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 "misc.h" 18 #include "ctree.h" 19 #include "extent_map.h" 20 #include "disk-io.h" 21 #include "transaction.h" 22 #include "print-tree.h" 23 #include "volumes.h" 24 #include "raid56.h" 25 #include "async-thread.h" 26 #include "check-integrity.h" 27 #include "rcu-string.h" 28 #include "dev-replace.h" 29 #include "sysfs.h" 30 #include "tree-checker.h" 31 #include "space-info.h" 32 #include "block-group.h" 33 #include "discard.h" 34 #include "zoned.h" 35 36 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { 37 [BTRFS_RAID_RAID10] = { 38 .sub_stripes = 2, 39 .dev_stripes = 1, 40 .devs_max = 0, /* 0 == as many as possible */ 41 .devs_min = 2, 42 .tolerated_failures = 1, 43 .devs_increment = 2, 44 .ncopies = 2, 45 .nparity = 0, 46 .raid_name = "raid10", 47 .bg_flag = BTRFS_BLOCK_GROUP_RAID10, 48 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, 49 }, 50 [BTRFS_RAID_RAID1] = { 51 .sub_stripes = 1, 52 .dev_stripes = 1, 53 .devs_max = 2, 54 .devs_min = 2, 55 .tolerated_failures = 1, 56 .devs_increment = 2, 57 .ncopies = 2, 58 .nparity = 0, 59 .raid_name = "raid1", 60 .bg_flag = BTRFS_BLOCK_GROUP_RAID1, 61 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, 62 }, 63 [BTRFS_RAID_RAID1C3] = { 64 .sub_stripes = 1, 65 .dev_stripes = 1, 66 .devs_max = 3, 67 .devs_min = 3, 68 .tolerated_failures = 2, 69 .devs_increment = 3, 70 .ncopies = 3, 71 .nparity = 0, 72 .raid_name = "raid1c3", 73 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, 74 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, 75 }, 76 [BTRFS_RAID_RAID1C4] = { 77 .sub_stripes = 1, 78 .dev_stripes = 1, 79 .devs_max = 4, 80 .devs_min = 4, 81 .tolerated_failures = 3, 82 .devs_increment = 4, 83 .ncopies = 4, 84 .nparity = 0, 85 .raid_name = "raid1c4", 86 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, 87 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, 88 }, 89 [BTRFS_RAID_DUP] = { 90 .sub_stripes = 1, 91 .dev_stripes = 2, 92 .devs_max = 1, 93 .devs_min = 1, 94 .tolerated_failures = 0, 95 .devs_increment = 1, 96 .ncopies = 2, 97 .nparity = 0, 98 .raid_name = "dup", 99 .bg_flag = BTRFS_BLOCK_GROUP_DUP, 100 .mindev_error = 0, 101 }, 102 [BTRFS_RAID_RAID0] = { 103 .sub_stripes = 1, 104 .dev_stripes = 1, 105 .devs_max = 0, 106 .devs_min = 1, 107 .tolerated_failures = 0, 108 .devs_increment = 1, 109 .ncopies = 1, 110 .nparity = 0, 111 .raid_name = "raid0", 112 .bg_flag = BTRFS_BLOCK_GROUP_RAID0, 113 .mindev_error = 0, 114 }, 115 [BTRFS_RAID_SINGLE] = { 116 .sub_stripes = 1, 117 .dev_stripes = 1, 118 .devs_max = 1, 119 .devs_min = 1, 120 .tolerated_failures = 0, 121 .devs_increment = 1, 122 .ncopies = 1, 123 .nparity = 0, 124 .raid_name = "single", 125 .bg_flag = 0, 126 .mindev_error = 0, 127 }, 128 [BTRFS_RAID_RAID5] = { 129 .sub_stripes = 1, 130 .dev_stripes = 1, 131 .devs_max = 0, 132 .devs_min = 2, 133 .tolerated_failures = 1, 134 .devs_increment = 1, 135 .ncopies = 1, 136 .nparity = 1, 137 .raid_name = "raid5", 138 .bg_flag = BTRFS_BLOCK_GROUP_RAID5, 139 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, 140 }, 141 [BTRFS_RAID_RAID6] = { 142 .sub_stripes = 1, 143 .dev_stripes = 1, 144 .devs_max = 0, 145 .devs_min = 3, 146 .tolerated_failures = 2, 147 .devs_increment = 1, 148 .ncopies = 1, 149 .nparity = 2, 150 .raid_name = "raid6", 151 .bg_flag = BTRFS_BLOCK_GROUP_RAID6, 152 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, 153 }, 154 }; 155 156 /* 157 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 158 * can be used as index to access btrfs_raid_array[]. 159 */ 160 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) 161 { 162 if (flags & BTRFS_BLOCK_GROUP_RAID10) 163 return BTRFS_RAID_RAID10; 164 else if (flags & BTRFS_BLOCK_GROUP_RAID1) 165 return BTRFS_RAID_RAID1; 166 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3) 167 return BTRFS_RAID_RAID1C3; 168 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4) 169 return BTRFS_RAID_RAID1C4; 170 else if (flags & BTRFS_BLOCK_GROUP_DUP) 171 return BTRFS_RAID_DUP; 172 else if (flags & BTRFS_BLOCK_GROUP_RAID0) 173 return BTRFS_RAID_RAID0; 174 else if (flags & BTRFS_BLOCK_GROUP_RAID5) 175 return BTRFS_RAID_RAID5; 176 else if (flags & BTRFS_BLOCK_GROUP_RAID6) 177 return BTRFS_RAID_RAID6; 178 179 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ 180 } 181 182 const char *btrfs_bg_type_to_raid_name(u64 flags) 183 { 184 const int index = btrfs_bg_flags_to_raid_index(flags); 185 186 if (index >= BTRFS_NR_RAID_TYPES) 187 return NULL; 188 189 return btrfs_raid_array[index].raid_name; 190 } 191 192 /* 193 * Fill @buf with textual description of @bg_flags, no more than @size_buf 194 * bytes including terminating null byte. 195 */ 196 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) 197 { 198 int i; 199 int ret; 200 char *bp = buf; 201 u64 flags = bg_flags; 202 u32 size_bp = size_buf; 203 204 if (!flags) { 205 strcpy(bp, "NONE"); 206 return; 207 } 208 209 #define DESCRIBE_FLAG(flag, desc) \ 210 do { \ 211 if (flags & (flag)) { \ 212 ret = snprintf(bp, size_bp, "%s|", (desc)); \ 213 if (ret < 0 || ret >= size_bp) \ 214 goto out_overflow; \ 215 size_bp -= ret; \ 216 bp += ret; \ 217 flags &= ~(flag); \ 218 } \ 219 } while (0) 220 221 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); 222 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); 223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); 224 225 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); 226 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) 227 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, 228 btrfs_raid_array[i].raid_name); 229 #undef DESCRIBE_FLAG 230 231 if (flags) { 232 ret = snprintf(bp, size_bp, "0x%llx|", flags); 233 size_bp -= ret; 234 } 235 236 if (size_bp < size_buf) 237 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ 238 239 /* 240 * The text is trimmed, it's up to the caller to provide sufficiently 241 * large buffer 242 */ 243 out_overflow:; 244 } 245 246 static int init_first_rw_device(struct btrfs_trans_handle *trans); 247 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); 248 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); 249 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); 250 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 251 enum btrfs_map_op op, 252 u64 logical, u64 *length, 253 struct btrfs_io_context **bioc_ret, 254 int mirror_num, int need_raid_map); 255 256 /* 257 * Device locking 258 * ============== 259 * 260 * There are several mutexes that protect manipulation of devices and low-level 261 * structures like chunks but not block groups, extents or files 262 * 263 * uuid_mutex (global lock) 264 * ------------------------ 265 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from 266 * the SCAN_DEV ioctl registration or from mount either implicitly (the first 267 * device) or requested by the device= mount option 268 * 269 * the mutex can be very coarse and can cover long-running operations 270 * 271 * protects: updates to fs_devices counters like missing devices, rw devices, 272 * seeding, structure cloning, opening/closing devices at mount/umount time 273 * 274 * global::fs_devs - add, remove, updates to the global list 275 * 276 * does not protect: manipulation of the fs_devices::devices list in general 277 * but in mount context it could be used to exclude list modifications by eg. 278 * scan ioctl 279 * 280 * btrfs_device::name - renames (write side), read is RCU 281 * 282 * fs_devices::device_list_mutex (per-fs, with RCU) 283 * ------------------------------------------------ 284 * protects updates to fs_devices::devices, ie. adding and deleting 285 * 286 * simple list traversal with read-only actions can be done with RCU protection 287 * 288 * may be used to exclude some operations from running concurrently without any 289 * modifications to the list (see write_all_supers) 290 * 291 * Is not required at mount and close times, because our device list is 292 * protected by the uuid_mutex at that point. 293 * 294 * balance_mutex 295 * ------------- 296 * protects balance structures (status, state) and context accessed from 297 * several places (internally, ioctl) 298 * 299 * chunk_mutex 300 * ----------- 301 * protects chunks, adding or removing during allocation, trim or when a new 302 * device is added/removed. Additionally it also protects post_commit_list of 303 * individual devices, since they can be added to the transaction's 304 * post_commit_list only with chunk_mutex held. 305 * 306 * cleaner_mutex 307 * ------------- 308 * a big lock that is held by the cleaner thread and prevents running subvolume 309 * cleaning together with relocation or delayed iputs 310 * 311 * 312 * Lock nesting 313 * ============ 314 * 315 * uuid_mutex 316 * device_list_mutex 317 * chunk_mutex 318 * balance_mutex 319 * 320 * 321 * Exclusive operations 322 * ==================== 323 * 324 * Maintains the exclusivity of the following operations that apply to the 325 * whole filesystem and cannot run in parallel. 326 * 327 * - Balance (*) 328 * - Device add 329 * - Device remove 330 * - Device replace (*) 331 * - Resize 332 * 333 * The device operations (as above) can be in one of the following states: 334 * 335 * - Running state 336 * - Paused state 337 * - Completed state 338 * 339 * Only device operations marked with (*) can go into the Paused state for the 340 * following reasons: 341 * 342 * - ioctl (only Balance can be Paused through ioctl) 343 * - filesystem remounted as read-only 344 * - filesystem unmounted and mounted as read-only 345 * - system power-cycle and filesystem mounted as read-only 346 * - filesystem or device errors leading to forced read-only 347 * 348 * The status of exclusive operation is set and cleared atomically. 349 * During the course of Paused state, fs_info::exclusive_operation remains set. 350 * A device operation in Paused or Running state can be canceled or resumed 351 * either by ioctl (Balance only) or when remounted as read-write. 352 * The exclusive status is cleared when the device operation is canceled or 353 * completed. 354 */ 355 356 DEFINE_MUTEX(uuid_mutex); 357 static LIST_HEAD(fs_uuids); 358 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) 359 { 360 return &fs_uuids; 361 } 362 363 /* 364 * alloc_fs_devices - allocate struct btrfs_fs_devices 365 * @fsid: if not NULL, copy the UUID to fs_devices::fsid 366 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid 367 * 368 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). 369 * The returned struct is not linked onto any lists and can be destroyed with 370 * kfree() right away. 371 */ 372 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, 373 const u8 *metadata_fsid) 374 { 375 struct btrfs_fs_devices *fs_devs; 376 377 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 378 if (!fs_devs) 379 return ERR_PTR(-ENOMEM); 380 381 mutex_init(&fs_devs->device_list_mutex); 382 383 INIT_LIST_HEAD(&fs_devs->devices); 384 INIT_LIST_HEAD(&fs_devs->alloc_list); 385 INIT_LIST_HEAD(&fs_devs->fs_list); 386 INIT_LIST_HEAD(&fs_devs->seed_list); 387 if (fsid) 388 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 389 390 if (metadata_fsid) 391 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); 392 else if (fsid) 393 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); 394 395 return fs_devs; 396 } 397 398 void btrfs_free_device(struct btrfs_device *device) 399 { 400 WARN_ON(!list_empty(&device->post_commit_list)); 401 rcu_string_free(device->name); 402 extent_io_tree_release(&device->alloc_state); 403 bio_put(device->flush_bio); 404 btrfs_destroy_dev_zone_info(device); 405 kfree(device); 406 } 407 408 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 409 { 410 struct btrfs_device *device; 411 WARN_ON(fs_devices->opened); 412 while (!list_empty(&fs_devices->devices)) { 413 device = list_entry(fs_devices->devices.next, 414 struct btrfs_device, dev_list); 415 list_del(&device->dev_list); 416 btrfs_free_device(device); 417 } 418 kfree(fs_devices); 419 } 420 421 void __exit btrfs_cleanup_fs_uuids(void) 422 { 423 struct btrfs_fs_devices *fs_devices; 424 425 while (!list_empty(&fs_uuids)) { 426 fs_devices = list_entry(fs_uuids.next, 427 struct btrfs_fs_devices, fs_list); 428 list_del(&fs_devices->fs_list); 429 free_fs_devices(fs_devices); 430 } 431 } 432 433 static noinline struct btrfs_fs_devices *find_fsid( 434 const u8 *fsid, const u8 *metadata_fsid) 435 { 436 struct btrfs_fs_devices *fs_devices; 437 438 ASSERT(fsid); 439 440 /* Handle non-split brain cases */ 441 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 442 if (metadata_fsid) { 443 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 444 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 445 BTRFS_FSID_SIZE) == 0) 446 return fs_devices; 447 } else { 448 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 449 return fs_devices; 450 } 451 } 452 return NULL; 453 } 454 455 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 456 struct btrfs_super_block *disk_super) 457 { 458 459 struct btrfs_fs_devices *fs_devices; 460 461 /* 462 * Handle scanned device having completed its fsid change but 463 * belonging to a fs_devices that was created by first scanning 464 * a device which didn't have its fsid/metadata_uuid changed 465 * at all and the CHANGING_FSID_V2 flag set. 466 */ 467 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 468 if (fs_devices->fsid_change && 469 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 470 BTRFS_FSID_SIZE) == 0 && 471 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 472 BTRFS_FSID_SIZE) == 0) { 473 return fs_devices; 474 } 475 } 476 /* 477 * Handle scanned device having completed its fsid change but 478 * belonging to a fs_devices that was created by a device that 479 * has an outdated pair of fsid/metadata_uuid and 480 * CHANGING_FSID_V2 flag set. 481 */ 482 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 483 if (fs_devices->fsid_change && 484 memcmp(fs_devices->metadata_uuid, 485 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 486 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 487 BTRFS_FSID_SIZE) == 0) { 488 return fs_devices; 489 } 490 } 491 492 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 493 } 494 495 496 static int 497 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 498 int flush, struct block_device **bdev, 499 struct btrfs_super_block **disk_super) 500 { 501 int ret; 502 503 *bdev = blkdev_get_by_path(device_path, flags, holder); 504 505 if (IS_ERR(*bdev)) { 506 ret = PTR_ERR(*bdev); 507 goto error; 508 } 509 510 if (flush) 511 filemap_write_and_wait((*bdev)->bd_inode->i_mapping); 512 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 513 if (ret) { 514 blkdev_put(*bdev, flags); 515 goto error; 516 } 517 invalidate_bdev(*bdev); 518 *disk_super = btrfs_read_dev_super(*bdev); 519 if (IS_ERR(*disk_super)) { 520 ret = PTR_ERR(*disk_super); 521 blkdev_put(*bdev, flags); 522 goto error; 523 } 524 525 return 0; 526 527 error: 528 *bdev = NULL; 529 return ret; 530 } 531 532 static bool device_path_matched(const char *path, struct btrfs_device *device) 533 { 534 int found; 535 536 rcu_read_lock(); 537 found = strcmp(rcu_str_deref(device->name), path); 538 rcu_read_unlock(); 539 540 return found == 0; 541 } 542 543 /* 544 * Search and remove all stale (devices which are not mounted) devices. 545 * When both inputs are NULL, it will search and release all stale devices. 546 * path: Optional. When provided will it release all unmounted devices 547 * matching this path only. 548 * skip_dev: Optional. Will skip this device when searching for the stale 549 * devices. 550 * Return: 0 for success or if @path is NULL. 551 * -EBUSY if @path is a mounted device. 552 * -ENOENT if @path does not match any device in the list. 553 */ 554 static int btrfs_free_stale_devices(const char *path, 555 struct btrfs_device *skip_device) 556 { 557 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 558 struct btrfs_device *device, *tmp_device; 559 int ret = 0; 560 561 lockdep_assert_held(&uuid_mutex); 562 563 if (path) 564 ret = -ENOENT; 565 566 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 567 568 mutex_lock(&fs_devices->device_list_mutex); 569 list_for_each_entry_safe(device, tmp_device, 570 &fs_devices->devices, dev_list) { 571 if (skip_device && skip_device == device) 572 continue; 573 if (path && !device->name) 574 continue; 575 if (path && !device_path_matched(path, device)) 576 continue; 577 if (fs_devices->opened) { 578 /* for an already deleted device return 0 */ 579 if (path && ret != 0) 580 ret = -EBUSY; 581 break; 582 } 583 584 /* delete the stale device */ 585 fs_devices->num_devices--; 586 list_del(&device->dev_list); 587 btrfs_free_device(device); 588 589 ret = 0; 590 } 591 mutex_unlock(&fs_devices->device_list_mutex); 592 593 if (fs_devices->num_devices == 0) { 594 btrfs_sysfs_remove_fsid(fs_devices); 595 list_del(&fs_devices->fs_list); 596 free_fs_devices(fs_devices); 597 } 598 } 599 600 return ret; 601 } 602 603 /* 604 * This is only used on mount, and we are protected from competing things 605 * messing with our fs_devices by the uuid_mutex, thus we do not need the 606 * fs_devices->device_list_mutex here. 607 */ 608 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 609 struct btrfs_device *device, fmode_t flags, 610 void *holder) 611 { 612 struct request_queue *q; 613 struct block_device *bdev; 614 struct btrfs_super_block *disk_super; 615 u64 devid; 616 int ret; 617 618 if (device->bdev) 619 return -EINVAL; 620 if (!device->name) 621 return -EINVAL; 622 623 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 624 &bdev, &disk_super); 625 if (ret) 626 return ret; 627 628 devid = btrfs_stack_device_id(&disk_super->dev_item); 629 if (devid != device->devid) 630 goto error_free_page; 631 632 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 633 goto error_free_page; 634 635 device->generation = btrfs_super_generation(disk_super); 636 637 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 638 if (btrfs_super_incompat_flags(disk_super) & 639 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 640 pr_err( 641 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 642 goto error_free_page; 643 } 644 645 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 646 fs_devices->seeding = true; 647 } else { 648 if (bdev_read_only(bdev)) 649 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 650 else 651 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 652 } 653 654 q = bdev_get_queue(bdev); 655 if (!blk_queue_nonrot(q)) 656 fs_devices->rotating = true; 657 658 device->bdev = bdev; 659 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 660 device->mode = flags; 661 662 fs_devices->open_devices++; 663 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 664 device->devid != BTRFS_DEV_REPLACE_DEVID) { 665 fs_devices->rw_devices++; 666 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 667 } 668 btrfs_release_disk_super(disk_super); 669 670 return 0; 671 672 error_free_page: 673 btrfs_release_disk_super(disk_super); 674 blkdev_put(bdev, flags); 675 676 return -EINVAL; 677 } 678 679 /* 680 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 681 * being created with a disk that has already completed its fsid change. Such 682 * disk can belong to an fs which has its FSID changed or to one which doesn't. 683 * Handle both cases here. 684 */ 685 static struct btrfs_fs_devices *find_fsid_inprogress( 686 struct btrfs_super_block *disk_super) 687 { 688 struct btrfs_fs_devices *fs_devices; 689 690 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 691 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 692 BTRFS_FSID_SIZE) != 0 && 693 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 694 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 695 return fs_devices; 696 } 697 } 698 699 return find_fsid(disk_super->fsid, NULL); 700 } 701 702 703 static struct btrfs_fs_devices *find_fsid_changed( 704 struct btrfs_super_block *disk_super) 705 { 706 struct btrfs_fs_devices *fs_devices; 707 708 /* 709 * Handles the case where scanned device is part of an fs that had 710 * multiple successful changes of FSID but currently device didn't 711 * observe it. Meaning our fsid will be different than theirs. We need 712 * to handle two subcases : 713 * 1 - The fs still continues to have different METADATA/FSID uuids. 714 * 2 - The fs is switched back to its original FSID (METADATA/FSID 715 * are equal). 716 */ 717 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 718 /* Changed UUIDs */ 719 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 720 BTRFS_FSID_SIZE) != 0 && 721 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 722 BTRFS_FSID_SIZE) == 0 && 723 memcmp(fs_devices->fsid, disk_super->fsid, 724 BTRFS_FSID_SIZE) != 0) 725 return fs_devices; 726 727 /* Unchanged UUIDs */ 728 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 729 BTRFS_FSID_SIZE) == 0 && 730 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 731 BTRFS_FSID_SIZE) == 0) 732 return fs_devices; 733 } 734 735 return NULL; 736 } 737 738 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 739 struct btrfs_super_block *disk_super) 740 { 741 struct btrfs_fs_devices *fs_devices; 742 743 /* 744 * Handle the case where the scanned device is part of an fs whose last 745 * metadata UUID change reverted it to the original FSID. At the same 746 * time * fs_devices was first created by another constitutent device 747 * which didn't fully observe the operation. This results in an 748 * btrfs_fs_devices created with metadata/fsid different AND 749 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 750 * fs_devices equal to the FSID of the disk. 751 */ 752 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 753 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 754 BTRFS_FSID_SIZE) != 0 && 755 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 756 BTRFS_FSID_SIZE) == 0 && 757 fs_devices->fsid_change) 758 return fs_devices; 759 } 760 761 return NULL; 762 } 763 /* 764 * Add new device to list of registered devices 765 * 766 * Returns: 767 * device pointer which was just added or updated when successful 768 * error pointer when failed 769 */ 770 static noinline struct btrfs_device *device_list_add(const char *path, 771 struct btrfs_super_block *disk_super, 772 bool *new_device_added) 773 { 774 struct btrfs_device *device; 775 struct btrfs_fs_devices *fs_devices = NULL; 776 struct rcu_string *name; 777 u64 found_transid = btrfs_super_generation(disk_super); 778 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 779 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 780 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 781 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 782 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 783 784 if (fsid_change_in_progress) { 785 if (!has_metadata_uuid) 786 fs_devices = find_fsid_inprogress(disk_super); 787 else 788 fs_devices = find_fsid_changed(disk_super); 789 } else if (has_metadata_uuid) { 790 fs_devices = find_fsid_with_metadata_uuid(disk_super); 791 } else { 792 fs_devices = find_fsid_reverted_metadata(disk_super); 793 if (!fs_devices) 794 fs_devices = find_fsid(disk_super->fsid, NULL); 795 } 796 797 798 if (!fs_devices) { 799 if (has_metadata_uuid) 800 fs_devices = alloc_fs_devices(disk_super->fsid, 801 disk_super->metadata_uuid); 802 else 803 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 804 805 if (IS_ERR(fs_devices)) 806 return ERR_CAST(fs_devices); 807 808 fs_devices->fsid_change = fsid_change_in_progress; 809 810 mutex_lock(&fs_devices->device_list_mutex); 811 list_add(&fs_devices->fs_list, &fs_uuids); 812 813 device = NULL; 814 } else { 815 struct btrfs_dev_lookup_args args = { 816 .devid = devid, 817 .uuid = disk_super->dev_item.uuid, 818 }; 819 820 mutex_lock(&fs_devices->device_list_mutex); 821 device = btrfs_find_device(fs_devices, &args); 822 823 /* 824 * If this disk has been pulled into an fs devices created by 825 * a device which had the CHANGING_FSID_V2 flag then replace the 826 * metadata_uuid/fsid values of the fs_devices. 827 */ 828 if (fs_devices->fsid_change && 829 found_transid > fs_devices->latest_generation) { 830 memcpy(fs_devices->fsid, disk_super->fsid, 831 BTRFS_FSID_SIZE); 832 833 if (has_metadata_uuid) 834 memcpy(fs_devices->metadata_uuid, 835 disk_super->metadata_uuid, 836 BTRFS_FSID_SIZE); 837 else 838 memcpy(fs_devices->metadata_uuid, 839 disk_super->fsid, BTRFS_FSID_SIZE); 840 841 fs_devices->fsid_change = false; 842 } 843 } 844 845 if (!device) { 846 if (fs_devices->opened) { 847 mutex_unlock(&fs_devices->device_list_mutex); 848 return ERR_PTR(-EBUSY); 849 } 850 851 device = btrfs_alloc_device(NULL, &devid, 852 disk_super->dev_item.uuid); 853 if (IS_ERR(device)) { 854 mutex_unlock(&fs_devices->device_list_mutex); 855 /* we can safely leave the fs_devices entry around */ 856 return device; 857 } 858 859 name = rcu_string_strdup(path, GFP_NOFS); 860 if (!name) { 861 btrfs_free_device(device); 862 mutex_unlock(&fs_devices->device_list_mutex); 863 return ERR_PTR(-ENOMEM); 864 } 865 rcu_assign_pointer(device->name, name); 866 867 list_add_rcu(&device->dev_list, &fs_devices->devices); 868 fs_devices->num_devices++; 869 870 device->fs_devices = fs_devices; 871 *new_device_added = true; 872 873 if (disk_super->label[0]) 874 pr_info( 875 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 876 disk_super->label, devid, found_transid, path, 877 current->comm, task_pid_nr(current)); 878 else 879 pr_info( 880 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 881 disk_super->fsid, devid, found_transid, path, 882 current->comm, task_pid_nr(current)); 883 884 } else if (!device->name || strcmp(device->name->str, path)) { 885 /* 886 * When FS is already mounted. 887 * 1. If you are here and if the device->name is NULL that 888 * means this device was missing at time of FS mount. 889 * 2. If you are here and if the device->name is different 890 * from 'path' that means either 891 * a. The same device disappeared and reappeared with 892 * different name. or 893 * b. The missing-disk-which-was-replaced, has 894 * reappeared now. 895 * 896 * We must allow 1 and 2a above. But 2b would be a spurious 897 * and unintentional. 898 * 899 * Further in case of 1 and 2a above, the disk at 'path' 900 * would have missed some transaction when it was away and 901 * in case of 2a the stale bdev has to be updated as well. 902 * 2b must not be allowed at all time. 903 */ 904 905 /* 906 * For now, we do allow update to btrfs_fs_device through the 907 * btrfs dev scan cli after FS has been mounted. We're still 908 * tracking a problem where systems fail mount by subvolume id 909 * when we reject replacement on a mounted FS. 910 */ 911 if (!fs_devices->opened && found_transid < device->generation) { 912 /* 913 * That is if the FS is _not_ mounted and if you 914 * are here, that means there is more than one 915 * disk with same uuid and devid.We keep the one 916 * with larger generation number or the last-in if 917 * generation are equal. 918 */ 919 mutex_unlock(&fs_devices->device_list_mutex); 920 return ERR_PTR(-EEXIST); 921 } 922 923 /* 924 * We are going to replace the device path for a given devid, 925 * make sure it's the same device if the device is mounted 926 */ 927 if (device->bdev) { 928 int error; 929 dev_t path_dev; 930 931 error = lookup_bdev(path, &path_dev); 932 if (error) { 933 mutex_unlock(&fs_devices->device_list_mutex); 934 return ERR_PTR(error); 935 } 936 937 if (device->bdev->bd_dev != path_dev) { 938 mutex_unlock(&fs_devices->device_list_mutex); 939 /* 940 * device->fs_info may not be reliable here, so 941 * pass in a NULL instead. This avoids a 942 * possible use-after-free when the fs_info and 943 * fs_info->sb are already torn down. 944 */ 945 btrfs_warn_in_rcu(NULL, 946 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 947 path, devid, found_transid, 948 current->comm, 949 task_pid_nr(current)); 950 return ERR_PTR(-EEXIST); 951 } 952 btrfs_info_in_rcu(device->fs_info, 953 "devid %llu device path %s changed to %s scanned by %s (%d)", 954 devid, rcu_str_deref(device->name), 955 path, current->comm, 956 task_pid_nr(current)); 957 } 958 959 name = rcu_string_strdup(path, GFP_NOFS); 960 if (!name) { 961 mutex_unlock(&fs_devices->device_list_mutex); 962 return ERR_PTR(-ENOMEM); 963 } 964 rcu_string_free(device->name); 965 rcu_assign_pointer(device->name, name); 966 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 967 fs_devices->missing_devices--; 968 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 969 } 970 } 971 972 /* 973 * Unmount does not free the btrfs_device struct but would zero 974 * generation along with most of the other members. So just update 975 * it back. We need it to pick the disk with largest generation 976 * (as above). 977 */ 978 if (!fs_devices->opened) { 979 device->generation = found_transid; 980 fs_devices->latest_generation = max_t(u64, found_transid, 981 fs_devices->latest_generation); 982 } 983 984 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 985 986 mutex_unlock(&fs_devices->device_list_mutex); 987 return device; 988 } 989 990 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 991 { 992 struct btrfs_fs_devices *fs_devices; 993 struct btrfs_device *device; 994 struct btrfs_device *orig_dev; 995 int ret = 0; 996 997 lockdep_assert_held(&uuid_mutex); 998 999 fs_devices = alloc_fs_devices(orig->fsid, NULL); 1000 if (IS_ERR(fs_devices)) 1001 return fs_devices; 1002 1003 fs_devices->total_devices = orig->total_devices; 1004 1005 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 1006 struct rcu_string *name; 1007 1008 device = btrfs_alloc_device(NULL, &orig_dev->devid, 1009 orig_dev->uuid); 1010 if (IS_ERR(device)) { 1011 ret = PTR_ERR(device); 1012 goto error; 1013 } 1014 1015 /* 1016 * This is ok to do without rcu read locked because we hold the 1017 * uuid mutex so nothing we touch in here is going to disappear. 1018 */ 1019 if (orig_dev->name) { 1020 name = rcu_string_strdup(orig_dev->name->str, 1021 GFP_KERNEL); 1022 if (!name) { 1023 btrfs_free_device(device); 1024 ret = -ENOMEM; 1025 goto error; 1026 } 1027 rcu_assign_pointer(device->name, name); 1028 } 1029 1030 list_add(&device->dev_list, &fs_devices->devices); 1031 device->fs_devices = fs_devices; 1032 fs_devices->num_devices++; 1033 } 1034 return fs_devices; 1035 error: 1036 free_fs_devices(fs_devices); 1037 return ERR_PTR(ret); 1038 } 1039 1040 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1041 struct btrfs_device **latest_dev) 1042 { 1043 struct btrfs_device *device, *next; 1044 1045 /* This is the initialized path, it is safe to release the devices. */ 1046 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1047 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1048 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1049 &device->dev_state) && 1050 !test_bit(BTRFS_DEV_STATE_MISSING, 1051 &device->dev_state) && 1052 (!