned long ptr; 1830 1831 path = btrfs_alloc_path(); 1832 if (!path) 1833 return -ENOMEM; 1834 1835 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1836 key.type = BTRFS_DEV_ITEM_KEY; 1837 key.offset = device->devid; 1838 1839 btrfs_reserve_chunk_metadata(trans, true); 1840 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1841 &key, sizeof(*dev_item)); 1842 btrfs_trans_release_chunk_metadata(trans); 1843 if (ret) 1844 goto out; 1845 1846 leaf = path->nodes[0]; 1847 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1848 1849 btrfs_set_device_id(leaf, dev_item, device->devid); 1850 btrfs_set_device_generation(leaf, dev_item, 0); 1851 btrfs_set_device_type(leaf, dev_item, device->type); 1852 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1853 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1854 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1855 btrfs_set_device_total_bytes(leaf, dev_item, 1856 btrfs_device_get_disk_total_bytes(device)); 1857 btrfs_set_device_bytes_used(leaf, dev_item, 1858 btrfs_device_get_bytes_used(device)); 1859 btrfs_set_device_group(leaf, dev_item, 0); 1860 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1861 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1862 btrfs_set_device_start_offset(leaf, dev_item, 0); 1863 1864 ptr = btrfs_device_uuid(dev_item); 1865 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1866 ptr = btrfs_device_fsid(dev_item); 1867 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1868 ptr, BTRFS_FSID_SIZE); 1869 btrfs_mark_buffer_dirty(leaf); 1870 1871 ret = 0; 1872 out: 1873 btrfs_free_path(path); 1874 return ret; 1875 } 1876 1877 /* 1878 * Function to update ctime/mtime for a given device path. 1879 * Mainly used for ctime/mtime based probe like libblkid. 1880 * 1881 * We don't care about errors here, this is just to be kind to userspace. 1882 */ 1883 static void update_dev_time(const char *device_path) 1884 { 1885 struct path path; 1886 struct timespec64 now; 1887 int ret; 1888 1889 ret = kern_path(device_path, LOOKUP_FOLLOW, &path); 1890 if (ret) 1891 return; 1892 1893 now = current_time(d_inode(path.dentry)); 1894 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); 1895 path_put(&path); 1896 } 1897 1898 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, 1899 struct btrfs_device *device) 1900 { 1901 struct btrfs_root *root = device->fs_info->chunk_root; 1902 int ret; 1903 struct btrfs_path *path; 1904 struct btrfs_key key; 1905 1906 path = btrfs_alloc_path(); 1907 if (!path) 1908 return -ENOMEM; 1909 1910 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1911 key.type = BTRFS_DEV_ITEM_KEY; 1912 key.offset = device->devid; 1913 1914 btrfs_reserve_chunk_metadata(trans, false); 1915 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1916 btrfs_trans_release_chunk_metadata(trans); 1917 if (ret) { 1918 if (ret > 0) 1919 ret = -ENOENT; 1920 goto out; 1921 } 1922 1923 ret = btrfs_del_item(trans, root, path); 1924 out: 1925 btrfs_free_path(path); 1926 return ret; 1927 } 1928 1929 /* 1930 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1931 * filesystem. It's up to the caller to adjust that number regarding eg. device 1932 * replace. 1933 */ 1934 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1935 u64 num_devices) 1936 { 1937 u64 all_avail; 1938 unsigned seq; 1939 int i; 1940 1941 do { 1942 seq = read_seqbegin(&fs_info->profiles_lock); 1943 1944 all_avail = fs_info->avail_data_alloc_bits | 1945 fs_info->avail_system_alloc_bits | 1946 fs_info->avail_metadata_alloc_bits; 1947 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1948 1949 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1950 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1951 continue; 1952 1953 if (num_devices < btrfs_raid_array[i].devs_min) 1954 return btrfs_raid_array[i].mindev_error; 1955 } 1956 1957 return 0; 1958 } 1959 1960 static struct btrfs_device * btrfs_find_next_active_device( 1961 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1962 { 1963 struct btrfs_device *next_device; 1964 1965 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1966 if (next_device != device && 1967 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1968 && next_device->bdev) 1969 return next_device; 1970 } 1971 1972 return NULL; 1973 } 1974 1975 /* 1976 * Helper function to check if the given device is part of s_bdev / latest_dev 1977 * and replace it with the provided or the next active device, in the context 1978 * where this function called, there should be always be another device (or 1979 * this_dev) which is active. 1980 */ 1981 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 1982 struct btrfs_device *next_device) 1983 { 1984 struct btrfs_fs_info *fs_info = device->fs_info; 1985 1986 if (!next_device) 1987 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 1988 device); 1989 ASSERT(next_device); 1990 1991 if (fs_info->sb->s_bdev && 1992 (fs_info->sb->s_bdev == device->bdev)) 1993 fs_info->sb->s_bdev = next_device->bdev; 1994 1995 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 1996 fs_info->fs_devices->latest_dev = next_device; 1997 } 1998 1999 /* 2000 * Return btrfs_fs_devices::num_devices excluding the device that's being 2001 * currently replaced. 2002 */ 2003 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2004 { 2005 u64 num_devices = fs_info->fs_devices->num_devices; 2006 2007 down_read(&fs_info->dev_replace.rwsem); 2008 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2009 ASSERT(num_devices > 1); 2010 num_devices--; 2011 } 2012 up_read(&fs_info->dev_replace.rwsem); 2013 2014 return num_devices; 2015 } 2016 2017 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2018 struct block_device *bdev, 2019 const char *device_path) 2020 { 2021 struct btrfs_super_block *disk_super; 2022 int copy_num; 2023 2024 if (!bdev) 2025 return; 2026 2027 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2028 struct page *page; 2029 int ret; 2030 2031 disk_super = btrfs_read_dev_one_super(bdev, copy_num); 2032 if (IS_ERR(disk_super)) 2033 continue; 2034 2035 if (bdev_is_zoned(bdev)) { 2036 btrfs_reset_sb_log_zones(bdev, copy_num); 2037 continue; 2038 } 2039 2040 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 2041 2042 page = virt_to_page(disk_super); 2043 set_page_dirty(page); 2044 lock_page(page); 2045 /* write_on_page() unlocks the page */ 2046 ret = write_one_page(page); 2047 if (ret) 2048 btrfs_warn(fs_info, 2049 "error clearing superblock number %d (%d)", 2050 copy_num, ret); 2051 btrfs_release_disk_super(disk_super); 2052 2053 } 2054 2055 /* Notify udev that device has changed */ 2056 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2057 2058 /* Update ctime/mtime for device path for libblkid */ 2059 update_dev_time(device_path); 2060 } 2061 2062 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2063 struct btrfs_dev_lookup_args *args, 2064 struct block_device **bdev, fmode_t *mode) 2065 { 2066 struct btrfs_trans_handle *trans; 2067 struct btrfs_device *device; 2068 struct btrfs_fs_devices *cur_devices; 2069 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2070 u64 num_devices; 2071 int ret = 0; 2072 2073 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2074 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); 2075 return -EINVAL; 2076 } 2077 2078 /* 2079 * The device list in fs_devices is accessed without locks (neither 2080 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2081 * filesystem and another device rm cannot run. 2082 */ 2083 num_devices = btrfs_num_devices(fs_info); 2084 2085 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2086 if (ret) 2087 return ret; 2088 2089 device = btrfs_find_device(fs_info->fs_devices, args); 2090 if (!device) { 2091 if (args->missing) 2092 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2093 else 2094 ret = -ENOENT; 2095 return ret; 2096 } 2097 2098 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2099 btrfs_warn_in_rcu(fs_info, 2100 "cannot remove device %s (devid %llu) due to active swapfile", 2101 rcu_str_deref(device->name), device->devid); 2102 return -ETXTBSY; 2103 } 2104 2105 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 2106 return BTRFS_ERROR_DEV_TGT_REPLACE; 2107 2108 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2109 fs_info->fs_devices->rw_devices == 1) 2110 return BTRFS_ERROR_DEV_ONLY_WRITABLE; 2111 2112 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2113 mutex_lock(&fs_info->chunk_mutex); 2114 list_del_init(&device->dev_alloc_list); 2115 device->fs_devices->rw_devices--; 2116 mutex_unlock(&fs_info->chunk_mutex); 2117 } 2118 2119 ret = btrfs_shrink_device(device, 0); 2120 if (ret) 2121 goto error_undo; 2122 2123 trans = btrfs_start_transaction(fs_info->chunk_root, 0); 2124 if (IS_ERR(trans)) { 2125 ret = PTR_ERR(trans); 2126 goto error_undo; 2127 } 2128 2129 ret = btrfs_rm_dev_item(trans, device); 2130 if (ret) { 2131 /* Any error in dev item removal is critical */ 2132 btrfs_crit(fs_info, 2133 "failed to remove device item for devid %llu: %d", 2134 device->devid, ret); 2135 btrfs_abort_transaction(trans, ret); 2136 btrfs_end_transaction(trans); 2137 return ret; 2138 } 2139 2140 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2141 btrfs_scrub_cancel_dev(device); 2142 2143 /* 2144 * the device list mutex makes sure that we don't change 2145 * the device list while someone else is writing out all 2146 * the device supers. Whoever is writing all supers, should 2147 * lock the device list mutex before getting the number of 2148 * devices in the super block (super_copy). Conversely, 2149 * whoever updates the number of devices in the super block 2150 * (super_copy) should hold the device list mutex. 2151 */ 2152 2153 /* 2154 * In normal cases the cur_devices == fs_devices. But in case 2155 * of deleting a seed device, the cur_devices should point to 2156 * its own fs_devices listed under the fs_devices->seed_list. 2157 */ 2158 cur_devices = device->fs_devices; 2159 mutex_lock(&fs_devices->device_list_mutex); 2160 list_del_rcu(&device->dev_list); 2161 2162 cur_devices->num_devices--; 2163 cur_devices->total_devices--; 2164 /* Update total_devices of the parent fs_devices if it's seed */ 2165 if (cur_devices != fs_devices) 2166 fs_devices->total_devices--; 2167 2168 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2169 cur_devices->missing_devices--; 2170 2171 btrfs_assign_next_active_device(device, NULL); 2172 2173 if (device->bdev) { 2174 cur_devices->open_devices--; 2175 /* remove sysfs entry */ 2176 btrfs_sysfs_remove_device(device); 2177 } 2178 2179 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2180 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2181 mutex_unlock(&fs_devices->device_list_mutex); 2182 2183 /* 2184 * At this point, the device is zero sized and detached from the 2185 * devices list. All that's left is to zero out the old supers and 2186 * free the device. 2187 * 2188 * We cannot call btrfs_close_bdev() here because we're holding the sb 2189 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2190 * block device and it's dependencies. Instead just flush the device 2191 * and let the caller do the final blkdev_put. 2192 */ 2193 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2194 btrfs_scratch_superblocks(fs_info, device->bdev, 2195 device->name->str); 2196 if (device->bdev) { 2197 sync_blockdev(device->bdev); 2198 invalidate_bdev(device->bdev); 2199 } 2200 } 2201 2202 *bdev = device->bdev; 2203 *mode = device->mode; 2204 synchronize_rcu(); 2205 btrfs_free_device(device); 2206 2207 /* 2208 * This can happen if cur_devices is the private seed devices list. We 2209 * cannot call close_fs_devices() here because it expects the uuid_mutex 2210 * to be held, but in fact we don't need that for the private 2211 * seed_devices, we can simply decrement cur_devices->opened and then 2212 * remove it from our list and free the fs_devices. 2213 */ 2214 if (cur_devices->num_devices == 0) { 2215 list_del_init(&cur_devices->seed_list); 2216 ASSERT(cur_devices->opened == 1); 2217 cur_devices->opened--; 2218 free_fs_devices(cur_devices); 2219 } 2220 2221 ret = btrfs_commit_transaction(trans); 2222 2223 return ret; 2224 2225 error_undo: 2226 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2227 mutex_lock(&fs_info->chunk_mutex); 2228 list_add(&device->dev_alloc_list, 2229 &fs_devices->alloc_list); 2230 device->fs_devices->rw_devices++; 2231 mutex_unlock(&fs_info->chunk_mutex); 2232 } 2233 return ret; 2234 } 2235 2236 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2237 { 2238 struct btrfs_fs_devices *fs_devices; 2239 2240 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2241 2242 /* 2243 * in case of fs with no seed, srcdev->fs_devices will point 2244 * to fs_devices of fs_info. However when the dev being replaced is 2245 * a seed dev it will point to the seed's local fs_devices. In short 2246 * srcdev will have its correct fs_devices in both the cases. 2247 */ 2248 fs_devices = srcdev->fs_devices; 2249 2250 list_del_rcu(&srcdev->dev_list); 2251 list_del(&srcdev->dev_alloc_list); 2252 fs_devices->num_devices--; 2253 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2254 fs_devices->missing_devices--; 2255 2256 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2257 fs_devices->rw_devices--; 2258 2259 if (srcdev->bdev) 2260 fs_devices->open_devices--; 2261 } 2262 2263 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2264 { 2265 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2266 2267 mutex_lock(&uuid_mutex); 2268 2269 btrfs_close_bdev(srcdev); 2270 synchronize_rcu(); 2271 btrfs_free_device(srcdev); 2272 2273 /* if this is no devs we rather delete the fs_devices */ 2274 if (!fs_devices->num_devices) { 2275 /* 2276 * On a mounted FS, num_devices can't be zero unless it's a 2277 * seed. In case of a seed device being replaced, the replace 2278 * target added to the sprout FS, so there will be no more 2279 * device left under the seed FS. 2280 */ 2281 ASSERT(fs_devices->seeding); 2282 2283 list_del_init(&fs_devices->seed_list); 2284 close_fs_devices(fs_devices); 2285 free_fs_devices(fs_devices); 2286 } 2287 mutex_unlock(&uuid_mutex); 2288 } 2289 2290 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2291 { 2292 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2293 2294 mutex_lock(&fs_devices->device_list_mutex); 2295 2296 btrfs_sysfs_remove_device(tgtdev); 2297 2298 if (tgtdev->bdev) 2299 fs_devices->open_devices--; 2300 2301 fs_devices->num_devices--; 2302 2303 btrfs_assign_next_active_device(tgtdev, NULL); 2304 2305 list_del_rcu(&tgtdev->dev_list); 2306 2307 mutex_unlock(&fs_devices->device_list_mutex); 2308 2309 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2310 tgtdev->name->str); 2311 2312 btrfs_close_bdev(tgtdev); 2313 synchronize_rcu(); 2314 btrfs_free_device(tgtdev); 2315 } 2316 2317 /** 2318 * Populate args from device at path 2319 * 2320 * @fs_info: the filesystem 2321 * @args: the args to populate 2322 * @path: the path to the device 2323 * 2324 * This will read the super block of the device at @path and populate @args with 2325 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2326 * lookup a device to operate on, but need to do it before we take any locks. 2327 * This properly handles the special case of "missing" that a user may pass in, 2328 * and does some basic sanity checks. The caller must make sure that @path is 2329 * properly NUL terminated before calling in, and must call 2330 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2331 * uuid buffers. 2332 * 2333 * Return: 0 for success, -errno for failure 2334 */ 2335 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2336 struct btrfs_dev_lookup_args *args, 2337 const char *path) 2338 { 2339 struct btrfs_super_block *disk_super; 2340 struct block_device *bdev; 2341 int ret; 2342 2343 if (!path || !path[0]) 2344 return -EINVAL; 2345 if (!strcmp(path, "missing")) { 2346 args->missing = true; 2347 return 0; 2348 } 2349 2350 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2351 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2352 if (!args->uuid || !args->fsid) { 2353 btrfs_put_dev_args_from_path(args); 2354 return -ENOMEM; 2355 } 2356 2357 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2358 &bdev, &disk_super); 2359 if (ret) 2360 return ret; 2361 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2362 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2363 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2364 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2365 else 2366 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2367 btrfs_release_disk_super(disk_super); 2368 blkdev_put(bdev, FMODE_READ); 2369 return 0; 2370 } 2371 2372 /* 2373 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2374 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2375 * that don't need to be freed. 2376 */ 2377 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2378 { 2379 kfree(args->uuid); 2380 kfree(args->fsid); 2381 args->uuid = NULL; 2382 args->fsid = NULL; 2383 } 2384 2385 struct btrfs_device *btrfs_find_device_by_devspec( 2386 struct btrfs_fs_info *fs_info, u64 devid, 2387 const char *device_path) 2388 { 2389 BTRFS_DEV_LOOKUP_ARGS(args); 2390 struct btrfs_device *device; 2391 int ret; 2392 2393 if (devid) { 2394 args.devid = devid; 2395 device = btrfs_find_device(fs_info->fs_devices, &args); 2396 if (!device) 2397 return ERR_PTR(-ENOENT); 2398 return device; 2399 } 2400 2401 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2402 if (ret) 2403 return ERR_PTR(ret); 2404 device = btrfs_find_device(fs_info->fs_devices, &args); 2405 btrfs_put_dev_args_from_path(&args); 2406 if (!device) 2407 return ERR_PTR(-ENOENT); 2408 return device; 2409 } 2410 2411 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) 2412 { 2413 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2414 struct btrfs_fs_devices *old_devices; 2415 struct btrfs_fs_devices *seed_devices; 2416 2417 lockdep_assert_held(&uuid_mutex); 2418 if (!fs_devices->seeding) 2419 return ERR_PTR(-EINVAL); 2420 2421 /* 2422 * Private copy of the seed devices, anchored at 2423 * fs_info->fs_devices->seed_list 2424 */ 2425 seed_devices = alloc_fs_devices(NULL, NULL); 2426 if (IS_ERR(seed_devices)) 2427 return seed_devices; 2428 2429 /* 2430 * It's necessary to retain a copy of the original seed fs_devices in 2431 * fs_uuids so that filesystems which have been seeded can successfully 2432 * reference the seed device from open_seed_devices. This also supports 2433 * multiple fs seed. 2434 */ 2435 old_devices = clone_fs_devices(fs_devices); 2436 if (IS_ERR(old_devices)) { 2437 kfree(seed_devices); 2438 return old_devices; 2439 } 2440 2441 list_add(&old_devices->fs_list, &fs_uuids); 2442 2443 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2444 seed_devices->opened = 1; 2445 INIT_LIST_HEAD(&seed_devices->devices); 2446 INIT_LIST_HEAD(&seed_devices->alloc_list); 2447 mutex_init(&seed_devices->device_list_mutex); 2448 2449 return seed_devices; 2450 } 2451 2452 /* 2453 * Splice seed devices into the sprout fs_devices. 2454 * Generate a new fsid for the sprouted read-write filesystem. 2455 */ 2456 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, 2457 struct btrfs_fs_devices *seed_devices) 2458 { 2459 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2460 struct btrfs_super_block *disk_super = fs_info->super_copy; 2461 struct btrfs_device *device; 2462 u64 super_flags; 2463 2464 /* 2465 * We are updating the fsid, the thread leading to device_list_add() 2466 * could race, so uuid_mutex is needed. 2467 */ 2468 lockdep_assert_held(&uuid_mutex); 2469 2470 /* 2471 * The threads listed below may traverse dev_list but can do that without 2472 * device_list_mutex: 2473 * - All device ops and balance - as we are in btrfs_exclop_start. 2474 * - Various dev_list readers - are using RCU. 2475 * - btrfs_ioctl_fitrim() - is using RCU. 2476 * 2477 * For-read threads as below are using device_list_mutex: 2478 * - Readonly scrub btrfs_scrub_dev() 2479 * - Readonly scrub btrfs_scrub_progress() 2480 * - btrfs_get_dev_stats() 2481 */ 2482 lockdep_assert_held(&fs_devices->device_list_mutex); 2483 2484 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2485 synchronize_rcu); 2486 list_for_each_entry(device, &seed_devices->devices, dev_list) 2487 device->fs_devices = seed_devices; 2488 2489 fs_devices->seeding = false; 2490 fs_devices->num_devices = 0; 2491 fs_devices->open_devices = 0; 2492 fs_devices->missing_devices = 0; 2493 fs_devices->rotating = false; 2494 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2495 2496 generate_random_uuid(fs_devices->fsid); 2497 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2498 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2499 2500 super_flags = btrfs_super_flags(disk_super) & 2501 ~BTRFS_SUPER_FLAG_SEEDING; 2502 btrfs_set_super_flags(disk_super, super_flags); 2503 } 2504 2505 /* 2506 * Store the expected generation for seed devices in device items. 