*latest_dev || 1053 device->generation > (*latest_dev)->generation)) { 1054 *latest_dev = device; 1055 } 1056 continue; 1057 } 1058 1059 /* 1060 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1061 * in btrfs_init_dev_replace() so just continue. 1062 */ 1063 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1064 continue; 1065 1066 if (device->bdev) { 1067 blkdev_put(device->bdev, device->mode); 1068 device->bdev = NULL; 1069 fs_devices->open_devices--; 1070 } 1071 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1072 list_del_init(&device->dev_alloc_list); 1073 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1074 fs_devices->rw_devices--; 1075 } 1076 list_del_init(&device->dev_list); 1077 fs_devices->num_devices--; 1078 btrfs_free_device(device); 1079 } 1080 1081 } 1082 1083 /* 1084 * After we have read the system tree and know devids belonging to this 1085 * filesystem, remove the device which does not belong there. 1086 */ 1087 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1088 { 1089 struct btrfs_device *latest_dev = NULL; 1090 struct btrfs_fs_devices *seed_dev; 1091 1092 mutex_lock(&uuid_mutex); 1093 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1094 1095 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1096 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1097 1098 fs_devices->latest_dev = latest_dev; 1099 1100 mutex_unlock(&uuid_mutex); 1101 } 1102 1103 static void btrfs_close_bdev(struct btrfs_device *device) 1104 { 1105 if (!device->bdev) 1106 return; 1107 1108 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1109 sync_blockdev(device->bdev); 1110 invalidate_bdev(device->bdev); 1111 } 1112 1113 blkdev_put(device->bdev, device->mode); 1114 } 1115 1116 static void btrfs_close_one_device(struct btrfs_device *device) 1117 { 1118 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1119 1120 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1121 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1122 list_del_init(&device->dev_alloc_list); 1123 fs_devices->rw_devices--; 1124 } 1125 1126 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1127 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1128 1129 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 1130 fs_devices->missing_devices--; 1131 1132 btrfs_close_bdev(device); 1133 if (device->bdev) { 1134 fs_devices->open_devices--; 1135 device->bdev = NULL; 1136 } 1137 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1138 btrfs_destroy_dev_zone_info(device); 1139 1140 device->fs_info = NULL; 1141 atomic_set(&device->dev_stats_ccnt, 0); 1142 extent_io_tree_release(&device->alloc_state); 1143 1144 /* 1145 * Reset the flush error record. We might have a transient flush error 1146 * in this mount, and if so we aborted the current transaction and set 1147 * the fs to an error state, guaranteeing no super blocks can be further 1148 * committed. However that error might be transient and if we unmount the 1149 * filesystem and mount it again, we should allow the mount to succeed 1150 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1151 * filesystem again we still get flush errors, then we will again abort 1152 * any transaction and set the error state, guaranteeing no commits of 1153 * unsafe super blocks. 1154 */ 1155 device->last_flush_error = 0; 1156 1157 /* Verify the device is back in a pristine state */ 1158 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1159 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1160 ASSERT(list_empty(&device->dev_alloc_list)); 1161 ASSERT(list_empty(&device->post_commit_list)); 1162 ASSERT(atomic_read(&device->reada_in_flight) == 0); 1163 } 1164 1165 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1166 { 1167 struct btrfs_device *device, *tmp; 1168 1169 lockdep_assert_held(&uuid_mutex); 1170 1171 if (--fs_devices->opened > 0) 1172 return; 1173 1174 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1175 btrfs_close_one_device(device); 1176 1177 WARN_ON(fs_devices->open_devices); 1178 WARN_ON(fs_devices->rw_devices); 1179 fs_devices->opened = 0; 1180 fs_devices->seeding = false; 1181 fs_devices->fs_info = NULL; 1182 } 1183 1184 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1185 { 1186 LIST_HEAD(list); 1187 struct btrfs_fs_devices *tmp; 1188 1189 mutex_lock(&uuid_mutex); 1190 close_fs_devices(fs_devices); 1191 if (!fs_devices->opened) 1192 list_splice_init(&fs_devices->seed_list, &list); 1193 1194 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1195 close_fs_devices(fs_devices); 1196 list_del(&fs_devices->seed_list); 1197 free_fs_devices(fs_devices); 1198 } 1199 mutex_unlock(&uuid_mutex); 1200 } 1201 1202 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1203 fmode_t flags, void *holder) 1204 { 1205 struct btrfs_device *device; 1206 struct btrfs_device *latest_dev = NULL; 1207 struct btrfs_device *tmp_device; 1208 1209 flags |= FMODE_EXCL; 1210 1211 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1212 dev_list) { 1213 int ret; 1214 1215 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1216 if (ret == 0 && 1217 (!latest_dev || device->generation > latest_dev->generation)) { 1218 latest_dev = device; 1219 } else if (ret == -ENODATA) { 1220 fs_devices->num_devices--; 1221 list_del(&device->dev_list); 1222 btrfs_free_device(device); 1223 } 1224 } 1225 if (fs_devices->open_devices == 0) 1226 return -EINVAL; 1227 1228 fs_devices->opened = 1; 1229 fs_devices->latest_dev = latest_dev; 1230 fs_devices->total_rw_bytes = 0; 1231 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1232 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1233 1234 return 0; 1235 } 1236 1237 static int devid_cmp(void *priv, const struct list_head *a, 1238 const struct list_head *b) 1239 { 1240 const struct btrfs_device *dev1, *dev2; 1241 1242 dev1 = list_entry(a, struct btrfs_device, dev_list); 1243 dev2 = list_entry(b, struct btrfs_device, dev_list); 1244 1245 if (dev1->devid < dev2->devid) 1246 return -1; 1247 else if (dev1->devid > dev2->devid) 1248 return 1; 1249 return 0; 1250 } 1251 1252 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1253 fmode_t flags, void *holder) 1254 { 1255 int ret; 1256 1257 lockdep_assert_held(&uuid_mutex); 1258 /* 1259 * The device_list_mutex cannot be taken here in case opening the 1260 * underlying device takes further locks like open_mutex. 1261 * 1262 * We also don't need the lock here as this is called during mount and 1263 * exclusion is provided by uuid_mutex 1264 */ 1265 1266 if (fs_devices->opened) { 1267 fs_devices->opened++; 1268 ret = 0; 1269 } else { 1270 list_sort(NULL, &fs_devices->devices, devid_cmp); 1271 ret = open_fs_devices(fs_devices, flags, holder); 1272 } 1273 1274 return ret; 1275 } 1276 1277 void btrfs_release_disk_super(struct btrfs_super_block *super) 1278 { 1279 struct page *page = virt_to_page(super); 1280 1281 put_page(page); 1282 } 1283 1284 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1285 u64 bytenr, u64 bytenr_orig) 1286 { 1287 struct btrfs_super_block *disk_super; 1288 struct page *page; 1289 void *p; 1290 pgoff_t index; 1291 1292 /* make sure our super fits in the device */ 1293 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) 1294 return ERR_PTR(-EINVAL); 1295 1296 /* make sure our super fits in the page */ 1297 if (sizeof(*disk_super) > PAGE_SIZE) 1298 return ERR_PTR(-EINVAL); 1299 1300 /* make sure our super doesn't straddle pages on disk */ 1301 index = bytenr >> PAGE_SHIFT; 1302 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1303 return ERR_PTR(-EINVAL); 1304 1305 /* pull in the page with our super */ 1306 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1307 1308 if (IS_ERR(page)) 1309 return ERR_CAST(page); 1310 1311 p = page_address(page); 1312 1313 /* align our pointer to the offset of the super block */ 1314 disk_super = p + offset_in_page(bytenr); 1315 1316 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1317 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1318 btrfs_release_disk_super(p); 1319 return ERR_PTR(-EINVAL); 1320 } 1321 1322 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1323 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1324 1325 return disk_super; 1326 } 1327 1328 int btrfs_forget_devices(const char *path) 1329 { 1330 int ret; 1331 1332 mutex_lock(&uuid_mutex); 1333 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); 1334 mutex_unlock(&uuid_mutex); 1335 1336 return ret; 1337 } 1338 1339 /* 1340 * Look for a btrfs signature on a device. This may be called out of the mount path 1341 * and we are not allowed to call set_blocksize during the scan. The superblock 1342 * is read via pagecache 1343 */ 1344 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1345 void *holder) 1346 { 1347 struct btrfs_super_block *disk_super; 1348 bool new_device_added = false; 1349 struct btrfs_device *device = NULL; 1350 struct block_device *bdev; 1351 u64 bytenr, bytenr_orig; 1352 int ret; 1353 1354 lockdep_assert_held(&uuid_mutex); 1355 1356 /* 1357 * we would like to check all the supers, but that would make 1358 * a btrfs mount succeed after a mkfs from a different FS. 1359 * So, we need to add a special mount option to scan for 1360 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1361 */ 1362 flags |= FMODE_EXCL; 1363 1364 bdev = blkdev_get_by_path(path, flags, holder); 1365 if (IS_ERR(bdev)) 1366 return ERR_CAST(bdev); 1367 1368 bytenr_orig = btrfs_sb_offset(0); 1369 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1370 if (ret) 1371 return ERR_PTR(ret); 1372 1373 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1374 if (IS_ERR(disk_super)) { 1375 device = ERR_CAST(disk_super); 1376 goto error_bdev_put; 1377 } 1378 1379 device = device_list_add(path, disk_super, &new_device_added); 1380 if (!IS_ERR(device)) { 1381 if (new_device_added) 1382 btrfs_free_stale_devices(path, device); 1383 } 1384 1385 btrfs_release_disk_super(disk_super); 1386 1387 error_bdev_put: 1388 blkdev_put(bdev, flags); 1389 1390 return device; 1391 } 1392 1393 /* 1394 * Try to find a chunk that intersects [start, start + len] range and when one 1395 * such is found, record the end of it in *start 1396 */ 1397 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1398 u64 len) 1399 { 1400 u64 physical_start, physical_end; 1401 1402 lockdep_assert_held(&device->fs_info->chunk_mutex); 1403 1404 if (!find_first_extent_bit(&device->alloc_state, *start, 1405 &physical_start, &physical_end, 1406 CHUNK_ALLOCATED, NULL)) { 1407 1408 if (in_range(physical_start, *start, len) || 1409 in_range(*start, physical_start, 1410 physical_end - physical_start)) { 1411 *start = physical_end + 1; 1412 return true; 1413 } 1414 } 1415 return false; 1416 } 1417 1418 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1419 { 1420 switch (device->fs_devices->chunk_alloc_policy) { 1421 case BTRFS_CHUNK_ALLOC_REGULAR: 1422 /* 1423 * We don't want to overwrite the superblock on the drive nor 1424 * any area used by the boot loader (grub for example), so we 1425 * make sure to start at an offset of at least 1MB. 1426 */ 1427 return max_t(u64, start, SZ_1M); 1428 case BTRFS_CHUNK_ALLOC_ZONED: 1429 /* 1430 * We don't care about the starting region like regular 1431 * allocator, because we anyway use/reserve the first two zones 1432 * for superblock logging. 1433 */ 1434 return ALIGN(start, device->zone_info->zone_size); 1435 default: 1436 BUG(); 1437 } 1438 } 1439 1440 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1441 u64 *hole_start, u64 *hole_size, 1442 u64 num_bytes) 1443 { 1444 u64 zone_size = device->zone_info->zone_size; 1445 u64 pos; 1446 int ret; 1447 bool changed = false; 1448 1449 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1450 1451 while (*hole_size > 0) { 1452 pos = btrfs_find_allocatable_zones(device, *hole_start, 1453 *hole_start + *hole_size, 1454 num_bytes); 1455 if (pos != *hole_start) { 1456 *hole_size = *hole_start + *hole_size - pos; 1457 *hole_start = pos; 1458 changed = true; 1459 if (*hole_size < num_bytes) 1460 break; 1461 } 1462 1463 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1464 1465 /* Range is ensured to be empty */ 1466 if (!ret) 1467 return changed; 1468 1469 /* Given hole range was invalid (outside of device) */ 1470 if (ret == -ERANGE) { 1471 *hole_start += *hole_size; 1472 *hole_size = 0; 1473 return true; 1474 } 1475 1476 *hole_start += zone_size; 1477 *hole_size -= zone_size; 1478 changed = true; 1479 } 1480 1481 return changed; 1482 } 1483 1484 /** 1485 * dev_extent_hole_check - check if specified hole is suitable for allocation 1486 * @device: the device which we have the hole 1487 * @hole_start: starting position of the hole 1488 * @hole_size: the size of the hole 1489 * @num_bytes: the size of the free space that we need 1490 * 1491 * This function may modify @hole_start and @hole_size to reflect the suitable 1492 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1493 */ 1494 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1495 u64 *hole_size, u64 num_bytes) 1496 { 1497 bool changed = false; 1498 u64 hole_end = *hole_start + *hole_size; 1499 1500 for (;;) { 1501 /* 1502 * Check before we set max_hole_start, otherwise we could end up 1503 * sending back this offset anyway. 1504 */ 1505 if (contains_pending_extent(device, hole_start, *hole_size)) { 1506 if (hole_end >= *hole_start) 1507 *hole_size = hole_end - *hole_start; 1508 else 1509 *hole_size = 0; 1510 changed = true; 1511 } 1512 1513 switch (device->fs_devices->chunk_alloc_policy) { 1514 case BTRFS_CHUNK_ALLOC_REGULAR: 1515 /* No extra check */ 1516 break; 1517 case BTRFS_CHUNK_ALLOC_ZONED: 1518 if (dev_extent_hole_check_zoned(device, hole_start, 1519 hole_size, num_bytes)) { 1520 changed = true; 1521 /* 1522 * The changed hole can contain pending extent. 1523 * Loop again to check that. 1524 */ 1525 continue; 1526 } 1527 break; 1528 default: 1529 BUG(); 1530 } 1531 1532 break; 1533 } 1534 1535 return changed; 1536 } 1537 1538 /* 1539 * find_free_dev_extent_start - find free space in the specified device 1540 * @device: the device which we search the free space in 1541 * @num_bytes: the size of the free space that we need 1542 * @search_start: the position from which to begin the search 1543 * @start: store the start of the free space. 1544 * @len: the size of the free space. that we find, or the size 1545 * of the max free space if we don't find suitable free space 1546 * 1547 * this uses a pretty simple search, the expectation is that it is 1548 * called very infrequently and that a given device has a small number 1549 * of extents 1550 * 1551 * @start is used to store the start of the free space if we find. But if we 1552 * don't find suitable free space, it will be used to store the start position 1553 * of the max free space. 1554 * 1555 * @len is used to store the size of the free space that we find. 1556 * But if we don't find suitable free space, it is used to store the size of 1557 * the max free space. 1558 * 1559 * NOTE: This function will search *commit* root of device tree, and does extra 1560 * check to ensure dev extents are not double allocated. 1561 * This makes the function safe to allocate dev extents but may not report 1562 * correct usable device space, as device extent freed in current transaction 1563 * is not reported as available. 1564 */ 1565 static int find_free_dev_extent_start(struct btrfs_device *device, 1566 u64 num_bytes, u64 search_start, u64 *start, 1567 u64 *len) 1568 { 1569 struct btrfs_fs_info *fs_info = device->fs_info; 1570 struct btrfs_root *root = fs_info->dev_root; 1571 struct btrfs_key key; 1572 struct btrfs_dev_extent *dev_extent; 1573 struct btrfs_path *path; 1574 u64 hole_size; 1575 u64 max_hole_start; 1576 u64 max_hole_size; 1577 u64 extent_end; 1578 u64 search_end = device->total_bytes; 1579 int ret; 1580 int slot; 1581 struct extent_buffer *l; 1582 1583 search_start = dev_extent_search_start(device, search_start); 1584 1585 WARN_ON(device->zone_info && 1586 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1587 1588 path = btrfs_alloc_path(); 1589 if (!path) 1590 return -ENOMEM; 1591 1592 max_hole_start = search_start; 1593 max_hole_size = 0; 1594 1595 again: 1596 if (search_start >= search_end || 1597 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1598 ret = -ENOSPC; 1599 goto out; 1600 } 1601 1602 path->reada = READA_FORWARD; 1603 path->search_commit_root = 1; 1604 path->skip_locking = 1; 1605 1606 key.objectid = device->devid; 1607 key.offset = search_start; 1608 key.type = BTRFS_DEV_EXTENT_KEY; 1609 1610 ret = btrfs_search_backwards(root, &key, path); 1611 if (ret < 0) 1612 goto out; 1613 1614 while (1) { 1615 l = path->nodes[0]; 1616 slot = path->slots[0]; 1617 if (slot >= btrfs_header_nritems(l)) { 1618 ret = btrfs_next_leaf(root, path); 1619 if (ret == 0) 1620 continue; 1621 if (ret < 0) 1622 goto out; 1623 1624 break; 1625 } 1626 btrfs_item_key_to_cpu(l, &key, slot); 1627 1628 if (key.objectid < device->devid) 1629 goto next; 1630 1631 if (key.objectid > device->devid) 1632 break; 1633 1634 if (key.type != BTRFS_DEV_EXTENT_KEY) 1635 goto next; 1636 1637 if (key.offset > search_start) { 1638 hole_size = key.offset - search_start; 1639 dev_extent_hole_check(device, &search_start, &hole_size, 1640 num_bytes); 1641 1642 if (hole_size > max_hole_size) { 1643 max_hole_start = search_start; 1644 max_hole_size = hole_size; 1645 } 1646 1647 /* 1648 * If this free space is greater than which we need, 1649 * it must be the max free space that we have found 1650 * until now, so max_hole_start must point to the start 1651 * of this free space and the length of this free space 1652 * is stored in max_hole_size. Thus, we return 1653 * max_hole_start and max_hole_size and go back to the 1654 * caller. 1655 */ 1656 if (hole_size >= num_bytes) { 1657 ret = 0; 1658 goto out; 1659 } 1660 } 1661 1662 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1663 extent_end = key.offset + btrfs_dev_extent_length(l, 1664 dev_extent); 1665 if (extent_end > search_start) 1666 search_start = extent_end; 1667 next: 1668 path->slots[0]++; 1669 cond_resched(); 1670 } 1671 1672 /* 1673 * At this point, search_start should be the end of 1674 * allocated dev extents, and when shrinking the device, 1675 * search_end may be smaller than search_start. 1676 */ 1677 if (search_end > search_start) { 1678 hole_size = search_end - search_start; 1679 if (dev_extent_hole_check(device, &search_start, &hole_size, 1680 num_bytes)) { 1681 btrfs_release_path(path); 1682 goto again; 1683 } 1684 1685 if (hole_size > max_hole_size) { 1686 max_hole_start = search_start; 1687 max_hole_size = hole_size; 1688 } 1689 } 1690 1691 /* See above. */ 1692 if (max_hole_size < num_bytes) 1693 ret = -ENOSPC; 1694 else 1695 ret = 0; 1696 1697 out: 1698 btrfs_free_path(path); 1699 *start = max_hole_start; 1700 if (len) 1701 *len = max_hole_size; 1702 return ret; 1703 } 1704 1705 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1706 u64 *start, u64 *len) 1707 { 1708 /* FIXME use last free of some kind */ 1709 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1710 } 1711 1712 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1713 struct btrfs_device *device, 1714 u64 start, u64 *dev_extent_len) 1715 { 1716 struct btrfs_fs_info *fs_info = device->fs_info; 1717 struct btrfs_root *root = fs_info->dev_root; 1718 int ret; 1719 struct btrfs_path *path; 1720 struct btrfs_key key; 1721 struct btrfs_key found_key; 1722 struct extent_buffer *leaf = NULL; 1723 struct btrfs_dev_extent *extent = NULL; 1724 1725 path = btrfs_alloc_path(); 1726 if (!path) 1727 return -ENOMEM; 1728 1729 key.objectid = device->devid; 1730 key.offset = start; 1731 key.type = BTRFS_DEV_EXTENT_KEY; 1732 again: 1733 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1734 if (ret > 0) { 1735 ret = btrfs_previous_item(root, path, key.objectid, 1736 BTRFS_DEV_EXTENT_KEY); 1737 if (ret) 1738 goto out; 1739 leaf = path->nodes[0]; 1740 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1741 extent = btrfs_item_ptr(leaf, path->slots[0], 1742 struct btrfs_dev_extent); 1743 BUG_ON(found_key.offset > start || found_key.offset + 1744 btrfs_dev_extent_length(leaf, extent) < start); 1745 key = found_key; 1746 btrfs_release_path(path); 1747 goto again; 1748 } else if (ret == 0) { 1749 leaf = path->nodes[0]; 1750 extent = btrfs_item_ptr(leaf, path->slots[0], 1751 struct btrfs_dev_extent); 1752 } else { 1753 goto out; 1754 } 1755 1756 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1757 1758 ret = btrfs_del_item(trans, root, path); 1759 if (ret == 0) 1760 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1761 out: 1762 btrfs_free_path(path); 1763 return ret; 1764 } 1765 1766 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1767 { 1768 struct extent_map_tree *em_tree; 1769 struct extent_map *em; 1770 struct rb_node *n; 1771 u64 ret = 0; 1772 1773 em_tree = &fs_info->mapping_tree; 1774 read_lock(&em_tree->lock); 1775 n = rb_last(&em_tree->map.rb_root); 1776 if (n) { 1777 em = rb_entry(n, struct extent_map, rb_node); 1778 ret = em->start + em->len; 1779 } 1780 read_unlock(&em_tree->lock); 1781 1782 return ret; 1783 } 1784 1785 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1786 u64 *devid_ret) 1787 { 1788 int ret; 1789 struct btrfs_key key; 1790 struct btrfs_key found_key; 1791 struct btrfs_path *path; 1792 1793 path = btrfs_alloc_path(); 1794 if (!path) 1795 return -ENOMEM; 1796 1797 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1798 key.type = BTRFS_DEV_ITEM_KEY; 1799 key.offset = (u64)-1; 1800 1801 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1802 if (ret < 0) 1803 goto error; 1804 1805 if (ret == 0) { 1806 /* Corruption */ 1807 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1808 ret = -EUCLEAN; 1809 goto error; 1810 } 1811 1812 ret = btrfs_previous_item(fs_info->chunk_root, path, 1813 BTRFS_DEV_ITEMS_OBJECTID, 1814 BTRFS_DEV_ITEM_KEY); 1815 if (ret) { 1816 *devid_ret = 1; 1817 } else { 1818 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1819 path->slots[0]); 1820 *devid_ret = found_key.offset + 1; 1821 } 1822 ret = 0; 1823 error: 1824 btrfs_free_path(path); 1825 return ret; 1826 } 1827 1828 /* 1829 * the device information is stored in the chunk root 1830 * the btrfs_device struct should be fully filled in 1831 */ 1832 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1833 struct btrfs_device *device) 1834 { 1835 int ret; 1836 struct btrfs_path *path; 1837 struct btrfs_dev_item *dev_item; 1838 struct extent_buffer *leaf; 1839 struct btrfs_key key; 1840 unsigned long ptr; 1841 1842 path = btrfs_alloc_path(); 1843 if (!path) 1844 return -ENOMEM; 1845 1846 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1847 key.type = BTRFS_DEV_ITEM_KEY; 1848 key.offset = device->devid; 1849 1850 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1851 &key, sizeof(*dev_item)); 1852 if (ret) 1853 goto out; 1854 1855 leaf = path->nodes[0]; 1856 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1857 1858 btrfs_set_device_id(leaf, dev_item, device->devid); 1859 btrfs_set_device_generation(leaf, dev_item, 0); 1860 btrfs_set_device_type(leaf, dev_item, device->type); 1861 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1862 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1863 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1864 btrfs_set_device_total_bytes(leaf, dev_item, 1865 btrfs_device_get_disk_total_bytes(device)); 1866 btrfs_set_device_bytes_used(leaf, dev_item, 1867 btrfs_device_get_bytes_used(device)); 1868 btrfs_set_device_group(leaf, dev_item, 0); 1869 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1870 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1871 btrfs_set_device_start_offset(leaf, dev_item, 0); 1872 1873 ptr = btrfs_device_uuid(dev_item); 1874 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1875 ptr = btrfs_device_fsid(dev_item); 1876 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1877 ptr, BTRFS_FSID_SIZE); 1878 btrfs_mark_buffer_dirty(leaf); 1879 1880 ret = 0; 1881 out: 1882 btrfs_free_path(path); 1883 return ret; 1884 } 1885 1886 /* 1887 * Function to update ctime/mtime for a given device path. 1888 * Mainly used for ctime/mtime based probe like libblkid. 1889 */ 1890 static void update_dev_time(struct block_device *bdev) 1891 { 1892 struct inode *inode = bdev->bd_inode; 1893 struct timespec64 now; 1894 1895 /* Shouldn't happen but just in case. */ 1896 if (!inode) 1897 return; 1898 1899 now = current_time(inode); 1900 generic_update_time(inode, &now, S_MTIME | S_CTIME); 1901 } 1902 1903 static int btrfs_rm_dev_item(struct btrfs_device *device) 1904 { 1905 struct btrfs_root *root = device->fs_info->chunk_root; 1906 int ret; 1907 struct btrfs_path *path; 1908 struct btrfs_key key; 1909 struct btrfs_trans_handle *trans; 1910 1911 path = btrfs_alloc_path(); 1912 if (!path) 1913 return -ENOMEM; 1914 1915 trans = btrfs_start_transaction(root, 0); 1916 if (IS_ERR(trans)) { 1917 btrfs_free_path(path); 1918 return PTR_ERR(trans); 1919 } 1920 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1921 key.type = BTRFS_DEV_ITEM_KEY; 1922 key.offset = device->devid; 1923 1924 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1925 if (ret) { 1926 if (ret > 0) 1927 ret = -ENOENT; 1928 btrfs_abort_transaction(trans, ret); 1929 btrfs_end_transaction(trans); 1930 goto out; 1931 } 1932 1933 ret = btrfs_del_item(trans, root, path); 1934 if (ret) { 1935 btrfs_abort_transaction(trans, ret); 1936 btrfs_end_transaction(trans); 1937 } 1938 1939 out: 1940 btrfs_free_path(path); 1941 if (!ret) 1942 ret = btrfs_commit_transaction(trans); 1943 return ret; 1944 } 1945 1946 /* 1947 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1948 * filesystem. It's up to the caller to adjust that number regarding eg. device 1949 * replace. 1950 */ 1951 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1952 u64 num_devices) 1953 { 1954 u64 all_avail; 1955 unsigned seq; 1956 int i; 1957 1958 do { 1959 seq = read_seqbegin(&fs_info->profiles_lock); 1960 1961 all_avail = fs_info->avail_data_alloc_bits | 1962 fs_info->avail_system_alloc_bits | 1963 fs_info->avail_metadata_alloc_bits; 1964 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1965 1966 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1967 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1968 continue; 1969 1970 if (num_devices < btrfs_raid_array[i].devs_min) 1971 return btrfs_raid_array[i].mindev_error; 1972 } 1973 1974 return 0; 1975 } 1976 1977 static struct btrfs_device * btrfs_find_next_active_device( 1978 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1979 { 1980 struct btrfs_device *next_device; 1981 1982 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1983 if (next_device != device && 1984 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1985 && next_device->bdev) 1986 return next_device; 1987 } 1988 1989 return NULL; 1990 } 1991 1992 /* 1993 * Helper function to check if the given device is part of s_bdev / latest_dev 1994 * and replace it with the provided or the next active device, in the context 1995 * where this function called, there should be always be another device (or 1996 * this_dev) which is active. 1997 */ 1998 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 1999 struct btrfs_device *next_device) 2000 { 2001 struct btrfs_fs_info *fs_info = device->fs_info; 2002 2003 if (!next_device) 2004 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 2005 device); 2006 ASSERT(next_device); 2007 2008 if (fs_info->sb->s_bdev && 2009 (fs_info->sb->s_bdev == device->bdev)) 2010 fs_info->sb->s_bdev = next_device->bdev; 2011 2012 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2013 fs_info->fs_devices->latest_dev = next_device; 2014 } 2015 2016 /* 2017 * Return btrfs_fs_devices::num_devices excluding the device that's being 2018 * currently replaced. 2019 */ 2020 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2021 { 2022 u64 num_devices = fs_info->fs_devices->num_devices; 2023 2024 down_read(&fs_info->dev_replace.rwsem); 2025 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2026 ASSERT(num_devices > 1); 2027 num_devices--; 2028 } 2029 up_read(&fs_info->dev_replace.rwsem); 2030 2031 return num_devices; 2032 } 2033 2034 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2035 struct block_device *bdev, 2036 const char *device_path) 2037 { 2038 struct btrfs_super_block *disk_super; 2039 int copy_num; 2040 2041 if (!bdev) 2042 return; 2043 2044 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2045 struct page *page; 2046 int ret; 2047 2048 disk_super = btrfs_read_dev_one_super(bdev, copy_num); 2049 if (IS_ERR(disk_super)) 2050 continue; 2051 2052 if (bdev_is_zoned(bdev)) { 2053 btrfs_reset_sb_log_zones(bdev, copy_num); 2054 continue; 2055 } 2056 2057 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 2058 2059 page = virt_to_page(disk_super); 2060 set_page_dirty(page); 2061 lock_page(page); 2062 /* write_on_page() unlocks the page */ 2063 ret = write_one_page(page); 2064 if (ret) 2065 btrfs_warn(fs_info, 2066 "error clearing superblock number %d (%d)", 2067 copy_num, ret); 2068 btrfs_release_disk_super(disk_super); 2069 2070 } 2071 2072 /* Notify udev that device has changed */ 2073 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2074 2075 /* Update ctime/mtime for device path for libblkid */ 2076 update_dev_time(bdev); 2077 } 2078 2079 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 2080 struct btrfs_dev_lookup_args *args, 2081 struct block_device **bdev, fmode_t *mode) 2082 { 2083 struct btrfs_device *device; 2084 struct btrfs_fs_devices *cur_devices; 2085 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2086 u64 num_devices; 2087 int ret = 0; 2088 2089 /* 2090 * The device list in fs_devices is accessed without locks (neither 2091 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2092 * filesystem and another device rm cannot run. 2093 */ 2094 num_devices = btrfs_num_devices(fs_info); 2095 2096 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2097 if (ret) 2098 goto out; 2099 2100 device = btrfs_find_device(fs_info->fs_devices, args); 2101 if (!device) { 2102 if (args->missing) 2103 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2104 else 2105 ret = -ENOENT; 2106 goto out; 2107 } 2108 2109 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2110 btrfs_warn_in_rcu(fs_info, 2111 "cannot remove device %s (devid %llu) due to active swapfile", 2112 rcu_str_deref(device->name), device->devid); 2113 ret = -ETXTBSY; 2114 goto out; 2115 } 2116 2117 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2118 ret = BTRFS_ERROR_DEV_TGT_REPLACE; 2119 goto out; 2120 } 2121 2122 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2123 fs_info->fs_devices->rw_devices == 1) { 2124 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; 2125 goto out; 2126 } 2127 2128 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2129 mutex_lock(&fs_info->chunk_mutex); 2130 list_del_init(&device->dev_alloc_list); 2131 device->fs_devices->rw_devices--; 2132 mutex_unlock(&fs_info->chunk_mutex); 2133 } 2134 2135 ret = btrfs_shrink_device(device, 0); 2136 if (!ret) 2137 btrfs_reada_remove_dev(device); 2138 if (ret) 2139 goto error_undo; 2140 2141 /* 2142 * TODO: the superblock still includes this device in its num_devices 2143 * counter although write_all_supers() is not locked out. This 2144 * could give a filesystem state which requires a degraded mount. 