2507 */ 2508 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2509 { 2510 BTRFS_DEV_LOOKUP_ARGS(args); 2511 struct btrfs_fs_info *fs_info = trans->fs_info; 2512 struct btrfs_root *root = fs_info->chunk_root; 2513 struct btrfs_path *path; 2514 struct extent_buffer *leaf; 2515 struct btrfs_dev_item *dev_item; 2516 struct btrfs_device *device; 2517 struct btrfs_key key; 2518 u8 fs_uuid[BTRFS_FSID_SIZE]; 2519 u8 dev_uuid[BTRFS_UUID_SIZE]; 2520 int ret; 2521 2522 path = btrfs_alloc_path(); 2523 if (!path) 2524 return -ENOMEM; 2525 2526 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2527 key.offset = 0; 2528 key.type = BTRFS_DEV_ITEM_KEY; 2529 2530 while (1) { 2531 btrfs_reserve_chunk_metadata(trans, false); 2532 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2533 btrfs_trans_release_chunk_metadata(trans); 2534 if (ret < 0) 2535 goto error; 2536 2537 leaf = path->nodes[0]; 2538 next_slot: 2539 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2540 ret = btrfs_next_leaf(root, path); 2541 if (ret > 0) 2542 break; 2543 if (ret < 0) 2544 goto error; 2545 leaf = path->nodes[0]; 2546 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2547 btrfs_release_path(path); 2548 continue; 2549 } 2550 2551 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2552 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2553 key.type != BTRFS_DEV_ITEM_KEY) 2554 break; 2555 2556 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2557 struct btrfs_dev_item); 2558 args.devid = btrfs_device_id(leaf, dev_item); 2559 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2560 BTRFS_UUID_SIZE); 2561 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2562 BTRFS_FSID_SIZE); 2563 args.uuid = dev_uuid; 2564 args.fsid = fs_uuid; 2565 device = btrfs_find_device(fs_info->fs_devices, &args); 2566 BUG_ON(!device); /* Logic error */ 2567 2568 if (device->fs_devices->seeding) { 2569 btrfs_set_device_generation(leaf, dev_item, 2570 device->generation); 2571 btrfs_mark_buffer_dirty(leaf); 2572 } 2573 2574 path->slots[0]++; 2575 goto next_slot; 2576 } 2577 ret = 0; 2578 error: 2579 btrfs_free_path(path); 2580 return ret; 2581 } 2582 2583 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2584 { 2585 struct btrfs_root *root = fs_info->dev_root; 2586 struct btrfs_trans_handle *trans; 2587 struct btrfs_device *device; 2588 struct block_device *bdev; 2589 struct super_block *sb = fs_info->sb; 2590 struct rcu_string *name; 2591 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2592 struct btrfs_fs_devices *seed_devices; 2593 u64 orig_super_total_bytes; 2594 u64 orig_super_num_devices; 2595 int ret = 0; 2596 bool seeding_dev = false; 2597 bool locked = false; 2598 2599 if (sb_rdonly(sb) && !fs_devices->seeding) 2600 return -EROFS; 2601 2602 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2603 fs_info->bdev_holder); 2604 if (IS_ERR(bdev)) 2605 return PTR_ERR(bdev); 2606 2607 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2608 ret = -EINVAL; 2609 goto error; 2610 } 2611 2612 if (fs_devices->seeding) { 2613 seeding_dev = true; 2614 down_write(&sb->s_umount); 2615 mutex_lock(&uuid_mutex); 2616 locked = true; 2617 } 2618 2619 sync_blockdev(bdev); 2620 2621 rcu_read_lock(); 2622 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2623 if (device->bdev == bdev) { 2624 ret = -EEXIST; 2625 rcu_read_unlock(); 2626 goto error; 2627 } 2628 } 2629 rcu_read_unlock(); 2630 2631 device = btrfs_alloc_device(fs_info, NULL, NULL); 2632 if (IS_ERR(device)) { 2633 /* we can safely leave the fs_devices entry around */ 2634 ret = PTR_ERR(device); 2635 goto error; 2636 } 2637 2638 name = rcu_string_strdup(device_path, GFP_KERNEL); 2639 if (!name) { 2640 ret = -ENOMEM; 2641 goto error_free_device; 2642 } 2643 rcu_assign_pointer(device->name, name); 2644 2645 device->fs_info = fs_info; 2646 device->bdev = bdev; 2647 ret = lookup_bdev(device_path, &device->devt); 2648 if (ret) 2649 goto error_free_device; 2650 2651 ret = btrfs_get_dev_zone_info(device, false); 2652 if (ret) 2653 goto error_free_device; 2654 2655 trans = btrfs_start_transaction(root, 0); 2656 if (IS_ERR(trans)) { 2657 ret = PTR_ERR(trans); 2658 goto error_free_zone; 2659 } 2660 2661 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2662 device->generation = trans->transid; 2663 device->io_width = fs_info->sectorsize; 2664 device->io_align = fs_info->sectorsize; 2665 device->sector_size = fs_info->sectorsize; 2666 device->total_bytes = 2667 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); 2668 device->disk_total_bytes = device->total_bytes; 2669 device->commit_total_bytes = device->total_bytes; 2670 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2671 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2672 device->mode = FMODE_EXCL; 2673 device->dev_stats_valid = 1; 2674 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2675 2676 if (seeding_dev) { 2677 btrfs_clear_sb_rdonly(sb); 2678 2679 /* GFP_KERNEL allocation must not be under device_list_mutex */ 2680 seed_devices = btrfs_init_sprout(fs_info); 2681 if (IS_ERR(seed_devices)) { 2682 ret = PTR_ERR(seed_devices); 2683 btrfs_abort_transaction(trans, ret); 2684 goto error_trans; 2685 } 2686 } 2687 2688 mutex_lock(&fs_devices->device_list_mutex); 2689 if (seeding_dev) { 2690 btrfs_setup_sprout(fs_info, seed_devices); 2691 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2692 device); 2693 } 2694 2695 device->fs_devices = fs_devices; 2696 2697 mutex_lock(&fs_info->chunk_mutex); 2698 list_add_rcu(&device->dev_list, &fs_devices->devices); 2699 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2700 fs_devices->num_devices++; 2701 fs_devices->open_devices++; 2702 fs_devices->rw_devices++; 2703 fs_devices->total_devices++; 2704 fs_devices->total_rw_bytes += device->total_bytes; 2705 2706 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2707 2708 if (!bdev_nonrot(bdev)) 2709 fs_devices->rotating = true; 2710 2711 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2712 btrfs_set_super_total_bytes(fs_info->super_copy, 2713 round_down(orig_super_total_bytes + device->total_bytes, 2714 fs_info->sectorsize)); 2715 2716 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2717 btrfs_set_super_num_devices(fs_info->super_copy, 2718 orig_super_num_devices + 1); 2719 2720 /* 2721 * we've got more storage, clear any full flags on the space 2722 * infos 2723 */ 2724 btrfs_clear_space_info_full(fs_info); 2725 2726 mutex_unlock(&fs_info->chunk_mutex); 2727 2728 /* Add sysfs device entry */ 2729 btrfs_sysfs_add_device(device); 2730 2731 mutex_unlock(&fs_devices->device_list_mutex); 2732 2733 if (seeding_dev) { 2734 mutex_lock(&fs_info->chunk_mutex); 2735 ret = init_first_rw_device(trans); 2736 mutex_unlock(&fs_info->chunk_mutex); 2737 if (ret) { 2738 btrfs_abort_transaction(trans, ret); 2739 goto error_sysfs; 2740 } 2741 } 2742 2743 ret = btrfs_add_dev_item(trans, device); 2744 if (ret) { 2745 btrfs_abort_transaction(trans, ret); 2746 goto error_sysfs; 2747 } 2748 2749 if (seeding_dev) { 2750 ret = btrfs_finish_sprout(trans); 2751 if (ret) { 2752 btrfs_abort_transaction(trans, ret); 2753 goto error_sysfs; 2754 } 2755 2756 /* 2757 * fs_devices now represents the newly sprouted filesystem and 2758 * its fsid has been changed by btrfs_sprout_splice(). 2759 */ 2760 btrfs_sysfs_update_sprout_fsid(fs_devices); 2761 } 2762 2763 ret = btrfs_commit_transaction(trans); 2764 2765 if (seeding_dev) { 2766 mutex_unlock(&uuid_mutex); 2767 up_write(&sb->s_umount); 2768 locked = false; 2769 2770 if (ret) /* transaction commit */ 2771 return ret; 2772 2773 ret = btrfs_relocate_sys_chunks(fs_info); 2774 if (ret < 0) 2775 btrfs_handle_fs_error(fs_info, ret, 2776 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2777 trans = btrfs_attach_transaction(root); 2778 if (IS_ERR(trans)) { 2779 if (PTR_ERR(trans) == -ENOENT) 2780 return 0; 2781 ret = PTR_ERR(trans); 2782 trans = NULL; 2783 goto error_sysfs; 2784 } 2785 ret = btrfs_commit_transaction(trans); 2786 } 2787 2788 /* 2789 * Now that we have written a new super block to this device, check all 2790 * other fs_devices list if device_path alienates any other scanned 2791 * device. 2792 * We can ignore the return value as it typically returns -EINVAL and 2793 * only succeeds if the device was an alien. 2794 */ 2795 btrfs_forget_devices(device->devt); 2796 2797 /* Update ctime/mtime for blkid or udev */ 2798 update_dev_time(device_path); 2799 2800 return ret; 2801 2802 error_sysfs: 2803 btrfs_sysfs_remove_device(device); 2804 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2805 mutex_lock(&fs_info->chunk_mutex); 2806 list_del_rcu(&device->dev_list); 2807 list_del(&device->dev_alloc_list); 2808 fs_info->fs_devices->num_devices--; 2809 fs_info->fs_devices->open_devices--; 2810 fs_info->fs_devices->rw_devices--; 2811 fs_info->fs_devices->total_devices--; 2812 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2813 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2814 btrfs_set_super_total_bytes(fs_info->super_copy, 2815 orig_super_total_bytes); 2816 btrfs_set_super_num_devices(fs_info->super_copy, 2817 orig_super_num_devices); 2818 mutex_unlock(&fs_info->chunk_mutex); 2819 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2820 error_trans: 2821 if (seeding_dev) 2822 btrfs_set_sb_rdonly(sb); 2823 if (trans) 2824 btrfs_end_transaction(trans); 2825 error_free_zone: 2826 btrfs_destroy_dev_zone_info(device); 2827 error_free_device: 2828 btrfs_free_device(device); 2829 error: 2830 blkdev_put(bdev, FMODE_EXCL); 2831 if (locked) { 2832 mutex_unlock(&uuid_mutex); 2833 up_write(&sb->s_umount); 2834 } 2835 return ret; 2836 } 2837 2838 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2839 struct btrfs_device *device) 2840 { 2841 int ret; 2842 struct btrfs_path *path; 2843 struct btrfs_root *root = device->fs_info->chunk_root; 2844 struct btrfs_dev_item *dev_item; 2845 struct extent_buffer *leaf; 2846 struct btrfs_key key; 2847 2848 path = btrfs_alloc_path(); 2849 if (!path) 2850 return -ENOMEM; 2851 2852 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2853 key.type = BTRFS_DEV_ITEM_KEY; 2854 key.offset = device->devid; 2855 2856 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2857 if (ret < 0) 2858 goto out; 2859 2860 if (ret > 0) { 2861 ret = -ENOENT; 2862 goto out; 2863 } 2864 2865 leaf = path->nodes[0]; 2866 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2867 2868 btrfs_set_device_id(leaf, dev_item, device->devid); 2869 btrfs_set_device_type(leaf, dev_item, device->type); 2870 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2871 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2872 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2873 btrfs_set_device_total_bytes(leaf, dev_item, 2874 btrfs_device_get_disk_total_bytes(device)); 2875 btrfs_set_device_bytes_used(leaf, dev_item, 2876 btrfs_device_get_bytes_used(device)); 2877 btrfs_mark_buffer_dirty(leaf); 2878 2879 out: 2880 btrfs_free_path(path); 2881 return ret; 2882 } 2883 2884 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2885 struct btrfs_device *device, u64 new_size) 2886 { 2887 struct btrfs_fs_info *fs_info = device->fs_info; 2888 struct btrfs_super_block *super_copy = fs_info->super_copy; 2889 u64 old_total; 2890 u64 diff; 2891 int ret; 2892 2893 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2894 return -EACCES; 2895 2896 new_size = round_down(new_size, fs_info->sectorsize); 2897 2898 mutex_lock(&fs_info->chunk_mutex); 2899 old_total = btrfs_super_total_bytes(super_copy); 2900 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2901 2902 if (new_size <= device->total_bytes || 2903 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2904 mutex_unlock(&fs_info->chunk_mutex); 2905 return -EINVAL; 2906 } 2907 2908 btrfs_set_super_total_bytes(super_copy, 2909 round_down(old_total + diff, fs_info->sectorsize)); 2910 device->fs_devices->total_rw_bytes += diff; 2911 2912 btrfs_device_set_total_bytes(device, new_size); 2913 btrfs_device_set_disk_total_bytes(device, new_size); 2914 btrfs_clear_space_info_full(device->fs_info); 2915 if (list_empty(&device->post_commit_list)) 2916 list_add_tail(&device->post_commit_list, 2917 &trans->transaction->dev_update_list); 2918 mutex_unlock(&fs_info->chunk_mutex); 2919 2920 btrfs_reserve_chunk_metadata(trans, false); 2921 ret = btrfs_update_device(trans, device); 2922 btrfs_trans_release_chunk_metadata(trans); 2923 2924 return ret; 2925 } 2926 2927 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2928 { 2929 struct btrfs_fs_info *fs_info = trans->fs_info; 2930 struct btrfs_root *root = fs_info->chunk_root; 2931 int ret; 2932 struct btrfs_path *path; 2933 struct btrfs_key key; 2934 2935 path = btrfs_alloc_path(); 2936 if (!path) 2937 return -ENOMEM; 2938 2939 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2940 key.offset = chunk_offset; 2941 key.type = BTRFS_CHUNK_ITEM_KEY; 2942 2943 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2944 if (ret < 0) 2945 goto out; 2946 else if (ret > 0) { /* Logic error or corruption */ 2947 btrfs_handle_fs_error(fs_info, -ENOENT, 2948 "Failed lookup while freeing chunk."); 2949 ret = -ENOENT; 2950 goto out; 2951 } 2952 2953 ret = btrfs_del_item(trans, root, path); 2954 if (ret < 0) 2955 btrfs_handle_fs_error(fs_info, ret, 2956 "Failed to delete chunk item."); 2957 out: 2958 btrfs_free_path(path); 2959 return ret; 2960 } 2961 2962 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2963 { 2964 struct btrfs_super_block *super_copy = fs_info->super_copy; 2965 struct btrfs_disk_key *disk_key; 2966 struct btrfs_chunk *chunk; 2967 u8 *ptr; 2968 int ret = 0; 2969 u32 num_stripes; 2970 u32 array_size; 2971 u32 len = 0; 2972 u32 cur; 2973 struct btrfs_key key; 2974 2975 lockdep_assert_held(&fs_info->chunk_mutex); 2976 array_size = btrfs_super_sys_array_size(super_copy); 2977 2978 ptr = super_copy->sys_chunk_array; 2979 cur = 0; 2980 2981 while (cur < array_size) { 2982 disk_key = (struct btrfs_disk_key *)ptr; 2983 btrfs_disk_key_to_cpu(&key, disk_key); 2984 2985 len = sizeof(*disk_key); 2986 2987 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 2988 chunk = (struct btrfs_chunk *)(ptr + len); 2989 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 2990 len += btrfs_chunk_item_size(num_stripes); 2991 } else { 2992 ret = -EIO; 2993 break; 2994 } 2995 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 2996 key.offset == chunk_offset) { 2997 memmove(ptr, ptr + len, array_size - (cur + len)); 2998 array_size -= len; 2999 btrfs_set_super_sys_array_size(super_copy, array_size); 3000 } else { 3001 ptr += len; 3002 cur += len; 3003 } 3004 } 3005 return ret; 3006 } 3007 3008 /* 3009 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 3010 * @logical: Logical block offset in bytes. 3011 * @length: Length of extent in bytes. 3012 * 3013 * Return: Chunk mapping or ERR_PTR. 3014 */ 3015 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 3016 u64 logical, u64 length) 3017 { 3018 struct extent_map_tree *em_tree; 3019 struct extent_map *em; 3020 3021 em_tree = &fs_info->mapping_tree; 3022 read_lock(&em_tree->lock); 3023 em = lookup_extent_mapping(em_tree, logical, length); 3024 read_unlock(&em_tree->lock); 3025 3026 if (!em) { 3027 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 3028 logical, length); 3029 return ERR_PTR(-EINVAL); 3030 } 3031 3032 if (em->start > logical || em->start + em->len < logical) { 3033 btrfs_crit(fs_info, 3034 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3035 logical, length, em->start, em->start + em->len); 3036 free_extent_map(em); 3037 return ERR_PTR(-EINVAL); 3038 } 3039 3040 /* callers are responsible for dropping em's ref. */ 3041 return em; 3042 } 3043 3044 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3045 struct map_lookup *map, u64 chunk_offset) 3046 { 3047 int i; 3048 3049 /* 3050 * Removing chunk items and updating the device items in the chunks btree 3051 * requires holding the chunk_mutex. 3052 * See the comment at btrfs_chunk_alloc() for the details. 3053 */ 3054 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3055 3056 for (i = 0; i < map->num_stripes; i++) { 3057 int ret; 3058 3059 ret = btrfs_update_device(trans, map->stripes[i].dev); 3060 if (ret) 3061 return ret; 3062 } 3063 3064 return btrfs_free_chunk(trans, chunk_offset); 3065 } 3066 3067 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3068 { 3069 struct btrfs_fs_info *fs_info = trans->fs_info; 3070 struct extent_map *em; 3071 struct map_lookup *map; 3072 u64 dev_extent_len = 0; 3073 int i, ret = 0; 3074 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3075 3076 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3077 if (IS_ERR(em)) { 3078 /* 3079 * This is a logic error, but we don't want to just rely on the 3080 * user having built with ASSERT enabled, so if ASSERT doesn't 3081 * do anything we still error out. 3082 */ 3083 ASSERT(0); 3084 return PTR_ERR(em); 3085 } 3086 map = em->map_lookup; 3087 3088 /* 3089 * First delete the device extent items from the devices btree. 3090 * We take the device_list_mutex to avoid racing with the finishing phase 3091 * of a device replace operation. See the comment below before acquiring 3092 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3093 * because that can result in a deadlock when deleting the device extent 3094 * items from the devices btree - COWing an extent buffer from the btree 3095 * may result in allocating a new metadata chunk, which would attempt to 3096 * lock again fs_info->chunk_mutex. 3097 */ 3098 mutex_lock(&fs_devices->device_list_mutex); 3099 for (i = 0; i < map->num_stripes; i++) { 3100 struct btrfs_device *device = map->stripes[i].dev; 3101 ret = btrfs_free_dev_extent(trans, device, 3102 map->stripes[i].physical, 3103 &dev_extent_len); 3104 if (ret) { 3105 mutex_unlock(&fs_devices->device_list_mutex); 3106 btrfs_abort_transaction(trans, ret); 3107 goto out; 3108 } 3109 3110 if (device->bytes_used > 0) { 3111 mutex_lock(&fs_info->chunk_mutex); 3112 btrfs_device_set_bytes_used(device, 3113 device->bytes_used - dev_extent_len); 3114 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3115 btrfs_clear_space_info_full(fs_info); 3116 mutex_unlock(&fs_info->chunk_mutex); 3117 } 3118 } 3119 mutex_unlock(&fs_devices->device_list_mutex); 3120 3121 /* 3122 * We acquire fs_info->chunk_mutex for 2 reasons: 3123 * 3124 * 1) Just like with the first phase of the chunk allocation, we must 3125 * reserve system space, do all chunk btree updates and deletions, and 3126 * update the system chunk array in the superblock while holding this 3127 * mutex. This is for similar reasons as explained on the comment at 3128 * the top of btrfs_chunk_alloc(); 3129 * 3130 * 2) Prevent races with the final phase of a device replace operation 3131 * that replaces the device object associated with the map's stripes, 3132 * because the device object's id can change at any time during that 3133 * final phase of the device replace operation 3134 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3135 * replaced device and then see it with an ID of 3136 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3137 * the device item, which does not exists on the chunk btree. 3138 * The finishing phase of device replace acquires both the 3139 * device_list_mutex and the chunk_mutex, in that order, so we are 3140 * safe by just acquiring the chunk_mutex. 3141 */ 3142 trans->removing_chunk = true; 3143 mutex_lock(&fs_info->chunk_mutex); 3144 3145 check_system_chunk(trans, map->type); 3146 3147 ret = remove_chunk_item(trans, map, chunk_offset); 3148 /* 3149 * Normally we should not get -ENOSPC since we reserved space before 3150 * through the call to check_system_chunk(). 3151 * 3152 * Despite our system space_info having enough free space, we may not 3153 * be able to allocate extents from its block groups, because all have 3154 * an incompatible profile, which will force us to allocate a new system 3155 * block group with the right profile, or right after we called 3156 * check_system_space() above, a scrub turned the only system block group 3157 * with enough free space into RO mode. 3158 * This is explained with more detail at do_chunk_alloc(). 3159 * 3160 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3161 */ 3162 if (ret == -ENOSPC) { 3163 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3164 struct btrfs_block_group *sys_bg; 3165 3166 sys_bg = btrfs_create_chunk(trans, sys_flags); 3167 if (IS_ERR(sys_bg)) { 3168 ret = PTR_ERR(sys_bg); 3169 btrfs_abort_transaction(trans, ret); 3170 goto out; 3171 } 3172 3173 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3174 if (ret) { 3175 btrfs_abort_transaction(trans, ret); 3176 goto out; 3177 } 3178 3179 ret = remove_chunk_item(trans, map, chunk_offset); 3180 if (ret) { 3181 btrfs_abort_transaction(trans, ret); 3182 goto out; 3183 } 3184 } else if (ret) { 3185 btrfs_abort_transaction(trans, ret); 3186 goto out; 3187 } 3188 3189 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3190 3191 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3192 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3193 if (ret) { 3194 btrfs_abort_transaction(trans, ret); 3195 goto out; 3196 } 3197 } 3198 3199 mutex_unlock(&fs_info->chunk_mutex); 3200 trans->removing_chunk = false; 3201 3202 /* 3203 * We are done with chunk btree updates and deletions, so release the 3204 * system space we previously reserved (with check_system_chunk()). 3205 */ 3206 btrfs_trans_release_chunk_metadata(trans); 3207 3208 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3209 if (ret) { 3210 btrfs_abort_transaction(trans, ret); 3211 goto out; 3212 } 3213 3214 out: 3215 if (trans->removing_chunk) { 3216 mutex_unlock(&fs_info->chunk_mutex); 3217 trans->removing_chunk = false; 3218 } 3219 /* once for us */ 3220 free_extent_map(em); 3221 return ret; 3222 } 3223 3224 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3225 { 3226 struct btrfs_root *root = fs_info->chunk_root; 3227 struct btrfs_trans_handle *trans; 3228 struct btrfs_block_group *block_group; 3229 u64 length; 3230 int ret; 3231 3232 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3233 btrfs_err(fs_info, 3234 "relocate: not supported on extent tree v2 yet"); 3235 return -EINVAL; 3236 } 3237 3238 /* 3239 * Prevent races with automatic removal of unused block groups. 3240 * After we relocate and before we remove the chunk with offset 3241 * chunk_offset, automatic removal of the block group can kick in, 3242 * resulting in a failure when calling btrfs_remove_chunk() below. 3243 * 3244 * Make sure to acquire this mutex before doing a tree search (dev 3245 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3246 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3247 * we release the path used to search the chunk/dev tree and before 3248 * the current task acquires this mutex and calls us. 3249 */ 3250 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3251 3252 /* step one, relocate all the extents inside this chunk */ 3253 btrfs_scrub_pause(fs_info); 3254 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3255 btrfs_scrub_continue(fs_info); 3256 if (ret) 3257 return ret; 3258 3259 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3260 if (!block_group) 3261 return -ENOENT; 3262 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3263 length = block_group->length; 3264 btrfs_put_block_group(block_group); 3265 3266 /* 3267 * On a zoned file system, discard the whole block group, this will 3268 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3269 * resetting the zone fails, don't treat it as a fatal problem from the 3270 * filesystem's point of view. 3271 */ 3272 if (btrfs_is_zoned(fs_info)) { 3273 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3274 if (ret) 3275 btrfs_info(fs_info, 3276 "failed to reset zone %llu after relocation", 3277 chunk_offset); 3278 } 3279 3280 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3281 chunk_offset); 3282 if (IS_ERR(trans)) { 3283 ret = PTR_ERR(trans); 3284 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3285 return ret; 3286 } 3287 3288 /* 3289 * step two, delete the device extents and the 3290 * chunk tree entries 3291 */ 3292 ret = btrfs_remove_chunk(trans, chunk_offset); 3293 btrfs_end_transaction(trans); 3294 return ret; 3295 } 3296 3297 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3298 { 3299 struct btrfs_root *chunk_root = fs_info->chunk_root; 3300 struct btrfs_path *path; 3301 struct extent_buffer *leaf; 3302 struct btrfs_chunk *chunk; 3303 struct btrfs_key key; 3304 struct btrfs_key found_key; 3305 u64 chunk_type; 3306 bool retried = false; 3307 int failed = 0; 3308 int ret; 3309 3310 path = btrfs_alloc_path(); 3311 if (!path) 3312 return -ENOMEM; 3313 3314 again: 3315 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3316 key.offset = (u64)-1; 3317 key.type = BTRFS_CHUNK_ITEM_KEY; 3318 3319 while (1) { 3320 mutex_lock(&fs_info->reclaim_bgs_lock); 3321 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3322 if (ret < 0) { 3323 mutex_unlock(&fs_info->reclaim_bgs_lock); 3324 goto error; 3325 } 3326 BUG_ON(ret == 0); /* Corruption */ 3327 3328 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3329 key.type); 3330 if (ret) 3331 mutex_unlock(&fs_info->reclaim_bgs_lock); 3332 if (ret < 0) 3333 goto error; 3334 if (ret > 0) 3335 break; 3336 3337 leaf = path->nodes[0]; 3338 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3339 3340 chunk = btrfs_item_ptr(leaf, path->slots[0], 3341 struct btrfs_chunk); 3342 chunk_type = btrfs_chunk_type(leaf, chunk); 3343 btrfs_release_path(path); 3344 3345 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3346 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3347 if (ret == -ENOSPC) 3348 failed++; 3349 else 3350 BUG_ON(ret); 3351 } 3352 mutex_unlock(&fs_info->reclaim_bgs_lock); 3353 3354 if (found_key.offset == 0) 3355 break; 3356 key.offset = found_key.offset - 1; 3357 } 3358 ret = 0; 3359 if (failed && !retried) { 3360 failed = 0; 3361 retried = true; 3362 goto again; 3363 } else if (WARN_ON(failed && retried)) { 3364 ret = -ENOSPC; 3365 } 3366 error: 3367 btrfs_free_path(path); 3368 return ret; 3369 } 3370 3371 /* 3372 * return 1 : allocate a data chunk successfully, 3373 * return <0: errors during allocating a data chunk, 3374 * return 0 : no need to allocate a data chunk. 