2145 */ 2146 ret = btrfs_rm_dev_item(device); 2147 if (ret) 2148 goto error_undo; 2149 2150 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2151 btrfs_scrub_cancel_dev(device); 2152 2153 /* 2154 * the device list mutex makes sure that we don't change 2155 * the device list while someone else is writing out all 2156 * the device supers. Whoever is writing all supers, should 2157 * lock the device list mutex before getting the number of 2158 * devices in the super block (super_copy). Conversely, 2159 * whoever updates the number of devices in the super block 2160 * (super_copy) should hold the device list mutex. 2161 */ 2162 2163 /* 2164 * In normal cases the cur_devices == fs_devices. But in case 2165 * of deleting a seed device, the cur_devices should point to 2166 * its own fs_devices listed under the fs_devices->seed_list. 2167 */ 2168 cur_devices = device->fs_devices; 2169 mutex_lock(&fs_devices->device_list_mutex); 2170 list_del_rcu(&device->dev_list); 2171 2172 cur_devices->num_devices--; 2173 cur_devices->total_devices--; 2174 /* Update total_devices of the parent fs_devices if it's seed */ 2175 if (cur_devices != fs_devices) 2176 fs_devices->total_devices--; 2177 2178 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2179 cur_devices->missing_devices--; 2180 2181 btrfs_assign_next_active_device(device, NULL); 2182 2183 if (device->bdev) { 2184 cur_devices->open_devices--; 2185 /* remove sysfs entry */ 2186 btrfs_sysfs_remove_device(device); 2187 } 2188 2189 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2190 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2191 mutex_unlock(&fs_devices->device_list_mutex); 2192 2193 /* 2194 * At this point, the device is zero sized and detached from the 2195 * devices list. All that's left is to zero out the old supers and 2196 * free the device. 2197 * 2198 * We cannot call btrfs_close_bdev() here because we're holding the sb 2199 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2200 * block device and it's dependencies. Instead just flush the device 2201 * and let the caller do the final blkdev_put. 2202 */ 2203 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2204 btrfs_scratch_superblocks(fs_info, device->bdev, 2205 device->name->str); 2206 if (device->bdev) { 2207 sync_blockdev(device->bdev); 2208 invalidate_bdev(device->bdev); 2209 } 2210 } 2211 2212 *bdev = device->bdev; 2213 *mode = device->mode; 2214 synchronize_rcu(); 2215 btrfs_free_device(device); 2216 2217 /* 2218 * This can happen if cur_devices is the private seed devices list. We 2219 * cannot call close_fs_devices() here because it expects the uuid_mutex 2220 * to be held, but in fact we don't need that for the private 2221 * seed_devices, we can simply decrement cur_devices->opened and then 2222 * remove it from our list and free the fs_devices. 2223 */ 2224 if (cur_devices->num_devices == 0) { 2225 list_del_init(&cur_devices->seed_list); 2226 ASSERT(cur_devices->opened == 1); 2227 cur_devices->opened--; 2228 free_fs_devices(cur_devices); 2229 } 2230 2231 out: 2232 return ret; 2233 2234 error_undo: 2235 btrfs_reada_undo_remove_dev(device); 2236 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2237 mutex_lock(&fs_info->chunk_mutex); 2238 list_add(&device->dev_alloc_list, 2239 &fs_devices->alloc_list); 2240 device->fs_devices->rw_devices++; 2241 mutex_unlock(&fs_info->chunk_mutex); 2242 } 2243 goto out; 2244 } 2245 2246 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2247 { 2248 struct btrfs_fs_devices *fs_devices; 2249 2250 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2251 2252 /* 2253 * in case of fs with no seed, srcdev->fs_devices will point 2254 * to fs_devices of fs_info. However when the dev being replaced is 2255 * a seed dev it will point to the seed's local fs_devices. In short 2256 * srcdev will have its correct fs_devices in both the cases. 2257 */ 2258 fs_devices = srcdev->fs_devices; 2259 2260 list_del_rcu(&srcdev->dev_list); 2261 list_del(&srcdev->dev_alloc_list); 2262 fs_devices->num_devices--; 2263 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2264 fs_devices->missing_devices--; 2265 2266 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2267 fs_devices->rw_devices--; 2268 2269 if (srcdev->bdev) 2270 fs_devices->open_devices--; 2271 } 2272 2273 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2274 { 2275 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2276 2277 mutex_lock(&uuid_mutex); 2278 2279 btrfs_close_bdev(srcdev); 2280 synchronize_rcu(); 2281 btrfs_free_device(srcdev); 2282 2283 /* if this is no devs we rather delete the fs_devices */ 2284 if (!fs_devices->num_devices) { 2285 /* 2286 * On a mounted FS, num_devices can't be zero unless it's a 2287 * seed. In case of a seed device being replaced, the replace 2288 * target added to the sprout FS, so there will be no more 2289 * device left under the seed FS. 2290 */ 2291 ASSERT(fs_devices->seeding); 2292 2293 list_del_init(&fs_devices->seed_list); 2294 close_fs_devices(fs_devices); 2295 free_fs_devices(fs_devices); 2296 } 2297 mutex_unlock(&uuid_mutex); 2298 } 2299 2300 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2301 { 2302 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2303 2304 mutex_lock(&fs_devices->device_list_mutex); 2305 2306 btrfs_sysfs_remove_device(tgtdev); 2307 2308 if (tgtdev->bdev) 2309 fs_devices->open_devices--; 2310 2311 fs_devices->num_devices--; 2312 2313 btrfs_assign_next_active_device(tgtdev, NULL); 2314 2315 list_del_rcu(&tgtdev->dev_list); 2316 2317 mutex_unlock(&fs_devices->device_list_mutex); 2318 2319 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2320 tgtdev->name->str); 2321 2322 btrfs_close_bdev(tgtdev); 2323 synchronize_rcu(); 2324 btrfs_free_device(tgtdev); 2325 } 2326 2327 /** 2328 * Populate args from device at path 2329 * 2330 * @fs_info: the filesystem 2331 * @args: the args to populate 2332 * @path: the path to the device 2333 * 2334 * This will read the super block of the device at @path and populate @args with 2335 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2336 * lookup a device to operate on, but need to do it before we take any locks. 2337 * This properly handles the special case of "missing" that a user may pass in, 2338 * and does some basic sanity checks. The caller must make sure that @path is 2339 * properly NUL terminated before calling in, and must call 2340 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2341 * uuid buffers. 2342 * 2343 * Return: 0 for success, -errno for failure 2344 */ 2345 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2346 struct btrfs_dev_lookup_args *args, 2347 const char *path) 2348 { 2349 struct btrfs_super_block *disk_super; 2350 struct block_device *bdev; 2351 int ret; 2352 2353 if (!path || !path[0]) 2354 return -EINVAL; 2355 if (!strcmp(path, "missing")) { 2356 args->missing = true; 2357 return 0; 2358 } 2359 2360 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2361 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2362 if (!args->uuid || !args->fsid) { 2363 btrfs_put_dev_args_from_path(args); 2364 return -ENOMEM; 2365 } 2366 2367 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2368 &bdev, &disk_super); 2369 if (ret) 2370 return ret; 2371 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2372 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2373 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2374 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2375 else 2376 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2377 btrfs_release_disk_super(disk_super); 2378 blkdev_put(bdev, FMODE_READ); 2379 return 0; 2380 } 2381 2382 /* 2383 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2384 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2385 * that don't need to be freed. 2386 */ 2387 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2388 { 2389 kfree(args->uuid); 2390 kfree(args->fsid); 2391 args->uuid = NULL; 2392 args->fsid = NULL; 2393 } 2394 2395 struct btrfs_device *btrfs_find_device_by_devspec( 2396 struct btrfs_fs_info *fs_info, u64 devid, 2397 const char *device_path) 2398 { 2399 BTRFS_DEV_LOOKUP_ARGS(args); 2400 struct btrfs_device *device; 2401 int ret; 2402 2403 if (devid) { 2404 args.devid = devid; 2405 device = btrfs_find_device(fs_info->fs_devices, &args); 2406 if (!device) 2407 return ERR_PTR(-ENOENT); 2408 return device; 2409 } 2410 2411 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2412 if (ret) 2413 return ERR_PTR(ret); 2414 device = btrfs_find_device(fs_info->fs_devices, &args); 2415 btrfs_put_dev_args_from_path(&args); 2416 if (!device) 2417 return ERR_PTR(-ENOENT); 2418 return device; 2419 } 2420 2421 /* 2422 * does all the dirty work required for changing file system's UUID. 2423 */ 2424 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) 2425 { 2426 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2427 struct btrfs_fs_devices *old_devices; 2428 struct btrfs_fs_devices *seed_devices; 2429 struct btrfs_super_block *disk_super = fs_info->super_copy; 2430 struct btrfs_device *device; 2431 u64 super_flags; 2432 2433 lockdep_assert_held(&uuid_mutex); 2434 if (!fs_devices->seeding) 2435 return -EINVAL; 2436 2437 /* 2438 * Private copy of the seed devices, anchored at 2439 * fs_info->fs_devices->seed_list 2440 */ 2441 seed_devices = alloc_fs_devices(NULL, NULL); 2442 if (IS_ERR(seed_devices)) 2443 return PTR_ERR(seed_devices); 2444 2445 /* 2446 * It's necessary to retain a copy of the original seed fs_devices in 2447 * fs_uuids so that filesystems which have been seeded can successfully 2448 * reference the seed device from open_seed_devices. This also supports 2449 * multiple fs seed. 2450 */ 2451 old_devices = clone_fs_devices(fs_devices); 2452 if (IS_ERR(old_devices)) { 2453 kfree(seed_devices); 2454 return PTR_ERR(old_devices); 2455 } 2456 2457 list_add(&old_devices->fs_list, &fs_uuids); 2458 2459 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2460 seed_devices->opened = 1; 2461 INIT_LIST_HEAD(&seed_devices->devices); 2462 INIT_LIST_HEAD(&seed_devices->alloc_list); 2463 mutex_init(&seed_devices->device_list_mutex); 2464 2465 mutex_lock(&fs_devices->device_list_mutex); 2466 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2467 synchronize_rcu); 2468 list_for_each_entry(device, &seed_devices->devices, dev_list) 2469 device->fs_devices = seed_devices; 2470 2471 fs_devices->seeding = false; 2472 fs_devices->num_devices = 0; 2473 fs_devices->open_devices = 0; 2474 fs_devices->missing_devices = 0; 2475 fs_devices->rotating = false; 2476 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2477 2478 generate_random_uuid(fs_devices->fsid); 2479 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2480 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2481 mutex_unlock(&fs_devices->device_list_mutex); 2482 2483 super_flags = btrfs_super_flags(disk_super) & 2484 ~BTRFS_SUPER_FLAG_SEEDING; 2485 btrfs_set_super_flags(disk_super, super_flags); 2486 2487 return 0; 2488 } 2489 2490 /* 2491 * Store the expected generation for seed devices in device items. 2492 */ 2493 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2494 { 2495 BTRFS_DEV_LOOKUP_ARGS(args); 2496 struct btrfs_fs_info *fs_info = trans->fs_info; 2497 struct btrfs_root *root = fs_info->chunk_root; 2498 struct btrfs_path *path; 2499 struct extent_buffer *leaf; 2500 struct btrfs_dev_item *dev_item; 2501 struct btrfs_device *device; 2502 struct btrfs_key key; 2503 u8 fs_uuid[BTRFS_FSID_SIZE]; 2504 u8 dev_uuid[BTRFS_UUID_SIZE]; 2505 int ret; 2506 2507 path = btrfs_alloc_path(); 2508 if (!path) 2509 return -ENOMEM; 2510 2511 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2512 key.offset = 0; 2513 key.type = BTRFS_DEV_ITEM_KEY; 2514 2515 while (1) { 2516 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2517 if (ret < 0) 2518 goto error; 2519 2520 leaf = path->nodes[0]; 2521 next_slot: 2522 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2523 ret = btrfs_next_leaf(root, path); 2524 if (ret > 0) 2525 break; 2526 if (ret < 0) 2527 goto error; 2528 leaf = path->nodes[0]; 2529 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2530 btrfs_release_path(path); 2531 continue; 2532 } 2533 2534 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2535 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2536 key.type != BTRFS_DEV_ITEM_KEY) 2537 break; 2538 2539 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2540 struct btrfs_dev_item); 2541 args.devid = btrfs_device_id(leaf, dev_item); 2542 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2543 BTRFS_UUID_SIZE); 2544 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2545 BTRFS_FSID_SIZE); 2546 args.uuid = dev_uuid; 2547 args.fsid = fs_uuid; 2548 device = btrfs_find_device(fs_info->fs_devices, &args); 2549 BUG_ON(!device); /* Logic error */ 2550 2551 if (device->fs_devices->seeding) { 2552 btrfs_set_device_generation(leaf, dev_item, 2553 device->generation); 2554 btrfs_mark_buffer_dirty(leaf); 2555 } 2556 2557 path->slots[0]++; 2558 goto next_slot; 2559 } 2560 ret = 0; 2561 error: 2562 btrfs_free_path(path); 2563 return ret; 2564 } 2565 2566 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2567 { 2568 struct btrfs_root *root = fs_info->dev_root; 2569 struct request_queue *q; 2570 struct btrfs_trans_handle *trans; 2571 struct btrfs_device *device; 2572 struct block_device *bdev; 2573 struct super_block *sb = fs_info->sb; 2574 struct rcu_string *name; 2575 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2576 u64 orig_super_total_bytes; 2577 u64 orig_super_num_devices; 2578 int seeding_dev = 0; 2579 int ret = 0; 2580 bool locked = false; 2581 2582 if (sb_rdonly(sb) && !fs_devices->seeding) 2583 return -EROFS; 2584 2585 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2586 fs_info->bdev_holder); 2587 if (IS_ERR(bdev)) 2588 return PTR_ERR(bdev); 2589 2590 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2591 ret = -EINVAL; 2592 goto error; 2593 } 2594 2595 if (fs_devices->seeding) { 2596 seeding_dev = 1; 2597 down_write(&sb->s_umount); 2598 mutex_lock(&uuid_mutex); 2599 locked = true; 2600 } 2601 2602 sync_blockdev(bdev); 2603 2604 rcu_read_lock(); 2605 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2606 if (device->bdev == bdev) { 2607 ret = -EEXIST; 2608 rcu_read_unlock(); 2609 goto error; 2610 } 2611 } 2612 rcu_read_unlock(); 2613 2614 device = btrfs_alloc_device(fs_info, NULL, NULL); 2615 if (IS_ERR(device)) { 2616 /* we can safely leave the fs_devices entry around */ 2617 ret = PTR_ERR(device); 2618 goto error; 2619 } 2620 2621 name = rcu_string_strdup(device_path, GFP_KERNEL); 2622 if (!name) { 2623 ret = -ENOMEM; 2624 goto error_free_device; 2625 } 2626 rcu_assign_pointer(device->name, name); 2627 2628 device->fs_info = fs_info; 2629 device->bdev = bdev; 2630 2631 ret = btrfs_get_dev_zone_info(device); 2632 if (ret) 2633 goto error_free_device; 2634 2635 trans = btrfs_start_transaction(root, 0); 2636 if (IS_ERR(trans)) { 2637 ret = PTR_ERR(trans); 2638 goto error_free_zone; 2639 } 2640 2641 q = bdev_get_queue(bdev); 2642 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2643 device->generation = trans->transid; 2644 device->io_width = fs_info->sectorsize; 2645 device->io_align = fs_info->sectorsize; 2646 device->sector_size = fs_info->sectorsize; 2647 device->total_bytes = round_down(i_size_read(bdev->bd_inode), 2648 fs_info->sectorsize); 2649 device->disk_total_bytes = device->total_bytes; 2650 device->commit_total_bytes = device->total_bytes; 2651 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2652 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2653 device->mode = FMODE_EXCL; 2654 device->dev_stats_valid = 1; 2655 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2656 2657 if (seeding_dev) { 2658 btrfs_clear_sb_rdonly(sb); 2659 ret = btrfs_prepare_sprout(fs_info); 2660 if (ret) { 2661 btrfs_abort_transaction(trans, ret); 2662 goto error_trans; 2663 } 2664 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2665 device); 2666 } 2667 2668 device->fs_devices = fs_devices; 2669 2670 mutex_lock(&fs_devices->device_list_mutex); 2671 mutex_lock(&fs_info->chunk_mutex); 2672 list_add_rcu(&device->dev_list, &fs_devices->devices); 2673 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2674 fs_devices->num_devices++; 2675 fs_devices->open_devices++; 2676 fs_devices->rw_devices++; 2677 fs_devices->total_devices++; 2678 fs_devices->total_rw_bytes += device->total_bytes; 2679 2680 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2681 2682 if (!blk_queue_nonrot(q)) 2683 fs_devices->rotating = true; 2684 2685 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2686 btrfs_set_super_total_bytes(fs_info->super_copy, 2687 round_down(orig_super_total_bytes + device->total_bytes, 2688 fs_info->sectorsize)); 2689 2690 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2691 btrfs_set_super_num_devices(fs_info->super_copy, 2692 orig_super_num_devices + 1); 2693 2694 /* 2695 * we've got more storage, clear any full flags on the space 2696 * infos 2697 */ 2698 btrfs_clear_space_info_full(fs_info); 2699 2700 mutex_unlock(&fs_info->chunk_mutex); 2701 2702 /* Add sysfs device entry */ 2703 btrfs_sysfs_add_device(device); 2704 2705 mutex_unlock(&fs_devices->device_list_mutex); 2706 2707 if (seeding_dev) { 2708 mutex_lock(&fs_info->chunk_mutex); 2709 ret = init_first_rw_device(trans); 2710 mutex_unlock(&fs_info->chunk_mutex); 2711 if (ret) { 2712 btrfs_abort_transaction(trans, ret); 2713 goto error_sysfs; 2714 } 2715 } 2716 2717 ret = btrfs_add_dev_item(trans, device); 2718 if (ret) { 2719 btrfs_abort_transaction(trans, ret); 2720 goto error_sysfs; 2721 } 2722 2723 if (seeding_dev) { 2724 ret = btrfs_finish_sprout(trans); 2725 if (ret) { 2726 btrfs_abort_transaction(trans, ret); 2727 goto error_sysfs; 2728 } 2729 2730 /* 2731 * fs_devices now represents the newly sprouted filesystem and 2732 * its fsid has been changed by btrfs_prepare_sprout 2733 */ 2734 btrfs_sysfs_update_sprout_fsid(fs_devices); 2735 } 2736 2737 ret = btrfs_commit_transaction(trans); 2738 2739 if (seeding_dev) { 2740 mutex_unlock(&uuid_mutex); 2741 up_write(&sb->s_umount); 2742 locked = false; 2743 2744 if (ret) /* transaction commit */ 2745 return ret; 2746 2747 ret = btrfs_relocate_sys_chunks(fs_info); 2748 if (ret < 0) 2749 btrfs_handle_fs_error(fs_info, ret, 2750 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2751 trans = btrfs_attach_transaction(root); 2752 if (IS_ERR(trans)) { 2753 if (PTR_ERR(trans) == -ENOENT) 2754 return 0; 2755 ret = PTR_ERR(trans); 2756 trans = NULL; 2757 goto error_sysfs; 2758 } 2759 ret = btrfs_commit_transaction(trans); 2760 } 2761 2762 /* 2763 * Now that we have written a new super block to this device, check all 2764 * other fs_devices list if device_path alienates any other scanned 2765 * device. 2766 * We can ignore the return value as it typically returns -EINVAL and 2767 * only succeeds if the device was an alien. 2768 */ 2769 btrfs_forget_devices(device_path); 2770 2771 /* Update ctime/mtime for blkid or udev */ 2772 update_dev_time(bdev); 2773 2774 return ret; 2775 2776 error_sysfs: 2777 btrfs_sysfs_remove_device(device); 2778 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2779 mutex_lock(&fs_info->chunk_mutex); 2780 list_del_rcu(&device->dev_list); 2781 list_del(&device->dev_alloc_list); 2782 fs_info->fs_devices->num_devices--; 2783 fs_info->fs_devices->open_devices--; 2784 fs_info->fs_devices->rw_devices--; 2785 fs_info->fs_devices->total_devices--; 2786 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2787 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2788 btrfs_set_super_total_bytes(fs_info->super_copy, 2789 orig_super_total_bytes); 2790 btrfs_set_super_num_devices(fs_info->super_copy, 2791 orig_super_num_devices); 2792 mutex_unlock(&fs_info->chunk_mutex); 2793 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2794 error_trans: 2795 if (seeding_dev) 2796 btrfs_set_sb_rdonly(sb); 2797 if (trans) 2798 btrfs_end_transaction(trans); 2799 error_free_zone: 2800 btrfs_destroy_dev_zone_info(device); 2801 error_free_device: 2802 btrfs_free_device(device); 2803 error: 2804 blkdev_put(bdev, FMODE_EXCL); 2805 if (locked) { 2806 mutex_unlock(&uuid_mutex); 2807 up_write(&sb->s_umount); 2808 } 2809 return ret; 2810 } 2811 2812 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2813 struct btrfs_device *device) 2814 { 2815 int ret; 2816 struct btrfs_path *path; 2817 struct btrfs_root *root = device->fs_info->chunk_root; 2818 struct btrfs_dev_item *dev_item; 2819 struct extent_buffer *leaf; 2820 struct btrfs_key key; 2821 2822 path = btrfs_alloc_path(); 2823 if (!path) 2824 return -ENOMEM; 2825 2826 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2827 key.type = BTRFS_DEV_ITEM_KEY; 2828 key.offset = device->devid; 2829 2830 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2831 if (ret < 0) 2832 goto out; 2833 2834 if (ret > 0) { 2835 ret = -ENOENT; 2836 goto out; 2837 } 2838 2839 leaf = path->nodes[0]; 2840 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2841 2842 btrfs_set_device_id(leaf, dev_item, device->devid); 2843 btrfs_set_device_type(leaf, dev_item, device->type); 2844 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2845 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2846 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2847 btrfs_set_device_total_bytes(leaf, dev_item, 2848 btrfs_device_get_disk_total_bytes(device)); 2849 btrfs_set_device_bytes_used(leaf, dev_item, 2850 btrfs_device_get_bytes_used(device)); 2851 btrfs_mark_buffer_dirty(leaf); 2852 2853 out: 2854 btrfs_free_path(path); 2855 return ret; 2856 } 2857 2858 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2859 struct btrfs_device *device, u64 new_size) 2860 { 2861 struct btrfs_fs_info *fs_info = device->fs_info; 2862 struct btrfs_super_block *super_copy = fs_info->super_copy; 2863 u64 old_total; 2864 u64 diff; 2865 2866 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2867 return -EACCES; 2868 2869 new_size = round_down(new_size, fs_info->sectorsize); 2870 2871 mutex_lock(&fs_info->chunk_mutex); 2872 old_total = btrfs_super_total_bytes(super_copy); 2873 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2874 2875 if (new_size <= device->total_bytes || 2876 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2877 mutex_unlock(&fs_info->chunk_mutex); 2878 return -EINVAL; 2879 } 2880 2881 btrfs_set_super_total_bytes(super_copy, 2882 round_down(old_total + diff, fs_info->sectorsize)); 2883 device->fs_devices->total_rw_bytes += diff; 2884 2885 btrfs_device_set_total_bytes(device, new_size); 2886 btrfs_device_set_disk_total_bytes(device, new_size); 2887 btrfs_clear_space_info_full(device->fs_info); 2888 if (list_empty(&device->post_commit_list)) 2889 list_add_tail(&device->post_commit_list, 2890 &trans->transaction->dev_update_list); 2891 mutex_unlock(&fs_info->chunk_mutex); 2892 2893 return btrfs_update_device(trans, device); 2894 } 2895 2896 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2897 { 2898 struct btrfs_fs_info *fs_info = trans->fs_info; 2899 struct btrfs_root *root = fs_info->chunk_root; 2900 int ret; 2901 struct btrfs_path *path; 2902 struct btrfs_key key; 2903 2904 path = btrfs_alloc_path(); 2905 if (!path) 2906 return -ENOMEM; 2907 2908 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2909 key.offset = chunk_offset; 2910 key.type = BTRFS_CHUNK_ITEM_KEY; 2911 2912 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2913 if (ret < 0) 2914 goto out; 2915 else if (ret > 0) { /* Logic error or corruption */ 2916 btrfs_handle_fs_error(fs_info, -ENOENT, 2917 "Failed lookup while freeing chunk."); 2918 ret = -ENOENT; 2919 goto out; 2920 } 2921 2922 ret = btrfs_del_item(trans, root, path); 2923 if (ret < 0) 2924 btrfs_handle_fs_error(fs_info, ret, 2925 "Failed to delete chunk item."); 2926 out: 2927 btrfs_free_path(path); 2928 return ret; 2929 } 2930 2931 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2932 { 2933 struct btrfs_super_block *super_copy = fs_info->super_copy; 2934 struct btrfs_disk_key *disk_key; 2935 struct btrfs_chunk *chunk; 2936 u8 *ptr; 2937 int ret = 0; 2938 u32 num_stripes; 2939 u32 array_size; 2940 u32 len = 0; 2941 u32 cur; 2942 struct btrfs_key key; 2943 2944 lockdep_assert_held(&fs_info->chunk_mutex); 2945 array_size = btrfs_super_sys_array_size(super_copy); 2946 2947 ptr = super_copy->sys_chunk_array; 2948 cur = 0; 2949 2950 while (cur < array_size) { 2951 disk_key = (struct btrfs_disk_key *)ptr; 2952 btrfs_disk_key_to_cpu(&key, disk_key); 2953 2954 len = sizeof(*disk_key); 2955 2956 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 2957 chunk = (struct btrfs_chunk *)(ptr + len); 2958 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 2959 len += btrfs_chunk_item_size(num_stripes); 2960 } else { 2961 ret = -EIO; 2962 break; 2963 } 2964 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 2965 key.offset == chunk_offset) { 2966 memmove(ptr, ptr + len, array_size - (cur + len)); 2967 array_size -= len; 2968 btrfs_set_super_sys_array_size(super_copy, array_size); 2969 } else { 2970 ptr += len; 2971 cur += len; 2972 } 2973 } 2974 return ret; 2975 } 2976 2977 /* 2978 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 2979 * @logical: Logical block offset in bytes. 2980 * @length: Length of extent in bytes. 2981 * 2982 * Return: Chunk mapping or ERR_PTR. 2983 */ 2984 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 2985 u64 logical, u64 length) 2986 { 2987 struct extent_map_tree *em_tree; 2988 struct extent_map *em; 2989 2990 em_tree = &fs_info->mapping_tree; 2991 read_lock(&em_tree->lock); 2992 em = lookup_extent_mapping(em_tree, logical, length); 2993 read_unlock(&em_tree->lock); 2994 2995 if (!em) { 2996 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 2997 logical, length); 2998 return ERR_PTR(-EINVAL); 2999 } 3000 3001 if (em->start > logical || em->start + em->len < logical) { 3002 btrfs_crit(fs_info, 3003 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3004 logical, length, em->start, em->start + em->len); 3005 free_extent_map(em); 3006 return ERR_PTR(-EINVAL); 3007 } 3008 3009 /* callers are responsible for dropping em's ref. */ 3010 return em; 3011 } 3012 3013 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3014 struct map_lookup *map, u64 chunk_offset) 3015 { 3016 int i; 3017 3018 /* 3019 * Removing chunk items and updating the device items in the chunks btree 3020 * requires holding the chunk_mutex. 3021 * See the comment at btrfs_chunk_alloc() for the details. 3022 */ 3023 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3024 3025 for (i = 0; i < map->num_stripes; i++) { 3026 int ret; 3027 3028 ret = btrfs_update_device(trans, map->stripes[i].dev); 3029 if (ret) 3030 return ret; 3031 } 3032 3033 return btrfs_free_chunk(trans, chunk_offset); 3034 } 3035 3036 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3037 { 3038 struct btrfs_fs_info *fs_info = trans->fs_info; 3039 struct extent_map *em; 3040 struct map_lookup *map; 3041 u64 dev_extent_len = 0; 3042 int i, ret = 0; 3043 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3044 3045 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3046 if (IS_ERR(em)) { 3047 /* 3048 * This is a logic error, but we don't want to just rely on the 3049 * user having built with ASSERT enabled, so if ASSERT doesn't 3050 * do anything we still error out. 3051 */ 3052 ASSERT(0); 3053 return PTR_ERR(em); 3054 } 3055 map = em->map_lookup; 3056 3057 /* 3058 * First delete the device extent items from the devices btree. 3059 * We take the device_list_mutex to avoid racing with the finishing phase 3060 * of a device replace operation. See the comment below before acquiring 3061 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3062 * because that can result in a deadlock when deleting the device extent 3063 * items from the devices btree - COWing an extent buffer from the btree 3064 * may result in allocating a new metadata chunk, which would attempt to 3065 * lock again fs_info->chunk_mutex. 3066 */ 3067 mutex_lock(&fs_devices->device_list_mutex); 3068 for (i = 0; i < map->num_stripes; i++) { 3069 struct btrfs_device *device = map->stripes[i].dev; 3070 ret = btrfs_free_dev_extent(trans, device, 3071 map->stripes[i].physical, 3072 &dev_extent_len); 3073 if (ret) { 3074 mutex_unlock(&fs_devices->device_list_mutex); 3075 btrfs_abort_transaction(trans, ret); 3076 goto out; 3077 } 3078 3079 if (device->bytes_used > 0) { 3080 mutex_lock(&fs_info->chunk_mutex); 3081 btrfs_device_set_bytes_used(device, 3082 device->bytes_used - dev_extent_len); 3083 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3084 btrfs_clear_space_info_full(fs_info); 3085 mutex_unlock(&fs_info->chunk_mutex); 3086 } 3087 } 3088 mutex_unlock(&fs_devices->device_list_mutex); 3089 3090 /* 3091 * We acquire fs_info->chunk_mutex for 2 reasons: 3092 * 3093 * 1) Just like with the first phase of the chunk allocation, we must 3094 * reserve system space, do all chunk btree updates and deletions, and 3095 * update the system chunk array in the superblock while holding this 3096 * mutex. This is for similar reasons as explained on the comment at 3097 * the top of btrfs_chunk_alloc(); 3098 * 3099 * 2) Prevent races with the final phase of a device replace operation 3100 * that replaces the device object associated with the map's stripes, 3101 * because the device object's id can change at any time during that 3102 * final phase of the device replace operation 3103 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3104 * replaced device and then see it with an ID of 3105 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3106 * the device item, which does not exists on the chunk btree. 3107 * The finishing phase of device replace acquires both the 3108 * device_list_mutex and the chunk_mutex, in that order, so we are 3109 * safe by just acquiring the chunk_mutex. 3110 */ 3111 trans->removing_chunk = true; 3112 mutex_lock(&fs_info->chunk_mutex); 3113 3114 check_system_chunk(trans, map->type); 3115 3116 ret = remove_chunk_item(trans, map, chunk_offset); 3117 /* 3118 * Normally we should not get -ENOSPC since we reserved space before 3119 * through the call to check_system_chunk(). 