3375 */ 3376 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3377 u64 chunk_offset) 3378 { 3379 struct btrfs_block_group *cache; 3380 u64 bytes_used; 3381 u64 chunk_type; 3382 3383 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3384 ASSERT(cache); 3385 chunk_type = cache->flags; 3386 btrfs_put_block_group(cache); 3387 3388 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3389 return 0; 3390 3391 spin_lock(&fs_info->data_sinfo->lock); 3392 bytes_used = fs_info->data_sinfo->bytes_used; 3393 spin_unlock(&fs_info->data_sinfo->lock); 3394 3395 if (!bytes_used) { 3396 struct btrfs_trans_handle *trans; 3397 int ret; 3398 3399 trans = btrfs_join_transaction(fs_info->tree_root); 3400 if (IS_ERR(trans)) 3401 return PTR_ERR(trans); 3402 3403 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3404 btrfs_end_transaction(trans); 3405 if (ret < 0) 3406 return ret; 3407 return 1; 3408 } 3409 3410 return 0; 3411 } 3412 3413 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3414 struct btrfs_balance_control *bctl) 3415 { 3416 struct btrfs_root *root = fs_info->tree_root; 3417 struct btrfs_trans_handle *trans; 3418 struct btrfs_balance_item *item; 3419 struct btrfs_disk_balance_args disk_bargs; 3420 struct btrfs_path *path; 3421 struct extent_buffer *leaf; 3422 struct btrfs_key key; 3423 int ret, err; 3424 3425 path = btrfs_alloc_path(); 3426 if (!path) 3427 return -ENOMEM; 3428 3429 trans = btrfs_start_transaction(root, 0); 3430 if (IS_ERR(trans)) { 3431 btrfs_free_path(path); 3432 return PTR_ERR(trans); 3433 } 3434 3435 key.objectid = BTRFS_BALANCE_OBJECTID; 3436 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3437 key.offset = 0; 3438 3439 ret = btrfs_insert_empty_item(trans, root, path, &key, 3440 sizeof(*item)); 3441 if (ret) 3442 goto out; 3443 3444 leaf = path->nodes[0]; 3445 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3446 3447 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3448 3449 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3450 btrfs_set_balance_data(leaf, item, &disk_bargs); 3451 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3452 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3453 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3454 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3455 3456 btrfs_set_balance_flags(leaf, item, bctl->flags); 3457 3458 btrfs_mark_buffer_dirty(leaf); 3459 out: 3460 btrfs_free_path(path); 3461 err = btrfs_commit_transaction(trans); 3462 if (err && !ret) 3463 ret = err; 3464 return ret; 3465 } 3466 3467 static int del_balance_item(struct btrfs_fs_info *fs_info) 3468 { 3469 struct btrfs_root *root = fs_info->tree_root; 3470 struct btrfs_trans_handle *trans; 3471 struct btrfs_path *path; 3472 struct btrfs_key key; 3473 int ret, err; 3474 3475 path = btrfs_alloc_path(); 3476 if (!path) 3477 return -ENOMEM; 3478 3479 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3480 if (IS_ERR(trans)) { 3481 btrfs_free_path(path); 3482 return PTR_ERR(trans); 3483 } 3484 3485 key.objectid = BTRFS_BALANCE_OBJECTID; 3486 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3487 key.offset = 0; 3488 3489 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3490 if (ret < 0) 3491 goto out; 3492 if (ret > 0) { 3493 ret = -ENOENT; 3494 goto out; 3495 } 3496 3497 ret = btrfs_del_item(trans, root, path); 3498 out: 3499 btrfs_free_path(path); 3500 err = btrfs_commit_transaction(trans); 3501 if (err && !ret) 3502 ret = err; 3503 return ret; 3504 } 3505 3506 /* 3507 * This is a heuristic used to reduce the number of chunks balanced on 3508 * resume after balance was interrupted. 3509 */ 3510 static void update_balance_args(struct btrfs_balance_control *bctl) 3511 { 3512 /* 3513 * Turn on soft mode for chunk types that were being converted. 3514 */ 3515 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3516 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3517 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3518 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3519 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3520 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3521 3522 /* 3523 * Turn on usage filter if is not already used. The idea is 3524 * that chunks that we have already balanced should be 3525 * reasonably full. Don't do it for chunks that are being 3526 * converted - that will keep us from relocating unconverted 3527 * (albeit full) chunks. 3528 */ 3529 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3530 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3531 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3532 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3533 bctl->data.usage = 90; 3534 } 3535 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3536 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3537 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3538 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3539 bctl->sys.usage = 90; 3540 } 3541 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3542 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3543 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3544 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3545 bctl->meta.usage = 90; 3546 } 3547 } 3548 3549 /* 3550 * Clear the balance status in fs_info and delete the balance item from disk. 3551 */ 3552 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3553 { 3554 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3555 int ret; 3556 3557 BUG_ON(!fs_info->balance_ctl); 3558 3559 spin_lock(&fs_info->balance_lock); 3560 fs_info->balance_ctl = NULL; 3561 spin_unlock(&fs_info->balance_lock); 3562 3563 kfree(bctl); 3564 ret = del_balance_item(fs_info); 3565 if (ret) 3566 btrfs_handle_fs_error(fs_info, ret, NULL); 3567 } 3568 3569 /* 3570 * Balance filters. Return 1 if chunk should be filtered out 3571 * (should not be balanced). 3572 */ 3573 static int chunk_profiles_filter(u64 chunk_type, 3574 struct btrfs_balance_args *bargs) 3575 { 3576 chunk_type = chunk_to_extended(chunk_type) & 3577 BTRFS_EXTENDED_PROFILE_MASK; 3578 3579 if (bargs->profiles & chunk_type) 3580 return 0; 3581 3582 return 1; 3583 } 3584 3585 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3586 struct btrfs_balance_args *bargs) 3587 { 3588 struct btrfs_block_group *cache; 3589 u64 chunk_used; 3590 u64 user_thresh_min; 3591 u64 user_thresh_max; 3592 int ret = 1; 3593 3594 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3595 chunk_used = cache->used; 3596 3597 if (bargs->usage_min == 0) 3598 user_thresh_min = 0; 3599 else 3600 user_thresh_min = div_factor_fine(cache->length, 3601 bargs->usage_min); 3602 3603 if (bargs->usage_max == 0) 3604 user_thresh_max = 1; 3605 else if (bargs->usage_max > 100) 3606 user_thresh_max = cache->length; 3607 else 3608 user_thresh_max = div_factor_fine(cache->length, 3609 bargs->usage_max); 3610 3611 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3612 ret = 0; 3613 3614 btrfs_put_block_group(cache); 3615 return ret; 3616 } 3617 3618 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3619 u64 chunk_offset, struct btrfs_balance_args *bargs) 3620 { 3621 struct btrfs_block_group *cache; 3622 u64 chunk_used, user_thresh; 3623 int ret = 1; 3624 3625 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3626 chunk_used = cache->used; 3627 3628 if (bargs->usage_min == 0) 3629 user_thresh = 1; 3630 else if (bargs->usage > 100) 3631 user_thresh = cache->length; 3632 else 3633 user_thresh = div_factor_fine(cache->length, bargs->usage); 3634 3635 if (chunk_used < user_thresh) 3636 ret = 0; 3637 3638 btrfs_put_block_group(cache); 3639 return ret; 3640 } 3641 3642 static int chunk_devid_filter(struct extent_buffer *leaf, 3643 struct btrfs_chunk *chunk, 3644 struct btrfs_balance_args *bargs) 3645 { 3646 struct btrfs_stripe *stripe; 3647 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3648 int i; 3649 3650 for (i = 0; i < num_stripes; i++) { 3651 stripe = btrfs_stripe_nr(chunk, i); 3652 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3653 return 0; 3654 } 3655 3656 return 1; 3657 } 3658 3659 static u64 calc_data_stripes(u64 type, int num_stripes) 3660 { 3661 const int index = btrfs_bg_flags_to_raid_index(type); 3662 const int ncopies = btrfs_raid_array[index].ncopies; 3663 const int nparity = btrfs_raid_array[index].nparity; 3664 3665 return (num_stripes - nparity) / ncopies; 3666 } 3667 3668 /* [pstart, pend) */ 3669 static int chunk_drange_filter(struct extent_buffer *leaf, 3670 struct btrfs_chunk *chunk, 3671 struct btrfs_balance_args *bargs) 3672 { 3673 struct btrfs_stripe *stripe; 3674 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3675 u64 stripe_offset; 3676 u64 stripe_length; 3677 u64 type; 3678 int factor; 3679 int i; 3680 3681 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3682 return 0; 3683 3684 type = btrfs_chunk_type(leaf, chunk); 3685 factor = calc_data_stripes(type, num_stripes); 3686 3687 for (i = 0; i < num_stripes; i++) { 3688 stripe = btrfs_stripe_nr(chunk, i); 3689 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3690 continue; 3691 3692 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3693 stripe_length = btrfs_chunk_length(leaf, chunk); 3694 stripe_length = div_u64(stripe_length, factor); 3695 3696 if (stripe_offset < bargs->pend && 3697 stripe_offset + stripe_length > bargs->pstart) 3698 return 0; 3699 } 3700 3701 return 1; 3702 } 3703 3704 /* [vstart, vend) */ 3705 static int chunk_vrange_filter(struct extent_buffer *leaf, 3706 struct btrfs_chunk *chunk, 3707 u64 chunk_offset, 3708 struct btrfs_balance_args *bargs) 3709 { 3710 if (chunk_offset < bargs->vend && 3711 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3712 /* at least part of the chunk is inside this vrange */ 3713 return 0; 3714 3715 return 1; 3716 } 3717 3718 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3719 struct btrfs_chunk *chunk, 3720 struct btrfs_balance_args *bargs) 3721 { 3722 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3723 3724 if (bargs->stripes_min <= num_stripes 3725 && num_stripes <= bargs->stripes_max) 3726 return 0; 3727 3728 return 1; 3729 } 3730 3731 static int chunk_soft_convert_filter(u64 chunk_type, 3732 struct btrfs_balance_args *bargs) 3733 { 3734 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3735 return 0; 3736 3737 chunk_type = chunk_to_extended(chunk_type) & 3738 BTRFS_EXTENDED_PROFILE_MASK; 3739 3740 if (bargs->target == chunk_type) 3741 return 1; 3742 3743 return 0; 3744 } 3745 3746 static int should_balance_chunk(struct extent_buffer *leaf, 3747 struct btrfs_chunk *chunk, u64 chunk_offset) 3748 { 3749 struct btrfs_fs_info *fs_info = leaf->fs_info; 3750 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3751 struct btrfs_balance_args *bargs = NULL; 3752 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3753 3754 /* type filter */ 3755 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3756 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3757 return 0; 3758 } 3759 3760 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3761 bargs = &bctl->data; 3762 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3763 bargs = &bctl->sys; 3764 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3765 bargs = &bctl->meta; 3766 3767 /* profiles filter */ 3768 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3769 chunk_profiles_filter(chunk_type, bargs)) { 3770 return 0; 3771 } 3772 3773 /* usage filter */ 3774 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3775 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3776 return 0; 3777 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3778 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3779 return 0; 3780 } 3781 3782 /* devid filter */ 3783 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3784 chunk_devid_filter(leaf, chunk, bargs)) { 3785 return 0; 3786 } 3787 3788 /* drange filter, makes sense only with devid filter */ 3789 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3790 chunk_drange_filter(leaf, chunk, bargs)) { 3791 return 0; 3792 } 3793 3794 /* vrange filter */ 3795 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3796 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3797 return 0; 3798 } 3799 3800 /* stripes filter */ 3801 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3802 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3803 return 0; 3804 } 3805 3806 /* soft profile changing mode */ 3807 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3808 chunk_soft_convert_filter(chunk_type, bargs)) { 3809 return 0; 3810 } 3811 3812 /* 3813 * limited by count, must be the last filter 3814 */ 3815 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3816 if (bargs->limit == 0) 3817 return 0; 3818 else 3819 bargs->limit--; 3820 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3821 /* 3822 * Same logic as the 'limit' filter; the minimum cannot be 3823 * determined here because we do not have the global information 3824 * about the count of all chunks that satisfy the filters. 3825 */ 3826 if (bargs->limit_max == 0) 3827 return 0; 3828 else 3829 bargs->limit_max--; 3830 } 3831 3832 return 1; 3833 } 3834 3835 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3836 { 3837 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3838 struct btrfs_root *chunk_root = fs_info->chunk_root; 3839 u64 chunk_type; 3840 struct btrfs_chunk *chunk; 3841 struct btrfs_path *path = NULL; 3842 struct btrfs_key key; 3843 struct btrfs_key found_key; 3844 struct extent_buffer *leaf; 3845 int slot; 3846 int ret; 3847 int enospc_errors = 0; 3848 bool counting = true; 3849 /* The single value limit and min/max limits use the same bytes in the */ 3850 u64 limit_data = bctl->data.limit; 3851 u64 limit_meta = bctl->meta.limit; 3852 u64 limit_sys = bctl->sys.limit; 3853 u32 count_data = 0; 3854 u32 count_meta = 0; 3855 u32 count_sys = 0; 3856 int chunk_reserved = 0; 3857 3858 path = btrfs_alloc_path(); 3859 if (!path) { 3860 ret = -ENOMEM; 3861 goto error; 3862 } 3863 3864 /* zero out stat counters */ 3865 spin_lock(&fs_info->balance_lock); 3866 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3867 spin_unlock(&fs_info->balance_lock); 3868 again: 3869 if (!counting) { 3870 /* 3871 * The single value limit and min/max limits use the same bytes 3872 * in the 3873 */ 3874 bctl->data.limit = limit_data; 3875 bctl->meta.limit = limit_meta; 3876 bctl->sys.limit = limit_sys; 3877 } 3878 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3879 key.offset = (u64)-1; 3880 key.type = BTRFS_CHUNK_ITEM_KEY; 3881 3882 while (1) { 3883 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3884 atomic_read(&fs_info->balance_cancel_req)) { 3885 ret = -ECANCELED; 3886 goto error; 3887 } 3888 3889 mutex_lock(&fs_info->reclaim_bgs_lock); 3890 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3891 if (ret < 0) { 3892 mutex_unlock(&fs_info->reclaim_bgs_lock); 3893 goto error; 3894 } 3895 3896 /* 3897 * this shouldn't happen, it means the last relocate 3898 * failed 3899 */ 3900 if (ret == 0) 3901 BUG(); /* FIXME break ? */ 3902 3903 ret = btrfs_previous_item(chunk_root, path, 0, 3904 BTRFS_CHUNK_ITEM_KEY); 3905 if (ret) { 3906 mutex_unlock(&fs_info->reclaim_bgs_lock); 3907 ret = 0; 3908 break; 3909 } 3910 3911 leaf = path->nodes[0]; 3912 slot = path->slots[0]; 3913 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3914 3915 if (found_key.objectid != key.objectid) { 3916 mutex_unlock(&fs_info->reclaim_bgs_lock); 3917 break; 3918 } 3919 3920 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3921 chunk_type = btrfs_chunk_type(leaf, chunk); 3922 3923 if (!counting) { 3924 spin_lock(&fs_info->balance_lock); 3925 bctl->stat.considered++; 3926 spin_unlock(&fs_info->balance_lock); 3927 } 3928 3929 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3930 3931 btrfs_release_path(path); 3932 if (!ret) { 3933 mutex_unlock(&fs_info->reclaim_bgs_lock); 3934 goto loop; 3935 } 3936 3937 if (counting) { 3938 mutex_unlock(&fs_info->reclaim_bgs_lock); 3939 spin_lock(&fs_info->balance_lock); 3940 bctl->stat.expected++; 3941 spin_unlock(&fs_info->balance_lock); 3942 3943 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3944 count_data++; 3945 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3946 count_sys++; 3947 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3948 count_meta++; 3949 3950 goto loop; 3951 } 3952 3953 /* 3954 * Apply limit_min filter, no need to check if the LIMITS 3955 * filter is used, limit_min is 0 by default 3956 */ 3957 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3958 count_data < bctl->data.limit_min) 3959 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3960 count_meta < bctl->meta.limit_min) 3961 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3962 count_sys < bctl->sys.limit_min)) { 3963 mutex_unlock(&fs_info->reclaim_bgs_lock); 3964 goto loop; 3965 } 3966 3967 if (!chunk_reserved) { 3968 /* 3969 * We may be relocating the only data chunk we have, 3970 * which could potentially end up with losing data's 3971 * raid profile, so lets allocate an empty one in 3972 * advance. 3973 */ 3974 ret = btrfs_may_alloc_data_chunk(fs_info, 3975 found_key.offset); 3976 if (ret < 0) { 3977 mutex_unlock(&fs_info->reclaim_bgs_lock); 3978 goto error; 3979 } else if (ret == 1) { 3980 chunk_reserved = 1; 3981 } 3982 } 3983 3984 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3985 mutex_unlock(&fs_info->reclaim_bgs_lock); 3986 if (ret == -ENOSPC) { 3987 enospc_errors++; 3988 } else if (ret == -ETXTBSY) { 3989 btrfs_info(fs_info, 3990 "skipping relocation of block group %llu due to active swapfile", 3991 found_key.offset); 3992 ret = 0; 3993 } else if (ret) { 3994 goto error; 3995 } else { 3996 spin_lock(&fs_info->balance_lock); 3997 bctl->stat.completed++; 3998 spin_unlock(&fs_info->balance_lock); 3999 } 4000 loop: 4001 if (found_key.offset == 0) 4002 break; 4003 key.offset = found_key.offset - 1; 4004 } 4005 4006 if (counting) { 4007 btrfs_release_path(path); 4008 counting = false; 4009 goto again; 4010 } 4011 error: 4012 btrfs_free_path(path); 4013 if (enospc_errors) { 4014 btrfs_info(fs_info, "%d enospc errors during balance", 4015 enospc_errors); 4016 if (!ret) 4017 ret = -ENOSPC; 4018 } 4019 4020 return ret; 4021 } 4022 4023 /** 4024 * alloc_profile_is_valid - see if a given profile is valid and reduced 4025 * @flags: profile to validate 4026 * @extended: if true @flags is treated as an extended profile 4027 */ 4028 static int alloc_profile_is_valid(u64 flags, int extended) 4029 { 4030 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 4031 BTRFS_BLOCK_GROUP_PROFILE_MASK); 4032 4033 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 4034 4035 /* 1) check that all other bits are zeroed */ 4036 if (flags & ~mask) 4037 return 0; 4038 4039 /* 2) see if profile is reduced */ 4040 if (flags == 0) 4041 return !extended; /* "0" is valid for usual profiles */ 4042 4043 return has_single_bit_set(flags); 4044 } 4045 4046 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4047 { 4048 /* cancel requested || normal exit path */ 4049 return atomic_read(&fs_info->balance_cancel_req) || 4050 (atomic_read(&fs_info->balance_pause_req) == 0 && 4051 atomic_read(&fs_info->balance_cancel_req) == 0); 4052 } 4053 4054 /* 4055 * Validate target profile against allowed profiles and return true if it's OK. 4056 * Otherwise print the error message and return false. 4057 */ 4058 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4059 const struct btrfs_balance_args *bargs, 4060 u64 allowed, const char *type) 4061 { 4062 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4063 return true; 4064 4065 if (fs_info->sectorsize < PAGE_SIZE && 4066 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) { 4067 btrfs_err(fs_info, 4068 "RAID56 is not yet supported for sectorsize %u with page size %lu", 4069 fs_info->sectorsize, PAGE_SIZE); 4070 return false; 4071 } 4072 /* Profile is valid and does not have bits outside of the allowed set */ 4073 if (alloc_profile_is_valid(bargs->target, 1) && 4074 (bargs->target & ~allowed) == 0) 4075 return true; 4076 4077 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4078 type, btrfs_bg_type_to_raid_name(bargs->target)); 4079 return false; 4080 } 4081 4082 /* 4083 * Fill @buf with textual description of balance filter flags @bargs, up to 4084 * @size_buf including the terminating null. The output may be trimmed if it 4085 * does not fit into the provided buffer. 4086 */ 4087 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4088 u32 size_buf) 4089 { 4090 int ret; 4091 u32 size_bp = size_buf; 4092 char *bp = buf; 4093 u64 flags = bargs->flags; 4094 char tmp_buf[128] = {'\0'}; 4095 4096 if (!flags) 4097 return; 4098 4099 #define CHECK_APPEND_NOARG(a) \ 4100 do { \ 4101 ret = snprintf(bp, size_bp, (a)); \ 4102 if (ret < 0 || ret >= size_bp) \ 4103 goto out_overflow; \ 4104 size_bp -= ret; \ 4105 bp += ret; \ 4106 } while (0) 4107 4108 #define CHECK_APPEND_1ARG(a, v1) \ 4109 do { \ 4110 ret = snprintf(bp, size_bp, (a), (v1)); \ 4111 if (ret < 0 || ret >= size_bp) \ 4112 goto out_overflow; \ 4113 size_bp -= ret; \ 4114 bp += ret; \ 4115 } while (0) 4116 4117 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4118 do { \ 4119 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4120 if (ret < 0 || ret >= size_bp) \ 4121 goto out_overflow; \ 4122 size_bp -= ret; \ 4123 bp += ret; \ 4124 } while (0) 4125 4126 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4127 CHECK_APPEND_1ARG("convert=%s,", 4128 btrfs_bg_type_to_raid_name(bargs->target)); 4129 4130 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4131 CHECK_APPEND_NOARG("soft,"); 4132 4133 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4134 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4135 sizeof(tmp_buf)); 4136 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4137 } 4138 4139 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4140 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4141 4142 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4143 CHECK_APPEND_2ARG("usage=%u..%u,", 4144 bargs->usage_min, bargs->usage_max); 4145 4146 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4147 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4148 4149 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4150 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4151 bargs->pstart, bargs->pend); 4152 4153 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4154 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4155 bargs->vstart, bargs->vend); 4156 4157 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4158 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4159 4160 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4161 CHECK_APPEND_2ARG("limit=%u..%u,", 4162 bargs->limit_min, bargs->limit_max); 4163 4164 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4165 CHECK_APPEND_2ARG("stripes=%u..%u,", 4166 bargs->stripes_min, bargs->stripes_max); 4167 4168 #undef CHECK_APPEND_2ARG 4169 #undef CHECK_APPEND_1ARG 4170 #undef CHECK_APPEND_NOARG 4171 4172 out_overflow: 4173 4174 if (size_bp < size_buf) 4175 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4176 else 4177 buf[0] = '\0'; 4178 } 4179 4180 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4181 { 4182 u32 size_buf = 1024; 4183 char tmp_buf[192] = {'\0'}; 4184 char *buf; 4185 char *bp; 4186 u32 size_bp = size_buf; 4187 int ret; 4188 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4189 4190 buf = kzalloc(size_buf, GFP_KERNEL); 4191 if (!buf) 4192 return; 4193 4194 bp = buf; 4195 4196 #define CHECK_APPEND_1ARG(a, v1) \ 4197 do { \ 4198 ret = snprintf(bp, size_bp, (a), (v1)); \ 4199 if (ret < 0 || ret >= size_bp) \ 4200 goto out_overflow; \ 4201 size_bp -= ret; \ 4202 bp += ret; \ 4203 } while (0) 4204 4205 if (bctl->flags & BTRFS_BALANCE_FORCE) 4206 CHECK_APPEND_1ARG("%s", "-f "); 4207 4208 if (bctl->flags & BTRFS_BALANCE_DATA) { 4209 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4210 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4211 } 4212 4213 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4214 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4215 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4216 } 4217 4218 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4219 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4220 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4221 } 4222 4223 #undef CHECK_APPEND_1ARG 4224 4225 out_overflow: 4226 4227 if (size_bp < size_buf) 4228 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4229 btrfs_info(fs_info, "balance: %s %s", 4230 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4231 "resume" : "start", buf); 4232 4233 kfree(buf); 4234 } 4235 4236 /* 4237 * Should be called with balance mutexe held 4238 */ 4239 int btrfs_balance(struct btrfs_fs_info *fs_info, 4240 struct btrfs_balance_control *bctl, 4241 struct btrfs_ioctl_balance_args *bargs) 4242 { 4243 u64 meta_target, data_target; 4244 u64 allowed; 4245 int mixed = 0; 4246 int ret; 4247 u64 num_devices; 4248 unsigned seq; 4249 bool reducing_redundancy; 4250 int i; 4251 4252 if (btrfs_fs_closing(fs_info) || 4253 atomic_read(&fs_info->balance_pause_req) || 4254 btrfs_should_cancel_balance(fs_info)) { 4255 ret = -EINVAL; 4256 goto out; 4257 } 4258 4259 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4260 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4261 mixed = 1; 4262 4263 /* 4264 * In case of mixed groups both data and meta should be picked, 4265 * and identical options should be given for both of them. 4266 */ 4267 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4268 if (mixed && (bctl->flags & allowed)) { 4269 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4270 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4271 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4272 btrfs_err(fs_info, 4273 "balance: mixed groups data and metadata options must be the same"); 4274 ret = -EINVAL; 4275 goto out; 4276 } 4277 } 4278 4279 /* 4280 * rw_devices will not change at the moment, device add/delete/replace 4281 * are exclusive 4282 */ 4283 num_devices = fs_info->fs_devices->rw_devices; 4284 4285 /* 4286 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4287 * special bit for it, to make it easier to distinguish. Thus we need 4288 * to set it manually, or balance would refuse the profile. 4289 */ 4290 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4291 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4292 if (num_devices >= btrfs_raid_array[i].devs_min) 4293 allowed |= btrfs_raid_array[i].bg_flag; 4294 4295 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4296 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4297 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4298 ret = -EINVAL; 4299 goto out; 4300 } 4301 4302 /* 4303 * Allow to reduce metadata or system integrity only if force set for 4304 * profiles with redundancy (copies, parity) 4305 */ 4306 allowed = 0; 4307 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4308 if (btrfs_raid_array[i].ncopies >= 2 || 4309 btrfs_raid_array[i].tolerated_failures >= 1) 4310 allowed |= btrfs_raid_array[i].bg_flag; 4311 } 4312 do { 4313 seq = read_seqbegin(&fs_info->profiles_lock); 4314 4315 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4316 (fs_info->avail_system_alloc_bits & allowed) && 4317 !(bctl->sys.target & allowed)) || 4318 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4319 (fs_info->avail_metadata_alloc_bits & allowed) && 4320 !(bctl->meta.target & allowed))) 4321 reducing_redundancy = true; 4322 else 4323 reducing_redundancy = false; 4324 4325 /* if we're not converting, the target field is uninitialized */ 4326 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4327 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4328 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4329 bctl->data.target : fs_info->avail_data_alloc_bits; 4330 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4331 4332 if (reducing_redundancy) { 4333 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4334 btrfs_info(fs_info, 4335 "balance: force reducing metadata redundancy"); 4336 } else { 4337 btrfs_err(fs_info, 4338 "balance: reduces metadata redundancy, use --force if you want this"); 4339 ret = -EINVAL; 4340 goto out; 4341 } 4342 } 4343 4344 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4345 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4346 btrfs_warn(fs_info, 4347 "balance: metadata profile %s has lower redundancy than data profile %s", 4348 btrfs_bg_type_to_raid_name(meta_target), 4349 btrfs_bg_type_to_raid_name(data_target)); 4350 } 4351 4352 ret = insert_balance_item(fs_info, bctl); 4353 if (ret && ret != -EEXIST) 4354 goto out; 4355 4356 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4357 BUG_ON(ret == -EEXIST); 4358 BUG_ON(fs_info->balance_ctl); 4359 spin_lock(&fs_info->balance_lock); 4360 fs_info->balance_ctl = bctl; 4361 spin_unlock(&fs_info->balance_lock); 4362 } else { 4363 BUG_ON(ret != -EEXIST); 4364 spin_lock(&fs_info->balance_lock); 4365 update_balance_args(bctl); 4366 spin_unlock(&fs_info->balance_lock); 4367 } 4368 4369 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4370 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4371 describe_balance_start_or_resume(fs_info); 4372 mutex_unlock(&fs_info->balance_mutex); 4373 4374 ret = __btrfs_balance(fs_info); 4375 4376 mutex_lock(&fs_info->balance_mutex); 4377 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { 4378 btrfs_info(fs_info, "balance: paused"); 4379 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 4380 } 4381 /* 4382 * Balance can be canceled by: 4383 * 4384 * - Regular cancel request 4385 * Then ret == -ECANCELED and balance_cancel_req > 0 4386 * 4387 * - Fatal signal to "btrfs" process 4388 * Either the signal caught by wait_reserve_ticket() and callers 4389 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4390 * got -ECANCELED. 4391 * Either way, in this case balance_cancel_req = 0, and 4392 * ret == -EINTR or ret == -ECANCELED. 4393 * 4394 * So here we only check the return value to catch canceled balance. 4395 */ 4396 else if (ret == -ECANCELED || ret == -EINTR) 4397 btrfs_info(fs_info, "balance: canceled"); 4398 else 4399 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4400 4401 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4402 4403 if (bargs) { 4404 memset(bargs, 0, sizeof(*bargs)); 4405 btrfs_update_ioctl_balance_args(fs_info, bargs); 4406 } 4407 4408 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4409 balance_need_close(fs_info)) { 4410 reset_balance_state(fs_info); 4411 btrfs_exclop_finish(fs_info); 4412 } 4413 4414 wake_up(&fs_info->balance_wait_q); 4415 4416 return ret; 4417 out: 4418 if (bctl->flags & BTRFS_BALANCE_RESUME) 4419 reset_balance_state(fs_info); 4420 else 4421 kfree(bctl); 4422 btrfs_exclop_finish(fs_info); 4423 4424 return ret; 4425 } 4426 4427 static int balance_kthread(void *data) 4428 { 4429 struct btrfs_fs_info *fs_info = data; 4430 int ret = 0; 4431 4432 sb_start_write(fs_info->sb); 4433 mutex_lock(&fs_info->balance_mutex); 4434 if (fs_info->balance_ctl) 4435 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4436 mutex_unlock(&fs_info->balance_mutex); 4437 sb_end_write(fs_info->sb); 4438 4439 return ret; 4440 } 4441 4442 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4443 { 4444 struct task_struct *tsk; 4445 4446 mutex_lock(&fs_info->balance_mutex); 4447 if (!fs_info->balance_ctl) { 4448 mutex_unlock(&fs_info->balance_mutex); 4449 return 0; 4450 } 4451 mutex_unlock(&fs_info->balance_mutex); 4452 4453 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4454 btrfs_info(fs_info, "balance: resume skipped"); 4455 return 0; 4456 } 4457 4458 spin_lock(&fs_info->super_lock); 4459 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 4460 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 4461 spin_unlock(&fs_info->super_lock); 4462 /* 4463 * A ro->rw remount sequence should continue with the paused balance 4464 * regardless of who pauses it, system or the user as of now, so set 4465 * the resume flag. 4466 */ 4467 spin_lock(&fs_info->balance_lock); 4468 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4469 spin_unlock(&fs_info->balance_lock); 4470 4471 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4472 return PTR_ERR_OR_ZERO(tsk); 4473 } 4474 4475 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4476 { 4477 struct btrfs_balance_control *bctl; 4478 struct btrfs_balance_item *item; 4479 struct btrfs_disk_balance_args disk_bargs; 4480 struct btrfs_path *path; 4481 struct extent_buffer *leaf; 4482 struct btrfs_key key; 4483 int ret; 4484 4485 path = btrfs_alloc_path(); 4486 if (!path) 4487 return -ENOMEM; 4488 4489 key.objectid = BTRFS_BALANCE_OBJECTID; 4490 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4491 key.offset = 0; 4492 4493 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4494 if (ret < 0) 4495 goto out; 4496 if (ret > 0) { /* ret = -ENOENT; */ 4497 ret = 0; 4498 goto out; 4499 } 4500 4501 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4502 if (!bctl) { 4503 ret = -ENOMEM; 4504 goto out; 4505 } 4506 4507 leaf = path->nodes[0]; 4508 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4509 4510 bctl->flags = btrfs_balance_flags(leaf, item); 4511 bctl->flags |= BTRFS_BALANCE_RESUME; 4512 4513 btrfs_balance_data(leaf, item, &disk_bargs); 4514 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4515 btrfs_balance_meta(leaf, item, &disk_bargs); 4516 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4517 btrfs_balance_sys(leaf, item, &disk_bargs); 4518 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4519 4520 /* 4521 * This should never happen, as the paused balance state is recovered 4522 * during mount without any chance of other exclusive ops to collide. 4523 * 4524 * This gives the exclusive op status to balance and keeps in paused 4525 * state until user intervention (cancel or umount). If the ownership 4526 * cannot be assigned, show a message but do not fail. The balance 4527 * is in a paused state and must have fs_info::balance_ctl properly 4528 * set up. 4529 */ 4530 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) 4531 btrfs_warn(fs_info, 4532 "balance: cannot set exclusive op status, resume manually"); 4533 4534 btrfs_release_path(path); 4535 4536 mutex_lock(&fs_info->balance_mutex); 4537 BUG_ON(fs_info->balance_ctl); 4538 spin_lock(&fs_info->balance_lock); 4539 fs_info->balance_ctl = bctl; 4540 spin_unlock(&fs_info->balance_lock); 4541 mutex_unlock(&fs_info->balance_mutex); 4542 out: 4543 btrfs_free_path(path); 4544 return ret; 4545 } 4546 4547 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4548 { 4549 int ret = 0; 4550 4551 mutex_lock(&fs_info->balance_mutex); 4552 if (!fs_info->balance_ctl) { 4553 mutex_unlock(&fs_info->balance_mutex); 4554 return -ENOTCONN; 4555 } 4556 4557 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4558 atomic_inc(&fs_info->balance_pause_req); 4559 mutex_unlock(&fs_info->balance_mutex); 4560 4561 wait_event(fs_info->balance_wait_q, 4562 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4563 4564 mutex_lock(&fs_info->balance_mutex); 4565 /* we are good with balance_ctl ripped off from under us */ 4566 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4567 atomic_dec(&fs_info->balance_pause_req); 4568 } else { 4569 ret = -ENOTCONN; 4570 } 4571 4572 mutex_unlock(&fs_info->balance_mutex); 4573 return ret; 4574 } 4575 4576 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4577 { 4578 mutex_lock(&fs_info->balance_mutex); 4579 if (!fs_info->balance_ctl) { 4580 mutex_unlock(&fs_info->balance_mutex); 4581 return -ENOTCONN; 4582 } 4583 4584 /* 4585 * A paused balance with the item stored on disk can be resumed at 4586 * mount time if the mount is read-write. Otherwise it's still paused 4587 * and we must not allow cancelling as it deletes the item. 4588 */ 4589 if (sb_rdonly(fs_info->sb)) { 4590 mutex_unlock(&fs_info->balance_mutex); 4591 return -EROFS; 4592 } 4593 4594 atomic_inc(&fs_info->balance_cancel_req); 4595 /* 4596 * if we are running just wait and return, balance item is 4597 * deleted in btrfs_balance in this case 4598 */ 4599 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4600 mutex_unlock(&fs_info->balance_mutex); 4601 wait_event(fs_info->balance_wait_q, 4602 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4603 mutex_lock(&fs_info->balance_mutex); 4604 } else { 4605 mutex_unlock(&fs_info->balance_mutex); 4606 /* 4607 * Lock released to allow other waiters to continue, we'll 4608 * reexamine the status again. 4609 */ 4610 mutex_lock(&fs_info->balance_mutex); 4611 4612 if (fs_info->balance_ctl) { 4613 reset_balance_state(fs_info); 4614 btrfs_exclop_finish(fs_info); 4615 btrfs_info(fs_info, "balance: canceled"); 4616 } 4617 } 4618 4619 BUG_ON(fs_info->balance_ctl || 4620 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4621 atomic_dec(&fs_info->balance_cancel_req); 4622 mutex_unlock(&fs_info->balance_mutex); 4623 return 0; 4624 } 4625 4626 int btrfs_uuid_scan_kthread(void *data) 4627 { 4628 struct btrfs_fs_info *fs_info = data; 4629 struct btrfs_root *root = fs_info->tree_root; 4630 struct btrfs_key key; 4631 struct btrfs_path *path = NULL; 4632 int ret = 0; 4633 struct extent_buffer *eb; 4634 int slot; 4635 struct btrfs_root_item root_item; 4636 u32 item_size; 4637 struct btrfs_trans_handle *trans = NULL; 4638 bool closing = false; 4639 4640 path = btrfs_alloc_path(); 4641 if (!path) { 4642 ret = -ENOMEM; 4643 goto out; 4644 } 4645 4646 key.objectid = 0; 4647 key.type = BTRFS_ROOT_ITEM_KEY; 4648 key.offset = 0; 4649 4650 while (1) { 4651 if (btrfs_fs_closing(fs_info)) { 4652 closing = true; 4653 break; 4654 } 4655 ret = btrfs_search_forward(root, &key, path, 4656 BTRFS_OLDEST_GENERATION); 4657 if (ret) { 4658 if (ret > 0) 4659 ret = 0; 4660 break; 4661 } 4662 4663 if (key.type != BTRFS_ROOT_ITEM_KEY || 4664 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4665 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4666 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4667 goto skip; 4668 4669 eb = path->nodes[0]; 4670 slot = path->slots[0]; 4671 item_size = btrfs_item_size(eb, slot); 4672 if (item_size < sizeof(root_item)) 4673 goto skip; 4674 4675 read_extent_buffer(eb, &root_item, 4676 btrfs_item_ptr_offset(eb, slot), 4677 (int)sizeof(root_item)); 4678 if (btrfs_root_refs(&root_item) == 0) 4679 goto skip; 4680 4681 if (!btrfs_is_empty_uuid(root_item.uuid) || 4682 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4683 if (trans) 4684 goto update_tree; 4685 4686 btrfs_release_path(path); 4687 /* 4688 * 1 - subvol uuid item 4689 * 1 - received_subvol uuid item 4690 */ 4691 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4692 if (IS_ERR(trans)) { 4693 ret = PTR_ERR(trans); 4694 break; 4695 } 4696 continue; 4697 } else { 4698 goto skip; 4699 } 4700 update_tree: 4701 btrfs_release_path(path); 4702 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4703 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4704 BTRFS_UUID_KEY_SUBVOL, 4705 key.objectid); 4706 if (ret < 0) { 4707 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4708 ret); 4709 break; 4710 } 4711 } 4712 4713 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4714 ret = btrfs_uuid_tree_add(trans, 4715 root_item.received_uuid, 4716 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4717 key.objectid); 4718 if (ret < 0) { 4719 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4720 ret); 4721 break; 4722 } 4723 } 4724 4725 skip: 4726 btrfs_release_path(path); 4727 if (trans) { 4728 ret = btrfs_end_transaction(trans); 4729 trans = NULL; 4730 if (ret) 4731 break; 4732 } 4733 4734 if (key.offset < (u64)-1) { 4735 key.offset++; 4736 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4737 key.offset = 0; 4738 key.type = BTRFS_ROOT_ITEM_KEY; 4739 } else if (key.objectid < (u64)-1) { 4740 key.offset = 0; 4741 key.type = BTRFS_ROOT_ITEM_KEY; 4742 key.objectid++; 4743 } else { 4744 break; 4745 } 4746 cond_resched(); 4747 } 4748 4749 out: 4750 btrfs_free_path(path); 4751 if (trans && !IS_ERR(trans)) 4752 btrfs_end_transaction(trans); 4753 if (ret) 4754 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4755 else if (!closing) 4756 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4757 up(&fs_info->uuid_tree_rescan_sem); 4758 return 0; 4759 } 4760 4761 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4762 { 4763 struct btrfs_trans_handle *trans; 4764 struct btrfs_root *tree_root = fs_info->tree_root; 4765 struct btrfs_root *uuid_root; 4766 struct task_struct *task; 4767 int ret; 4768 4769 /* 4770 * 1 - root node 4771 * 1 - root item 4772 */ 4773 trans = btrfs_start_transaction(tree_root, 2); 4774 if (IS_ERR(trans)) 4775 return PTR_ERR(trans); 4776 4777 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4778 if (IS_ERR(uuid_root)) { 4779 ret = PTR_ERR(uuid_root); 4780 btrfs_abort_transaction(trans, ret); 4781 btrfs_end_transaction(trans); 4782 return ret; 4783 } 4784 4785 fs_info->uuid_root = uuid_root; 4786 4787 ret = btrfs_commit_transaction(trans); 4788 if (ret) 4789 return ret; 4790 4791 down(&fs_info->uuid_tree_rescan_sem); 4792 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4793 if (IS_ERR(task)) { 4794 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4795 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4796 up(&fs_info->uuid_tree_rescan_sem); 4797 return PTR_ERR(task); 4798 } 4799 4800 return 0; 4801 } 4802 4803 /* 4804 * shrinking a device means finding all of the device extents past 4805 * the new size, and then following the back refs to the chunks. 4806 * The chunk relocation code actually frees the device extent 4807 */ 4808 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4809 { 4810 struct btrfs_fs_info *fs_info = device->fs_info; 4811 struct btrfs_root *root = fs_info->dev_root; 4812 struct btrfs_trans_handle *trans; 4813 struct btrfs_dev_extent *dev_extent = NULL; 4814 struct btrfs_path *path; 4815 u64 length; 4816 u64 chunk_offset; 4817 int ret; 4818 int slot; 4819 int failed = 0; 4820 bool retried = false; 4821 struct extent_buffer *l; 4822 struct btrfs_key key; 4823 struct btrfs_super_block *super_copy = fs_info->super_copy; 4824 u64 old_total = btrfs_super_total_bytes(super_copy); 4825 u64 old_size = btrfs_device_get_total_bytes(device); 4826 u64 diff; 4827 u64 start; 4828 4829 new_size = round_down(new_size, fs_info->sectorsize); 4830 start = new_size; 4831 diff = round_down(old_size - new_size, fs_info->sectorsize); 4832 4833 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4834 return -EINVAL; 4835 4836 path = btrfs_alloc_path(); 4837 if (!path) 4838 return -ENOMEM; 4839 4840 path->reada = READA_BACK; 4841 4842 trans = btrfs_start_transaction(root, 0); 4843 if (IS_ERR(trans)) { 4844 btrfs_free_path(path); 4845 return PTR_ERR(trans); 4846 } 4847 4848 mutex_lock(&fs_info->chunk_mutex); 4849 4850 btrfs_device_set_total_bytes(device, new_size); 4851 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4852 device->fs_devices->total_rw_bytes -= diff; 4853 atomic64_sub(diff, &fs_info->free_chunk_space); 4854 } 4855 4856 /* 4857 * Once the device's size has been set to the new size, ensure all 4858 * in-memory chunks are synced to disk so that the loop below sees them 4859 * and relocates them accordingly. 4860 */ 4861 if (contains_pending_extent(device, &start, diff)) { 4862 mutex_unlock(&fs_info->chunk_mutex); 4863 ret = btrfs_commit_transaction(trans); 4864 if (ret) 4865 goto done; 4866 } else { 4867 mutex_unlock(&fs_info->chunk_mutex); 4868 btrfs_end_transaction(trans); 4869 } 4870 4871 again: 4872 key.objectid = device->devid; 4873 key.offset = (u64)-1; 4874 key.type = BTRFS_DEV_EXTENT_KEY; 4875 4876 do { 4877 mutex_lock(&fs_info->reclaim_bgs_lock); 4878 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4879 if (ret < 0) { 4880 mutex_unlock(&fs_info->reclaim_bgs_lock); 4881 goto done; 4882 } 4883 4884 ret = btrfs_previous_item(root, path, 0, key.type); 4885 if (ret) { 4886 mutex_unlock(&fs_info->reclaim_bgs_lock); 4887 if (ret < 0) 4888 goto done; 4889 ret = 0; 4890 btrfs_release_path(path); 4891 break; 4892 } 4893 4894 l = path->nodes[0]; 4895 slot = path->slots[0]; 4896 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4897 4898 if (key.objectid != device->devid) { 4899 mutex_unlock(&fs_info->reclaim_bgs_lock); 4900 btrfs_release_path(path); 4901 break; 4902 } 4903 4904 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4905 length = btrfs_dev_extent_length(l, dev_extent); 4906 4907 if (key.offset + length <= new_size) { 4908 mutex_unlock(&fs_info->reclaim_bgs_lock); 4909 btrfs_release_path(path); 4910 break; 4911 } 4912 4913 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4914 btrfs_release_path(path); 4915 4916 /* 4917 * We may be relocating the only data chunk we have, 4918 * which could potentially end up with losing data's 4919 * raid profile, so lets allocate an empty one in 4920 * advance. 4921 */ 4922 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4923 if (ret < 0) { 4924 mutex_unlock(&fs_info->reclaim_bgs_lock); 4925 goto done; 4926 } 4927 4928 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4929 mutex_unlock(&fs_info->reclaim_bgs_lock); 4930 if (ret == -ENOSPC) { 4931 failed++; 4932 } else if (ret) { 4933 if (ret == -ETXTBSY) { 4934 btrfs_warn(fs_info, 4935 "could not shrink block group %llu due to active swapfile", 4936 chunk_offset); 4937 } 4938 goto done; 4939 } 4940 } while (key.offset-- > 0); 4941 4942 if (failed && !retried) { 4943 failed = 0; 4944 retried = true; 4945 goto again; 4946 } else if (failed && retried) { 4947 ret = -ENOSPC; 4948 goto done; 4949 } 4950 4951 /* Shrinking succeeded, else we would be at "done". */ 4952 trans = btrfs_start_transaction(root, 0); 4953 if (IS_ERR(trans)) { 4954 ret = PTR_ERR(trans); 4955 goto done; 4956 } 4957 4958 mutex_lock(&fs_info->chunk_mutex); 4959 /* Clear all state bits beyond the shrunk device size */ 4960 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4961 CHUNK_STATE_MASK); 4962 4963 btrfs_device_set_disk_total_bytes(device, new_size); 4964 if (list_empty(&device->post_commit_list)) 4965 list_add_tail(&device->post_commit_list, 4966 &trans->transaction->dev_update_list); 4967 4968 WARN_ON(diff > old_total); 4969 btrfs_set_super_total_bytes(super_copy, 4970 round_down(old_total - diff, fs_info->sectorsize)); 4971 mutex_unlock(&fs_info->chunk_mutex); 4972 4973 btrfs_reserve_chunk_metadata(trans, false); 4974 /* Now btrfs_update_device() will change the on-disk size. */ 4975 ret = btrfs_update_device(trans, device); 4976 btrfs_trans_release_chunk_metadata(trans); 4977 if (ret < 0) { 4978 btrfs_abort_transaction(trans, ret); 4979 btrfs_end_transaction(trans); 4980 } else { 4981 ret = btrfs_commit_transaction(trans); 4982 } 4983 done: 4984 btrfs_free_path(path); 4985 if (ret) { 4986 mutex_lock(&fs_info->chunk_mutex); 4987 btrfs_device_set_total_bytes(device, old_size); 4988 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 4989 device->fs_devices->total_rw_bytes += diff; 4990 atomic64_add(diff, &fs_info->free_chunk_space); 4991 mutex_unlock(&fs_info->chunk_mutex); 4992 } 4993 return ret; 4994 } 4995 4996 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 4997 struct btrfs_key *key, 4998 struct btrfs_chunk *chunk, int item_size) 4999 { 5000 struct btrfs_super_block *super_copy = fs_info->super_copy; 5001 struct btrfs_disk_key disk_key; 5002 u32 array_size; 5003 u8 *ptr; 5004 5005 lockdep_assert_held(&fs_info->chunk_mutex); 5006 5007 array_size = btrfs_super_sys_array_size(super_copy); 5008 if (array_size + item_size + sizeof(disk_key) 5009 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 5010 return -EFBIG; 5011 5012 ptr = super_copy->sys_chunk_array + array_size; 5013 btrfs_cpu_key_to_disk(&disk_key, key); 5014 memcpy(ptr, &disk_key, sizeof(disk_key)); 5015 ptr += sizeof(disk_key); 5016 memcpy(ptr, chunk, item_size); 5017 item_size += sizeof(disk_key); 5018 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 5019 5020 return 0; 5021 } 5022 5023 /* 5024 * sort the devices in descending order by max_avail, total_avail 5025 */ 5026 static int btrfs_cmp_device_info(const void *a, const void *b) 5027 { 5028 const struct btrfs_device_info *di_a = a; 5029 const struct btrfs_device_info *di_b = b; 5030 5031 if (di_a->max_avail > di_b->max_avail) 5032 return -1; 5033 if (di_a->max_avail < di_b->max_avail) 5034 return 1; 5035 if (di_a->total_avail > di_b->total_avail) 5036 return -1; 5037 if (di_a->total_avail < di_b->total_avail) 5038 return 1; 5039 return 0; 5040 } 5041 5042 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 5043 { 5044 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 5045 return; 5046 5047 btrfs_set_fs_incompat(info, RAID56); 5048 } 5049 5050 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5051 { 5052 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5053 return; 5054 5055 btrfs_set_fs_incompat(info, RAID1C34); 5056 } 5057 5058 /* 5059 * Structure used internally for btrfs_create_chunk() function. 