3120 * 3121 * Despite our system space_info having enough free space, we may not 3122 * be able to allocate extents from its block groups, because all have 3123 * an incompatible profile, which will force us to allocate a new system 3124 * block group with the right profile, or right after we called 3125 * check_system_space() above, a scrub turned the only system block group 3126 * with enough free space into RO mode. 3127 * This is explained with more detail at do_chunk_alloc(). 3128 * 3129 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3130 */ 3131 if (ret == -ENOSPC) { 3132 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3133 struct btrfs_block_group *sys_bg; 3134 3135 sys_bg = btrfs_create_chunk(trans, sys_flags); 3136 if (IS_ERR(sys_bg)) { 3137 ret = PTR_ERR(sys_bg); 3138 btrfs_abort_transaction(trans, ret); 3139 goto out; 3140 } 3141 3142 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3143 if (ret) { 3144 btrfs_abort_transaction(trans, ret); 3145 goto out; 3146 } 3147 3148 ret = remove_chunk_item(trans, map, chunk_offset); 3149 if (ret) { 3150 btrfs_abort_transaction(trans, ret); 3151 goto out; 3152 } 3153 } else if (ret) { 3154 btrfs_abort_transaction(trans, ret); 3155 goto out; 3156 } 3157 3158 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3159 3160 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3161 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3162 if (ret) { 3163 btrfs_abort_transaction(trans, ret); 3164 goto out; 3165 } 3166 } 3167 3168 mutex_unlock(&fs_info->chunk_mutex); 3169 trans->removing_chunk = false; 3170 3171 /* 3172 * We are done with chunk btree updates and deletions, so release the 3173 * system space we previously reserved (with check_system_chunk()). 3174 */ 3175 btrfs_trans_release_chunk_metadata(trans); 3176 3177 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3178 if (ret) { 3179 btrfs_abort_transaction(trans, ret); 3180 goto out; 3181 } 3182 3183 out: 3184 if (trans->removing_chunk) { 3185 mutex_unlock(&fs_info->chunk_mutex); 3186 trans->removing_chunk = false; 3187 } 3188 /* once for us */ 3189 free_extent_map(em); 3190 return ret; 3191 } 3192 3193 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3194 { 3195 struct btrfs_root *root = fs_info->chunk_root; 3196 struct btrfs_trans_handle *trans; 3197 struct btrfs_block_group *block_group; 3198 u64 length; 3199 int ret; 3200 3201 /* 3202 * Prevent races with automatic removal of unused block groups. 3203 * After we relocate and before we remove the chunk with offset 3204 * chunk_offset, automatic removal of the block group can kick in, 3205 * resulting in a failure when calling btrfs_remove_chunk() below. 3206 * 3207 * Make sure to acquire this mutex before doing a tree search (dev 3208 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3209 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3210 * we release the path used to search the chunk/dev tree and before 3211 * the current task acquires this mutex and calls us. 3212 */ 3213 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3214 3215 /* step one, relocate all the extents inside this chunk */ 3216 btrfs_scrub_pause(fs_info); 3217 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3218 btrfs_scrub_continue(fs_info); 3219 if (ret) 3220 return ret; 3221 3222 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3223 if (!block_group) 3224 return -ENOENT; 3225 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3226 length = block_group->length; 3227 btrfs_put_block_group(block_group); 3228 3229 /* 3230 * On a zoned file system, discard the whole block group, this will 3231 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3232 * resetting the zone fails, don't treat it as a fatal problem from the 3233 * filesystem's point of view. 3234 */ 3235 if (btrfs_is_zoned(fs_info)) { 3236 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3237 if (ret) 3238 btrfs_info(fs_info, 3239 "failed to reset zone %llu after relocation", 3240 chunk_offset); 3241 } 3242 3243 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3244 chunk_offset); 3245 if (IS_ERR(trans)) { 3246 ret = PTR_ERR(trans); 3247 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3248 return ret; 3249 } 3250 3251 /* 3252 * step two, delete the device extents and the 3253 * chunk tree entries 3254 */ 3255 ret = btrfs_remove_chunk(trans, chunk_offset); 3256 btrfs_end_transaction(trans); 3257 return ret; 3258 } 3259 3260 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3261 { 3262 struct btrfs_root *chunk_root = fs_info->chunk_root; 3263 struct btrfs_path *path; 3264 struct extent_buffer *leaf; 3265 struct btrfs_chunk *chunk; 3266 struct btrfs_key key; 3267 struct btrfs_key found_key; 3268 u64 chunk_type; 3269 bool retried = false; 3270 int failed = 0; 3271 int ret; 3272 3273 path = btrfs_alloc_path(); 3274 if (!path) 3275 return -ENOMEM; 3276 3277 again: 3278 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3279 key.offset = (u64)-1; 3280 key.type = BTRFS_CHUNK_ITEM_KEY; 3281 3282 while (1) { 3283 mutex_lock(&fs_info->reclaim_bgs_lock); 3284 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3285 if (ret < 0) { 3286 mutex_unlock(&fs_info->reclaim_bgs_lock); 3287 goto error; 3288 } 3289 BUG_ON(ret == 0); /* Corruption */ 3290 3291 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3292 key.type); 3293 if (ret) 3294 mutex_unlock(&fs_info->reclaim_bgs_lock); 3295 if (ret < 0) 3296 goto error; 3297 if (ret > 0) 3298 break; 3299 3300 leaf = path->nodes[0]; 3301 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3302 3303 chunk = btrfs_item_ptr(leaf, path->slots[0], 3304 struct btrfs_chunk); 3305 chunk_type = btrfs_chunk_type(leaf, chunk); 3306 btrfs_release_path(path); 3307 3308 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3309 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3310 if (ret == -ENOSPC) 3311 failed++; 3312 else 3313 BUG_ON(ret); 3314 } 3315 mutex_unlock(&fs_info->reclaim_bgs_lock); 3316 3317 if (found_key.offset == 0) 3318 break; 3319 key.offset = found_key.offset - 1; 3320 } 3321 ret = 0; 3322 if (failed && !retried) { 3323 failed = 0; 3324 retried = true; 3325 goto again; 3326 } else if (WARN_ON(failed && retried)) { 3327 ret = -ENOSPC; 3328 } 3329 error: 3330 btrfs_free_path(path); 3331 return ret; 3332 } 3333 3334 /* 3335 * return 1 : allocate a data chunk successfully, 3336 * return <0: errors during allocating a data chunk, 3337 * return 0 : no need to allocate a data chunk. 3338 */ 3339 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3340 u64 chunk_offset) 3341 { 3342 struct btrfs_block_group *cache; 3343 u64 bytes_used; 3344 u64 chunk_type; 3345 3346 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3347 ASSERT(cache); 3348 chunk_type = cache->flags; 3349 btrfs_put_block_group(cache); 3350 3351 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3352 return 0; 3353 3354 spin_lock(&fs_info->data_sinfo->lock); 3355 bytes_used = fs_info->data_sinfo->bytes_used; 3356 spin_unlock(&fs_info->data_sinfo->lock); 3357 3358 if (!bytes_used) { 3359 struct btrfs_trans_handle *trans; 3360 int ret; 3361 3362 trans = btrfs_join_transaction(fs_info->tree_root); 3363 if (IS_ERR(trans)) 3364 return PTR_ERR(trans); 3365 3366 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3367 btrfs_end_transaction(trans); 3368 if (ret < 0) 3369 return ret; 3370 return 1; 3371 } 3372 3373 return 0; 3374 } 3375 3376 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3377 struct btrfs_balance_control *bctl) 3378 { 3379 struct btrfs_root *root = fs_info->tree_root; 3380 struct btrfs_trans_handle *trans; 3381 struct btrfs_balance_item *item; 3382 struct btrfs_disk_balance_args disk_bargs; 3383 struct btrfs_path *path; 3384 struct extent_buffer *leaf; 3385 struct btrfs_key key; 3386 int ret, err; 3387 3388 path = btrfs_alloc_path(); 3389 if (!path) 3390 return -ENOMEM; 3391 3392 trans = btrfs_start_transaction(root, 0); 3393 if (IS_ERR(trans)) { 3394 btrfs_free_path(path); 3395 return PTR_ERR(trans); 3396 } 3397 3398 key.objectid = BTRFS_BALANCE_OBJECTID; 3399 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3400 key.offset = 0; 3401 3402 ret = btrfs_insert_empty_item(trans, root, path, &key, 3403 sizeof(*item)); 3404 if (ret) 3405 goto out; 3406 3407 leaf = path->nodes[0]; 3408 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3409 3410 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3411 3412 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3413 btrfs_set_balance_data(leaf, item, &disk_bargs); 3414 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3415 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3416 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3417 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3418 3419 btrfs_set_balance_flags(leaf, item, bctl->flags); 3420 3421 btrfs_mark_buffer_dirty(leaf); 3422 out: 3423 btrfs_free_path(path); 3424 err = btrfs_commit_transaction(trans); 3425 if (err && !ret) 3426 ret = err; 3427 return ret; 3428 } 3429 3430 static int del_balance_item(struct btrfs_fs_info *fs_info) 3431 { 3432 struct btrfs_root *root = fs_info->tree_root; 3433 struct btrfs_trans_handle *trans; 3434 struct btrfs_path *path; 3435 struct btrfs_key key; 3436 int ret, err; 3437 3438 path = btrfs_alloc_path(); 3439 if (!path) 3440 return -ENOMEM; 3441 3442 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3443 if (IS_ERR(trans)) { 3444 btrfs_free_path(path); 3445 return PTR_ERR(trans); 3446 } 3447 3448 key.objectid = BTRFS_BALANCE_OBJECTID; 3449 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3450 key.offset = 0; 3451 3452 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3453 if (ret < 0) 3454 goto out; 3455 if (ret > 0) { 3456 ret = -ENOENT; 3457 goto out; 3458 } 3459 3460 ret = btrfs_del_item(trans, root, path); 3461 out: 3462 btrfs_free_path(path); 3463 err = btrfs_commit_transaction(trans); 3464 if (err && !ret) 3465 ret = err; 3466 return ret; 3467 } 3468 3469 /* 3470 * This is a heuristic used to reduce the number of chunks balanced on 3471 * resume after balance was interrupted. 3472 */ 3473 static void update_balance_args(struct btrfs_balance_control *bctl) 3474 { 3475 /* 3476 * Turn on soft mode for chunk types that were being converted. 3477 */ 3478 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3479 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3480 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3481 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3482 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3483 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3484 3485 /* 3486 * Turn on usage filter if is not already used. The idea is 3487 * that chunks that we have already balanced should be 3488 * reasonably full. Don't do it for chunks that are being 3489 * converted - that will keep us from relocating unconverted 3490 * (albeit full) chunks. 3491 */ 3492 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3493 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3494 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3495 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3496 bctl->data.usage = 90; 3497 } 3498 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3499 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3500 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3501 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3502 bctl->sys.usage = 90; 3503 } 3504 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3505 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3506 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3507 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3508 bctl->meta.usage = 90; 3509 } 3510 } 3511 3512 /* 3513 * Clear the balance status in fs_info and delete the balance item from disk. 3514 */ 3515 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3516 { 3517 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3518 int ret; 3519 3520 BUG_ON(!fs_info->balance_ctl); 3521 3522 spin_lock(&fs_info->balance_lock); 3523 fs_info->balance_ctl = NULL; 3524 spin_unlock(&fs_info->balance_lock); 3525 3526 kfree(bctl); 3527 ret = del_balance_item(fs_info); 3528 if (ret) 3529 btrfs_handle_fs_error(fs_info, ret, NULL); 3530 } 3531 3532 /* 3533 * Balance filters. Return 1 if chunk should be filtered out 3534 * (should not be balanced). 3535 */ 3536 static int chunk_profiles_filter(u64 chunk_type, 3537 struct btrfs_balance_args *bargs) 3538 { 3539 chunk_type = chunk_to_extended(chunk_type) & 3540 BTRFS_EXTENDED_PROFILE_MASK; 3541 3542 if (bargs->profiles & chunk_type) 3543 return 0; 3544 3545 return 1; 3546 } 3547 3548 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3549 struct btrfs_balance_args *bargs) 3550 { 3551 struct btrfs_block_group *cache; 3552 u64 chunk_used; 3553 u64 user_thresh_min; 3554 u64 user_thresh_max; 3555 int ret = 1; 3556 3557 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3558 chunk_used = cache->used; 3559 3560 if (bargs->usage_min == 0) 3561 user_thresh_min = 0; 3562 else 3563 user_thresh_min = div_factor_fine(cache->length, 3564 bargs->usage_min); 3565 3566 if (bargs->usage_max == 0) 3567 user_thresh_max = 1; 3568 else if (bargs->usage_max > 100) 3569 user_thresh_max = cache->length; 3570 else 3571 user_thresh_max = div_factor_fine(cache->length, 3572 bargs->usage_max); 3573 3574 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3575 ret = 0; 3576 3577 btrfs_put_block_group(cache); 3578 return ret; 3579 } 3580 3581 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3582 u64 chunk_offset, struct btrfs_balance_args *bargs) 3583 { 3584 struct btrfs_block_group *cache; 3585 u64 chunk_used, user_thresh; 3586 int ret = 1; 3587 3588 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3589 chunk_used = cache->used; 3590 3591 if (bargs->usage_min == 0) 3592 user_thresh = 1; 3593 else if (bargs->usage > 100) 3594 user_thresh = cache->length; 3595 else 3596 user_thresh = div_factor_fine(cache->length, bargs->usage); 3597 3598 if (chunk_used < user_thresh) 3599 ret = 0; 3600 3601 btrfs_put_block_group(cache); 3602 return ret; 3603 } 3604 3605 static int chunk_devid_filter(struct extent_buffer *leaf, 3606 struct btrfs_chunk *chunk, 3607 struct btrfs_balance_args *bargs) 3608 { 3609 struct btrfs_stripe *stripe; 3610 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3611 int i; 3612 3613 for (i = 0; i < num_stripes; i++) { 3614 stripe = btrfs_stripe_nr(chunk, i); 3615 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3616 return 0; 3617 } 3618 3619 return 1; 3620 } 3621 3622 static u64 calc_data_stripes(u64 type, int num_stripes) 3623 { 3624 const int index = btrfs_bg_flags_to_raid_index(type); 3625 const int ncopies = btrfs_raid_array[index].ncopies; 3626 const int nparity = btrfs_raid_array[index].nparity; 3627 3628 return (num_stripes - nparity) / ncopies; 3629 } 3630 3631 /* [pstart, pend) */ 3632 static int chunk_drange_filter(struct extent_buffer *leaf, 3633 struct btrfs_chunk *chunk, 3634 struct btrfs_balance_args *bargs) 3635 { 3636 struct btrfs_stripe *stripe; 3637 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3638 u64 stripe_offset; 3639 u64 stripe_length; 3640 u64 type; 3641 int factor; 3642 int i; 3643 3644 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3645 return 0; 3646 3647 type = btrfs_chunk_type(leaf, chunk); 3648 factor = calc_data_stripes(type, num_stripes); 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 continue; 3654 3655 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3656 stripe_length = btrfs_chunk_length(leaf, chunk); 3657 stripe_length = div_u64(stripe_length, factor); 3658 3659 if (stripe_offset < bargs->pend && 3660 stripe_offset + stripe_length > bargs->pstart) 3661 return 0; 3662 } 3663 3664 return 1; 3665 } 3666 3667 /* [vstart, vend) */ 3668 static int chunk_vrange_filter(struct extent_buffer *leaf, 3669 struct btrfs_chunk *chunk, 3670 u64 chunk_offset, 3671 struct btrfs_balance_args *bargs) 3672 { 3673 if (chunk_offset < bargs->vend && 3674 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3675 /* at least part of the chunk is inside this vrange */ 3676 return 0; 3677 3678 return 1; 3679 } 3680 3681 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3682 struct btrfs_chunk *chunk, 3683 struct btrfs_balance_args *bargs) 3684 { 3685 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3686 3687 if (bargs->stripes_min <= num_stripes 3688 && num_stripes <= bargs->stripes_max) 3689 return 0; 3690 3691 return 1; 3692 } 3693 3694 static int chunk_soft_convert_filter(u64 chunk_type, 3695 struct btrfs_balance_args *bargs) 3696 { 3697 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3698 return 0; 3699 3700 chunk_type = chunk_to_extended(chunk_type) & 3701 BTRFS_EXTENDED_PROFILE_MASK; 3702 3703 if (bargs->target == chunk_type) 3704 return 1; 3705 3706 return 0; 3707 } 3708 3709 static int should_balance_chunk(struct extent_buffer *leaf, 3710 struct btrfs_chunk *chunk, u64 chunk_offset) 3711 { 3712 struct btrfs_fs_info *fs_info = leaf->fs_info; 3713 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3714 struct btrfs_balance_args *bargs = NULL; 3715 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3716 3717 /* type filter */ 3718 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3719 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3720 return 0; 3721 } 3722 3723 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3724 bargs = &bctl->data; 3725 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3726 bargs = &bctl->sys; 3727 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3728 bargs = &bctl->meta; 3729 3730 /* profiles filter */ 3731 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3732 chunk_profiles_filter(chunk_type, bargs)) { 3733 return 0; 3734 } 3735 3736 /* usage filter */ 3737 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3738 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3739 return 0; 3740 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3741 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3742 return 0; 3743 } 3744 3745 /* devid filter */ 3746 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3747 chunk_devid_filter(leaf, chunk, bargs)) { 3748 return 0; 3749 } 3750 3751 /* drange filter, makes sense only with devid filter */ 3752 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3753 chunk_drange_filter(leaf, chunk, bargs)) { 3754 return 0; 3755 } 3756 3757 /* vrange filter */ 3758 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3759 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3760 return 0; 3761 } 3762 3763 /* stripes filter */ 3764 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3765 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3766 return 0; 3767 } 3768 3769 /* soft profile changing mode */ 3770 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3771 chunk_soft_convert_filter(chunk_type, bargs)) { 3772 return 0; 3773 } 3774 3775 /* 3776 * limited by count, must be the last filter 3777 */ 3778 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3779 if (bargs->limit == 0) 3780 return 0; 3781 else 3782 bargs->limit--; 3783 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3784 /* 3785 * Same logic as the 'limit' filter; the minimum cannot be 3786 * determined here because we do not have the global information 3787 * about the count of all chunks that satisfy the filters. 3788 */ 3789 if (bargs->limit_max == 0) 3790 return 0; 3791 else 3792 bargs->limit_max--; 3793 } 3794 3795 return 1; 3796 } 3797 3798 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3799 { 3800 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3801 struct btrfs_root *chunk_root = fs_info->chunk_root; 3802 u64 chunk_type; 3803 struct btrfs_chunk *chunk; 3804 struct btrfs_path *path = NULL; 3805 struct btrfs_key key; 3806 struct btrfs_key found_key; 3807 struct extent_buffer *leaf; 3808 int slot; 3809 int ret; 3810 int enospc_errors = 0; 3811 bool counting = true; 3812 /* The single value limit and min/max limits use the same bytes in the */ 3813 u64 limit_data = bctl->data.limit; 3814 u64 limit_meta = bctl->meta.limit; 3815 u64 limit_sys = bctl->sys.limit; 3816 u32 count_data = 0; 3817 u32 count_meta = 0; 3818 u32 count_sys = 0; 3819 int chunk_reserved = 0; 3820 3821 path = btrfs_alloc_path(); 3822 if (!path) { 3823 ret = -ENOMEM; 3824 goto error; 3825 } 3826 3827 /* zero out stat counters */ 3828 spin_lock(&fs_info->balance_lock); 3829 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3830 spin_unlock(&fs_info->balance_lock); 3831 again: 3832 if (!counting) { 3833 /* 3834 * The single value limit and min/max limits use the same bytes 3835 * in the 3836 */ 3837 bctl->data.limit = limit_data; 3838 bctl->meta.limit = limit_meta; 3839 bctl->sys.limit = limit_sys; 3840 } 3841 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3842 key.offset = (u64)-1; 3843 key.type = BTRFS_CHUNK_ITEM_KEY; 3844 3845 while (1) { 3846 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3847 atomic_read(&fs_info->balance_cancel_req)) { 3848 ret = -ECANCELED; 3849 goto error; 3850 } 3851 3852 mutex_lock(&fs_info->reclaim_bgs_lock); 3853 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3854 if (ret < 0) { 3855 mutex_unlock(&fs_info->reclaim_bgs_lock); 3856 goto error; 3857 } 3858 3859 /* 3860 * this shouldn't happen, it means the last relocate 3861 * failed 3862 */ 3863 if (ret == 0) 3864 BUG(); /* FIXME break ? */ 3865 3866 ret = btrfs_previous_item(chunk_root, path, 0, 3867 BTRFS_CHUNK_ITEM_KEY); 3868 if (ret) { 3869 mutex_unlock(&fs_info->reclaim_bgs_lock); 3870 ret = 0; 3871 break; 3872 } 3873 3874 leaf = path->nodes[0]; 3875 slot = path->slots[0]; 3876 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3877 3878 if (found_key.objectid != key.objectid) { 3879 mutex_unlock(&fs_info->reclaim_bgs_lock); 3880 break; 3881 } 3882 3883 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3884 chunk_type = btrfs_chunk_type(leaf, chunk); 3885 3886 if (!counting) { 3887 spin_lock(&fs_info->balance_lock); 3888 bctl->stat.considered++; 3889 spin_unlock(&fs_info->balance_lock); 3890 } 3891 3892 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3893 3894 btrfs_release_path(path); 3895 if (!ret) { 3896 mutex_unlock(&fs_info->reclaim_bgs_lock); 3897 goto loop; 3898 } 3899 3900 if (counting) { 3901 mutex_unlock(&fs_info->reclaim_bgs_lock); 3902 spin_lock(&fs_info->balance_lock); 3903 bctl->stat.expected++; 3904 spin_unlock(&fs_info->balance_lock); 3905 3906 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3907 count_data++; 3908 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3909 count_sys++; 3910 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3911 count_meta++; 3912 3913 goto loop; 3914 } 3915 3916 /* 3917 * Apply limit_min filter, no need to check if the LIMITS 3918 * filter is used, limit_min is 0 by default 3919 */ 3920 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3921 count_data < bctl->data.limit_min) 3922 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3923 count_meta < bctl->meta.limit_min) 3924 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3925 count_sys < bctl->sys.limit_min)) { 3926 mutex_unlock(&fs_info->reclaim_bgs_lock); 3927 goto loop; 3928 } 3929 3930 if (!chunk_reserved) { 3931 /* 3932 * We may be relocating the only data chunk we have, 3933 * which could potentially end up with losing data's 3934 * raid profile, so lets allocate an empty one in 3935 * advance. 3936 */ 3937 ret = btrfs_may_alloc_data_chunk(fs_info, 3938 found_key.offset); 3939 if (ret < 0) { 3940 mutex_unlock(&fs_info->reclaim_bgs_lock); 3941 goto error; 3942 } else if (ret == 1) { 3943 chunk_reserved = 1; 3944 } 3945 } 3946 3947 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3948 mutex_unlock(&fs_info->reclaim_bgs_lock); 3949 if (ret == -ENOSPC) { 3950 enospc_errors++; 3951 } else if (ret == -ETXTBSY) { 3952 btrfs_info(fs_info, 3953 "skipping relocation of block group %llu due to active swapfile", 3954 found_key.offset); 3955 ret = 0; 3956 } else if (ret) { 3957 goto error; 3958 } else { 3959 spin_lock(&fs_info->balance_lock); 3960 bctl->stat.completed++; 3961 spin_unlock(&fs_info->balance_lock); 3962 } 3963 loop: 3964 if (found_key.offset == 0) 3965 break; 3966 key.offset = found_key.offset - 1; 3967 } 3968 3969 if (counting) { 3970 btrfs_release_path(path); 3971 counting = false; 3972 goto again; 3973 } 3974 error: 3975 btrfs_free_path(path); 3976 if (enospc_errors) { 3977 btrfs_info(fs_info, "%d enospc errors during balance", 3978 enospc_errors); 3979 if (!ret) 3980 ret = -ENOSPC; 3981 } 3982 3983 return ret; 3984 } 3985 3986 /** 3987 * alloc_profile_is_valid - see if a given profile is valid and reduced 3988 * @flags: profile to validate 3989 * @extended: if true @flags is treated as an extended profile 3990 */ 3991 static int alloc_profile_is_valid(u64 flags, int extended) 3992 { 3993 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 3994 BTRFS_BLOCK_GROUP_PROFILE_MASK); 3995 3996 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 3997 3998 /* 1) check that all other bits are zeroed */ 3999 if (flags & ~mask) 4000 return 0; 4001 4002 /* 2) see if profile is reduced */ 4003 if (flags == 0) 4004 return !extended; /* "0" is valid for usual profiles */ 4005 4006 return has_single_bit_set(flags); 4007 } 4008 4009 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4010 { 4011 /* cancel requested || normal exit path */ 4012 return atomic_read(&fs_info->balance_cancel_req) || 4013 (atomic_read(&fs_info->balance_pause_req) == 0 && 4014 atomic_read(&fs_info->balance_cancel_req) == 0); 4015 } 4016 4017 /* 4018 * Validate target profile against allowed profiles and return true if it's OK. 4019 * Otherwise print the error message and return false. 4020 */ 4021 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4022 const struct btrfs_balance_args *bargs, 4023 u64 allowed, const char *type) 4024 { 4025 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4026 return true; 4027 4028 if (fs_info->sectorsize < PAGE_SIZE && 4029 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) { 4030 btrfs_err(fs_info, 4031 "RAID56 is not yet supported for sectorsize %u with page size %lu", 4032 fs_info->sectorsize, PAGE_SIZE); 4033 return false; 4034 } 4035 /* Profile is valid and does not have bits outside of the allowed set */ 4036 if (alloc_profile_is_valid(bargs->target, 1) && 4037 (bargs->target & ~allowed) == 0) 4038 return true; 4039 4040 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4041 type, btrfs_bg_type_to_raid_name(bargs->target)); 4042 return false; 4043 } 4044 4045 /* 4046 * Fill @buf with textual description of balance filter flags @bargs, up to 4047 * @size_buf including the terminating null. The output may be trimmed if it 4048 * does not fit into the provided buffer. 4049 */ 4050 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4051 u32 size_buf) 4052 { 4053 int ret; 4054 u32 size_bp = size_buf; 4055 char *bp = buf; 4056 u64 flags = bargs->flags; 4057 char tmp_buf[128] = {'\0'}; 4058 4059 if (!flags) 4060 return; 4061 4062 #define CHECK_APPEND_NOARG(a) \ 4063 do { \ 4064 ret = snprintf(bp, size_bp, (a)); \ 4065 if (ret < 0 || ret >= size_bp) \ 4066 goto out_overflow; \ 4067 size_bp -= ret; \ 4068 bp += ret; \ 4069 } while (0) 4070 4071 #define CHECK_APPEND_1ARG(a, v1) \ 4072 do { \ 4073 ret = snprintf(bp, size_bp, (a), (v1)); \ 4074 if (ret < 0 || ret >= size_bp) \ 4075 goto out_overflow; \ 4076 size_bp -= ret; \ 4077 bp += ret; \ 4078 } while (0) 4079 4080 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4081 do { \ 4082 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4083 if (ret < 0 || ret >= size_bp) \ 4084 goto out_overflow; \ 4085 size_bp -= ret; \ 4086 bp += ret; \ 4087 } while (0) 4088 4089 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4090 CHECK_APPEND_1ARG("convert=%s,", 4091 btrfs_bg_type_to_raid_name(bargs->target)); 4092 4093 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4094 CHECK_APPEND_NOARG("soft,"); 4095 4096 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4097 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4098 sizeof(tmp_buf)); 4099 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4100 } 4101 4102 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4103 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4104 4105 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4106 CHECK_APPEND_2ARG("usage=%u..%u,", 4107 bargs->usage_min, bargs->usage_max); 4108 4109 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4110 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4111 4112 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4113 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4114 bargs->pstart, bargs->pend); 4115 4116 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4117 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4118 bargs->vstart, bargs->vend); 4119 4120 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4121 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4122 4123 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4124 CHECK_APPEND_2ARG("limit=%u..%u,", 4125 bargs->limit_min, bargs->limit_max); 4126 4127 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4128 CHECK_APPEND_2ARG("stripes=%u..