5060 * Wraps needed parameters. 5061 */ 5062 struct alloc_chunk_ctl { 5063 u64 start; 5064 u64 type; 5065 /* Total number of stripes to allocate */ 5066 int num_stripes; 5067 /* sub_stripes info for map */ 5068 int sub_stripes; 5069 /* Stripes per device */ 5070 int dev_stripes; 5071 /* Maximum number of devices to use */ 5072 int devs_max; 5073 /* Minimum number of devices to use */ 5074 int devs_min; 5075 /* ndevs has to be a multiple of this */ 5076 int devs_increment; 5077 /* Number of copies */ 5078 int ncopies; 5079 /* Number of stripes worth of bytes to store parity information */ 5080 int nparity; 5081 u64 max_stripe_size; 5082 u64 max_chunk_size; 5083 u64 dev_extent_min; 5084 u64 stripe_size; 5085 u64 chunk_size; 5086 int ndevs; 5087 }; 5088 5089 static void init_alloc_chunk_ctl_policy_regular( 5090 struct btrfs_fs_devices *fs_devices, 5091 struct alloc_chunk_ctl *ctl) 5092 { 5093 u64 type = ctl->type; 5094 5095 if (type & BTRFS_BLOCK_GROUP_DATA) { 5096 ctl->max_stripe_size = SZ_1G; 5097 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; 5098 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5099 /* For larger filesystems, use larger metadata chunks */ 5100 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) 5101 ctl->max_stripe_size = SZ_1G; 5102 else 5103 ctl->max_stripe_size = SZ_256M; 5104 ctl->max_chunk_size = ctl->max_stripe_size; 5105 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5106 ctl->max_stripe_size = SZ_32M; 5107 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5108 ctl->devs_max = min_t(int, ctl->devs_max, 5109 BTRFS_MAX_DEVS_SYS_CHUNK); 5110 } else { 5111 BUG(); 5112 } 5113 5114 /* We don't want a chunk larger than 10% of writable space */ 5115 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5116 ctl->max_chunk_size); 5117 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5118 } 5119 5120 static void init_alloc_chunk_ctl_policy_zoned( 5121 struct btrfs_fs_devices *fs_devices, 5122 struct alloc_chunk_ctl *ctl) 5123 { 5124 u64 zone_size = fs_devices->fs_info->zone_size; 5125 u64 limit; 5126 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5127 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5128 u64 min_chunk_size = min_data_stripes * zone_size; 5129 u64 type = ctl->type; 5130 5131 ctl->max_stripe_size = zone_size; 5132 if (type & BTRFS_BLOCK_GROUP_DATA) { 5133 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5134 zone_size); 5135 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5136 ctl->max_chunk_size = ctl->max_stripe_size; 5137 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5138 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5139 ctl->devs_max = min_t(int, ctl->devs_max, 5140 BTRFS_MAX_DEVS_SYS_CHUNK); 5141 } else { 5142 BUG(); 5143 } 5144 5145 /* We don't want a chunk larger than 10% of writable space */ 5146 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5147 zone_size), 5148 min_chunk_size); 5149 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5150 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5151 } 5152 5153 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5154 struct alloc_chunk_ctl *ctl) 5155 { 5156 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5157 5158 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5159 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5160 ctl->devs_max = btrfs_raid_array[index].devs_max; 5161 if (!ctl->devs_max) 5162 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5163 ctl->devs_min = btrfs_raid_array[index].devs_min; 5164 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5165 ctl->ncopies = btrfs_raid_array[index].ncopies; 5166 ctl->nparity = btrfs_raid_array[index].nparity; 5167 ctl->ndevs = 0; 5168 5169 switch (fs_devices->chunk_alloc_policy) { 5170 case BTRFS_CHUNK_ALLOC_REGULAR: 5171 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5172 break; 5173 case BTRFS_CHUNK_ALLOC_ZONED: 5174 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5175 break; 5176 default: 5177 BUG(); 5178 } 5179 } 5180 5181 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5182 struct alloc_chunk_ctl *ctl, 5183 struct btrfs_device_info *devices_info) 5184 { 5185 struct btrfs_fs_info *info = fs_devices->fs_info; 5186 struct btrfs_device *device; 5187 u64 total_avail; 5188 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5189 int ret; 5190 int ndevs = 0; 5191 u64 max_avail; 5192 u64 dev_offset; 5193 5194 /* 5195 * in the first pass through the devices list, we gather information 5196 * about the available holes on each device. 5197 */ 5198 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5199 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5200 WARN(1, KERN_ERR 5201 "BTRFS: read-only device in alloc_list\n"); 5202 continue; 5203 } 5204 5205 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5206 &device->dev_state) || 5207 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5208 continue; 5209 5210 if (device->total_bytes > device->bytes_used) 5211 total_avail = device->total_bytes - device->bytes_used; 5212 else 5213 total_avail = 0; 5214 5215 /* If there is no space on this device, skip it. */ 5216 if (total_avail < ctl->dev_extent_min) 5217 continue; 5218 5219 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5220 &max_avail); 5221 if (ret && ret != -ENOSPC) 5222 return ret; 5223 5224 if (ret == 0) 5225 max_avail = dev_extent_want; 5226 5227 if (max_avail < ctl->dev_extent_min) { 5228 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5229 btrfs_debug(info, 5230 "%s: devid %llu has no free space, have=%llu want=%llu", 5231 __func__, device->devid, max_avail, 5232 ctl->dev_extent_min); 5233 continue; 5234 } 5235 5236 if (ndevs == fs_devices->rw_devices) { 5237 WARN(1, "%s: found more than %llu devices\n", 5238 __func__, fs_devices->rw_devices); 5239 break; 5240 } 5241 devices_info[ndevs].dev_offset = dev_offset; 5242 devices_info[ndevs].max_avail = max_avail; 5243 devices_info[ndevs].total_avail = total_avail; 5244 devices_info[ndevs].dev = device; 5245 ++ndevs; 5246 } 5247 ctl->ndevs = ndevs; 5248 5249 /* 5250 * now sort the devices by hole size / available space 5251 */ 5252 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5253 btrfs_cmp_device_info, NULL); 5254 5255 return 0; 5256 } 5257 5258 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5259 struct btrfs_device_info *devices_info) 5260 { 5261 /* Number of stripes that count for block group size */ 5262 int data_stripes; 5263 5264 /* 5265 * The primary goal is to maximize the number of stripes, so use as 5266 * many devices as possible, even if the stripes are not maximum sized. 5267 * 5268 * The DUP profile stores more than one stripe per device, the 5269 * max_avail is the total size so we have to adjust. 5270 */ 5271 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5272 ctl->dev_stripes); 5273 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5274 5275 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5276 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5277 5278 /* 5279 * Use the number of data stripes to figure out how big this chunk is 5280 * really going to be in terms of logical address space, and compare 5281 * that answer with the max chunk size. If it's higher, we try to 5282 * reduce stripe_size. 5283 */ 5284 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5285 /* 5286 * Reduce stripe_size, round it up to a 16MB boundary again and 5287 * then use it, unless it ends up being even bigger than the 5288 * previous value we had already. 5289 */ 5290 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5291 data_stripes), SZ_16M), 5292 ctl->stripe_size); 5293 } 5294 5295 /* Align to BTRFS_STRIPE_LEN */ 5296 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5297 ctl->chunk_size = ctl->stripe_size * data_stripes; 5298 5299 return 0; 5300 } 5301 5302 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5303 struct btrfs_device_info *devices_info) 5304 { 5305 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5306 /* Number of stripes that count for block group size */ 5307 int data_stripes; 5308 5309 /* 5310 * It should hold because: 5311 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5312 */ 5313 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5314 5315 ctl->stripe_size = zone_size; 5316 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5317 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5318 5319 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5320 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5321 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5322 ctl->stripe_size) + ctl->nparity, 5323 ctl->dev_stripes); 5324 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5325 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5326 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5327 } 5328 5329 ctl->chunk_size = ctl->stripe_size * data_stripes; 5330 5331 return 0; 5332 } 5333 5334 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5335 struct alloc_chunk_ctl *ctl, 5336 struct btrfs_device_info *devices_info) 5337 { 5338 struct btrfs_fs_info *info = fs_devices->fs_info; 5339 5340 /* 5341 * Round down to number of usable stripes, devs_increment can be any 5342 * number so we can't use round_down() that requires power of 2, while 5343 * rounddown is safe. 5344 */ 5345 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5346 5347 if (ctl->ndevs < ctl->devs_min) { 5348 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5349 btrfs_debug(info, 5350 "%s: not enough devices with free space: have=%d minimum required=%d", 5351 __func__, ctl->ndevs, ctl->devs_min); 5352 } 5353 return -ENOSPC; 5354 } 5355 5356 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5357 5358 switch (fs_devices->chunk_alloc_policy) { 5359 case BTRFS_CHUNK_ALLOC_REGULAR: 5360 return decide_stripe_size_regular(ctl, devices_info); 5361 case BTRFS_CHUNK_ALLOC_ZONED: 5362 return decide_stripe_size_zoned(ctl, devices_info); 5363 default: 5364 BUG(); 5365 } 5366 } 5367 5368 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5369 struct alloc_chunk_ctl *ctl, 5370 struct btrfs_device_info *devices_info) 5371 { 5372 struct btrfs_fs_info *info = trans->fs_info; 5373 struct map_lookup *map = NULL; 5374 struct extent_map_tree *em_tree; 5375 struct btrfs_block_group *block_group; 5376 struct extent_map *em; 5377 u64 start = ctl->start; 5378 u64 type = ctl->type; 5379 int ret; 5380 int i; 5381 int j; 5382 5383 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5384 if (!map) 5385 return ERR_PTR(-ENOMEM); 5386 map->num_stripes = ctl->num_stripes; 5387 5388 for (i = 0; i < ctl->ndevs; ++i) { 5389 for (j = 0; j < ctl->dev_stripes; ++j) { 5390 int s = i * ctl->dev_stripes + j; 5391 map->stripes[s].dev = devices_info[i].dev; 5392 map->stripes[s].physical = devices_info[i].dev_offset + 5393 j * ctl->stripe_size; 5394 } 5395 } 5396 map->stripe_len = BTRFS_STRIPE_LEN; 5397 map->io_align = BTRFS_STRIPE_LEN; 5398 map->io_width = BTRFS_STRIPE_LEN; 5399 map->type = type; 5400 map->sub_stripes = ctl->sub_stripes; 5401 5402 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5403 5404 em = alloc_extent_map(); 5405 if (!em) { 5406 kfree(map); 5407 return ERR_PTR(-ENOMEM); 5408 } 5409 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5410 em->map_lookup = map; 5411 em->start = start; 5412 em->len = ctl->chunk_size; 5413 em->block_start = 0; 5414 em->block_len = em->len; 5415 em->orig_block_len = ctl->stripe_size; 5416 5417 em_tree = &info->mapping_tree; 5418 write_lock(&em_tree->lock); 5419 ret = add_extent_mapping(em_tree, em, 0); 5420 if (ret) { 5421 write_unlock(&em_tree->lock); 5422 free_extent_map(em); 5423 return ERR_PTR(ret); 5424 } 5425 write_unlock(&em_tree->lock); 5426 5427 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5428 if (IS_ERR(block_group)) 5429 goto error_del_extent; 5430 5431 for (i = 0; i < map->num_stripes; i++) { 5432 struct btrfs_device *dev = map->stripes[i].dev; 5433 5434 btrfs_device_set_bytes_used(dev, 5435 dev->bytes_used + ctl->stripe_size); 5436 if (list_empty(&dev->post_commit_list)) 5437 list_add_tail(&dev->post_commit_list, 5438 &trans->transaction->dev_update_list); 5439 } 5440 5441 atomic64_sub(ctl->stripe_size * map->num_stripes, 5442 &info->free_chunk_space); 5443 5444 free_extent_map(em); 5445 check_raid56_incompat_flag(info, type); 5446 check_raid1c34_incompat_flag(info, type); 5447 5448 return block_group; 5449 5450 error_del_extent: 5451 write_lock(&em_tree->lock); 5452 remove_extent_mapping(em_tree, em); 5453 write_unlock(&em_tree->lock); 5454 5455 /* One for our allocation */ 5456 free_extent_map(em); 5457 /* One for the tree reference */ 5458 free_extent_map(em); 5459 5460 return block_group; 5461 } 5462 5463 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5464 u64 type) 5465 { 5466 struct btrfs_fs_info *info = trans->fs_info; 5467 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5468 struct btrfs_device_info *devices_info = NULL; 5469 struct alloc_chunk_ctl ctl; 5470 struct btrfs_block_group *block_group; 5471 int ret; 5472 5473 lockdep_assert_held(&info->chunk_mutex); 5474 5475 if (!alloc_profile_is_valid(type, 0)) { 5476 ASSERT(0); 5477 return ERR_PTR(-EINVAL); 5478 } 5479 5480 if (list_empty(&fs_devices->alloc_list)) { 5481 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5482 btrfs_debug(info, "%s: no writable device", __func__); 5483 return ERR_PTR(-ENOSPC); 5484 } 5485 5486 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5487 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5488 ASSERT(0); 5489 return ERR_PTR(-EINVAL); 5490 } 5491 5492 ctl.start = find_next_chunk(info); 5493 ctl.type = type; 5494 init_alloc_chunk_ctl(fs_devices, &ctl); 5495 5496 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5497 GFP_NOFS); 5498 if (!devices_info) 5499 return ERR_PTR(-ENOMEM); 5500 5501 ret = gather_device_info(fs_devices, &ctl, devices_info); 5502 if (ret < 0) { 5503 block_group = ERR_PTR(ret); 5504 goto out; 5505 } 5506 5507 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5508 if (ret < 0) { 5509 block_group = ERR_PTR(ret); 5510 goto out; 5511 } 5512 5513 block_group = create_chunk(trans, &ctl, devices_info); 5514 5515 out: 5516 kfree(devices_info); 5517 return block_group; 5518 } 5519 5520 /* 5521 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5522 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5523 * chunks. 5524 * 5525 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5526 * phases. 5527 */ 5528 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5529 struct btrfs_block_group *bg) 5530 { 5531 struct btrfs_fs_info *fs_info = trans->fs_info; 5532 struct btrfs_root *chunk_root = fs_info->chunk_root; 5533 struct btrfs_key key; 5534 struct btrfs_chunk *chunk; 5535 struct btrfs_stripe *stripe; 5536 struct extent_map *em; 5537 struct map_lookup *map; 5538 size_t item_size; 5539 int i; 5540 int ret; 5541 5542 /* 5543 * We take the chunk_mutex for 2 reasons: 5544 * 5545 * 1) Updates and insertions in the chunk btree must be done while holding 5546 * the chunk_mutex, as well as updating the system chunk array in the 5547 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5548 * details; 5549 * 5550 * 2) To prevent races with the final phase of a device replace operation 5551 * that replaces the device object associated with the map's stripes, 5552 * because the device object's id can change at any time during that 5553 * final phase of the device replace operation 5554 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5555 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5556 * which would cause a failure when updating the device item, which does 5557 * not exists, or persisting a stripe of the chunk item with such ID. 5558 * Here we can't use the device_list_mutex because our caller already 5559 * has locked the chunk_mutex, and the final phase of device replace 5560 * acquires both mutexes - first the device_list_mutex and then the 5561 * chunk_mutex. Using any of those two mutexes protects us from a 5562 * concurrent device replace. 5563 */ 5564 lockdep_assert_held(&fs_info->chunk_mutex); 5565 5566 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5567 if (IS_ERR(em)) { 5568 ret = PTR_ERR(em); 5569 btrfs_abort_transaction(trans, ret); 5570 return ret; 5571 } 5572 5573 map = em->map_lookup; 5574 item_size = btrfs_chunk_item_size(map->num_stripes); 5575 5576 chunk = kzalloc(item_size, GFP_NOFS); 5577 if (!chunk) { 5578 ret = -ENOMEM; 5579 btrfs_abort_transaction(trans, ret); 5580 goto out; 5581 } 5582 5583 for (i = 0; i < map->num_stripes; i++) { 5584 struct btrfs_device *device = map->stripes[i].dev; 5585 5586 ret = btrfs_update_device(trans, device); 5587 if (ret) 5588 goto out; 5589 } 5590 5591 stripe = &chunk->stripe; 5592 for (i = 0; i < map->num_stripes; i++) { 5593 struct btrfs_device *device = map->stripes[i].dev; 5594 const u64 dev_offset = map->stripes[i].physical; 5595 5596 btrfs_set_stack_stripe_devid(stripe, device->devid); 5597 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5598 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5599 stripe++; 5600 } 5601 5602 btrfs_set_stack_chunk_length(chunk, bg->length); 5603 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); 5604 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5605 btrfs_set_stack_chunk_type(chunk, map->type); 5606 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5607 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5608 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5609 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5610 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5611 5612 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5613 key.type = BTRFS_CHUNK_ITEM_KEY; 5614 key.offset = bg->start; 5615 5616 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5617 if (ret) 5618 goto out; 5619 5620 bg->chunk_item_inserted = 1; 5621 5622 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5623 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5624 if (ret) 5625 goto out; 5626 } 5627 5628 out: 5629 kfree(chunk); 5630 free_extent_map(em); 5631 return ret; 5632 } 5633 5634 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5635 { 5636 struct btrfs_fs_info *fs_info = trans->fs_info; 5637 u64 alloc_profile; 5638 struct btrfs_block_group *meta_bg; 5639 struct btrfs_block_group *sys_bg; 5640 5641 /* 5642 * When adding a new device for sprouting, the seed device is read-only 5643 * so we must first allocate a metadata and a system chunk. But before 5644 * adding the block group items to the extent, device and chunk btrees, 5645 * we must first: 5646 * 5647 * 1) Create both chunks without doing any changes to the btrees, as 5648 * otherwise we would get -ENOSPC since the block groups from the 5649 * seed device are read-only; 5650 * 5651 * 2) Add the device item for the new sprout device - finishing the setup 5652 * of a new block group requires updating the device item in the chunk 5653 * btree, so it must exist when we attempt to do it. The previous step 5654 * ensures this does not fail with -ENOSPC. 5655 * 5656 * After that we can add the block group items to their btrees: 5657 * update existing device item in the chunk btree, add a new block group 5658 * item to the extent btree, add a new chunk item to the chunk btree and 5659 * finally add the new device extent items to the devices btree. 5660 */ 5661 5662 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5663 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5664 if (IS_ERR(meta_bg)) 5665 return PTR_ERR(meta_bg); 5666 5667 alloc_profile = btrfs_system_alloc_profile(fs_info); 5668 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5669 if (IS_ERR(sys_bg)) 5670 return PTR_ERR(sys_bg); 5671 5672 return 0; 5673 } 5674 5675 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5676 { 5677 const int index = btrfs_bg_flags_to_raid_index(map->type); 5678 5679 return btrfs_raid_array[index].tolerated_failures; 5680 } 5681 5682 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5683 { 5684 struct extent_map *em; 5685 struct map_lookup *map; 5686 int miss_ndevs = 0; 5687 int i; 5688 bool ret = true; 5689 5690 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5691 if (IS_ERR(em)) 5692 return false; 5693 5694 map = em->map_lookup; 5695 for (i = 0; i < map->num_stripes; i++) { 5696 if (test_bit(BTRFS_DEV_STATE_MISSING, 5697 &map->stripes[i].dev->dev_state)) { 5698 miss_ndevs++; 5699 continue; 5700 } 5701 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5702 &map->stripes[i].dev->dev_state)) { 5703 ret = false; 5704 goto end; 5705 } 5706 } 5707 5708 /* 5709 * If the number of missing devices is larger than max errors, we can 5710 * not write the data into that chunk successfully. 5711 */ 5712 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5713 ret = false; 5714 end: 5715 free_extent_map(em); 5716 return ret; 5717 } 5718 5719 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5720 { 5721 struct extent_map *em; 5722 5723 while (1) { 5724 write_lock(&tree->lock); 5725 em = lookup_extent_mapping(tree, 0, (u64)-1); 5726 if (em) 5727 remove_extent_mapping(tree, em); 5728 write_unlock(&tree->lock); 5729 if (!em) 5730 break; 5731 /* once for us */ 5732 free_extent_map(em); 5733 /* once for the tree */ 5734 free_extent_map(em); 5735 } 5736 } 5737 5738 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5739 { 5740 struct extent_map *em; 5741 struct map_lookup *map; 5742 int ret; 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 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) 5756 ret = map->num_stripes; 5757 else if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5758 ret = map->sub_stripes; 5759 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5760 ret = 2; 5761 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5762 /* 5763 * There could be two corrupted data stripes, we need 5764 * to loop retry in order to rebuild the correct data. 5765 * 5766 * Fail a stripe at a time on every retry except the 5767 * stripe under reconstruction. 5768 */ 5769 ret = map->num_stripes; 5770 else 5771 ret = 1; 5772 free_extent_map(em); 5773 5774 down_read(&fs_info->dev_replace.rwsem); 5775 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5776 fs_info->dev_replace.tgtdev) 5777 ret++; 5778 up_read(&fs_info->dev_replace.rwsem); 5779 5780 return ret; 5781 } 5782 5783 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5784 u64 logical) 5785 { 5786 struct extent_map *em; 5787 struct map_lookup *map; 5788 unsigned long len = fs_info->sectorsize; 5789 5790 em = btrfs_get_chunk_map(fs_info, logical, len); 5791 5792 if (!WARN_ON(IS_ERR(em))) { 5793 map = em->map_lookup; 5794 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5795 len = map->stripe_len * nr_data_stripes(map); 5796 free_extent_map(em); 5797 } 5798 return len; 5799 } 5800 5801 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5802 { 5803 struct extent_map *em; 5804 struct map_lookup *map; 5805 int ret = 0; 5806 5807 em = btrfs_get_chunk_map(fs_info, logical, len); 5808 5809 if(!WARN_ON(IS_ERR(em))) { 5810 map = em->map_lookup; 5811 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5812 ret = 1; 5813 free_extent_map(em); 5814 } 5815 return ret; 5816 } 5817 5818 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5819 struct map_lookup *map, int first, 5820 int dev_replace_is_ongoing) 5821 { 5822 int i; 5823 int num_stripes; 5824 int preferred_mirror; 5825 int tolerance; 5826 struct btrfs_device *srcdev; 5827 5828 ASSERT((map->type & 5829 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5830 5831 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5832 num_stripes = map->sub_stripes; 5833 else 5834 num_stripes = map->num_stripes; 5835 5836 switch (fs_info->fs_devices->read_policy) { 5837 default: 5838 /* Shouldn't happen, just warn and use pid instead of failing */ 5839 btrfs_warn_rl(fs_info, 5840 "unknown read_policy type %u, reset to pid", 5841 fs_info->fs_devices->read_policy); 5842 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5843 fallthrough; 5844 case BTRFS_READ_POLICY_PID: 5845 preferred_mirror = first + (current->pid % num_stripes); 5846 break; 5847 } 5848 5849 if (dev_replace_is_ongoing && 5850 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5851 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5852 srcdev = fs_info->dev_replace.srcdev; 5853 else 5854 srcdev = NULL; 5855 5856 /* 5857 * try to avoid the drive that is the source drive for a 5858 * dev-replace procedure, only choose it if no other non-missing 5859 * mirror is available 5860 */ 5861 for (tolerance = 0; tolerance < 2; tolerance++) { 5862 if (map->stripes[preferred_mirror].dev->bdev && 5863 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5864 return preferred_mirror; 5865 for (i = first; i < first + num_stripes; i++) { 5866 if (map->stripes[i].dev->bdev && 5867 (tolerance || map->stripes[i].dev != srcdev)) 5868 return i; 5869 } 5870 } 5871 5872 /* we couldn't find one that doesn't fail. Just return something 5873 * and the io error handling code will clean up eventually 5874 */ 5875 return preferred_mirror; 5876 } 5877 5878 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5879 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5880 { 5881 int i; 5882 int again = 1; 5883 5884 while (again) { 5885 again = 0; 5886 for (i = 0; i < num_stripes - 1; i++) { 5887 /* Swap if parity is on a smaller index */ 5888 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5889 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5890 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5891 again = 1; 5892 } 5893 } 5894 } 5895 } 5896 5897 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5898 int total_stripes, 5899 int real_stripes) 5900 { 5901 struct btrfs_io_context *bioc = kzalloc( 5902 /* The size of btrfs_io_context */ 5903 sizeof(struct btrfs_io_context) + 5904 /* Plus the variable array for the stripes */ 5905 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5906 /* Plus the variable array for the tgt dev */ 5907 sizeof(int) * (real_stripes) + 5908 /* 5909 * Plus the raid_map, which includes both the tgt dev 5910 * and the stripes. 5911 */ 5912 sizeof(u64) * (total_stripes), 5913 GFP_NOFS|__GFP_NOFAIL); 5914 5915 atomic_set(&bioc->error, 0); 5916 refcount_set(&bioc->refs, 1); 5917 5918 bioc->fs_info = fs_info; 5919 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5920 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5921 5922 return bioc; 5923 } 5924 5925 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5926 { 5927 WARN_ON(!refcount_read(&bioc->refs)); 5928 refcount_inc(&bioc->refs); 5929 } 5930 5931 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5932 { 5933 if (!bioc) 5934 return; 5935 if (refcount_dec_and_test(&bioc->refs)) 5936 kfree(bioc); 5937 } 5938 5939 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ 5940 /* 5941 * Please note that, discard won't be sent to target device of device 5942 * replace. 5943 */ 5944 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, 5945 u64 logical, u64 *length_ret, 5946 struct btrfs_io_context **bioc_ret) 5947 { 5948 struct extent_map *em; 5949 struct map_lookup *map; 5950 struct btrfs_io_context *bioc; 5951 u64 length = *length_ret; 5952 u64 offset; 5953 u64 stripe_nr; 5954 u64 stripe_nr_end; 5955 u64 stripe_end_offset; 5956 u64 stripe_cnt; 5957 u64 stripe_len; 5958 u64 stripe_offset; 5959 u64 num_stripes; 5960 u32 stripe_index; 5961 u32 factor = 0; 5962 u32 sub_stripes = 0; 5963 u64 stripes_per_dev = 0; 5964 u32 remaining_stripes = 0; 5965 u32 last_stripe = 0; 5966 int ret = 0; 5967 int i; 5968 5969 /* Discard always returns a bioc. */ 5970 ASSERT(bioc_ret); 5971 5972 em = btrfs_get_chunk_map(fs_info, logical, length); 5973 if (IS_ERR(em)) 5974 return PTR_ERR(em); 5975 5976 map = em->map_lookup; 5977 /* we don't discard raid56 yet */ 5978 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5979 ret = -EOPNOTSUPP; 5980 goto out; 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 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0); 6034 if (!bioc) { 6035 ret = -ENOMEM; 6036 goto out; 6037 } 6038 6039 for (i = 0; i < num_stripes; i++) { 6040 bioc->stripes[i].physical = 6041 map->stripes[stripe_index].physical + 6042 stripe_offset + stripe_nr * map->stripe_len; 6043 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 6044 6045 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6046 BTRFS_BLOCK_GROUP_RAID10)) { 6047 bioc->stripes[i].length = stripes_per_dev * 6048 map->stripe_len; 6049 6050 if (i / sub_stripes < remaining_stripes) 6051 bioc->stripes[i].length += map->stripe_len; 6052 6053 /* 6054 * Special for the first stripe and 6055 * the last stripe: 6056 * 6057 * |-------|...|-------| 6058 * |----------| 6059 * off end_off 6060 */ 6061 if (i < sub_stripes) 6062 bioc->stripes[i].length -= stripe_offset; 6063 6064 if (stripe_index >= last_stripe && 6065 stripe_index <= (last_stripe + 6066 sub_stripes - 1)) 6067 bioc->stripes[i].length -= stripe_end_offset; 6068 6069 if (i == sub_stripes - 1) 6070 stripe_offset = 0; 6071 } else { 6072 bioc->stripes[i].length = length; 6073 } 6074 6075 stripe_index++; 6076 if (stripe_index == map->num_stripes) { 6077 stripe_index = 0; 6078 stripe_nr++; 6079 } 6080 } 6081 6082 *bioc_ret = bioc; 6083 bioc->map_type = map->type; 6084 bioc->num_stripes = num_stripes; 6085 out: 6086 free_extent_map(em); 6087 return ret; 6088 } 6089 6090 /* 6091 * In dev-replace case, for repair case (that's the only case where the mirror 6092 * is selected explicitly when calling btrfs_map_block), blocks left of the 6093 * left cursor can also be read from the target drive. 6094 * 6095 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6096 * array of stripes. 6097 * For READ, it also needs to be supported using the same mirror number. 6098 * 6099 * If the requested block is not left of the left cursor, EIO is returned. This 6100 * can happen because btrfs_num_copies() returns one more in the dev-replace 6101 * case. 6102 */ 6103 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6104 u64 logical, u64 length, 6105 u64 srcdev_devid, int *mirror_num, 6106 u64 *physical) 6107 { 6108 struct btrfs_io_context *bioc = NULL; 6109 int num_stripes; 6110 int index_srcdev = 0; 6111 int found = 0; 6112 u64 physical_of_found = 0; 6113 int i; 6114 int ret = 0; 6115 6116 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6117 logical, &length, &bioc, 0, 0); 6118 if (ret) { 6119 ASSERT(bioc == NULL); 6120 return ret; 6121 } 6122 6123 num_stripes = bioc->num_stripes; 6124 if (*mirror_num > num_stripes) { 6125 /* 6126 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6127 * that means that the requested area is not left of the left 6128 * cursor 6129 */ 6130 btrfs_put_bioc(bioc); 6131 return -EIO; 6132 } 6133 6134 /* 6135 * process the rest of the function using the mirror_num of the source 6136 * drive. Therefore look it up first. At the end, patch the device 6137 * pointer to the one of the target drive. 6138 */ 6139 for (i = 0; i < num_stripes; i++) { 6140 if (bioc->stripes[i].dev->devid != srcdev_devid) 6141 continue; 6142 6143 /* 6144 * In case of DUP, in order to keep it simple, only add the 6145 * mirror with the lowest physical address 6146 */ 6147 if (found && 6148 physical_of_found <= bioc->stripes[i].physical) 6149 continue; 6150 6151 index_srcdev = i; 6152 found = 1; 6153 physical_of_found = bioc->stripes[i].physical; 6154 } 6155 6156 btrfs_put_bioc(bioc); 6157 6158 ASSERT(found); 6159 if (!found) 6160 return -EIO; 6161 6162 *mirror_num = index_srcdev + 1; 6163 *physical = physical_of_found; 6164 return ret; 6165 } 6166 6167 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6168 { 6169 struct btrfs_block_group *cache; 6170 bool ret; 6171 6172 /* Non zoned filesystem does not use "to_copy" flag */ 6173 if (!btrfs_is_zoned(fs_info)) 6174 return false; 6175 6176 cache = btrfs_lookup_block_group(fs_info, logical); 6177 6178 spin_lock(&cache->lock); 6179 ret = cache->to_copy; 6180 spin_unlock(&cache->lock); 6181 6182 btrfs_put_block_group(cache); 6183 return ret; 6184 } 6185 6186 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6187 struct btrfs_io_context **bioc_ret, 6188 struct btrfs_dev_replace *dev_replace, 6189 u64 logical, 6190 int *num_stripes_ret, int *max_errors_ret) 6191 { 6192 struct btrfs_io_context *bioc = *bioc_ret; 6193 u64 srcdev_devid = dev_replace->srcdev->devid; 6194 int tgtdev_indexes = 0; 6195 int num_stripes = *num_stripes_ret; 6196 int max_errors = *max_errors_ret; 6197 int i; 6198 6199 if (op == BTRFS_MAP_WRITE) { 6200 int index_where_to_add; 6201 6202 /* 6203 * A block group which have "to_copy" set will eventually 6204 * copied by dev-replace process. We can avoid cloning IO here. 6205 */ 6206 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6207 return; 6208 6209 /* 6210 * duplicate the write operations while the dev replace 6211 * procedure is running. Since the copying of the old disk to 6212 * the new disk takes place at run time while the filesystem is 6213 * mounted writable, the regular write operations to the old 6214 * disk have to be duplicated to go to the new disk as well. 6215 * 6216 * Note that device->missing is handled by the caller, and that 6217 * the write to the old disk is already set up in the stripes 6218 * array. 6219 */ 6220 index_where_to_add = num_stripes; 6221 for (i = 0; i < num_stripes; i++) { 6222 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6223 /* write to new disk, too */ 6224 struct btrfs_io_stripe *new = 6225 bioc->stripes + index_where_to_add; 6226 struct btrfs_io_stripe *old = 6227 bioc->stripes + i; 6228 6229 new->physical = old->physical; 6230 new->length = old->length; 6231 new->dev = dev_replace->tgtdev; 6232 bioc->tgtdev_map[i] = index_where_to_add; 6233 index_where_to_add++; 6234 max_errors++; 6235 tgtdev_indexes++; 6236 } 6237 } 6238 num_stripes = index_where_to_add; 6239 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6240 int index_srcdev = 0; 6241 int found = 0; 6242 u64 physical_of_found = 0; 6243 6244 /* 6245 * During the dev-replace procedure, the target drive can also 6246 * be used to read data in case it is needed to repair a corrupt 6247 * block elsewhere. This is possible if the requested area is 6248 * left of the left cursor. In this area, the target drive is a 6249 * full copy of the source drive. 6250 */ 6251 for (i = 0; i < num_stripes; i++) { 6252 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6253 /* 6254 * In case of DUP, in order to keep it simple, 6255 * only add the mirror with the lowest physical 6256 * address 6257 */ 6258 if (found && 6259 physical_of_found <= bioc->stripes[i].physical) 6260 continue; 6261 index_srcdev = i; 6262 found = 1; 6263 physical_of_found = bioc->stripes[i].physical; 6264 } 6265 } 6266 if (found) { 6267 struct btrfs_io_stripe *tgtdev_stripe = 6268 bioc->stripes + num_stripes; 6269 6270 tgtdev_stripe->physical = physical_of_found; 6271 tgtdev_stripe->length = 6272 bioc->stripes[index_srcdev].length; 6273 tgtdev_stripe->dev = dev_replace->tgtdev; 6274 bioc->tgtdev_map[index_srcdev] = num_stripes; 6275 6276 tgtdev_indexes++; 6277 num_stripes++; 6278 } 6279 } 6280 6281 *num_stripes_ret = num_stripes; 6282 *max_errors_ret = max_errors; 6283 bioc->num_tgtdevs = tgtdev_indexes; 6284 *bioc_ret = bioc; 6285 } 6286 6287 static bool need_full_stripe(enum btrfs_map_op op) 6288 { 6289 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6290 } 6291 6292 /* 6293 * Calculate the geometry of a particular (address, len) tuple. This 6294 * information is used to calculate how big a particular bio can get before it 6295 * straddles a stripe. 6296 * 6297 * @fs_info: the filesystem 6298 * @em: mapping containing the logical extent 6299 * @op: type of operation - write or read 6300 * @logical: address that we want to figure out the geometry of 6301 * @io_geom: pointer used to return values 6302 * 6303 * Returns < 0 in case a chunk for the given logical address cannot be found, 6304 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6305 */ 6306 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6307 enum btrfs_map_op op, u64 logical, 6308 struct btrfs_io_geometry *io_geom) 6309 { 6310 struct map_lookup *map; 6311 u64 len; 6312 u64 offset; 6313 u64 stripe_offset; 6314 u64 stripe_nr; 6315 u64 stripe_len; 6316 u64 raid56_full_stripe_start = (u64)-1; 6317 int data_stripes; 6318 6319 ASSERT(op != BTRFS_MAP_DISCARD); 6320 6321 map = em->map_lookup; 6322 /* Offset of this logical address in the chunk */ 6323 offset = logical - em->start; 6324 /* Len of a stripe in a chunk */ 6325 stripe_len = map->stripe_len; 6326 /* Stripe where this block falls in */ 6327 stripe_nr = div64_u64(offset, stripe_len); 6328 /* Offset of stripe in the chunk */ 6329 stripe_offset = stripe_nr * stripe_len; 6330 if (offset < stripe_offset) { 6331 btrfs_crit(fs_info, 6332 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", 6333 stripe_offset, offset, em->start, logical, stripe_len); 6334 return -EINVAL; 6335 } 6336 6337 /* stripe_offset is the offset of this block in its stripe */ 6338 stripe_offset = offset - stripe_offset; 6339 data_stripes = nr_data_stripes(map); 6340 6341 /* Only stripe based profiles needs to check against stripe length. */ 6342 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) { 6343 u64 max_len = stripe_len - stripe_offset; 6344 6345 /* 6346 * In case of raid56, we need to know the stripe aligned start 6347 */ 6348 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6349 unsigned long full_stripe_len = stripe_len * data_stripes; 6350 raid56_full_stripe_start = offset; 6351 6352 /* 6353 * Allow a write of a full stripe, but make sure we 6354 * don't allow straddling of stripes 6355 */ 6356 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6357 full_stripe_len); 6358 raid56_full_stripe_start *= full_stripe_len; 6359 6360 /* 6361 * For writes to RAID[56], allow a full stripeset across 6362 * all disks. For other RAID types and for RAID[56] 6363 * reads, just allow a single stripe (on a single disk). 6364 */ 6365 if (op == BTRFS_MAP_WRITE) { 6366 max_len = stripe_len * data_stripes - 6367 (offset - raid56_full_stripe_start); 6368 } 6369 } 6370 len = min_t(u64, em->len - offset, max_len); 6371 } else { 6372 len = em->len - offset; 6373 } 6374 6375 io_geom->len = len; 6376 io_geom->offset = offset; 6377 io_geom->stripe_len = stripe_len; 6378 io_geom->stripe_nr = stripe_nr; 6379 io_geom->stripe_offset = stripe_offset; 6380 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6381 6382 return 0; 6383 } 6384 6385 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6386 enum btrfs_map_op op, 6387 u64 logical, u64 *length, 6388 struct btrfs_io_context **bioc_ret, 6389 int mirror_num, int need_raid_map) 6390 { 6391 struct extent_map *em; 6392 struct map_lookup *map; 6393 u64 stripe_offset; 6394 u64 stripe_nr; 6395 u64 stripe_len; 6396 u32 stripe_index; 6397 int data_stripes; 6398 int i; 6399 int ret = 0; 6400 int num_stripes; 6401 int max_errors = 0; 6402 int tgtdev_indexes = 0; 6403 struct btrfs_io_context *bioc = NULL; 6404 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6405 int dev_replace_is_ongoing = 0; 6406 int num_alloc_stripes; 6407 int patch_the_first_stripe_for_dev_replace = 0; 6408 u64 physical_to_patch_in_first_stripe = 0; 6409 u64 raid56_full_stripe_start = (u64)-1; 6410 struct btrfs_io_geometry geom; 6411 6412 ASSERT(bioc_ret); 6413 ASSERT(op != BTRFS_MAP_DISCARD); 6414 6415 em = btrfs_get_chunk_map(fs_info, logical, *length); 6416 ASSERT(!IS_ERR(em)); 6417 6418 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6419 if (ret < 0) 6420 return ret; 6421 6422 map = em->map_lookup; 6423 6424 *length = geom.len; 6425 stripe_len = geom.stripe_len; 6426 stripe_nr = geom.stripe_nr; 6427 stripe_offset = geom.stripe_offset; 6428 raid56_full_stripe_start = geom.raid56_stripe_offset; 6429 data_stripes = nr_data_stripes(map); 6430 6431 down_read(&dev_replace->rwsem); 6432 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6433 /* 6434 * Hold the semaphore for read during the whole operation, write is 6435 * requested at commit time but must wait. 6436 */ 6437 if (!dev_replace_is_ongoing) 6438 up_read(&dev_replace->rwsem); 6439 6440 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6441 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6442 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6443 dev_replace->srcdev->devid, 6444 &mirror_num, 6445 &physical_to_patch_in_first_stripe); 6446 if (ret) 6447 goto out; 6448 else 6449 patch_the_first_stripe_for_dev_replace = 1; 6450 } else if (mirror_num > map->num_stripes) { 6451 mirror_num = 0; 6452 } 6453 6454 num_stripes = 1; 6455 stripe_index = 0; 6456 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6457 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6458 &stripe_index); 6459 if (!need_full_stripe(op)) 6460 mirror_num = 1; 6461 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6462 if (need_full_stripe(op)) 6463 num_stripes = map->num_stripes; 6464 else if (mirror_num) 6465 stripe_index = mirror_num - 1; 6466 else { 6467 stripe_index = find_live_mirror(fs_info, map, 0, 6468 dev_replace_is_ongoing); 6469 mirror_num = stripe_index + 1; 6470 } 6471 6472 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6473 if (need_full_stripe(op)) { 6474 num_stripes = map->num_stripes; 6475 } else if (mirror_num) { 6476 stripe_index = mirror_num - 1; 6477 } else { 6478 mirror_num = 1; 6479 } 6480 6481 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6482 u32 factor = map->num_stripes / map->sub_stripes; 6483 6484 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6485 stripe_index *= map->sub_stripes; 6486 6487 if (need_full_stripe(op)) 6488 num_stripes = map->sub_stripes; 6489 else if (mirror_num) 6490 stripe_index += mirror_num - 1; 6491 else { 6492 int old_stripe_index = stripe_index; 6493 stripe_index = find_live_mirror(fs_info, map, 6494 stripe_index, 6495 dev_replace_is_ongoing); 6496 mirror_num = stripe_index - old_stripe_index + 1; 6497 } 6498 6499 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6500 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6501 /* push stripe_nr back to the start of the full stripe */ 6502 stripe_nr = div64_u64(raid56_full_stripe_start, 6503 stripe_len * data_stripes); 6504 6505 /* RAID[56] write or recovery. Return all stripes */ 6506 num_stripes = map->num_stripes; 6507 max_errors = nr_parity_stripes(map); 6508 6509 *length = map->stripe_len; 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 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6557 if (!bioc) { 6558 ret = -ENOMEM; 6559 goto out; 6560 } 6561 6562 for (i = 0; i < num_stripes; i++) { 6563 bioc->stripes[i].physical = map->stripes[stripe_index].physical + 6564 stripe_offset + stripe_nr * map->stripe_len; 6565 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 6566 stripe_index++; 6567 } 6568 6569 /* Build raid_map */ 6570 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6571 (need_full_stripe(op) || mirror_num > 1)) { 6572 u64 tmp; 6573 unsigned rot; 6574 6575 /* Work out the disk rotation on this stripe-set */ 6576 div_u64_rem(stripe_nr, num_stripes, &rot); 6577 6578 /* Fill in the logical address of each stripe */ 6579 tmp = stripe_nr * data_stripes; 6580 for (i = 0; i < data_stripes; i++) 6581 bioc->raid_map[(i + rot) % num_stripes] = 6582 em->start + (tmp + i) * map->stripe_len; 6583 6584 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6585 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6586 bioc->raid_map[(i + rot + 1) % num_stripes] = 6587 RAID6_Q_STRIPE; 6588 6589 sort_parity_stripes(bioc, num_stripes); 6590 } 6591 6592 if (need_full_stripe(op)) 6593 max_errors = btrfs_chunk_max_errors(map); 6594 6595 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6596 need_full_stripe(op)) { 6597 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6598 &num_stripes, &max_errors); 6599 } 6600 6601 *bioc_ret = bioc; 6602 bioc->map_type = map->type; 6603 bioc->num_stripes = num_stripes; 6604 bioc->max_errors = max_errors; 6605 bioc->mirror_num = mirror_num; 6606 6607 /* 6608 * this is the case that REQ_READ && dev_replace_is_ongoing && 6609 * mirror_num == num_stripes + 1 && dev_replace target drive is 6610 * available as a mirror 6611 */ 6612 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6613 WARN_ON(num_stripes > 1); 6614 bioc->stripes[0].dev = dev_replace->tgtdev; 6615 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6616 bioc->mirror_num = map->num_stripes + 1; 6617 } 6618 out: 6619 if (dev_replace_is_ongoing) { 6620 lockdep_assert_held(&dev_replace->rwsem); 6621 /* Unlock and let waiting writers proceed */ 6622 up_read(&dev_replace->rwsem); 6623 } 6624 free_extent_map(em); 6625 return ret; 6626 } 6627 6628 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6629 u64 logical, u64 *length, 6630 struct btrfs_io_context **bioc_ret, int mirror_num) 6631 { 6632 if (op == BTRFS_MAP_DISCARD) 6633 return __btrfs_map_block_for_discard(fs_info, logical, 6634 length, bioc_ret); 6635 6636 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6637 mirror_num, 0); 6638 } 6639 6640 /* For Scrub/replace */ 6641 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6642 u64 logical, u64 *length, 6643 struct btrfs_io_context **bioc_ret) 6644 { 6645 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1); 6646 } 6647 6648 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio) 6649 { 6650 bio->bi_private = bioc->private; 6651 bio->bi_end_io = bioc->end_io; 6652 bio_endio(bio); 6653 6654 btrfs_put_bioc(bioc); 6655 } 6656 6657 static void btrfs_end_bio(struct bio *bio) 6658 { 6659 struct btrfs_io_context *bioc = bio->bi_private; 6660 int is_orig_bio = 0; 6661 6662 if (bio->bi_status) { 6663 atomic_inc(&bioc->error); 6664 if (bio->bi_status == BLK_STS_IOERR || 6665 bio->bi_status == BLK_STS_TARGET) { 6666 struct btrfs_device *dev = btrfs_bio(bio)->device; 6667 6668 ASSERT(dev->bdev); 6669 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6670 btrfs_dev_stat_inc_and_print(dev, 6671 BTRFS_DEV_STAT_WRITE_ERRS); 6672 else if (!