%u,", 4129 bargs->stripes_min, bargs->stripes_max); 4130 4131 #undef CHECK_APPEND_2ARG 4132 #undef CHECK_APPEND_1ARG 4133 #undef CHECK_APPEND_NOARG 4134 4135 out_overflow: 4136 4137 if (size_bp < size_buf) 4138 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4139 else 4140 buf[0] = '\0'; 4141 } 4142 4143 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4144 { 4145 u32 size_buf = 1024; 4146 char tmp_buf[192] = {'\0'}; 4147 char *buf; 4148 char *bp; 4149 u32 size_bp = size_buf; 4150 int ret; 4151 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4152 4153 buf = kzalloc(size_buf, GFP_KERNEL); 4154 if (!buf) 4155 return; 4156 4157 bp = buf; 4158 4159 #define CHECK_APPEND_1ARG(a, v1) \ 4160 do { \ 4161 ret = snprintf(bp, size_bp, (a), (v1)); \ 4162 if (ret < 0 || ret >= size_bp) \ 4163 goto out_overflow; \ 4164 size_bp -= ret; \ 4165 bp += ret; \ 4166 } while (0) 4167 4168 if (bctl->flags & BTRFS_BALANCE_FORCE) 4169 CHECK_APPEND_1ARG("%s", "-f "); 4170 4171 if (bctl->flags & BTRFS_BALANCE_DATA) { 4172 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4173 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4174 } 4175 4176 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4177 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4178 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4179 } 4180 4181 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4182 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4183 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4184 } 4185 4186 #undef CHECK_APPEND_1ARG 4187 4188 out_overflow: 4189 4190 if (size_bp < size_buf) 4191 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4192 btrfs_info(fs_info, "balance: %s %s", 4193 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4194 "resume" : "start", buf); 4195 4196 kfree(buf); 4197 } 4198 4199 /* 4200 * Should be called with balance mutexe held 4201 */ 4202 int btrfs_balance(struct btrfs_fs_info *fs_info, 4203 struct btrfs_balance_control *bctl, 4204 struct btrfs_ioctl_balance_args *bargs) 4205 { 4206 u64 meta_target, data_target; 4207 u64 allowed; 4208 int mixed = 0; 4209 int ret; 4210 u64 num_devices; 4211 unsigned seq; 4212 bool reducing_redundancy; 4213 int i; 4214 4215 if (btrfs_fs_closing(fs_info) || 4216 atomic_read(&fs_info->balance_pause_req) || 4217 btrfs_should_cancel_balance(fs_info)) { 4218 ret = -EINVAL; 4219 goto out; 4220 } 4221 4222 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4223 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4224 mixed = 1; 4225 4226 /* 4227 * In case of mixed groups both data and meta should be picked, 4228 * and identical options should be given for both of them. 4229 */ 4230 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4231 if (mixed && (bctl->flags & allowed)) { 4232 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4233 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4234 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4235 btrfs_err(fs_info, 4236 "balance: mixed groups data and metadata options must be the same"); 4237 ret = -EINVAL; 4238 goto out; 4239 } 4240 } 4241 4242 /* 4243 * rw_devices will not change at the moment, device add/delete/replace 4244 * are exclusive 4245 */ 4246 num_devices = fs_info->fs_devices->rw_devices; 4247 4248 /* 4249 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4250 * special bit for it, to make it easier to distinguish. Thus we need 4251 * to set it manually, or balance would refuse the profile. 4252 */ 4253 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4254 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4255 if (num_devices >= btrfs_raid_array[i].devs_min) 4256 allowed |= btrfs_raid_array[i].bg_flag; 4257 4258 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4259 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4260 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4261 ret = -EINVAL; 4262 goto out; 4263 } 4264 4265 /* 4266 * Allow to reduce metadata or system integrity only if force set for 4267 * profiles with redundancy (copies, parity) 4268 */ 4269 allowed = 0; 4270 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4271 if (btrfs_raid_array[i].ncopies >= 2 || 4272 btrfs_raid_array[i].tolerated_failures >= 1) 4273 allowed |= btrfs_raid_array[i].bg_flag; 4274 } 4275 do { 4276 seq = read_seqbegin(&fs_info->profiles_lock); 4277 4278 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4279 (fs_info->avail_system_alloc_bits & allowed) && 4280 !(bctl->sys.target & allowed)) || 4281 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4282 (fs_info->avail_metadata_alloc_bits & allowed) && 4283 !(bctl->meta.target & allowed))) 4284 reducing_redundancy = true; 4285 else 4286 reducing_redundancy = false; 4287 4288 /* if we're not converting, the target field is uninitialized */ 4289 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4290 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4291 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4292 bctl->data.target : fs_info->avail_data_alloc_bits; 4293 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4294 4295 if (reducing_redundancy) { 4296 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4297 btrfs_info(fs_info, 4298 "balance: force reducing metadata redundancy"); 4299 } else { 4300 btrfs_err(fs_info, 4301 "balance: reduces metadata redundancy, use --force if you want this"); 4302 ret = -EINVAL; 4303 goto out; 4304 } 4305 } 4306 4307 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4308 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4309 btrfs_warn(fs_info, 4310 "balance: metadata profile %s has lower redundancy than data profile %s", 4311 btrfs_bg_type_to_raid_name(meta_target), 4312 btrfs_bg_type_to_raid_name(data_target)); 4313 } 4314 4315 ret = insert_balance_item(fs_info, bctl); 4316 if (ret && ret != -EEXIST) 4317 goto out; 4318 4319 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4320 BUG_ON(ret == -EEXIST); 4321 BUG_ON(fs_info->balance_ctl); 4322 spin_lock(&fs_info->balance_lock); 4323 fs_info->balance_ctl = bctl; 4324 spin_unlock(&fs_info->balance_lock); 4325 } else { 4326 BUG_ON(ret != -EEXIST); 4327 spin_lock(&fs_info->balance_lock); 4328 update_balance_args(bctl); 4329 spin_unlock(&fs_info->balance_lock); 4330 } 4331 4332 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4333 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4334 describe_balance_start_or_resume(fs_info); 4335 mutex_unlock(&fs_info->balance_mutex); 4336 4337 ret = __btrfs_balance(fs_info); 4338 4339 mutex_lock(&fs_info->balance_mutex); 4340 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) 4341 btrfs_info(fs_info, "balance: paused"); 4342 /* 4343 * Balance can be canceled by: 4344 * 4345 * - Regular cancel request 4346 * Then ret == -ECANCELED and balance_cancel_req > 0 4347 * 4348 * - Fatal signal to "btrfs" process 4349 * Either the signal caught by wait_reserve_ticket() and callers 4350 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4351 * got -ECANCELED. 4352 * Either way, in this case balance_cancel_req = 0, and 4353 * ret == -EINTR or ret == -ECANCELED. 4354 * 4355 * So here we only check the return value to catch canceled balance. 4356 */ 4357 else if (ret == -ECANCELED || ret == -EINTR) 4358 btrfs_info(fs_info, "balance: canceled"); 4359 else 4360 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4361 4362 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4363 4364 if (bargs) { 4365 memset(bargs, 0, sizeof(*bargs)); 4366 btrfs_update_ioctl_balance_args(fs_info, bargs); 4367 } 4368 4369 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4370 balance_need_close(fs_info)) { 4371 reset_balance_state(fs_info); 4372 btrfs_exclop_finish(fs_info); 4373 } 4374 4375 wake_up(&fs_info->balance_wait_q); 4376 4377 return ret; 4378 out: 4379 if (bctl->flags & BTRFS_BALANCE_RESUME) 4380 reset_balance_state(fs_info); 4381 else 4382 kfree(bctl); 4383 btrfs_exclop_finish(fs_info); 4384 4385 return ret; 4386 } 4387 4388 static int balance_kthread(void *data) 4389 { 4390 struct btrfs_fs_info *fs_info = data; 4391 int ret = 0; 4392 4393 mutex_lock(&fs_info->balance_mutex); 4394 if (fs_info->balance_ctl) 4395 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4396 mutex_unlock(&fs_info->balance_mutex); 4397 4398 return ret; 4399 } 4400 4401 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4402 { 4403 struct task_struct *tsk; 4404 4405 mutex_lock(&fs_info->balance_mutex); 4406 if (!fs_info->balance_ctl) { 4407 mutex_unlock(&fs_info->balance_mutex); 4408 return 0; 4409 } 4410 mutex_unlock(&fs_info->balance_mutex); 4411 4412 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4413 btrfs_info(fs_info, "balance: resume skipped"); 4414 return 0; 4415 } 4416 4417 /* 4418 * A ro->rw remount sequence should continue with the paused balance 4419 * regardless of who pauses it, system or the user as of now, so set 4420 * the resume flag. 4421 */ 4422 spin_lock(&fs_info->balance_lock); 4423 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4424 spin_unlock(&fs_info->balance_lock); 4425 4426 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4427 return PTR_ERR_OR_ZERO(tsk); 4428 } 4429 4430 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4431 { 4432 struct btrfs_balance_control *bctl; 4433 struct btrfs_balance_item *item; 4434 struct btrfs_disk_balance_args disk_bargs; 4435 struct btrfs_path *path; 4436 struct extent_buffer *leaf; 4437 struct btrfs_key key; 4438 int ret; 4439 4440 path = btrfs_alloc_path(); 4441 if (!path) 4442 return -ENOMEM; 4443 4444 key.objectid = BTRFS_BALANCE_OBJECTID; 4445 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4446 key.offset = 0; 4447 4448 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4449 if (ret < 0) 4450 goto out; 4451 if (ret > 0) { /* ret = -ENOENT; */ 4452 ret = 0; 4453 goto out; 4454 } 4455 4456 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4457 if (!bctl) { 4458 ret = -ENOMEM; 4459 goto out; 4460 } 4461 4462 leaf = path->nodes[0]; 4463 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4464 4465 bctl->flags = btrfs_balance_flags(leaf, item); 4466 bctl->flags |= BTRFS_BALANCE_RESUME; 4467 4468 btrfs_balance_data(leaf, item, &disk_bargs); 4469 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4470 btrfs_balance_meta(leaf, item, &disk_bargs); 4471 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4472 btrfs_balance_sys(leaf, item, &disk_bargs); 4473 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4474 4475 /* 4476 * This should never happen, as the paused balance state is recovered 4477 * during mount without any chance of other exclusive ops to collide. 4478 * 4479 * This gives the exclusive op status to balance and keeps in paused 4480 * state until user intervention (cancel or umount). If the ownership 4481 * cannot be assigned, show a message but do not fail. The balance 4482 * is in a paused state and must have fs_info::balance_ctl properly 4483 * set up. 4484 */ 4485 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) 4486 btrfs_warn(fs_info, 4487 "balance: cannot set exclusive op status, resume manually"); 4488 4489 btrfs_release_path(path); 4490 4491 mutex_lock(&fs_info->balance_mutex); 4492 BUG_ON(fs_info->balance_ctl); 4493 spin_lock(&fs_info->balance_lock); 4494 fs_info->balance_ctl = bctl; 4495 spin_unlock(&fs_info->balance_lock); 4496 mutex_unlock(&fs_info->balance_mutex); 4497 out: 4498 btrfs_free_path(path); 4499 return ret; 4500 } 4501 4502 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4503 { 4504 int ret = 0; 4505 4506 mutex_lock(&fs_info->balance_mutex); 4507 if (!fs_info->balance_ctl) { 4508 mutex_unlock(&fs_info->balance_mutex); 4509 return -ENOTCONN; 4510 } 4511 4512 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4513 atomic_inc(&fs_info->balance_pause_req); 4514 mutex_unlock(&fs_info->balance_mutex); 4515 4516 wait_event(fs_info->balance_wait_q, 4517 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4518 4519 mutex_lock(&fs_info->balance_mutex); 4520 /* we are good with balance_ctl ripped off from under us */ 4521 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4522 atomic_dec(&fs_info->balance_pause_req); 4523 } else { 4524 ret = -ENOTCONN; 4525 } 4526 4527 mutex_unlock(&fs_info->balance_mutex); 4528 return ret; 4529 } 4530 4531 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4532 { 4533 mutex_lock(&fs_info->balance_mutex); 4534 if (!fs_info->balance_ctl) { 4535 mutex_unlock(&fs_info->balance_mutex); 4536 return -ENOTCONN; 4537 } 4538 4539 /* 4540 * A paused balance with the item stored on disk can be resumed at 4541 * mount time if the mount is read-write. Otherwise it's still paused 4542 * and we must not allow cancelling as it deletes the item. 4543 */ 4544 if (sb_rdonly(fs_info->sb)) { 4545 mutex_unlock(&fs_info->balance_mutex); 4546 return -EROFS; 4547 } 4548 4549 atomic_inc(&fs_info->balance_cancel_req); 4550 /* 4551 * if we are running just wait and return, balance item is 4552 * deleted in btrfs_balance in this case 4553 */ 4554 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4555 mutex_unlock(&fs_info->balance_mutex); 4556 wait_event(fs_info->balance_wait_q, 4557 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4558 mutex_lock(&fs_info->balance_mutex); 4559 } else { 4560 mutex_unlock(&fs_info->balance_mutex); 4561 /* 4562 * Lock released to allow other waiters to continue, we'll 4563 * reexamine the status again. 4564 */ 4565 mutex_lock(&fs_info->balance_mutex); 4566 4567 if (fs_info->balance_ctl) { 4568 reset_balance_state(fs_info); 4569 btrfs_exclop_finish(fs_info); 4570 btrfs_info(fs_info, "balance: canceled"); 4571 } 4572 } 4573 4574 BUG_ON(fs_info->balance_ctl || 4575 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4576 atomic_dec(&fs_info->balance_cancel_req); 4577 mutex_unlock(&fs_info->balance_mutex); 4578 return 0; 4579 } 4580 4581 int btrfs_uuid_scan_kthread(void *data) 4582 { 4583 struct btrfs_fs_info *fs_info = data; 4584 struct btrfs_root *root = fs_info->tree_root; 4585 struct btrfs_key key; 4586 struct btrfs_path *path = NULL; 4587 int ret = 0; 4588 struct extent_buffer *eb; 4589 int slot; 4590 struct btrfs_root_item root_item; 4591 u32 item_size; 4592 struct btrfs_trans_handle *trans = NULL; 4593 bool closing = false; 4594 4595 path = btrfs_alloc_path(); 4596 if (!path) { 4597 ret = -ENOMEM; 4598 goto out; 4599 } 4600 4601 key.objectid = 0; 4602 key.type = BTRFS_ROOT_ITEM_KEY; 4603 key.offset = 0; 4604 4605 while (1) { 4606 if (btrfs_fs_closing(fs_info)) { 4607 closing = true; 4608 break; 4609 } 4610 ret = btrfs_search_forward(root, &key, path, 4611 BTRFS_OLDEST_GENERATION); 4612 if (ret) { 4613 if (ret > 0) 4614 ret = 0; 4615 break; 4616 } 4617 4618 if (key.type != BTRFS_ROOT_ITEM_KEY || 4619 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4620 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4621 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4622 goto skip; 4623 4624 eb = path->nodes[0]; 4625 slot = path->slots[0]; 4626 item_size = btrfs_item_size_nr(eb, slot); 4627 if (item_size < sizeof(root_item)) 4628 goto skip; 4629 4630 read_extent_buffer(eb, &root_item, 4631 btrfs_item_ptr_offset(eb, slot), 4632 (int)sizeof(root_item)); 4633 if (btrfs_root_refs(&root_item) == 0) 4634 goto skip; 4635 4636 if (!btrfs_is_empty_uuid(root_item.uuid) || 4637 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4638 if (trans) 4639 goto update_tree; 4640 4641 btrfs_release_path(path); 4642 /* 4643 * 1 - subvol uuid item 4644 * 1 - received_subvol uuid item 4645 */ 4646 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4647 if (IS_ERR(trans)) { 4648 ret = PTR_ERR(trans); 4649 break; 4650 } 4651 continue; 4652 } else { 4653 goto skip; 4654 } 4655 update_tree: 4656 btrfs_release_path(path); 4657 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4658 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4659 BTRFS_UUID_KEY_SUBVOL, 4660 key.objectid); 4661 if (ret < 0) { 4662 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4663 ret); 4664 break; 4665 } 4666 } 4667 4668 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4669 ret = btrfs_uuid_tree_add(trans, 4670 root_item.received_uuid, 4671 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4672 key.objectid); 4673 if (ret < 0) { 4674 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4675 ret); 4676 break; 4677 } 4678 } 4679 4680 skip: 4681 btrfs_release_path(path); 4682 if (trans) { 4683 ret = btrfs_end_transaction(trans); 4684 trans = NULL; 4685 if (ret) 4686 break; 4687 } 4688 4689 if (key.offset < (u64)-1) { 4690 key.offset++; 4691 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4692 key.offset = 0; 4693 key.type = BTRFS_ROOT_ITEM_KEY; 4694 } else if (key.objectid < (u64)-1) { 4695 key.offset = 0; 4696 key.type = BTRFS_ROOT_ITEM_KEY; 4697 key.objectid++; 4698 } else { 4699 break; 4700 } 4701 cond_resched(); 4702 } 4703 4704 out: 4705 btrfs_free_path(path); 4706 if (trans && !IS_ERR(trans)) 4707 btrfs_end_transaction(trans); 4708 if (ret) 4709 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4710 else if (!closing) 4711 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4712 up(&fs_info->uuid_tree_rescan_sem); 4713 return 0; 4714 } 4715 4716 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4717 { 4718 struct btrfs_trans_handle *trans; 4719 struct btrfs_root *tree_root = fs_info->tree_root; 4720 struct btrfs_root *uuid_root; 4721 struct task_struct *task; 4722 int ret; 4723 4724 /* 4725 * 1 - root node 4726 * 1 - root item 4727 */ 4728 trans = btrfs_start_transaction(tree_root, 2); 4729 if (IS_ERR(trans)) 4730 return PTR_ERR(trans); 4731 4732 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4733 if (IS_ERR(uuid_root)) { 4734 ret = PTR_ERR(uuid_root); 4735 btrfs_abort_transaction(trans, ret); 4736 btrfs_end_transaction(trans); 4737 return ret; 4738 } 4739 4740 fs_info->uuid_root = uuid_root; 4741 4742 ret = btrfs_commit_transaction(trans); 4743 if (ret) 4744 return ret; 4745 4746 down(&fs_info->uuid_tree_rescan_sem); 4747 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4748 if (IS_ERR(task)) { 4749 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4750 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4751 up(&fs_info->uuid_tree_rescan_sem); 4752 return PTR_ERR(task); 4753 } 4754 4755 return 0; 4756 } 4757 4758 /* 4759 * shrinking a device means finding all of the device extents past 4760 * the new size, and then following the back refs to the chunks. 4761 * The chunk relocation code actually frees the device extent 4762 */ 4763 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4764 { 4765 struct btrfs_fs_info *fs_info = device->fs_info; 4766 struct btrfs_root *root = fs_info->dev_root; 4767 struct btrfs_trans_handle *trans; 4768 struct btrfs_dev_extent *dev_extent = NULL; 4769 struct btrfs_path *path; 4770 u64 length; 4771 u64 chunk_offset; 4772 int ret; 4773 int slot; 4774 int failed = 0; 4775 bool retried = false; 4776 struct extent_buffer *l; 4777 struct btrfs_key key; 4778 struct btrfs_super_block *super_copy = fs_info->super_copy; 4779 u64 old_total = btrfs_super_total_bytes(super_copy); 4780 u64 old_size = btrfs_device_get_total_bytes(device); 4781 u64 diff; 4782 u64 start; 4783 4784 new_size = round_down(new_size, fs_info->sectorsize); 4785 start = new_size; 4786 diff = round_down(old_size - new_size, fs_info->sectorsize); 4787 4788 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4789 return -EINVAL; 4790 4791 path = btrfs_alloc_path(); 4792 if (!path) 4793 return -ENOMEM; 4794 4795 path->reada = READA_BACK; 4796 4797 trans = btrfs_start_transaction(root, 0); 4798 if (IS_ERR(trans)) { 4799 btrfs_free_path(path); 4800 return PTR_ERR(trans); 4801 } 4802 4803 mutex_lock(&fs_info->chunk_mutex); 4804 4805 btrfs_device_set_total_bytes(device, new_size); 4806 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4807 device->fs_devices->total_rw_bytes -= diff; 4808 atomic64_sub(diff, &fs_info->free_chunk_space); 4809 } 4810 4811 /* 4812 * Once the device's size has been set to the new size, ensure all 4813 * in-memory chunks are synced to disk so that the loop below sees them 4814 * and relocates them accordingly. 4815 */ 4816 if (contains_pending_extent(device, &start, diff)) { 4817 mutex_unlock(&fs_info->chunk_mutex); 4818 ret = btrfs_commit_transaction(trans); 4819 if (ret) 4820 goto done; 4821 } else { 4822 mutex_unlock(&fs_info->chunk_mutex); 4823 btrfs_end_transaction(trans); 4824 } 4825 4826 again: 4827 key.objectid = device->devid; 4828 key.offset = (u64)-1; 4829 key.type = BTRFS_DEV_EXTENT_KEY; 4830 4831 do { 4832 mutex_lock(&fs_info->reclaim_bgs_lock); 4833 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4834 if (ret < 0) { 4835 mutex_unlock(&fs_info->reclaim_bgs_lock); 4836 goto done; 4837 } 4838 4839 ret = btrfs_previous_item(root, path, 0, key.type); 4840 if (ret) { 4841 mutex_unlock(&fs_info->reclaim_bgs_lock); 4842 if (ret < 0) 4843 goto done; 4844 ret = 0; 4845 btrfs_release_path(path); 4846 break; 4847 } 4848 4849 l = path->nodes[0]; 4850 slot = path->slots[0]; 4851 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4852 4853 if (key.objectid != device->devid) { 4854 mutex_unlock(&fs_info->reclaim_bgs_lock); 4855 btrfs_release_path(path); 4856 break; 4857 } 4858 4859 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4860 length = btrfs_dev_extent_length(l, dev_extent); 4861 4862 if (key.offset + length <= new_size) { 4863 mutex_unlock(&fs_info->reclaim_bgs_lock); 4864 btrfs_release_path(path); 4865 break; 4866 } 4867 4868 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4869 btrfs_release_path(path); 4870 4871 /* 4872 * We may be relocating the only data chunk we have, 4873 * which could potentially end up with losing data's 4874 * raid profile, so lets allocate an empty one in 4875 * advance. 4876 */ 4877 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4878 if (ret < 0) { 4879 mutex_unlock(&fs_info->reclaim_bgs_lock); 4880 goto done; 4881 } 4882 4883 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4884 mutex_unlock(&fs_info->reclaim_bgs_lock); 4885 if (ret == -ENOSPC) { 4886 failed++; 4887 } else if (ret) { 4888 if (ret == -ETXTBSY) { 4889 btrfs_warn(fs_info, 4890 "could not shrink block group %llu due to active swapfile", 4891 chunk_offset); 4892 } 4893 goto done; 4894 } 4895 } while (key.offset-- > 0); 4896 4897 if (failed && !retried) { 4898 failed = 0; 4899 retried = true; 4900 goto again; 4901 } else if (failed && retried) { 4902 ret = -ENOSPC; 4903 goto done; 4904 } 4905 4906 /* Shrinking succeeded, else we would be at "done". */ 4907 trans = btrfs_start_transaction(root, 0); 4908 if (IS_ERR(trans)) { 4909 ret = PTR_ERR(trans); 4910 goto done; 4911 } 4912 4913 mutex_lock(&fs_info->chunk_mutex); 4914 /* Clear all state bits beyond the shrunk device size */ 4915 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4916 CHUNK_STATE_MASK); 4917 4918 btrfs_device_set_disk_total_bytes(device, new_size); 4919 if (list_empty(&device->post_commit_list)) 4920 list_add_tail(&device->post_commit_list, 4921 &trans->transaction->dev_update_list); 4922 4923 WARN_ON(diff > old_total); 4924 btrfs_set_super_total_bytes(super_copy, 4925 round_down(old_total - diff, fs_info->sectorsize)); 4926 mutex_unlock(&fs_info->chunk_mutex); 4927 4928 /* Now btrfs_update_device() will change the on-disk size. */ 4929 ret = btrfs_update_device(trans, device); 4930 if (ret < 0) { 4931 btrfs_abort_transaction(trans, ret); 4932 btrfs_end_transaction(trans); 4933 } else { 4934 ret = btrfs_commit_transaction(trans); 4935 } 4936 done: 4937 btrfs_free_path(path); 4938 if (ret) { 4939 mutex_lock(&fs_info->chunk_mutex); 4940 btrfs_device_set_total_bytes(device, old_size); 4941 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 4942 device->fs_devices->total_rw_bytes += diff; 4943 atomic64_add(diff, &fs_info->free_chunk_space); 4944 mutex_unlock(&fs_info->chunk_mutex); 4945 } 4946 return ret; 4947 } 4948 4949 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 4950 struct btrfs_key *key, 4951 struct btrfs_chunk *chunk, int item_size) 4952 { 4953 struct btrfs_super_block *super_copy = fs_info->super_copy; 4954 struct btrfs_disk_key disk_key; 4955 u32 array_size; 4956 u8 *ptr; 4957 4958 lockdep_assert_held(&fs_info->chunk_mutex); 4959 4960 array_size = btrfs_super_sys_array_size(super_copy); 4961 if (array_size + item_size + sizeof(disk_key) 4962 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 4963 return -EFBIG; 4964 4965 ptr = super_copy->sys_chunk_array + array_size; 4966 btrfs_cpu_key_to_disk(&disk_key, key); 4967 memcpy(ptr, &disk_key, sizeof(disk_key)); 4968 ptr += sizeof(disk_key); 4969 memcpy(ptr, chunk, item_size); 4970 item_size += sizeof(disk_key); 4971 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 4972 4973 return 0; 4974 } 4975 4976 /* 4977 * sort the devices in descending order by max_avail, total_avail 4978 */ 4979 static int btrfs_cmp_device_info(const void *a, const void *b) 4980 { 4981 const struct btrfs_device_info *di_a = a; 4982 const struct btrfs_device_info *di_b = b; 4983 4984 if (di_a->max_avail > di_b->max_avail) 4985 return -1; 4986 if (di_a->max_avail < di_b->max_avail) 4987 return 1; 4988 if (di_a->total_avail > di_b->total_avail) 4989 return -1; 4990 if (di_a->total_avail < di_b->total_avail) 4991 return 1; 4992 return 0; 4993 } 4994 4995 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 4996 { 4997 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 4998 return; 4999 5000 btrfs_set_fs_incompat(info, RAID56); 5001 } 5002 5003 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5004 { 5005 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5006 return; 5007 5008 btrfs_set_fs_incompat(info, RAID1C34); 5009 } 5010 5011 /* 5012 * Structure used internally for btrfs_create_chunk() function. 5013 * Wraps needed parameters. 5014 */ 5015 struct alloc_chunk_ctl { 5016 u64 start; 5017 u64 type; 5018 /* Total number of stripes to allocate */ 5019 int num_stripes; 5020 /* sub_stripes info for map */ 5021 int sub_stripes; 5022 /* Stripes per device */ 5023 int dev_stripes; 5024 /* Maximum number of devices to use */ 5025 int devs_max; 5026 /* Minimum number of devices to use */ 5027 int devs_min; 5028 /* ndevs has to be a multiple of this */ 5029 int devs_increment; 5030 /* Number of copies */ 5031 int ncopies; 5032 /* Number of stripes worth of bytes to store parity information */ 5033 int nparity; 5034 u64 max_stripe_size; 5035 u64 max_chunk_size; 5036 u64 dev_extent_min; 5037 u64 stripe_size; 5038 u64 chunk_size; 5039 int ndevs; 5040 }; 5041 5042 static void init_alloc_chunk_ctl_policy_regular( 5043 struct btrfs_fs_devices *fs_devices, 5044 struct alloc_chunk_ctl *ctl) 5045 { 5046 u64 type = ctl->type; 5047 5048 if (type & BTRFS_BLOCK_GROUP_DATA) { 5049 ctl->max_stripe_size = SZ_1G; 5050 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; 5051 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5052 /* For larger filesystems, use larger metadata chunks */ 5053 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) 5054 ctl->max_stripe_size = SZ_1G; 5055 else 5056 ctl->max_stripe_size = SZ_256M; 5057 ctl->max_chunk_size = ctl->max_stripe_size; 5058 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5059 ctl->max_stripe_size = SZ_32M; 5060 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5061 ctl->devs_max = min_t(int, ctl->devs_max, 5062 BTRFS_MAX_DEVS_SYS_CHUNK); 5063 } else { 5064 BUG(); 5065 } 5066 5067 /* We don't want a chunk larger than 10% of writable space */ 5068 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5069 ctl->max_chunk_size); 5070 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5071 } 5072 5073 static void init_alloc_chunk_ctl_policy_zoned( 5074 struct btrfs_fs_devices *fs_devices, 5075 struct alloc_chunk_ctl *ctl) 5076 { 5077 u64 zone_size = fs_devices->fs_info->zone_size; 5078 u64 limit; 5079 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5080 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5081 u64 min_chunk_size = min_data_stripes * zone_size; 5082 u64 type = ctl->type; 5083 5084 ctl->max_stripe_size = zone_size; 5085 if (type & BTRFS_BLOCK_GROUP_DATA) { 5086 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5087 zone_size); 5088 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5089 ctl->max_chunk_size = ctl->max_stripe_size; 5090 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5091 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5092 ctl->devs_max = min_t(int, ctl->devs_max, 5093 BTRFS_MAX_DEVS_SYS_CHUNK); 5094 } else { 5095 BUG(); 5096 } 5097 5098 /* We don't want a chunk larger than 10% of writable space */ 5099 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5100 zone_size), 5101 min_chunk_size); 5102 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5103 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5104 } 5105 5106 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5107 struct alloc_chunk_ctl *ctl) 5108 { 5109 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5110 5111 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5112 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5113 ctl->devs_max = btrfs_raid_array[index].devs_max; 5114 if (!ctl->devs_max) 5115 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5116 ctl->devs_min = btrfs_raid_array[index].devs_min; 5117 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5118 ctl->ncopies = btrfs_raid_array[index].ncopies; 5119 ctl->nparity = btrfs_raid_array[index].nparity; 5120 ctl->ndevs = 0; 5121 5122 switch (fs_devices->chunk_alloc_policy) { 5123 case BTRFS_CHUNK_ALLOC_REGULAR: 5124 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5125 break; 5126 case BTRFS_CHUNK_ALLOC_ZONED: 5127 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5128 break; 5129 default: 5130 BUG(); 5131 } 5132 } 5133 5134 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5135 struct alloc_chunk_ctl *ctl, 5136 struct btrfs_device_info *devices_info) 5137 { 5138 struct btrfs_fs_info *info = fs_devices->fs_info; 5139 struct btrfs_device *device; 5140 u64 total_avail; 5141 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5142 int ret; 5143 int ndevs = 0; 5144 u64 max_avail; 5145 u64 dev_offset; 5146 5147 /* 5148 * in the first pass through the devices list, we gather information 5149 * about the available holes on each device. 5150 */ 5151 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5152 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5153 WARN(1, KERN_ERR 5154 "BTRFS: read-only device in alloc_list\n"); 5155 continue; 5156 } 5157 5158 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5159 &device->dev_state) || 5160 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5161 continue; 5162 5163 if (device->total_bytes > device->bytes_used) 5164 total_avail = device->total_bytes - device->bytes_used; 5165 else 5166 total_avail = 0; 5167 5168 /* If there is no space on this device, skip it. */ 5169 if (total_avail < ctl->dev_extent_min) 5170 continue; 5171 5172 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5173 &max_avail); 5174 if (ret && ret != -ENOSPC) 5175 return ret; 5176 5177 if (ret == 0) 5178 max_avail = dev_extent_want; 5179 5180 if (max_avail < ctl->dev_extent_min) { 5181 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5182 btrfs_debug(info, 5183 "%s: devid %llu has no free space, have=%llu want=%llu", 5184 __func__, device->devid, max_avail, 5185 ctl->dev_extent_min); 5186 continue; 5187 } 5188 5189 if (ndevs == fs_devices->rw_devices) { 5190 WARN(1, "%s: found more than %llu devices\n", 5191 __func__, fs_devices->rw_devices); 5192 break; 5193 } 5194 devices_info[ndevs].dev_offset = dev_offset; 5195 devices_info[ndevs].max_avail = max_avail; 5196 devices_info[ndevs].total_avail = total_avail; 5197 devices_info[ndevs].dev = device; 5198 ++ndevs; 5199 } 5200 ctl->ndevs = ndevs; 5201 5202 /* 5203 * now sort the devices by hole size / available space 5204 */ 5205 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5206 btrfs_cmp_device_info, NULL); 5207 5208 return 0; 5209 } 5210 5211 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5212 struct btrfs_device_info *devices_info) 5213 { 5214 /* Number of stripes that count for block group size */ 5215 int data_stripes; 5216 5217 /* 5218 * The primary goal is to maximize the number of stripes, so use as 5219 * many devices as possible, even if the stripes are not maximum sized. 5220 * 5221 * The DUP profile stores more than one stripe per device, the 5222 * max_avail is the total size so we have to adjust. 