(bio->bi_opf & REQ_RAHEAD)) 6673 btrfs_dev_stat_inc_and_print(dev, 6674 BTRFS_DEV_STAT_READ_ERRS); 6675 if (bio->bi_opf & REQ_PREFLUSH) 6676 btrfs_dev_stat_inc_and_print(dev, 6677 BTRFS_DEV_STAT_FLUSH_ERRS); 6678 } 6679 } 6680 6681 if (bio == bioc->orig_bio) 6682 is_orig_bio = 1; 6683 6684 btrfs_bio_counter_dec(bioc->fs_info); 6685 6686 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6687 if (!is_orig_bio) { 6688 bio_put(bio); 6689 bio = bioc->orig_bio; 6690 } 6691 6692 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6693 /* only send an error to the higher layers if it is 6694 * beyond the tolerance of the btrfs bio 6695 */ 6696 if (atomic_read(&bioc->error) > bioc->max_errors) { 6697 bio->bi_status = BLK_STS_IOERR; 6698 } else { 6699 /* 6700 * this bio is actually up to date, we didn't 6701 * go over the max number of errors 6702 */ 6703 bio->bi_status = BLK_STS_OK; 6704 } 6705 6706 btrfs_end_bioc(bioc, bio); 6707 } else if (!is_orig_bio) { 6708 bio_put(bio); 6709 } 6710 } 6711 6712 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio, 6713 u64 physical, struct btrfs_device *dev) 6714 { 6715 struct btrfs_fs_info *fs_info = bioc->fs_info; 6716 6717 bio->bi_private = bioc; 6718 btrfs_bio(bio)->device = dev; 6719 bio->bi_end_io = btrfs_end_bio; 6720 bio->bi_iter.bi_sector = physical >> 9; 6721 /* 6722 * For zone append writing, bi_sector must point the beginning of the 6723 * zone 6724 */ 6725 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6726 if (btrfs_dev_is_sequential(dev, physical)) { 6727 u64 zone_start = round_down(physical, fs_info->zone_size); 6728 6729 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6730 } else { 6731 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6732 bio->bi_opf |= REQ_OP_WRITE; 6733 } 6734 } 6735 btrfs_debug_in_rcu(fs_info, 6736 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6737 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6738 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6739 dev->devid, bio->bi_iter.bi_size); 6740 bio_set_dev(bio, dev->bdev); 6741 6742 btrfs_bio_counter_inc_noblocked(fs_info); 6743 6744 btrfsic_submit_bio(bio); 6745 } 6746 6747 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical) 6748 { 6749 atomic_inc(&bioc->error); 6750 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6751 /* Should be the original bio. */ 6752 WARN_ON(bio != bioc->orig_bio); 6753 6754 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6755 bio->bi_iter.bi_sector = logical >> 9; 6756 if (atomic_read(&bioc->error) > bioc->max_errors) 6757 bio->bi_status = BLK_STS_IOERR; 6758 else 6759 bio->bi_status = BLK_STS_OK; 6760 btrfs_end_bioc(bioc, bio); 6761 } 6762 } 6763 6764 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 6765 int mirror_num) 6766 { 6767 struct btrfs_device *dev; 6768 struct bio *first_bio = bio; 6769 u64 logical = bio->bi_iter.bi_sector << 9; 6770 u64 length = 0; 6771 u64 map_length; 6772 int ret; 6773 int dev_nr; 6774 int total_devs; 6775 struct btrfs_io_context *bioc = NULL; 6776 6777 length = bio->bi_iter.bi_size; 6778 map_length = length; 6779 6780 btrfs_bio_counter_inc_blocked(fs_info); 6781 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, 6782 &map_length, &bioc, mirror_num, 1); 6783 if (ret) { 6784 btrfs_bio_counter_dec(fs_info); 6785 return errno_to_blk_status(ret); 6786 } 6787 6788 total_devs = bioc->num_stripes; 6789 bioc->orig_bio = first_bio; 6790 bioc->private = first_bio->bi_private; 6791 bioc->end_io = first_bio->bi_end_io; 6792 atomic_set(&bioc->stripes_pending, bioc->num_stripes); 6793 6794 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && 6795 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) { 6796 /* In this case, map_length has been set to the length of 6797 a single stripe; not the whole write */ 6798 if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 6799 ret = raid56_parity_write(bio, bioc, map_length); 6800 } else { 6801 ret = raid56_parity_recover(bio, bioc, map_length, 6802 mirror_num, 1); 6803 } 6804 6805 btrfs_bio_counter_dec(fs_info); 6806 return errno_to_blk_status(ret); 6807 } 6808 6809 if (map_length < length) { 6810 btrfs_crit(fs_info, 6811 "mapping failed logical %llu bio len %llu len %llu", 6812 logical, length, map_length); 6813 BUG(); 6814 } 6815 6816 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { 6817 dev = bioc->stripes[dev_nr].dev; 6818 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, 6819 &dev->dev_state) || 6820 (btrfs_op(first_bio) == BTRFS_MAP_WRITE && 6821 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6822 bioc_error(bioc, first_bio, logical); 6823 continue; 6824 } 6825 6826 if (dev_nr < total_devs - 1) 6827 bio = btrfs_bio_clone(first_bio); 6828 else 6829 bio = first_bio; 6830 6831 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev); 6832 } 6833 btrfs_bio_counter_dec(fs_info); 6834 return BLK_STS_OK; 6835 } 6836 6837 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6838 const struct btrfs_fs_devices *fs_devices) 6839 { 6840 if (args->fsid == NULL) 6841 return true; 6842 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6843 return true; 6844 return false; 6845 } 6846 6847 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6848 const struct btrfs_device *device) 6849 { 6850 ASSERT((args->devid != (u64)-1) || args->missing); 6851 6852 if ((args->devid != (u64)-1) && device->devid != args->devid) 6853 return false; 6854 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6855 return false; 6856 if (!args->missing) 6857 return true; 6858 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6859 !device->bdev) 6860 return true; 6861 return false; 6862 } 6863 6864 /* 6865 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6866 * return NULL. 6867 * 6868 * If devid and uuid are both specified, the match must be exact, otherwise 6869 * only devid is used. 6870 */ 6871 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6872 const struct btrfs_dev_lookup_args *args) 6873 { 6874 struct btrfs_device *device; 6875 struct btrfs_fs_devices *seed_devs; 6876 6877 if (dev_args_match_fs_devices(args, fs_devices)) { 6878 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6879 if (dev_args_match_device(args, device)) 6880 return device; 6881 } 6882 } 6883 6884 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6885 if (!dev_args_match_fs_devices(args, seed_devs)) 6886 continue; 6887 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6888 if (dev_args_match_device(args, device)) 6889 return device; 6890 } 6891 } 6892 6893 return NULL; 6894 } 6895 6896 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6897 u64 devid, u8 *dev_uuid) 6898 { 6899 struct btrfs_device *device; 6900 unsigned int nofs_flag; 6901 6902 /* 6903 * We call this under the chunk_mutex, so we want to use NOFS for this 6904 * allocation, however we don't want to change btrfs_alloc_device() to 6905 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6906 * places. 6907 */ 6908 nofs_flag = memalloc_nofs_save(); 6909 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 6910 memalloc_nofs_restore(nofs_flag); 6911 if (IS_ERR(device)) 6912 return device; 6913 6914 list_add(&device->dev_list, &fs_devices->devices); 6915 device->fs_devices = fs_devices; 6916 fs_devices->num_devices++; 6917 6918 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6919 fs_devices->missing_devices++; 6920 6921 return device; 6922 } 6923 6924 /** 6925 * btrfs_alloc_device - allocate struct btrfs_device 6926 * @fs_info: used only for generating a new devid, can be NULL if 6927 * devid is provided (i.e. @devid != NULL). 6928 * @devid: a pointer to devid for this device. If NULL a new devid 6929 * is generated. 6930 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6931 * is generated. 6932 * 6933 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6934 * on error. Returned struct is not linked onto any lists and must be 6935 * destroyed with btrfs_free_device. 6936 */ 6937 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6938 const u64 *devid, 6939 const u8 *uuid) 6940 { 6941 struct btrfs_device *dev; 6942 u64 tmp; 6943 6944 if (WARN_ON(!devid && !fs_info)) 6945 return ERR_PTR(-EINVAL); 6946 6947 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6948 if (!dev) 6949 return ERR_PTR(-ENOMEM); 6950 6951 INIT_LIST_HEAD(&dev->dev_list); 6952 INIT_LIST_HEAD(&dev->dev_alloc_list); 6953 INIT_LIST_HEAD(&dev->post_commit_list); 6954 6955 atomic_set(&dev->dev_stats_ccnt, 0); 6956 btrfs_device_data_ordered_init(dev); 6957 extent_io_tree_init(fs_info, &dev->alloc_state, 6958 IO_TREE_DEVICE_ALLOC_STATE, NULL); 6959 6960 if (devid) 6961 tmp = *devid; 6962 else { 6963 int ret; 6964 6965 ret = find_next_devid(fs_info, &tmp); 6966 if (ret) { 6967 btrfs_free_device(dev); 6968 return ERR_PTR(ret); 6969 } 6970 } 6971 dev->devid = tmp; 6972 6973 if (uuid) 6974 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6975 else 6976 generate_random_uuid(dev->uuid); 6977 6978 return dev; 6979 } 6980 6981 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6982 u64 devid, u8 *uuid, bool error) 6983 { 6984 if (error) 6985 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6986 devid, uuid); 6987 else 6988 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6989 devid, uuid); 6990 } 6991 6992 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) 6993 { 6994 const int data_stripes = calc_data_stripes(type, num_stripes); 6995 6996 return div_u64(chunk_len, data_stripes); 6997 } 6998 6999 #if BITS_PER_LONG == 32 7000 /* 7001 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 7002 * can't be accessed on 32bit systems. 7003 * 7004 * This function do mount time check to reject the fs if it already has 7005 * metadata chunk beyond that limit. 7006 */ 7007 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7008 u64 logical, u64 length, u64 type) 7009 { 7010 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7011 return 0; 7012 7013 if (logical + length < MAX_LFS_FILESIZE) 7014 return 0; 7015 7016 btrfs_err_32bit_limit(fs_info); 7017 return -EOVERFLOW; 7018 } 7019 7020 /* 7021 * This is to give early warning for any metadata chunk reaching 7022 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 7023 * Although we can still access the metadata, it's not going to be possible 7024 * once the limit is reached. 7025 */ 7026 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 7027 u64 logical, u64 length, u64 type) 7028 { 7029 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 7030 return; 7031 7032 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 7033 return; 7034 7035 btrfs_warn_32bit_limit(fs_info); 7036 } 7037 #endif 7038 7039 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, 7040 u64 devid, u8 *uuid) 7041 { 7042 struct btrfs_device *dev; 7043 7044 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7045 btrfs_report_missing_device(fs_info, devid, uuid, true); 7046 return ERR_PTR(-ENOENT); 7047 } 7048 7049 dev = add_missing_dev(fs_info->fs_devices, devid, uuid); 7050 if (IS_ERR(dev)) { 7051 btrfs_err(fs_info, "failed to init missing device %llu: %ld", 7052 devid, PTR_ERR(dev)); 7053 return dev; 7054 } 7055 btrfs_report_missing_device(fs_info, devid, uuid, false); 7056 7057 return dev; 7058 } 7059 7060 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 7061 struct btrfs_chunk *chunk) 7062 { 7063 BTRFS_DEV_LOOKUP_ARGS(args); 7064 struct btrfs_fs_info *fs_info = leaf->fs_info; 7065 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7066 struct map_lookup *map; 7067 struct extent_map *em; 7068 u64 logical; 7069 u64 length; 7070 u64 devid; 7071 u64 type; 7072 u8 uuid[BTRFS_UUID_SIZE]; 7073 int num_stripes; 7074 int ret; 7075 int i; 7076 7077 logical = key->offset; 7078 length = btrfs_chunk_length(leaf, chunk); 7079 type = btrfs_chunk_type(leaf, chunk); 7080 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 7081 7082 #if BITS_PER_LONG == 32 7083 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 7084 if (ret < 0) 7085 return ret; 7086 warn_32bit_meta_chunk(fs_info, logical, length, type); 7087 #endif 7088 7089 /* 7090 * Only need to verify chunk item if we're reading from sys chunk array, 7091 * as chunk item in tree block is already verified by tree-checker. 7092 */ 7093 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 7094 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 7095 if (ret) 7096 return ret; 7097 } 7098 7099 read_lock(&map_tree->lock); 7100 em = lookup_extent_mapping(map_tree, logical, 1); 7101 read_unlock(&map_tree->lock); 7102 7103 /* already mapped? */ 7104 if (em && em->start <= logical && em->start + em->len > logical) { 7105 free_extent_map(em); 7106 return 0; 7107 } else if (em) { 7108 free_extent_map(em); 7109 } 7110 7111 em = alloc_extent_map(); 7112 if (!em) 7113 return -ENOMEM; 7114 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7115 if (!map) { 7116 free_extent_map(em); 7117 return -ENOMEM; 7118 } 7119 7120 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7121 em->map_lookup = map; 7122 em->start = logical; 7123 em->len = length; 7124 em->orig_start = 0; 7125 em->block_start = 0; 7126 em->block_len = em->len; 7127 7128 map->num_stripes = num_stripes; 7129 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7130 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7131 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7132 map->type = type; 7133 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); 7134 map->verified_stripes = 0; 7135 em->orig_block_len = calc_stripe_length(type, em->len, 7136 map->num_stripes); 7137 for (i = 0; i < num_stripes; i++) { 7138 map->stripes[i].physical = 7139 btrfs_stripe_offset_nr(leaf, chunk, i); 7140 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7141 args.devid = devid; 7142 read_extent_buffer(leaf, uuid, (unsigned long) 7143 btrfs_stripe_dev_uuid_nr(chunk, i), 7144 BTRFS_UUID_SIZE); 7145 args.uuid = uuid; 7146 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 7147 if (!map->stripes[i].dev) { 7148 map->stripes[i].dev = handle_missing_device(fs_info, 7149 devid, uuid); 7150 if (IS_ERR(map->stripes[i].dev)) { 7151 free_extent_map(em); 7152 return PTR_ERR(map->stripes[i].dev); 7153 } 7154 } 7155 7156 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7157 &(map->stripes[i].dev->dev_state)); 7158 } 7159 7160 write_lock(&map_tree->lock); 7161 ret = add_extent_mapping(map_tree, em, 0); 7162 write_unlock(&map_tree->lock); 7163 if (ret < 0) { 7164 btrfs_err(fs_info, 7165 "failed to add chunk map, start=%llu len=%llu: %d", 7166 em->start, em->len, ret); 7167 } 7168 free_extent_map(em); 7169 7170 return ret; 7171 } 7172 7173 static void fill_device_from_item(struct extent_buffer *leaf, 7174 struct btrfs_dev_item *dev_item, 7175 struct btrfs_device *device) 7176 { 7177 unsigned long ptr; 7178 7179 device->devid = btrfs_device_id(leaf, dev_item); 7180 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7181 device->total_bytes = device->disk_total_bytes; 7182 device->commit_total_bytes = device->disk_total_bytes; 7183 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7184 device->commit_bytes_used = device->bytes_used; 7185 device->type = btrfs_device_type(leaf, dev_item); 7186 device->io_align = btrfs_device_io_align(leaf, dev_item); 7187 device->io_width = btrfs_device_io_width(leaf, dev_item); 7188 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7189 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7190 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7191 7192 ptr = btrfs_device_uuid(dev_item); 7193 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7194 } 7195 7196 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7197 u8 *fsid) 7198 { 7199 struct btrfs_fs_devices *fs_devices; 7200 int ret; 7201 7202 lockdep_assert_held(&uuid_mutex); 7203 ASSERT(fsid); 7204 7205 /* This will match only for multi-device seed fs */ 7206 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7207 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7208 return fs_devices; 7209 7210 7211 fs_devices = find_fsid(fsid, NULL); 7212 if (!fs_devices) { 7213 if (!btrfs_test_opt(fs_info, DEGRADED)) 7214 return ERR_PTR(-ENOENT); 7215 7216 fs_devices = alloc_fs_devices(fsid, NULL); 7217 if (IS_ERR(fs_devices)) 7218 return fs_devices; 7219 7220 fs_devices->seeding = true; 7221 fs_devices->opened = 1; 7222 return fs_devices; 7223 } 7224 7225 /* 7226 * Upon first call for a seed fs fsid, just create a private copy of the 7227 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7228 */ 7229 fs_devices = clone_fs_devices(fs_devices); 7230 if (IS_ERR(fs_devices)) 7231 return fs_devices; 7232 7233 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7234 if (ret) { 7235 free_fs_devices(fs_devices); 7236 return ERR_PTR(ret); 7237 } 7238 7239 if (!fs_devices->seeding) { 7240 close_fs_devices(fs_devices); 7241 free_fs_devices(fs_devices); 7242 return ERR_PTR(-EINVAL); 7243 } 7244 7245 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7246 7247 return fs_devices; 7248 } 7249 7250 static int read_one_dev(struct extent_buffer *leaf, 7251 struct btrfs_dev_item *dev_item) 7252 { 7253 BTRFS_DEV_LOOKUP_ARGS(args); 7254 struct btrfs_fs_info *fs_info = leaf->fs_info; 7255 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7256 struct btrfs_device *device; 7257 u64 devid; 7258 int ret; 7259 u8 fs_uuid[BTRFS_FSID_SIZE]; 7260 u8 dev_uuid[BTRFS_UUID_SIZE]; 7261 7262 devid = args.devid = btrfs_device_id(leaf, dev_item); 7263 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7264 BTRFS_UUID_SIZE); 7265 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7266 BTRFS_FSID_SIZE); 7267 args.uuid = dev_uuid; 7268 args.fsid = fs_uuid; 7269 7270 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7271 fs_devices = open_seed_devices(fs_info, fs_uuid); 7272 if (IS_ERR(fs_devices)) 7273 return PTR_ERR(fs_devices); 7274 } 7275 7276 device = btrfs_find_device(fs_info->fs_devices, &args); 7277 if (!device) { 7278 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7279 btrfs_report_missing_device(fs_info, devid, 7280 dev_uuid, true); 7281 return -ENOENT; 7282 } 7283 7284 device = add_missing_dev(fs_devices, devid, dev_uuid); 7285 if (IS_ERR(device)) { 7286 btrfs_err(fs_info, 7287 "failed to add missing dev %llu: %ld", 7288 devid, PTR_ERR(device)); 7289 return PTR_ERR(device); 7290 } 7291 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7292 } else { 7293 if (!device->bdev) { 7294 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7295 btrfs_report_missing_device(fs_info, 7296 devid, dev_uuid, true); 7297 return -ENOENT; 7298 } 7299 btrfs_report_missing_device(fs_info, devid, 7300 dev_uuid, false); 7301 } 7302 7303 if (!device->bdev && 7304 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7305 /* 7306 * this happens when a device that was properly setup 7307 * in the device info lists suddenly goes bad. 7308 * device->bdev is NULL, and so we have to set 7309 * device->missing to one here 7310 */ 7311 device->fs_devices->missing_devices++; 7312 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7313 } 7314 7315 /* Move the device to its own fs_devices */ 7316 if (device->fs_devices != fs_devices) { 7317 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7318 &device->dev_state)); 7319 7320 list_move(&device->dev_list, &fs_devices->devices); 7321 device->fs_devices->num_devices--; 7322 fs_devices->num_devices++; 7323 7324 device->fs_devices->missing_devices--; 7325 fs_devices->missing_devices++; 7326 7327 device->fs_devices = fs_devices; 7328 } 7329 } 7330 7331 if (device->fs_devices != fs_info->fs_devices) { 7332 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7333 if (device->generation != 7334 btrfs_device_generation(leaf, dev_item)) 7335 return -EINVAL; 7336 } 7337 7338 fill_device_from_item(leaf, dev_item, device); 7339 if (device->bdev) { 7340 u64 max_total_bytes = bdev_nr_bytes(device->bdev); 7341 7342 if (device->total_bytes > max_total_bytes) { 7343 btrfs_err(fs_info, 7344 "device total_bytes should be at most %llu but found %llu", 7345 max_total_bytes, device->total_bytes); 7346 return -EINVAL; 7347 } 7348 } 7349 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7350 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7351 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7352 device->fs_devices->total_rw_bytes += device->total_bytes; 7353 atomic64_add(device->total_bytes - device->bytes_used, 7354 &fs_info->free_chunk_space); 7355 } 7356 ret = 0; 7357 return ret; 7358 } 7359 7360 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7361 { 7362 struct btrfs_root *root = fs_info->tree_root; 7363 struct btrfs_super_block *super_copy = fs_info->super_copy; 7364 struct extent_buffer *sb; 7365 struct btrfs_disk_key *disk_key; 7366 struct btrfs_chunk *chunk; 7367 u8 *array_ptr; 7368 unsigned long sb_array_offset; 7369 int ret = 0; 7370 u32 num_stripes; 7371 u32 array_size; 7372 u32 len = 0; 7373 u32 cur_offset; 7374 u64 type; 7375 struct btrfs_key key; 7376 7377 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7378 /* 7379 * This will create extent buffer of nodesize, superblock size is 7380 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will 7381 * overallocate but we can keep it as-is, only the first page is used. 7382 */ 7383 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET, 7384 root->root_key.objectid, 0); 7385 if (IS_ERR(sb)) 7386 return PTR_ERR(sb); 7387 set_extent_buffer_uptodate(sb); 7388 /* 7389 * The sb extent buffer is artificial and just used to read the system array. 7390 * set_extent_buffer_uptodate() call does not properly mark all it's 7391 * pages up-to-date when the page is larger: extent does not cover the 7392 * whole page and consequently check_page_uptodate does not find all 7393 * the page's extents up-to-date (the hole beyond sb), 7394 * write_extent_buffer then triggers a WARN_ON. 7395 * 7396 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, 7397 * but sb spans only this function. Add an explicit SetPageUptodate call 7398 * to silence the warning eg. on PowerPC 64. 7399 */ 7400 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) 7401 SetPageUptodate(sb->pages[0]); 7402 7403 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7404 array_size = btrfs_super_sys_array_size(super_copy); 7405 7406 array_ptr = super_copy->sys_chunk_array; 7407 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7408 cur_offset = 0; 7409 7410 while (cur_offset < array_size) { 7411 disk_key = (struct btrfs_disk_key *)array_ptr; 7412 len = sizeof(*disk_key); 7413 if (cur_offset + len > array_size) 7414 goto out_short_read; 7415 7416 btrfs_disk_key_to_cpu(&key, disk_key); 7417 7418 array_ptr += len; 7419 sb_array_offset += len; 7420 cur_offset += len; 7421 7422 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7423 btrfs_err(fs_info, 7424 "unexpected item type %u in sys_array at offset %u", 7425 (u32)key.type, cur_offset); 7426 ret = -EIO; 7427 break; 7428 } 7429 7430 chunk = (struct btrfs_chunk *)sb_array_offset; 7431 /* 7432 * At least one btrfs_chunk with one stripe must be present, 7433 * exact stripe count check comes afterwards 7434 */ 7435 len = btrfs_chunk_item_size(1); 7436 if (cur_offset + len > array_size) 7437 goto out_short_read; 7438 7439 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7440 if (!num_stripes) { 7441 btrfs_err(fs_info, 7442 "invalid number of stripes %u in sys_array at offset %u", 7443 num_stripes, cur_offset); 7444 ret = -EIO; 7445 break; 7446 } 7447 7448 type = btrfs_chunk_type(sb, chunk); 7449 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7450 btrfs_err(fs_info, 7451 "invalid chunk type %llu in sys_array at offset %u", 7452 type, cur_offset); 7453 ret = -EIO; 7454 break; 7455 } 7456 7457 len = btrfs_chunk_item_size(num_stripes); 7458 if (cur_offset + len > array_size) 7459 goto out_short_read; 7460 7461 ret = read_one_chunk(&key, sb, chunk); 7462 if (ret) 7463 break; 7464 7465 array_ptr += len; 7466 sb_array_offset += len; 7467 cur_offset += len; 7468 } 7469 clear_extent_buffer_uptodate(sb); 7470 free_extent_buffer_stale(sb); 7471 return ret; 7472 7473 out_short_read: 7474 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7475 len, cur_offset); 7476 clear_extent_buffer_uptodate(sb); 7477 free_extent_buffer_stale(sb); 7478 return -EIO; 7479 } 7480 7481 /* 7482 * Check if all chunks in the fs are OK for read-write degraded mount 7483 * 7484 * If the @failing_dev is specified, it's accounted as missing. 7485 * 7486 * Return true if all chunks meet the minimal RW mount requirements. 7487 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7488 */ 7489 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7490 struct btrfs_device *failing_dev) 7491 { 7492 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7493 struct extent_map *em; 7494 u64 next_start = 0; 7495 bool ret = true; 7496 7497 read_lock(&map_tree->lock); 7498 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7499 read_unlock(&map_tree->lock); 7500 /* No chunk at all? Return false anyway */ 7501 if (!em) { 7502 ret = false; 7503 goto out; 7504 } 7505 while (em) { 7506 struct map_lookup *map; 7507 int missing = 0; 7508 int max_tolerated; 7509 int i; 7510 7511 map = em->map_lookup; 7512 max_tolerated = 7513 btrfs_get_num_tolerated_disk_barrier_failures( 7514 map->type); 7515 for (i = 0; i < map->num_stripes; i++) { 7516 struct btrfs_device *dev = map->stripes[i].dev; 7517 7518 if (!dev || !dev->bdev || 7519 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7520 dev->last_flush_error) 7521 missing++; 7522 else if (failing_dev && failing_dev == dev) 7523 missing++; 7524 } 7525 if (missing > max_tolerated) { 7526 if (!failing_dev) 7527 btrfs_warn(fs_info, 7528 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7529 em->start, missing, max_tolerated); 7530 free_extent_map(em); 7531 ret = false; 7532 goto out; 7533 } 7534 next_start = extent_map_end(em); 7535 free_extent_map(em); 7536 7537 read_lock(&map_tree->lock); 7538 em = lookup_extent_mapping(map_tree, next_start, 7539 (u64)(-1) - next_start); 7540 read_unlock(&map_tree->lock); 7541 } 7542 out: 7543 return ret; 7544 } 7545 7546 static void readahead_tree_node_children(struct extent_buffer *node) 7547 { 7548 int i; 7549 const int nr_items = btrfs_header_nritems(node); 7550 7551 for (i = 0; i < nr_items; i++) 7552 btrfs_readahead_node_child(node, i); 7553 } 7554 7555 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7556 { 7557 struct btrfs_root *root = fs_info->chunk_root; 7558 struct btrfs_path *path; 7559 struct extent_buffer *leaf; 7560 struct btrfs_key key; 7561 struct btrfs_key found_key; 7562 int ret; 7563 int slot; 7564 u64 total_dev = 0; 7565 u64 last_ra_node = 0; 7566 7567 path = btrfs_alloc_path(); 7568 if (!path) 7569 return -ENOMEM; 7570 7571 /* 7572 * uuid_mutex is needed only if we are mounting a sprout FS 7573 * otherwise we don't need it. 7574 */ 7575 mutex_lock(&uuid_mutex); 7576 7577 /* 7578 * It is possible for mount and umount to race in such a way that 7579 * we execute this code path, but open_fs_devices failed to clear 7580 * total_rw_bytes. We certainly want it cleared before reading the 7581 * device items, so clear it here. 7582 */ 7583 fs_info->fs_devices->total_rw_bytes = 0; 7584 7585 /* 7586 * Lockdep complains about possible circular locking dependency between 7587 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores 7588 * used for freeze procection of a fs (struct super_block.s_writers), 7589 * which we take when starting a transaction, and extent buffers of the 7590 * chunk tree if we call read_one_dev() while holding a lock on an 7591 * extent buffer of the chunk tree. Since we are mounting the filesystem 7592 * and at this point there can't be any concurrent task modifying the 7593 * chunk tree, to keep it simple, just skip locking on the chunk tree. 7594 */ 7595 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7596 path->skip_locking = 1; 7597 7598 /* 7599 * Read all device items, and then all the chunk items. All 7600 * device items are found before any chunk item (their object id 7601 * is smaller than the lowest possible object id for a chunk 7602 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7603 */ 7604 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7605 key.offset = 0; 7606 key.type = 0; 7607 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7608 if (ret < 0) 7609 goto error; 7610 while (1) { 7611 struct extent_buffer *node; 7612 7613 leaf = path->nodes[0]; 7614 slot = path->slots[0]; 7615 if (slot >= btrfs_header_nritems(leaf)) { 7616 ret = btrfs_next_leaf(root, path); 7617 if (ret == 0) 7618 continue; 7619 if (ret < 0) 7620 goto error; 7621 break; 7622 } 7623 node = path->nodes[1]; 7624 if (node) { 7625 if (last_ra_node != node->start) { 7626 readahead_tree_node_children(node); 7627 last_ra_node = node->start; 7628 } 7629 } 7630 btrfs_item_key_to_cpu(leaf, &found_key, slot); 7631 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7632 struct btrfs_dev_item *dev_item; 7633 dev_item = btrfs_item_ptr(leaf, slot, 7634 struct btrfs_dev_item); 7635 ret = read_one_dev(leaf, dev_item); 7636 if (ret) 7637 goto error; 7638 total_dev++; 7639 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7640 struct btrfs_chunk *chunk; 7641 7642 /* 7643 * We are only called at mount time, so no need to take 7644 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7645 * we always lock first fs_info->chunk_mutex before 7646 * acquiring any locks on the chunk tree. This is a 7647 * requirement for chunk allocation, see the comment on 7648 * top of btrfs_chunk_alloc() for details. 7649 */ 7650 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7651 ret = read_one_chunk(&found_key, leaf, chunk); 7652 if (ret) 7653 goto error; 7654 } 7655 path->slots[0]++; 7656 } 7657 7658 /* 7659 * After loading chunk tree, we've got all device information, 7660 * do another round of validation checks. 7661 */ 7662 if (total_dev != fs_info->fs_devices->total_devices) { 7663 btrfs_err(fs_info, 7664 "super_num_devices %llu mismatch with num_devices %llu found here", 7665 btrfs_super_num_devices(fs_info->super_copy), 7666 total_dev); 7667 ret = -EINVAL; 7668 goto error; 7669 } 7670 if (btrfs_super_total_bytes(fs_info->super_copy) < 7671 fs_info->fs_devices->total_rw_bytes) { 7672 btrfs_err(fs_info, 7673 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7674 btrfs_super_total_bytes(fs_info->super_copy), 7675 fs_info->fs_devices->total_rw_bytes); 7676 ret = -EINVAL; 7677 goto error; 7678 } 7679 ret = 0; 7680 error: 7681 mutex_unlock(&uuid_mutex); 7682 7683 btrfs_free_path(path); 7684 return ret; 7685 } 7686 7687 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7688 { 7689 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7690 struct btrfs_device *device; 7691 7692 fs_devices->fs_info = fs_info; 7693 7694 mutex_lock(&fs_devices->device_list_mutex); 7695 list_for_each_entry(device, &fs_devices->devices, dev_list) 7696 device->fs_info = fs_info; 7697 7698 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7699 list_for_each_entry(device, &seed_devs->devices, dev_list) 7700 device->fs_info = fs_info; 7701 7702 seed_devs->fs_info = fs_info; 7703 } 7704 mutex_unlock(&fs_devices->device_list_mutex); 7705 } 7706 7707 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7708 const struct btrfs_dev_stats_item *ptr, 7709 int index) 7710 { 7711 u64 val; 7712 7713 read_extent_buffer(eb, &val, 7714 offsetof(struct btrfs_dev_stats_item, values) + 7715 ((unsigned long)ptr) + (index * sizeof(u64)), 7716 sizeof(val)); 7717 return val; 7718 } 7719 7720 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7721 struct btrfs_dev_stats_item *ptr, 7722 int index, u64 val) 7723 { 7724 write_extent_buffer(eb, &val, 7725 offsetof(struct btrfs_dev_stats_item, values) + 7726 ((unsigned long)ptr) + (index * sizeof(u64)), 7727 sizeof(val)); 7728 } 7729 7730 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7731 struct btrfs_path *path) 7732 { 7733 struct btrfs_dev_stats_item *ptr; 7734 struct extent_buffer *eb; 7735 struct btrfs_key key; 7736 int item_size; 7737 int i, ret, slot; 7738 7739 if (!device->fs_info->dev_root) 7740 return 0; 7741 7742 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7743 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7744 key.offset = device->devid; 7745 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7746 if (ret) { 7747 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7748 btrfs_dev_stat_set(device, i, 0); 7749 device->dev_stats_valid = 1; 7750 btrfs_release_path(path); 7751 return ret < 0 ? ret : 0; 7752 } 7753 slot = path->slots[0]; 7754 eb = path->nodes[0]; 7755 item_size = btrfs_item_size(eb, slot); 7756 7757 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7758 7759 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7760 if (item_size >= (1 + i) * sizeof(__le64)) 7761 btrfs_dev_stat_set(device, i, 7762 btrfs_dev_stats_value(eb, ptr, i)); 7763 else 7764 btrfs_dev_stat_set(device, i, 0); 7765 } 7766 7767 device->dev_stats_valid = 1; 7768 btrfs_dev_stat_print_on_load(device); 7769 btrfs_release_path(path); 7770 7771 return 0; 7772 } 7773 7774 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7775 { 7776 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7777 struct btrfs_device *device; 7778 struct btrfs_path *path = NULL; 7779 int ret = 0; 7780 7781 path = btrfs_alloc_path(); 7782 if (!path) 7783 return -ENOMEM; 7784 7785 mutex_lock(&fs_devices->device_list_mutex); 7786 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7787 ret = btrfs_device_init_dev_stats(device, path); 7788 if (ret) 7789 goto out; 7790 } 7791 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7792 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7793 ret = btrfs_device_init_dev_stats(device, path); 7794 if (ret) 7795 goto out; 7796 } 7797 } 7798 out: 7799 mutex_unlock(&fs_devices->device_list_mutex); 7800 7801 btrfs_free_path(path); 7802 return ret; 7803 } 7804 7805 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7806 struct btrfs_device *device) 7807 { 7808 struct btrfs_fs_info *fs_info = trans->fs_info; 7809 struct btrfs_root *dev_root = fs_info->dev_root; 7810 struct btrfs_path *path; 7811 struct btrfs_key key; 7812 struct extent_buffer *eb; 7813 struct btrfs_dev_stats_item *ptr; 7814 int ret; 7815 int i; 7816 7817 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7818 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7819 key.offset = device->devid; 7820 7821 path = btrfs_alloc_path(); 7822 if (!path) 7823 return -ENOMEM; 7824 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7825 if (ret < 0) { 7826 btrfs_warn_in_rcu(fs_info, 7827 "error %d while searching for dev_stats item for device %s", 7828 ret, rcu_str_deref(device->name)); 7829 goto out; 7830 } 7831 7832 if (ret == 0 && 7833 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7834 /* need to delete old one and insert a new one */ 7835 ret = btrfs_del_item(trans, dev_root, path); 7836 if (ret != 0) { 7837 btrfs_warn_in_rcu(fs_info, 7838 "delete too small dev_stats item for device %s failed %d", 7839 rcu_str_deref(device->name), ret); 7840 goto out; 7841 } 7842 ret = 1; 7843 } 7844 7845 if (ret == 1) { 7846 /* need to insert a new item */ 7847 btrfs_release_path(path); 7848 ret = btrfs_insert_empty_item(trans, dev_root, path, 7849 &key, sizeof(*ptr)); 7850 if (ret < 0) { 7851 btrfs_warn_in_rcu(fs_info, 7852 "insert dev_stats item for device %s failed %d", 7853 rcu_str_deref(device->name), ret); 7854 goto out; 7855 } 7856 } 7857 7858 eb = path->nodes[0]; 7859 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7860 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7861 btrfs_set_dev_stats_value(eb, ptr, i, 7862 btrfs_dev_stat_read(device, i)); 7863 btrfs_mark_buffer_dirty(eb); 7864 7865 out: 7866 btrfs_free_path(path); 7867 return ret; 7868 } 7869 7870 /* 7871 * called from commit_transaction. Writes all changed device stats to disk. 7872 */ 7873 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7874 { 7875 struct btrfs_fs_info *fs_info = trans->fs_info; 7876 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7877 struct btrfs_device *device; 7878 int stats_cnt; 7879 int ret = 0; 7880 7881 mutex_lock(&fs_devices->device_list_mutex); 7882 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7883 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7884 if (!device->dev_stats_valid || stats_cnt == 0) 7885 continue; 7886 7887 7888 /* 7889 * There is a LOAD-LOAD control dependency between the value of 7890 * dev_stats_ccnt and updating the on-disk values which requires 7891 * reading the in-memory counters. Such control dependencies 7892 * require explicit read memory barriers. 7893 * 7894 * This memory barriers pairs with smp_mb__before_atomic in 7895 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7896 * barrier implied by atomic_xchg in 7897 * btrfs_dev_stats_read_and_reset 7898 */ 7899 smp_rmb(); 7900 7901 ret = update_dev_stat_item(trans, device); 7902 if (!ret) 7903 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7904 } 7905 mutex_unlock(&fs_devices->device_list_mutex); 7906 7907 return ret; 7908 } 7909 7910 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7911 { 7912 btrfs_dev_stat_inc(dev, index); 7913 btrfs_dev_stat_print_on_error(dev); 7914 } 7915 7916 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) 7917 { 7918 if (!dev->dev_stats_valid) 7919 return; 7920 btrfs_err_rl_in_rcu(dev->fs_info, 7921 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7922 rcu_str_deref(dev->name), 7923 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7924 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7925 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7926 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7927 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7928 } 7929 7930 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7931 { 7932 int i; 7933 7934 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7935 if (btrfs_dev_stat_read(dev, i) != 0) 7936 break; 7937 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7938 return; /* all values == 0, suppress message */ 7939 7940 btrfs_info_in_rcu(dev->fs_info, 7941 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7942 rcu_str_deref(dev->name), 7943 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7944 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7945 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7946 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7947 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7948 } 7949 7950 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7951 struct btrfs_ioctl_get_dev_stats *stats) 7952 { 7953 BTRFS_DEV_LOOKUP_ARGS(args); 7954 struct btrfs_device *dev; 7955 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7956 int i; 7957 7958 mutex_lock(&fs_devices->device_list_mutex); 7959 args.devid = stats->devid; 7960 dev = btrfs_find_device(fs_info->fs_devices, &args); 7961 mutex_unlock(&fs_devices->device_list_mutex); 7962 7963 if (!dev) { 7964 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7965 return -ENODEV; 7966 } else if (!dev->dev_stats_valid) { 7967 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7968 return -ENODEV; 7969 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7970 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7971 if (stats->nr_items > i) 7972 stats->values[i] = 7973 btrfs_dev_stat_read_and_reset(dev, i); 7974 else 7975 btrfs_dev_stat_set(dev, i, 0); 7976 } 7977 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7978 current->comm, task_pid_nr(current)); 7979 } else { 7980 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7981 if (stats->nr_items > i) 7982 stats->values[i] = btrfs_dev_stat_read(dev, i); 7983 } 7984 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7985 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7986 return 0; 7987 } 7988 7989 /* 7990 * Update the size and bytes used for each device where it changed. This is 7991 * delayed since we would otherwise get errors while writing out the 7992 * superblocks. 7993 * 7994 * Must be invoked during transaction commit. 7995 */ 7996 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7997 { 7998 struct btrfs_device *curr, *next; 7999 8000 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 8001 8002 if (list_empty(&trans->dev_update_list)) 8003 return; 8004 8005 /* 8006 * We don't need the device_list_mutex here. This list is owned by the 8007 * transaction and the transaction must complete before the device is 8008 * released. 8009 */ 8010 mutex_lock(&trans->fs_info->chunk_mutex); 8011 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 8012 post_commit_list) { 8013 list_del_init(&curr->post_commit_list); 8014 curr->commit_total_bytes = curr->disk_total_bytes; 8015 curr->commit_bytes_used = curr->bytes_used; 8016 } 8017 mutex_unlock(&trans->fs_info->chunk_mutex); 8018 } 8019 8020 /* 8021 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 8022 */ 8023 int btrfs_bg_type_to_factor(u64 flags) 8024 { 8025 const int index = btrfs_bg_flags_to_raid_index(flags); 8026 8027 return btrfs_raid_array[index].ncopies; 8028 } 8029 8030 8031 8032 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 8033 u64 chunk_offset, u64 devid, 8034 u64 physical_offset, u64 physical_len) 8035 { 8036 struct btrfs_dev_lookup_args args = { .devid = devid }; 8037 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8038 struct extent_map *em; 8039 struct map_lookup *map; 8040 struct btrfs_device *dev; 8041 u64 stripe_len; 8042 bool found = false; 8043 int ret = 0; 8044 int i; 8045 8046 read_lock(&em_tree->lock); 8047 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 8048 read_unlock(&em_tree->lock); 8049 8050 if (!em) { 8051 btrfs_err(fs_info, 8052 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 8053 physical_offset, devid); 8054 ret = -EUCLEAN; 8055 goto out; 8056 } 8057 8058 map = em->map_lookup; 8059 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); 8060 if (physical_len != stripe_len) { 8061 btrfs_err(fs_info, 8062 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 8063 physical_offset, devid, em->start, physical_len, 8064 stripe_len); 8065 ret = -EUCLEAN; 8066 goto out; 8067 } 8068 8069 for (i = 0; i < map->num_stripes; i++) { 8070 if (map->stripes[i].dev->devid == devid && 8071 map->stripes[i].physical == physical_offset) { 8072 found = true; 8073 if (map->verified_stripes >= map->num_stripes) { 8074 btrfs_err(fs_info, 8075 "too many dev extents for chunk %llu found", 8076 em->start); 8077 ret = -EUCLEAN; 8078 goto out; 8079 } 8080 map->verified_stripes++; 8081 break; 8082 } 8083 } 8084 if (!found) { 8085 btrfs_err(fs_info, 8086 "dev extent physical offset %llu devid %llu has no corresponding chunk", 8087 physical_offset, devid); 8088 ret = -EUCLEAN; 8089 } 8090 8091 /* Make sure no dev extent is beyond device boundary */ 8092 dev = btrfs_find_device(fs_info->fs_devices, &args); 8093 if (!dev) { 8094 btrfs_err(fs_info, "failed to find devid %llu", devid); 8095 ret = -EUCLEAN; 8096 goto out; 8097 } 8098 8099 if (physical_offset + physical_len > dev->disk_total_bytes) { 8100 btrfs_err(fs_info, 8101 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 8102 devid, physical_offset, physical_len, 8103 dev->disk_total_bytes); 8104 ret = -EUCLEAN; 8105 goto out; 8106 } 8107 8108 if (dev->zone_info) { 8109 u64 zone_size = dev->zone_info->zone_size; 8110 8111 if (!IS_ALIGNED(physical_offset, zone_size) || 8112 !IS_ALIGNED(physical_len, zone_size)) { 8113 btrfs_err(fs_info, 8114 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8115 devid, physical_offset, physical_len); 8116 ret = -EUCLEAN; 8117 goto out; 8118 } 8119 } 8120 8121 out: 8122 free_extent_map(em); 8123 return ret; 8124 } 8125 8126 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8127 { 8128 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8129 struct extent_map *em; 8130 struct rb_node *node; 8131 int ret = 0; 8132 8133 read_lock(&em_tree->lock); 8134 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8135 em = rb_entry(node, struct extent_map, rb_node); 8136 if (em->map_lookup->num_stripes != 8137 em->map_lookup->verified_stripes) { 8138 btrfs_err(fs_info, 8139 "chunk %llu has missing dev extent, have %d expect %d", 8140 em->start, em->map_lookup->verified_stripes, 8141 em->map_lookup->num_stripes); 8142 ret = -EUCLEAN; 8143 goto out; 8144 } 8145 } 8146 out: 8147 read_unlock(&em_tree->lock); 8148 return ret; 8149 } 8150 8151 /* 8152 * Ensure that all dev extents are mapped to correct chunk, otherwise 8153 * later chunk allocation/free would cause unexpected behavior. 8154 * 8155 * NOTE: This will iterate through the whole device tree, which should be of 8156 * the same size level as the chunk tree. This slightly increases mount time. 8157 */ 8158 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8159 { 8160 struct btrfs_path *path; 8161 struct btrfs_root *root = fs_info->dev_root; 8162 struct btrfs_key key; 8163 u64 prev_devid = 0; 8164 u64 prev_dev_ext_end = 0; 8165 int ret = 0; 8166 8167 /* 8168 * We don't have a dev_root because we mounted with ignorebadroots and 8169 * failed to load the root, so we want to skip the verification in this 8170 * case for sure. 8171 * 8172 * However if the dev root is fine, but the tree itself is corrupted 8173 * we'd still fail to mount. This verification is only to make sure 8174 * writes can happen safely, so instead just bypass this check 8175 * completely in the case of IGNOREBADROOTS. 8176 */ 8177 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8178 return 0; 8179 8180 key.objectid = 1; 8181 key.type = BTRFS_DEV_EXTENT_KEY; 8182 key.offset = 0; 8183 8184 path = btrfs_alloc_path(); 8185 if (!path) 8186 return -ENOMEM; 8187 8188 path->reada = READA_FORWARD; 8189 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8190 if (ret < 0) 8191 goto out; 8192 8193 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8194 ret = btrfs_next_leaf(root, path); 8195 if (ret < 0) 8196 goto out; 8197 /* No dev extents at all? Not good */ 8198 if (ret > 0) { 8199 ret = -EUCLEAN; 8200 goto out; 8201 } 8202 } 8203 while (1) { 8204 struct extent_buffer *leaf = path->nodes[0]; 8205 struct btrfs_dev_extent *dext; 8206 int slot = path->slots[0]; 8207 u64 chunk_offset; 8208 u64 physical_offset; 8209 u64 physical_len; 8210 u64 devid; 8211 8212 btrfs_item_key_to_cpu(leaf, &key, slot); 8213 if (key.type != BTRFS_DEV_EXTENT_KEY) 8214 break; 8215 devid = key.objectid; 8216 physical_offset = key.offset; 8217 8218 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8219 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8220 physical_len = btrfs_dev_extent_length(leaf, dext); 8221 8222 /* Check if this dev extent overlaps with the previous one */ 8223 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8224 btrfs_err(fs_info, 8225 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8226 devid, physical_offset, prev_dev_ext_end); 8227 ret = -EUCLEAN; 8228 goto out; 8229 } 8230 8231 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8232 physical_offset, physical_len); 8233 if (ret < 0) 8234 goto out; 8235 prev_devid = devid; 8236 prev_dev_ext_end = physical_offset + physical_len; 8237 8238 ret = btrfs_next_item(root, path); 8239 if (ret < 0) 8240 goto out; 8241 if (ret > 0) { 8242 ret = 0; 8243 break; 8244 } 8245 } 8246 8247 /* Ensure all chunks have corresponding dev extents */ 8248 ret = verify_chunk_dev_extent_mapping(fs_info); 8249 out: 8250 btrfs_free_path(path); 8251 return ret; 8252 } 8253 8254 /* 8255 * Check whether the given block group or device is pinned by any inode being 8256 * used as a swapfile. 8257 */ 8258 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8259 { 8260 struct btrfs_swapfile_pin *sp; 8261 struct rb_node *node; 8262 8263 spin_lock(&fs_info->swapfile_pins_lock); 8264 node = fs_info->swapfile_pins.rb_node; 8265 while (node) { 8266 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8267 if (ptr < sp->ptr) 8268 node = node->rb_left; 8269 else if (ptr > sp->ptr) 8270 node = node->rb_right; 8271 else 8272 break; 8273 } 8274 spin_unlock(&fs_info->swapfile_pins_lock); 8275 return node != NULL; 8276 } 8277 8278 static int relocating_repair_kthread(void *data) 8279 { 8280 struct btrfs_block_group *cache = (struct btrfs_block_group *)data; 8281 struct btrfs_fs_info *fs_info = cache->fs_info; 8282 u64 target; 8283 int ret = 0; 8284 8285 target = cache->start; 8286 btrfs_put_block_group(cache); 8287 8288 sb_start_write(fs_info->sb); 8289 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8290 btrfs_info(fs_info, 8291 "zoned: skip relocating block group %llu to repair: EBUSY", 8292 target); 8293 sb_end_write(fs_info->sb); 8294 return -EBUSY; 8295 } 8296 8297 mutex_lock(&fs_info->reclaim_bgs_lock); 8298 8299 /* Ensure block group still exists */ 8300 cache = btrfs_lookup_block_group(fs_info, target); 8301 if (!cache) 8302 goto out; 8303 8304 if (!cache->relocating_repair) 8305 goto out; 8306 8307 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8308 if (ret < 0) 8309 goto out; 8310 8311 btrfs_info(fs_info, 8312 "zoned: relocating block group %llu to repair IO failure", 8313 target); 8314 ret = btrfs_relocate_chunk(fs_info, target); 8315 8316 out: 8317 if (cache) 8318 btrfs_put_block_group(cache); 8319 mutex_unlock(&fs_info->reclaim_bgs_lock); 8320 btrfs_exclop_finish(fs_info); 8321 sb_end_write(fs_info->sb); 8322 8323 return ret; 8324 } 8325 8326 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8327 { 8328 struct btrfs_block_group *cache; 8329 8330 if (!btrfs_is_zoned(fs_info)) 8331 return false; 8332 8333 /* Do not attempt to repair in degraded state */ 8334 if (btrfs_test_opt(fs_info, DEGRADED)) 8335 return true; 8336 8337 cache = btrfs_lookup_block_group(fs_info, logical); 8338 if (!cache) 8339 return true; 8340 8341 spin_lock(&cache->lock); 8342 if (cache->relocating_repair) { 8343 spin_unlock(&cache->lock); 8344 btrfs_put_block_group(cache); 8345 return true; 8346 } 8347 cache->relocating_repair = 1; 8348 spin_unlock(&cache->lock); 8349 8350 kthread_run(relocating_repair_kthread, cache, 8351 "btrfs-relocating-repair"); 8352 8353 return true; 8354 } 8355