5223 */ 5224 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5225 ctl->dev_stripes); 5226 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5227 5228 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5229 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5230 5231 /* 5232 * Use the number of data stripes to figure out how big this chunk is 5233 * really going to be in terms of logical address space, and compare 5234 * that answer with the max chunk size. If it's higher, we try to 5235 * reduce stripe_size. 5236 */ 5237 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5238 /* 5239 * Reduce stripe_size, round it up to a 16MB boundary again and 5240 * then use it, unless it ends up being even bigger than the 5241 * previous value we had already. 5242 */ 5243 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5244 data_stripes), SZ_16M), 5245 ctl->stripe_size); 5246 } 5247 5248 /* Align to BTRFS_STRIPE_LEN */ 5249 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5250 ctl->chunk_size = ctl->stripe_size * data_stripes; 5251 5252 return 0; 5253 } 5254 5255 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5256 struct btrfs_device_info *devices_info) 5257 { 5258 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5259 /* Number of stripes that count for block group size */ 5260 int data_stripes; 5261 5262 /* 5263 * It should hold because: 5264 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5265 */ 5266 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5267 5268 ctl->stripe_size = zone_size; 5269 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5270 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5271 5272 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5273 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5274 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5275 ctl->stripe_size) + ctl->nparity, 5276 ctl->dev_stripes); 5277 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5278 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5279 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5280 } 5281 5282 ctl->chunk_size = ctl->stripe_size * data_stripes; 5283 5284 return 0; 5285 } 5286 5287 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5288 struct alloc_chunk_ctl *ctl, 5289 struct btrfs_device_info *devices_info) 5290 { 5291 struct btrfs_fs_info *info = fs_devices->fs_info; 5292 5293 /* 5294 * Round down to number of usable stripes, devs_increment can be any 5295 * number so we can't use round_down() that requires power of 2, while 5296 * rounddown is safe. 5297 */ 5298 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5299 5300 if (ctl->ndevs < ctl->devs_min) { 5301 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5302 btrfs_debug(info, 5303 "%s: not enough devices with free space: have=%d minimum required=%d", 5304 __func__, ctl->ndevs, ctl->devs_min); 5305 } 5306 return -ENOSPC; 5307 } 5308 5309 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5310 5311 switch (fs_devices->chunk_alloc_policy) { 5312 case BTRFS_CHUNK_ALLOC_REGULAR: 5313 return decide_stripe_size_regular(ctl, devices_info); 5314 case BTRFS_CHUNK_ALLOC_ZONED: 5315 return decide_stripe_size_zoned(ctl, devices_info); 5316 default: 5317 BUG(); 5318 } 5319 } 5320 5321 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5322 struct alloc_chunk_ctl *ctl, 5323 struct btrfs_device_info *devices_info) 5324 { 5325 struct btrfs_fs_info *info = trans->fs_info; 5326 struct map_lookup *map = NULL; 5327 struct extent_map_tree *em_tree; 5328 struct btrfs_block_group *block_group; 5329 struct extent_map *em; 5330 u64 start = ctl->start; 5331 u64 type = ctl->type; 5332 int ret; 5333 int i; 5334 int j; 5335 5336 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5337 if (!map) 5338 return ERR_PTR(-ENOMEM); 5339 map->num_stripes = ctl->num_stripes; 5340 5341 for (i = 0; i < ctl->ndevs; ++i) { 5342 for (j = 0; j < ctl->dev_stripes; ++j) { 5343 int s = i * ctl->dev_stripes + j; 5344 map->stripes[s].dev = devices_info[i].dev; 5345 map->stripes[s].physical = devices_info[i].dev_offset + 5346 j * ctl->stripe_size; 5347 } 5348 } 5349 map->stripe_len = BTRFS_STRIPE_LEN; 5350 map->io_align = BTRFS_STRIPE_LEN; 5351 map->io_width = BTRFS_STRIPE_LEN; 5352 map->type = type; 5353 map->sub_stripes = ctl->sub_stripes; 5354 5355 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5356 5357 em = alloc_extent_map(); 5358 if (!em) { 5359 kfree(map); 5360 return ERR_PTR(-ENOMEM); 5361 } 5362 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5363 em->map_lookup = map; 5364 em->start = start; 5365 em->len = ctl->chunk_size; 5366 em->block_start = 0; 5367 em->block_len = em->len; 5368 em->orig_block_len = ctl->stripe_size; 5369 5370 em_tree = &info->mapping_tree; 5371 write_lock(&em_tree->lock); 5372 ret = add_extent_mapping(em_tree, em, 0); 5373 if (ret) { 5374 write_unlock(&em_tree->lock); 5375 free_extent_map(em); 5376 return ERR_PTR(ret); 5377 } 5378 write_unlock(&em_tree->lock); 5379 5380 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5381 if (IS_ERR(block_group)) 5382 goto error_del_extent; 5383 5384 for (i = 0; i < map->num_stripes; i++) { 5385 struct btrfs_device *dev = map->stripes[i].dev; 5386 5387 btrfs_device_set_bytes_used(dev, 5388 dev->bytes_used + ctl->stripe_size); 5389 if (list_empty(&dev->post_commit_list)) 5390 list_add_tail(&dev->post_commit_list, 5391 &trans->transaction->dev_update_list); 5392 } 5393 5394 atomic64_sub(ctl->stripe_size * map->num_stripes, 5395 &info->free_chunk_space); 5396 5397 free_extent_map(em); 5398 check_raid56_incompat_flag(info, type); 5399 check_raid1c34_incompat_flag(info, type); 5400 5401 return block_group; 5402 5403 error_del_extent: 5404 write_lock(&em_tree->lock); 5405 remove_extent_mapping(em_tree, em); 5406 write_unlock(&em_tree->lock); 5407 5408 /* One for our allocation */ 5409 free_extent_map(em); 5410 /* One for the tree reference */ 5411 free_extent_map(em); 5412 5413 return block_group; 5414 } 5415 5416 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5417 u64 type) 5418 { 5419 struct btrfs_fs_info *info = trans->fs_info; 5420 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5421 struct btrfs_device_info *devices_info = NULL; 5422 struct alloc_chunk_ctl ctl; 5423 struct btrfs_block_group *block_group; 5424 int ret; 5425 5426 lockdep_assert_held(&info->chunk_mutex); 5427 5428 if (!alloc_profile_is_valid(type, 0)) { 5429 ASSERT(0); 5430 return ERR_PTR(-EINVAL); 5431 } 5432 5433 if (list_empty(&fs_devices->alloc_list)) { 5434 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5435 btrfs_debug(info, "%s: no writable device", __func__); 5436 return ERR_PTR(-ENOSPC); 5437 } 5438 5439 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5440 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5441 ASSERT(0); 5442 return ERR_PTR(-EINVAL); 5443 } 5444 5445 ctl.start = find_next_chunk(info); 5446 ctl.type = type; 5447 init_alloc_chunk_ctl(fs_devices, &ctl); 5448 5449 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5450 GFP_NOFS); 5451 if (!devices_info) 5452 return ERR_PTR(-ENOMEM); 5453 5454 ret = gather_device_info(fs_devices, &ctl, devices_info); 5455 if (ret < 0) { 5456 block_group = ERR_PTR(ret); 5457 goto out; 5458 } 5459 5460 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5461 if (ret < 0) { 5462 block_group = ERR_PTR(ret); 5463 goto out; 5464 } 5465 5466 block_group = create_chunk(trans, &ctl, devices_info); 5467 5468 out: 5469 kfree(devices_info); 5470 return block_group; 5471 } 5472 5473 /* 5474 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5475 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5476 * chunks. 5477 * 5478 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5479 * phases. 5480 */ 5481 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5482 struct btrfs_block_group *bg) 5483 { 5484 struct btrfs_fs_info *fs_info = trans->fs_info; 5485 struct btrfs_root *extent_root = fs_info->extent_root; 5486 struct btrfs_root *chunk_root = fs_info->chunk_root; 5487 struct btrfs_key key; 5488 struct btrfs_chunk *chunk; 5489 struct btrfs_stripe *stripe; 5490 struct extent_map *em; 5491 struct map_lookup *map; 5492 size_t item_size; 5493 int i; 5494 int ret; 5495 5496 /* 5497 * We take the chunk_mutex for 2 reasons: 5498 * 5499 * 1) Updates and insertions in the chunk btree must be done while holding 5500 * the chunk_mutex, as well as updating the system chunk array in the 5501 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5502 * details; 5503 * 5504 * 2) To prevent races with the final phase of a device replace operation 5505 * that replaces the device object associated with the map's stripes, 5506 * because the device object's id can change at any time during that 5507 * final phase of the device replace operation 5508 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5509 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5510 * which would cause a failure when updating the device item, which does 5511 * not exists, or persisting a stripe of the chunk item with such ID. 5512 * Here we can't use the device_list_mutex because our caller already 5513 * has locked the chunk_mutex, and the final phase of device replace 5514 * acquires both mutexes - first the device_list_mutex and then the 5515 * chunk_mutex. Using any of those two mutexes protects us from a 5516 * concurrent device replace. 5517 */ 5518 lockdep_assert_held(&fs_info->chunk_mutex); 5519 5520 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5521 if (IS_ERR(em)) { 5522 ret = PTR_ERR(em); 5523 btrfs_abort_transaction(trans, ret); 5524 return ret; 5525 } 5526 5527 map = em->map_lookup; 5528 item_size = btrfs_chunk_item_size(map->num_stripes); 5529 5530 chunk = kzalloc(item_size, GFP_NOFS); 5531 if (!chunk) { 5532 ret = -ENOMEM; 5533 btrfs_abort_transaction(trans, ret); 5534 goto out; 5535 } 5536 5537 for (i = 0; i < map->num_stripes; i++) { 5538 struct btrfs_device *device = map->stripes[i].dev; 5539 5540 ret = btrfs_update_device(trans, device); 5541 if (ret) 5542 goto out; 5543 } 5544 5545 stripe = &chunk->stripe; 5546 for (i = 0; i < map->num_stripes; i++) { 5547 struct btrfs_device *device = map->stripes[i].dev; 5548 const u64 dev_offset = map->stripes[i].physical; 5549 5550 btrfs_set_stack_stripe_devid(stripe, device->devid); 5551 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5552 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5553 stripe++; 5554 } 5555 5556 btrfs_set_stack_chunk_length(chunk, bg->length); 5557 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); 5558 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5559 btrfs_set_stack_chunk_type(chunk, map->type); 5560 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5561 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5562 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5563 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5564 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5565 5566 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5567 key.type = BTRFS_CHUNK_ITEM_KEY; 5568 key.offset = bg->start; 5569 5570 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5571 if (ret) 5572 goto out; 5573 5574 bg->chunk_item_inserted = 1; 5575 5576 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5577 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5578 if (ret) 5579 goto out; 5580 } 5581 5582 out: 5583 kfree(chunk); 5584 free_extent_map(em); 5585 return ret; 5586 } 5587 5588 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5589 { 5590 struct btrfs_fs_info *fs_info = trans->fs_info; 5591 u64 alloc_profile; 5592 struct btrfs_block_group *meta_bg; 5593 struct btrfs_block_group *sys_bg; 5594 5595 /* 5596 * When adding a new device for sprouting, the seed device is read-only 5597 * so we must first allocate a metadata and a system chunk. But before 5598 * adding the block group items to the extent, device and chunk btrees, 5599 * we must first: 5600 * 5601 * 1) Create both chunks without doing any changes to the btrees, as 5602 * otherwise we would get -ENOSPC since the block groups from the 5603 * seed device are read-only; 5604 * 5605 * 2) Add the device item for the new sprout device - finishing the setup 5606 * of a new block group requires updating the device item in the chunk 5607 * btree, so it must exist when we attempt to do it. The previous step 5608 * ensures this does not fail with -ENOSPC. 5609 * 5610 * After that we can add the block group items to their btrees: 5611 * update existing device item in the chunk btree, add a new block group 5612 * item to the extent btree, add a new chunk item to the chunk btree and 5613 * finally add the new device extent items to the devices btree. 5614 */ 5615 5616 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5617 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5618 if (IS_ERR(meta_bg)) 5619 return PTR_ERR(meta_bg); 5620 5621 alloc_profile = btrfs_system_alloc_profile(fs_info); 5622 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5623 if (IS_ERR(sys_bg)) 5624 return PTR_ERR(sys_bg); 5625 5626 return 0; 5627 } 5628 5629 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5630 { 5631 const int index = btrfs_bg_flags_to_raid_index(map->type); 5632 5633 return btrfs_raid_array[index].tolerated_failures; 5634 } 5635 5636 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5637 { 5638 struct extent_map *em; 5639 struct map_lookup *map; 5640 int miss_ndevs = 0; 5641 int i; 5642 bool ret = true; 5643 5644 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5645 if (IS_ERR(em)) 5646 return false; 5647 5648 map = em->map_lookup; 5649 for (i = 0; i < map->num_stripes; i++) { 5650 if (test_bit(BTRFS_DEV_STATE_MISSING, 5651 &map->stripes[i].dev->dev_state)) { 5652 miss_ndevs++; 5653 continue; 5654 } 5655 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5656 &map->stripes[i].dev->dev_state)) { 5657 ret = false; 5658 goto end; 5659 } 5660 } 5661 5662 /* 5663 * If the number of missing devices is larger than max errors, we can 5664 * not write the data into that chunk successfully. 5665 */ 5666 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5667 ret = false; 5668 end: 5669 free_extent_map(em); 5670 return ret; 5671 } 5672 5673 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5674 { 5675 struct extent_map *em; 5676 5677 while (1) { 5678 write_lock(&tree->lock); 5679 em = lookup_extent_mapping(tree, 0, (u64)-1); 5680 if (em) 5681 remove_extent_mapping(tree, em); 5682 write_unlock(&tree->lock); 5683 if (!em) 5684 break; 5685 /* once for us */ 5686 free_extent_map(em); 5687 /* once for the tree */ 5688 free_extent_map(em); 5689 } 5690 } 5691 5692 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5693 { 5694 struct extent_map *em; 5695 struct map_lookup *map; 5696 int ret; 5697 5698 em = btrfs_get_chunk_map(fs_info, logical, len); 5699 if (IS_ERR(em)) 5700 /* 5701 * We could return errors for these cases, but that could get 5702 * ugly and we'd probably do the same thing which is just not do 5703 * anything else and exit, so return 1 so the callers don't try 5704 * to use other copies. 5705 */ 5706 return 1; 5707 5708 map = em->map_lookup; 5709 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) 5710 ret = map->num_stripes; 5711 else if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5712 ret = map->sub_stripes; 5713 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5714 ret = 2; 5715 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5716 /* 5717 * There could be two corrupted data stripes, we need 5718 * to loop retry in order to rebuild the correct data. 5719 * 5720 * Fail a stripe at a time on every retry except the 5721 * stripe under reconstruction. 5722 */ 5723 ret = map->num_stripes; 5724 else 5725 ret = 1; 5726 free_extent_map(em); 5727 5728 down_read(&fs_info->dev_replace.rwsem); 5729 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5730 fs_info->dev_replace.tgtdev) 5731 ret++; 5732 up_read(&fs_info->dev_replace.rwsem); 5733 5734 return ret; 5735 } 5736 5737 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5738 u64 logical) 5739 { 5740 struct extent_map *em; 5741 struct map_lookup *map; 5742 unsigned long len = fs_info->sectorsize; 5743 5744 em = btrfs_get_chunk_map(fs_info, logical, len); 5745 5746 if (!WARN_ON(IS_ERR(em))) { 5747 map = em->map_lookup; 5748 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5749 len = map->stripe_len * nr_data_stripes(map); 5750 free_extent_map(em); 5751 } 5752 return len; 5753 } 5754 5755 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5756 { 5757 struct extent_map *em; 5758 struct map_lookup *map; 5759 int ret = 0; 5760 5761 em = btrfs_get_chunk_map(fs_info, logical, len); 5762 5763 if(!WARN_ON(IS_ERR(em))) { 5764 map = em->map_lookup; 5765 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5766 ret = 1; 5767 free_extent_map(em); 5768 } 5769 return ret; 5770 } 5771 5772 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5773 struct map_lookup *map, int first, 5774 int dev_replace_is_ongoing) 5775 { 5776 int i; 5777 int num_stripes; 5778 int preferred_mirror; 5779 int tolerance; 5780 struct btrfs_device *srcdev; 5781 5782 ASSERT((map->type & 5783 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5784 5785 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5786 num_stripes = map->sub_stripes; 5787 else 5788 num_stripes = map->num_stripes; 5789 5790 switch (fs_info->fs_devices->read_policy) { 5791 default: 5792 /* Shouldn't happen, just warn and use pid instead of failing */ 5793 btrfs_warn_rl(fs_info, 5794 "unknown read_policy type %u, reset to pid", 5795 fs_info->fs_devices->read_policy); 5796 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5797 fallthrough; 5798 case BTRFS_READ_POLICY_PID: 5799 preferred_mirror = first + (current->pid % num_stripes); 5800 break; 5801 } 5802 5803 if (dev_replace_is_ongoing && 5804 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5805 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5806 srcdev = fs_info->dev_replace.srcdev; 5807 else 5808 srcdev = NULL; 5809 5810 /* 5811 * try to avoid the drive that is the source drive for a 5812 * dev-replace procedure, only choose it if no other non-missing 5813 * mirror is available 5814 */ 5815 for (tolerance = 0; tolerance < 2; tolerance++) { 5816 if (map->stripes[preferred_mirror].dev->bdev && 5817 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5818 return preferred_mirror; 5819 for (i = first; i < first + num_stripes; i++) { 5820 if (map->stripes[i].dev->bdev && 5821 (tolerance || map->stripes[i].dev != srcdev)) 5822 return i; 5823 } 5824 } 5825 5826 /* we couldn't find one that doesn't fail. Just return something 5827 * and the io error handling code will clean up eventually 5828 */ 5829 return preferred_mirror; 5830 } 5831 5832 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5833 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5834 { 5835 int i; 5836 int again = 1; 5837 5838 while (again) { 5839 again = 0; 5840 for (i = 0; i < num_stripes - 1; i++) { 5841 /* Swap if parity is on a smaller index */ 5842 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5843 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5844 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5845 again = 1; 5846 } 5847 } 5848 } 5849 } 5850 5851 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5852 int total_stripes, 5853 int real_stripes) 5854 { 5855 struct btrfs_io_context *bioc = kzalloc( 5856 /* The size of btrfs_io_context */ 5857 sizeof(struct btrfs_io_context) + 5858 /* Plus the variable array for the stripes */ 5859 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5860 /* Plus the variable array for the tgt dev */ 5861 sizeof(int) * (real_stripes) + 5862 /* 5863 * Plus the raid_map, which includes both the tgt dev 5864 * and the stripes. 5865 */ 5866 sizeof(u64) * (total_stripes), 5867 GFP_NOFS|__GFP_NOFAIL); 5868 5869 atomic_set(&bioc->error, 0); 5870 refcount_set(&bioc->refs, 1); 5871 5872 bioc->fs_info = fs_info; 5873 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5874 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5875 5876 return bioc; 5877 } 5878 5879 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5880 { 5881 WARN_ON(!refcount_read(&bioc->refs)); 5882 refcount_inc(&bioc->refs); 5883 } 5884 5885 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5886 { 5887 if (!bioc) 5888 return; 5889 if (refcount_dec_and_test(&bioc->refs)) 5890 kfree(bioc); 5891 } 5892 5893 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ 5894 /* 5895 * Please note that, discard won't be sent to target device of device 5896 * replace. 5897 */ 5898 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, 5899 u64 logical, u64 *length_ret, 5900 struct btrfs_io_context **bioc_ret) 5901 { 5902 struct extent_map *em; 5903 struct map_lookup *map; 5904 struct btrfs_io_context *bioc; 5905 u64 length = *length_ret; 5906 u64 offset; 5907 u64 stripe_nr; 5908 u64 stripe_nr_end; 5909 u64 stripe_end_offset; 5910 u64 stripe_cnt; 5911 u64 stripe_len; 5912 u64 stripe_offset; 5913 u64 num_stripes; 5914 u32 stripe_index; 5915 u32 factor = 0; 5916 u32 sub_stripes = 0; 5917 u64 stripes_per_dev = 0; 5918 u32 remaining_stripes = 0; 5919 u32 last_stripe = 0; 5920 int ret = 0; 5921 int i; 5922 5923 /* Discard always returns a bioc. */ 5924 ASSERT(bioc_ret); 5925 5926 em = btrfs_get_chunk_map(fs_info, logical, length); 5927 if (IS_ERR(em)) 5928 return PTR_ERR(em); 5929 5930 map = em->map_lookup; 5931 /* we don't discard raid56 yet */ 5932 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5933 ret = -EOPNOTSUPP; 5934 goto out; 5935 } 5936 5937 offset = logical - em->start; 5938 length = min_t(u64, em->start + em->len - logical, length); 5939 *length_ret = length; 5940 5941 stripe_len = map->stripe_len; 5942 /* 5943 * stripe_nr counts the total number of stripes we have to stride 5944 * to get to this block 5945 */ 5946 stripe_nr = div64_u64(offset, stripe_len); 5947 5948 /* stripe_offset is the offset of this block in its stripe */ 5949 stripe_offset = offset - stripe_nr * stripe_len; 5950 5951 stripe_nr_end = round_up(offset + length, map->stripe_len); 5952 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); 5953 stripe_cnt = stripe_nr_end - stripe_nr; 5954 stripe_end_offset = stripe_nr_end * map->stripe_len - 5955 (offset + length); 5956 /* 5957 * after this, stripe_nr is the number of stripes on this 5958 * device we have to walk to find the data, and stripe_index is 5959 * the number of our device in the stripe array 5960 */ 5961 num_stripes = 1; 5962 stripe_index = 0; 5963 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5964 BTRFS_BLOCK_GROUP_RAID10)) { 5965 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 5966 sub_stripes = 1; 5967 else 5968 sub_stripes = map->sub_stripes; 5969 5970 factor = map->num_stripes / sub_stripes; 5971 num_stripes = min_t(u64, map->num_stripes, 5972 sub_stripes * stripe_cnt); 5973 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 5974 stripe_index *= sub_stripes; 5975 stripes_per_dev = div_u64_rem(stripe_cnt, factor, 5976 &remaining_stripes); 5977 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); 5978 last_stripe *= sub_stripes; 5979 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 5980 BTRFS_BLOCK_GROUP_DUP)) { 5981 num_stripes = map->num_stripes; 5982 } else { 5983 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 5984 &stripe_index); 5985 } 5986 5987 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0); 5988 if (!bioc) { 5989 ret = -ENOMEM; 5990 goto out; 5991 } 5992 5993 for (i = 0; i < num_stripes; i++) { 5994 bioc->stripes[i].physical = 5995 map->stripes[stripe_index].physical + 5996 stripe_offset + stripe_nr * map->stripe_len; 5997 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 5998 5999 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6000 BTRFS_BLOCK_GROUP_RAID10)) { 6001 bioc->stripes[i].length = stripes_per_dev * 6002 map->stripe_len; 6003 6004 if (i / sub_stripes < remaining_stripes) 6005 bioc->stripes[i].length += map->stripe_len; 6006 6007 /* 6008 * Special for the first stripe and 6009 * the last stripe: 6010 * 6011 * |-------|...|-------| 6012 * |----------| 6013 * off end_off 6014 */ 6015 if (i < sub_stripes) 6016 bioc->stripes[i].length -= stripe_offset; 6017 6018 if (stripe_index >= last_stripe && 6019 stripe_index <= (last_stripe + 6020 sub_stripes - 1)) 6021 bioc->stripes[i].length -= stripe_end_offset; 6022 6023 if (i == sub_stripes - 1) 6024 stripe_offset = 0; 6025 } else { 6026 bioc->stripes[i].length = length; 6027 } 6028 6029 stripe_index++; 6030 if (stripe_index == map->num_stripes) { 6031 stripe_index = 0; 6032 stripe_nr++; 6033 } 6034 } 6035 6036 *bioc_ret = bioc; 6037 bioc->map_type = map->type; 6038 bioc->num_stripes = num_stripes; 6039 out: 6040 free_extent_map(em); 6041 return ret; 6042 } 6043 6044 /* 6045 * In dev-replace case, for repair case (that's the only case where the mirror 6046 * is selected explicitly when calling btrfs_map_block), blocks left of the 6047 * left cursor can also be read from the target drive. 6048 * 6049 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6050 * array of stripes. 6051 * For READ, it also needs to be supported using the same mirror number. 6052 * 6053 * If the requested block is not left of the left cursor, EIO is returned. This 6054 * can happen because btrfs_num_copies() returns one more in the dev-replace 6055 * case. 6056 */ 6057 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6058 u64 logical, u64 length, 6059 u64 srcdev_devid, int *mirror_num, 6060 u64 *physical) 6061 { 6062 struct btrfs_io_context *bioc = NULL; 6063 int num_stripes; 6064 int index_srcdev = 0; 6065 int found = 0; 6066 u64 physical_of_found = 0; 6067 int i; 6068 int ret = 0; 6069 6070 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6071 logical, &length, &bioc, 0, 0); 6072 if (ret) { 6073 ASSERT(bioc == NULL); 6074 return ret; 6075 } 6076 6077 num_stripes = bioc->num_stripes; 6078 if (*mirror_num > num_stripes) { 6079 /* 6080 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6081 * that means that the requested area is not left of the left 6082 * cursor 6083 */ 6084 btrfs_put_bioc(bioc); 6085 return -EIO; 6086 } 6087 6088 /* 6089 * process the rest of the function using the mirror_num of the source 6090 * drive. Therefore look it up first. At the end, patch the device 6091 * pointer to the one of the target drive. 6092 */ 6093 for (i = 0; i < num_stripes; i++) { 6094 if (bioc->stripes[i].dev->devid != srcdev_devid) 6095 continue; 6096 6097 /* 6098 * In case of DUP, in order to keep it simple, only add the 6099 * mirror with the lowest physical address 6100 */ 6101 if (found && 6102 physical_of_found <= bioc->stripes[i].physical) 6103 continue; 6104 6105 index_srcdev = i; 6106 found = 1; 6107 physical_of_found = bioc->stripes[i].physical; 6108 } 6109 6110 btrfs_put_bioc(bioc); 6111 6112 ASSERT(found); 6113 if (!found) 6114 return -EIO; 6115 6116 *mirror_num = index_srcdev + 1; 6117 *physical = physical_of_found; 6118 return ret; 6119 } 6120 6121 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6122 { 6123 struct btrfs_block_group *cache; 6124 bool ret; 6125 6126 /* Non zoned filesystem does not use "to_copy" flag */ 6127 if (!btrfs_is_zoned(fs_info)) 6128 return false; 6129 6130 cache = btrfs_lookup_block_group(fs_info, logical); 6131 6132 spin_lock(&cache->lock); 6133 ret = cache->to_copy; 6134 spin_unlock(&cache->lock); 6135 6136 btrfs_put_block_group(cache); 6137 return ret; 6138 } 6139 6140 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6141 struct btrfs_io_context **bioc_ret, 6142 struct btrfs_dev_replace *dev_replace, 6143 u64 logical, 6144 int *num_stripes_ret, int *max_errors_ret) 6145 { 6146 struct btrfs_io_context *bioc = *bioc_ret; 6147 u64 srcdev_devid = dev_replace->srcdev->devid; 6148 int tgtdev_indexes = 0; 6149 int num_stripes = *num_stripes_ret; 6150 int max_errors = *max_errors_ret; 6151 int i; 6152 6153 if (op == BTRFS_MAP_WRITE) { 6154 int index_where_to_add; 6155 6156 /* 6157 * A block group which have "to_copy" set will eventually 6158 * copied by dev-replace process. We can avoid cloning IO here. 6159 */ 6160 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6161 return; 6162 6163 /* 6164 * duplicate the write operations while the dev replace 6165 * procedure is running. Since the copying of the old disk to 6166 * the new disk takes place at run time while the filesystem is 6167 * mounted writable, the regular write operations to the old 6168 * disk have to be duplicated to go to the new disk as well. 6169 * 6170 * Note that device->missing is handled by the caller, and that 6171 * the write to the old disk is already set up in the stripes 6172 * array. 6173 */ 6174 index_where_to_add = num_stripes; 6175 for (i = 0; i < num_stripes; i++) { 6176 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6177 /* write to new disk, too */ 6178 struct btrfs_io_stripe *new = 6179 bioc->stripes + index_where_to_add; 6180 struct btrfs_io_stripe *old = 6181 bioc->stripes + i; 6182 6183 new->physical = old->physical; 6184 new->length = old->length; 6185 new->dev = dev_replace->tgtdev; 6186 bioc->tgtdev_map[i] = index_where_to_add; 6187 index_where_to_add++; 6188 max_errors++; 6189 tgtdev_indexes++; 6190 } 6191 } 6192 num_stripes = index_where_to_add; 6193 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6194 int index_srcdev = 0; 6195 int found = 0; 6196 u64 physical_of_found = 0; 6197 6198 /* 6199 * During the dev-replace procedure, the target drive can also 6200 * be used to read data in case it is needed to repair a corrupt 6201 * block elsewhere. This is possible if the requested area is 6202 * left of the left cursor. In this area, the target drive is a 6203 * full copy of the source drive. 6204 */ 6205 for (i = 0; i < num_stripes; i++) { 6206 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6207 /* 6208 * In case of DUP, in order to keep it simple, 6209 * only add the mirror with the lowest physical 6210 * address 6211 */ 6212 if (found && 6213 physical_of_found <= bioc->stripes[i].physical) 6214 continue; 6215 index_srcdev = i; 6216 found = 1; 6217 physical_of_found = bioc->stripes[i].physical; 6218 } 6219 } 6220 if (found) { 6221 struct btrfs_io_stripe *tgtdev_stripe = 6222 bioc->stripes + num_stripes; 6223 6224 tgtdev_stripe->physical = physical_of_found; 6225 tgtdev_stripe->length = 6226 bioc->stripes[index_srcdev].length; 6227 tgtdev_stripe->dev = dev_replace->tgtdev; 6228 bioc->tgtdev_map[index_srcdev] = num_stripes; 6229 6230 tgtdev_indexes++; 6231 num_stripes++; 6232 } 6233 } 6234 6235 *num_stripes_ret = num_stripes; 6236 *max_errors_ret = max_errors; 6237 bioc->num_tgtdevs = tgtdev_indexes; 6238 *bioc_ret = bioc; 6239 } 6240 6241 static bool need_full_stripe(enum btrfs_map_op op) 6242 { 6243 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6244 } 6245 6246 /* 6247 * Calculate the geometry of a particular (address, len) tuple. This 6248 * information is used to calculate how big a particular bio can get before it 6249 * straddles a stripe. 6250 * 6251 * @fs_info: the filesystem 6252 * @em: mapping containing the logical extent 6253 * @op: type of operation - write or read 6254 * @logical: address that we want to figure out the geometry of 6255 * @io_geom: pointer used to return values 6256 * 6257 * Returns < 0 in case a chunk for the given logical address cannot be found, 6258 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6259 */ 6260 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6261 enum btrfs_map_op op, u64 logical, 6262 struct btrfs_io_geometry *io_geom) 6263 { 6264 struct map_lookup *map; 6265 u64 len; 6266 u64 offset; 6267 u64 stripe_offset; 6268 u64 stripe_nr; 6269 u64 stripe_len; 6270 u64 raid56_full_stripe_start = (u64)-1; 6271 int data_stripes; 6272 6273 ASSERT(op != BTRFS_MAP_DISCARD); 6274 6275 map = em->map_lookup; 6276 /* Offset of this logical address in the chunk */ 6277 offset = logical - em->start; 6278 /* Len of a stripe in a chunk */ 6279 stripe_len = map->stripe_len; 6280 /* Stripe where this block falls in */ 6281 stripe_nr = div64_u64(offset, stripe_len); 6282 /* Offset of stripe in the chunk */ 6283 stripe_offset = stripe_nr * stripe_len; 6284 if (offset < stripe_offset) { 6285 btrfs_crit(fs_info, 6286 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", 6287 stripe_offset, offset, em->start, logical, stripe_len); 6288 return -EINVAL; 6289 } 6290 6291 /* stripe_offset is the offset of this block in its stripe */ 6292 stripe_offset = offset - stripe_offset; 6293 data_stripes = nr_data_stripes(map); 6294 6295 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { 6296 u64 max_len = stripe_len - stripe_offset; 6297 6298 /* 6299 * In case of raid56, we need to know the stripe aligned start 6300 */ 6301 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6302 unsigned long full_stripe_len = stripe_len * data_stripes; 6303 raid56_full_stripe_start = offset; 6304 6305 /* 6306 * Allow a write of a full stripe, but make sure we 6307 * don't allow straddling of stripes 6308 */ 6309 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6310 full_stripe_len); 6311 raid56_full_stripe_start *= full_stripe_len; 6312 6313 /* 6314 * For writes to RAID[56], allow a full stripeset across 6315 * all disks. For other RAID types and for RAID[56] 6316 * reads, just allow a single stripe (on a single disk). 6317 */ 6318 if (op == BTRFS_MAP_WRITE) { 6319 max_len = stripe_len * data_stripes - 6320 (offset - raid56_full_stripe_start); 6321 } 6322 } 6323 len = min_t(u64, em->len - offset, max_len); 6324 } else { 6325 len = em->len - offset; 6326 } 6327 6328 io_geom->len = len; 6329 io_geom->offset = offset; 6330 io_geom->stripe_len = stripe_len; 6331 io_geom->stripe_nr = stripe_nr; 6332 io_geom->stripe_offset = stripe_offset; 6333 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6334 6335 return 0; 6336 } 6337 6338 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6339 enum btrfs_map_op op, 6340 u64 logical, u64 *length, 6341 struct btrfs_io_context **bioc_ret, 6342 int mirror_num, int need_raid_map) 6343 { 6344 struct extent_map *em; 6345 struct map_lookup *map; 6346 u64 stripe_offset; 6347 u64 stripe_nr; 6348 u64 stripe_len; 6349 u32 stripe_index; 6350 int data_stripes; 6351 int i; 6352 int ret = 0; 6353 int num_stripes; 6354 int max_errors = 0; 6355 int tgtdev_indexes = 0; 6356 struct btrfs_io_context *bioc = NULL; 6357 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6358 int dev_replace_is_ongoing = 0; 6359 int num_alloc_stripes; 6360 int patch_the_first_stripe_for_dev_replace = 0; 6361 u64 physical_to_patch_in_first_stripe = 0; 6362 u64 raid56_full_stripe_start = (u64)-1; 6363 struct btrfs_io_geometry geom; 6364 6365 ASSERT(bioc_ret); 6366 ASSERT(op != BTRFS_MAP_DISCARD); 6367 6368 em = btrfs_get_chunk_map(fs_info, logical, *length); 6369 ASSERT(!IS_ERR(em)); 6370 6371 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6372 if (ret < 0) 6373 return ret; 6374 6375 map = em->map_lookup; 6376 6377 *length = geom.len; 6378 stripe_len = geom.stripe_len; 6379 stripe_nr = geom.stripe_nr; 6380 stripe_offset = geom.stripe_offset; 6381 raid56_full_stripe_start = geom.raid56_stripe_offset; 6382 data_stripes = nr_data_stripes(map); 6383 6384 down_read(&dev_replace->rwsem); 6385 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6386 /* 6387 * Hold the semaphore for read during the whole operation, write is 6388 * requested at commit time but must wait. 6389 */ 6390 if (!dev_replace_is_ongoing) 6391 up_read(&dev_replace->rwsem); 6392 6393 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6394 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6395 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6396 dev_replace->srcdev->devid, 6397 &mirror_num, 6398 &physical_to_patch_in_first_stripe); 6399 if (ret) 6400 goto out; 6401 else 6402 patch_the_first_stripe_for_dev_replace = 1; 6403 } else if (mirror_num > map->num_stripes) { 6404 mirror_num = 0; 6405 } 6406 6407 num_stripes = 1; 6408 stripe_index = 0; 6409 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6410 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6411 &stripe_index); 6412 if (!need_full_stripe(op)) 6413 mirror_num = 1; 6414 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6415 if (need_full_stripe(op)) 6416 num_stripes = map->num_stripes; 6417 else if (mirror_num) 6418 stripe_index = mirror_num - 1; 6419 else { 6420 stripe_index = find_live_mirror(fs_info, map, 0, 6421 dev_replace_is_ongoing); 6422 mirror_num = stripe_index + 1; 6423 } 6424 6425 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6426 if (need_full_stripe(op)) { 6427 num_stripes = map->num_stripes; 6428 } else if (mirror_num) { 6429 stripe_index = mirror_num - 1; 6430 } else { 6431 mirror_num = 1; 6432 } 6433 6434 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6435 u32 factor = map->num_stripes / map->sub_stripes; 6436 6437 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6438 stripe_index *= map->sub_stripes; 6439 6440 if (need_full_stripe(op)) 6441 num_stripes = map->sub_stripes; 6442 else if (mirror_num) 6443 stripe_index += mirror_num - 1; 6444 else { 6445 int old_stripe_index = stripe_index; 6446 stripe_index = find_live_mirror(fs_info, map, 6447 stripe_index, 6448 dev_replace_is_ongoing); 6449 mirror_num = stripe_index - old_stripe_index + 1; 6450 } 6451 6452 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6453 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6454 /* push stripe_nr back to the start of the full stripe */ 6455 stripe_nr = div64_u64(raid56_full_stripe_start, 6456 stripe_len * data_stripes); 6457 6458 /* RAID[56] write or recovery. Return all stripes */ 6459 num_stripes = map->num_stripes; 6460 max_errors = nr_parity_stripes(map); 6461 6462 *length = map->stripe_len; 6463 stripe_index = 0; 6464 stripe_offset = 0; 6465 } else { 6466 /* 6467 * Mirror #0 or #1 means the original data block. 6468 * Mirror #2 is RAID5 parity block. 6469 * Mirror #3 is RAID6 Q block. 6470 */ 6471 stripe_nr = div_u64_rem(stripe_nr, 6472 data_stripes, &stripe_index); 6473 if (mirror_num > 1) 6474 stripe_index = data_stripes + mirror_num - 2; 6475 6476 /* We distribute the parity blocks across stripes */ 6477 div_u64_rem(stripe_nr + stripe_index, map->num_stripes, 6478 &stripe_index); 6479 if (!need_full_stripe(op) && mirror_num <= 1) 6480 mirror_num = 1; 6481 } 6482 } else { 6483 /* 6484 * after this, stripe_nr is the number of stripes on this 6485 * device we have to walk to find the data, and stripe_index is 6486 * the number of our device in the stripe array 6487 */ 6488 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6489 &stripe_index); 6490 mirror_num = stripe_index + 1; 6491 } 6492 if (stripe_index >= map->num_stripes) { 6493 btrfs_crit(fs_info, 6494 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6495 stripe_index, map->num_stripes); 6496 ret = -EINVAL; 6497 goto out; 6498 } 6499 6500 num_alloc_stripes = num_stripes; 6501 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6502 if (op == BTRFS_MAP_WRITE) 6503 num_alloc_stripes <<= 1; 6504 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6505 num_alloc_stripes++; 6506 tgtdev_indexes = num_stripes; 6507 } 6508 6509 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6510 if (!bioc) { 6511 ret = -ENOMEM; 6512 goto out; 6513 } 6514 6515 for (i = 0; i < num_stripes; i++) { 6516 bioc->stripes[i].physical = map->stripes[stripe_index].physical + 6517 stripe_offset + stripe_nr * map->stripe_len; 6518 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 6519 stripe_index++; 6520 } 6521 6522 /* Build raid_map */ 6523 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6524 (need_full_stripe(op) || mirror_num > 1)) { 6525 u64 tmp; 6526 unsigned rot; 6527 6528 /* Work out the disk rotation on this stripe-set */ 6529 div_u64_rem(stripe_nr, num_stripes, &rot); 6530 6531 /* Fill in the logical address of each stripe */ 6532 tmp = stripe_nr * data_stripes; 6533 for (i = 0; i < data_stripes; i++) 6534 bioc->raid_map[(i + rot) % num_stripes] = 6535 em->start + (tmp + i) * map->stripe_len; 6536 6537 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6538 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6539 bioc->raid_map[(i + rot + 1) % num_stripes] = 6540 RAID6_Q_STRIPE; 6541 6542 sort_parity_stripes(bioc, num_stripes); 6543 } 6544 6545 if (need_full_stripe(op)) 6546 max_errors = btrfs_chunk_max_errors(map); 6547 6548 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6549 need_full_stripe(op)) { 6550 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6551 &num_stripes, &max_errors); 6552 } 6553 6554 *bioc_ret = bioc; 6555 bioc->map_type = map->type; 6556 bioc->num_stripes = num_stripes; 6557 bioc->max_errors = max_errors; 6558 bioc->mirror_num = mirror_num; 6559 6560 /* 6561 * this is the case that REQ_READ && dev_replace_is_ongoing && 6562 * mirror_num == num_stripes + 1 && dev_replace target drive is 6563 * available as a mirror 6564 */ 6565 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6566 WARN_ON(num_stripes > 1); 6567 bioc->stripes[0].dev = dev_replace->tgtdev; 6568 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6569 bioc->mirror_num = map->num_stripes + 1; 6570 } 6571 out: 6572 if (dev_replace_is_ongoing) { 6573 lockdep_assert_held(&dev_replace->rwsem); 6574 /* Unlock and let waiting writers proceed */ 6575 up_read(&dev_replace->rwsem); 6576 } 6577 free_extent_map(em); 6578 return ret; 6579 } 6580 6581 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6582 u64 logical, u64 *length, 6583 struct btrfs_io_context **bioc_ret, int mirror_num) 6584 { 6585 if (op == BTRFS_MAP_DISCARD) 6586 return __btrfs_map_block_for_discard(fs_info, logical, 6587 length, bioc_ret); 6588 6589 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6590 mirror_num, 0); 6591 } 6592 6593 /* For Scrub/replace */ 6594 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6595 u64 logical, u64 *length, 6596 struct btrfs_io_context **bioc_ret) 6597 { 6598 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1); 6599 } 6600 6601 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio) 6602 { 6603 bio->bi_private = bioc->private; 6604 bio->bi_end_io = bioc->end_io; 6605 bio_endio(bio); 6606 6607 btrfs_put_bioc(bioc); 6608 } 6609 6610 static void btrfs_end_bio(struct bio *bio) 6611 { 6612 struct btrfs_io_context *bioc = bio->bi_private; 6613 int is_orig_bio = 0; 6614 6615 if (bio->bi_status) { 6616 atomic_inc(&bioc->error); 6617 if (bio->bi_status == BLK_STS_IOERR || 6618 bio->bi_status == BLK_STS_TARGET) { 6619 struct btrfs_device *dev = btrfs_bio(bio)->device; 6620 6621 ASSERT(dev->bdev); 6622 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6623 btrfs_dev_stat_inc_and_print(dev, 6624 BTRFS_DEV_STAT_WRITE_ERRS); 6625 else if (!(bio->bi_opf & REQ_RAHEAD)) 6626 btrfs_dev_stat_inc_and_print(dev, 6627 BTRFS_DEV_STAT_READ_ERRS); 6628 if (bio->bi_opf & REQ_PREFLUSH) 6629 btrfs_dev_stat_inc_and_print(dev, 6630 BTRFS_DEV_STAT_FLUSH_ERRS); 6631 } 6632 } 6633 6634 if (bio == bioc->orig_bio) 6635 is_orig_bio = 1; 6636 6637 btrfs_bio_counter_dec(bioc->fs_info); 6638 6639 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6640 if (!is_orig_bio) { 6641 bio_put(bio); 6642 bio = bioc->orig_bio; 6643 } 6644 6645 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6646 /* only send an error to the higher layers if it is 6647 * beyond the tolerance of the btrfs bio 6648 */ 6649 if (atomic_read(&bioc->error) > bioc->max_errors) { 6650 bio->bi_status = BLK_STS_IOERR; 6651 } else { 6652 /* 6653 * this bio is actually up to date, we didn't 6654 * go over the max number of errors 6655 */ 6656 bio->bi_status = BLK_STS_OK; 6657 } 6658 6659 btrfs_end_bioc(bioc, bio); 6660 } else if (!is_orig_bio) { 6661 bio_put(bio); 6662 } 6663 } 6664 6665 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio, 6666 u64 physical, struct btrfs_device *dev) 6667 { 6668 struct btrfs_fs_info *fs_info = bioc->fs_info; 6669 6670 bio->bi_private = bioc; 6671 btrfs_bio(bio)->device = dev; 6672 bio->bi_end_io = btrfs_end_bio; 6673 bio->bi_iter.bi_sector = physical >> 9; 6674 /* 6675 * For zone append writing, bi_sector must point the beginning of the 6676 * zone 6677 */ 6678 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6679 if (btrfs_dev_is_sequential(dev, physical)) { 6680 u64 zone_start = round_down(physical, fs_info->zone_size); 6681 6682 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6683 } else { 6684 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6685 bio->bi_opf |= REQ_OP_WRITE; 6686 } 6687 } 6688 btrfs_debug_in_rcu(fs_info, 6689 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6690 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6691 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6692 dev->devid, bio->bi_iter.bi_size); 6693 bio_set_dev(bio, dev->bdev); 6694 6695 btrfs_bio_counter_inc_noblocked(fs_info); 6696 6697 btrfsic_submit_bio(bio); 6698 } 6699 6700 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical) 6701 { 6702 atomic_inc(&bioc->error); 6703 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6704 /* Should be the original bio. */ 6705 WARN_ON(bio != bioc->orig_bio); 6706 6707 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6708 bio->bi_iter.bi_sector = logical >> 9; 6709 if (atomic_read(&bioc->error) > bioc->max_errors) 6710 bio->bi_status = BLK_STS_IOERR; 6711 else 6712 bio->bi_status = BLK_STS_OK; 6713 btrfs_end_bioc(bioc, bio); 6714 } 6715 } 6716 6717 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 6718 int mirror_num) 6719 { 6720 struct btrfs_device *dev; 6721 struct bio *first_bio = bio; 6722 u64 logical = bio->bi_iter.bi_sector << 9; 6723 u64 length = 0; 6724 u64 map_length; 6725 int ret; 6726 int dev_nr; 6727 int total_devs; 6728 struct btrfs_io_context *bioc = NULL; 6729 6730 length = bio->bi_iter.bi_size; 6731 map_length = length; 6732 6733 btrfs_bio_counter_inc_blocked(fs_info); 6734 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, 6735 &map_length, &bioc, mirror_num, 1); 6736 if (ret) { 6737 btrfs_bio_counter_dec(fs_info); 6738 return errno_to_blk_status(ret); 6739 } 6740 6741 total_devs = bioc->num_stripes; 6742 bioc->orig_bio = first_bio; 6743 bioc->private = first_bio->bi_private; 6744 bioc->end_io = first_bio->bi_end_io; 6745 atomic_set(&bioc->stripes_pending, bioc->num_stripes); 6746 6747 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && 6748 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) { 6749 /* In this case, map_length has been set to the length of 6750 a single stripe; not the whole write */ 6751 if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 6752 ret = raid56_parity_write(bio, bioc, map_length); 6753 } else { 6754 ret = raid56_parity_recover(bio, bioc, map_length, 6755 mirror_num, 1); 6756 } 6757 6758 btrfs_bio_counter_dec(fs_info); 6759 return errno_to_blk_status(ret); 6760 } 6761 6762 if (map_length < length) { 6763 btrfs_crit(fs_info, 6764 "mapping failed logical %llu bio len %llu len %llu", 6765 logical, length, map_length); 6766 BUG(); 6767 } 6768 6769 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { 6770 dev = bioc->stripes[dev_nr].dev; 6771 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, 6772 &dev->dev_state) || 6773 (btrfs_op(first_bio) == BTRFS_MAP_WRITE && 6774 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6775 bioc_error(bioc, first_bio, logical); 6776 continue; 6777 } 6778 6779 if (dev_nr < total_devs - 1) 6780 bio = btrfs_bio_clone(first_bio); 6781 else 6782 bio = first_bio; 6783 6784 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev); 6785 } 6786 btrfs_bio_counter_dec(fs_info); 6787 return BLK_STS_OK; 6788 } 6789 6790 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6791 const struct btrfs_fs_devices *fs_devices) 6792 { 6793 if (args->fsid == NULL) 6794 return true; 6795 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6796 return true; 6797 return false; 6798 } 6799 6800 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6801 const struct btrfs_device *device) 6802 { 6803 ASSERT((args->devid != (u64)-1) || args->missing); 6804 6805 if ((args->devid != (u64)-1) && device->devid != args->devid) 6806 return false; 6807 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6808 return false; 6809 if (!args->missing) 6810 return true; 6811 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6812 !device->bdev) 6813 return true; 6814 return false; 6815 } 6816 6817 /* 6818 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6819 * return NULL. 6820 * 6821 * If devid and uuid are both specified, the match must be exact, otherwise 6822 * only devid is used. 6823 */ 6824 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6825 const struct btrfs_dev_lookup_args *args) 6826 { 6827 struct btrfs_device *device; 6828 struct btrfs_fs_devices *seed_devs; 6829 6830 if (dev_args_match_fs_devices(args, fs_devices)) { 6831 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6832 if (dev_args_match_device(args, device)) 6833 return device; 6834 } 6835 } 6836 6837 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6838 if (!dev_args_match_fs_devices(args, seed_devs)) 6839 continue; 6840 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6841 if (dev_args_match_device(args, device)) 6842 return device; 6843 } 6844 } 6845 6846 return NULL; 6847 } 6848 6849 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6850 u64 devid, u8 *dev_uuid) 6851 { 6852 struct btrfs_device *device; 6853 unsigned int nofs_flag; 6854 6855 /* 6856 * We call this under the chunk_mutex, so we want to use NOFS for this 6857 * allocation, however we don't want to change btrfs_alloc_device() to 6858 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6859 * places. 6860 */ 6861 nofs_flag = memalloc_nofs_save(); 6862 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 6863 memalloc_nofs_restore(nofs_flag); 6864 if (IS_ERR(device)) 6865 return device; 6866 6867 list_add(&device->dev_list, &fs_devices->devices); 6868 device->fs_devices = fs_devices; 6869 fs_devices->num_devices++; 6870 6871 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6872 fs_devices->missing_devices++; 6873 6874 return device; 6875 } 6876 6877 /** 6878 * btrfs_alloc_device - allocate struct btrfs_device 6879 * @fs_info: used only for generating a new devid, can be NULL if 6880 * devid is provided (i.e. @devid != NULL). 6881 * @devid: a pointer to devid for this device. If NULL a new devid 6882 * is generated. 6883 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6884 * is generated. 6885 * 6886 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6887 * on error. Returned struct is not linked onto any lists and must be 6888 * destroyed with btrfs_free_device. 6889 */ 6890 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6891 const u64 *devid, 6892 const u8 *uuid) 6893 { 6894 struct btrfs_device *dev; 6895 u64 tmp; 6896 6897 if (WARN_ON(!devid && !fs_info)) 6898 return ERR_PTR(-EINVAL); 6899 6900 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6901 if (!dev) 6902 return ERR_PTR(-ENOMEM); 6903 6904 /* 6905 * Preallocate a bio that's always going to be used for flushing device 6906 * barriers and matches the device lifespan 6907 */ 6908 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0); 6909 if (!dev->flush_bio) { 6910 kfree(dev); 6911 return ERR_PTR(-ENOMEM); 6912 } 6913 6914 INIT_LIST_HEAD(&dev->dev_list); 6915 INIT_LIST_HEAD(&dev->dev_alloc_list); 6916 INIT_LIST_HEAD(&dev->post_commit_list); 6917 6918 atomic_set(&dev->reada_in_flight, 0); 6919 atomic_set(&dev->dev_stats_ccnt, 0); 6920 btrfs_device_data_ordered_init(dev); 6921 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6922 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6923 extent_io_tree_init(fs_info, &dev->alloc_state, 6924 IO_TREE_DEVICE_ALLOC_STATE, NULL); 6925 6926 if (devid) 6927 tmp = *devid; 6928 else { 6929 int ret; 6930 6931 ret = find_next_devid(fs_info, &tmp); 6932 if (ret) { 6933 btrfs_free_device(dev); 6934 return ERR_PTR(ret); 6935 } 6936 } 6937 dev->devid = tmp; 6938 6939 if (uuid) 6940 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6941 else 6942 generate_random_uuid(dev->uuid); 6943 6944 return dev; 6945 } 6946 6947 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6948 u64 devid, u8 *uuid, bool error) 6949 { 6950 if (error) 6951 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6952 devid, uuid); 6953 else 6954 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6955 devid, uuid); 6956 } 6957 6958 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) 6959 { 6960 const int data_stripes = calc_data_stripes(type, num_stripes); 6961 6962 return div_u64(chunk_len, data_stripes); 6963 } 6964 6965 #if BITS_PER_LONG == 32 6966 /* 6967 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6968 * can't be accessed on 32bit systems. 6969 * 6970 * This function do mount time check to reject the fs if it already has 6971 * metadata chunk beyond that limit. 6972 */ 6973 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6974 u64 logical, u64 length, u64 type) 6975 { 6976 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6977 return 0; 6978 6979 if (logical + length < MAX_LFS_FILESIZE) 6980 return 0; 6981 6982 btrfs_err_32bit_limit(fs_info); 6983 return -EOVERFLOW; 6984 } 6985 6986 /* 6987 * This is to give early warning for any metadata chunk reaching 6988 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6989 * Although we can still access the metadata, it's not going to be possible 6990 * once the limit is reached. 6991 */ 6992 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6993 u64 logical, u64 length, u64 type) 6994 { 6995 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6996 return; 6997 6998 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 6999 return; 7000 7001 btrfs_warn_32bit_limit(fs_info); 7002 } 7003 #endif 7004 7005 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 7006 struct btrfs_chunk *chunk) 7007 { 7008 BTRFS_DEV_LOOKUP_ARGS(args); 7009 struct btrfs_fs_info *fs_info = leaf->fs_info; 7010 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7011 struct map_lookup *map; 7012 struct extent_map *em; 7013 u64 logical; 7014 u64 length; 7015 u64 devid; 7016 u64 type; 7017 u8 uuid[BTRFS_UUID_SIZE]; 7018 int num_stripes; 7019 int ret; 7020 int i; 7021 7022 logical = key->offset; 7023 length = btrfs_chunk_length(leaf, chunk); 7024 type = btrfs_chunk_type(leaf, chunk); 7025 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 7026 7027 #if BITS_PER_LONG == 32 7028 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 7029 if (ret < 0) 7030 return ret; 7031 warn_32bit_meta_chunk(fs_info, logical, length, type); 7032 #endif 7033 7034 /* 7035 * Only need to verify chunk item if we're reading from sys chunk array, 7036 * as chunk item in tree block is already verified by tree-checker. 7037 */ 7038 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 7039 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 7040 if (ret) 7041 return ret; 7042 } 7043 7044 read_lock(&map_tree->lock); 7045 em = lookup_extent_mapping(map_tree, logical, 1); 7046 read_unlock(&map_tree->lock); 7047 7048 /* already mapped? */ 7049 if (em && em->start <= logical && em->start + em->len > logical) { 7050 free_extent_map(em); 7051 return 0; 7052 } else if (em) { 7053 free_extent_map(em); 7054 } 7055 7056 em = alloc_extent_map(); 7057 if (!em) 7058 return -ENOMEM; 7059 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7060 if (!map) { 7061 free_extent_map(em); 7062 return -ENOMEM; 7063 } 7064 7065 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7066 em->map_lookup = map; 7067 em->start = logical; 7068 em->len = length; 7069 em->orig_start = 0; 7070 em->block_start = 0; 7071 em->block_len = em->len; 7072 7073 map->num_stripes = num_stripes; 7074 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7075 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7076 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7077 map->type = type; 7078 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); 7079 map->verified_stripes = 0; 7080 em->orig_block_len = calc_stripe_length(type, em->len, 7081 map->num_stripes); 7082 for (i = 0; i < num_stripes; i++) { 7083 map->stripes[i].physical = 7084 btrfs_stripe_offset_nr(leaf, chunk, i); 7085 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7086 args.devid = devid; 7087 read_extent_buffer(leaf, uuid, (unsigned long) 7088 btrfs_stripe_dev_uuid_nr(chunk, i), 7089 BTRFS_UUID_SIZE); 7090 args.uuid = uuid; 7091 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 7092 if (!map->stripes[i].dev && 7093 !btrfs_test_opt(fs_info, DEGRADED)) { 7094 free_extent_map(em); 7095 btrfs_report_missing_device(fs_info, devid, uuid, true); 7096 return -ENOENT; 7097 } 7098 if (!map->stripes[i].dev) { 7099 map->stripes[i].dev = 7100 add_missing_dev(fs_info->fs_devices, devid, 7101 uuid); 7102 if (IS_ERR(map->stripes[i].dev)) { 7103 free_extent_map(em); 7104 btrfs_err(fs_info, 7105 "failed to init missing dev %llu: %ld", 7106 devid, PTR_ERR(map->stripes[i].dev)); 7107 return PTR_ERR(map->stripes[i].dev); 7108 } 7109 btrfs_report_missing_device(fs_info, devid, uuid, false); 7110 } 7111 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7112 &(map->stripes[i].dev->dev_state)); 7113 7114 } 7115 7116 write_lock(&map_tree->lock); 7117 ret = add_extent_mapping(map_tree, em, 0); 7118 write_unlock(&map_tree->lock); 7119 if (ret < 0) { 7120 btrfs_err(fs_info, 7121 "failed to add chunk map, start=%llu len=%llu: %d", 7122 em->start, em->len, ret); 7123 } 7124 free_extent_map(em); 7125 7126 return ret; 7127 } 7128 7129 static void fill_device_from_item(struct extent_buffer *leaf, 7130 struct btrfs_dev_item *dev_item, 7131 struct btrfs_device *device) 7132 { 7133 unsigned long ptr; 7134 7135 device->devid = btrfs_device_id(leaf, dev_item); 7136 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7137 device->total_bytes = device->disk_total_bytes; 7138 device->commit_total_bytes = device->disk_total_bytes; 7139 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7140 device->commit_bytes_used = device->bytes_used; 7141 device->type = btrfs_device_type(leaf, dev_item); 7142 device->io_align = btrfs_device_io_align(leaf, dev_item); 7143 device->io_width = btrfs_device_io_width(leaf, dev_item); 7144 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7145 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7146 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7147 7148 ptr = btrfs_device_uuid(dev_item); 7149 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7150 } 7151 7152 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7153 u8 *fsid) 7154 { 7155 struct btrfs_fs_devices *fs_devices; 7156 int ret; 7157 7158 lockdep_assert_held(&uuid_mutex); 7159 ASSERT(fsid); 7160 7161 /* This will match only for multi-device seed fs */ 7162 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7163 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7164 return fs_devices; 7165 7166 7167 fs_devices = find_fsid(fsid, NULL); 7168 if (!fs_devices) { 7169 if (!btrfs_test_opt(fs_info, DEGRADED)) 7170 return ERR_PTR(-ENOENT); 7171 7172 fs_devices = alloc_fs_devices(fsid, NULL); 7173 if (IS_ERR(fs_devices)) 7174 return fs_devices; 7175 7176 fs_devices->seeding = true; 7177 fs_devices->opened = 1; 7178 return fs_devices; 7179 } 7180 7181 /* 7182 * Upon first call for a seed fs fsid, just create a private copy of the 7183 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7184 */ 7185 fs_devices = clone_fs_devices(fs_devices); 7186 if (IS_ERR(fs_devices)) 7187 return fs_devices; 7188 7189 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7190 if (ret) { 7191 free_fs_devices(fs_devices); 7192 return ERR_PTR(ret); 7193 } 7194 7195 if (!fs_devices->seeding) { 7196 close_fs_devices(fs_devices); 7197 free_fs_devices(fs_devices); 7198 return ERR_PTR(-EINVAL); 7199 } 7200 7201 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7202 7203 return fs_devices; 7204 } 7205 7206 static int read_one_dev(struct extent_buffer *leaf, 7207 struct btrfs_dev_item *dev_item) 7208 { 7209 BTRFS_DEV_LOOKUP_ARGS(args); 7210 struct btrfs_fs_info *fs_info = leaf->fs_info; 7211 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7212 struct btrfs_device *device; 7213 u64 devid; 7214 int ret; 7215 u8 fs_uuid[BTRFS_FSID_SIZE]; 7216 u8 dev_uuid[BTRFS_UUID_SIZE]; 7217 7218 devid = args.devid = btrfs_device_id(leaf, dev_item); 7219 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7220 BTRFS_UUID_SIZE); 7221 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7222 BTRFS_FSID_SIZE); 7223 args.uuid = dev_uuid; 7224 args.fsid = fs_uuid; 7225 7226 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7227 fs_devices = open_seed_devices(fs_info, fs_uuid); 7228 if (IS_ERR(fs_devices)) 7229 return PTR_ERR(fs_devices); 7230 } 7231 7232 device = btrfs_find_device(fs_info->fs_devices, &args); 7233 if (!device) { 7234 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7235 btrfs_report_missing_device(fs_info, devid, 7236 dev_uuid, true); 7237 return -ENOENT; 7238 } 7239 7240 device = add_missing_dev(fs_devices, devid, dev_uuid); 7241 if (IS_ERR(device)) { 7242 btrfs_err(fs_info, 7243 "failed to add missing dev %llu: %ld", 7244 devid, PTR_ERR(device)); 7245 return PTR_ERR(device); 7246 } 7247 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7248 } else { 7249 if (!device->bdev) { 7250 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7251 btrfs_report_missing_device(fs_info, 7252 devid, dev_uuid, true); 7253 return -ENOENT; 7254 } 7255 btrfs_report_missing_device(fs_info, devid, 7256 dev_uuid, false); 7257 } 7258 7259 if (!device->bdev && 7260 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7261 /* 7262 * this happens when a device that was properly setup 7263 * in the device info lists suddenly goes bad. 7264 * device->bdev is NULL, and so we have to set 7265 * device->missing to one here 7266 */ 7267 device->fs_devices->missing_devices++; 7268 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7269 } 7270 7271 /* Move the device to its own fs_devices */ 7272 if (device->fs_devices != fs_devices) { 7273 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7274 &device->dev_state)); 7275 7276 list_move(&device->dev_list, &fs_devices->devices); 7277 device->fs_devices->num_devices--; 7278 fs_devices->num_devices++; 7279 7280 device->fs_devices->missing_devices--; 7281 fs_devices->missing_devices++; 7282 7283 device->fs_devices = fs_devices; 7284 } 7285 } 7286 7287 if (device->fs_devices != fs_info->fs_devices) { 7288 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7289 if (device->generation != 7290 btrfs_device_generation(leaf, dev_item)) 7291 return -EINVAL; 7292 } 7293 7294 fill_device_from_item(leaf, dev_item, device); 7295 if (device->bdev) { 7296 u64 max_total_bytes = i_size_read(device->bdev->bd_inode); 7297 7298 if (device->total_bytes > max_total_bytes) { 7299 btrfs_err(fs_info, 7300 "device total_bytes should be at most %llu but found %llu", 7301 max_total_bytes, device->total_bytes); 7302 return -EINVAL; 7303 } 7304 } 7305 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7306 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7307 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7308 device->fs_devices->total_rw_bytes += device->total_bytes; 7309 atomic64_add(device->total_bytes - device->bytes_used, 7310 &fs_info->free_chunk_space); 7311 } 7312 ret = 0; 7313 return ret; 7314 } 7315 7316 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7317 { 7318 struct btrfs_root *root = fs_info->tree_root; 7319 struct btrfs_super_block *super_copy = fs_info->super_copy; 7320 struct extent_buffer *sb; 7321 struct btrfs_disk_key *disk_key; 7322 struct btrfs_chunk *chunk; 7323 u8 *array_ptr; 7324 unsigned long sb_array_offset; 7325 int ret = 0; 7326 u32 num_stripes; 7327 u32 array_size; 7328 u32 len = 0; 7329 u32 cur_offset; 7330 u64 type; 7331 struct btrfs_key key; 7332 7333 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7334 /* 7335 * This will create extent buffer of nodesize, superblock size is 7336 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will 7337 * overallocate but we can keep it as-is, only the first page is used. 7338 */ 7339 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET, 7340 root->root_key.objectid, 0); 7341 if (IS_ERR(sb)) 7342 return PTR_ERR(sb); 7343 set_extent_buffer_uptodate(sb); 7344 /* 7345 * The sb extent buffer is artificial and just used to read the system array. 7346 * set_extent_buffer_uptodate() call does not properly mark all it's 7347 * pages up-to-date when the page is larger: extent does not cover the 7348 * whole page and consequently check_page_uptodate does not find all 7349 * the page's extents up-to-date (the hole beyond sb), 7350 * write_extent_buffer then triggers a WARN_ON. 7351 * 7352 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, 7353 * but sb spans only this function. Add an explicit SetPageUptodate call 7354 * to silence the warning eg. on PowerPC 64. 7355 */ 7356 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) 7357 SetPageUptodate(sb->pages[0]); 7358 7359 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7360 array_size = btrfs_super_sys_array_size(super_copy); 7361 7362 array_ptr = super_copy->sys_chunk_array; 7363 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7364 cur_offset = 0; 7365 7366 while (cur_offset < array_size) { 7367 disk_key = (struct btrfs_disk_key *)array_ptr; 7368 len = sizeof(*disk_key); 7369 if (cur_offset + len > array_size) 7370 goto out_short_read; 7371 7372 btrfs_disk_key_to_cpu(&key, disk_key); 7373 7374 array_ptr += len; 7375 sb_array_offset += len; 7376 cur_offset += len; 7377 7378 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7379 btrfs_err(fs_info, 7380 "unexpected item type %u in sys_array at offset %u", 7381 (u32)key.type, cur_offset); 7382 ret = -EIO; 7383 break; 7384 } 7385 7386 chunk = (struct btrfs_chunk *)sb_array_offset; 7387 /* 7388 * At least one btrfs_chunk with one stripe must be present, 7389 * exact stripe count check comes afterwards 7390 */ 7391 len = btrfs_chunk_item_size(1); 7392 if (cur_offset + len > array_size) 7393 goto out_short_read; 7394 7395 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7396 if (!num_stripes) { 7397 btrfs_err(fs_info, 7398 "invalid number of stripes %u in sys_array at offset %u", 7399 num_stripes, cur_offset); 7400 ret = -EIO; 7401 break; 7402 } 7403 7404 type = btrfs_chunk_type(sb, chunk); 7405 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7406 btrfs_err(fs_info, 7407 "invalid chunk type %llu in sys_array at offset %u", 7408 type, cur_offset); 7409 ret = -EIO; 7410 break; 7411 } 7412 7413 len = btrfs_chunk_item_size(num_stripes); 7414 if (cur_offset + len > array_size) 7415 goto out_short_read; 7416 7417 ret = read_one_chunk(&key, sb, chunk); 7418 if (ret) 7419 break; 7420 7421 array_ptr += len; 7422 sb_array_offset += len; 7423 cur_offset += len; 7424 } 7425 clear_extent_buffer_uptodate(sb); 7426 free_extent_buffer_stale(sb); 7427 return ret; 7428 7429 out_short_read: 7430 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7431 len, cur_offset); 7432 clear_extent_buffer_uptodate(sb); 7433 free_extent_buffer_stale(sb); 7434 return -EIO; 7435 } 7436 7437 /* 7438 * Check if all chunks in the fs are OK for read-write degraded mount 7439 * 7440 * If the @failing_dev is specified, it's accounted as missing. 7441 * 7442 * Return true if all chunks meet the minimal RW mount requirements. 7443 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7444 */ 7445 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7446 struct btrfs_device *failing_dev) 7447 { 7448 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7449 struct extent_map *em; 7450 u64 next_start = 0; 7451 bool ret = true; 7452 7453 read_lock(&map_tree->lock); 7454 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7455 read_unlock(&map_tree->lock); 7456 /* No chunk at all? Return false anyway */ 7457 if (!em) { 7458 ret = false; 7459 goto out; 7460 } 7461 while (em) { 7462 struct map_lookup *map; 7463 int missing = 0; 7464 int max_tolerated; 7465 int i; 7466 7467 map = em->map_lookup; 7468 max_tolerated = 7469 btrfs_get_num_tolerated_disk_barrier_failures( 7470 map->type); 7471 for (i = 0; i < map->num_stripes; i++) { 7472 struct btrfs_device *dev = map->stripes[i].dev; 7473 7474 if (!dev || !dev->bdev || 7475 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7476 dev->last_flush_error) 7477 missing++; 7478 else if (failing_dev && failing_dev == dev) 7479 missing++; 7480 } 7481 if (missing > max_tolerated) { 7482 if (!failing_dev) 7483 btrfs_warn(fs_info, 7484 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7485 em->start, missing, max_tolerated); 7486 free_extent_map(em); 7487 ret = false; 7488 goto out; 7489 } 7490 next_start = extent_map_end(em); 7491 free_extent_map(em); 7492 7493 read_lock(&map_tree->lock); 7494 em = lookup_extent_mapping(map_tree, next_start, 7495 (u64)(-1) - next_start); 7496 read_unlock(&map_tree->lock); 7497 } 7498 out: 7499 return ret; 7500 } 7501 7502 static void readahead_tree_node_children(struct extent_buffer *node) 7503 { 7504 int i; 7505 const int nr_items = btrfs_header_nritems(node); 7506 7507 for (i = 0; i < nr_items; i++) 7508 btrfs_readahead_node_child(node, i); 7509 } 7510 7511 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7512 { 7513 struct btrfs_root *root = fs_info->chunk_root; 7514 struct btrfs_path *path; 7515 struct extent_buffer *leaf; 7516 struct btrfs_key key; 7517 struct btrfs_key found_key; 7518 int ret; 7519 int slot; 7520 u64 total_dev = 0; 7521 u64 last_ra_node = 0; 7522 7523 path = btrfs_alloc_path(); 7524 if (!path) 7525 return -ENOMEM; 7526 7527 /* 7528 * uuid_mutex is needed only if we are mounting a sprout FS 7529 * otherwise we don't need it. 7530 */ 7531 mutex_lock(&uuid_mutex); 7532 7533 /* 7534 * It is possible for mount and umount to race in such a way that 7535 * we execute this code path, but open_fs_devices failed to clear 7536 * total_rw_bytes. We certainly want it cleared before reading the 7537 * device items, so clear it here. 7538 */ 7539 fs_info->fs_devices->total_rw_bytes = 0; 7540 7541 /* 7542 * Read all device items, and then all the chunk items. All 7543 * device items are found before any chunk item (their object id 7544 * is smaller than the lowest possible object id for a chunk 7545 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7546 */ 7547 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7548 key.offset = 0; 7549 key.type = 0; 7550 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7551 if (ret < 0) 7552 goto error; 7553 while (1) { 7554 struct extent_buffer *node; 7555 7556 leaf = path->nodes[0]; 7557 slot = path->slots[0]; 7558 if (slot >= btrfs_header_nritems(leaf)) { 7559 ret = btrfs_next_leaf(root, path); 7560 if (ret == 0) 7561 continue; 7562 if (ret < 0) 7563 goto error; 7564 break; 7565 } 7566 /* 7567 * The nodes on level 1 are not locked but we don't need to do 7568 * that during mount time as nothing else can access the tree 7569 */ 7570 node = path->nodes[1]; 7571 if (node) { 7572 if (last_ra_node != node->start) { 7573 readahead_tree_node_children(node); 7574 last_ra_node = node->start; 7575 } 7576 } 7577 btrfs_item_key_to_cpu(leaf, &found_key, slot); 7578 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7579 struct btrfs_dev_item *dev_item; 7580 dev_item = btrfs_item_ptr(leaf, slot, 7581 struct btrfs_dev_item); 7582 ret = read_one_dev(leaf, dev_item); 7583 if (ret) 7584 goto error; 7585 total_dev++; 7586 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7587 struct btrfs_chunk *chunk; 7588 7589 /* 7590 * We are only called at mount time, so no need to take 7591 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7592 * we always lock first fs_info->chunk_mutex before 7593 * acquiring any locks on the chunk tree. This is a 7594 * requirement for chunk allocation, see the comment on 7595 * top of btrfs_chunk_alloc() for details. 7596 */ 7597 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7598 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7599 ret = read_one_chunk(&found_key, leaf, chunk); 7600 if (ret) 7601 goto error; 7602 } 7603 path->slots[0]++; 7604 } 7605 7606 /* 7607 * After loading chunk tree, we've got all device information, 7608 * do another round of validation checks. 7609 */ 7610 if (total_dev != fs_info->fs_devices->total_devices) { 7611 btrfs_err(fs_info, 7612 "super_num_devices %llu mismatch with num_devices %llu found here", 7613 btrfs_super_num_devices(fs_info->super_copy), 7614 total_dev); 7615 ret = -EINVAL; 7616 goto error; 7617 } 7618 if (btrfs_super_total_bytes(fs_info->super_copy) < 7619 fs_info->fs_devices->total_rw_bytes) { 7620 btrfs_err(fs_info, 7621 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7622 btrfs_super_total_bytes(fs_info->super_copy), 7623 fs_info->fs_devices->total_rw_bytes); 7624 ret = -EINVAL; 7625 goto error; 7626 } 7627 ret = 0; 7628 error: 7629 mutex_unlock(&uuid_mutex); 7630 7631 btrfs_free_path(path); 7632 return ret; 7633 } 7634 7635 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7636 { 7637 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7638 struct btrfs_device *device; 7639 7640 fs_devices->fs_info = fs_info; 7641 7642 mutex_lock(&fs_devices->device_list_mutex); 7643 list_for_each_entry(device, &fs_devices->devices, dev_list) 7644 device->fs_info = fs_info; 7645 7646 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7647 list_for_each_entry(device, &seed_devs->devices, dev_list) 7648 device->fs_info = fs_info; 7649 7650 seed_devs->fs_info = fs_info; 7651 } 7652 mutex_unlock(&fs_devices->device_list_mutex); 7653 } 7654 7655 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7656 const struct btrfs_dev_stats_item *ptr, 7657 int index) 7658 { 7659 u64 val; 7660 7661 read_extent_buffer(eb, &val, 7662 offsetof(struct btrfs_dev_stats_item, values) + 7663 ((unsigned long)ptr) + (index * sizeof(u64)), 7664 sizeof(val)); 7665 return val; 7666 } 7667 7668 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7669 struct btrfs_dev_stats_item *ptr, 7670 int index, u64 val) 7671 { 7672 write_extent_buffer(eb, &val, 7673 offsetof(struct btrfs_dev_stats_item, values) + 7674 ((unsigned long)ptr) + (index * sizeof(u64)), 7675 sizeof(val)); 7676 } 7677 7678 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7679 struct btrfs_path *path) 7680 { 7681 struct btrfs_dev_stats_item *ptr; 7682 struct extent_buffer *eb; 7683 struct btrfs_key key; 7684 int item_size; 7685 int i, ret, slot; 7686 7687 if (!device->fs_info->dev_root) 7688 return 0; 7689 7690 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7691 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7692 key.offset = device->devid; 7693 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7694 if (ret) { 7695 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7696 btrfs_dev_stat_set(device, i, 0); 7697 device->dev_stats_valid = 1; 7698 btrfs_release_path(path); 7699 return ret < 0 ? ret : 0; 7700 } 7701 slot = path->slots[0]; 7702 eb = path->nodes[0]; 7703 item_size = btrfs_item_size_nr(eb, slot); 7704 7705 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7706 7707 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7708 if (item_size >= (1 + i) * sizeof(__le64)) 7709 btrfs_dev_stat_set(device, i, 7710 btrfs_dev_stats_value(eb, ptr, i)); 7711 else 7712 btrfs_dev_stat_set(device, i, 0); 7713 } 7714 7715 device->dev_stats_valid = 1; 7716 btrfs_dev_stat_print_on_load(device); 7717 btrfs_release_path(path); 7718 7719 return 0; 7720 } 7721 7722 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7723 { 7724 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7725 struct btrfs_device *device; 7726 struct btrfs_path *path = NULL; 7727 int ret = 0; 7728 7729 path = btrfs_alloc_path(); 7730 if (!path) 7731 return -ENOMEM; 7732 7733 mutex_lock(&fs_devices->device_list_mutex); 7734 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7735 ret = btrfs_device_init_dev_stats(device, path); 7736 if (ret) 7737 goto out; 7738 } 7739 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7740 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7741 ret = btrfs_device_init_dev_stats(device, path); 7742 if (ret) 7743 goto out; 7744 } 7745 } 7746 out: 7747 mutex_unlock(&fs_devices->device_list_mutex); 7748 7749 btrfs_free_path(path); 7750 return ret; 7751 } 7752 7753 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7754 struct btrfs_device *device) 7755 { 7756 struct btrfs_fs_info *fs_info = trans->fs_info; 7757 struct btrfs_root *dev_root = fs_info->dev_root; 7758 struct btrfs_path *path; 7759 struct btrfs_key key; 7760 struct extent_buffer *eb; 7761 struct btrfs_dev_stats_item *ptr; 7762 int ret; 7763 int i; 7764 7765 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7766 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7767 key.offset = device->devid; 7768 7769 path = btrfs_alloc_path(); 7770 if (!path) 7771 return -ENOMEM; 7772 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7773 if (ret < 0) { 7774 btrfs_warn_in_rcu(fs_info, 7775 "error %d while searching for dev_stats item for device %s", 7776 ret, rcu_str_deref(device->name)); 7777 goto out; 7778 } 7779 7780 if (ret == 0 && 7781 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7782 /* need to delete old one and insert a new one */ 7783 ret = btrfs_del_item(trans, dev_root, path); 7784 if (ret != 0) { 7785 btrfs_warn_in_rcu(fs_info, 7786 "delete too small dev_stats item for device %s failed %d", 7787 rcu_str_deref(device->name), ret); 7788 goto out; 7789 } 7790 ret = 1; 7791 } 7792 7793 if (ret == 1) { 7794 /* need to insert a new item */ 7795 btrfs_release_path(path); 7796 ret = btrfs_insert_empty_item(trans, dev_root, path, 7797 &key, sizeof(*ptr)); 7798 if (ret < 0) { 7799 btrfs_warn_in_rcu(fs_info, 7800 "insert dev_stats item for device %s failed %d", 7801 rcu_str_deref(device->name), ret); 7802 goto out; 7803 } 7804 } 7805 7806 eb = path->nodes[0]; 7807 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7808 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7809 btrfs_set_dev_stats_value(eb, ptr, i, 7810 btrfs_dev_stat_read(device, i)); 7811 btrfs_mark_buffer_dirty(eb); 7812 7813 out: 7814 btrfs_free_path(path); 7815 return ret; 7816 } 7817 7818 /* 7819 * called from commit_transaction. Writes all changed device stats to disk. 7820 */ 7821 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7822 { 7823 struct btrfs_fs_info *fs_info = trans->fs_info; 7824 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7825 struct btrfs_device *device; 7826 int stats_cnt; 7827 int ret = 0; 7828 7829 mutex_lock(&fs_devices->device_list_mutex); 7830 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7831 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7832 if (!device->dev_stats_valid || stats_cnt == 0) 7833 continue; 7834 7835 7836 /* 7837 * There is a LOAD-LOAD control dependency between the value of 7838 * dev_stats_ccnt and updating the on-disk values which requires 7839 * reading the in-memory counters. Such control dependencies 7840 * require explicit read memory barriers. 7841 * 7842 * This memory barriers pairs with smp_mb__before_atomic in 7843 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7844 * barrier implied by atomic_xchg in 7845 * btrfs_dev_stats_read_and_reset 7846 */ 7847 smp_rmb(); 7848 7849 ret = update_dev_stat_item(trans, device); 7850 if (!ret) 7851 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7852 } 7853 mutex_unlock(&fs_devices->device_list_mutex); 7854 7855 return ret; 7856 } 7857 7858 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7859 { 7860 btrfs_dev_stat_inc(dev, index); 7861 btrfs_dev_stat_print_on_error(dev); 7862 } 7863 7864 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) 7865 { 7866 if (!dev->dev_stats_valid) 7867 return; 7868 btrfs_err_rl_in_rcu(dev->fs_info, 7869 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7870 rcu_str_deref(dev->name), 7871 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7872 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7873 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7874 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7875 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7876 } 7877 7878 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7879 { 7880 int i; 7881 7882 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7883 if (btrfs_dev_stat_read(dev, i) != 0) 7884 break; 7885 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7886 return; /* all values == 0, suppress message */ 7887 7888 btrfs_info_in_rcu(dev->fs_info, 7889 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7890 rcu_str_deref(dev->name), 7891 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7892 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7893 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7894 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7895 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7896 } 7897 7898 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7899 struct btrfs_ioctl_get_dev_stats *stats) 7900 { 7901 BTRFS_DEV_LOOKUP_ARGS(args); 7902 struct btrfs_device *dev; 7903 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7904 int i; 7905 7906 mutex_lock(&fs_devices->device_list_mutex); 7907 args.devid = stats->devid; 7908 dev = btrfs_find_device(fs_info->fs_devices, &args); 7909 mutex_unlock(&fs_devices->device_list_mutex); 7910 7911 if (!dev) { 7912 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7913 return -ENODEV; 7914 } else if (!dev->dev_stats_valid) { 7915 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7916 return -ENODEV; 7917 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7918 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7919 if (stats->nr_items > i) 7920 stats->values[i] = 7921 btrfs_dev_stat_read_and_reset(dev, i); 7922 else 7923 btrfs_dev_stat_set(dev, i, 0); 7924 } 7925 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7926 current->comm, task_pid_nr(current)); 7927 } else { 7928 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7929 if (stats->nr_items > i) 7930 stats->values[i] = btrfs_dev_stat_read(dev, i); 7931 } 7932 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7933 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7934 return 0; 7935 } 7936 7937 /* 7938 * Update the size and bytes used for each device where it changed. This is 7939 * delayed since we would otherwise get errors while writing out the 7940 * superblocks. 7941 * 7942 * Must be invoked during transaction commit. 7943 */ 7944 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7945 { 7946 struct btrfs_device *curr, *next; 7947 7948 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7949 7950 if (list_empty(&trans->dev_update_list)) 7951 return; 7952 7953 /* 7954 * We don't need the device_list_mutex here. This list is owned by the 7955 * transaction and the transaction must complete before the device is 7956 * released. 7957 */ 7958 mutex_lock(&trans->fs_info->chunk_mutex); 7959 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7960 post_commit_list) { 7961 list_del_init(&curr->post_commit_list); 7962 curr->commit_total_bytes = curr->disk_total_bytes; 7963 curr->commit_bytes_used = curr->bytes_used; 7964 } 7965 mutex_unlock(&trans->fs_info->chunk_mutex); 7966 } 7967 7968 /* 7969 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7970 */ 7971 int btrfs_bg_type_to_factor(u64 flags) 7972 { 7973 const int index = btrfs_bg_flags_to_raid_index(flags); 7974 7975 return btrfs_raid_array[index].ncopies; 7976 } 7977 7978 7979 7980 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7981 u64 chunk_offset, u64 devid, 7982 u64 physical_offset, u64 physical_len) 7983 { 7984 struct btrfs_dev_lookup_args args = { .devid = devid }; 7985 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7986 struct extent_map *em; 7987 struct map_lookup *map; 7988 struct btrfs_device *dev; 7989 u64 stripe_len; 7990 bool found = false; 7991 int ret = 0; 7992 int i; 7993 7994 read_lock(&em_tree->lock); 7995 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7996 read_unlock(&em_tree->lock); 7997 7998 if (!em) { 7999 btrfs_err(fs_info, 8000 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 8001 physical_offset, devid); 8002 ret = -EUCLEAN; 8003 goto out; 8004 } 8005 8006 map = em->map_lookup; 8007 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); 8008 if (physical_len != stripe_len) { 8009 btrfs_err(fs_info, 8010 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 8011 physical_offset, devid, em->start, physical_len, 8012 stripe_len); 8013 ret = -EUCLEAN; 8014 goto out; 8015 } 8016 8017 for (i = 0; i < map->num_stripes; i++) { 8018 if (map->stripes[i].dev->devid == devid && 8019 map->stripes[i].physical == physical_offset) { 8020 found = true; 8021 if (map->verified_stripes >= map->num_stripes) { 8022 btrfs_err(fs_info, 8023 "too many dev extents for chunk %llu found", 8024 em->start); 8025 ret = -EUCLEAN; 8026 goto out; 8027 } 8028 map->verified_stripes++; 8029 break; 8030 } 8031 } 8032 if (!found) { 8033 btrfs_err(fs_info, 8034 "dev extent physical offset %llu devid %llu has no corresponding chunk", 8035 physical_offset, devid); 8036 ret = -EUCLEAN; 8037 } 8038 8039 /* Make sure no dev extent is beyond device boundary */ 8040 dev = btrfs_find_device(fs_info->fs_devices, &args); 8041 if (!dev) { 8042 btrfs_err(fs_info, "failed to find devid %llu", devid); 8043 ret = -EUCLEAN; 8044 goto out; 8045 } 8046 8047 if (physical_offset + physical_len > dev->disk_total_bytes) { 8048 btrfs_err(fs_info, 8049 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 8050 devid, physical_offset, physical_len, 8051 dev->disk_total_bytes); 8052 ret = -EUCLEAN; 8053 goto out; 8054 } 8055 8056 if (dev->zone_info) { 8057 u64 zone_size = dev->zone_info->zone_size; 8058 8059 if (!IS_ALIGNED(physical_offset, zone_size) || 8060 !IS_ALIGNED(physical_len, zone_size)) { 8061 btrfs_err(fs_info, 8062 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8063 devid, physical_offset, physical_len); 8064 ret = -EUCLEAN; 8065 goto out; 8066 } 8067 } 8068 8069 out: 8070 free_extent_map(em); 8071 return ret; 8072 } 8073 8074 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8075 { 8076 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8077 struct extent_map *em; 8078 struct rb_node *node; 8079 int ret = 0; 8080 8081 read_lock(&em_tree->lock); 8082 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8083 em = rb_entry(node, struct extent_map, rb_node); 8084 if (em->map_lookup->num_stripes != 8085 em->map_lookup->verified_stripes) { 8086 btrfs_err(fs_info, 8087 "chunk %llu has missing dev extent, have %d expect %d", 8088 em->start, em->map_lookup->verified_stripes, 8089 em->map_lookup->num_stripes); 8090 ret = -EUCLEAN; 8091 goto out; 8092 } 8093 } 8094 out: 8095 read_unlock(&em_tree->lock); 8096 return ret; 8097 } 8098 8099 /* 8100 * Ensure that all dev extents are mapped to correct chunk, otherwise 8101 * later chunk allocation/free would cause unexpected behavior. 8102 * 8103 * NOTE: This will iterate through the whole device tree, which should be of 8104 * the same size level as the chunk tree. This slightly increases mount time. 8105 */ 8106 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8107 { 8108 struct btrfs_path *path; 8109 struct btrfs_root *root = fs_info->dev_root; 8110 struct btrfs_key key; 8111 u64 prev_devid = 0; 8112 u64 prev_dev_ext_end = 0; 8113 int ret = 0; 8114 8115 /* 8116 * We don't have a dev_root because we mounted with ignorebadroots and 8117 * failed to load the root, so we want to skip the verification in this 8118 * case for sure. 8119 * 8120 * However if the dev root is fine, but the tree itself is corrupted 8121 * we'd still fail to mount. This verification is only to make sure 8122 * writes can happen safely, so instead just bypass this check 8123 * completely in the case of IGNOREBADROOTS. 8124 */ 8125 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8126 return 0; 8127 8128 key.objectid = 1; 8129 key.type = BTRFS_DEV_EXTENT_KEY; 8130 key.offset = 0; 8131 8132 path = btrfs_alloc_path(); 8133 if (!path) 8134 return -ENOMEM; 8135 8136 path->reada = READA_FORWARD; 8137 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8138 if (ret < 0) 8139 goto out; 8140 8141 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8142 ret = btrfs_next_leaf(root, path); 8143 if (ret < 0) 8144 goto out; 8145 /* No dev extents at all? Not good */ 8146 if (ret > 0) { 8147 ret = -EUCLEAN; 8148 goto out; 8149 } 8150 } 8151 while (1) { 8152 struct extent_buffer *leaf = path->nodes[0]; 8153 struct btrfs_dev_extent *dext; 8154 int slot = path->slots[0]; 8155 u64 chunk_offset; 8156 u64 physical_offset; 8157 u64 physical_len; 8158 u64 devid; 8159 8160 btrfs_item_key_to_cpu(leaf, &key, slot); 8161 if (key.type != BTRFS_DEV_EXTENT_KEY) 8162 break; 8163 devid = key.objectid; 8164 physical_offset = key.offset; 8165 8166 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8167 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8168 physical_len = btrfs_dev_extent_length(leaf, dext); 8169 8170 /* Check if this dev extent overlaps with the previous one */ 8171 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8172 btrfs_err(fs_info, 8173 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8174 devid, physical_offset, prev_dev_ext_end); 8175 ret = -EUCLEAN; 8176 goto out; 8177 } 8178 8179 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8180 physical_offset, physical_len); 8181 if (ret < 0) 8182 goto out; 8183 prev_devid = devid; 8184 prev_dev_ext_end = physical_offset + physical_len; 8185 8186 ret = btrfs_next_item(root, path); 8187 if (ret < 0) 8188 goto out; 8189 if (ret > 0) { 8190 ret = 0; 8191 break; 8192 } 8193 } 8194 8195 /* Ensure all chunks have corresponding dev extents */ 8196 ret = verify_chunk_dev_extent_mapping(fs_info); 8197 out: 8198 btrfs_free_path(path); 8199 return ret; 8200 } 8201 8202 /* 8203 * Check whether the given block group or device is pinned by any inode being 8204 * used as a swapfile. 8205 */ 8206 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8207 { 8208 struct btrfs_swapfile_pin *sp; 8209 struct rb_node *node; 8210 8211 spin_lock(&fs_info->swapfile_pins_lock); 8212 node = fs_info->swapfile_pins.rb_node; 8213 while (node) { 8214 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8215 if (ptr < sp->ptr) 8216 node = node->rb_left; 8217 else if (ptr > sp->ptr) 8218 node = node->rb_right; 8219 else 8220 break; 8221 } 8222 spin_unlock(&fs_info->swapfile_pins_lock); 8223 return node != NULL; 8224 } 8225 8226 static int relocating_repair_kthread(void *data) 8227 { 8228 struct btrfs_block_group *cache = (struct btrfs_block_group *)data; 8229 struct btrfs_fs_info *fs_info = cache->fs_info; 8230 u64 target; 8231 int ret = 0; 8232 8233 target = cache->start; 8234 btrfs_put_block_group(cache); 8235 8236 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8237 btrfs_info(fs_info, 8238 "zoned: skip relocating block group %llu to repair: EBUSY", 8239 target); 8240 return -EBUSY; 8241 } 8242 8243 mutex_lock(&fs_info->reclaim_bgs_lock); 8244 8245 /* Ensure block group still exists */ 8246 cache = btrfs_lookup_block_group(fs_info, target); 8247 if (!cache) 8248 goto out; 8249 8250 if (!cache->relocating_repair) 8251 goto out; 8252 8253 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8254 if (ret < 0) 8255 goto out; 8256 8257 btrfs_info(fs_info, 8258 "zoned: relocating block group %llu to repair IO failure", 8259 target); 8260 ret = btrfs_relocate_chunk(fs_info, target); 8261 8262 out: 8263 if (cache) 8264 btrfs_put_block_group(cache); 8265 mutex_unlock(&fs_info->reclaim_bgs_lock); 8266 btrfs_exclop_finish(fs_info); 8267 8268 return ret; 8269 } 8270 8271 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8272 { 8273 struct btrfs_block_group *cache; 8274 8275 /* Do not attempt to repair in degraded state */ 8276 if (btrfs_test_opt(fs_info, DEGRADED)) 8277 return 0; 8278 8279 cache = btrfs_lookup_block_group(fs_info, logical); 8280 if (!cache) 8281 return 0; 8282 8283 spin_lock(&cache->lock); 8284 if (cache->relocating_repair) { 8285 spin_unlock(&cache->lock); 8286 btrfs_put_block_group(cache); 8287 return 0; 8288 } 8289 cache->relocating_repair = 1; 8290 spin_unlock(&cache->lock); 8291 8292 kthread_run(relocating_repair_kthread, cache, 8293 "btrfs-relocating-repair"); 8294 8295 return 0; 8296 } 8297